3 Data Analytics Report

Keywords

Occupational Accidents, Workplace Accidents, Accidents at work, Workplace injuries, Determinants, Factors, Cost, Occupational Safety, Occupational Risk, Commuting Accidents, Accident Frequency, Accident Severity

3.1 Representativeness of the employer data

Unlike National Social Security Office (NSSO) and Federaal agentschap voor beroepsrisico’s (FEDRIS), two Belgian federal services, Liantis does not have access to employment and Occupational Accident (OA) data of the full Belgian population. Therefore, we need to assess the representativeness of the Liantis employer data in comparison to the full Belgian employer population.

Since Liantis has access to basic Crossroads Bank for Enterprises (KBO in Dutch) (CBE) information for all Belgian employers (e.g. Belgian version of the the Europese activiteitennomenclatuur (NACE) (NACE-BEL) 2008 classification and municipality per CBE number) it is possible to make the comparison with the Liantis External Service for Prevention and Protection at work (EDPB in Dutch) (ESPP) and Payroll Services (PS) mutual employers, for which the same information is present together with employment and OA data. This will allow us to make a representative sample of the Liantis mutual customers and to correct for representativeness in the further analysis when needed.

As the Liantis PS and ESPP mutual customers are primarely located in Flanders, we will focus on the representativeness of the Liantis mutual customers in Flanders.

We start with loading a summary table of the NACE-BEL 2008 structure. The table (in long format) contains all codes (in 5 levels) with their corresponding labels in four languages (Dutch, French, English and German).

level 1: sections, of which there are 21, identified by alphabetical letters (A to U)
level 2: divisions, of which there are 88, identified by two-digit numerical codes ranging from 01 to 99
level 3: groups, of which there are 272, identified by three-digit numerical codes ranging from from 011 to 990
level 4: classes, of which there are 615, identified by four-digit numerical codes ranging from 0111 to 9900
level 5: subclasses, of which there are 943, identified by five-digit numerical codes ranging from 01110 to 99000

We restructure this table into a hierarchical (broad) format. This will allow us to easily join more general codes (level 3 and level 1) to more specific codes (level 5) present in our data.

Table 3.1 shows a sample of ten rows of the codes and labels at level 5,3 and 1.

Table 3.1: Sample of ten rows of the NACEBEL-2008 level 5, 3 and 1 codes

L5_CD	L3_CD	L3_EN	L1_CD	L1_EN
01290	012	Growing of perennial crops	A	Agriculture, forestry and fishing
15110	151	Tanning and dressing of leather; manufacture of luggage, handbags,saddlery and harness; dressing and dyeing of fur	C	Manufacturing
27900	279	Manufacture of other electrical equipment	C	Manufacturing
38211	382	Waste treatment and disposal	E	Water supply; sewerage; waste management and remediation activities
46710	467	Other specialised wholesale	G	Wholesale and retail trade; repair of motor vehicles and motorcycles
47230	472	Retail sale of food, beverages and tobacco in specialised stores	G	Wholesale and retail trade; repair of motor vehicles and motorcycles
47252	472	Retail sale of food, beverages and tobacco in specialised stores	G	Wholesale and retail trade; repair of motor vehicles and motorcycles
45202	452	Maintenance and repair of motor vehicles	G	Wholesale and retail trade; repair of motor vehicles and motorcycles
46120	461	Wholesale on a fee or contract basis	G	Wholesale and retail trade; repair of motor vehicles and motorcycles
77296	772	Renting and leasing of personal and household goods	N	Administrative and support service activities

In a next step, we load a summary table ‘datam’ in which all Belgian municipalities with corresponding postal code (see Section 2.7.3.3) are mapped to their Province and Region.

Once we have the sector (NACE-BEL 2008 codes) and geographical (postal codes, Nationaal Instituut voor de Statistiek (NIS) codes & municipalities) classification systems available, we load a table in which all Belgian active employers with NSSO activities (paid employees) are present with their CBE number, NACE-BEL 2008 level 5 code and postal code. This information is obtained trough the Liantis CBE-COPY-SERVICE which synchronizes a few basic elements from the CBE register to the Liantis data lake. Data from NACE-BEL 2008 classification system ‘nacesum’ as wel as the geographical information ‘datam’ are left joined. The object was generated and stored 09/04/2025 with the name ‘crbsnapshot’. See ‘functions/constructionCBE_datasource.R’ for details.

We load the object, filter on Flanders, and group by province and level 3 NACE-BEL 2008 code to count the number of employers per province and level 3 NACE-BEL 2008 code.

Table 3.2 shows the top ten rows of this counts as an example.

Table 3.2: Number of active Belgian companies with NSSO activities (ncbe) per Flemish Province per NACEBEL-2008 level 3 sector (top 10 rows)

Region	Province	L3_CD	ncbe
Vlaanderen	Antwerpen	011	293
Vlaanderen	Antwerpen	012	97
Vlaanderen	Antwerpen	013	67
Vlaanderen	Antwerpen	014	267
Vlaanderen	Antwerpen	015	42
Vlaanderen	Antwerpen	016	82
Vlaanderen	Antwerpen	021	6
Vlaanderen	Antwerpen	022	5
Vlaanderen	Antwerpen	024	1
Vlaanderen	Antwerpen	031	1

We take the Liantis ESPP and PS mutual customers and classify them into two segments: smaller companies up to 49 employees and larger companies from 50 employees on. The result is shown in Table 3.3.

Table 3.3: Number and percentage of mutual Liantis ESPP and PS employers by size (<50 or >=50 employees)

segment	n	percentage
small	47394	99.1
large	426	0.9

In the Liantis dataset of mutual customers, only 0.9% of mutual employers are considered large employers. This is lower than the 3.4% calculated from the number of employers, defined under “De werkgever” in the last quarter of 2023, as data from ‘Employment_valAANTALW_NL_20234.xlsx’ (downloaded from the archive, sheet 2 private sector, show (<50 (221545) and \(\geq\) 50 (7831), Figure 3.1). Hence, it will also be a good idea to include also this criterion in the representativeness check.

Figure 3.1: Number of employers with less than or 50 and more employees

Within the mutual customers, we clean the NACE-BEL 2008 codes where necessary (removal of dots, adding of leading zero’s), join the structured NACE-BEL 2008 information and the geographical information. Subsequently we filter on Flanders (Liantis PS has a much smaller portion of customers in Wallonia and Brussels than Liantis ESPP).

If we look at the number of Liantis mutual ESPP and PS customers per Flemish province and NACE-BEL 2008 level 1 sector in Table 3.4, we see that there are a few sectors with very few customers (B, D, E, O, T & U). If we want to report on employers distributed representatively across Flemish provinces and NACE-BEL 2008 level 1 sectors, those sectors will need to be filtered out because of lack of representativeness.

Table 3.4: Number of Liantis mutual ESPP customers per Flemish province and sector

	Antwerpen	Limburg	Oost-Vlaanderen	Vlaams-Brabant	West-Vlaanderen
A	86	76	85	60	458
B	0	2	0	0	0
C	372	404	390	247	1396
D	3	2	0	2	3
E	12	16	12	7	57
F	1287	1187	1007	879	3134
G	1603	1442	1219	1083	3747
H	263	172	239	192	552
I	1340	963	1034	940	3151
J	166	100	157	141	219
K	212	132	193	158	491
L	143	131	104	112	448
M	539	436	461	365	1117
N	357	265	321	302	885
O	0	5	8	1	8
P	65	55	80	33	198
Q	312	294	265	232	776
R	169	122	136	108	284
S	471	372	381	307	986
T	29	15	13	23	56
U	2	0	0	0	0

Data Quality Alert: some (primarly public) sectors have to be filtered out from further analysis because of lack of representativeness

The combined Belgian workforce and employer presence in sectors B, D, E, O, U, and T is predominantly associated with the public sector. This is primarily due to the substantial representation of sectors O (public administration) and U (international organisations), despite the partially private nature of sectors B, D, and E. Given that Liantis primarily serves private sector employers, it is unsurprising that these sectors are underrepresented in the Liantis database. This underrepresentation may introduce bias if these sectors are included in the analysis. Therefore, to safeguard the accuracy and reliability of our findings, we have chosen to exclude these sectors from further analysis.

After filtering out the sectors B, D, E, O, T, U (respectively Mining and quarrying; Electricity, gas, steam and air conditioning supply; Water supply; sewerage; waste management and remediation activities; Public administration and defence; compulsory social security; Activities of households as employers; undifferentiated goods- and services-producing activities of households for own use; Activities of extraterritorial organisations and bodies), no missing NACE-BEL 2008 level 1 or postal codes remain.

An overview of NACE-BEL 2008 level 1 sectors labels and there meaning (in English and Dutch) can be found in Table 3.5.

Table 3.5: NACE-BEL 2008 level 1 values and corresponding labels

NID1	SECTIONS	SECTIES
A	Agriculture, forestry and fishing	Landbouw, bosbouw en visserij
B	Mining and quarrying	Winning van delfstoffen
C	Manufacturing	Industrie
D	Electricity, gas, steam and air conditioning supply	Productie en distributie van elektriciteit, gas, stoom en gekoelde lucht
E	Water supply; sewerage; waste management and remediation activities	Distributie van water; afval- en afvalwaterbeheer en sanering
F	Construction	Bouwnijverheid
G	Wholesale and retail trade; repair of motor vehicles and motorcycles	Groot- en detailhandel; reparatie van auto’s en motorfietsen
H	Transportation and storage	Vervoer en opslag
I	Accommodation and food service activities	Verschaffen van accommodatie en maaltijden
J	Information and communication	Informatie en communicatie
K	Financial and insurance activities	Financiële activiteiten en verzekeringen
L	Real estate activities	Exploitatie van en handel in onroerend goed
M	Professional, scientific and technical activities	Vrije beroepen en wetenschappelijke en technische activiteiten
N	Administrative and support service activities	Administratieve en ondersteunende diensten
O	Public administration and defence; compulsory social security	Openbaar bestuur en defensie; verplichte sociale verzekeringen
P	Education	Onderwijs
Q	Human health and social work activities	Menselijke gezondheidszorg en maatschappelijke dienstverlening
R	Arts, entertainment and recreation	Kunst, amusement en recreatie
S	Other service activities	Overige diensten
T	Activities of households as employers; undifferentiated goods- and services-producing activities of households for own use	Huishoudens als werkgever; niet-gedifferentieerde productie van goederen en diensten door huishoudens voor eigen gebruik
U	Activities of extraterritorial organisations and bodies	Extraterritoriale organisaties en lichamen

Since we had customer data and CBE data available at NACE-BEL 2008 level 3 and postal code, but we want to take a representative sample at NACE-BEL 2008 level 1 and Flemish province, we further summarise our employer data to level 1 and Province.

The first ten rows of the result of this summarization are shown in Table 3.6.

Table 3.6: NACE-BEL 2008 level 1 counts per province (CBE and Liantis mutual customers)

Region	Province	NID1	L1_EN	nLiantis	nCBE
Vlaanderen	Antwerpen	A	Agriculture, forestry and fishing	86	859
Vlaanderen	Antwerpen	C	Manufacturing	370	2200
Vlaanderen	Antwerpen	F	Construction	1287	4783
Vlaanderen	Antwerpen	G	Wholesale and retail trade; repair of motor vehicles and motorcycles	1603	8797
Vlaanderen	Antwerpen	H	Transportation and storage	263	1951
Vlaanderen	Antwerpen	I	Accommodation and food service activities	1340	4676
Vlaanderen	Antwerpen	J	Information and communication	165	1536
Vlaanderen	Antwerpen	K	Financial and insurance activities	212	1243
Vlaanderen	Antwerpen	L	Real estate activities	143	1053
Vlaanderen	Antwerpen	M	Professional, scientific and technical activities	539	3828

Subsequently, we calculate the necessary proportions to be able compile a list from which we can randomly draw a sample.

propLiantis is the percentage of companies within Liantis in the combination of Flemish province and NACE-BEL 2008 1 level
propCBE is the percentage of companies within the CBE in the combination of Flemish province and NACE-BEL 2008 1 level
ratioSampleCBE is the ratio of the proportions within the sample and the CBE and should be around 1 to be representative. Ratios much larger than 1 indicate an over-representation in our customer base, ratios much smaller than 1 indicate an under-representation in our customer base
if we propose to construct a representative sample of 10000 employers, we can calculate the number of companies to be sampled per province and NACE-BEL 2008 level 1 sector combination (and size <50 or \(\geq\) 50 employees).

This gives us the necessary proportions and numbers to sample over the combinations of Flemish provinces and NACE-BEL 2008 1 levels. The first and last five rows as shown as an example in Table 3.7.

Table 3.7: Numbers and proportions needed to construct a representative sample (first and last five rows)

Province	NID1	nLiantis	nCBE	nSample	propLiantis	propCBE	propSample	ratioLiantisCBE	ratioSampleCBE	sampleOK
West-Vlaanderen	I	3151	3612	274	0.074	0.027	0.027	2.7	1	TRUE
West-Vlaanderen	S	986	1115	84	0.023	0.008	0.008	2.7	1	TRUE
West-Vlaanderen	P	198	245	19	0.005	0.002	0.002	2.5	1	TRUE
West-Vlaanderen	F	3134	4134	313	0.074	0.031	0.031	2.3	1	TRUE
West-Vlaanderen	G	3747	5829	442	0.088	0.044	0.044	2.0	1	TRUE
Vlaams-Brabant	A	60	473	36	0.001	0.004	0.004	0.4	1	TRUE
Vlaams-Brabant	P	33	258	20	0.001	0.002	0.002	0.4	1	TRUE
Antwerpen	A	86	859	65	0.002	0.007	0.007	0.3	1	TRUE
Antwerpen	J	165	1536	116	0.004	0.012	0.012	0.3	1	TRUE
Limburg	A	75	759	57	0.002	0.006	0.006	0.3	1	TRUE

Next, we generate the file ‘datas’ with the representative sample of employers and save it.

In the table Table 3.8 we show the result of the sampling procedure on the mutual customers.

Table 3.8: Summary of counts and percentages of number of employers in Flemish provinces and NACE-BEL 2008 level 1 sectors in the CBE and the sample

variable	Province	A	C	F	G	H	I	J	K	L	M	N	P	Q	R	S
nKBO	Antwerpen	862.00	2376.00	4783.00	8797.00	1968.00	4676.00	1550.00	1253.00	1053.00	3828.00	1986.00	431.00	2070.00	1090.00	1554.00
nSAM	Antwerpen	63.00	164.00	360.00	656.00	147.00	348.00	113.00	91.00	80.00	284.00	149.00	35.00	156.00	83.00	112.00
percKBO	Antwerpen	0.65	1.78	3.59	6.61	1.48	3.51	1.16	0.94	0.79	2.87	1.49	0.32	1.55	0.82	1.17
percSAM	Antwerpen	0.63	1.64	3.60	6.56	1.47	3.48	1.13	0.91	0.80	2.84	1.49	0.35	1.56	0.83	1.12
nKBO	Limburg	759.00	1402.00	2849.00	3788.00	598.00	2053.00	414.00	583.00	305.00	1449.00	802.00	200.00	1008.00	386.00	753.00
nSAM	Limburg	57.00	100.00	208.00	282.00	46.00	151.00	31.00	44.00	23.00	109.00	62.00	19.00	74.00	29.00	54.00
percKBO	Limburg	0.57	1.05	2.14	2.84	0.45	1.54	0.31	0.44	0.23	1.09	0.60	0.15	0.76	0.29	0.57
percSAM	Limburg	0.57	1.00	2.08	2.82	0.46	1.51	0.31	0.44	0.23	1.09	0.62	0.19	0.74	0.29	0.54
nKBO	Oost-Vlaanderen	709.00	2207.00	4796.00	6092.00	1275.00	3121.00	926.00	1032.00	645.00	2773.00	1473.00	349.00	1659.00	838.00	1256.00
nSAM	Oost-Vlaanderen	50.00	159.00	352.00	458.00	98.00	233.00	69.00	75.00	47.00	207.00	104.00	30.00	123.00	61.00	92.00
percKBO	Oost-Vlaanderen	0.53	1.66	3.60	4.57	0.96	2.34	0.70	0.77	0.48	2.08	1.11	0.26	1.25	0.63	0.94
percSAM	Oost-Vlaanderen	0.50	1.59	3.52	4.58	0.98	2.33	0.69	0.75	0.47	2.07	1.04	0.30	1.23	0.61	0.92
nKBO	Vlaams-Brabant	473.00	918.00	2982.00	4628.00	1284.00	2490.00	882.00	565.00	475.00	2096.00	1224.00	272.00	1141.00	524.00	1040.00
nSAM	Vlaams-Brabant	35.00	60.00	221.00	344.00	96.00	182.00	61.00	42.00	35.00	152.00	87.00	20.00	91.00	40.00	77.00
percKBO	Vlaams-Brabant	0.36	0.69	2.24	3.47	0.96	1.87	0.66	0.42	0.36	1.57	0.92	0.20	0.86	0.39	0.78
percSAM	Vlaams-Brabant	0.35	0.60	2.21	3.44	0.96	1.82	0.61	0.42	0.35	1.52	0.87	0.20	0.91	0.40	0.77
nKBO	West-Vlaanderen	1170.00	2538.00	4134.00	5829.00	1079.00	3612.00	472.00	1003.00	796.00	2129.00	1352.00	245.00	1336.00	596.00	1124.00
nSAM	West-Vlaanderen	85.00	206.00	319.00	442.00	81.00	272.00	33.00	72.00	57.00	162.00	110.00	43.00	158.00	46.00	83.00
percKBO	West-Vlaanderen	0.88	1.91	3.10	4.38	0.81	2.71	0.35	0.75	0.60	1.60	1.02	0.18	1.00	0.45	0.84
percSAM	West-Vlaanderen	0.85	2.06	3.19	4.42	0.81	2.72	0.33	0.72	0.57	1.62	1.10	0.43	1.58	0.46	0.83

The result of the procedure was tested with a Chi-squared test of independence. Effect sizes were labelled following Funder’s (2019) recommendations. The Pearson’s Chi-squared test of independence between sample and CBE suggests that the effect is statistically not significant (\(\chi^2 = 55.59\), \(p = 0.946\)).

From the representative sample of 10000 employers we can extract the CBE number, and subsequently label which employers in the complete dataset of mutual ESPP and PS customers make part of this representative subset (for NACE-BEL 2008 level 1 code, Flemish province and size <50 or \(\geq\) 50 employees) and which do not.

At this point, it is possible to indicate which OAs took place in the representative subset and which took place in other companies as well as to examine which variables which might be influenced. An example is shown in Table 3.9: the seriousness of an OA does not seem to differ between the complement and the representative subset, but biological sex does. Our raw dataset contains a lot more male workers experiencing an OA, but in a representative subset biological sex is more equally balanced. Hence, correcting for representativeness is important when analysing OAs.

Table 3.9: Counts and percentages of the number of occupational accidents by their employer being part of the representative subset of employers, seriousness of the accident or biological sex of the victim

category	OAnotinrep	OAinrep	percOAnotinrep	perOAinrep
total	188998	27948	87.12	12.88
normal oa	121189	23931	95.41	95.54
potential serious oa	1082	219	0.85	0.87
serious oa	4753	899	3.74	3.59
F	44668	11857	35.24	47.39
M	82070	13165	64.76	52.61

A representative sample of 10,000 employers

We were able to draw a representative sample of 10,000 employers based on the following characteristics:

Province: representative across the Flemish Region (excluding Brussels),
Sector: classified according to the NACE level 1 code (excluding sectors B, D, E, O, U, and T),
Company size: categorized as either fewer than 50 employees or 50 employees and more,
Establishment count: the number of companies established within each unique combination of these factors (province × sector × size) was taken into account to ensure proportional representation.

To account for potential differences in outcomes due to representativeness issues, a binary variable indicating whether an observation belongs to the representative sample can be included in the models.

« previous

3.2 Statistical methodology

The data analysis is conducted in two stages. First, descriptive statistics of the determinants are presented. When applicable, external references (e.g., NSSO, FEDRIS) are included. In the second stage, each determinant is modelled in relation to various outcome variables (see Section 1.2). Modelling is done in a univariable (per single determinant, see Section 3.4) as well as a multivariable (all determinants combined, see Section 3.5) context.

3.2.1 Descriptives

Stage one presents descriptive data on determinants related to OAs. External sources, such as NSSO and FEDRIS, provide standardized classifications for determinants attributable to both employees and employers (age, gender, nsso category, sector,…). This data, being publicly accessible through statistical quarter or year reports, is collected and structured into categories to support comparative analysis.

Where applicable, data is presented for all employees, combining public and private sectors, as well as for private sector employees separately. The numbers include total OAs, with a further breakdown into commuting and workplace accidents. When possible, three tables are provided for each category (total, commuting, and workplace accidents), each shown both in absolute numbers and as percentages across categories. For comparability, a representative sample of mutual Liantis ESPP and PS customers was created. Based on census data, companies were sampled by sector (NACE-BEL 2008 section), Flemish province, and employee count (<50 or \(\geq\) 50). All relevant employee records from these firms, over the relevant time period, are included.

The descriptive tables include a maximum of eight columns, each representing a specific set.

category: the various categories of the determinant.
Rijksdienst voor Sociale Zekerheid (RSZ) all: aggregated totals or percentages of employees in both public and private sectors, based on quarterly data from 2014 to 2023 from the NSSO (in Dutch RSZ). This acts as an external reference.
RSZ private: aggregated totals or percentages of employees in the private sector, based on quarterly data from 2014 to 2023 from the NSSO (in Dutch RSZ. This acts as an external reference.
Liantis all: aggregated totals or percentages of employees from Liantis client employers, specifically those using both PS and ESPP services in the private sector, based on monthly data from 2014 to 2023.
FEDRIS All: notifications of OAs in the private sector, as reported to FEDRIS, combining commuting and workplace accidents, aggregated over the years 2014 to 2023. This acts as an external reference.
TotalNot: OA notifications among all Liantis customers using ESPP services in the private sector, aggregated over the years 2014 to 2023.
TotalNotMut: OA notifications among mutual Liantis customers using both PS and ESPP services in the private sector, aggregated over the years 2014 to 2023.
TotalNotMutRep: OA notifications among mutual Liantis customers using both PS and ESPP services in the private sector, limited to a representative sample of employers (by Flemish province, sector, and company size), aggregated over the years 2014 to 2023.

3.2.2 Modelling

3.2.2.1 Datasets

The datasets used for modelling include variables such as company identifiers and individual identifiers, year, month, and the respective outcome variable. All statistical models were constructed using the glmmTMB package (Brooks et al., 2017) using approximations to estimate the Intraclass Correlation Coefficient (ICC) where necessary, based on methods proposed by Nakagawa et al. (2017). Only mutual Liantis ESPP and PS customers are included over the respective timeframe. Observations corresponding to individuals employed by these companies during the relevant time periods are included. A representative sample, aligned with the one used in the descriptive statistics, is present as a subset. An additional variable identifies whether an observation is part of this sample or not, to account for potential differences in outcomes.

It is important to note that the inclusion of this variable to correct for representativeness (as a fixed effect) did not significantly alter the effect sizes or significance levels of the other determinants, suggesting that its presence does not bias the results found.

Fallacies

When interpreting our findings, it is essential to be mindful of the ecological and atomistic fallacies, which can lead to incorrect conclusions if inferences are made at the wrong level of analysis. The ecological fallacy involves applying group-level associations to individuals, while the atomistic fallacy occurs when individual-level relationships are assumed to hold at the group level. These fallacies are particularly relevant in our study, where both individual and organisational-level data are analysed.

For example, at the individual level, we find that younger workers are more likely to experience occupational accidents. However, at the group level, organisations with a higher proportion of younger employees tend to have lower rates of occupational accidents. This apparent contradiction illustrates how relationships can differ, or even reverse, depending on the level of analysis. One possible explanation is that organisations employing more young workers may also have better safety cultures, more dynamic work environments, or more proactive prevention policies, which mitigate risk at the group level despite individual vulnerabilities. This example underscores the importance of aligning interpretations with the appropriate level of data and avoiding simplistic generalizations across levels. Recognizing these fallacies helps ensure that conclusions are both accurate and contextually valid.

« previous

3.2.2.2 Occurence

The first set of models examines the occurrence of OAs. Four models (Model 1, Model 1.1, Model 2 and Model 3) address individual-level outcomes; a fifth one (Model 4) focuses on a company-level outcome. A multilevel modelling approach is generally applied, recognizing the hierarchical structure of the data, where observations are nested within individuals and individuals within companies. Theoretically, this structure would justify a three-level model. However, due to convergence issues encountered during estimation, a two-level model is used instead, with both individual and company-level clustering treated at the same level. For the fifth model (Model 4), a multilevel modelling approach for a company-level ouctome is employed to account for repeated yearly observations at company-level.

Model 1 (hadOA): likelihood to notify an OA, both commuting and workplace notifications. Each observation represents a specific month for an individual and indicates whether an OA notification was reported during that period in relation to the determinant.
Model 1.1 (statusOA): likelihood an OA insurer refuses an OA, both commuting and workplace notifications. Each observation represents whether a specific OA notification was refused in relation to the determinant. This dataset is a subset of the dataset used in Model 1 (only workers with OA notifications).
Model 2 (comOA): likelihood to experience an accepted commuting OA. Each observation represents a specific month for an individual and indicates whether an accepted OA notification during commuting was noted during that period in relation to the determinant. This dataset is similar to the dataset used in Model 1, but workplace OAs are in this set not counted (since these are non-commuting accidents).
Model 3 (wplOA): likelihood to experience an accepted workplace OA. Each observation represents a specific month for an individual and indicates whether an accepted OA notification for the workplace was noted during that period in relation to the determinant. This dataset is similar to the dataset used in Model 1, but commuting OAs are in this set not counted (since these are non-workplace accidents).
Model 4 (freqOA): this model focuses on a company-level outcome: the frequency degree of OAs. This metric is calculated by dividing the number of accepted workplace OAs by the total hours of exposure, multiplied by a constant (1,000,000). Determinants are expressed as fractions (ranging from 0 to 1) representing the proportion of employees in specific categories. For instance, when age is used as a determinant, the workforce is divided into six age groups, and the model includes the proportion of employees (some categories can be left out due to rank deficiency issues). Each observation corresponds to a company, with repeated yearly measures clustered at the company level.

Five models related to occurrence

A series of five models will address the likelihood and frequency of occupational accidents, at both individual and company levels. A multilevel modelling approach is used to account for the hierarchical structure of the data.

« previous

3.2.2.3 Severity

The second group of models examines the severity of OAs using individual-level data. The first four models focus on (monthly) individual outcomes, the last two on a (yearly) company level outcome.

The first model (Model 5) models OA seriousness in a two-level structure, treating individual and company clustering equivalently.

The next two (Model 6 and Model 6.1) model absence days. To address the count nature of this number of days as well as overdispersion, a negative binomial distribution is used. Specifically, the NB2 parametrization is applied, where the conditional variance increases quadratically with the mean. This approach offers greater flexibility in modelling scenarios where overdispersion grows with the expected value, making it more appropriate than a Poisson model in this context.

For the fourth model (Model 7), a multilevel modelling approach is applied to account for clustering at the company level, reflecting repeated observations within the same organisational context.

The last two (Model 8 and Model 8.1) shift to company-level outcomes and use multilevel modelling to incorporate repeated yearly observations of company variables. Determinants in this models are operationalized as fractions (ranging from 0 to 1) of employees falling into specific categories. For example, when using age as a determinant, the workforce is divided into six age groups, and the model includes the proportion of employees in each group per company. Observations are thus companies with their fractions. In this model the observations are clustered on a company level to account for repeated measures through time (in maximum the 10 different years of the study between 2014 and 2023).

Model 5 (seriousOA): chance on a severe accepted workplace OA (see Section 2.10). The binary outcome indicates whether a workplace accident was classified as severe in relation to the determinant. Only workplace OA are considered, as commuting accidents are (according to Belgian law) never to be regarded as serious Liantis database information (in certain cases modified data in comparison with the original relevant notification fields) is used in combination with the original commuting information from the notification the decide on seriousness. Each observation represents a single OA.
Model 6 (nDaysTAO): duration of absence. This outcome captures the (validated) number of days an individual is absent due to an OA. Absence duration is operationalized in two ways. This first way reflects validated days of absence, as reported by FEDRIS (and, by extension, insurers), representing officially recognized durations of absence.
Model 6.1 (lengtecor): duration of absence. This outcome captures the (calculated) number of days an individual is absent due to an OA. Absence duration is operationalized in two ways. This second way reflects duration of absence by using Liantis PS data calculating absence days based on wage codes linked to OAs, offering an alternative measure of absence duration.
Model 7 (costOA): cost of OAs. This cost is calculated based on the actual wage expenses incurred during the period of absence, as recorded in payroll data. Each observation represents a single OA, with the corresponding wage cost for the absence period. Since the costs are higly skewed to the right, a log10-transformation is applied to the original cost + €1 (to avoid minus infinity results) to normalize the distribution.
Model 8 (sevOAval): severity degree (validated days). This collective measure reflects the proportion of the sum of the number of lost calendar days from employees within a company who experienced a workplace OA. The sum of the total number lost calendar days is calculated in two ways. This first way uses the sum of the validated days of absence as reported by FEDRIS (and, by extension, insurers), representing officially recognized durations of absence.
Model 8.1 (sevOAcal): severity degree (calculated days). This collective measure reflects the proportion of the sum of the number of lost calendar days from employees within a company who experienced an workplace OA. The sum of the total number lost calendar days is calculated in two ways. This second way to operationalize the severity degree is by using payroll-reported days of absence based on wage codes provided through Liantis PS data, offering an alternative measure of absence duration.

Six models related to severity

Six models will address severity of occupational accidents at both individual and company levels. They examine the likelihood of severe accidents, duration of absence, and associated costs, using validated and calculated days absent. A multilevel modelling approach is used to account for the hierarchical structure of the data.

« previous

3.3 Outcomes

The outcomes analyzed in this study encompass a range of indicators related to OA and incidents, including:

notifications;
accepted accidents;
commuting accidents;
workplace accidents;
frecuency degree;
seriousness of OA;
length of absence following an OA;
cost of OA;
severity degree.

These outcomes are discussed in detail in Section 3.5, where they are integrated into multivariable models. A key consideration is that many of these outcomes are inherently tied to a temporal dimension (e.g. amount of notifications in a certain month of a given year). As such, providing standalone descriptive statistics outside the modelling context would be both methodologically limiting and potentially misleading.

Additionally, several outcomes, particularly frequency degree and severity degree, are derived measures that rely on assumptions about exposure. These assumptions can vary significantly depending on the chosen methodology. For instance, exposure hours may be estimated using a flat calculation (e.g., standard definition) or derived from actual payroll data, which reflects real-time presence and activity. These methodological choices directly influence the distribution and interpretation of the outcomes, and thus must be considered within the modelling framework.

Another important factor is data completeness. Different models exclude varying subsets of data due to missing values in key variables. This means that the distribution of the final modelled outcomes may differ from the raw distributions, further complicating the presentation of general descriptives.

Taken together, these considerations, temporal dependencies, methodological variability in derived measures, and data exclusions, have led to the decision not to present standalone descriptive statistics for these outcomes in this section. Instead, the outcomes are more appropriately interpreted within the context of the multivariable models, where these complexities are accounted for.

For readers who still like to find some more detailed descriptive statistics on some of the outcomes, we refer to Section 3.4.5.1. In the descriptives section of this univariable analysis by year, counts and percentages are presented for each of the ten years of the study.

Counting and reporting a number of OA per year might not be as simple as it seems

An apparently simple parameter like the number of OA per year can be counted in very different ways:

a total number of notifications as a proxy for all OAs (last declarations, before any acceptance or refusal, as well commuting as workplace accidents)
not all those declarations of OAs will be accepted by the insurers
accepted OAs are further classified for reporting in
- commuting accidents (which are by law always defined as “normal” accepted accidents)
- workplace accidents (which may be “normal”, “severe” or “very severe” accepted accidents with different types of consequences)
from the accepted workplace accidents, the accidents with temporary (an absence of at least one calendar day excluding the day of the workplace accidents), permanent of fatal consequences are counted and used to calculate frequency and severity degrees

Rather than reporting many detailed descriptives here, we would like to focus on the modelling of these outcomes in Section 3.5. For readers who still like to find some more detailed counts per year, we refer to Section 3.4.5.1.

« previous

3.4.1 Individual factors

3.4.1.1 Age

3.4.1.1.1 Descriptives

Age categories are provided in the NSSO and FEDRIS statistical reports. See Section 2.16 to Section 2.19 for more information. Since the FEDRIS categories are broader, this categorisation was chosen.

Age categories are not directly available in Liantis PS and Liantis ESPP data, nor in the FEDRIS notifications. So, these categories were calculated from the Identificatienummer Sociale Zekerheid (rijksregisternummer of BIS-registernummer) (INSZ) number of the employee. BIS numbers were excluded. See Section 2.7.1 for more information.

Since age categories are not provided in the notifications, the fit with the present categorizationbased on the INSZ number cannot be made.

In the next 6 tables, an overview of all (accepted) OAs (Table 3.10 and Table 3.11), the commuting OAs (Table 3.12 and Table 3.13) and the workplace OAs (Table 3.14 and Table 3.15) are given.

In the first tables (Table 3.10, Table 3.12 and Table 3.14), the absolute numbers are provided stratified per age category, next to the total number of employees, as available in the PS Liantis database, and the total number of employees provided by the NSSO (a separate column is given for the private sector).

In the second tables (Table 3.11, Table 3.13 and Table 3.15), the relative numbers of the OAs per age category are provided as percentages. See Section 3.2 for more information.

From the Tables Table 3.11, Table 3.13 and Table 3.15, an overrepresentation of workers in the age category 20-29 years can be noticed in the OA notifications. For instance, in Table 3.11, workers between 20-29 years account for about 19% of the workforce, while they account for 26-28% of all occupational notifications. The same pattern can be noticed for the commuting (Table 3.13) and workplace accidents (Table 3.15).

all accepted occupational accidents

Table 3.10: Numbers of employees per age category (nsso and liantis) and numbers of all occupational accident notifications per age category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
15 - 19	1138108	1081215	492930	40020	4476	1906	673
20 - 29	28627755	23214645	3856018	377200	50440	17289	7088
30 - 39	39264068	29622113	4044646	340298	48228	14764	6063
40 - 49	38802753	28465193	3863957	304885	45385	13503	5904
50 - 59	36769043	25920567	3647163	254074	40279	11718	5444
60 and older	8566415	6131093	1277762	36900	5655	1822	779

Table 3.11: Percentages of employees per age category (nsso and liantis) and percentages of all occupational accident notifications per age category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
15 - 19	0.74	0.94	2.87	2.96	2.30	3.12	2.59
20 - 29	18.69	20.29	22.44	27.87	25.94	28.34	27.31
30 - 39	25.63	25.89	23.54	25.14	24.80	24.20	23.36
40 - 49	25.33	24.87	22.49	22.53	23.34	22.14	22.75
50 - 59	24.01	22.65	21.23	18.77	20.71	19.21	20.98
60 and older	5.59	5.36	7.44	2.73	2.91	2.99	3.00

commuting accepted occupational accidents

Fedris changed its reporting methodology in 2023 and gives since that year statistics for the public and private sector together. Commuting accidents for the private sector separately by age category were not directly available in the published public statistics but were obtained through personal communication with the FEDRIS Stats team.

Table 3.12: Numbers of employees per age category (nsso and liantis) and numbers of commuting occupational accident notifications per age category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
15 - 19	1138108	1081215	492930	5689	732	348	130
20 - 29	28627755	23214645	3856018	61418	7016	2334	1064
30 - 39	39264068	29622113	4044646	56851	6970	2065	972
40 - 49	38802753	28465193	3863957	49861	6373	1810	927
50 - 59	36769043	25920567	3647163	44985	6191	1696	848
60 and older	8566415	6131093	1277762	7506	991	301	152

Table 3.13: Percentages of employees per age category (nsso and liantis) and percentages of commuting occupational accident notifications per age category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
15 - 19	0.74	0.94	2.87	2.51	2.59	4.07	3.18
20 - 29	18.69	20.29	22.44	27.14	24.82	27.29	26.00
30 - 39	25.63	25.89	23.54	25.12	24.65	24.14	23.75
40 - 49	25.33	24.87	22.49	22.03	22.54	21.16	22.65
50 - 59	24.01	22.65	21.23	19.88	21.90	19.83	20.72
60 and older	5.59	5.36	7.44	3.32	3.51	3.52	3.71

workplace accepted occupational accidents

Table 3.14: Numbers of employees per age category (nsso and liantis) and numbers of workplace occupational accident notifications per age category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
15 - 19	1138108	1081215	492930	34331	3744	1558	543
20 - 29	28627755	23214645	3856018	315782	43424	14955	6024
30 - 39	39264068	29622113	4044646	283447	41258	12699	5091
40 - 49	38802753	28465193	3863957	255024	39012	11693	4977
50 - 59	36769043	25920567	3647163	209089	34088	10022	4596
60 and older	8566415	6131093	1277762	29394	4664	1521	627

Table 3.15: Percentages of employees per age category (nsso and liantis) and percentages of workplace occupational accident notifications per age category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
15 - 19	0.74	0.94	2.87	3.05	2.25	2.97	2.48
20 - 29	18.69	20.29	22.44	28.02	26.13	28.51	27.56
30 - 39	25.63	25.89	23.54	25.15	24.83	24.21	23.29
40 - 49	25.33	24.87	22.49	22.63	23.47	22.29	22.77
50 - 59	24.01	22.65	21.23	18.55	20.51	19.11	21.03
60 and older	5.59	5.36	7.44	2.61	2.81	2.90	2.87

3.4.1.1.2 Models

The following Table 3.16 summarizes the results of the models assessing the relationship between age categories and the nine outcomes. Below the table, you can find the data on which each model is based. The results of the nine models are then presented both graphically and in a table.

Table 3.16: Overview of the results of the different models examining the relation between age and Occupational Accidents

catage	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
<=19	REF	REF	REF	REF	ns	REF	REF	REF	ns
20-29	>	ns	ns	>	>	ns	>/>	>	ns
30-39	>	ns	ns	>	>	ns	>/>	>	>/>
40-49	ns	ns	ns	>	>	ns	>/>	>	ns/>
50-59	ns	ns	ns	ns	ns	ns	>/>	>	>/ns
>=60	<	ns	ns	<	<	ns	>/>	>	ns

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Age and occupational accidents

Workers from the age group 20-29 have a higher chance for a workplace occupational accident, while workers older than 60 years have a lower chance (in comparison with the reference category). There was no association between age category and the chance for an accepted commuting accident.
There was no relation between age category and the chance for an occupational accident to be categorized as severe.
If a company has a higher proportion of workers aged 60 and above, it may have a lower frequency degree of occupational accidents.
The number of days off and the wage cost related to an occupational accident increases significantly with age.
A higher proportion of workers aged 30-59 (significantly) increases the severity degree of a company.

« previous

Figure 3.2 shows that workers aged 20-29 (and 30-39) have a 37% (and 14%) higher chance of reporting an OA (both commuting and workplace accidents), while workers aged 60 and above have a 43% lower chance compared to the reference category (under 20), after accounting for significant differences in company representativeness (Section 3.2). The results suggest a trend where older age groups experience lower chance for reporting workplace accidents.

Figure 3.3 shows that age does not significantly determines whether an insurer accepts or rejects an OA claim. This suggests that there is no age-based discrimination in insurance decisions.

The results presented in Figure 3.4 do not support earlier findings from literature showing that workers between 45-65 year have a higher incidence of commuting accidents in comparison with the younger workers.

Figure 3.5 shows that workers aged 20-29 have a 47% higher chance of having an accepted workplace accident, while workers aged 60 and above have a 41% lower chance compared to the reference category (under 20). This underscores the findings of several researchers (Yong Jeong (1999), Takahashi & Miura (2016), Kaur et al. (2023)).

Figure 3.6 (a) and Figure 3.6 (b) present a model estimating the frequency degree of OAs at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures show that a higher proportion of workers in the age categories 20-29, 30-39, and 40-49 significantly increases the frequency degree of OAs within a company. Conversely, a higher proportion of workers aged 60 and older is associated with a significant decrease in the frequency degree of a company. It is important to note that this trend does not completely align with the relationship between age categories and the likelihood of a workplace OA (where a lower chance of a workplace OA is observed from the age of 30). However, for the calculation of the frequency rate of OAs, accidents without consequences are excluded.

Figure 3.7 (a) and Figure 3.7 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that age does not significantly determine whether an workplace OA is classified as serious.

Figure 3.8 (a) and Figure 3.8 (b) are the results of a model assessing the relation between age and the number of absenteeism days. These first 2 figures are based on the absenteeism data provided by FEDRIS (while Figure 3.9 (a) and Figure 3.9 (b) present the same model based on the Liantis PS data). These figures show that age is an important factor in determining the validated number of absenteeism days related to the OA. The number of accepted days off increases significantly with age (e.g., 26% more days for ages 20-29, 88% more days for ages 30-39,…). This aligns with Salminen (2004), which states that older workers experience more severe consequences from OAs.

As mentioned above, Figure 3.9 (a) and Figure 3.9 (b) are the results of the model assessing the relation between age category and number of absenteeism days related to the workplace accident, based on the Liantis PS data. The results are very similar to these of Figure 3.8 above.

Figure 3.10 (a) and Figure 3.10 (b) show the results of the model that is assessing the relation between age category and direct wage cost of an OA are displayed. It indicates that age seems to be a significant factor in determining the direct wage cost of an OA for an employer. This can be attributed to the higher number of absenteeism days in older age groups, as shown in previous models, as well as the higher wages of older employees.

The next 4 figures (Figure 3.11 (a) and Figure 3.11 (b), Figure 3.12 (a) and Figure 3.12 (b)) are displaying the results of a model assessing the relation between proportion of workers within an age category and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.11), the validated number of absenteeism days from FEDRIS was used, while the second model (Figure 3.12) was based on the numbers of days off calculated from the Liantis PS data. These figures show that a higher proportion of workers aged 30-39 and 50-59 significantly increases the severity level within a company.

Figure 3.12 shows that a higher proportion of workers aged 30-39 and 40-49 significantly increases the severity level within a company, based on the number of calculated days from Lantis PS data. These results are generally in line with those from the model, which used validated FEDRIS data.

3.4.1.2 Biological sex

3.4.1.2.1 Descriptives

Both NSSO and FEDRIS provide biological sex categories. Liantis PS and ESPP data do contain biological sex variables, but since the FEDRIS notifications do not contain biological age categories, the variable was calculated from the INSZ number (except bis numbers) of the employee. Since sex categories are absent in the notifications, the fit between the derived biological sex of an employee experiencing an OA cannot be compared with the corresponding sex classification in the notification of the accident.

From the tables Table 3.18, Table 3.20 and Table 3.22, some important findings can be noticed. While male workers account for about 50% of the workforce, they account for 60% of all OA notifications. This is a significant overrepresentation.

all accepted occupational accidents

Table 3.17: Numbers of employees per sex category (nsso and liantis) and numbers of occupational accident notifications per sex category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Mannen	78213055	62065095	8831706	788224	126380	39103	13193
Vrouwen	74955087	52369731	8999383	439553	68405	22048	12821

Table 3.18: Percentages of employees per sex category (nsso and liantis) and percentages of occupational accident notifications per sex category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Mannen	51.06	54.24	49.53	64.2	64.88	63.94	50.71
Vrouwen	48.94	45.76	50.47	35.8	35.12	36.06	49.29

commuting accepted occupational accidents

Table 3.19: Numbers of employees per sex category (nsso and liantis) and numbers of occupational accident notifications per sex category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Mannen	78213055	62065095	8831706	93986	12886	3596	1424
Vrouwen	74955087	52369731	8999383	107551	15482	5001	2687

Table 3.20: Percentages of employees per sex category (nsso and liantis) and percentages of occupational accident notifications per sex category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Mannen	51.06	54.24	49.53	46.63	45.42	41.83	34.64
Vrouwen	48.94	45.76	50.47	53.37	54.58	58.17	65.36

workplace accepted occupational accidents

Table 3.21: Numbers of employees per sex category (nsso and liantis) and numbers of occupational accident notifications per sex category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Mannen	78213055	62065095	8831706	761309	113494	35507	11769
Vrouwen	74955087	52369731	8999383	365728	52923	17047	10134

Table 3.22: Percentages of employees per sex category (nsso and liantis) and percentages of occupational accident notifications per sex category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Mannen	51.06	54.24	49.53	67.55	68.2	67.56	53.73
Vrouwen	48.94	45.76	50.47	32.45	31.8	32.44	46.27

3.4.1.2.2 Models

The following Table 3.23 summarizes the results of the models assessing the relationship between biological sex (Female or Male) and the nine outcomes. Below the table, you can find the data on which each model is based. The results of the nine models are then presented both graphically and in a table.

Table 3.23: Overview of the results of the different models examining the relation between biological sex and Occupational Accidents

catsex	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Female	REF	REF	REF	REF	<	REF	REF	REF	ns
Male	>	ns	<	>	>	>	ns/>	>	>/>

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Biological sex and occupational accidents

We notice a higher occupational accident chance for males. This is only noticed for accepted workplace occupational accidents and not for accepted commuting accidents.
There was no relation between biological sex and the chance for an occupational accident to be categorized as severe.
If a company has a higher proportion of male workers, it may have a higher frequency degree and severity degree of occupational accidents.
The number of days off and the wage cost related to an occupational accident is significantly higher for male workers.

« previous

From Figure 3.13 (a) and Figure 3.13 (b) we learn that workers in the category male have a 76% higher chance to notify an OA (both commuting and workplace accidents) than the female reference category (after accounting for significant differences in representativeness of the company).

From Figure 3.14 (a) and Figure 3.14 (b) we learn that sex does not not seem to be an important determinant for acceptance or refusal of an OA by an insurer.

From Figure 3.15 (a) and Figure 3.15 (b) we learn that the effect of biological sex is significant in the current study. Also the direction of the observed lower chance for males is consistent with literature which does state that women are at higher risk for occupational commuting accidents (Craig et al., 2023; Federal Public Service Health, Food Chain Safety and Environment, 2022; López et al., 2017; Salerno & Giliberti, 2019, 2021).

From Figure 3.16 (a) and Figure 3.16 (b) we learn that male workers have a 103% higher chance on an accepted workplace accident than female workers (after accounting for significant differences in representativeness of the company). This is in line with previous findings (Feijó et al., 2018; Hendricks et al., 2024; Neves & Fonseca, 2023).

From Figure 3.17 (a) and Figure 3.17 (b) we learn that the proportion of male workers has a significant increasing effect on the frequency degree of a company, while the proportion of female workers has a significant decreasing effect on the frequency degree of a company. Even if a higher proportion of male workers in a company is generally associated with a higher chance of OAs, we would not expect this to be attributed to gender ratio alone, but rather due to the types of jobs men and women typically perform. The specific risk would also depend on job roles, industry, and organisational factors, not just gender ratio alone.

From Figure 3.18 (a) and Figure 3.18 (b) we learn that sex does seem to be an important determinant for classifying an workplace OA as serious: men tend to have a 89% higher chance on a serious accident.

Figure 3.19 (a) and Figure 3.19 (b) indicate that biological sex does not appear to be a significant determinant of the validated number of days of work unavailability. This aligns with the broader literature, which reports inconsistent findings regarding the role of sex in recovery duration. In our study, the number of accepted days of absence does not differ significantly between men and women, suggesting that men do not experience longer absences due to workplace accidents, in contrary to the findings of the Medex study (Federal Public Service Health, Food Chain Safety and Environment, 2022). Our results also diverge from those of Fontaneda et al. (2019), who reported longer absences among women. Even if women tend to have longer recovery periods following OAs, particularly in lower-responsibility or female-dominated roles, this pattern likely varies depending on occupation, job responsibilities, and workplace context.

From Figure 3.20 (a) and Figure 3.20 (b) we learn that sex seems to be a determinant for the calculated number of days of unavailability for work from Liantis PS data. The number of calculated days increases significantly for male workers compared to female workers. The results diverge from the results of model6, where validated FEDRIS data are used.

From Figure 3.21 (a) and Figure 3.21 (b) we learn that sex does seem to be an significant determinant for the direct wage cost of an OA for an employer: the cost for male workers is significantly higher. Higher costs for men may seem plausible since they are more likely to work in higher-wage, higher-risk jobs and receive greater wage compensation for risk. This pattern is reinforced by occupational segregation and wage structures across industries.

From Figure 3.22 (a) and Figure 3.22 (b) we learn that the proportion of male workers has a signicant increasing effect on the severity degree of a company.

From Figure 3.23 (a) and Figure 3.23 (b) we learn that the proportion of male workers has a signicant increasing effect on the severity degree of a company using the number of calculated days from Liantis PS data. The results are equivalent to the results of model8 where validated FEDRIS data are used.

3.4.1.3 Work experience (seniority with the employer)

3.4.1.3.1 Descriptives

Literature generally indicates a correlation between limited work experience and a higher incidence of OA. However, since work experience data is often unavailable, seniority (the number of years an employee has worked with a specific employer) is used as a proxy.

Unfortunately, NSSO does not provide data on seniority with the employer, but FEDRIS does. Liantis PS data includes the start and end dates of an employee’s contracts with a specific employer. So, the proxy for experience was derived by calculating the difference between the last day of the last contract and the first day of the first contract with the same employer. This time difference was categorized using the same categories as in the FEDRIS reports and notifications.

Since seniority categories are included in the notifications, the calculated seniority of an employee involved in an OA can be compared with the corresponding seniority classification in the accident notification. The results of this comparison are displayed in Figure 3.24. The correlation between the two datasets is 0.85, which indicates a strong positive relationship. This suggests that the seniority categories derived from Liantis PS data align well with those in FEDRIS data, making them both suitable for further analysis.

Figure 3.24: Concordance between FEDRIS provided and Liantis PS calculated seniority of an employee

In the next 6 tables, an overview of all (accepted) OAs (Table 3.24 and Table 3.25), the commuting OAs (Table 3.26 and Table 3.27) and the workplace OAs (Table 3.28 and Table 3.29) are given. In the first table ((Table 3.24, Table 3.26 and Table 3.28), the absolute numbers are provided stratified per experience category, next to the total number of employees, as available in the Liantis PS database. As mentioned above, the NSSO does not provide information about seniority. In the second table (Table 3.25, Table 3.27 and Table 3.29), the relative numbers of the OAs per seniority category are provided as percentages.

From the tables Table 3.25, Table 3.27 and Table 3.29, some important findings can be noticed. While workers with a seniority less than 1 year account for >33% of employees within liantis customers, they account for <33% of OA notifications.

all accepted occupational accidents

Table 3.24: Numbers of employees per work experience category (nsso and liantis) and numbers of all occupational accident notifications per work experience category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catexp	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
less than 1 year	NA	NA	6337684	389735	41605	15470	5970
between 1 and 4 years	NA	NA	5389165	361530	57288	18908	7366
between 5 and 10 years	NA	NA	3252844	210007	33513	9779	4057
between 11 and 20 years	NA	NA	2120661	155635	25239	6135	2869
21 years or more	NA	NA	955512	97582	16228	3119	1704
unknown	NA	NA	NA	138888	21346	7824	4064

Table 3.25: Percentages of employees per work experience category (nsso and liantis) and percentages of all occupational accident notifications per work experience category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catexp	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
less than 1 year	NA	NA	35.1	32.1	23.9	29.0	27.2
between 1 and 4 years	NA	NA	29.8	29.8	32.9	35.4	33.5
between 5 and 10 years	NA	NA	18.0	17.3	19.3	18.3	18.5
between 11 and 20 years	NA	NA	11.7	12.8	14.5	11.5	13.1
21 years or more	NA	NA	5.3	8.0	9.3	5.8	7.8
unknown	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.26: Numbers of employees per work experience category (nsso and liantis) and numbers of commuting occupational accident notifications per work experience category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catexp	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
less than 1 year	NA	NA	6337684	62403	6319	2438	1093
between 1 and 4 years	NA	NA	5389165	60578	7820	2530	1159
between 5 and 10 years	NA	NA	3252844	35759	4680	1244	590
between 11 and 20 years	NA	NA	2120661	28140	3859	874	442
21 years or more	NA	NA	955512	20404	3036	537	319
unknown	NA	NA	NA	19026	2673	976	509

Table 3.27: Percentages of employees per work experience category (nsso and liantis) and percentages of commuting occupational accident notifications per work experience category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catexp	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
less than 1 year	NA	NA	35.1	30.1	24.6	32.0	30.3
between 1 and 4 years	NA	NA	29.8	29.2	30.4	33.2	32.2
between 5 and 10 years	NA	NA	18.0	17.3	18.2	16.3	16.4
between 11 and 20 years	NA	NA	11.7	13.6	15.0	11.5	12.3
21 years or more	NA	NA	5.3	9.8	11.8	7.0	8.9
unknown	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.28: Numbers of employees per work experience category (nsso and liantis) and numbers of workplace occupational accident notifications per work experience category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catexp	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
less than 1 year	NA	NA	6337684	327332	35286	13032	4877
between 1 and 4 years	NA	NA	5389165	300952	49468	16378	6207
between 5 and 10 years	NA	NA	3252844	174248	28833	8535	3467
between 11 and 20 years	NA	NA	2120661	127495	21380	5261	2427
21 years or more	NA	NA	955512	77178	13192	2582	1385
unknown	NA	NA	NA	119862	18673	6848	3555

Table 3.29: Percentages of employees per work experience category (nsso and liantis) and percentages of workplace occupational accident notifications per work experience category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catexp	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
less than 1 year	NA	NA	35.1	32.5	23.8	28.5	26.6
between 1 and 4 years	NA	NA	29.8	29.9	33.4	35.8	33.8
between 5 and 10 years	NA	NA	18.0	17.3	19.5	18.6	18.9
between 11 and 20 years	NA	NA	11.7	12.7	14.4	11.5	13.2
21 years or more	NA	NA	5.3	7.7	8.9	5.6	7.5
unknown	NA	NA	NA	NA	NA	NA	NA

3.4.1.3.2 Models

The following Table 3.30 summarizes the results of the models assessing the relationship between work experience categories (seniority with the employer) and the nine outcomes. Below the table, you can find the data on which each model is based. The results of the nine models are then presented both graphically and in a table.

Table 3.30: Overview of the results of the different models examining the relation between work experience (seniority with the employer) and Occupational Accidents

catsen	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
less than 1	REF	REF	REF	REF	<	REF	REF	REF	ns/<
1 to 4	ns	ns	<	ns	ns	>	>/>	>	ns
5 to 10	<	ns	<	<	ns	>	>/>	>	ns
11 to 20	<	ns	ns	<	ns	ns	>/>	>	ns
21 and more	<	ns	<	<	ns	ns	>/>	>	ns

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Work experience (seniority with the employer) and occupational accidents

We notice a trend of lower occupational accident chances with higher seniority. This is only noticed for workplace occupational accidents and not for commuting accidents.
There was some evidence that workers in the categories 1-4 and 5-10 years of seniority have a higher chance for an occupational accident to be categorized as serious.
If a company has a higher proportion of workers with less than 1 year seniority, it may have a lower frequency degree of occupational accidents.
The number of days off and the wage cost related to an occupational accident increases significantly with seniority.

« previous

Figure 3.25 (a) and Figure 3.25 (b) shows that workers in with 1-4 years seniority do not have a lower chance of reporting an OA (both commuting and workplace accidents). Workers with higher seniority although do have a lower chance (12% to 39%) compared to the reference category <1 year (after accounting for significant differences in company representativeness). Although research is inconsistent, also previous studies generally report a trend of lower OA rates with increased experience (Breslin, 2006; Jeong, 2021; Morassaei et al., 2012).

From Figure 3.26 (a) and Figure 3.26 (b), it can be concluded that seniority does not not seem to be an important determinant for acceptance or refusal of an OA by an insurer. This is in line with expectations: we do not expect that based on seniority an OA would be accepted or refused.

According to Figure 3.28 (a) and Figure 3.28 (b), seniority does not appear to be a significant factor in determining the likelihood of an accepted commuting accident.

Figure 3.29 (a) and Figure 3.29 (b) demonstrate that workers with more seniority (in categories 5-10 and higher) have a 19% to 36% lower chance (after accounting for significant differences in representativeness of the company) on workplace OA. This is in line with a number of former studies, reporting a lower chance for OAs with higher experience (Breslin, 2006; Jeong, 2021; Morassaei et al., 2012).

Figure 3.30 (a) and Figure 3.30 (b) show that the proportion of workers in the categories <1 year has a significant decreasing effect on the frequency degree of a company. At first sight, this in contrast with expectations. Possible explanations may lie in how both the outcome and the independent variable are operationalized. For calculating the frequency degree, OAs without consequences are excluded. Since younger (and generally less experienced) workers are known to have a higher incidence of OA, but less severe ones, this may partly explain the findings. Another explanation could be the use of the concept of ‘proportion,’ which indirectly considers the size of the company. In a large company, the same number of workers in this category will result in a much lower proportion compared to a small company. A third explanation could be the inconsistency in literature regarding this relationship; since some studies did not find support for this relation between work experience and OA incidence (Monteiro Ferreira et al., 2020).

Figure 3.31 (a) and Figure 3.31 (b) show that work experience seems to be a determinant for classifying a workplace OA as serious, at least for workers with up to ten years of work experience. This is partially in line with findings of Yedla et al. (2020), who suggests that job experience is an important variable to predict the seriousness of an injury. We should keep in mind that the definition of a serious workplace accident in Belgium may have played a role since not every OA with injury is classified as serious. Moreover, there exist also methodological differences with existing literature: we modelled the likelihood of an OA being categorized as serious, whereas previous research mostly focused on the likelihood of an individual worker experiencing a serious accident.

Figure 3.32 (a) and Figure 3.32 (b) are the results of a model assessing the relation between seniority and the number of absenteeism days. These first 2 figures are based on the absenteeism data provided by FEDRIS (while Figure 3.33 (a) and Figure 3.33 (b) present the same model based on the Liantis PS data). These figures show that work seniority is an important factor in determining the validated number of absenteeism days related to the OA.

According to Figure 3.32 (a) and Figure 3.32 (b), the number of accepted days increases significantly with seniority category by 6-9%.

According to Figure 3.33 (a) and Figure 3.33 (b), which present the model results estimating the relationship between seniority category and the calculated number of absenteeism days based on Liantis PS data, the number of absenteeism days increases significantly with seniority, ranging from 19% to 80%.

The observed trend -where a higher seniority corresponds to more days lost from work- aligns with previous research findings (Margolis, 2010; Yedla et al., 2020). Additionally, the slightly larger effect size found in calculations using the Liantis PS data compared to the FEDRIS absenteeism data might be due to the Liantis PS dataset being considerably smaller, with fewer than half the number of employers and less than one-fifth the number of employees.

Figure 3.34 (a) and Figure 3.34 (b) demonstrate that seniority does seem to be a significant determinant for the direct wage cost of an OA for an employer.

This finding logically follows from previous models that established a significant relationship with absenteeism. Additionally, the relationship between seniority and pay level reinforces this result.

The next 4 figures (Figure 3.35 (a), Figure 3.35 (b), Figure 3.36 (a) and Figure 3.36 (b)) are displaying the results of a model assessing the relation between proportion of workers within a seniority category and the severity degree of a company, which is again a variable at the company level. In the first model ((Figure 3.35 (a) and Figure 3.35 (b)), validated number of absenteeism days from FEDRIS were used, while the second model (Figure 3.36 (a) and Figure 3.36 (b)) was based on the numbers of days off in the Liantis PS data.

From Figure 3.35 (a) and Figure 3.35 (b), a non-signicant trend that increasing fractions of people with higher seniority is associated higher severity degrees in companies can be observed.

From Figure 3.36 (a) and Figure 3.36 (b), the same trend, but somewhat clearer than in former model (based on FEDRIS absenteeism days), can be noticed. This is in line with the findings of models 6 and 6.1.

3.4.1.4 Nationality (foreign workers)

3.4.1.4.1 Descriptives

International research demonstrate that foreign workers have an increased risk of (both non-fatal and fatal) occupational injures compared to the native workers (Hargreaves et al., 2019).

We used several variables as proxies to determine ‘immigrant worker’ status. First, the statistical reports from FEDRIS provide data on employee nationality in four categories: Belgium, Neighbouring Country, Other Country, and Undetermined/Unknown. Unfortunately, NSSO does not provide data on employee nationality. The Liantis PS data includes the country of birth and the INSZ (RN/BIS) number, allowing us to categorize employees into the same four categories as above. While the notifications do not provide a country of birth, they do include an INSZ (RN/BIS) number, enabling us to use them at least as a proxy for the nationality categories. Consequently, no direct match can be made between the derived category and the corresponding nationality classification.

In the next 6 tables, an overview of all (accepted) OAs (Table 3.31 and Table 3.32), the commuting OAs (Table 3.33 and Table 3.34) and the workplace OAs (Table 3.35 and Table 3.36) are given. In the first table ((Table 3.31, Table 3.33 and Table 3.35), the absolute numbers are provided stratified per nationality category, next to the total number of employees, as available in the Liantis PS database. As mentioned above, the NSSO does not provide information about nationality. In the second table (Table 3.32, Table 3.34 and Table 3.36), the relative numbers of the OAs per nationality category are provided as percentages.

From the table Table 3.32, it can be noticed that while workers from neighbours countries and other countries account for 2.5% and 7% of employees within Liantis customers, they account for about 3 and 9% of OA notifications.

all accepted occupational accidents

Table 3.31: Numbers of employees per work nationality category (nsso and liantis) and numbers of occupational accident notifications per work nationality category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
nation	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Belgium	NA	NA	15073793	1163793	36778	36778	16272
Neighbour country	NA	NA	454276	64086	1433	1433	455
Other country	NA	NA	1583601	125211	4359	4359	1939
Undetermined/Unknown	NA	NA	944196	287	1310	1310	408

Table 3.32: Percentages of employees per work nationality category (nsso and liantis) and percentages of occupational accident notifications per work nationality category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
nation	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Belgium	NA	NA	83.48	85.99	83.81	83.81	85.31
Neighbour country	NA	NA	2.52	4.74	3.27	3.27	2.39
Other country	NA	NA	8.77	9.25	9.93	9.93	10.17
Undetermined/Unknown	NA	NA	5.23	0.02	2.99	2.99	2.14

commuting accepted occupational accidents

Table 3.33: Numbers of employees per work nationality category (nsso and liantis) and numbers of occupational accident notifications per work nationality category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
nation	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Belgium	NA	NA	15073793	198019	5185	5185	2566
Neighbour country	NA	NA	454276	8244	128	128	45
Other country	NA	NA	1583601	19996	625	625	315
Undetermined/Unknown	NA	NA	944196	51	179	179	68

Table 3.34: Percentages of employees per work nationality category (nsso and liantis) and percentages of occupational accident notifications per work nationality category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
nation	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Belgium	NA	NA	83.48	87.50	84.76	84.76	85.70
Neighbour country	NA	NA	2.52	3.64	2.09	2.09	1.50
Other country	NA	NA	8.77	8.84	10.22	10.22	10.52
Undetermined/Unknown	NA	NA	5.23	0.02	2.93	2.93	2.27

workplace accepted occupational accidents

Table 3.35: Numbers of employees per work nationality category (nsso and liantis) and numbers of occupational accident notifications per work nationality category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
nation	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Belgium	NA	NA	15073793	965774	31593	31593	13706
Neighbour country	NA	NA	454276	55842	1305	1305	410
Other country	NA	NA	1583601	105215	3734	3734	1624
Undetermined/Unknown	NA	NA	944196	236	1131	1131	340

Table 3.36: Percentages of employees per work nationality category (nsso and liantis) and percentages of occupational accident notifications per work nationality category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
nation	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Belgium	NA	NA	83.48	85.69	83.66	83.66	85.24
Neighbour country	NA	NA	2.52	4.95	3.46	3.46	2.55
Other country	NA	NA	8.77	9.34	9.89	9.89	10.10
Undetermined/Unknown	NA	NA	5.23	0.02	2.99	2.99	2.11

3.4.1.4.2 Models

The following Table 3.37 summarizes the results of the models assessing the relationship between nationality categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.37: Overview of the results of the different models examining the relation between nationality and Occupational Accidents

catnation	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Belgian	REF	REF	REF	REF	ns	REF	REF	REF	ns/>
Neighbour	>	ns	ns	>	ns	ns	>/ns	>	ns/ns
Other	>	ns	ns	>	ns	ns	>/ns	>	>/>
Undet	<	ns	<	<	rank def	ns	ns/ns	<	rank def

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Nationality (foreign workers) and occupational accidents

Workers from other countries seem to have a higher likelihood of experiencing workplace accidents in comparison with the nationals.
There was no relation between nationality and the chance for an occupational accident to be categorized as serious.
For the companies, we did not find a relation between proportion of foreign workers and the frequency degree. There was a positive relation between a higher proportion of workers from countries apart from neighbours and a higher severity degree.
The (validated but not calculated) number of days off related to an occupational accident is significantly higher in workers from a neighbour or other country and also reflected in the direct wage cost.

« previous

According to Figure 3.37 (a) and Figure 3.37 (b), workers from another country have a 20% to 22% higher chance to report an OA (both commuting and workplace accidents) than the reference category of Belgian workers (after accounting for significant differences in representativeness of the company). This is in line with literature that highlights that migrants are more often involved in OAs than natives (Ahonen, 2006; Ahonen & Benavides, 2016; Guerreschi et al., 2024; Hargreaves et al., 2019; Salminen, 2011).

From Figure 3.38 (a) and Figure 3.38 (b) it can be concluded that nationality does not seem to be an important determinant for acceptance or refusal of an OA by an insurer. This is in line with expectations: we do not expect that based on nationality an OA would be accepted or refused.

Figure 3.39 (a) and Figure 3.39 (b) shows that workers from other countries do not experience an higher chance for an accepted commuting accident in comparison with nationals. The literature review did not provide a clear conclusion regarding this relationship, nor do our results.

According Figure 3.40 (a) and Figure 3.40 (b), workers from countries show a 20% to 24% higher chance on an accepted workplace OA than the nationals (after accounting for significant differences in representativeness of the company). As previously mentioned, this aligns with the findings from the literature on this topic (Ahonen, 2006; Ahonen & Benavides, 2016; Guerreschi et al., 2024; Hargreaves et al., 2019; Salminen, 2011).

Figure 3.41 (a) and Figure 3.41 (b) present a model estimating the frequency degree of OAs at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures show that the proportion of workers who are nationals, from neighbouring countries, or from other countries is not related to the company’s accident frequency. While one might hypothesize that companies employing many immigrant workers or those with a mother tongue other than Dutch or French might experience more accidents, this analysis did not reveal such a pattern.

According to Figure 3.42 (a) and Figure 3.42 (b), nationality does not appear to be a significant factor in classifying an workplace OA as serious. This contrasts somewhat with literature findings that suggest a higher risk of serious and fatal OAs among migrant workers (Ahonen, 2006; Ahonen & Benavides, 2016; Guerreschi et al., 2024; Hargreaves et al., 2019). However, it’s important to note that our analysis defines a serious accident according to Belgian law, which may differ considerably from definitions used in the literature. Additionally, our analysis focused on workers with accepted OAs (modelling the risk that an OA is classified as serious), whereas the literature often models the likelihood of experiencing a serious OA.

Figure 3.43 (a) and Figure 3.43 (b) are the results of a model assessing the relation between nationality and the number of absenteeism days. These first 2 figures are based on the absenteeism data provided by FEDRIS (while Figure 3.44 (a) and Figure 3.44 (b) present the same model based on the Liantis PS data). These figures show that nationality is an important factor in determining the validated number of absenteeism days related to the OA. A worker from a neighbourhood country has on average a 20% increase in absenteeism days in comparison with the Belgian worker, while workers from other than neighbour countries show 12% more absenteeism days.

The next two figures (Figure 3.44 (a) and Figure 3.44 (b)) showing the calculated number of absenteeism days (from the Liantis PS data) related to the OA however did not confirm the previous significant findings between nationality and validated absenteeism days (from FEDRIS). These inconsistent findings may be due to differences in the study populations between the Liantis PS data and the FEDRIS absenteeism data (e.g. the Liantis PS dataset being considerably smaller).

According to Figure 3.45 (a) and Figure 3.45 (b), workers from other countries incur significantly higher direct wage costs (in comparison with nationals) for employers due to OAs.

The next 4 figures (Figure 3.46 (a), Figure 3.46 (b), Figure 3.47 (a) and Figure 3.47 (b)) are displaying the results of a model assessing the relation between the proportion of workers within a nationality category and the severity degree of a company, which is again a variable at the company level. In the first model ((Figure 3.46 (a) and Figure 3.46 (b)), validated number of absenteeism days from FEDRIS were used for the calculation of the severity degree, while the second model (Figure 3.47 (a) and Figure 3.47 (b)) was based on the numbers of days off in the Liantis PS data.

From Figure 3.46 (a) and Figure 3.46 (b), it can be inferred that a higher proportion of workers from countries other than neighbouring ones is associated with a higher severity degree in companies.

The analysis based on the number of absenteeism days in the Liantis PS data (Figure 3.47 (a) and Figure 3.47 (b)) also shows a significant positive relationship between the proportion of workers from non-neighbouring countries and the severity degree of a company. Additionally, a positive relationship was found for the proportion of Belgian workers. However, these findings may be influenced by differences in the sample populations.

3.4.1.5 Salary level

3.4.1.5.1 Descriptives

We used the day wage as a proxy for salary. The NSSO provides statistics on the day wage of employees in 12 categories, which we regrouped into seven categories (<€50, €50-99, €100-124, €125-149, €150-199, €200-249, \(\geq\) €250). The Liantis PS data includes the hourly wage (derived through the division of the brute wage and the effectively worked hours), which allows us to derive the day wage of employees by multiplying the hourly wage by 7.6 (equivalent to the day wage of a full-time worker). However, the notifications do not provide any salary or wage information of the employee. Since this information is absent in the notifications passed through to Liantis, the comparison with the derived wage of an employee experiencing an OA cannot be made.

A reference category of day wages below €50 should be interpreted with caution

Employees with a day wage of <€50 are likely to be part-time workers, temporary workers, or workers with very low working hours. Therefore, this category may not accurately represent low-wage full-time employees. Caution is advised when interpreting results related to this wage category. It was challenging to join and compare globalised NSSO data with detailed Liantis data. The risk exists that we are trying to compare very different things. The authors do believe that the relative differences between the categories are informative, but caution is needed when interpreting and certain data quality issues will be need to considered. For more details, see Section 3.6 on limitations and strenghts.

« previous

In the next 6 tables, an overview of all (accepted) OAs (Table 3.38 and Table 3.39), the commuting OAs (Table 3.40 and Table 3.41) and the workplace OAs (Table 3.42 and Table 3.43) are given. In the first table ((Table 3.38, Table 3.40 and Table 3.42), the absolute numbers are provided stratified per wage category, next to the total number of employees, as available in the Liantis PS database. In the second table (Table 3.39, Table 3.41 and Table 3.43), the relative numbers of the OAs per wage category are provided as percentages.

From the table Table 3.39, it can be noticed that while workers in the wage category <50 euro, and \(\geq\) 205 euro, account for respectively 19% and 9% of the employees within Liantis customers, they account for 4.4% and 4.6% of the employees within the notifications. The opposite pattern can be noticed for the workers in the wage category 100-124 euro and 150-199 euro: they account for respectively 20 and 12% of the employees within Liantis customers, while they account for 32.9 and 18.3% of the workers within the notifications.

all accepted occupational accidents

Table 3.38: Numbers of employees per wage category (nsso and liantis) and numbers of occupational accident notifications per wage category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catwage	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
< 50	1244217.79	1134679.33	3404294	NA	1910	1910	657
50 - 99	19876144.04	17520360.55	4275137	NA	11059	11059	4455
100 - 124	28050533.65	22741692.40	3625308	NA	14445	14445	5598
125 - 149	24670336.11	18466319.81	2227516	NA	8015	8015	3427
150 - 199	29640188.47	19089420.83	2040008	NA	4775	4775	2733
200 - 249	12467321.44	8014278.56	796203	NA	1653	1653	1110
>= 250	10098764.90	7273469.82	1687400	NA	2023	2023	1094
Unknown	38353.87	37133.66	0	NA	151339	17355	6956

Table 3.39: Percentages of employees per wage category (nsso and liantis) and percentages of occupational accident notifications per wage category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catwage	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
< 50	1.0	1.2	18.9	NA	4.4	4.4	3.4
50 - 99	15.8	18.6	23.7	NA	25.2	25.2	23.4
100 - 124	22.3	24.1	20.1	NA	32.9	32.9	29.3
125 - 149	19.6	19.6	12.3	NA	18.3	18.3	18.0
150 - 199	23.5	20.3	11.3	NA	10.9	10.9	14.3
200 - 249	9.9	8.5	4.4	NA	3.8	3.8	5.8
>= 250	8.0	7.7	9.3	NA	4.6	4.6	5.7
Unknown	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.40: Numbers of employees per wage category (nsso and liantis) and numbers of occupational accident notifications per wage category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catwage	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
< 50	1244217.79	1134679.33	3404294	NA	277	277	113
50 - 99	19876144.04	17520360.55	4275137	NA	1850	1850	802
100 - 124	28050533.65	22741692.40	3625308	NA	1563	1563	717
125 - 149	24670336.11	18466319.81	2227516	NA	889	889	454
150 - 199	29640188.47	19089420.83	2040008	NA	858	858	483
200 - 249	12467321.44	8014278.56	796203	NA	347	347	233
>= 250	10098764.90	7273469.82	1687400	NA	333	333	192
Unknown	38353.87	37133.66	0	NA	22270	2482	1118

Table 3.41: Percentages of employees per wage category (nsso and liantis) and percentages of occupational accident notifications per wage category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catwage	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
< 50	1.0	1.2	18.9	NA	4.5	4.5	3.8
50 - 99	15.8	18.6	23.7	NA	30.2	30.2	26.8
100 - 124	22.3	24.1	20.1	NA	25.6	25.6	23.9
125 - 149	19.6	19.6	12.3	NA	14.5	14.5	15.2
150 - 199	23.5	20.3	11.3	NA	14.0	14.0	16.1
200 - 249	9.9	8.5	4.4	NA	5.7	5.7	7.8
>= 250	8.0	7.7	9.3	NA	5.4	5.4	6.4
Unknown	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.42: Numbers of employees per wage category (nsso and liantis) and numbers of occupational accident notifications per wage category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catwage	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
< 50	1244217.79	1134679.33	3404294	NA	1633	1633	544
50 - 99	19876144.04	17520360.55	4275137	NA	9209	9209	3653
100 - 124	28050533.65	22741692.40	3625308	NA	12882	12882	4881
125 - 149	24670336.11	18466319.81	2227516	NA	7126	7126	2973
150 - 199	29640188.47	19089420.83	2040008	NA	3917	3917	2250
200 - 249	12467321.44	8014278.56	796203	NA	1306	1306	877
>= 250	10098764.90	7273469.82	1687400	NA	1690	1690	902
Unknown	38353.87	37133.66	0	NA	129069	14873	5838

Table 3.43: Percentages of employees per wage category (nsso and liantis) and percentages of occupational accident notifications per wage category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catwage	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
< 50	1.0	1.2	18.9	NA	4.3	4.3	3.4
50 - 99	15.8	18.6	23.7	NA	24.4	24.4	22.7
100 - 124	22.3	24.1	20.1	NA	34.1	34.1	30.4
125 - 149	19.6	19.6	12.3	NA	18.9	18.9	18.5
150 - 199	23.5	20.3	11.3	NA	10.4	10.4	14.0
200 - 249	9.9	8.5	4.4	NA	3.5	3.5	5.5
>= 250	8.0	7.7	9.3	NA	4.5	4.5	5.6
Unknown	NA	NA	NA	NA	NA	NA	NA

3.4.1.5.2 Models

The following Table 3.44 table summarizes the results of the models assessing the relationship between wage categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.44: Overview of the results of the different models examining the relation between wage and Occupational Accidents

catwage	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
<50	REF	REF	REF	REF	ns	REF	REF	REF	>/>
50-99	>	ns	>	>	>	ns	>/ns	>	ns/>
100-124	>	ns	>	>	>	ns	>/ns	>	>/>
125-149	>	ns	>	>	>	ns	ns/ns	>	>/>
150-199	>	ns	>	>	>	ns	ns/ns	>	ns/ns
200-249	>	ns	>	>	ns	ns	ns/>	>	ns/ns
>=250	>	ns	>	>	rank def	ns	</ns	>	rank def

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Salary and occupational accidents

Workers earning between €100-124 per day have the highest probability of experiencing a commuting or workplace accident. For higher wage categories, the odds decrease as the wage category increases.
A higher proportion of workers earning between €100-124 and €125-149 significantly increases the frequency degree of occupational accidents within a company.
Wage category does not significantly determine whether an occupational accident is classified as serious.
Wage category is a significant factor in determining the number of absenteeism days due to occupational accidents: the number of days off is significantly higher for lower wage categories.
A higher proportion of workers in the lowest wage categories increases the severity degree within a company.

« previous

Figure 3.48 (a) and Figure 3.48 (b) illustrate that all wage categories have a higher likelihood of reporting an OA (both commuting and workplace accidents) compared to the lowest category (<€50), even after accounting for significant differences in company representativeness. Workers earning between €100-124/day have the highest probability of reporting an OA, with odds more than seven times higher than those in the lowest wage group.

In literature, inconsistent results are found regarding the relationship between wages and the incidence of OAs. Two opposing theories are proposed: some authors suggest that higher wage groups experience a higher incidence of OAs, viewing wages as compensation for the risky nature of their jobs (Lalive, Rafael and Ruf, Oliver and Zweimuller, Josef, 2006). Conversely, other findings indicate that workers with higher wages have a lower incidence of OAs, reflecting their preference for less risky jobs (Cabrera-Flores, 2023). Both theories may be relevant to our results.

Figure 3.49 (a) and Figure 3.49 (b) show that wage category does not significantly determine whether an insurer accepts or rejects an OA claim. This is in line with expectations.

Figure 3.50 (a) and Figure 3.50 (b) show that all wage categories have a higher likelihood of an accepted commuting accident compared to the lowest wage category (<€50). However, workers earning between €100-124/day have the highest probability of experiencing a commuting accident, with odds nearly 7 times higher than those in the lowest wage group. For higher wage categories, the trend shows that the odds decrease as the wage category increases. We found no studies examining this specific relationship between wage and the likelihood of commuting occupational risk.

Figure 3.51 (a) and Figure 3.51 (b) show the estimates for the relation between wage categories and an accepted workplace OA. The relation is comparable of that noticed in the figures for reporting an OA. In summary, all wage categories have a higher likelihood of an accepted workplace OA, with the highest odds for the category between €100-124/day, compared to the lowest category (<€50). For higher wage categories, the trend shows that the odds decrease as the wage category increases.

Figure 3.52 (a) and Figure 3.52 (b) present a model estimating the frequency degree of OAs at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures indicate that a higher proportion of workers earning between €100-124 and €125-149 significantly increases the frequency degree of OAs within a company.

Figure 3.53 (a) and Figure 3.53 (b) present a model estimating the chance for an accepted OA being serious (according to the definition of the Belgian law). These figures indicate that wage category does not significantly determine whether a workplace OA is classified as serious. This contrasts with the wage premium theory, which suggests that workers in riskier jobs are offered higher wages. According to this theory, these higher-paid workers tend to experience more serious or fatal accidents, but fewer non-serious ones (Lalive, Rafael and Ruf, Oliver and Zweimuller, Josef, 2006). However, it’s important to note that the definition of a serious OA under Belgian law may differ from those used in literature.

Figure 3.54 (a) and Figure 3.54 (b) show the results of a model assessing the relationship between wage and the number of absenteeism days. These figures are based on absenteeism data from FEDRIS, while Figure 3.55 (a) and Figure 3.55 (b) present the same model using data from Liantis PS. The figures indicate that wage category is a significant factor in determining the validated number of absenteeism days due to OAs. The number of accepted days off increases significantly for lower wage categories (e.g., 21% more days for €50-99 and 14% more days for €100-124). These wage categories likely represent jobs with higher exposure to occupational risks.

As mentioned above, Figure 3.55 (a) and Figure 3.55 (b) are the results of the model assessing the relation between wage category and number of absenteeism days related to the workplace OA, based on the Liantis PS data. The results are somewhat different to these of the model above. The number of accepted days off increases significantly for a higher wage category: 27% more days for €200-249 wage category, while no significant results were found for the other categories. These inconsistent findings may be due to differences in the study populations between the Liantis PS data and the FEDRIS absenteeism data, the Liantis PS dataset being considerably smaller.

Figure 3.56 (a) and Figure 3.56 (b) display the results of a model assessing the relationship between wage category and the direct wage cost of an OA. The results indicate that wage is a significant factor in determining the direct wage cost for an employer. How higher wage categories however, how smaller the cost in comparison with the reference. This is likely due to fewer and less serious accidents in the higher wage categories (when we compare with the results from the previous models).

The next 4 figures (Figure 3.57 (a) and Figure 3.57 (b), Figure 3.58 (a) and Figure 3.58 (b)) are displaying the results of a model assessing the relation between proportion of workers within a wage category and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.57 (a) and Figure 3.57 (b)), validated number of absenteeism days from FEDRIS were used, while the second model (Figure 3.58 (a) and Figure 3.58 (b)) was based on the numbers of days off in the Liantis PS data. The initial figures indicate that a higher proportion of workers in three of the four lowest wage categories (<€50, €100-124, and €125-149) significantly increases the severity level within a company. However, the figures based on Liantis PS absenteeism data (Figure 3.58 (a) and Figure 3.58 (b)) show a slightly different pattern: a higher proportion of workers in all four lowest wage categories significantly increases the severity level within a company.

3.4.1.6 Residential location (province)

3.4.1.6.1 Descriptives

The NSSO provides data about the employee’s residence, including the province where they live. FEDRIS statistical reports offer the same information on residence: the province where the employee resides. From Liantis PS data, it’s possible to determine the employee’s province through their home address postal code. However, the notification unfortunately does not include the postal code of the employee’s residence. Since this information is missing in the notification passed to Liantis, the derived residential location of an employee experiencing an OA cannot be compared with the original notification’s residential location classification.

In the next 6 tables, an overview of all (accepted) OAs (Table 3.45 and Table 3.46), the commuting OAs (Table 3.47 and Table 3.48) and the workplace OAs (Table 3.49 and Table 3.50) are given.

In the first table ((Table 3.45, Table 3.47 and Table 3.49), the absolute numbers are provided stratified per province, next to the total number of employees, as available in the Liantis PS database.

In the second table (Table 3.46, Table 3.48 and Table 3.50), the relative numbers of the OAs per province are provided as percentages. In this tables, no information is available from the notifications.

From tables Table 3.46, Table 3.48, and Table 3.50, a number of observations can be made. We observe a discrepancy between NSSO and Liantis data: Liantis reports a significantly higher percentage of workers in West Flanders (43%) compared to NSSO figures (10%). In Limburg, East Flanders, and Flemish Brabant, the percentages are more consistent. For the other provinces, the proportion of Liantis workers is lower than the NSSO figures.

all accepted occupational accidents

Table 3.45: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
province	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Antwerpen	25957449	20922766	2513258	246383	NA	NA	NA
Brussel	12692609	9885254	416523	80045	NA	NA	NA
Henegouwen	16060454	11302421	179155	129779	NA	NA	NA
Limburg	11871370	9248788	1343655	112247	NA	NA	NA
Luik	13448311	9263038	74145	120404	NA	NA	NA
Luxemburg	2671126	1671443	17809	19659	NA	NA	NA
Namen	6533638	4385366	60317	51896	NA	NA	NA
Oost-Vlaanderen	22366805	17033724	2441520	205517	NA	NA	NA
Unknown	3494077	3223122	1488183	59915	NA	NA	NA
Vlaams Brabant	16365498	12449407	1640213	115393	NA	NA	NA
Waals Brabant	5063215	3702801	73997	30005	NA	NA	NA
West-Vlaanderen	16643590	13273829	7807091	182134	NA	NA	NA

Table 3.46: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
province	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Antwerpen	16.95	17.98	13.92	18.21	NA	NA	NA
Brussel	8.29	8.50	2.31	5.91	NA	NA	NA
Henegouwen	10.49	9.71	0.99	9.59	NA	NA	NA
Limburg	7.75	7.95	7.44	8.29	NA	NA	NA
Luik	8.78	7.96	0.41	8.90	NA	NA	NA
Luxemburg	1.74	1.44	0.10	1.45	NA	NA	NA
Namen	4.27	3.77	0.33	3.83	NA	NA	NA
Oost-Vlaanderen	14.60	14.64	13.52	15.19	NA	NA	NA
Unknown	2.28	2.77	8.24	4.43	NA	NA	NA
Vlaams Brabant	10.68	10.70	9.08	8.53	NA	NA	NA
Waals Brabant	3.31	3.18	0.41	2.22	NA	NA	NA
West-Vlaanderen	10.87	11.41	43.24	13.46	NA	NA	NA

commuting accepted occupational accidents

Table 3.47: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
province	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Antwerpen	25957449	20922766	2513258	53566	NA	NA	NA
Brussel	12692609	9885254	416523	17193	NA	NA	NA
Henegouwen	16060454	11302421	179155	13749	NA	NA	NA
Limburg	11871370	9248788	1343655	18266	NA	NA	NA
Luik	13448311	9263038	74145	12266	NA	NA	NA
Luxemburg	2671126	1671443	17809	1846	NA	NA	NA
Namen	6533638	4385366	60317	6090	NA	NA	NA
Oost-Vlaanderen	22366805	17033724	2441520	39787	NA	NA	NA
Unknown	3494077	3223122	1488183	5201	NA	NA	NA
Vlaams Brabant	16365498	12449407	1640213	25839	NA	NA	NA
Waals Brabant	5063215	3702801	73997	5311	NA	NA	NA
West-Vlaanderen	16643590	13273829	7807091	27196	NA	NA	NA

Table 3.48: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
province	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Antwerpen	16.95	17.98	13.92	23.67	NA	NA	NA
Brussel	8.29	8.50	2.31	7.60	NA	NA	NA
Henegouwen	10.49	9.71	0.99	6.08	NA	NA	NA
Limburg	7.75	7.95	7.44	8.07	NA	NA	NA
Luik	8.78	7.96	0.41	5.42	NA	NA	NA
Luxemburg	1.74	1.44	0.10	0.82	NA	NA	NA
Namen	4.27	3.77	0.33	2.69	NA	NA	NA
Oost-Vlaanderen	14.60	14.64	13.52	17.58	NA	NA	NA
Unknown	2.28	2.77	8.24	2.30	NA	NA	NA
Vlaams Brabant	10.68	10.70	9.08	11.42	NA	NA	NA
Waals Brabant	3.31	3.18	0.41	2.35	NA	NA	NA
West-Vlaanderen	10.87	11.41	43.24	12.02	NA	NA	NA

workplace accepted occupational accidents

Table 3.49: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
province	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Antwerpen	25957449	20922766	2513258	192817	NA	NA	NA
Brussel	12692609	9885254	416523	62852	NA	NA	NA
Henegouwen	16060454	11302421	179155	116030	NA	NA	NA
Limburg	11871370	9248788	1343655	93981	NA	NA	NA
Luik	13448311	9263038	74145	108138	NA	NA	NA
Luxemburg	2671126	1671443	17809	17813	NA	NA	NA
Namen	6533638	4385366	60317	45806	NA	NA	NA
Oost-Vlaanderen	22366805	17033724	2441520	165730	NA	NA	NA
Unknown	3494077	3223122	1488183	54714	NA	NA	NA
Vlaams Brabant	16365498	12449407	1640213	89554	NA	NA	NA
Waals Brabant	5063215	3702801	73997	24694	NA	NA	NA
West-Vlaanderen	16643590	13273829	7807091	154938	NA	NA	NA

Table 3.50: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
province	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Antwerpen	16.95	17.98	13.92	17.11	NA	NA	NA
Brussel	8.29	8.50	2.31	5.58	NA	NA	NA
Henegouwen	10.49	9.71	0.99	10.29	NA	NA	NA
Limburg	7.75	7.95	7.44	8.34	NA	NA	NA
Luik	8.78	7.96	0.41	9.59	NA	NA	NA
Luxemburg	1.74	1.44	0.10	1.58	NA	NA	NA
Namen	4.27	3.77	0.33	4.06	NA	NA	NA
Oost-Vlaanderen	14.60	14.64	13.52	14.70	NA	NA	NA
Unknown	2.28	2.77	8.24	4.85	NA	NA	NA
Vlaams Brabant	10.68	10.70	9.08	7.95	NA	NA	NA
Waals Brabant	3.31	3.18	0.41	2.19	NA	NA	NA
West-Vlaanderen	10.87	11.41	43.24	13.75	NA	NA	NA

3.4.1.6.2 Models

The following Table 3.51 table summarizes the results of the models assessing the relationship between residential location categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.51: Overview of the results of the different models examining the relation between place of residence (province) and Occupational Accidents

catresiloc	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Antwerp	REF	REF	REF	REF	ns	REF	REF	REF	ns
Brussel	<	ns	ns	<	<	ns	>/>	>	ns
Hainaut	>	ns	ns	>	>	ns	ns	ns	ns
Limburg	<	ns	ns	<	<	ns	>/ns	ns	ns
Liège	ns	ns	ns	ns	ns	ns	ns	>	ns
Luxembourg	>	ns	ns	>	ns	ns	ns	ns	ns
Namur	ns	ns	ns	ns	ns	ns	ns	ns	ns
East Fl	>	ns	ns	ns	ns	ns	ns	ns	ns
Fl Brabant	<	ns	ns	<	ns	ns	>/>	ns	ns
Wal Brabant	ns	ns	ns	ns	ns	ns	ns/>	ns	ns
West-Fl	>	ns	>	>	>	ns	ns	ns	ns

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Residential location (province) and occupational accidents

Workers residing in West-Flanders, Luxembourg, and Hainaut have a significantly higher likelihood of reporting occupational accidents and experiencing occupational workplace accidents compared to those living in Antwerp
The province in which an employee is living does not significantly determine the likelihood to have an accepted commuting occupational accident.
The figures suggest that workers who are living in Brussels and Flemish Brabant have a higher number of absenteeism days due to occupational accidents.
The findings suggest that workers living in Brussel have a significant higher direct wage cost related to the occupational accident, for employers in comparison with the direct wage cost of workers living in Antwerp.
The proportion of workers living in West-Flanders and Hainaut significantly increases the frequency degree (but not the severity degree) of occupational accidents within a company.

« previous

According to Figure 3.59 (a) and Figure 3.59 (b), workers residing in Luxembourg, West-Flanders and Hainaut have a significantly higher likelihood of reporting OAs (both commuting and workplace accidents) compared to those living in Antwerp, even after accounting for differences in company representativeness (link to methodology).

According to Figure 3.60 (a) and Figure 3.60 (b), the province in which a worker lives does not significantly determine the likelihood of a reported OA being rejected by the insurer.

According to Figure 3.61 (a) and Figure 3.61 (b), the province in which an employee is living does not significantly determine the likelihood to have an accepted commuting accident.

Figure 3.62 (a) and Figure 3.62 (b) show the estimates for the relation between residence (province) and an accepted workplace OA. The results indicate that workers living in West-Flanders, Hainout and Luxembourg have a significantly higher likelihood of having an accepted workplace OA.

Figure 3.63 (a) and Figure 3.63 (b) present a model estimating the frequency degree of OAs at the company level, contrasting with previous models that focused on individual employee accident probabilities. These figures show that a higher proportion of workers living in West-Flanders and Hainaut significantly increases the frequency degree of OAs within a company. On the contrary, a higher proportion workers from Limbourg and Brussels, significantly decreases the frequency degree of a company.

Figure 3.64 (a) and Figure 3.64 (b) present a model estimating the likelihood of an accepted accident being classified as serious according to Belgian law. These figures indicate that the province a worker is living in does not significantly determine the likelihood an OA is classified as serious.

Figure 3.65 (a) and Figure 3.65 (b) show the results of a model assessing the relationship between residence and the number of absenteeism days. These figures are based on absenteeism data from FEDRIS, while Figure 3.66 (a) and Figure 3.66 (b) present the same model using data from Liantis PS. The figures indicate that workers who are living in Brussels, Flemish Brabant or Limbourg have a higher validated number of absenteeism days due to OAs. The number of accepted days off increases significantly for workers in Brussels, Flemish Brabant and Limbourg, with respectively 66%, 15% and 21% in comparison with workers living in Antwerp.

As mentioned above, Figure 3.66 (a) and Figure 3.66 (b) are the results of the model assessing the relation between residence and the number of absenteeism days related to the workplace accident, based on the Liantis PS data. The findings are more or less in line with those of the model based on the FEDRIS absenteeism data: workers living in Brussels and Flemish Brabant have more absenteeism days related to OAs then workers from Antwerp. However, in this sample, also workers from Walloon Brabant have a significant higher number of absenteeism days (61%) related to OAs in comparison with workers from Antwerp.

Figure 3.67 (a) and Figure 3.67 (b) show the results of a model evaluating the relationship between residence and the direct wage cost of an OA. The findings indicate that workers living in Brussels have a significant higher direct wage cost related to the OA, for employers in comparison with the direct wage cost of workers living in Antwerp.

The next 4 figures (Figure 3.68 (a) and Figure 3.68 (b), Figure 3.69 (a) and Figure 3.69 (b)) are displaying the results of a model assessing the relation between proportion of workers living within a province and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.68 (a) and Figure 3.68 (b)), validated number of absenteeism days from FEDRIS were used, while the second model Figure 3.69 (a) and Figure 3.69 (b)) was based on the numbers of days off in the Liantis PS data. The results of both models indicate that the proportion of workers living in a particular province does not determine the severity level within a company.

3.4.1.7 Commuting distance

3.4.1.7.1 Descriptives

NSSO nor FEDRIS provide statistics on commuting distance. From Liantis PS data however, it also possible to roughly estimate the commuting distance of a worker by calculating the distance between the residential location and the postal code of the employer.

From the tables Table 3.53, Table 3.55 and Table 3.57, some important findings can be noticed. Employees commuting 10–19 km have the highest probability of experiencing an OA (both commuting and workplace accidents), those commuting 5–9 km have the lowest probability. Interestingly, the shortest (<5 km) and longest (\(\geq\) 20 km) commutes are associated with moderate accident probabilities. This suggests that medium-range commuters (10–19 km) might be at greater risk, possibly due to factors like higher traffic exposure or fatigue without the compensating benefits of very short or very long commutes (e.g., more cautious behaviour or different transport modes).

all accepted occupational accidents

Table 3.52: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catdist	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
<5 km	NA	NA	5946135	NA	11236	11236	4643
5 - 9 km	NA	NA	2592608	NA	6743	6743	2917
10 - 19 km	NA	NA	4006374	NA	11655	11655	5317
>=20 km	NA	NA	3964647	NA	10600	10600	5194
NA	NA	NA	1546102	NA	154985	21001	7959

Table 3.53: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catdist	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
<5 km	NA	NA	36.0	NA	27.9	27.9	25.7
5 - 9 km	NA	NA	15.7	NA	16.8	16.8	16.1
10 - 19 km	NA	NA	24.3	NA	29.0	29.0	29.4
>=20 km	NA	NA	24.0	NA	26.3	26.3	28.7
NA	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.54: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catdist	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
<5 km	NA	NA	5946135	NA	1484	1484	674
5 - 9 km	NA	NA	2592608	NA	972	972	473
10 - 19 km	NA	NA	4006374	NA	1561	1561	849
>=20 km	NA	NA	3964647	NA	1675	1675	853
NA	NA	NA	1546102	NA	22695	2907	1263

Table 3.55: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catdist	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
<5 km	NA	NA	36.0	NA	26.1	26.1	23.7
5 - 9 km	NA	NA	15.7	NA	17.1	17.1	16.6
10 - 19 km	NA	NA	24.3	NA	27.4	27.4	29.8
>=20 km	NA	NA	24.0	NA	29.4	29.4	29.9
NA	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.56: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catdist	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
<5 km	NA	NA	5946135	NA	9752	9752	3969
5 - 9 km	NA	NA	2592608	NA	5771	5771	2444
10 - 19 km	NA	NA	4006374	NA	10094	10094	4468
>=20 km	NA	NA	3964647	NA	8925	8925	4341
NA	NA	NA	1546102	NA	132290	18094	6696

Table 3.57: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catdist	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
<5 km	NA	NA	36.0	NA	28.2	28.2	26.1
5 - 9 km	NA	NA	15.7	NA	16.7	16.7	16.1
10 - 19 km	NA	NA	24.3	NA	29.2	29.2	29.4
>=20 km	NA	NA	24.0	NA	25.8	25.8	28.5
NA	NA	NA	NA	NA	NA	NA	NA

3.4.1.7.2 Models

The following Table 3.58 table summarizes the results of the models assessing the relationship between commuting distance categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.58: Overview of the results of the different models examining the relation between commuting distance and Occupational Accidents

catdist	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
less than 5	REF	REF	REF	REF	<	REF	REF	REF	</ns
5 to 9	>	ns	>	>	>	ns	ns	>	ns
10 to 19	>	ns	>	>	>	ns	ns	ns	ns
20 and more	>	ns	>	>	>	ns	ns	>	ns/<

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Commuting distance and occupational accidents

Workers with commuting distances of 5 km and more all have higher chances to report occupational accidents (commuting and workplace accidents) compared to those living less than 5 km from work.
Odds for accepted commuting occupational accidents show an increasing trend for larger commuting distance categories.
Workers working between 10 and 20 km from home show the highest odds to experience an accepted workplace accident. A multivariable model should confirm this finding whether commuting distances do not only impact the chance to experience commuting accidents, but as well the chance to experience workplace accidents.
Higher proportions of workers working more than 5 km from home significantly increase the frequency degree of a company while a higher proportion of workers working less than 5 km from home significantly decreases it.
Commuting distance categories do not do not significantly determine whether an occupational accident is classified as serious nor the number of absenteeism days due to occupational accidents.
All commuting distance categories of 5 and more km show a significant higher direct wage cost than the reference category.
Severity degrees tend to be lower for companies with higher proportions of workers in the commuting distance categories less than 5 and 20 and more km.

« previous

Figure 3.70 (a) and Figure 3.70 (b) show that people working 5 and more km from home have a higher chance to report an OA. People from the category 10 - 19 km have a 34% higher chance compared to the reference category.

According to Figure 3.71 (a) and Figure 3.71 (b), the distance category between the residential and workplace location does not significantly determine the likelihood of a reported OA being rejected by the insurer.

According to Figure 3.72 (a) and Figure 3.72 (b), the commuting distance category significantly determines the likelihood to have an accepted commuting accident. Workers falling into categories of 5 km and more show higher odds for commuting accidents with a trend for higher odds for longer distances. While the increasing trend is in line with the findings of (Ponsin et al., 2020), the relative differences between the categories are not of the same order of magnitude (we find 50% and not 400% more chance for commuting distances \(\geq\) 20 km). We do not have any information on the perception of the risk nor the mode of transport (by bicycle, on foot, by car,…) and thus cannot compare or results with other studies (Fadzil & Bulgiba, 2023; Finnish Institute of Occupational Health, 2025).

Figure 3.73 (a) and Figure 3.73 (b) show that the commuting distance category significantly determines the likelihood to experience an accepted workplace accident. Workers falling into categories of 5 km and more show higher odds for workplace accidents with the highest ratio (35% higher) occurring in the category 10 - 19 km.

Figure 3.74 (a) and Figure 3.74 (b) present a model estimating the frequency degree of OAs at the company level, contrasting with previous models that focused on the individual employee accident probabilities. These figures show that a higher proportion of workers living between ten and twenty km from work significantly increases the frequency degree of OAs within a company. On the contrary, a higher proportion workers living less than 5 km from work, significantly decreases the frequency degree of a company.

Figure 3.75 (a) and Figure 3.75 (b) present a model estimating the likelihood of an accepted accident being classified as serious according to Belgian law. These figures indicate that the commuting distance category does not significantly determine the likelihood an OA is classified as serious.

Figure 3.76 (a) and Figure 3.76 (b) show the results of a model assessing the relationship between commuting distance categories and absenteeism days. These figures are based on absenteeism data FEDRIS, while Figure 3.77 (a) and Figure 3.77 (b) present the same model using data from Liantis PS. The figures indicate that the commuting distance category of a worker does not significantly determine the number of days the worker is absent after experiencing an accepted workplace accident.

Figure 3.78 (a) and Figure 3.78 (b) show the results of a model evaluating the relationship between commuting distance category and the direct wage cost of an OA. The findings indicate that workers living 5-9 and 20 and more km away from their workplace have a significant higher direct wage cost related to the OA that workers from the reference category living less than 5 km from their workplace.

The next 4 figures (Figure 3.79 (a) and Figure 3.79 (b), Figure 3.80 (a) and Figure 3.80 (b)) are displaying the results of a model assessing the relation between proportion of workers falling into a certain commuting distance category and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.79 (a) and Figure 3.79 (b)), vailidated numbers of absenteeism days from FEDRIS were used, while the second model (Figure 3.80 (a) and Figure 3.80 (b)) was based on the numbers of days off in the Liantis PS data. The results of both models indicate a trend for lower severity degrees for companies with higher proportions of workers in the commuting distance categories less than 5 and 20 and more km.

3.4.1.8 Use of Personal Protective Equipment (PPE)

Data Quality Alert: Use of PPE

Unfortunately and as described in Section 2.15, given the limited data, the lack of longitudinal tracking, the uneven distribution in time and other data quality issues, linking the Liantis ESPP PPE evaluation data to occupational accidents would provide a very limited and potentially misleading picture considering the small number of employees for whom data is available. As a result, no further analyses or modelling efforts were undertaken in relation to use of PPE.

« previous

3.4.1.9 Health problems: quality of hearing

Based on the parameters outlined in Section 1.3.2.9, hearing impairment qualifies as a potential determinant with objective test results available through the medical preventive consults for further investigation. The other parameters mentioned are not systematically captured in the Liantis ESPP dataset via technical test results and will not be explored further.

NSSO nor FEDRIS provide statistics concerning health problems. For this reason, a descriptives section analogous to the other univariable analyses sections does not make sense here. We will only present results of the models examining the relation between the presence of noise dips and the different outcomes of OA under investigation.

Data Quality Alert: quality of hearing

As described in Section 2.13, after initial testing, in general only employees with assigned risks related to noise exposure qualify for follow up audiometric testing during a medical prevention consult. Thus, the results will not be representative of the entire working population within (or outside) Liantis customers through the whole ten-year period of the study.

Moreover, only data from employees from mutual Liantis ESPP and PS customers can be used for modelling, further reducing the potential number of cases described by 70%.

Additionally, an important assumption was made. Because the likelihood in the first three models is assessed on a monthly basis -i.e., whether an OA occurred for a given individual in a given month- and audiometric test results are not available monthly, we interpolated the noise dip variable within individuals. Specifically, we applied an “updown” approach: once a noise dip is observed (or not), its status is assumed to persist until the next available test result for the same person. For the first available result, we assumed the condition (presence or absence of a noise dip) was already present prior to testing. Employees without any audiometric test data were excluded from the analysis.

« previous

3.4.1.9.1 Models

The following Table 3.59 summarizes the results of the models assessing the relationship between absence (0) or presence (1) of noise dips and the nine outcomes. Below the table, you can find the data on which each model is based. The results of the nine models are then presented both graphically and in a table.

Table 3.59: Overview of the results of the different models examining the relation between noise dips and Occupational Accidents

cataudio	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree
noise dip 0	REF	REF	REF	REF	<	REF	REF	REF	</ns
noise dip 1	>	ns	ns	>	rank def	ns	ns/>	>	rank def

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Health problems: quality of hearing (ESPP audiometric test results)

Workers experiencing noise dips show a higher likelihood of workplace occupational accidents. There was no association between the noise dip category and the likelihood of an accepted commuting occupational accident, nor with the refusal of a notified occupational accident.
There was no observed relationship between noise dips and the likelihood of an occupational accident being categorized as serious.
Companies with a higher proportion of workers without noise dips tend to show lower frequency and severity degrees, based on the validated number of days absent.
The cost of an occupational accident appears to be higher for individuals experiencing noise dips, as does the number of days off calculated by Liantis PS.
Since hearing loss worsens with age -and wage, costs, and recovery time also increase with age- the associations found are not surprising. Further research using multivariable modelling is needed to clarify whether these relationships still hold when age is included as a confounding factor.

« previous

Figure 3.81 shows that workers facing audiometric test results in which noise dips are detected, have a 7% higher chance of reporting an OA (both commuting and workplace accidents), after accounting for significant differences in company representativeness (see Section 3.2). The results suggest that hearing impairment influences the likelihood to report an OA.

Figure 3.82 shows that the presence of noise dips does not significantly determine whether an insurer accepts or rejects an OA claim.

The results presented in Figure 3.83 show no association between hearing impairment described through the presence of noise dips and the likelihood for an accepted commuting OA.

Figure 3.84 shows that workers with noise dips have a 7% higher chance of having an accepted workplace accident compared to the reference category (without noise dips). This result indicates that hearing impairment indeed may influence the likelihood to experience an accepted workplace OA.

Figure 3.85 (a) and Figure 3.85 (b) present a model estimating the frequency degree of occupational accidents at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures show that a higher proportion of workers without noise dips decreases the frequency degree of OA within a company.

Figure 3.86 (a) and Figure 3.86 (b) present a model estimating the chance for an accepted workplace OA being classified as serious (according to the definition in the Belgian law). These figures indicate that the noise dips do not significantly determine whether an occupational workplace accident is classified as serious.

Figure 3.87 (a) and Figure 3.87 (b) show the results of a model assessing the relation between noise dips and the number of absenteeism days. These first two figures are based on the absenteeism data provided by FEDRIS (while Figure 3.88 (a) and Figure 3.88 (b) present the same model based on the Liantis PS data). The figures show that noise dips are not important in determining the validated number of absenteeism days related to an accepted workplace OA.

As mentioned above, Figure 3.88 (a) and Figure 3.88 (b) are the results of the model assessing the relation between noise dip category and number of absenteeism days related to the workplace accident, based on the Liantis PS calculated data. The results are in contrast with the results of Figure 3.87 above (but the dataset used for the present model is only half in size).

Figure 3.89 (a) and Figure 3.89 (b) show the results of the model that is assessing the relation between noise dip category and direct wage cost of an OA. It indicates that the presence of noise dips seems to be significantly related to a higher direct wage cost of an OA for an employer.

The next 4 figures (Figure 3.90 (a) and Figure 3.90 (b), Figure 3.91 (a) and Figure 3.91 (b)) are displaying the results of a model assessing the relation between proportion of workers with (and without) noise dips and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.90), the validated number of absenteeism days from FEDRIS was used, while the second model (Figure 3.91) was based on the numbers of days off calculated from the Liantis PS data. These figures show that a higher proportion of workers without noise dips decreases the severity degree within a company.

Figure 3.91 shows that a higher proportion of workers without noise dips does not significantly decreases the severity degree within a company, based on the number of calculated days from Liantis PS data. The direction of the (insignificant) effect is however in line with the conclusion from the previous model using validated FEDRIS data.

3.4.1.10 Sleeping disorders

Data Quality Alert: sleeping disorders

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the General Medical Questionnaire (AMV in Dutch) (GMQ) (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported sleep problems and notifications of occupational accidents are presented in Table 3.60. A Pearson’s Chi-squared test of independence between hadOA and SLEEP PROBLEMS suggests that the effect is statistically not significant \(\chi^2 = 6.47\), \(df = 4\), \(p = 0.167\).

Table 3.60: Notifications of occupational accidents in a certain month by self reported time duration of sleep problems among employees completing the GMQ in 2022 and 2023

(a) counts

	0 days	< 1 week	1 week - 1 month	1 month - 3 months	> 3 months
0	487157	116954	87634	43796	93339
1	2394	567	471	219	503

(b) percentages

	0 days	< 1 week	1 week - 1 month	1 month - 3 months	> 3 months
0	99.51	99.52	99.47	99.5	99.46
1	0.49	0.48	0.53	0.5	0.54

3.4.1.11 Substance abuse

Data Quality Alert: substance abuse (alcohol)

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the GMQ (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported alcohol consumption and notifications of occupational accidents are presented in Table 3.61. A Pearson’s Chi-squared test of independence between hadOA and ALCOHOL CONSUMPTION suggests that the effect is statistically not significant \(\chi^2 = 6.58\), \(df = 5\), \(p = 0.254\).

Table 3.61: Notifications of occupational accidents in a certain month by self reported alcohol consumption among employees completing the GMQ in 2022 and 2023

(a) counts (glasses a week)

	No	<= 5	6 - 10	11 - 20	> 20	I do not wish to answer
0	16837	20333	5846	1452	317	1060
1	98	95	22	10	2	3

(b) percentages

	No	<= 5	6 - 10	11 - 20	> 20	I do not wish to answer
0	99.42	99.53	99.63	99.32	99.37	99.72
1	0.58	0.47	0.37	0.68	0.63	0.28

Data Quality Alert: substance abuse (drugs)

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the GMQ (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported drug use and notifications of occupational accidents are presented in Table 3.62. A Pearson’s Chi-squared test of independence between hadOA and DRUG USE suggests that the effect is statistically not significant \(\chi^2 = 4.77\), \(df = 3\), \(p = 0.189\).

Table 3.62: Notifications of occupational accidents in a certain month by self reported drug use among employees completing the GMQ in 2022 and 2023

(a) counts

	No	Occasionally	Every day	I do not wish to answer
0	44367	954	147	377
1	222	8	0	0

(b) percentages

	No	Occasionally	Every day	I do not wish to answer
0	99.5	99.17	100	100
1	0.5	0.83	0	0

Data Quality Alert: medication use

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the GMQ (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported medication use and notifications of occupational accidents are presented in Table 3.63. A Pearson’s Chi-squared test of independence between hadOA and MEDICATION suggests that the effect is statistically not significant \(\chi^2 = 3.52\), \(df = 3\), \(p = 0.318\).

Table 3.63: Notifications of occupational accidents in a certain month by self reported medication use among employees completing the GMQ in 2022 and 2023

(a) counts

	Nee	Af en toe	Dagelijks	Wil niet antwoorden
0	39147	3966	2440	292
1	191	26	13	0

(b) percentages

	Nee	Af en toe	Dagelijks	Wil niet antwoorden
0	99.51	99.35	99.47	100
1	0.49	0.65	0.53	0

3.4.1.12 Personality

Data Quality Alert: personality

Within Liantis, no systematically maintained variables related to personality traits are currently available that would allow for investigation of a potential association with occupational accidents. Similarly, the FEDRIS notification data does not include any variables related to personality traits. As a result, no further analyses or modelling efforts were undertaken in relation to personality characteristics.

« previous

3.4.1.13 Number of years of schooling

Data Quality Alert: number of years of schooling

Within Liantis, no systematically maintained variables related to the number of years of schooling are currently available that would allow for investigation of a potential association with occupational accidents. Similarly, the FEDRIS notification data does not include any variables related to number of years of schooling. As a result, no further analyses or modelling efforts were undertaken in relation to number of years of schooling.

« previous

3.4.2.1 Occupation and workforce segment

Data Quality Alert: Occupation and workforce segment

In the data quality report section, we examined three variables on their suitability as a proxy for occupation and workforce segment: catnsso, catprof, and catisco (see Section 2.7.2.3, Section 2.7.2.5 and Section 2.7.2.6). Although there are also some issues with the data from catnsso variable in the notifications, it seems a better choice than catprof and certainly catisco (since the tag is almost always missing for this last one, a mapping the the International Standard Classification of Occupations (ISCO) system will not be possible). With catnsso we see less NA or “Onbekend” values and a slightly better mapping to blue and white collar workers. For this reason, we will only examine the catnsso variable in further analyses and modelling.

« previous

3.4.2.1.1 Descriptives

The statute or employment type generally refers to the distinction between white and blue collars. The NSSO does provide data about the statute in three categories in their broadest categorisation: white collar workers, blue collar workers and civil servants). FEDRIS statistical reports and notifications provide information about the statute, although they include a lot more categories. Since Liantis PS data also contains the indication of a white collar, blue collar or other worker, this variable could be categorised in the same categories that are used in the NSSO reports. On the other hand, the FEDRIS statistical reports and the FEDRIS notification data had to be recategorized into larger categories. Hereafter, the fit between the derived statute category of an employee experiencing an OA can be compared with the corresponding statute category classification in the notification of the accident.

From Figure 3.92 we learn that the nature of the statute categories of Liantis PS data and FEDRIS data are quite comparable. This suggests that the statute categories derived from Liantis PS data might align well with those in FEDRIS data, making them suitable for further analysis. The accuracy is >98%.

Figure 3.92: Concordance between FEDRIS provided and Liantis PS statute of an employee

In the next 6 tables, an overview of all (accepted) OAs (Table 3.64 and Table 3.65), the commuting OAs (Table 3.66 and Table 3.67) and the workplace OAs (Table 3.68 and Table 3.69) are given. In the first table ((Table 3.64, Table 3.66 and Table 3.68), the absolute numbers are provided stratified per statute category, next to the total number of employees, as available in the Liantis database. In the second table (Table 3.65, Table 3.67 and Table 3.69), the relative numbers of the OAs per statute category are provided as percentages.

From tables Table 3.65, Table 3.67, and Table 3.69, several important findings emerge. We observe a difference between NSSO and Liantis data: NSSO has more white-collar workers, while Liantis has more blue-collar workers. However, the proportion of accidents is consistently higher among blue-collar workers than expected based on the NSSO or Liantis population distribution. Conversely, the proportion of accidents among white-collar workers is lower than expected. This suggests that the risk of accidents is higher for blue-collar workers, which aligns with expectations. Additionally, the proportion of accidents among civil servants is lower than expected based on the NSSO or Liantis population distribution. This suggests that blue-collar workers are at greater risk for OAs.

all accepted occupational accidents

Table 3.64: Numbers of employees per statute category (nsso and liantis) and numbers of occupational accident notifications per statute category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
andere	NA	NA	1611584	23380	21293	3215	2047
arbeiders	52888328	48618314	9189273	849555	121324	43085	15954
bedienden	79911613	65816512	7242032	480442	52602	14935	8029

Table 3.65: Percentages of employees per statute category (nsso and liantis) and percentages of occupational accident notifications per statute category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
andere	NA	NA	NA	NA	NA	NA	NA
arbeiders	39.83	42.49	55.9	63.9	69.8	74.3	66.5
bedienden	60.17	57.51	44.1	36.1	30.2	25.7	33.5

commuting accepted occupational accidents

Table 3.66: Numbers of employees per statute category (nsso and liantis) and numbers of occupational accident notifications per statute category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
andere	NA	NA	1611584	4083	3663	407	240
arbeiders	52888328	48618314	9189273	95916	11783	4436	1951
bedienden	79911613	65816512	7242032	126311	12941	3756	1921

Table 3.67: Percentages of employees per statute category (nsso and liantis) and percentages of occupational accident notifications per statute category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
andere	NA	NA	NA	NA	NA	NA	NA
arbeiders	39.83	42.49	55.9	43.2	47.7	54.2	50.4
bedienden	60.17	57.51	44.1	56.8	52.3	45.8	49.6

workplace accepted occupational accidents

Table 3.68: Numbers of employees per statute category (nsso and liantis) and numbers of occupational accident notifications per statute category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
andere	NA	NA	1611584	19297	17630	2808	1807
arbeiders	52888328	48618314	9189273	753639	109541	38649	14003
bedienden	79911613	65816512	7242032	354131	39661	11179	6108

Table 3.69: Percentages of employees per statute category (nsso and liantis) and percentages of occupational accident notifications per statute category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
andere	NA	NA	NA	NA	NA	NA	NA
arbeiders	39.83	42.49	55.9	68	73.4	77.6	69.6
bedienden	60.17	57.51	44.1	32	26.6	22.4	30.4

3.4.2.1.2 Models

The following Table 3.70 summarizes the results of the models assessing the relationship between contract and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both in a table and graphically.

Table 3.70: Overview of the results of the different models examining the relation between statute category and Occupational Accidents

catnsso	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
blue collars	REF	REF	REF	REF	>	REF	REF	REF	>/>
white collars	<	ns	>	<	ns	<	</<	<	</<

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Occupation and workforce segment and occupational accidents

Blue collars have a higher chance to report an occupational accident (and have a higher chance for an accepted workplace occupational accident)
Blue collars have a higher chance for an occupational accident being serious.
Blue collars have more absenteeism days related to an occupational accident, including more wage related costs for the employer.
A higher proportion of blue collars increases the frequency and severity degree within a company.

« previous

According to Figure 3.93 (a) and Figure 3.93 (b), white collars are 63% less likely to report OAs (both commuting and workplace accidents) compared to blue collar workers, even after accounting for significant differences in company representativeness. This is in line with expectations and former research (Abatemarco et al., 1995; Hägg et al., 2015).

According to Figure 3.94 (a) and Figure 3.94 (b), the statute does not significantly determine the likelihood of a reported OA being rejected by the insurer.

According to Figure 3.96 (a) and Figure 3.96 (b), white-collar workers show a 21% higher chance to experience an accepted commuting accident.

Figure 3.97 (a) and Figure 3.97 (b) show the estimates for the relation between statute categories and an accepted workplace OA. White collars have 69% less chance to experience accepted workplace accidents compared to blue collars. This is in line with expectations and former research ((Abatemarco et al., 1995; Hägg et al., 2015).

Figure 3.98 (a) and Figure 3.98 (b) present a model estimating the frequency degree of OAs at the company level, contrasting with previous models that focused on individual employee accident probabilities. These figures show that a higher proportion of blue-collar workers significantly increases the frequency degree of OAs within a company. This aligns with expectations, as blue-collar workers are generally more exposed to safety risks.

Figure 3.99 (a) and Figure 3.99 (b) present a model estimating the likelihood of an accepted accident being classified as serious according to Belgian law. These figures indicate that OAs among white-collar workers are significantly less likely to be classified as serious compared to those among blue-collar workers. This aligns with expectations: as mentioned above, blue collars are are generally more exposed to safety risks.

Figure 3.100 (a) and Figure 3.100 (b) show the results of a model assessing the relationship between statute and the number of absenteeism days. These figures are based on absenteeism data from FEDRIS, while Figure 3.101 (a) and Figure 3.101 (b) present the same model using data from Liantis PS. The figures indicate that statute is a significant factor in determining the validated number of absenteeism days due to OAs. The number of accepted days off decreases significantly for white collars (31% less days for white collars in comparison with blue collars).

As mentioned above, Figure 3.101 (a) and Figure 3.101 (b) are the results of the model assessing the relation between contract type and number of absenteeism days related to the workplace accident, based on the Liantis PS data. The findings aligns with those of the model based on the FEDRIS absenteeism data, showing white collars having less absenteeism days related to OAs then blue collars. However, the estimate is much smaller (only 6% less days for white collars, in comparison with 31% in the first model). This may be due to the smaller sample size of the Liantis PS data. In any case, these findings are consistent with expectations and the results of previous models.

Figure 3.102 (a) and Figure 3.102 (b) show the results of a model evaluating the relationship between statute and the direct wage cost of an OA. The findings indicate that the statute significantly impacts the direct wage cost for employers: white collars incur lower costs compared to blue collars.

The next 4 figures (Figure 3.103 (a) and Figure 3.103 (b), Figure 3.104 (a) and Figure 3.104 (b)) are displaying the results of a model assessing the relation between proportion of workers within a statute category and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.103 (a) and Figure 3.103 (b)), validated number of absenteeism days from FEDRIS were used, while the second model Figure 3.104 (a) and Figure 3.104 (b)) was based on the numbers of days off in the Liantis PS data. The results of both models indicate that a higher proportion of blue collars significantly increases the severity level within a company. On the other hand, a higher proportion of white collars significantly decreases the severity level within a company.

3.4.2.2 Contract type (permanent/indefinite/unlimited versus temporary/definite/limited) (precarious employment)

3.4.2.2.1 Descriptives

When we refer to contract type, we mean the duration of the contract. Generally, the distinction is made between permanent (indefinite or unlimited) and temporary (definite, fixed-term, or limited) contracts. The NSSO does not provide data on contract type. FEDRIS statistical reports and notifications include information about contract type in three categories: limited, unlimited, and unknown. In the Liantis PS dataset available for this project, contract type is not explicitly mentioned in terms of duration. Therefore, we categorize workers who have more than one contract per month as having a temporary (or limited duration) contract, while others are considered to have a permanent (or unlimited) contract.

Data Quality Alert: Contract type (permanent/indefinite/unlimited versus temporary/definite/limited) (precarious employment)

In the Liantis PS dataset available for this project, contract type is not explicitly mentioned in terms of duration. Therefore, we categorize workers who have more than one contract per month as having a temporary (or limited duration) contract, while others are considered to have a permanent (or unlimited) contract.

This is an assumption that may not always hold true, results have to be interpreted with caution.

« previous

In the next 6 tables, an overview of all (accepted) OAs (Table 3.71 and Table 3.72), the commuting OAs (Table 3.73 and Table 3.74) and the workplace OAs (Table 3.75 and Table 3.76) are given.

In the first table ((Table 3.71, Table 3.73 and Table 3.75), the absolute numbers are provided stratified per contract type category, next to the total number of employees, as available in the Liantis database.
In the second table (Table 3.72, Table 3.74 and Table 3.76), the relative numbers of the OAs per contract type category are provided as percentages.

From the table Table 3.72, it can be noticed that workers with an unlimited contract account for 79% of the employees within Liantis customers, while they account for about 88% of the employees within the notifications.

all accepted occupational accidents

Table 3.71: Numbers of employees per limdur of contract category (nsso and liantis) and numbers of occupational accident notifications per limdur of contract category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
unlimited	NA	NA	14333967	NA	149146	42873	19094
limited	NA	NA	3721899	NA	17857	5881	2676
unknown	NA	NA	NA	NA	28216	12481	4260

Table 3.72: Percentages of employees per limdur of contract category (nsso and liantis) and percentages of occupational accident notifications per limdur of contract category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
unlimited	NA	NA	79.4	NA	89.3	87.9	87.7
limited	NA	NA	20.6	NA	10.7	12.1	12.3
unknown	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.73: Numbers of employees per limdur of contract category (nsso and liantis) and numbers of occupational accident notifications per limdur of contract category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
unlimited	NA	NA	14333967	NA	22070	5996	3015
limited	NA	NA	3721899	NA	2876	1056	515
unknown	NA	NA	NA	NA	3441	1547	582

Table 3.74: Percentages of employees per limdur of contract category (nsso and liantis) and percentages of occupational accident notifications per limdur of contract category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
unlimited	NA	NA	79.4	NA	88.5	85	85.4
limited	NA	NA	20.6	NA	11.5	15	14.6
unknown	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.75: Numbers of employees per limdur of contract category (nsso and liantis) and numbers of occupational accident notifications per limdur of contract category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
unlimited	NA	NA	14333967	696665	127076	36877	16079
limited	NA	NA	3721899	198781	14981	4825	2161
unknown	NA	NA	NA	130824	24775	10934	3678

Table 3.76: Percentages of employees per limdur of contract category (nsso and liantis) and percentages of occupational accident notifications per limdur of contract category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
unlimited	NA	NA	79.4	77.8	89.5	88.4	88.2
limited	NA	NA	20.6	22.2	10.5	11.6	11.8
unknown	NA	NA	NA	NA	NA	NA	NA

3.4.2.2.2 Models

The following Table 3.77 table summarizes the results of the models assessing the relationship between contract type categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.77: Overview of the results of the different models examining the relation between contract type and Occupational Accidents

catlimdur	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
unlimited	REF	REF	REF	REF	>	REF	REF	REF	>/>
limited	<	ns	<	<	rank def	ns	</<	<	rank def

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Contract type (permanent/unlimited versus temporary/limited) (precarious employment) and occupational accidents

Workers with a contract of limited duration (fixed-term contract) have a lower chance to report an occupational accident (and have a lower chance for both an accepted commuting as an accepted workplace occupational accident)
Workers with contract of limited duration do not have a higher chance for an occupational accident being serious.
Workers with a contract of limited duration have less absenteeism days related to an occupational accident (with a lower direct wage cost), in comparison with workers with a contract of unlimited duration.
A higher proportion of workers with a contract of unlimited duration significantly increases the frequency and severity level within a company.
The findings should be interpreted with caution due to the uncertainty in classifying workers into fixed-term and permanent contract categories.
It is somewhat surprising that individual-level findings contradict company-level findings; under-reporting may play a role. Moreover, we also have to keep in mind the ecological and atomistic fallacies.

« previous

According to Figure 3.105 (a) and Figure 3.105 (b), workers with limited contracts are 30% less likely to report OAs (both commuting and workplace accidents) compared to those with unlimited contracts, even after accounting for significant differences in company representativeness. This contrasts with findings in the literature, which report a positive relationship between temporary employment and occupational injuries (Koranyi et al., 2018). However, the same authors caution against using registered data as an outcome, as this may bias the results due to underreporting. Other possible explanations for this inconsistency could include uncertainty in how contract types are defined in the current analyses or the specific jobs performed by workers with fixed-term contracts.

According to Figure 3.106 (a) and Figure 3.106 (b), contract type does not significantly determine the likelihood of a reported OA being rejected by the insurer.

According to Figure 3.108 (a) and Figure 3.108 (b), workers with fixed-term contracts are less likely to have accepted commuting accidents compared to those with permanent contracts. This topic has not been extensively studied in scientific research yet. Possible explanations could include bias due to underreporting or differences in commuting distance or mode of transportation.

Figure 3.109 (a) and Figure 3.109 (b) show the estimates for the relation between contract type categories and an accepted workplace OA. Once again, workers with fixed-term/limited contracts are less likely to experience accepted workplace accidents compared to those with permanent/unlimited contracts. This finding contradicts most previous research (Koranyi et al., 2018). Possible explanations for these inconsistencies include underreporting, uncertainty in how contract types are defined in the current analyses, or the specific jobs performed by workers with fixed-term contracts.

Figure 3.110 (a) and Figure 3.110 (b) present a model estimating the frequency degree of OAs at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures indicate that a higher proportion of workers with a contract of unlimited duration significantly increases the frequency of OAs within a company. This is in line with the models above, which suggest that workers with a fixed-term contract have a lower chance of both commuting and workplace OAs. It is plausible that such contracts are more common in companies with higher risks for OAs. Additionally, underreporting of accidents in this specific group may also play a role in the explanation of these findings.

Figure 3.111 (a) and Figure 3.111 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that contract type does not significantly determine whether an workplace OA is classified as serious. This contradicts earlier findings in the literature, which indicate that workers with temporary contracts are more likely to suffer severe injuries from workplace accidents compared to permanent workers (Picchio & Ours, 2016).

Figure 3.112 (a) and Figure 3.112 (b) show the results of a model assessing the relationship between contract type and the number of absenteeism days. These figures are based on absenteeism data from FEDRIS, while Figure 3.113 (a) and Figure 3.113 (b) present the same model using data from Liantis PS. The figures indicate that contract category is a significant factor in determining the validated number of absenteeism days due to OAs. The number of accepted days off decreases significantly for a contract with limited duration (e.g., 18% less days for a contract with limited duration).

As mentioned above, Figure 3.113 (a) and Figure 3.113 (b) are the results of the model assessing the relation between contract type and number of absenteeism days related to the workplace accident, based on the Liantis PS data. Although the second model is based on a smaller sample size, findings aligns those of the model based on the FEDRIS absenteeims data, showing a lower number of absenteeism days for workers with a contract of limited duration. This again contradicts research that expect workers with fixed-term contracts having more absenteeism days due to OAs. However, the authors caution against using absenteeism as an outcome due to the risk of underreporting (Koranyi et al., 2018).

Figure 3.114 (a) and Figure 3.114 (b) show the results of a model evaluating the relationship between contract type and the direct wage cost of an OA. The findings indicate that contract type significantly impacts the direct wage cost for employers: contracts with limited duration incur lower costs compared to unlimited contracts. This aligns with expectations given the results of the former models regarding the relation with absenteeism.

The next 4 figures (Figure 3.115 (a) and Figure 3.115 (b), Figure 3.116 (a) and Figure 3.116 (b)) are displaying the results of a model assessing the relation between proportion of workers within a contract type and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.115 (a) and Figure 3.115 (b)), validated number of absenteeism days from FEDRIS were used, while the second model Figure 3.116 (a) and Figure 3.116 (b)) was based on the numbers of days off in the Liantis PS data. The results of both models indicate that a higher proportion of workers with a contract of unlimited duration significantly increases the severity degree within a company.

3.4.2.3 Nature of the contract (full-time versus part-time employment)

3.4.2.3.1 Descriptives

The NSSO provides data on the nature of contracts in four categories: full-time, part-time, seasonal, or undetermined. Additionally, the FEDRIS statistical reports and notifications include information about the nature of contracts in three categories: full-time, part-time, and unknown.

Since Liantis PS data contains the number of hours an employee is expected to work per week, as well as the number of weekly hours an employee in the same role would work (the numerator and denominator of the so-called employment quotient), this quotient can be used as a proxy for full-time or part-time employment (\(\geq\) 1.00 is considered full-time, <1.00 as part-time, and unknown if the numerator and/or denominator are missing).

This allows us to categorize the nature of contracts in the same categories used in the FEDRIS reports and notifications. Given that the nature of contract categories is present in the notifications, the fit between the derived nature of the contract for an employee experiencing an OA can be compared with the corresponding nature of the contract classification in the accident notification.

According to Figure 3.117, the contract categories in Liantis PS data and FEDRIS data are quite comparable. This indicates that the contract categories derived from Liantis PS data might align well with those in FEDRIS data, making them suitable for further analysis. The combined accuracy is 85%.

Figure 3.117: Concordance between FEDRIS provided and Liantis PS nature of the contract of an employee

In the next 6 tables, an overview of all (accepted) OAs (Table 3.78 and Table 3.79), the commuting OAs (Table 3.80 and Table 3.81) and the workplace OAs (Table 3.82 and Table 3.83) are given.

In the first table ((Table 3.78, Table 3.80 and Table 3.82), the absolute numbers are provided stratified per contract category, next to the total number of employees, as available in the Liantis database.

In the second table (Table 3.79, Table 3.81 and Table 3.83), the relative numbers of the OAs per contract category are provided as percentages.

From the table Table 3.79, it can be noticed that full-time workers account for 65% of the employees within Liantis customers, while they account for about 75% of the employees within the notifications. For the commuting OAs (Table 3.81), these percentage are somewhat more balanced.

all accepted occupational accidents

Table 3.78: Numbers of employees per nature of contract category (nsso and liantis) and numbers of occupational accident notifications per nature of contract category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
fulltime	96371160	NA	11797777	0	119078	33177	13717
parttime	51035685	NA	6257299	0	40305	12959	7190
seasonal	5504361	NA	NA	NA	NA	NA	NA
unknown	256936	NA	790	0	35836	15099	5123

Table 3.79: Percentages of employees per nature of contract category (nsso and liantis) and percentages of occupational accident notifications per nature of contract category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
fulltime	65.4	NA	65.3	NaN	74.7	71.9	65.6
parttime	34.6	NA	34.7	NaN	25.3	28.1	34.4
seasonal	NA	NA	NA	NA	NA	NA	NA
unknown	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.80: Numbers of employees per nature of contract category (nsso and liantis) and numbers of occupational accident notifications per nature of contract category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
fulltime	96371160	NA	11797777	0	15927	4045	1946
parttime	51035685	NA	6257299	0	7972	2581	1391
seasonal	5504361	NA	NA	NA	NA	NA	NA
unknown	256936	NA	790	0	4488	1973	775

Table 3.81: Percentages of employees per nature of contract category (nsso and liantis) and percentages of occupational accident notifications per nature of contract category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
fulltime	65.4	NA	65.3	NaN	66.6	61	58.3
parttime	34.6	NA	34.7	NaN	33.4	39	41.7
seasonal	NA	NA	NA	NA	NA	NA	NA
unknown	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.82: Numbers of employees per nature of contract category (nsso and liantis) and numbers of occupational accident notifications per nature of contract category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
name	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
fulltime	96371160	NA	11797777	691920	103151	29132	11771
parttime	51035685	NA	6257299	243576	32333	10378	5799
seasonal	5504361	NA	NA	NA	NA	NA	NA
unknown	256936	NA	790	191571	31348	13126	4348

Table 3.83: Percentages of employees per nature of contract category (nsso and liantis) and percentages of occupational accident notifications per nature of contract category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
name	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
fulltime	65.4	NA	65.3	74	76.1	73.7	67
parttime	34.6	NA	34.7	26	23.9	26.3	33
seasonal	NA	NA	NA	NA	NA	NA	NA
unknown	NA	NA	NA	NA	NA	NA	NA

3.4.2.3.2 Models

The following Table 3.84 summarizes the results of the models assessing the relationship between contract and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both in a table and graphically.

Table 3.84: Overview of the results of the different models examining the relation between nature of the contract category and Occupational Accidents

catnature	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Part-time	REF	REF/>	REF	REF	rank def	REF	</ns	<	rank def
Full-time	>	</REF	ns	>	>	>	REF	REF	>/>

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Nature of the contract (full-time versus part-time employment) and occupational accidents

Full-time workers have a higher chance to report an occupational accident (and have a higher chance for both an accepted commuting as an accepted workplace occupational accident).
Part-time workers seem to show a higher chance that a reported occupational accident is refused by an insurer. A multivariable model should confirm this finding.
Full-time workers have a higher chance for an occupational accident being serious in comparison with part-time workers.
Inconsistent results were found for the relation between nature of contract type and number of absenteeism days.
A higher proportion of full-time workers significantly increases the frequency and severity degree within a company.

« previous

Figure 3.118 (a) and Figure 3.118 (b) illustrate that workers with a full-time contract have 55% more chance of reporting an OA (both commuting and workplace accidents) compared to the category with a part-time contract, even after accounting for significant differences in company representativeness. It should be noted that in literature, findings are inconsistent: some literature reports higher likelihood for OAs for full-time workers (Alamgir et al., 2008; Picchio & Ours, 2017), while other findings suggest the opposite (Alali et al., 2016).

It is possible that part-time workers are less familiar with workplace routines, which may increase the risk of accidents, although they may be less likely to report them. Moreover, a substantial body of literature treats part-time status as a proxy for fixed-term or precarious contracts, which may also be concentrated in specific sectors.

According to Figure 3.119 (a) and Figure 3.119 (b) (Liantis PS data) and Figure 3.120 (a) and Figure 3.120 (b) (Fedris notification data), the workers with a full-time contract have a lower chance for a reported OA being refused by the insurer. This may be because part-time workers are less familiar with the procedures for reporting OAs, potentially leading to late reports or incorrect procedures, which makes the claims more disputable (Kreshpaj et al., 2021).

According to Figure 3.121 (a) and Figure 3.121 (b), full-time employment does not seem to determine the chance of experiencing an accepted commuting accident. We are not aware of specific studies examining this relationship (more commonly studied variables include commuting time and work-home distance).

Figure 3.122 (a) and Figure 3.122 (b) show the estimates for the relation between contract categories and an accepted workplace OA. Full-time workers have a higher likelihood (65% more than part-time workers) for experiencing an accepted workplace OA.

Figure 3.123 (a) and Figure 3.123 (b) present a model estimating the frequency degree of OAs at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures indicate that a higher proportion of full-time workers significantly increases the frequency of OAs within a company.

Figure 3.124 (a) and Figure 3.124 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that the nature of the contract category does significantly determine whether an workplace OA is classified as serious: for full-time workers, the chance for an OA being classified as serious is 81% higher than for part-time workers.

Figure 3.125 (a) and Figure 3.125 (b) show the results of a model assessing the relationship between contract category and the number of absenteeism days. These figures are based on absenteeism data from FEDRIS, while Figure 3.126 (a) and Figure 3.126 (b) present the same model using data from Liantis PS. The figures indicate that contract category is a significant factor in determining the validated number of absenteeism days due to OAs. The number of accepted days off decreases significantly for a part-time contract (e.g., 6% less days for a part-time contract).

As mentioned above, Figure 3.126 (a) and Figure 3.126 (b) are the results of the model assessing the relation between contract category and number of absenteeism days related to the workplace accident, based on the Liantis PS data. The results conflict with those of the previous model. They indicate that part-time workers have slightly more accepted days off than full-time workers, although the results are not significant. These inconsistent findings may be due to differences in the study populations between the Liantis PS data and the FEDRIS absenteeism data. The Liantis PS dataset is considerably smaller, with fewer than half the number of employers and less than one-fifth the number of employees.

Figure 3.127 (a) and Figure 3.127 (b) show the results of a model evaluating the relationship between contract category and the direct wage cost of an OA. The findings indicate that contract type significantly impacts the direct wage cost for employers: part-time contracts incur lower costs compared to full-time contracts, which aligns with expectations.

The next 4 figures (Figure 3.128 (a) and Figure 3.128 (b), Figure 3.129 (a) and Figure 3.129 (b)) are displaying the results of a model assessing the relation between proportion of workers within a contract category and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.128 (a) and Figure 3.128 (b)), validated number of absenteeism days from FEDRIS were used, while the second model (Figure 3.129 (a) and Figure 3.129 (b)) was based on the numbers of days off in the Liantis PS data. The results of both models indicate that a higher proportion of full-time workers significantly increases the severity level within a company.

3.4.2.4 Working in shifts and long working hours

Data Quality Alert: working in shifts

Liantis ESPP risk codes for night and shift work are available but not always explicitly negatively validated
Many assumptions would have to be made to use the available information in modelling

Within Liantis, no systematically maintained variables related to working in shifts are currently available that would allow for investigation of a potential association with occupational accidents. Similarly, the FEDRIS notification data does not include any variables related to working in shifts. As a result, no further analyses or modelling efforts were undertaken in relation to working in shifts.

« previous

Data Quality Alert: long working hours

To our knowledge, no data quantitative data on long working hours is at this moment easily accessible in the Liantis ESPP or PS data
In the FEDRIS occupational accident data, information on the duration of the shift is indirectly present and be calculated (in 60% of the notifications) from the difference between the end and start of the work shift (see Section 2.7.2.7)
Many assumptions would have to be made to use the available information in modelling

Within Liantis, no systematically maintained variables related to long working hours are currently available that would allow for investigation of a potential association with occupational accidents. Additionally, the FEDRIS notification data do not include detailed quantitative information on the length of shifts. As a result, no further analyses or modelling efforts were undertaken in relation to long working hours.

« previous

3.4.2.5.1 Usual professional work activity

Data Quality Alert: usual professional work activity

Liantis (neither ESPP nor PS) does not have information on the frequencies with which all employees are exposed to work activities deviating from the usual professional work activity like FEDRIS describes in the context of occupational accidents
Information on deviation of the usual professional work activity is only present in occupational accident notifications received via FEDRIS (see Section 2.7.2.10)

Within Liantis, no systematically maintained variables related to deviations from usual professional work activities are currently available that would allow for investigation of a potential association with occupational accidents. As a result, no further analyses or modelling efforts were undertaken in relation to the usual professional work activity.

« previous

3.4.2.5.2 Specific activity

Data Quality Alert: specific activity

Liantis (neither ESPP nor PS) does not have information on the frequencies with which all employees exhibit specific activities or work processes in a structured way like FEDRIS or European Statistics on Accidents at Work (ESAW) describe in the context of occupational accidents
Information on specific activity is only present in occupational accident notifications received via FEDRIS (see Section 2.7.2.8)

Within Liantis, no systematically maintained variables related to specific activities are currently available that would allow for investigation of a potential association with occupational accidents. As a result, no further analyses or modelling efforts were undertaken in relation to the specific activity.

« previous

3.4.2.5.3 Type of workstation

Data Quality Alert: type of workstation

Liantis (neither ESPP nor PS) does not have information on the frequencies with which all employees exhibit activities at a specific type of workstation in a structured way like FEDRIS or ESAW describe in the context of occupational accidents
Information on type of workstation is only present in occupational accident notifications received via FEDRIS (see Section 2.7.2.9)

Within Liantis, no systematically maintained variables related to type of workstation are currently available that would allow for investigation of a potential association with occupational accidents. As a result, no further analyses or modelling efforts were undertaken in relation to type of workstation.

« previous

3.4.2.5.4 Risks (ESPP)

Data Quality Alert: risks

Liantis ESPP assigns risk codes (see Section 2.11) to employees. However, assumptions must be made to use these risk codes in statistical modelling.
When a certain risk is present, the corresponding risk code is assigned. Positive or negative validation is not performed monthly, so we assume the employee is exposed to the risk only if the risk code is valid.
Since there are more than 40 risk code groups, with large variation in both the number of codes and their frequency of use (some are assigned very often, others almost never), our models struggled to converge. Many risk code groups had to be excluded to achieve convergence.
Therefore, we chose to focus only on comparing univariable models for the likelihood of experiencing an accepted commuting versus an accepted workplace OA.

« previous

Only screen work is significantly associated with a lower likelihood of workplace occupational accidents

Most predictors in the univariable models show wide confidence intervals and non-significant p-values, indicating low statistical power and high uncertainty.
Screen work is the only statistically significant predictor for workplace occupational accidents, showing a protective effect (OR = 0.35, p = 0.001); no predictors were significant for commuting occupational accidents.
The safety function variable shows a non-significant trend: possibly protective for commuting OA (OR = 0.55) and slightly risk-enhancing for workplace OA (OR = 1.23), suggesting role-specific dynamics.
Model quality is dominated by random effects (conditional R² and ICC > 0.96), meaning contextual factors explain most of the variance, while fixed-effect predictors contribute very little (marginal R² < 0.01).
Because of the low added value of the risk code predictors in these univariable models, the convergence problems, and the risk of further waterfall loss of cases, multivariable models using the risk code predictor will not be pursued further.

« previous

In Figure 3.130, a graphical representation of the commuting and workplace model is provided.

Both models are shown side-by-side in Figure 3.131 for easier comparison. The explained variance is very similar in both models, with conditional R² values above 0.96 and marginal R² values below 0.01. This indicates that the random effects (company and employee) explain most of the variance, while the fixed effects (risk codes) contribute very little. Screen work is the only risk code showing a statistically significant association, specifically a protective effect against workplace OA (OR = 0.35, p = 0.001). No risk codes are significantly associated with a higher likelihood of commuting or workplace OA in our ten-year dataset.

3.4.3 Organisation related factors

3.4.3.1 Work location

3.4.3.1.1 Descriptives

Both NSSO and FEDRIS provide statistics on the work location of the employee (Province). From Liantis PS data it is also possible to derive the Province of the employer through the postal code of an employer adress. The notification in many cases also contains the postal code of the employer. Since this information is also present in the notification information passed through to Liantis, the fit between the derived work location of the employer of an employee experiencing an OA can be compared with the corresponding work location classification in the original notification of the accident.

From Figure 3.132 we learn that the work location provinces derived from Liantis PS data and FEDRIS data are highly comparable when both are available. The correlation between the two datasets is >99%. This suggests that the work provinces derived from Liantis PS data align well with those in FEDRIS data, making them suitable for further analysis.

Figure 3.132: Concordance between FEDRIS provided and Liantis PS location of an employer

all accepted occupational accidents

Table 3.85: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
province	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Antwerpen	NA	NA	2567346	246074	23472	7938	3897
Brussel	NA	NA	577831	223134	6641	1821	149
Henegouwen	NA	NA	97651	102403	6010	967	NA
Limburg	NA	NA	1262033	98040	9081	5512	1585
Luik	NA	NA	42563	91345	20099	195	NA
Luxemburg	NA	NA	11524	17764	1268	56	NA
Namen	NA	NA	49528	40733	7799	263	NA
Oost-Vlaanderen	NA	NA	2078211	159068	31016	6759	3540
Unknown	NA	NA	1465100	29171	13324	71	10
Vlaams Brabant	NA	NA	1803293	123953	10710	5343	2380
Waals Brabant	NA	NA	61516	28287	1574	256	NA
West-Vlaanderen	NA	NA	8039270	193405	64225	32054	14469

Table 3.86: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
province	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Antwerpen	NA	NA	14.22	18.18	12.02	12.96	14.97
Brussel	NA	NA	3.20	16.49	3.40	2.97	0.57
Henegouwen	NA	NA	0.54	7.57	3.08	1.58	NA
Limburg	NA	NA	6.99	7.24	4.65	9.00	6.09
Luik	NA	NA	0.24	6.75	10.30	0.32	NA
Luxemburg	NA	NA	0.06	1.31	0.65	0.09	NA
Namen	NA	NA	0.27	3.01	4.00	0.43	NA
Oost-Vlaanderen	NA	NA	11.51	11.75	15.89	11.04	13.60
Unknown	NA	NA	8.11	2.16	6.83	0.12	0.04
Vlaams Brabant	NA	NA	9.99	9.16	5.49	8.73	9.14
Waals Brabant	NA	NA	0.34	2.09	0.81	0.42	NA
West-Vlaanderen	NA	NA	44.52	14.29	32.90	52.35	55.59

commuting accepted occupational accidents

Table 3.87: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
province	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Antwerpen	NA	NA	2567346	50144	3810	1308	713
Brussel	NA	NA	577831	51871	1853	322	21
Henegouwen	NA	NA	97651	9139	500	73	NA
Limburg	NA	NA	1262033	14967	1150	763	245
Luik	NA	NA	42563	8995	1429	24	NA
Luxemburg	NA	NA	11524	1549	107	5	NA
Namen	NA	NA	49528	4508	851	20	NA
Oost-Vlaanderen	NA	NA	2078211	26746	5598	961	521
Unknown	NA	NA	1465100	4954	2607	22	2
Vlaams Brabant	NA	NA	1803293	23539	1720	740	385
Waals Brabant	NA	NA	61516	3914	219	36	NA
West-Vlaanderen	NA	NA	8039270	25984	8543	4325	2225

Table 3.88: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
province	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Antwerpen	NA	NA	14.22	22.16	13.42	15.21	17.34
Brussel	NA	NA	3.20	22.92	6.53	3.74	0.51
Henegouwen	NA	NA	0.54	4.04	1.76	0.85	NA
Limburg	NA	NA	6.99	6.61	4.05	8.87	5.96
Luik	NA	NA	0.24	3.97	5.03	0.28	NA
Luxemburg	NA	NA	0.06	0.68	0.38	0.06	NA
Namen	NA	NA	0.27	1.99	3.00	0.23	NA
Oost-Vlaanderen	NA	NA	11.51	11.82	19.72	11.18	12.67
Unknown	NA	NA	8.11	2.19	9.18	0.26	0.05
Vlaams Brabant	NA	NA	9.99	10.40	6.06	8.61	9.36
Waals Brabant	NA	NA	0.34	1.73	0.77	0.42	NA
West-Vlaanderen	NA	NA	44.52	11.48	30.09	50.30	54.11

workplace accepted occupational accidents

Table 3.89: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
province	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Antwerpen	NA	NA	2567346	195930	19662	6630	3184
Brussel	NA	NA	577831	171263	4788	1499	128
Henegouwen	NA	NA	97651	93264	5510	894	NA
Limburg	NA	NA	1262033	83073	7931	4749	1340
Luik	NA	NA	42563	82350	18670	171	NA
Luxemburg	NA	NA	11524	16215	1161	51	NA
Namen	NA	NA	49528	36225	6948	243	NA
Oost-Vlaanderen	NA	NA	2078211	132322	25418	5798	3019
Unknown	NA	NA	1465100	24217	10717	49	8
Vlaams Brabant	NA	NA	1803293	100414	8990	4603	1995
Waals Brabant	NA	NA	61516	24373	1355	220	NA
West-Vlaanderen	NA	NA	8039270	167421	55682	27729	12244

Table 3.90: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
province	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Antwerpen	NA	NA	14.22	17.38	11.79	12.60	14.53
Brussel	NA	NA	3.20	15.20	2.87	2.85	0.58
Henegouwen	NA	NA	0.54	8.27	3.30	1.70	NA
Limburg	NA	NA	6.99	7.37	4.75	9.02	6.11
Luik	NA	NA	0.24	7.31	11.19	0.32	NA
Luxemburg	NA	NA	0.06	1.44	0.70	0.10	NA
Namen	NA	NA	0.27	3.21	4.16	0.46	NA
Oost-Vlaanderen	NA	NA	11.51	11.74	15.24	11.02	13.77
Unknown	NA	NA	8.11	2.15	6.42	0.09	0.04
Vlaams Brabant	NA	NA	9.99	8.91	5.39	8.74	9.10
Waals Brabant	NA	NA	0.34	2.16	0.81	0.42	NA
West-Vlaanderen	NA	NA	44.52	14.85	33.38	52.68	55.86

3.4.3.1.2 Models

The following Table 3.91 table summarizes the results of the models assessing the relationship between work location categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.91: Overview of the results of the different models examining the relation between work location and Occupational Accidents

catworkloc	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Antwerp	REF	REF	REF	REF	ns	REF	REF	REF	ns
Brussel	<	ns	ns	<	<	ns	>/>	ns	ns
Hainaut	>	ns	ns	>	>	ns	ns	ns	ns
Limburg	ns	ns	ns	ns	<	ns	ns	<	ns
Liège	ns	ns	ns	ns	ns	ns	ns	ns	ns
Luxembourg	ns	ns	ns	>	ns	ns	ns	ns	ns
Namur	ns	ns	ns	ns	ns	ns	ns	ns	ns
East Fl	>	ns	ns	ns	ns	ns	ns	ns	ns
Fl Brabant	ns	ns	ns	ns	ns	ns	>/ns	ns	ns
Wal Brabant	ns	ns	ns	ns	ns	ns	ns	ns	ns
West-Fl	>	ns	>	>	>	ns	</ns	<	ns

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Work location (province) and occupational accidents

Employees working in West-Flanders have a significantly higher likelihood of reporting occupational accidents (both commuting and workplace accidents) compared to those working in Antwerp
The province in which an employee is working does not significantly determine the likelihood of refusal of the occupational accident.
The figures suggest that employees who are working in Brussels (and Flemish Brabant) have a higher number of absenteeism days due to occupational accidents but that the location (province) of the employer does not significantly determine the direct wage cost related to an occupational accident.
A higher proportion of employees working in Brussels and West-Flanders significantly lowers the frequency degree of an employer.

« previous

From Figure 3.133 (a) and Figure 3.133 (b) we learn that -relative to employees working in Antwerp- employees with company locations in Hainaut, East Flanders and West Flanders have a significant higher chance to notify an OA (but employees working in Brussels a lower one). The lower odds for Brussels found in the present study is not in line with Federal Agency for Occupational Risks (2023) and might partially be explained by a higher proportion of white-collar workers in a smaller proportion of Liantis customers in this Region.

According to Figure 3.134 (a) and Figure 3.134 (b), the province in which a worker works does not significantly determine the likelihood of a reported OA being rejected by the insurer.

According to Figure 3.135 (a) and Figure 3.135 (b), working in West Flanders significantly elevates the likelihood to have an accepted commuting accident.

Figure 3.136 (a) and Figure 3.136 (b) show that working in West Flanders, Luxembourg and Hainaut significantly increases the likelihood to have an accepted workplace accident, while working in Brussels decreases it. The lower odds for Brussels found in the present study is not in line with Federal Agency for Occupational Risks (2023) and might partially be explained by a higher proportion of white-collar workers in a smaller proportion of Liantis customers in this Region.

Figure 3.137 (a) and Figure 3.137 (b) present a model estimating the frequency degree of OAs at the company level, contrasting with previous models that focused on individual employee accident probabilities. These figures show that a higher proportion of workers working in West-Flanders and Hainaut significantly increases the frequency degree of OAs within a company. On the contrary, a higher proportion workers from Limbourg and Brussels, significantly decreases the frequency degree of a company.

Figure 3.138 (a) and Figure 3.138 (b) present a model estimating the likelihood of an accepted accident being classified as serious according to Belgian law. These figures indicate that the province an employee is working in does not significantly determine the likelihood an OA is classified as serious.

Figure 3.139 (a) and Figure 3.139 (b) show the results of a model assessing the relationship between work location and the number of absenteeism days. These figures are based on absenteeism data from FEDRIS, while Figure 3.140 (a) and Figure 3.140 (b) present the same model using data from Liantis PS. The figures indicate that workers who are working in Brussels and Flemish Brabant have a higher validated number of absenteeism days due to OAs (but not in West-Flanders). The number of validated accepted days off increases significantly for workers in Brussels, Flemish Brabant and Limbourg, with respectively 50% and 13% (but decreases in West-Flanders with 9%) in comparison with employees working in Antwerp.

The model using Liantis PS calculated days partially confirms the previous model and shows a 34% increase in days off for working in Brussels.

Figure 3.141 (a) and Figure 3.141 (b) show the results of a model evaluating the relationship between work location and the direct wage cost of an OA. The findings indicate some significant differences in direct wage cost related to an OA by the province an employer is located: for Limburg and West Flanders a significant lower result is found.

The next 4 figures (Figure 3.142 (a) and Figure 3.142 (b), Figure 3.143 (a) and Figure 3.143 (b)) are displaying the results of a model assessing the relation between proportion of workers living within a province and the severity degree of a company, which is again a variable at the company level. In the first model (Figure 3.142 (a) and Figure 3.142 (b)), validated number of absenteeism days from FEDRIS were used, while the second model Figure 3.143 (a) and Figure 3.143 (b)) was based on the numbers of days off in the Liantis PS data. The results of both models do not show an impact of the proportion of workers working in certain province on the severity degree of an employer.

3.4.3.2 Company size

3.4.3.2.1 Descriptives

Both NSSO and FEDRIS provide statistics on size of the company. From Liantis PS data it is also possible to derive the size of the company through the number of employees. The notification in many cases also contains the size category of the employer. Since this information is also present in the notification information passed through to Liantis, the fit between the derived size of the employer of an employee experiencing an OA can be compared with the corresponding size classification in the original notification of the accident.

From Figure 3.144 we learn that the size derived from Liantis PS data and FEDRIS data are highly comparable when both are available. The correlation between the two datasets is >77%. This suggests that the size derived from Liantis PS data align well with those in FEDRIS data, making them suitable for further analysis.

Figure 3.144: Concordance between FEDRIS provided and Liantis PS derived size of an employer

all accepted occupational accidents

Table 3.92: Numbers of employees per size category (nsso and liantis) and numbers of occupational accident notifications per size category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catsizef	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
1-4 werknemers	7802224	7713792	3928642	79600	14015	10020	1973
5-9 werknemers	6645987	6568088	2277552	68291	12397	7594	1517
10-19 werknemers	8167741	8054595	3191538	93495	16278	8428	1885
20-49 werknemers	12750195	12436450	3421511	161909	28428	10819	3360
50-99 werknemers	9191192	8338008	1435969	121021	21174	6032	3578
100-199 werknemers	10164557	8593060	1219028	129500	23599	5382	3533
200-499 werknemers	13610892	11198182	1060267	164524	29712	4508	2872
500-999 werknemers	9821180	8117406	879660	115253	18818	5095	3966
≥ 1000 werknemers	47931892	23257774	641699	404541	30714	3338	3338
Onbekend	NA	NA	NA	15243	84	19	8

Table 3.93: Percentages of employees per size category (nsso and liantis) and percentages of occupational accident notifications per size category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catsizef	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
1-4 werknemers	6.2	8.2	21.8	5.9	7.2	16.4	7.6
5-9 werknemers	5.3	7.0	12.6	5.1	6.4	12.4	5.8
10-19 werknemers	6.5	8.5	17.7	7.0	8.3	13.8	7.2
20-49 werknemers	10.1	13.2	18.9	12.1	14.6	17.7	12.9
50-99 werknemers	7.3	8.8	8.0	9.0	10.9	9.9	13.7
100-199 werknemers	8.1	9.1	6.8	9.7	12.1	8.8	13.6
200-499 werknemers	10.8	11.9	5.9	12.3	15.2	7.4	11.0
500-999 werknemers	7.8	8.6	4.9	8.6	9.6	8.3	15.2
≥ 1000 werknemers	38.0	24.7	3.6	30.2	15.7	5.5	12.8
Onbekend	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.94: Numbers of employees per size category (nsso and liantis) and numbers of occupational accident notifications per size category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catsizef	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
1 - 4 werknemers	7802224	7713792	3928642	10480	1629	1202	259
5 - 9 werknemers	6645987	6568088	2277552	8708	1315	860	160
10 - 19 werknemers	8167741	8054595	3191538	11837	1673	970	273
20 - 49 werknemers	12750195	12436450	3421511	21823	3099	1285	430
50 - 99 werknemers	9191192	8338008	1435969	18213	2712	959	541
100 - 199 werknemers	10164557	8593060	1219028	21010	3477	998	706
200 - 499 werknemers	13610892	11198182	1060267	29600	4807	863	524
500 - 999 werknemers	9821180	8117406	879660	23039	3244	957	715
≥ 1000 werknemers	47931892	23257774	641699	79522	6420	502	502
Onbekend	NA	NA	NA	2078	11	3	2

Table 3.95: Percentages of employees per size category (nsso and liantis) and percentages of occupational accident notifications per size category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catsizef	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
1 - 4 werknemers	6.2	8.2	21.8	4.7	5.7	14.0	6.3
5 - 9 werknemers	5.3	7.0	12.6	3.9	4.6	10.0	3.9
10 - 19 werknemers	6.5	8.5	17.7	5.3	5.9	11.3	6.6
20 - 49 werknemers	10.1	13.2	18.9	9.7	10.9	14.9	10.5
50 - 99 werknemers	7.3	8.8	8.0	8.1	9.6	11.2	13.2
100 - 199 werknemers	8.1	9.1	6.8	9.4	12.3	11.6	17.2
200 - 499 werknemers	10.8	11.9	5.9	13.2	16.9	10.0	12.7
500 - 999 werknemers	7.8	8.6	4.9	10.3	11.4	11.1	17.4
≥ 1000 werknemers	38.0	24.7	3.6	35.5	22.6	5.8	12.2
Onbekend	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.96: Numbers of employees per size category (nsso and liantis) and numbers of occupational accident notifications per size category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catsizef	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
1-4 werknemers	7802224	7713792	3928642	69120	12386	8818	1714
5-9 werknemers	6645987	6568088	2277552	59583	11082	6734	1357
10-19 werknemers	8167741	8054595	3191538	81658	14605	7458	1612
20-49 werknemers	12750195	12436450	3421511	140086	25329	9534	2930
50-99 werknemers	9191192	8338008	1435969	102808	18462	5073	3037
100-199 werknemers	10164557	8593060	1219028	108490	20122	4384	2827
200-499 werknemers	13610892	11198182	1060267	134924	24905	3645	2348
500-999 werknemers	9821180	8117406	879660	92214	15574	4138	3251
≥ 1000 werknemers	47931892	23257774	641699	325019	24294	2836	2836
Onbekend	NA	NA	NA	13165	73	16	6

Table 3.97: Percentages of employees per size category (nsso and liantis) and percentages of occupational accident notifications per size category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catsizef	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
1-4 werknemers	6.2	8.2	21.8	6.2	7.4	16.8	7.8
5-9 werknemers	5.3	7.0	12.6	5.3	6.6	12.8	6.2
10-19 werknemers	6.5	8.5	17.7	7.3	8.8	14.2	7.4
20-49 werknemers	10.1	13.2	18.9	12.6	15.2	18.1	13.4
50-99 werknemers	7.3	8.8	8.0	9.2	11.1	9.6	13.9
100-199 werknemers	8.1	9.1	6.8	9.7	12.1	8.3	12.9
200-499 werknemers	10.8	11.9	5.9	12.1	14.9	6.9	10.7
500-999 werknemers	7.8	8.6	4.9	8.3	9.3	7.9	14.8
≥ 1000 werknemers	38.0	24.7	3.6	29.2	14.6	5.4	12.9
Onbekend	NA	NA	NA	NA	NA	NA	NA

3.4.3.2.2 Models

The following Table 3.98 table summarizes the results of the models assessing the relationship between company size categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.98: Overview of the results of the different models examining the relation between company size and Occupational Accidents

catsize	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
< 5	REF	REF	REF	REF	ns	REF	REF	REF	ns
5 to 9	>	ns	>	>	>	ns	ns/ns	ns	ns
10 to 19	>	ns	>	>	>	ns	</<	ns	ns
20 to 49	>	ns	>	>	>	ns	</<	<	ns
50 to 99	>	ns	>	>	>	<	</<	<	ns
100 to 199	>	>	>	>	>	<	</<	ns	ns
200 to 499	>	ns	>	>	>	<	</<	ns	ns
500 to 999	>	ns	>	>	>	<	</ns	<	ns
>= 1000	>	ns	>	ns	ns	<	</<	<	ns

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Company size and occupational accidents

The likelihood for an accepted commuting accident gradually and significantly increases with company size. The overall likelihood to report an occupational accident and the likelihood for an accepted workplace accident increases for companies up up to 199 workers and then decreases again (with wider confidence intervals).
Except for companies in the category of 100 to 199 workers, size of the company does not seem to determine the chance on refusal of an occupational accident by the insurer.
Company size significantly determines whether an workplace occupational accident is classified as serious: the larger the company, the lower the likelihood.
The number of absence days associated with an accepted workplace accidents significantly decreases with company size, as well as the direct wage cost for the employer.
At company level, company size does significantly determine the frequency degree of a company, but not the severity degree.

« previous

From Figure 3.145 (a) and Figure 3.145 (b) we learn that likelihood of notifying an OA varies substantially across employer size categories. Compared to the reference employer group (<5 employees), larger employers (up to 200 employees) generally show higher odds, the odds ratio increasing progressively with size. Beyond 200 employees, the odds ratios tend to stabilize and to show wider confidence intervals, indicating variability and potential uncertainty in the estimates.

According to Figure 3.146 (a) and Figure 3.146 (b), companies with 100 to 199 employees have a significantly higher likelihood of having an OA refused by the insurer compared to companies with fewer than 5 employees (the reference category). For other company size categories, the odds ratios do not show any significant differences.

According to Figure 3.147 (a) and Figure 3.147 (b), working within larger employers implicates a higher likelihood for employees experiencing accepted commuting accidents.

Figure 3.148 (a) and Figure 3.148 (b) show that for employers up to 200 employees, the odds ratios for accepted workplace accidents increase with company size. For larger companies, odds ratios decrease again, although the confidence intervals are wider. For companies with 1000 or more employees, the odds ratio is not significantly different from the reference category.

Figure 3.149 (a) and Figure 3.149 (b) present a model estimating the frequency degree of OAs at the company level, in contrast to previous models that focused on individual employee accident probabilities. These figures show that a larger company size corresponds to a higher frequency degree within a company.

Figure 3.150 (a) and Figure 3.150 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that company size does significantly determine whether an workplace OA is classified as serious: the larger the company, the lower the odds.

Figure 3.151 (a) and Figure 3.151 (b) and Figure 3.152 (a) and Figure 3.152 (b) do not confirm previous literature (Fabiano et al., 2004; Kines & Mikkelsen, 2003; McVittie et al., 1997) that smaller companies report a lower number of days absent from work due to an accepted workplace accidents. The trend in our results is rather that number of days decreases when company sizes get larger.

Figure 3.153 (a) and Figure 3.153 (b) demonstrate that company size significantly influences the direct wage cost incurred by an employer following an accepted workplace accident. Among companies with 50 or more employees, a cler trend emerges: the larger the company, the lower the direct wage cost. However, the confidence intervals also widen as company size increases.

The next 4 figures (Figure 3.154 (a) and Figure 3.154 (b), Figure 3.155 (a) and Figure 3.155 (b)) indicate that company size does not significantly determine the severity degree of a company.

3.4.3.3 Precarious employment

3.4.3.3.1 Subcontracting

Data Quality Alert: subcontracting

Liantis (neither ESPP nor PS) does not have information on the frequencies with which all employees are exposed to work activities as subcontractors like FEDRIS describes in the context of occupational accidents
Information on subcontracting would only be present in occupational accident notifications received via FEDRIS but almost no positive validations are detected (see Section 2.7.2.4)

Within Liantis, no systematically maintained variables related to subcontracting are currently available that would allow for investigation of a potential association with occupational accidents. As a result, no further analyses or modelling efforts were undertaken in relation to subcontracting.

« previous

3.4.3.4.1 Work stress

Data Quality Alert: work stress

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the GMQ (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported psychosocial stress and notifications of occupational accidents are presented in Table 3.99. A Pearson’s Chi-squared test of independence between hadOA and PSSTRESS suggests that the effect is statistically not significant \(\chi^2 = 0.90\), \(df = 4\), \(p = 0.924\).

Table 3.99: Notifications of occupational accidents in a certain month by self reported psychosocial stress among employees completing the GMQ in 2022 and 2023

(a) counts

	0 days	< 1 week	1 week - 1 month	1 month - 3 months	> 3 months
0	21684	9220	7163	3244	4534
1	108	49	38	16	19

(b) percentages

	0 days	< 1 week	1 week - 1 month	1 month - 3 months	> 3 months
0	99.5	99.47	99.47	99.51	99.58
1	0.5	0.53	0.53	0.49	0.42

3.4.3.4.2 Job satisfaction

Data Quality Alert: job satisfaction

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the GMQ (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported satisfaction and notifications of occupational accidents are presented in Table 3.100. A Pearson’s Chi-squared test of independence between hadOA and SATISFACTION suggests that the effect is statistically not significant \(\chi^2 = 5.07\), \(df = 4\), \(p = 0.280\).

Table 3.100: Notifications of occupational accidents in a certain month by self reported satisfaction among employees completing the GMQ in 2022 and 2023

(a) counts

	Zeer ontevreden	Ontevreden	Neutraal	Tevreden	Zeer tevreden
0	119	750	7862	25221	11893
1	0	7	46	123	54

(b) percentages

	Zeer ontevreden	Ontevreden	Neutraal	Tevreden	Zeer tevreden
0	100	99.08	99.42	99.51	99.55
1	0	0.92	0.58	0.49	0.45

3.4.3.4.3 Work relationships

Data Quality Alert: work relations

Only limited data is available (from March 2022 to December 2023, see Section 2.14)
Information comes from the GMQ (self-reported)
An assumption was made (fill downup between questionnaire moments within person) to complete missing monthly data
We only present a (preliminary) result without undertaking any further modelling efforts

« previous

Descriptives on the relation between self-reported work relations and notifications of occupational accidents are presented in Table 3.101 A Pearson’s Chi-squared test of independence between hadOA and WORKREL suggests that the effect is statistically significant \(\chi^2 = 11.53\), \(df = 4\), \(p = 0.021\).

Table 3.101: Notifications of occupational accidents in a certain month by self reported work relations among employees completing the GMQ in 2022 and 2023

(a) counts

	0 days	< 1 week	1 week - 1 month	1 month - 3 months	> 3 months
0	5107	1254	949	484	937
1	23	11	9	0	10

(b) percentages

	0 days	< 1 week	1 week - 1 month	1 month - 3 months	> 3 months
0	99.55	99.13	99.06	100	98.94
1	0.45	0.87	0.94	0	1.06

3.4.3.4.4 Safety climate and safety culture

Data Quality Alert: safety climate and safety culture

Within Liantis, no systematically maintained variables related to safety climate and safety culture are currently available that would allow for investigation of a potential association with occupational accidents. Similarly, the FEDRIS notification data does not include any variables related to safety climate and safety culture. As a result, no further analyses or modelling efforts were undertaken in relation to safety climate and safety culture.

« previous

3.4.3.5.1 Descriptives

Neither NSSO nor FEDRIS provide statistics on the insurer of the employee. However, each OA notification includes the legal insurer’s identification number. Through Liantis Risk Solutions (RS), it is possible to link annual insurer fee payments for Liantis customers.

To estimate the insurer for employees who did not experience an OA, we applied the following method: we grouped data by employer, sorted it chronologically, and propagated the most recently known insurer number forward until a new insurer number was identified. Either through an employee’s OA notification or through Liantis RS data.

If no Liantis RS data was available for an employer and no OA occurred during the study period, we were unable to infer the insurer. These cases were excluded from the analysis.

Data Quality Alert: insurance companies

Liantis (neither via ESPP nor PS) has structurally available data on the occupational accident (OA) insurer for all its client employers, as described by FEDRIS in the context of occupational accidents.
However, since insurer information is consistently included in OA notifications received via FEDRIS (see Section 2.7.3.9), and OA insurance is typically managed at the employer level rather than the employee level, it is often possible to infer the insurer for employees who did not experience an OA.

Currently, Liantis does not maintain any systematically structured variables related to the occupational accident insurer. Nevertheless, with minimal assumptions, it may be feasible to explore whether the insurer has an influence on the occurrence and severity of occupational accidents.

« previous

all accepted occupational accidents

Table 3.102: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catnrins	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
pseudins01	NA	NA	NA	NA	9715	4470	1601
pseudins02	NA	NA	NA	NA	18340	9459	4455
pseudins03	NA	NA	NA	NA	15804	4231	1900
pseudins04	NA	NA	NA	NA	3951	1756	651
pseudins05	NA	NA	NA	NA	18396	9881	3664
pseudins06	NA	NA	NA	NA	28	NA	NA
pseudins07	NA	NA	NA	NA	12799	2	NA
pseudins08	NA	NA	NA	NA	35943	9209	3533
pseudins09	NA	NA	NA	NA	7632	3138	1669
pseudins10	NA	NA	NA	NA	7421	2383	750
pseudins11	NA	NA	NA	NA	21080	5813	3581
pseudins12	NA	NA	NA	NA	44086	10877	4226
pseudins13	NA	NA	NA	NA	24	16	NA

Table 3.103: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catnrins	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
pseudins01	NA	NA	NA	NA	4.98	7.30	6.15
pseudins02	NA	NA	NA	NA	9.39	15.45	17.11
pseudins03	NA	NA	NA	NA	8.10	6.91	7.30
pseudins04	NA	NA	NA	NA	2.02	2.87	2.50
pseudins05	NA	NA	NA	NA	9.42	16.14	14.08
pseudins06	NA	NA	NA	NA	0.01	NA	NA
pseudins07	NA	NA	NA	NA	6.56	0.00	NA
pseudins08	NA	NA	NA	NA	18.41	15.04	13.57
pseudins09	NA	NA	NA	NA	3.91	5.12	6.41
pseudins10	NA	NA	NA	NA	3.80	3.89	2.88
pseudins11	NA	NA	NA	NA	10.80	9.49	13.76
pseudins12	NA	NA	NA	NA	22.58	17.76	16.24
pseudins13	NA	NA	NA	NA	0.01	0.03	NA

commuting accepted occupational accidents

Table 3.104: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catnrins	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
pseudins01	NA	NA	NA	NA	1470	698	337
pseudins02	NA	NA	NA	NA	2942	1452	755
pseudins03	NA	NA	NA	NA	2089	548	291
pseudins04	NA	NA	NA	NA	526	215	89
pseudins05	NA	NA	NA	NA	2512	1338	526
pseudins07	NA	NA	NA	NA	2501	1	NA
pseudins08	NA	NA	NA	NA	5708	1322	656
pseudins09	NA	NA	NA	NA	468	243	170
pseudins10	NA	NA	NA	NA	1291	409	114
pseudins11	NA	NA	NA	NA	3225	1004	621
pseudins12	NA	NA	NA	NA	5654	1368	553
pseudins13	NA	NA	NA	NA	1	1	NA

Table 3.105: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catnrins	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
pseudins01	NA	NA	NA	NA	5.18	8.12	8.20
pseudins02	NA	NA	NA	NA	10.36	16.89	18.36
pseudins03	NA	NA	NA	NA	7.36	6.37	7.08
pseudins04	NA	NA	NA	NA	1.85	2.50	2.16
pseudins05	NA	NA	NA	NA	8.85	15.56	12.79
pseudins07	NA	NA	NA	NA	8.81	0.01	NA
pseudins08	NA	NA	NA	NA	20.11	15.37	15.95
pseudins09	NA	NA	NA	NA	1.65	2.83	4.13
pseudins10	NA	NA	NA	NA	4.55	4.76	2.77
pseudins11	NA	NA	NA	NA	11.36	11.68	15.10
pseudins12	NA	NA	NA	NA	19.92	15.91	13.45
pseudins13	NA	NA	NA	NA	0.00	0.01	NA

workplace accepted occupational accidents

Table 3.106: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
catnrins	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
pseudins01	NA	NA	NA	NA	8245	3772	1264
pseudins02	NA	NA	NA	NA	15398	8007	3700
pseudins03	NA	NA	NA	NA	13715	3683	1609
pseudins04	NA	NA	NA	NA	3425	1541	562
pseudins05	NA	NA	NA	NA	15884	8543	3138
pseudins06	NA	NA	NA	NA	28	NA	NA
pseudins07	NA	NA	NA	NA	10298	1	NA
pseudins08	NA	NA	NA	NA	30235	7887	2877
pseudins09	NA	NA	NA	NA	7164	2895	1499
pseudins10	NA	NA	NA	NA	6130	1974	636
pseudins11	NA	NA	NA	NA	17855	4809	2960
pseudins12	NA	NA	NA	NA	38432	9509	3673
pseudins13	NA	NA	NA	NA	23	15	NA

Table 3.107: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
catnrins	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
pseudins01	NA	NA	NA	NA	4.94	7.17	5.77
pseudins02	NA	NA	NA	NA	9.23	15.21	16.88
pseudins03	NA	NA	NA	NA	8.22	7.00	7.34
pseudins04	NA	NA	NA	NA	2.05	2.93	2.56
pseudins05	NA	NA	NA	NA	9.52	16.23	14.32
pseudins06	NA	NA	NA	NA	0.02	NA	NA
pseudins07	NA	NA	NA	NA	6.17	0.00	NA
pseudins08	NA	NA	NA	NA	18.12	14.98	13.13
pseudins09	NA	NA	NA	NA	4.29	5.50	6.84
pseudins10	NA	NA	NA	NA	3.67	3.75	2.90
pseudins11	NA	NA	NA	NA	10.70	9.14	13.50
pseudins12	NA	NA	NA	NA	23.04	18.07	16.76
pseudins13	NA	NA	NA	NA	0.01	0.03	NA

3.4.3.5.2 Models

The following Table 3.108 table summarizes the results of the models assessing the relationship between OA insurers and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.108: Overview of the results of the different models examining the relation between insurer and Occupational Accidents

catinsurer	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
pseudins01	REF	REF	REF	REF	REF	REF	REF	REF	REF
pseudins02	>	<	>	>	>	ns	</<	<	>/>
pseudins03	>	ns	ns	>	>	ns	ns/ns	ns	>/>
pseudins04	>	ns	ns	>	>	ns	</ns	ns	ns/ns
pseudins05	<	ns	ns	<	<	ns	</<	ns	</<
pseudins08	>	ns	>	>	>	ns	ns/ns	ns	>/>
pseudins09	>	<	ns	>	>	ns	ns/ns	>	>/>
pseudins10	>	ns	>	>	ns	<	</ns	ns	ns/ns
pseudins11	>	ns	>	>	>	<	</<	ns	ns/ns
pseudins12	>	ns	ns	>	>	ns	ns/ns	ns	>/>

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Insurance companies and occupational accidents

Compared to the pseudonymised reference insurer 1…
Likelihoods to notify seem to be lower with insurer 5, higher with all other insurers (9 showing the highest estimate)
Likelihoods for refusal seem to be lower with insurers 2 and 9
Likelihoods for experiencing commuting accidents appear to be higher with insurers 2, 8, 10 and 11, for workplace accidents for all insurers except 5 where it appears to be lower
Frequency degrees at company level appear to be higher for all insurers (3 and 9 showing the highest estimates) except 5 where it appears to be lower
Likelihoods for serious accidents appear to be lower for insurers 10 and 11
Number of absence days associated with an accepted workplace accident appear to be lower with insurers 2, 5 and 11
Costs appear to be lower with insurers 2, higher with insurer 9
Severity degrees at company level appear to be higher for insurers 3, 9, 8 and 12
To our knowledge, no literature is available to compare our results with (see Section 1.3.4.4)
If certain insurers are specialised in specific sectors or types of work, this could partially explain the differences observed, therefore we will also investigate the influence of the insurer as determinant in the multivariable analysis (see Section 3.5)
In any case, the present research needs to be replicated independently to confirm our results

« previous

Figure 3.156 (a) and Figure 3.156 (b) reveal substantial variation in the likelihood of notifying an OA across insurers. Compared to the reference insurer (1), most exhibit significantly higher odds ratios -with insurer 9 showing the highest value- while insurer 5 stands out with a lower odds ratio.

According to Figure 3.157 (a) and Figure 3.157 (b), only insurers 2 and 9 show significantly lower odds ratios for having an OA refused by the insurer compared to the reference insurer (1). Other insurers do not show significant differences.

According to Figure 3.158 (a) and Figure 3.158 (b), employees insured with insurers 2, 8, 10 and 11 have significantly higher odds of experiencing accepted commuting accidents compared to those insured with the reference insurer (1).

Figure 3.159 (a) and Figure 3.159 (b) indicate that, for all insurers except insurer 5, the odds ratios for accepted workplace accidents are significantly higher compared to the reference insurer (1). Insurer 5 shows a notably lower odds ratio—approximately 10% less—while insurer 9 exhibits the highest odds ratio, at around 115% more. Given the substantial differences observed, further investigation is warranted (see Section 3.5) to assess whether these disparities persist when controlling for other variables.

Figure 3.160 (a) and Figure 3.160 (b) present a model estimating the frequency degree of OA at the company level, in contrast to earlier models that focused on individual employee accident probabilities. These figures show that all insurers -except insurer 5- have significantly higher frequency degree estimates compared to the reference insurer (1). Insurers 3 and 9 stand out with the highest frequency levels, while insurer 5 shows a notably lower result. Given the magnitude of these differences, further analysis is needed (see Section 3.5) to determine whether these patterns persist when controlling for other explanatory variables.

Figure 3.161 (a) and Figure 3.161 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that only insurers 10 and 11 show significantly lower odds ratios for having an accepted workplace OA classified as serious compared to the reference insurer (1). Other insurers do not show significant differences.

Figure 3.162 (a) and Figure 3.162 (b) show that employees insured with insurers 2, 5 and 11 (and into a lesser extent 4 and 10) have significantly lower numbers of (FEDRIS validated) absence days associated with an accepted workplace OA compared to those insured with the reference insurer (1).

Figure 3.163 (a) and Figure 3.163 (b) confirm that employees insured with insurers 2, 5 and 11 have significantly lower numbers of (Liantis PS calculated) absence days associated with an accepted workplace OA compared to those insured with the reference insurer (1).

Figure 3.164 (a) and Figure 3.164 (b) show that employees insured with insurer 2 have significantly lower direct wage costs associated with an accepted workplace OA compared to those insured with the reference insurer (1). In contrast, employees insured with insurer 9 exhibit significantly higher direct wage costs.

The next 4 figures (Figure 3.165 (a) and Figure 3.165 (b), Figure 3.166 (a) and Figure 3.166 (b)) are displaying the results of models assessing the severity degree of a company in relation to its insurer. According to Figure 3.165 (a) and Figure 3.165 (b), insurers 2, 3, 8, 9 and 12 have significantly higher severity degree estimates compared to the reference insurer (1).

Figure 3.166 (a) and Figure 3.166 (b) confirm the results of the previous model with validated days.

3.4.3.6 Investments in prevention

Analysing the relationship between time invested in prevention and the occurrence of OA presents several methodological challenges, particularly when considering prevention time as a leading or predictive factor. To meaningfully assess this relationship, a sufficiently large and diverse dataset across multiple companies is required. This is especially important given that many organizations (fortunately) report zero occupational accidents, which limits the variability needed to detect patterns or vulnerabilities.

A preliminary aggregate-level analysis was performed using data exclusively from Liantis ESPP clients. This dataset comprises monthly records per company, including the number of employees, commuting and workplace OA notifications, and the total registered prevention time. Importantly, the recorded prevention time excludes any hours spent responding to OA after their occurrence. For further details on the time registration data, refer to Section 2.12.

Prior to aggregation, the dataset contains the total number of:

employees (per company, per year, per month);
commuting OA (per company, per year, per month);
workplace OA (per company, per year, per month);
hours invested by Liantis ESPP in preventive activities (per company, per year, per month).

By aggregating across all Liantis ESPP clients for each year and month, we construct a time series of 120 monthly totals for these four variables. To account for fluctuations in workforce size -whether due to internal changes within companies or shifts in the Liantis ESPP client portfolio- we normalize the three outcome variables (commuting OA, workplace OA, and time invested in prevention of OA) per 1,000 employees.

Figure 3.167 presents the three normalized time series: the number of commuting OA, the number of workplace OA, and the number of hours invested in prevention (excluding post-accident response time).

Figure 3.167: Occupational *commuting* and *workplace* and hours invested in prevention (pre OA) per 1000 employees in time

At first sight, a downward trend for workplace OA (middle panel) and an upward trend for time invested in prevention (lower panel) can be noticed, while the image for commuting OA seems less clear.

Due to the pronounced fluctuations and irregular patterns in the three timeseries, further interpretation can be challenging. To address this, a time series decomposition technique, Signal Extraction in Arima Time Series (SEATS) (Gómez & Maravall, 1997), is applied using the X_13ARIMA_SEATS function of the feasts package (O’Hara-Wild et al., 2024). The SEATS method separates the series into trend, seasonal, and irregular components, offering a clearer view of the underlying dynamics.

Results from the time series decomposition analyses using SEATS are presented in the upper parts of Figure 3.168, Figure 3.169, and Figure 3.170. The upper panel of each figure displays the same original time series as shown in Figure 3.167. The second panel shows the overall trend across the ten-year study.

For commuting OA, a modest upward trend is observed starting from 2020 (see Figure 3.168 (a)).
For workplace OA, a gradual and consistent downward trend is noticed over the ten-year period. The decline appears almost linear, suggesting a series of (long-term) improvements in workplace safety. These improvements may be attributed to enhanced preventive measures, better training, or changes in notification and reporting practices, being indicative of structural changes rather than short-term interventions (see Figure 3.169 (a)).
For time invested in prevention of OA, the trend shows an increase from early 2014 to late 2019, followed by a decline until the end of 2021, and a renewed increase from late 2021 to the end of 2023. The impact of the Coronavirus Disease 2019 (COVID-19) pandemic, which is prominently visible in the raw data (upper panels comOA, wplOA, and prevOA of the three figures), only manifests as a prolonged two-year trend effect in the trend panel of Figure 3.170 (a)).

The lower parts of Figure 3.168, Figure 3.169, and Figure 3.170 display the seasonal components extracted by the SEATS method over the ten-year study period and correspond to the third panel of the upper part of each figure.

For commuting OA, a pronounced seasonal effect is observed in January, with consistently elevated accident rates during this month (see Figure 3.168 (b)).
For workplace OA, accident rates tend to peak in March, June, September, and October, while lower rates are typically seen in May, July, and December—possibly reflecting reduced activity during summer holidays or year-end periods. The seasonal pattern remains relatively stable across years, suggesting that certain months inherently carry higher risk, potentially due to weather conditions, workload fluctuations, or staffing dynamics (see Figure 3.169 (b)).
For time invested in prevention of OA, seasonal peaks are consistently observed in the first quarter of each year, followed by a dip around mid-year (see Figure 3.170 (b)). This pattern may reflect organizational planning cycles, with increased allocation of resources to prevention activities early in the year—possibly linked to budget planning or strategic initiatives—and a quieter period during July and August.

The irregular component (fourth and lowest panel of the upper part of Figure 3.168, Figure 3.169, and Figure 3.170) captures short-term fluctuations that cannot be explained by trend or seasonality. These may be due to isolated incidents or anomalies, large and sudden changes in reporting or data collection, external shocks like policy changes or pandemics,…

Investments in prevention increase, workplace occupational accidents numbers decrease

Time series decomposition using SEATS reveals distinct seasonal peaks in both commuting and workplace occupational accidents, suggesting potential for targeted interventions during high-risk months. The analysis confirms that the observed patterns are not purely random. Notably, a slight upward trend in commuting occupational accidents emerges from mid-2020 onward. In contrast, workplace OA show a clear and encouraging downward trend, which appears to coincide with a collective increase in preventive investments across the Liantis ESPP client portfolio. This raises the question whether continued investment in prevention could sustain or further reinforce both trajectories.

« previous

(a) Time series decomposition of *commuting* OA using SEATS

(a) Time series decomposition of *workplace* OA using SEATS

(a) Time series decomposition of *time invested in prevention* of OA using SEATS

Although it is impossible to establish any causal relationships from the raw and decomposed time series, the normalised increased time invested in prevention coincides with lower normalised numbers of workplace OA (but not with lower numbers of commuting OA).

Due to the limited number of observations (120 months with aggregated data available in the dataset), it is not possible to conduct a robust statistical analysis.

A scatterplot of the 120 data points (see Figure 3.171) however confirms (in blue) the negative relation between investment in prevention and workplace OA observed in time series trends: a higher amount of time invested in prevention OA is related to a lower number of workplace OA, but not to a lower number of commuting OA (in red). While these observations are based on a simplified analysis, they offer an indication that prevention investments might have a positive impact on workplace OA occurrence.

Figure 3.171: Occupational accidents and hours invested in prevention (pre OA) per 1000 employees

Again, it is important to emphasize that this analysis does not allow for causal inference. We cannot conclude, based on this data alone, that companies investing more time in prevention necessarily experience fewer workplace OA. The observed patterns warrant further investigation using more advanced statistical techniques.

Prevention investments might have a positive impact on workplace OA occurrence

While our initial findings suggest a potential link between time invested in prevention and worplace occupational accident notification rates, the limitations of the dataset and the aggregate nature of the analysis mean that no definitive conclusions can be drawn at this stage. Future research should aim to explore this relationship in greater depth, using robust modelling approaches and more detailed and high quality datasets.

« previous

3.4.4.1 Section NACE-BEL 2008

3.4.4.1.1 Descriptives

Both NSSO and FEDRIS provide statistics on the sector of the employee (NSSO section level, 1 digit and FEDRIS division level, 2 digits). From Liantis PS data it is also possible to derive the section of the employer through the subclass level (5 digits) of an employer. The notification in many cases also contains the subclass level of the employer (5 digits). Since this information is also present in the notification information passed through to Liantis, the fit between the derived section of the employer of an employee experiencing an OA can be compared with the corresponding section derived from the original notification of the accident.

From Figure 3.172 we learn that the sections derived from Liantis PS data and FEDRIS data are highly comparable when both are available. The correlation between the two datasets is 98%. This suggests that the sections derived from Liantis PS data align well with those in FEDRIS data, making them suitable for further analysis.

Figure 3.172: Correlation between Liantis PS and FEDRIS sections for the same employer

all accepted occupational accidents

Table 3.109: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
nace	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
A	953206	953206	238640	1720	1190	774	234
B	96322	96322	486	330	405	51	NA
C	18938833	18938253	1859279	84924	43836	9377	4001
D	742516	699656	5600	1251	158	14	NA
E	1094598	853266	56820	4887	3254	565	NA
F	8197964	8197964	1754346	27208	28060	12006	2747
G	19742342	19742342	2965033	57134	17293	7518	2082
H	9119531	7503286	545779	36513	11277	3121	1258
I	4471088	4470459	2537607	10891	4208	2875	498
J	4317974	4121508	211759	7154	710	235	91
K	4825380	4823556	246944	9494	1580	217	60
L	847838	841704	169100	1499	470	210	57
M	6851246	6780059	782264	13135	2273	975	301
N	15699074	15560973	924668	63615	8486	3617	1155
O	15493922	7320334	9007	3057	9878	15	NA
P	15957634	6642834	701217	6421	2819	1677	908
Q	21268309	20726706	2621790	73128	52179	14876	11719
R	1460856	1439992	248602	4731	2670	1272	354
S	2815390	2813402	631306	5988	2527	1266	327
T	147863	147863	11613	58	18	9	NA
U	126256	126256	12224	221	7	7	NA
NA	NA	NA	1521782	439697	1921	558	238

Table 3.110: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
nace	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
A	0.6	0.7	1.4	0.4	0.6	1.3	0.9
B	0.1	0.1	0.0	0.1	0.2	0.1	NA
C	12.4	14.3	11.2	20.5	22.7	15.5	15.5
D	0.5	0.5	0.0	0.3	0.1	0.0	NA
E	0.7	0.6	0.3	1.2	1.7	0.9	NA
F	5.4	6.2	10.6	6.6	14.5	19.8	10.7
G	12.9	14.9	17.9	13.8	8.9	12.4	8.1
H	6.0	5.7	3.3	8.8	5.8	5.1	4.9
I	2.9	3.4	15.3	2.6	2.2	4.7	1.9
J	2.8	3.1	1.3	1.7	0.4	0.4	0.4
K	3.2	3.6	1.5	2.3	0.8	0.4	0.2
L	0.6	0.6	1.0	0.4	0.2	0.3	0.2
M	4.5	5.1	4.7	3.2	1.2	1.6	1.2
N	10.2	11.7	5.6	15.4	4.4	6.0	4.5
O	10.1	5.5	0.1	0.7	5.1	0.0	NA
P	10.4	5.0	4.2	1.6	1.5	2.8	3.5
Q	13.9	15.6	15.9	17.7	27.0	24.5	45.4
R	1.0	1.1	1.5	1.1	1.4	2.1	1.4
S	1.8	2.1	3.8	1.4	1.3	2.1	1.3
T	0.1	0.1	0.1	0.0	0.0	0.0	NA
U	0.1	0.1	0.1	0.1	0.0	0.0	NA
NA	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.111: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
nace	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
A	953206	953206	238640	677	80	54	19
B	96322	96322	486	102	21	NA	NA
C	18938833	18938253	1859279	31170	4588	940	402
D	742516	699656	5600	1007	84	1	NA
E	1094598	853266	56820	1041	268	33	NA
F	8197964	8197964	1754346	8040	1532	591	152
G	19742342	19742342	2965033	28330	2395	1187	391
H	9119531	7503286	545779	12996	1062	150	61
I	4471088	4470459	2537607	5888	658	467	68
J	4317974	4121508	211759	5703	320	113	50
K	4825380	4823556	246944	8897	904	105	23
L	847838	841704	169100	1030	99	41	18
M	6851246	6780059	782264	10219	720	341	110
N	15699074	15560973	924668	42351	1660	717	311
O	15493922	7320334	9007	2586	2007	3	NA
P	15957634	6642834	701217	4294	567	367	182
Q	21268309	20726706	2621790	52663	10130	2922	2152
R	1460856	1439992	248602	2045	206	97	24
S	2815390	2813402	631306	4706	757	393	112
T	147863	147863	11613	36	1	1	NA
U	126256	126256	12224	138	4	4	NA
NA	NA	NA	1521782	228699	324	72	37

Table 3.112: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
nace	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
A	0.6	0.7	1.4	0.3	0.3	0.6	0.5
B	0.1	0.1	0.0	0.0	0.1	NA	NA
C	12.4	14.3	11.2	13.9	16.3	11.0	9.9
D	0.5	0.5	0.0	0.4	0.3	0.0	NA
E	0.7	0.6	0.3	0.5	1.0	0.4	NA
F	5.4	6.2	10.6	3.6	5.5	6.9	3.7
G	12.9	14.9	17.9	12.7	8.5	13.9	9.6
H	6.0	5.7	3.3	5.8	3.8	1.8	1.5
I	2.9	3.4	15.3	2.6	2.3	5.5	1.7
J	2.8	3.1	1.3	2.5	1.1	1.3	1.2
K	3.2	3.6	1.5	4.0	3.2	1.2	0.6
L	0.6	0.6	1.0	0.5	0.4	0.5	0.4
M	4.5	5.1	4.7	4.6	2.6	4.0	2.7
N	10.2	11.7	5.6	18.9	5.9	8.4	7.6
O	10.1	5.5	0.1	1.2	7.2	0.0	NA
P	10.4	5.0	4.2	1.9	2.0	4.3	4.5
Q	13.9	15.6	15.9	23.5	36.1	34.3	52.8
R	1.0	1.1	1.5	0.9	0.7	1.1	0.6
S	1.8	2.1	3.8	2.1	2.7	4.6	2.7
T	0.1	0.1	0.1	0.0	0.0	0.0	NA
U	0.1	0.1	0.1	0.1	0.0	0.0	NA
NA	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.113: Numbers of employees per province (nsso and liantis) and numbers of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
nace	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
A	953206	953206	238640	1043	1110	720	215
B	96322	96322	486	228	384	51	NA
C	18938833	18938253	1859279	53754	39248	8437	3599
D	742516	699656	5600	244	74	13	NA
E	1094598	853266	56820	3846	2986	532	NA
F	8197964	8197964	1754346	19168	26528	11415	2595
G	19742342	19742342	2965033	28804	14898	6331	1691
H	9119531	7503286	545779	23517	10215	2971	1197
I	4471088	4470459	2537607	5003	3550	2408	430
J	4317974	4121508	211759	1451	390	122	41
K	4825380	4823556	246944	597	676	112	37
L	847838	841704	169100	469	371	169	39
M	6851246	6780059	782264	2916	1553	634	191
N	15699074	15560973	924668	21264	6826	2900	844
O	15493922	7320334	9007	471	7871	12	NA
P	15957634	6642834	701217	2127	2252	1310	726
Q	21268309	20726706	2621790	20465	42049	11954	9567
R	1460856	1439992	248602	2686	2464	1175	330
S	2815390	2813402	631306	1282	1770	873	215
T	147863	147863	11613	22	17	8	NA
U	126256	126256	12224	83	3	3	NA
NA	NA	NA	1521782	210998	1597	486	201

Table 3.114: Percentages of employees per province (nsso and liantis) and percentages of occupational accident notifications per province (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
nace	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
A	0.6	0.7	1.4	0.6	0.7	1.4	1.0
B	0.1	0.1	0.0	0.1	0.2	0.1	NA
C	12.4	14.3	11.2	28.4	23.8	16.2	16.6
D	0.5	0.5	0.0	0.1	0.0	0.0	NA
E	0.7	0.6	0.3	2.0	1.8	1.0	NA
F	5.4	6.2	10.6	10.1	16.1	21.9	11.9
G	12.9	14.9	17.9	15.2	9.0	12.1	7.8
H	6.0	5.7	3.3	12.4	6.2	5.7	5.5
I	2.9	3.4	15.3	2.6	2.1	4.6	2.0
J	2.8	3.1	1.3	0.8	0.2	0.2	0.2
K	3.2	3.6	1.5	0.3	0.4	0.2	0.2
L	0.6	0.6	1.0	0.2	0.2	0.3	0.2
M	4.5	5.1	4.7	1.5	0.9	1.2	0.9
N	10.2	11.7	5.6	11.2	4.1	5.6	3.9
O	10.1	5.5	0.1	0.2	4.8	0.0	NA
P	10.4	5.0	4.2	1.1	1.4	2.5	3.3
Q	13.9	15.6	15.9	10.8	25.4	22.9	44.1
R	1.0	1.1	1.5	1.4	1.5	2.3	1.5
S	1.8	2.1	3.8	0.7	1.1	1.7	1.0
T	0.1	0.1	0.1	0.0	0.0	0.0	NA
U	0.1	0.1	0.1	0.0	0.0	0.0	NA
NA	NA	NA	NA	NA	NA	NA	NA

3.4.4.1.2 Models

The following Table 3.115 table summarizes the results of the models assessing the relationship between NACE-BEL 2008 categories and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.115: Overview of the results of the different models examining the relation between NACE-BEL 2008 and Occupational Accidents

catnace	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
A	REF	REF	REF	REF	ns	REF	REF	REF	REF
C	ns	ns	ns	ns	>	ns	ns	>	</<
F	>	ns	ns	>	>	ns	ns	>	>/ns
G	<	ns	ns	<	<	ns	ns	ns	</<
H	ns	ns	ns	ns	ns	ns	>/ns	>	ns
I	<	ns	ns	<	<	ns	</<	<	</<
J	<	ns	ns	<	<	ns	</ns	<	</<
K	<	ns	ns	<	<	ns	ns	<	</<
L	<	ns	ns	<	<	ns	</ns	ns	</<
M	<	ns	ns	<	<	ns	</<	<	</<
N	ns	ns	>	ns	ns	ns	ns/<	>	ns
P	<	ns	ns	<	<	ns	</<	<	</<
Q	<	ns	>	<	<	<	</<	<	</<
R	<	ns	ns	<	<	ns	ns	<	</<
S	<	ns	ns	<	<	ns	</ns	<	</<

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Section NACE-BEL 2008 and occupational accidents

Compared to reference sector A (Agriculture, forestry and fishing)
Likelihoods to notify seem to be higher in sector F (Construction) and equal or lower in almost all other sectors
Likelihoods for refusal do not seem to differ significantly between sectors
Likelihoods for experiencing commuting accidents appear to be higher in sectors N (Administrative and support service activities) and Q (Human health and social work activities)
Likelihoods for experiencing workplace accidents appear to be higher in sector F (Construction) and lower in almost all other sectors
Frequency degrees at company level appear to be higher in sectors C (Manufacturing) and F (Construction)
Likelihoods for serious accidents only seems to be lower in sector Q (Human health and social work activities)
Number of absence days associated with an accepted workplace accident seem to be lower in M (Professional, scientific and technical activities) and higher in H (Transportation and storage)
Costs appear to be higher in sectors C (Manufacturing), F (Construction) and H (Transportation and storage) and lower in almost all other sectors
Severity degrees at company level appear to be higher in sectors F (Construction) and lower in almost all other sectors
influence of sector as determinant needs to be further assessed in the multivariable analysis (see Section 3.5) to clarify whether the observed differences persist when adjusting for other variables

« previous

From Figure 3.173 (a) and Figure 3.173 (b) we learn that, compared to the reference sector A (Agriculture, forestry and fishing), sector F (Construction) shows significantly higher odds ratios for notifying an OA, while most other sectors exhibit significantly lower odds ratios.

According to Figure 3.174 (a) and Figure 3.174 (b), the sector of a company is not significantly associated with the odds of having a reported OA refused by the insurer.

NACE-BEL 2008 Level 1 categories were introduced as fixed effects, and companies and employees as random effects. The model is based on data of all employees. The model is adjusted for representativeness of the company (fixed effect inmutrep) regarding to sector, size and Flemish province.

Linear mixed-effects model 1.1 assessing the impact of NACE-BEL 2008 Level 1 categories on the probability of a reported OA being refused by the insurer

According to Figure 3.175 (a) and Figure 3.175 (b), the likelihoods for experiencing commuting accidents appear to be higher in sectors N (Administrative and support service activities) and Q (Human health and social work activities) compared to the reference sector A (Agriculture, forestry and fishing).

Figure 3.176 (a) and Figure 3.176 (b) show that, compared to the reference sector A (Agriculture, forestry and fishing), sector F (Construction) exhibits significantly higher odds ratios for accepted workplace accidents, while most other sectors display significantly lower odds ratios.

Figure 3.177 (a) and Figure 3.177 (b) present a model estimating the frequency degree of OA at the company level. These figures indicate that sectors C (Manufacturing) and F (Construction) have significantly higher frequency degree estimates compared to the reference sector A (Agriculture, forestry and fishing).

Figure 3.178 (a) and Figure 3.178 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that only sector Q (Human health and social work activities) shows significantly lower odds ratios for having an accepted workplace OA classified as serious compared to the reference sector A (Agriculture, forestry and fishing). Other sectors do not show significant differences.

Figure 3.179 (a) and Figure 3.179 (b) show that, compared to the reference sector A (Agriculture, forestry and fishing), sector M (Professional, scientific and technical activities) exhibits significantly lower numbers of (FEDRIS validated) absence days associated with an accepted workplace OA, while sector H (Transportation and storage) shows significantly higher numbers.

Figure 3.180 (a) and Figure 3.180 (b) largely confirm the results of the previous model with validated days.

Figure 3.181 (a) and Figure 3.181 (b) show that, compared to the reference sector A (Agriculture, forestry and fishing), sectors C (Manufacturing), F (Construction), H (Transportation and storage) and to a lesser extent N (Administrative and support service activities) exhibit significantly higher direct wage costs associated with an accepted workplace OA, while most other sectors display significantly lower direct wage costs

The next 4 figures (Figure 3.182 (a) and Figure 3.182 (b), Figure 3.183 (a) and Figure 3.183 (b)) are displaying the results of models assessing the severity degree of a company in relation to its NACE-BEL 2008 sector. According to Figure 3.182 (a) and Figure 3.182 (b), sector F (Construction) has significantly higher severity degree estimates compared to the reference sector A (Agriculture, forestry and fishing), while most other sectors exhibit significantly lower severity degree estimates

3.4.4.2 Joint labour committees

3.4.4.2.1 Descriptives

Both NSSO and FEDRIS provide statistics on the joint labour committee sector group (in Dutch Paritair Comité (joint collective agreement or sectoral committee) (PC)) of the employee (NSSO 19 groups of joint labour committees, FEDRIS 106 joint labour committees). From Liantis PS data it is also possible to derive the section of the employer through the combined committee and subcommittee codes(6 digits, two times three, first three digits indicate the comittee) of an employer. The notification however does not contain information on the joint labour committee. This information has to be joined from Liantis Payroll data. Since this information is not present in the notification information, the fit between the derived joint labour committee of the employer of an employee experiencing an OA from Liantis data cannot be compared with the corresponding joint labour committee known by NSSO and or FEDRIS and linked to the original notification of the accident.

all accepted occupational accidents

Table 3.116: Numbers of employees per parcom (nsso and liantis) and numbers of occupational accident notifications per parcom (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
parcom	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Agriculture, horticulture, forestry and sea fishing	1303435	1303435	271548	NA	954	954	184
Chemicals and petroleum	5267391	5267391	239413	NA	391	391	108
Clothing and textile industry	1523952	1523952	177402	NA	365	365	80
Construction	5713318	5713318	1100346	NA	4631	4631	1072
Distribution	10456440	10456440	1512671	NA	2027	2027	470
Financial sector	4228631	4228631	161373	NA	73	73	11
Food industry	3690570	3690570	589514	NA	2787	2787	1726
Gas and electricity companies	702678	702678	2776	NA	3	3	NA
Hospitality, sports and recreation	4782410	4782410	2766727	NA	2192	2192	312
Media, printing and publishing sector	573638	573638	37341	NA	38	38	10
Metal industry	11225704	11225704	1135693	NA	3768	3768	1134
Other sectors	19437471	19437471	1658501	NA	2446	2446	682
Paper and cardboard industry	474283	474283	21256	NA	52	52	NA
Services to businesses and individuals	15623907	15623907	1163038	NA	1315	1315	379
Social profit	22664565	22664565	3701139	NA	12370	12370	9594
Stone and glass industry	668386	668386	7027	NA	5	5	3
Transport and logistics	7259434	7259434	477448	NA	1506	1506	498
Wood industry	765746	765746	111925	NA	414	414	94
National Labour Council	NA	NA	NA	NA	4	4	2
NA	NA	NA	NA	NA	159878	25894	9671

Table 3.117: Percentages of employees per parcom (nsso and liantis) and percentages of occupational accident notifications per parcom (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
parcom	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Agriculture, horticulture, forestry and sea fishing	1.1	1.1	1.8	NA	2.7	2.7	1.1
Chemicals and petroleum	4.5	4.5	1.6	NA	1.1	1.1	0.7
Clothing and textile industry	1.3	1.3	1.2	NA	1.0	1.0	0.5
Construction	4.9	4.9	7.3	NA	13.1	13.1	6.6
Distribution	9.0	9.0	10.0	NA	5.7	5.7	2.9
Financial sector	3.6	3.6	1.1	NA	0.2	0.2	0.1
Food industry	3.2	3.2	3.9	NA	7.9	7.9	10.6
Gas and electricity companies	0.6	0.6	0.0	NA	0.0	0.0	NA
Hospitality, sports and recreation	4.1	4.1	18.3	NA	6.2	6.2	1.9
Media, printing and publishing sector	0.5	0.5	0.2	NA	0.1	0.1	0.1
Metal industry	9.6	9.6	7.5	NA	10.7	10.7	6.9
Other sectors	16.7	16.7	11.0	NA	6.9	6.9	4.2
Paper and cardboard industry	0.4	0.4	0.1	NA	0.1	0.1	NA
Services to businesses and individuals	13.4	13.4	7.7	NA	3.7	3.7	2.3
Social profit	19.5	19.5	24.5	NA	35.0	35.0	58.7
Stone and glass industry	0.6	0.6	0.0	NA	0.0	0.0	0.0
Transport and logistics	6.2	6.2	3.2	NA	4.3	4.3	3.0
Wood industry	0.7	0.7	0.7	NA	1.2	1.2	0.6
National Labour Council	NA	NA	NA	NA	NA	NA	NA
NA	NA	NA	NA	NA	NA	NA	NA

commuting accepted occupational accidents

Table 3.118: Numbers of employees per parcom (nsso and liantis) and numbers of occupational accident notifications per parcom (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
parcom	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Agriculture, horticulture, forestry and sea fishing	1303435	1303435	271548	NA	52	52	12
Chemicals and petroleum	5267391	5267391	239413	NA	51	51	18
Clothing and textile industry	1523952	1523952	177402	NA	47	47	10
Construction	5713318	5713318	1100346	NA	223	223	56
Distribution	10456440	10456440	1512671	NA	382	382	92
Financial sector	4228631	4228631	161373	NA	46	46	4
Food industry	3690570	3690570	589514	NA	259	259	148
Gas and electricity companies	702678	702678	2776	NA	1	1	NA
Hospitality, sports and recreation	4782410	4782410	2766727	NA	378	378	50
Media, printing and publishing sector	573638	573638	37341	NA	12	12	3
Metal industry	11225704	11225704	1135693	NA	323	323	131
Other sectors	19437471	19437471	1658501	NA	478	478	170
Paper and cardboard industry	474283	474283	21256	NA	4	4	NA
Services to businesses and individuals	15623907	15623907	1163038	NA	404	404	121
Social profit	22664565	22664565	3701139	NA	2348	2348	1710
Stone and glass industry	668386	668386	7027	NA	1	1	NA
Transport and logistics	7259434	7259434	477448	NA	68	68	28
Wood industry	765746	765746	111925	NA	40	40	7
National Labour Council	NA	NA	NA	NA	1	1	1
NA	NA	NA	NA	NA	23269	3481	1551

Table 3.119: Percentages of employees per parcom (nsso and liantis) and percentages of occupational accident notifications per parcom (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
parcom	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Agriculture, horticulture, forestry and sea fishing	1.1	1.1	1.8	NA	1.0	1.0	0.5
Chemicals and petroleum	4.5	4.5	1.6	NA	1.0	1.0	0.7
Clothing and textile industry	1.3	1.3	1.2	NA	0.9	0.9	0.4
Construction	4.9	4.9	7.3	NA	4.4	4.4	2.2
Distribution	9.0	9.0	10.0	NA	7.5	7.5	3.6
Financial sector	3.6	3.6	1.1	NA	0.9	0.9	0.2
Food industry	3.2	3.2	3.9	NA	5.1	5.1	5.8
Gas and electricity companies	0.6	0.6	0.0	NA	0.0	0.0	NA
Hospitality, sports and recreation	4.1	4.1	18.3	NA	7.4	7.4	2.0
Media, printing and publishing sector	0.5	0.5	0.2	NA	0.2	0.2	0.1
Metal industry	9.6	9.6	7.5	NA	6.3	6.3	5.1
Other sectors	16.7	16.7	11.0	NA	9.3	9.3	6.6
Paper and cardboard industry	0.4	0.4	0.1	NA	0.1	0.1	NA
Services to businesses and individuals	13.4	13.4	7.7	NA	7.9	7.9	4.7
Social profit	19.5	19.5	24.5	NA	45.9	45.9	66.8
Stone and glass industry	0.6	0.6	0.0	NA	0.0	0.0	NA
Transport and logistics	6.2	6.2	3.2	NA	1.3	1.3	1.1
Wood industry	0.7	0.7	0.7	NA	0.8	0.8	0.3
National Labour Council	NA	NA	NA	NA	NA	NA	NA
NA	NA	NA	NA	NA	NA	NA	NA

workplace accepted occupational accidents

Table 3.120: Numbers of employees per parcom (nsso and liantis) and numbers of occupational accident notifications per parcom (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
parcom	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
Agriculture, horticulture, forestry and sea fishing	1303435	1303435	271548	14241	902	902	172
Chemicals and petroleum	5267391	5267391	239413	29552	340	340	90
Clothing and textile industry	1523952	1523952	177402	16919	318	318	70
Construction	5713318	5713318	1100346	116241	4408	4408	1016
Distribution	10456440	10456440	1512671	81235	1645	1645	378
Financial sector	4228631	4228631	161373	4627	27	27	7
Food industry	3690570	3690570	589514	44893	2528	2528	1578
Gas and electricity companies	702678	702678	2776	2333	2	2	NA
Hospitality, sports and recreation	4782410	4782410	2766727	34583	1814	1814	262
Media, printing and publishing sector	573638	573638	37341	3732	26	26	7
Metal industry	11225704	11225704	1135693	142027	3445	3445	1003
Other sectors	19437471	19437471	1658501	48430	1968	1968	512
Paper and cardboard industry	474283	474283	21256	4163	48	48	NA
Services to businesses and individuals	15623907	15623907	1163038	160062	911	911	258
Social profit	22664565	22664565	3701139	215425	10022	10022	7884
Stone and glass industry	668386	668386	7027	12567	4	4	3
Transport and logistics	7259434	7259434	477448	76809	1438	1438	470
Wood industry	765746	765746	111925	12549	374	374	87
National Labour Council	NA	NA	NA	NA	3	3	1
NA	NA	NA	NA	NA	136609	22413	8120

Table 3.121: Percentages of employees per parcom (nsso and liantis) and percentages of occupational accident notifications per parcom (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
parcom	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
Agriculture, horticulture, forestry and sea fishing	1.1	1.1	1.8	1.4	3.0	3.0	1.2
Chemicals and petroleum	4.5	4.5	1.6	2.9	1.1	1.1	0.7
Clothing and textile industry	1.3	1.3	1.2	1.7	1.1	1.1	0.5
Construction	4.9	4.9	7.3	11.4	14.6	14.6	7.4
Distribution	9.0	9.0	10.0	8.0	5.4	5.4	2.7
Financial sector	3.6	3.6	1.1	0.5	0.1	0.1	0.1
Food industry	3.2	3.2	3.9	4.4	8.4	8.4	11.4
Gas and electricity companies	0.6	0.6	0.0	0.2	0.0	0.0	NA
Hospitality, sports and recreation	4.1	4.1	18.3	3.4	6.0	6.0	1.9
Media, printing and publishing sector	0.5	0.5	0.2	0.4	0.1	0.1	0.1
Metal industry	9.6	9.6	7.5	13.9	11.4	11.4	7.3
Other sectors	16.7	16.7	11.0	4.7	6.5	6.5	3.7
Paper and cardboard industry	0.4	0.4	0.1	0.4	0.2	0.2	NA
Services to businesses and individuals	13.4	13.4	7.7	15.7	3.0	3.0	1.9
Social profit	19.5	19.5	24.5	21.1	33.2	33.2	57.1
Stone and glass industry	0.6	0.6	0.0	1.2	0.0	0.0	0.0
Transport and logistics	6.2	6.2	3.2	7.5	4.8	4.8	3.4
Wood industry	0.7	0.7	0.7	1.2	1.2	1.2	0.6
National Labour Council	NA	NA	NA	NA	NA	NA	NA
NA	NA	NA	NA	NA	NA	NA	NA

3.4.4.2.2 Models

The following Table 3.122 table summarizes the results of the models assessing the relationship between the joint labour committees and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.122: Overview of the results of the different models examining the relation between the joint labour committees and Occupational Accidents

catparcom	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Agriculture…	REF	REF	REF	REF	REF	REF	REF	REF	REF
Chemicals…	<	ns	ns	<	<	ns	ns	ns	</<
Clothing…	<	ns	ns	<	<	ns	>/>	>	</<
Construction	>	ns	ns	>	>	ns	>/ns	>	ns
Distribution	<	ns	ns	<	<	ns	ns	<	</<
Financial…	<	ns	ns	<	<	ns	ns	<	</<
Food industry	ns	ns	>	<	ns	ns	ns	>	</<
Hospitality…	<	ns	ns	<	<	ns	</ns	<	</<
Metal industry	ns	ns	ns	ns	ns	ns	ns	>	</<
Other sectors	<	ns	ns	<	<	ns	ns	>	</<
Paper…	ns	ns	ns	ns	ns	ns	ns	ns	ns
Services…	<	ns	>	<	<	ns	ns	ns	</<
Social profit	<	ns	>	<	<	<	</ns	<	</<
Transport…	<	ns	ns	<	ns	ns	>/>	>	</<
Wood industry	ns	ns	>	ns	>	ns	ns	>	ns

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Joint labour committee and occupational accidents

Compared to reference joint labour committee Agriculture, horticulture, forestry and sea fishing
Likelihoods to notify seem to be higher Construction and equal or lower in almost all other committees
Likelihoods for refusal do not seem to differ significantly between committees
Likelihoods for experiencing commuting accidents appear to be higher in Food industry and Social profit
Likelihoods for experiencing workplace accidents appear to be higher in Construction and lower in almost all other sectors (lowest likelihood in the Financial sector)
Frequency degrees at company level appear to be higher in Construction and Wood industry
Likelihoods for serious accidents only seems to be lower in Social profit
Number of absence days associated with an accepted workplace accident seem to be higher in Transport and logistics and Clothing and textile industry and lower in Hospitality, sports and recreation
Costs appear to be higher in Construction, Transport and logistics and Clothing and textile industry and lower in Financial sector
Severity degrees at company level appear to be lower in many other committees except Construction
The influence of joint labour committee as determinant could be further assessed in the multivariable analysis (see Section 3.5) to clarify whether the observed differences persist when adjusting for other variables, but since the correlation between sector (NACE-BEL 2008) and joint labour committee is quite high, multicollinearity issues arise when both variables are included in the same model

« previous

From Figure 3.184 (a) and Figure 3.184 (b) we learn that, compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing, the committee Construction shows significantly higher odds ratios for notifying an OA, while most other committees exhibit significantly lower odds ratios.

According to Figure 3.185 (a) and Figure 3.185 (b), the joint labour committee of a company is not significantly associated with the odds of having a reported OA being refused by the insurer.

According to Figure 3.186 (a) and Figure 3.186 (b), the likelihoods for experiencing commuting accidents appear to be higher in Food industry, Social profit and Services to businesses and individuals compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing.

Figure 3.187 (a) and Figure 3.187 (b) show that, compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing, the committee Construction exhibits significantly higher odds ratios for accepted workplace accidents, while most other committees display significantly lower odds ratios.

Figure 3.188 (a) and Figure 3.188 (b) present a model estimating the frequency degree of OA at the company level. These figures indicate that the committee Construction and Wood industry have significantly higher frequency degree estimates compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing.

Figure 3.189 (a) and Figure 3.189 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that only committee Social profit shows significantly lower odds ratios for having an accepted workplace OA classified as serious compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing. Other committees do not show significant differences.

Figure 3.190 (a) and Figure 3.190 (b) show that, compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing, committee Transport and logistics and Clothing and textile industry exhibit significantly higher numbers of (FEDRIS validated) absence days associated with an accepted workplace OA, while committee Hospitality, sports and recreation and Social profit show significantly lower numbers.

Figure 3.191 (a) and Figure 3.191 (b) confirm the results of the previous model with validated days for Clothing and textile industry and Transport and logistics.

Figure 3.192 (a) and Figure 3.192 (b) show that, compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing, committees Construction, Transport and logistics and Clothing and textile industry exhibit significantly higher direct wage costs associated with an accepted workplace OA, while Financial sector and social profit display significantly lower direct wage costs.

The next 4 figures (Figure 3.193 (a) and Figure 3.193 (b), Figure 3.194 (a) and Figure 3.194 (b)) are displaying the results of models assessing the severity degree of a company in relation to its joint labour committee. According to Figure 3.193 (a) and Figure 3.193 (b), committee Construction has significantly higher severity degree estimates compared to the reference joint labour committee Agriculture, horticulture, forestry and sea fishing, while most other committees exhibit significantly lower severity degree estimates

The model below Figure 3.194 (a) and Figure 3.194 (b) largely confirms the results of the previous model with validated days.

3.4.5 Temporal patterns

3.4.5.1 Time trend (years)

3.4.5.1.1 Descriptives

Both NSSO and FEDRIS provide yearly statistics with counts of employees and employers. The same counts can be made for Liantis PS data. The notification contains the date (and thus year) of the OA.

all accepted occupational accidents

Table 3.123: Numbers of employees per year category (nsso and liantis) and numbers of occupational accident notifications per year category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
year	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
2014	13618344	10818747	1421509	141865	18309	5736	2287
2015	13726629	10926121	1516903	137219	18205	5729	2350
2016	13885634	11087143	1545854	142229	19289	6212	2551
2017	15495893	11256351	1644027	145538	19547	6475	2680
2018	15686057	11433166	1730173	147124	21901	6315	2608
2019	15916663	11619292	1828135	146507	22074	6423	2752
2020	15858180	11536574	1832210	114086	18120	5414	2346
2021	16115691	11741334	1948030	125946	19403	6110	2648
2022	16385592	11977770	2169919	127296	19170	6422	2848
2023	16479459	12038328	2419106	125567	19201	6399	2960

Table 3.124: Percentages of employees per year category (nsso and liantis) and percentages of occupational accident notifications per year category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
year	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
2014	8.89	9.45	7.87	10.48	9.38	9.37	8.79
2015	8.96	9.55	8.40	10.14	9.33	9.36	9.03
2016	9.07	9.69	8.56	10.51	9.88	10.14	9.80
2017	10.12	9.84	9.11	10.75	10.01	10.57	10.30
2018	10.24	9.99	9.58	10.87	11.22	10.31	10.02
2019	10.39	10.15	10.12	10.83	11.31	10.49	10.57
2020	10.35	10.08	10.15	8.43	9.28	8.84	9.01
2021	10.52	10.26	10.79	9.31	9.94	9.98	10.17
2022	10.70	10.47	12.02	9.41	9.82	10.49	10.94
2023	10.76	10.52	13.40	9.28	9.84	10.45	11.37

commuting accepted occupational accidents

Table 3.125: Numbers of employees per year category (nsso and liantis) and numbers of occupational accident notifications per year category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
year	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
2014	13618344	10818747	1421509	20670	2228	668	298
2015	13726629	10926121	1516903	20772	2424	761	333
2016	13885634	11087143	1545854	22347	2562	759	347
2017	15495893	11256351	1644027	24627	2945	960	431
2018	15686057	11433166	1730173	24389	3102	797	356
2019	15916663	11619292	1828135	26429	3408	935	426
2020	15858180	11536574	1832210	17920	2504	708	351
2021	16115691	11741334	1948030	20660	2879	863	451
2022	16385592	11977770	2169919	23726	3111	1047	516
2023	16479459	12038328	2419106	24770	3224	1101	603

Table 3.126: Percentages of employees per year category (nsso and liantis) and percentages of occupational accident notifications per year category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
year	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
2014	8.89	9.45	7.87	9.13	7.85	7.77	7.25
2015	8.96	9.55	8.40	9.18	8.54	8.85	8.10
2016	9.07	9.69	8.56	9.87	9.03	8.83	8.44
2017	10.12	9.84	9.11	10.88	10.37	11.16	10.48
2018	10.24	9.99	9.58	10.78	10.93	9.27	8.66
2019	10.39	10.15	10.12	11.68	12.01	10.87	10.36
2020	10.35	10.08	10.15	7.92	8.82	8.23	8.54
2021	10.52	10.26	10.79	9.13	10.14	10.04	10.97
2022	10.70	10.47	12.02	10.48	10.96	12.18	12.55
2023	10.76	10.52	13.40	10.95	11.36	12.80	14.66

workplace accepted occupational accidents

Table 3.127: Numbers of employees per year category (nsso and liantis) and numbers of occupational accident notifications per year category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
year	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
2014	13618344	10818747	1421509	121195	16081	5068	1989
2015	13726629	10926121	1516903	116447	15781	4968	2017
2016	13885634	11087143	1545854	119882	16727	5453	2204
2017	15495893	11256351	1644027	120911	16602	5515	2249
2018	15686057	11433166	1730173	122735	18799	5518	2252
2019	15916663	11619292	1828135	120078	18666	5488	2326
2020	15858180	11536574	1832210	96166	15616	4706	1995
2021	16115691	11741334	1948030	105286	16524	5247	2197
2022	16385592	11977770	2169919	103570	16059	5375	2332
2023	16479459	12038328	2419106	100797	15977	5298	2357

Table 3.128: Percentages of employees per year category (nsso and liantis) and percentages of occupational accident notifications per year category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
year	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
2014	8.89	9.45	7.87	10.75	9.64	9.63	9.07
2015	8.96	9.55	8.40	10.33	9.46	9.44	9.20
2016	9.07	9.69	8.56	10.64	10.03	10.36	10.06
2017	10.12	9.84	9.11	10.73	9.95	10.48	10.26
2018	10.24	9.99	9.58	10.89	11.27	10.48	10.27
2019	10.39	10.15	10.12	10.65	11.19	10.43	10.61
2020	10.35	10.08	10.15	8.53	9.36	8.94	9.10
2021	10.52	10.26	10.79	9.34	9.90	9.97	10.02
2022	10.70	10.47	12.02	9.19	9.63	10.21	10.64
2023	10.76	10.52	13.40	8.94	9.58	10.07	10.75

3.4.5.1.2 Models

The following Table 3.129 table summarizes the results of the models assessing the relationship between the different years of the study and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.129: Overview of the results of the different models examining the relation between the different years of the study and Occupational Accidents

catyear	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
2014	REF	REF	REF	REF	REF	REF	REF	REF	REF
2015	<	ns	ns	<	<	ns	ns/ns	>	ns
2016	ns	ns	ns	ns	<	ns	>/ns	>	ns
2017	ns	ns	>	ns	<	ns	ns/ns	>	ns
2018	<	ns	ns	<	<	ns	ns/ns	>	ns
2019	<	ns	ns	<	<	ns	>/>	>	ns
2020	<	ns	<	<	<	>	>/ns	>	ns/<
2021	<	ns	ns	<	<	ns	>/>	>	ns/<
2022	<	>	ns	<	<	ns	>/>	>	ns/<
2023	<	>	ns	<	<	>	ns/ns	>	</<

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Time trend (years) and occupational accidents

Compared to reference year 2014, the likelihoods to notify were significantly lower in almost all years except 2016 and 2017
The likelihoods for refusal were significantly higher in 2022 and 2023
The likelihoods for experiencing commuting occupational accidents were significantly lower in 2020, but significantly higher in 2017
The likelihoods for experiencing workplace occupational accidents were significantly lower in almost all years with exception of 2016; a general decreasing trend in odds ratios is observed over the years
The same decreasing trend is observed for the frequency degree estimates at company level
The likelihoods for experiencing serious accidents were significantly higher in 2020 and 2023
The numbers of absence days associated with an accepted workplace OA were significantly higher in 2021 and 2022
Estimates of cost related to occupational accidents were significantly higher for all years and, with exception of 2018, seemed to be gradually increasing up tot 2021
The severity degree estimates at company level were significantly lower in 2023 (validated) and in 2020 to 2023 (calculated), a decreasing trend -however less clear than for the frequency degrees- can be noticed

« previous

From Figure 3.195 (a) and Figure 3.195 (b) we learn that, compared to the reference year 2014, the likelihoods to notify were significantly lower in almost all years except 2016 and 2017.

According to Figure 3.196 (a) and Figure 3.196 (b), the likelihoods for refusal were significantly higher in 2022 and 2023 compared to the reference year 2014.

According to Figure 3.197 (a) and Figure 3.197 (b), the likelihoods for experiencing commuting occupational accidents were significantly lower in 2020, but significantly higher in 2017 compared to the reference year 2014. The steep drop in 2020 can likely be attributed to the COVID-19 pandemic and associated lockdowns, which reduced commuting activities (see Figure 2.61).

Figure 3.198 (a) and Figure 3.198 (b) show that, compared to the reference year 2014, the likelihoods for experiencing workplace occupational accidents were significantly lower in almost all years with exception of 2016; a general decreasing trend in odds ratios is observed over the years.

Figure 3.199 (a) and Figure 3.199 (b) present a model estimating the frequency degree of OA at the company level. These figures indicate a general and significant decreasing trend in frequency degree estimates over the years compared to the reference year 2014.

Figure 3.200 (a) and Figure 3.200 (b) present a model estimating the chance for an accepted accident being serious (according to the definition of the Belgian law). These figures indicate that the likelihoods for experiencing serious accidents were significantly higher in 2020 and 2023 compared to the reference year 2014.

Figure 3.201 (a) and Figure 3.201 (b) show that, compared to the reference year 2014, the numbers of absence days associated with an accepted workplace OA were significantly higher in 2019, 2021 and 2022.

Figure 3.202 (a) and Figure 3.202 (b) confirm the results of the previous model with validated days, showing that the numbers of absence days associated with an accepted workplace OA were significantly higher in 2019, 2021 and 2022.

As shown in Figure 3.203 (a) and Figure 3.203 (b), the estimated costs associated with OA were significantly higher in all years compared to the reference year 2014. With the exception of 2018, these costs exhibited a generally increasing trend up to 2021. Although a slight decline is observed in 2022 and 2023, the estimates remain substantially above the 2014 baseline.

The next 4 figures (Figure 3.204 (a) and Figure 3.204 (b), Figure 3.205 (a) and Figure 3.205 (b)) are displaying the results of models assessing the severity degree of a company in relation to the different years of the study. According to Figure 3.204 (a) and Figure 3.204 (b), only 2023 shows a significantly lower severity degree estimate (based on the validated number of absence days) compared to the reference year 2014.

Figure 3.205 (a) and Figure 3.205 (b) (using Liantis PS calculated absence days) partially confirm the results of the previous model with validated days. Here, years 2020 to 2023 exhibit significantly lower severity degree estimates compared to the reference year 2014.

3.4.5.2 Time of the year (months)

3.4.5.2.1 Descriptives

Fedris provides statistics on month of accident. From Liantis PS data it is also possible to derive the month of the wage calculations. The notification contains the date of the OA from which month can be extracted.

all accepted occupational accidents

Table 3.130: Numbers of employees per month category (nsso and liantis) and numbers of occupational accident notifications per month category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
monthOAf	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
January	NA	NA	1417084	124572	17642	5600	2476
February	NA	NA	1431180	114448	16359	5129	2152
March	NA	NA	1453499	119297	17104	5360	2297
April	NA	NA	1489048	100418	14670	4560	1912
May	NA	NA	1493678	106381	15315	4734	2006
June	NA	NA	1523104	122525	17756	5548	2311
July	NA	NA	1562970	97404	13948	4311	1909
August	NA	NA	1565079	107211	15530	4938	2065
September	NA	NA	1547325	122914	18259	5729	2320
October	NA	NA	1536877	124753	18391	5788	2406
November	NA	NA	1519519	112742	15996	4948	2158
December	NA	NA	1516503	100712	14249	4590	2018

Table 3.131: Percentages of employees per month category (nsso and liantis) and percentages of occupational accident notifications per month category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
monthOAf	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
January	NA	NA	7.85	9.20	9.04	9.15	9.51
February	NA	NA	7.93	8.46	8.38	8.38	8.27
March	NA	NA	8.05	8.81	8.76	8.75	8.82
April	NA	NA	8.25	7.42	7.51	7.45	7.35
May	NA	NA	8.27	7.86	7.85	7.73	7.71
June	NA	NA	8.44	9.05	9.10	9.06	8.88
July	NA	NA	8.66	7.20	7.14	7.04	7.33
August	NA	NA	8.67	7.92	7.96	8.06	7.93
September	NA	NA	8.57	9.08	9.35	9.36	8.91
October	NA	NA	8.51	9.22	9.42	9.45	9.24
November	NA	NA	8.42	8.33	8.19	8.08	8.29
December	NA	NA	8.40	7.44	7.30	7.50	7.75

commuting accepted occupational accidents

Table 3.132: Numbers of employees per month category (nsso and liantis) and numbers of occupational accident notifications per month category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
monthOAf	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
January	NA	NA	1417084	26825	3460	1022	525
February	NA	NA	1431180	19726	2511	751	366
March	NA	NA	1453499	17935	2203	679	314
April	NA	NA	1489048	14433	1789	531	230
May	NA	NA	1493678	16193	2022	611	303
June	NA	NA	1523104	18887	2300	686	317
July	NA	NA	1562970	14265	1615	478	228
August	NA	NA	1565079	15637	2020	610	291
September	NA	NA	1547325	19919	2453	754	368
October	NA	NA	1536877	21508	2754	835	379
November	NA	NA	1519519	20864	2561	778	379
December	NA	NA	1516503	20118	2699	864	412

Table 3.133: Percentages of employees per month category (nsso and liantis) and percentages of occupational accident notifications per month category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
monthOAf	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
January	NA	NA	7.85	11.85	12.19	11.89	12.77
February	NA	NA	7.93	8.72	8.85	8.73	8.90
March	NA	NA	8.05	7.92	7.76	7.90	7.64
April	NA	NA	8.25	6.38	6.30	6.18	5.59
May	NA	NA	8.27	7.16	7.12	7.11	7.37
June	NA	NA	8.44	8.35	8.10	7.98	7.71
July	NA	NA	8.66	6.30	5.69	5.56	5.54
August	NA	NA	8.67	6.91	7.12	7.09	7.08
September	NA	NA	8.57	8.80	8.64	8.77	8.95
October	NA	NA	8.51	9.50	9.70	9.71	9.22
November	NA	NA	8.42	9.22	9.02	9.05	9.22
December	NA	NA	8.40	8.89	9.51	10.05	10.02

workplace accepted occupational accidents

Table 3.134: Numbers of employees per month category (nsso and liantis) and numbers of occupational accident notifications per month category (FEDRIS and Liantis)

Category	Employees (n)			Occupational Accidents Notifications (n)
monthOAf	rszall	rszpriv	liaall	fedrisall	totalnot	totalnotmut	totalnotrepmutflan
January	NA	NA	1417084	97747	14182	4578	1951
February	NA	NA	1431180	94722	13848	4378	1786
March	NA	NA	1453499	101362	14901	4681	1983
April	NA	NA	1489048	85985	12881	4029	1682
May	NA	NA	1493678	90188	13293	4123	1703
June	NA	NA	1523104	103638	15456	4862	1994
July	NA	NA	1562970	83139	12333	3833	1681
August	NA	NA	1565079	91574	13510	4328	1774
September	NA	NA	1547325	102995	15806	4975	1952
October	NA	NA	1536877	103245	15637	4953	2027
November	NA	NA	1519519	91878	13435	4170	1779
December	NA	NA	1516503	80594	11550	3726	1606

Table 3.135: Percentages of employees per month category (nsso and liantis) and percentages of occupational accident notifications per month category (FEDRIS and Liantis)

Category	Employees (%)			Occupational Accidents Notifications (%)
monthOAf	percrszall	percrszpriv	percliaall	percfedrisall	perctotalnot	perctotalnotmut	perctotalnotrepmutflan
January	NA	NA	7.85	8.67	8.50	8.70	8.90
February	NA	NA	7.93	8.40	8.30	8.32	8.15
March	NA	NA	8.05	8.99	8.93	8.89	9.05
April	NA	NA	8.25	7.63	7.72	7.65	7.67
May	NA	NA	8.27	8.00	7.97	7.83	7.77
June	NA	NA	8.44	9.20	9.26	9.24	9.10
July	NA	NA	8.66	7.38	7.39	7.28	7.67
August	NA	NA	8.67	8.12	8.10	8.22	8.09
September	NA	NA	8.57	9.14	9.47	9.45	8.91
October	NA	NA	8.51	9.16	9.37	9.41	9.25
November	NA	NA	8.42	8.15	8.05	7.92	8.12
December	NA	NA	8.40	7.15	6.92	7.08	7.33

3.4.5.2.2 Models

The following Table 3.136 table summarizes the results of the models assessing the relationship between the different years of the study and the several outcomes related to OAs. Below the table, details are provided about the data used for the analyses. The results of the nine models are then presented both graphically and in a table.

Table 3.136: Overview of the results of the different models examining the relation between the different months within the years of the study and Occupational Accidents

catmonth	chance to notify	chance on refusal	chance commuting	chance workplace	freq degree	chance serious	nDays TAO V/C	cost	sev degree V/C
Jan	REF	REF	REF	REF	ns	REF	REF	REF	REF
Feb	ns	ns	ns	ns	>	ns	ns	ns	ns
Mar	<	ns	<	ns	>	<	ns	<	>
Apr	<	ns	<	<	>	ns	ns/<	ns	>
May	<	ns	<	<	>	ns	</ns	<	ns
Jun	ns	ns	<	ns	>	ns	ns/<	ns	ns
Jul	<	ns	<	<	>	ns	</<	<	ns
Aug	<	ns	<	<	>	ns	</<	<	ns
Sep	ns	ns	<	ns	>	ns	ns	ns	ns
Oct	ns	ns	<	>	>	ns	ns	<	>
Nov	ns	ns	<	ns	>	ns	ns	ns	ns
Dec	<	ns	<	<	rank def	ns	ns	<	rank def

	occur	accept	occur	occur	occur	severe	severe	severe	severe
	empl	empl	empl	empl	comp	empl	empl	empl	comp
	c/w all	c/w ref	c acc	w acc	w acc	w acc	w acc	w acc	w acc

occur: data include all workers, chance for occurrence is calculated for the outcome
accept: data include only workers with a notification of OA
severe: data include only workers with an accepted OA
c/w all: commuting and workplace accidents, including refused and accidents without decisions
c/w ref: accepted or refused commuting and workplace accidents
c acc: accepted commuting accidents
w acc: accepted workplace accidents
empl: the outcome is situated at the individual level
comp: the outcome is situated at the company level
nDaysTAO: number of days absenteeism related to the OA
V/C: Validated number of days (available from FEDRIS) / Calculated number of days (based on the Liantis PS data)

Time of the year (months) and occupational accidents

Compared to reference month January, the likelihoods to notify were significantly lower in March, April, May, July, August and December
No significant differences were observed for the likelihoods for refusal across months
The likelihoods for experiencing commuting occupational accidents were significantly lower in all months but February
For workplace occupational accidents, the likelihoods were significantly lower in April, May, July, August and December; highest estimates were observed September, February, June and October, but only for October the increase was significant in comparison with January
The likelihoods for experiencing serious accidents were significantly lower in March compared to January
Absence days (validated and calculated) were lower in July and August
Costs related to occupational accidents were significantly lower in March, May, July, August, October and December; these are also the months with 31 days and smaller confidence intervals

« previous

From Figure 3.206 (a) and Figure 3.206 (b) we learn that, compared to the reference month January, the likelihoods to notify were significantly lower in March, April, May, July, August and December.

According to Figure 3.207 (a) and Figure 3.207 (b), no significant differences were observed for the likelihoods for refusal across months compared to the reference month January.

According to Figure 3.208 (a) and Figure 3.208 (b), the likelihoods for experiencing commuting occupational accidents were significantly lower in all months but February compared to the reference month January.

Figure 3.209 (a) and Figure 3.209 (b) show that, compared to the reference month January, the likelihoods for experiencing workplace occupational accidents were significantly lower in April, May, July, August and December; highest estimates were observed September, February, June and October, but only for October the increase was significant in comparison with January.

Figure 3.210 (a) and Figure 3.210 (b) present a model estimating the frequency degree of OA at the company level. These figures indicate that the frequency degree estimates are significantly influenced by fractions of employees at work in all months but January and December.

Figure 3.211 (a) and Figure 3.211 (b) present a model estimating the chance for an accepted workplace OA being serious (according to the definition of the Belgian law). These figures indicate that the likelihoods for experiencing serious accidents were significantly lower in March compared to January.

Figure 3.212 (a) and Figure 3.212 (b) show that, compared to the reference month January, absence days (validated and calculated) were lower in July and August.

Figure 3.213 (a) and Figure 3.213 (b) largely confirm the results of the previous model with validated days.

Figure 3.214 (a) and Figure 3.214 (b) show that, compared to the reference month January, costs related to occupational accidents were significantly lower in March, May, July, August, October and December; these are also the months with 31 days and smaller confidence intervals on the estimates.

The next 4 figures (Figure 3.215 (a) and Figure 3.215 (b), Figure 3.216 (a) and Figure 3.216 (b)) are displaying the results of models assessing the severity degree of a company using employment fractions accross the different months of the year. According to Figure 3.215 (a) and Figure 3.215 (b), employment fractions from March, April and October seem to have a larger impact.

Figure 3.216 (a) and Figure 3.216 (b) (using Liantis PS calculated absence days) show an image analogous to the previous model with validated days, although no significant results are found here.

3.4.5.3 Time of the week (days)

Data Quality Alert: time of the week (days)

Although Liantis holds detailed data on the number of people actually at work on each individual day across the ten-year study period, this level of granularity was not accessible within the scope of the present project.
For previous models, monthly-level data on wages and worked hours provided sufficient detail. However, to construct a model assessing the likelihood of experiencing an occupational accident on a specific weekday, this granularity is insufficient.
As a result, it was not possible to build weekday-level models comparable to earlier analyses. Specifically, no models can be constructed to investigate the impact of the day of the week on the likelihood to notify (or experience) an occupational (commuting or workplace) accident.
Instead, for the current and subsequent models exploring temporal and calendar effects, we opted to use all available Liantis ESPP OA notification data or for commuting or for workplace accidents.
This represents a fundamentally different approach, and it is important to remain aware of this methodological shift in this part of the report.

« previous

The number of OA notifications by weekday for commuting and workplace accidents is summarized in Table 3.137.

Table 3.137: Numbers and percentages of OA notifications by weekday (commuting and workplace)

wdayOA	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
Monday	6574	37402	20.2	20.3	14.9	85.1
Tuesday	6685	36713	20.6	19.9	15.4	84.6
Wednesday	5936	34610	18.3	18.8	14.6	85.4
Thursday	6174	33566	19.0	18.2	15.5	84.5
Friday	5203	28141	16.0	15.3	15.6	84.4
Saturday	1151	7974	3.5	4.3	12.6	87.4
Sunday	801	6016	2.5	3.3	11.8	88.2

In absolute terms, both commuting and workplace OA notifications peak on Mondays and are least frequent on Saturdays and Sundays. The weekday distribution of OA notifications differs significantly between commuting and workplace accidents (\(\chi^2 = 125.53\), \(df = 6\), \(p < 0.001\)). This pattern aligns with previous findings from Medex on OA in the public sector (Federal Public Service Health, Food Chain Safety and Environment, 2022). Notably, the proportion of workplace OA increases during weekends compared to weekdays, a trend also observed in the Medex study. However, the relative shares differ: approximately 15% commuting and 85% workplace in the present study (2014–2023) versus 25% and 75% in Medex (2016–2020). The lower number of OA on Wednesdays reported by Medex -attributed to higher telework and/or part-time work- was not confirmed in our study. Overall, our findings align with the broader literature discussed in Section 1.3.6.3.

Time of the week (days) and occupational accidents

Most occupational accidents -both commuting and workplace- occur on Mondays and Tuesdays.
The fewest occupational accidents (commuting and workplace) occur on Saturdays and Sundays.
However, the proportion of workplace accidents increases during weekends compared to weekdays.

« previous

3.4.5.4 Time of the day (hours)

Data Quality Alert: time of the day (hours)

Although Liantis holds detailed data on the number of people actually at work on each individual day across the ten-year study period, this level of granularity was not accessible within the scope of the present project.
As a result, it was not possible to build within-day-level models. Specifically, no models can be constructed to investigate the impact of the hour of the day on the likelihood to notify (or experience) an occupational (commuting or workplace) accident.
Instead, for the current and subsequent models exploring temporal and calendar effects, we opted to use all available Liantis ESPP OA notification data or for commuting or for workplace accidents.
This represents a fundamentally different approach, and it is important to remain aware of this methodological shift in this part of the report.

« previous

The number of OA notifications by hour of the day for commuting and workplace accidents is summarized in Table 3.138.

Table 3.138: Numbers and percentages of OA notifications by hour of the day (commuting and workplace)

hourOA	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
00:00:00	176	1872	0.6	1.0	8.6	91.4
01:00:00	53	1093	0.2	0.6	4.6	95.4
02:00:00	43	929	0.1	0.5	4.4	95.6
03:00:00	68	988	0.2	0.6	6.4	93.6
04:00:00	413	1155	1.3	0.6	26.3	73.7
05:00:00	1143	1646	3.6	0.9	41.0	59.0
06:00:00	2861	3422	9.0	1.9	45.5	54.5
07:00:00	6346	7529	20.0	4.2	45.7	54.3
08:00:00	4371	14167	13.8	7.9	23.6	76.4
09:00:00	970	16725	3.1	9.4	5.5	94.5
10:00:00	508	22054	1.6	12.4	2.3	97.7
11:00:00	617	20515	1.9	11.5	2.9	97.1
12:00:00	1796	9765	5.7	5.5	15.5	84.5
13:00:00	1422	12897	4.5	7.2	9.9	90.1
14:00:00	905	16406	2.8	9.2	5.2	94.8
15:00:00	1205	15729	3.8	8.8	7.1	92.9
16:00:00	2677	10263	8.4	5.8	20.7	79.3
17:00:00	2918	5359	9.2	3.0	35.3	64.7
18:00:00	1332	4172	4.2	2.3	24.2	75.8
19:00:00	443	3297	1.4	1.8	11.8	88.2
20:00:00	428	3036	1.3	1.7	12.4	87.6
21:00:00	515	2347	1.6	1.3	18.0	82.0
22:00:00	428	1636	1.3	0.9	20.7	79.3
23:00:00	122	1467	0.4	0.8	7.7	92.3
NA	764	5953	NA	NA	11.4	88.6

The result is also graphically presented in Figure 3.217.

Figure 3.217: Numbers of notifications by hour of the occupational accident (commuting in blue and workplace in orange)

Commuting accidents peak in the early morning (before 08:00) and evening (after 17:00), with a smaller peak around noon. Workplace accidents increase sharply from 06:00, reaching peaks at 10:00–11:00 and 14:00–15:00, and decline around noon and after 16:00. Between 05:00–08:00, at 12:00, and after 16:00, the proportion of commuting OA rises relative to workplace OA (see Table 3.138), reflecting typical home-to-work and return travel times. This pattern is consistent with the literature overview in Section 1.3.6.4. The difference in hourly distribution between commuting and workplace accidents is statistically significant (\(\chi^2 = 30483\), \(df = 24\), \(p < 0.001\)).

Time of the day (hours) and occupational accidents

Commuting accidents peak in the early morning (before 08:00) and evening (after 17:00), with a smaller peak around noon.
Workplace accidents increase sharply from 06:00, reaching peaks at 10:00–11:00 and 14:00–15:00, and decline around noon and after 16:00.
The proportion of commuting accidents rises relative to workplace accidents during typical home-to-work and return travel times (05:00–08:00, 12:00, and after 16:00).

« previous

3.4.5.5 Weather

In Section 2.18.3 is described how weather data were collected. Table 2.106 showed that more extreme weather conditions such as freezing temperatures and heatwaves are relatively uncommon (1.94% of days and 2.67% of days respectively in the source dataset).

We take those two conditions as an example to illustrate whether, when these conditions do occur, they influence the distribution of commuting versus workplace accidents.

The number of OA notifications by freezing conditions (maximum temperatures below zero) for commuting and workplace accidents is summarized in Table 3.139.

Table 3.139: Numbers and percentages of OA notifications by freezing conditions (commuting and workplace)

temp_freezing	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
0	30923	164196	97.2	98.2	15.8	84.2
1	899	3022	2.8	1.8	22.9	77.1
NA	421	14948	NA	NA	2.7	97.3

The number of OA notifications by heatwave conditions (maximum temperatures in Uccle \(\geq\) 25°C for at least five consecutive days, with at least three \(\geq\) 30°C) for commuting and workplace OA is summarized in Table 3.140.

Table 3.140: Numbers and percentages of OA notifications by heat wave conditions (commuting and workplace)

heatwave	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
0	31007	162595	97.4	97.2	16.0	84.0
1	815	4623	2.6	2.8	15.0	85.0
NA	421	14948	NA	NA	2.7	97.3

Freezing conditions are associated with a higher proportion of commuting OA (22.9%) compared to non-freezing conditions (15.8%) (see Table 3.139), a difference that is statistically significant (\(\chi^2 = 142.9\), \(df = 1\), \(p < 0.001\)). Heatwave conditions also show a significant difference in the proportion of commuting OA (15.0%) compared to non-heatwave conditions (16.0%) (see Table 3.140) (\(\chi^2 = 4.1\), \(df = 1\), \(p = 0.043\)). These findings suggest that freezing weather may increase the risk of commuting OA, whereas heatwaves may shift risk toward more workplace OA. Both observations are consistent with previous research discussed in Section 1.3.6.5. Further research should examine the mechanisms underlying these patterns and assess whether they persist after adjusting for other variables. Such a model is presented in Section 3.4.5.10.

Weather and occupational accidents

Extreme weather conditions appear to influence the distribution of commuting and workplace accidents.
Heatwaves are associated with a relative increase in workplace accidents, whereas freezing conditions are linked to a relative increase in commuting accidents.

« previous

3.4.5.6 Holidays & vacations

In Section 2.18.1 is described how the calendar with Belgian holidays was constructed (see Table 2.100, Table 2.101 and Table 2.102).

We take the Belgian school holidays (Herfstvakantie, Kerstvakantie, Krokusvakantie, Paasvakantie, Zomervakantie) and legal holidays (Allerheiligen, Dag van de arbeid, Kerstmis, Nationale feestdag van België, Nieuwjaar, Onze-Lieve-Heer-Hemelvaart, Onze-Lieve-Vrouw-Hemelvaart/Moederdag Antwerpen, Paasmaandag, Pasen, Pinksteren, Pinkstermaandag, Wapenstilstand/Sint-Maarten) as examples to illustrate whether, when these conditions do occur, they influence the distribution of commuting versus workplace accidents.

The distribution of commuting and workplace OA across school holiday periods is summarized in Table 3.141.

Table 3.141: Numbers and percentages of OA notifications by days falling in schoolholidays (commuting and workplace)

schoolhol	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
0	25596	139865	78.7	75.8	15.5	84.5
1	6928	44557	21.3	24.2	13.5	86.5

The distribution of commuting and workplace OA across legal holidays and non-holidays is summarized in Table 3.142.

Table 3.142: Numbers and percentages of OA notifications by heat wave conditions (commuting and workplace)

legalhol	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
0	32354	183083	99.5	99.3	15.0	85.0
1	170	1339	0.5	0.7	11.3	88.7

Both school holidays and legal holidays are associated with a lower proportion of commuting OA compared to non-holiday periods (see Table 3.141 and Table 3.142), with differences that are statistically significant (school holidays: \(\chi^2 = 124.7\), \(df = 1\), \(p < 0.001\); legal holidays: \(\chi^2 = 16.3\), \(df = 1\), \(p < 0.001\)). The absolute number of accidents is much lower on legal holidays, as most of the Belgian workforce does not work on these days, whereas during school holidays, variation occurs depending on which workers take leave. These findings suggest that both school and legal holidays reduce the risk of commuting OA, likely due to decreased travel during these periods. This is consistent with previous research discussed in Section 1.3.6.6. Further research should examine the mechanisms underlying these patterns and assess whether they persist after adjusting for other variables.

Holidays & vacations and occupational accidents

Holidays, including school and legal holidays, influence the distribution of commuting and workplace accidents.
School holidays occur more frequently and are associated with a relative decrease in commuting accidents from 15.5% to 13.5%.
Legal holidays occur less frequently and are associated with a relative decrease in commuting accidents from 15.0% to 11.3%.

« previous

3.4.5.7 Summer- and wintertime changes

In Section 2.18.1 is described how summer- and wintertime changes were included in the calendar (see Table 2.103).

In the present paragraph we analyse whether, when these conditions do occur, they influence the distribution of commuting versus workplace accidents.

The distribution of commuting and workplace OA across timeregimes is summarized in Table 3.143. Chi-square tests were conducted for the second and third rows and the fifth and sixth rows to assess differences in commuting versus workplace accidents in the week before before and after time changes. The relative increase in commuting accidents from 12.3% to 12.5% (week before and after the switch to summertime) and the decrease from 15.0% to 13.5% (week before and after the switch to wintertime) were not statistically significant (\(\chi^2 = 0.05\), \(df = 1\), \(p = 0.819\); \(\chi^2 = 3.27\), \(df = 1\), \(p = 0.071\), respectively). It should be noted however that time changes affect only one hour on two weekend days per year and often coincide with school holidays, limiting statistical power and introducing potential confounding.

Table 3.143: Numbers and percentages of OA notifications by timeregimes (commuting and workplace)

timechangeregime	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
winter (normal)	14304	68213	44.0	37.0	17.3	82.7
end winter (normal)	516	3667	1.6	2.0	12.3	87.7
begin summer (less sleep)	477	3330	1.5	1.8	12.5	87.5
summer (normal)	16061	102302	49.4	55.5	13.6	86.4
end summer (normal)	740	4189	2.3	2.3	15.0	85.0
begin winter (more sleep)	426	2721	1.3	1.5	13.5	86.5

Summer- and wintertime changes and occupational accidents

Based on the data collected in this study, no significant effect of summer- or wintertime changes on the occurrence of commuting or workplace accidents was detected.
This does not imply the absence of an effect; rather, any effect is likely small and difficult to detect given the limited observation window (one week before and after two specific dates per year) and the potential confounding influence of school holidays.

« previous

3.4.5.8 COVID-19 pandemic

In Section 2.18.2 is described how the COVID-19 measures taken in Belgium were included in the calendar using the COVID-19 stringency index (Hale et al., 2021). See Figure 2.61 for an illustration of the effect.

We arbitrarily categorised the stringency index between 2020-01-01 and 2022-12-31 (3 years) into values above and below 50 to assess its effect on the occurrence of commuting versus workplace accidents and called the period before 2020 (6 years) and after 2023 (1 year) pre and post COVID-19 respectively.

The distribution of commuting and workplace OA across COVID-19 measures, categorized by stringency index, is summarized in Table 3.144. A chi-square test confirmed that the observed differences in commuting versus workplace accidents across categories are statistically significant (\(\chi^2 = 234.01\), \(df = 3\), \(p < 0.001\)). Although not shown in the table, cases before and after COVID-19 account for roughly 60% and 10% of the total (over seven years) for both commuting and workplace OA. Within the remaining 30% of cases (three years), the effect of high versus low stringency index is clear and aligns with expectations.

Table 3.144: Numbers and percentages of OA notifications by COVID-19 stringency index category (commuting and workplace)

covid_stringency	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
pre COVID-19	18844	112583	57.9	61.0	14.3	85.7
high stringency	3716	22795	11.4	12.4	14.0	86.0
low stringency	6182	30928	19.0	16.8	16.7	83.3
post COVID-19	3782	18116	11.6	9.8	17.3	82.7

COVID-19 pandemic and occupational accidents

The COVID-19 pandemic and related public health measures significantly affected the distribution of commuting and workplace accidents.
Periods of high stringency were associated with a marked decrease in commuting accidents (from 16.7% to 14.0%) and a corresponding increase in the proportion of workplace accidents (from 83.3% to 86.0%).
In absolute terms, higher stringency index values (\(\geq\) 50 was chosen here) coincided with lower overall numbers of OA, for both commuting and workplace accidents.

« previous

3.4.5.9 Terrorist attacks

On 22 March 2016, terrorist attacks occurred in Belgium at Brussels Airport and Maelbeek metro station, followed by another attack in Liège on 29 May 2018. A previous report by Medex noted potential effects of these events on OA in the public sector on those days (Federal Public Service Health, Food Chain Safety and Environment, 2022).

Although the list of terrorist attacks in Belgium during the study period is longer, those to attacks to examine whether such events influence the distribution of commuting versus workplace accidents.

The distribution of commuting and workplace OA incidents on days when terrorist attacks occurred is summarized in Table 3.145. Chi-square tests were not conducted due to the exceptional nature of these events and the already established influence of factors such as weekday (see Section 3.4.5.3). For instance, attacks occurring during weekends are expected to coincide with a lower proportion of commuting accidents.

Notably, 22 March 2016 was a Tuesday, a day on which approximately 15.4% of commuting accidents would typically be expected (see Table 3.137). However, following the terrorist attacks at Brussels Airport and Maelbeek metro station, the airport departure hall was evacuated and metro services were suspended. The crisis center advised the public to avoid Brussels and remain at home. Schools, government offices, and many businesses sent staff home, and public transport was largely halted. The observed lower percentage of commuting accidents on that day aligns with these extraordinary circumstances.

Table 3.145: Numbers and percentages of OA notifications by attacks (commuting and workplace)

attack	dateTA	wdayTA	n_comOA	n_wplOA	perc_comOA_c	perc_wplOA_c	perc_comOA_r	perc_wplOA_r
attack 01	2014-05-24	Saturday	1	12	0.0	0.0	7.7	92.3
attack 02	2014-12-14	Sunday	1	8	0.0	0.0	11.1	88.9
attack 03	2015-01-15	Thursday	8	60	0.0	0.0	11.8	88.2
attack 04	2016-03-15	Tuesday	5	75	0.0	0.0	6.2	93.8
attack 05	2016-03-22	Tuesday	13	79	0.0	0.0	14.1	85.9
attack 06	2016-08-06	Saturday	2	18	0.0	0.0	10.0	90.0
attack 07	2017-06-20	Tuesday	9	81	0.0	0.0	10.0	90.0
attack 08	2017-08-25	Friday	3	57	0.0	0.0	5.0	95.0
attack 09	2018-05-29	Tuesday	10	114	0.0	0.1	8.1	91.9
attack 10	2022-11-10	Thursday	5	46	0.0	0.0	9.8	90.2
attack 11	2023-10-16	Monday	18	64	0.1	0.0	22.0	78.0
no attack	NA	NA	32449	183808	99.8	99.7	15.0	85.0

3.4.5.10 Interplay of temporal patterns

In the next section, the temporal patterns discussed under Section 3.4.5.3, Section 3.4.5.4, Section 3.4.5.5, Section 3.4.5.6, Section 3.4.5.7, Section 3.4.5.8 and Section 3.4.5.9 are combined into a single model.

Generalized additive models were constructed separately for commuting and workplace OA. First, a complete time series by hour from 2014-01-01 to 2023-12-31 was created, with counts of commuting and workplace OA per hour left joined. Hours without any notified commuting and/or workplace OA were assigned (a) zero value(s). Weekdays were coded as Monday to Sunday. Holiday periods (school holidays, legal holidays, general holidays and extra holidays) were included as binary indicators and merged by date. Extreme weather conditions (snow, dust, fog, thunderstorm, haze,…) were included as binary indicators and merged by date, just like the variable indicating on which days terrorist attacks took place. The COVID-19 stringency index value was merged by date, as well as time change regimes (summer/winter times and their changes). These time change regimes (summer/winter time and the weeks before and after the changes) were included as a six level factor variable. Negative binomial regression models were fitted to account for overdispersion in the count data. To model the effect of time of day on accident risk, we used a cyclic cubic spline for the variable hour. This approach is appropriate for variables that follow a natural daily cycle, such as working hours. Specifically, we specified a smooth term using s(hour, bs = “cc”, k = 20) and defined the cycle boundaries with knots = list(hour = c(0, 24)). This ensures that the spline treats the start and end of the day as connected, allowing for a smooth transition between hour 0 and hour 24. The cyclic spline captures periodic patterns without introducing artificial discontinuities at the boundaries, making it well-suited for analysing phenomena that repeat over a 24-hour period.

In Figure 3.218 (a), a graphical representation of the commuting model is provided. The model reveals markedly lower Incidence Rate Ratio (IRR) for commuting accidents on Saturdays and Sundays, indicating a substantial reduction in activity during weekends. Reductions observed on school and legal holidays are of a similar magnitude to those seen on Fridays, suggesting consistent patterns of decreased commuting during non-working days. Most other predictors do not reach statistical significance. Notable exceptions include heavy precipitation and winter/freezing temperature conditions, which are associated with an increased number of commuting accidents. In contrast, cold temperatures and a higher COVID-19 stringency index are linked to a decrease in commuting accident rates. The model does not provide evidence that time changs are associated with increased numbers of commuting accidents.

In Figure 3.218 (b), a graphical representation of the workplace model is provided. The model indicates significantly lower IRR for workplace accidents on Saturdays and Sundays, reflecting reduced workplace activity during weekends. Similar reductions are observed during school and legal holidays, comparable to those seen on Fridays. Where for commuting accidents, estimates for school and legal holidays are quite similar, the model for workplace accidents indicates a stronger reduction during legal holidays compared to school holidays. Most other predictors do not reach statistical significance. Temperature effects seem to be less pronounced in comparison with the commuting model. Additionally, a higher COVID-19 stringency index correlates with reduced workplace accidents. The model does not indicate any significant association between time changes and higher numbers of workplace accidents.

Both models are shown side-by-side in Figure 3.219 for easier comparison. Overall, the temporal patterns for commuting and workplace OA are quite similar, with both models showing significant reductions on weekends and holidays. However, some differences are noted, such as the more pronounced effect of legal holidays on workplace accidents compared to commuting accidents. Weather-related factors also appear to have a stronger influence on commuting accidents. These findings suggest that while general temporal trends are consistent across both types of OA, specific factors may differentially impact commuting versus workplace incidents.

Both models incorporate the same set of predictors, including day-of-week effects, holiday indicators, extreme weather condition indicators, COVID-19 stringency index (C19si), time change regimes, and terrorist attacks. This consistency in predictor inclusion allows for a direct comparison of their effects across the two contexts. The IRR for weekdays show a similar decreasing trend from Tuesday to Sunday in both models, with the lowest IRR observed during the weekend, indicating reduced activity levels. Similarly, legal and school holidays are associated with significantly lower OA numbers in both commuting and workplace settings, suggesting a general reduction in mobility during these periods.

However, the models differ in their explanatory power and the statistical significance of certain predictors. The workplace OA model shows a substantially higher R² value (0.945) compared to the commuting OA model (0.703), indicating that it explains a greater proportion of variance in the outcome (number of accidents). This suggests that workplace OA numbers are more systematically influenced by the included predictors -possibly due to more structured behavioural patterns in work environments- whereas commuting may be more variable and shaped by personal choices. Although some temporal patterns may exert more systematic influence in the workplace model, their effects appear less pronounced, as the IRRs in the workplace model are generally closer to 1.

Differences also emerge in the significance levels of specific predictors. For instance, freezing conditions have a stronger effect in the commuting model (IRR = 1.26, p < 0.001) than in the workplace model (IRR = 1.04, p = 0.017), suggesting that abnormal cold may influence commuting accidents more than workplace accidents. The COVID-19 stringency index (C19si) is significant in both models, but again with slightly stronger effects in the commuting context (IRR = 0.85 in the commuting vs. 0.89 in the workplace model), potentially reflecting greater sensitivity of commuting patterns to public health restrictions. For the summer hour period as a whole (excluding the weeks before and after the time changes), the effect even appears to be opposite: a small but significant decrease in commuting OA numbers (IRR = 0.86, p < 0.001) versus a small but significant increase in workplace OA numbers (IRR = 1.05, p = 0.003).

Temporal patterns in commuting and workplace occupational accidents

Fewer occupational accidents on weekends and holidays. Both commuting and workplace accidents drop significantly on Saturdays, Sundays, and during school/legal holidays—reflecting reduced activity on these days.
Weather matters more for commuting occupational accidents. Harsh weather (like snow or freezing temperatures) increases commuting accidents more than workplace ones.
COVID-19 restrictions lowered all occupational accidents. Stricter COVID-19 measures were linked to fewer accidents, especially for commuting.
No evidence for negative impact of time changes from our analysis. Shifts to and from daylight saving time showed no clear link to more occupational accidents.

« previous

3.5.1 Likelihood to notify an OA (hadOA, model 1)

In this first multilevel logistic multivariable regression model, we examine the factors associated with the likelihood that an individual working in a company reports an OA in a given month during the ten-year study period (hadOA).

The model comprises 7,938,537 monthly observations (coded as 0 for no OA notified and 1 for an OA notified for a given individual in a given month), covering 291,661 unique individuals across 16,362 mutual ESPP/PS companies for which complete information on all model determinants is available.

Overall, 31,822 notifications are counted (out of total of 7,938,537 observations), corresponding to an overall monthly reporting probability of 0.4% per individual.

The model includes both fixed effects (individual and company level predictors) and random effects (group-level variation across companies and individual clusters in time) and explains a substantial portion of the variance, with a marginal R² of 0.136 and a conditional R² of 0.328. This indicates that both individual and group-level factors contribute meaningfully to the outcome. The random intercepts show that group level heterogeneity lies with individuals (individual variance = 0.76) as well as between companies (company variance = 0.19). The estimated ICC was 0.22. This result suggests that 22% of the variance is attributable to differences between individuals and between companies. Since the ICC is much larger than 0.10, the use of a multilevel approach is justified (Section 3.2.2.2). Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.220. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.221.

Significant determinants identified in the multivariable regression model, confirming the results of the corresponding univariable regression models, are ranked using the univariable regression model’s highest marginal/conditional R² ratios and reported with their multivariable regression Odds Ratio (OR).

Determinants associated with the likelihood of reporting an OA include the following.

Salary level. Wage categories were strongly associated with the odds of reporting an OA. Compared to the reference group earning less than <€50 per day, the highest odds were observed in the €100–124 bracket (OR = 4.44) with a gradual decline towards the highest wage category (\(\geq\) €250, OR = 2.43). Although the magnitude differs, our findings support the work of Cabrera-Flores et al. (2023), which found that among Peruvian industries between 2016 and 2021, a 10% increase in average wages was associated with a 19% decrease in the rate of work accidents. Additionally, systematic reviews and empirical studies consistently show that low-wage workers are more likely to underreport workplace injuries and illnesses. Barriers include fear of retaliation, job loss, or reduced hours, as well as lack of knowledge about reporting procedures and perceptions that injuries are “part of the job” (Gholamizadeh et al., 2023; Mudenha et al., 2022). This might explain the discrepancy of the high estimates of all categories compared to the reference category. Alternatively, the theory that higher wages may reflect compensation for increased job risk could also be relevant (Lalive, Rafael and Ruf, Oliver and Zweimuller, Josef, 2006). See Section 1.3.2.5 and Section 3.4.1.5.
Sector. Compared to the agriculture, forestry and fishing reference category (level A), only construction (level F) and human health and social work activities (level Q) show a significant higher likelihood of reporting an OA, with odds ratios of OR=1.31 and OR=1.28, respectively. In contrast, accommodation and food service activities (OR=0.49) and information and communication (OR=0.53), levels I and J, show significantly lower odds of reporting an OA. The higher odds for sector levels F and Q are in line with the high incidence rates reported by Eurostat. See Section 1.3.5 and Section 3.4.4.
Workforce segment. Employees with white-collar status had substantially lower odds for reporting an OA (OR = 0.47). This is in line with expectations and former research. See Section 1.3.3.1 and Section 3.4.2.1.
Biological sex. Men were more likely than women to report an OA (OR = 1.18), confirming findings from former studies on the topic. See Section 1.3.2.2 and Section 3.4.1.2.
Insurance companies. The model reveals significant differences in the likelihood of reporting an OA, depending on the insurer. The odds ratios range from 0.93 to1.37. This suggests that the insurer may play a role in the reporting behaviour and/or notification process. This could reflect differences in how insurers communicate with employers and workers, or facilitate the reporting process. We could hypothesize for example that some insurers may have more proactive procedures, better digital infrastructure, or promote stronger safety cultures that encourage reporting. As this is a novel observation, it warrants further investigation to confirm and explain the underlying mechanisms. See Section 1.3.4.4 and Section 3.4.3.5.
Age. Individuals aged 20–59 had significantly higher odds of reporting an OA compared to the reference group (under 20), with the highest odds observed among workers aged 20-29 (OR = 1.27). This confirms the commonly observed trend that younger workers are more likely to experience OAs. See Section 1.3.2.1 and Section 3.4.1.1.
Nature of the contract. Full-time workers had significantly higher odds of reporting an OA (OR = 1.39) which confirms previous literature on the topic. See Section 1.3.3.3 and Section 3.4.2.3.
Work experience. Longer work experience was associated with lower odds of reporting an OA, showing a decreasing trend from 1–4 years (OR = 0.94) to \(\geq\) 21 years (OR = 0.69). In this study, work experience is defined as the duration of the employment contract with the current employer which may relfect a mix of job-specific and employer specific experience, but not overall career experience. These findings are consistent with former research stating that lower job experience correlates with a higher likelihood of OA. See Section 1.3.2.3 and Section 3.4.1.3.
Company size. Larger companies (\(\geq\) 200 employees) were associated with significantly higher odds of OA occurrence (OR = 1.70) in the univariable model, compared to the reference group (<5 employees). However, this pattern was reversed in the multivariable model, which revealed a more nuanced relationship. Companies with up to 49 employees showed decreasing odds ratios, confirming earlier findings that smaller companies tend to report relatively more OAs. Between 50 and 200 employees, the odds ratios increased, approaching the level of the reference group, and continued to rise beyond 200 employees. This shift suggests a complex organisational effect, likely influenced by confounding factors such as sector, which are accounted for in the multivariable model. The discrepancy between the univariable and multivariable results highlights the importance of adjusting for such determinants. While the univariable model presents a straightforward positive association with company size, the multivariable analysis uncovers a negative relationship up to 50 employees, followed by a reversal in the higher brackets. These findings align with previous research and underscore the multifaceted nature of the relationship between company size and OA occurrence. See Section 1.3.4.2 and Section 3.4.3.2.
Work location. Certain Belgian provinces such as Limburg, Namur, East Flanders, and West Flanders showed significantly higher odds, compared to the reference province Antwerp. In contrast, Brussels and Hainaut did not show significantly higher odds. While the overall pattern in the multivariable model is similar to the univariable model, confidence intervals tend to be wider than those of other determinants. Additionally, the low univariable R² ratio of 1% (0.004/0.405) suggests that variation in the geographic location of the employer has only a limited impact on the model’s overall explanatory power. See Section 1.3.4.1 and Section 3.4.3.1.
Temporal effects. The likelihood to report an OA in 2023 remains still significantly lower than in 2014. While there was adeclining trend from 2014 to 2019, this pattern reverted in recent years (2020–2023). Across all years, reporting likelihoods also varied by month. Compared to January, the odds for reporting OAs were significantly lower in all months (except March, June, September and October). See Section 1.3.6 and Section 3.4.5.

The multivariable model highlights that individual demographic characteristics (age, gender), employment conditions (wage, experience, company size), and temporal factors (year and month) are significant determinants of reporting an OA. The effects of wage and company size suggest that structural and economic factors play a critical role, while the temporal decline may reflect broader labour market trends or policy changes.

Wage, sector and workforce segment, but also biological sex and insurer play an import role in notification patterns of occupational accidents

The multivariable model demonstrates substantial explanatory power for predicting the notification of an occupational accident. Wage, sector, and workforce segment show the highest marginal-to-conditional R² proportions in their respective univariable models, closely followed by biological sex and insurer. The model highlights the importance of accounting for both individual- and company-level variation when attempting to explain the likelihood of reporting an occupational accident.

« previous

Figure 3.221 shows the different effect sizes of the categories included in the model. The figure immediately indicates large effects for wage, workforce segment and sector since their estimates (dots with whiskers) lie relatively further from the OR=1 reference line compared to the other variables included in the model.

3.5.2 Likelihood to refuse an OA (statusOA, model 1.1)

This multilevel logistic multivariable regression model estimates the probability that a reported OA is refused by an OA insurer in a given month during the ten-year study period (statusOA).

The model comprises 31,645 observations or notifications of single OA (coded as 0 for a not refused OA and 1 for a refused OA), across 25,200 unique individuals and 7,218 mutual ESPP/PS companies for which complete information on all model determinants is available.

Overall, 2,797 refusals are counted out of 31,645 observations, indicating an overall refusal rate of 8.8%. This is somewhat lower as the figures reported in FEDRIS’ year report 2023 Fig 1.2b -for the private sector- ranging from 10.6% in 2014 to 16.2.3% in 2023 (Federal Agency for Occupational Risks, 2023).

The model includes fixed effects for a wide range of individual (victim-level) and collective (company-level) variables, as well as random effects for both individuals (personal identification number within NSSO (INSZ number) (insznr)) and companies (enterprise identification number within CBE (crbnr)). The variable pseudins refers to a pseudonymized identifier of an OA insurer and is included as a fixed effect, reflecting the limited and predefined group of 13 OA insurers in Belgium. Of these, 10 are included in the model. The remaining three were excluded due to either an insufficient number of cases in certain category combinations or a high proportion of unknown decisions (data not shown).

The random effect variance for individuals (insznr) is high (\(\tau_{00\,\text{insz}}\) = 329.80), indicating substantial unexplained heterogeneity at the individual level. Although the company-level variance is near zero (\(\tau_{00\,\text{crbnr}}\) = 0.00), suggesting that company affiliation does not contribute meaningfully to the likelihood of refusal by an insurer, we retain the clustering to respect the data’s hierarchical structure and ensure accurate standard error estimation.

The model shows a conditional R² of 0.990, indicating that nearly all variation in refusal outcomes is explained when both fixed and random effects are considered. However, the marginal R² is only 0.001, showing that the fixed effects, such as insurer identity, contribute virtually nothing in explaining the outcome of refusal. This suggests that the model’s explanatory power lies almost entirely in the between-individual variation, with the individual-level random effects accounting for nearly all of the explained variance. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.222. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.223.

The indications of slightly higher refusal chances for part-time workers (see Section 3.4.2.3) or workers in the company segment of 100 to 199 employees (see Section 3.4.3.2), or a lower refusal chance for insurer 9 (see Section 3.4.3.5), do not withhold in the multivariable model after correction for all other potential determinants.

Although the fixed effects for insurers (pseudins) show some variation in odds ratios (4, 2 and 9 somewhat lower versus 11 and 12 somewhat higher), none of these effects reach statistical significance. This means that -while differences in refusal behaviour still may occur- the model does not suggest any significant systematic insurer-level bias.

The model includes over 100 categorical levels, yet only one (the year 2017) appears statistically significant at the conventional 5% level. However, without correcting for multiple testing, this result may be misleading. With so many tests, it is statistically expected that around five categories could appear significant purely by chance. Therefore, the observed effect for 2017, although consistent with a broader trend of lower refusal rates between 2016 and 2017, may represent a false positive. Since our study does not cover the public sector, we cannot confirm the decreasing trend from 2020 on for the public sector shown in the FEDRIS year report 2023 Fig 1.2b (Federal Agency for Occupational Risks, 2023), but with our model showing a rather increasing trend from 2017 on, we can confirm the FEDRIS results for the private sector.

Since the model primarily includes demographic and employment-related variables (e.g. age, wage, nationality, company size) and lacks information on the circumstances of the accident itself (such as severity, type, or claim completeness, delay between the moment of the accident and it’s notification,…) it has only very limited ability to explain refusal decisions. This limitation reflects the model’s scope rather than a flaw, and suggests that refusals may be more closely tied to accident-specific or procedural factors not captured here.

Without inclusion of occupational accident specific determinants and procedural factors, no significant determinants for refusals are found

The model demonstrates strong fit at the individual level but weak explanatory power from fixed effects, including insurer identity. The dominant source of variation is individual-level heterogeneity, with company affiliation playing no measurable role. The model’s utility for identifying systematic refusal patterns is limited by the absence of occupational accident specific variables. Like for all other determinants included in the model, with exception of the year 2017, no significant differences in refusal patterns between insurers are noticed.

« previous

Figure 3.223 shows that the studied set of variables does not significantly determine whether an insurer accepts or rejects an OA claim. Hence, the model does not show indications of discrimination in insurance decisions, nor significant differences in refusal chances between insurers.

Potentential for further investigation

Some procedural accident meta variables (see Section 2.8.1.2) currently not included in the model may be worth further investigation. Examples include the time between the occupational accident and the first declaration to the insurer, the number of notifications, the similarity degree between multiple notifications, whether serious accidents were investigated within ten days after occurrence,…

« previous

3.5.3 Likelihood to experience an accepted commuting OA (comOA, model 2)

This multilevel logistic multivariable regression model estimates the probability of an insurer-accepted commuting OA (comOA) in a certain month during the ten-year study and explores the associated risk factors.

The model includes 7,938,537 monthly observations (0 no accepted commuting OA, 1 an accepted commuting OA for a given individual in a certain month), across 291,661 unique individuals and 16,362 mutual ESPP/PS companies for which information on all determinants included in the model is available.

Overall, 4,182 accepted commuting OA are counted under the whole of 7,938,537 observations, which implies an overall individual chance on an accepted commuting OA in a certain month of 0,1%. On a total of 28,848 accepted OA, the commuting accidents represent 14,5% of all accepted OA in our dataset.

The model includes fixed effects for individual- and company-level predictors and random effects to account for variation across individuals and companies over time. The random intercepts show that nearly all unexplained heterogeneity lies between individuals (individual variance = 52.45) rather than between companies (company variance = 0.00), resulting in an estimated ICC of approximately 0.94. Fixed covariates alone explain about 18.3% of the variance, while adding individual-level random intercepts increases the explained variance up to 94.6% (inferred from the null model since the conditional R² could not be calculated directly for the multivariable model), indicating strong person-specific baseline propensities for experiencing commuting OA. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.224. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.225.

Determinants of an insurer-accepted commuting OA include the following.

Salary level. Wage categories were associated with odds of an insurer-accepted commuting OA; compared to the reference category of <50 euro day wage, odds were highest in the 50-99 bracket (OR = 8.67) and gradually decreased towards the highest bracket (\(\geq\) 250, OR = 5.05. The discrepancy of the high estimates of all categories in comparison to the reference category of <50 needs further investigation. See Section 1.3.2.5 and Section 3.4.1.5.
Insurancy companies. 5 lowest (OR = 0.88), 8 highest (OR = 1.36). This suggests that differences between OA insurers in some way might affect the likelihood on an insurer accepted commuting accident. This is a new finding that needs confirmation in future studies. See Section 1.3.4.4 and Section 3.4.3.5.
Temporal effects. The model reveals seasonal variation with the highest relative risk in the January reference category and much lower odds in July (OR = 0.42) and August (OR = 0.50). Significantly elevated odds occur in 2017, 2022 and 2023 and a clear trend of increasing odds is apparent from 2020 (the Corona-crisis induced a steep drop in commuting displacements) to 2023. This year trend likely reflects a broader mobility trend rather than company-level changes (Assuralia, 2025b, 2025a). This pattern may be observed due to changes in commuting habits, such as the growing use of bicycles, differences in workforce composition, reporting practices, or operational characteristics such altering work patterns with a potential impact on commuting patterns. See Section 1.3.6 and Section 3.4.5.
Work experience. Longer work experience was associated with lower odds on an insurer-accepted commuting OA, with a decreasing trend from 1–4 years (OR = 0.87) to \(\geq\) 21 years (OR = 0.70). Tt is unclear how job or employer specific work experience would be related to commuting accidents. One possible explanation is that work experience may be acting as a proxy for younger drivers who are known to be at higher risk in traffic incidents. The current age categories may be too broad to capture this nuance, whereas the tenure variable -especially the “less than one year” category- might better reflect the elevated risk among young and inexperienced commuters. Future research could benefit from refining age categorization (e.g., isolating the 18–24 group) and comparing it more directly with tenure to disentangle experience from age-related risk. See Section 1.3.2.3 and Section 3.4.1.3.
Nature of work. Full-time workers had significantly higher odds (OR = 1.26). See Section 1.3.3.3 and Section 3.4.2.3.

Distance to work does not appear to influence risk after adjustment, but this does not rule out differences by mode of transport, which could not be included in the model. Like demonstrated in other reports, adding information on commuting mode could provide clearer insights (Assuralia, 2025b, 2025a). See Section 1.3.2.7 and Section 3.4.1.7.

The same applies to biological sex. After adjustment in the multivariable model, the previously significant association -suggesting that men experience significantly fewer commuting-related OAs- no longer holds. It is important to note that findings from other studies, such as the Medex study (Federal Public Service Health, Food Chain Safety and Environment, 2022), are often based on univariable analyses and limited to the subset of workers who have experienced (commuting) OAs. Additionally, previous research has highlighted that women are underrepresented as drivers and passengers in cars (VIAS Institute, 2018), which may influence observed patterns. While future studies may again identify a statistically significant effect of biological sex, we anticipate that both the effect sizes and the practical relevance of sex as a determinant of commuting-related OAs will remain limited. See Section 1.3.2.2 and Section 3.4.1.2.

In the present study, no evidence was found to support that age would be a key determinant of the likelihood of experiencing an accepted commuting accident. Previous research has reported higher incidence rates among older workers, particularly those aged 45–65, compared to younger age groups. However, the pattern observed in our data -although not statistically significant- suggested a decreasing rather than increasing trend between ages 20 and 60. Earlier studies have also examined the interaction between age and biological sex, indicating that older women may be especially vulnerable to commuting accidents, while older men are more frequently involved in work-related traffic accidents during working hours (Delgado-Fernández et al., 2022; Nenonen, 2011; Salminen, 2000). This interaction was not explored in the present study. See Section 1.3.2.1 and Section 3.4.1.1.

Limited impact of sex and age when other factors are considered

Although previous studies often identified biological sex and age as important risk factors for commuting accidents, our findings suggest their influence is relatively minor when other factors are considered across a broader worker population. We conclude that being male or older does not substantially increase or decrease the likelihood of experiencing a commuting occupational accident once additional variables within the exposed population are taken into account.

« previous

In 2017, Belgium experienced a relative increase in the number of traffic accidents occurring during commuting (home-to-work travel), as reported by the VIAS institute. This rise was notable despite a general decline in overall traffic fatalities and injury-related accidents, suggesting that commuting incidents constituted a proportionally larger share of total traffic accidents that year. Several contributing factors were identified, including increased mobility, longer commuting distances, the growing use of alternative transport modes such as bicycles and electric scooters, and improved reporting mechanisms for traffic OA. However, data from 2018 indicate that this upward trend was not sustained. The VIAS institute’s 2018 Traffic Safety Barometer, reported a continued decline in traffic fatalities and a moderate decrease in the share of commuting-related accidents, implying that the spike observed in 2017 may have been influenced by temporary or situational factors rather than representing a structural shift.

Overall, the individual and company-level variables included in the model have limited explanatory power compared to the strong influence of individual-level heterogeneity. These results imply that prevention strategies should focus on person-level factors rather than broad company-wide measures.

Companies however can play a valuable role in targeted risk communication. Young aged (and/or early-tenure, see higher) and full-time employees represent different risk groups, and interventions should account for seasonal patterns, with increased efforts during winter and early spring. Large firms may warrant commuting-risk audits despite negligible random company effects, as fixed effects suggest meaningful differences in exposure and reporting contexts. Distance alone is not informative, so future analyses incorporating commuting mode and exposure measures could improve interpretability and actionability.

Individual-level heterogeneity is more important in explaining insurer-accepted commuting occupational accidents over wage, time and some other fixed effects

Our research shows that the risk of having an insurer-accepted commuting accident depends much more on personal differences between individuals than on factors like salary, time of year, or other characteristics. This means that some people are simply more likely to have such accidents, regardless of where they work or how much they earn. As a result, prevention efforts should focus more on individual risk factors rather than on broad company-wide measures. Companies however can play a valuable role in targeted risk communication e.g. focusing on different risk groups with increased efforts in winter and early spring.

« previous

Figure 3.225 illustrates some significant differences for wage, insurer, month,… however, since individual risk appears to be substantially more influential than differences between companies, the observed effects should be interpreted with caution.

3.5.4 Likelihood to experience an accepted workplace OA (wplOA, model 3)

This multilevel logistic multivariable regression model estimates the probability of an insurer-accepted workplace OA (wplOA) in a certain month during the ten-year study and explores the associated risk factors.

The model includes 7,938,537 monthly observations (0 no accepted workplace OA, 1 an accepted workplace OA for a given individual in a certain month), across 291,661 unique individuals and 16,362 mutual ESPP/PS companies for which information on all determinants included in the model is available.

Overall, 24,666 accepted workplace OA are counted under the whole of 7,938,537 observations, which implies an overall individual chance on an insurer-accepted workplace OA in a certain month of 0.3%. On a total of 28,848 accepted OA, the workplace accidents represent 85.5% of all accepted OA in our dataset.

The model includes fixed effects at the individual and company level over time and random intercepts for persons and firms. The random effects indicate substantial heterogeneity at both levels (company variance = 0.20; individual variance = 0.91), with an ICC of approximately 0.25. Since the ICC is much larger than 0.10, the use of a multilevel approach is justified (Section 3.2.2.2). Fixed covariates explain 14.6% of the variance, while the full model (fixed plus random effects) explains 36.1%, suggesting that both observed characteristics and unmeasured person- or firm-level factors contribute more or less equally to variation in risk. This is in strong contrast with the results of the previously discussed model on commuting OAs. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.226. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.227.

Several individual characteristics are associated with the outcome. In the univariable models, wage (17%) and workforce class (13%) showed the highest marginal-to-conditional R² ratios, followed by sex (5%) and age (2%). Nature of contract, experience, and distance also explained some variance, but their ratios were close to 1% or less. Individual level determinants of an insurer-accepted workplace OA include the following.

Salary level. Daily wage (in euros) is strongly associated with higher odds across wage bands compared with the lowest reference. Although the effect decreases at the highest bands (e.g., 50–99 OR = 3.67; 100–124 OR = 4.42; \(\geq\) 250 OR = 2.42). See Section 1.3.2.5 and Section 3.4.1.5.
Workforce segment. White-collar employees (bedienden) have substantially lower odds (OR = 0.41) compared with blue-collar workers (arbeiders). See Section 1.3.3.1 and Section 3.4.2.1.
Biological sex. Men have higher odds than women (OR = 1.27). See Section 1.3.2.2 and Section 3.4.1.2.
Age. Compared with the reference age group, workers aged 20–59 show increased odds (e.g., 20–29 OR = 1.41; 30–39 OR = 1.27; 40–49 OR = 1.29; 50–59 OR = 1.28), while those aged 60+ do not differ significantly. See Section 1.3.2.1 and Section 3.4.1.1.
Nature of the contract. Full-time contracts are associated with higher odds (OR = 1.40). See Section 1.3.3.3 and Section 3.4.2.3.
Contract type. Limited-duration contracts show lower odds (OR = 0.82). See Section 1.3.3.2 and Section 3.4.2.2.
Work experience. Tenure within the current employer is protective, with progressively lower odds at longer tenures (1–4 years OR = 0.96; 5–10 years OR = 0.80; 11–20 years OR = 0.75; \(\geq\) 21 years OR = 0.72). See Section 1.3.2.3 and Section 3.4.1.3.
Commuting distance. Distance to work shows only minor associations (10–19 km OR = 1.07). See Section 1.3.2.7 and Section 3.4.1.7.

Company and contextual characteristics also influence the likelihood. In univariable models, NACE-BEL 2008 sector and insurer fixed effects showed high marginal-to-conditional R² ratios (14% and 4%, respectively). Company-level determinants of an insurer-accepted workplace OA include the following.

Sector. Several NACE-BEL 2008 sections have lower odds compared with the reference (e.g., I OR = 0.42; J OR = 0.26; K OR = 0.64; M OR = 0.57; R OR = 0.67; S OR = 0.56), while section F (construction, OR=1.25) and Q (human health and social work activities care, OR=1.21) are the only sectors with a higher likelihood compared to agriculture, forestry an and fishing (level A). See Section 1.3.5 and Section 3.4.4.
Insurance companies. Insurer category (pseudins) also matters, some categories show higher odds (e.g., 9 and 11 OR = 1.31) and others lower (e.g. 5 OR = 0.92). See Section 1.3.4.4 and Section 3.4.3.5.
Work location. Geographic location of the employer. Some provinces have higher odds (e.g., Limburg OR = 1.46; Namur OR = 1.83; West Flanders OR = 1.15), others lower (e.g., Brussels OR = 0.80), while several show no meaningful difference. See Section 1.3.4.1 and Section 3.4.3.1.
Company size. Odds are lower for firms with 5–49 employees (OR 0.75 to 0.82), similar for 50–99, and substantially higher for larger firms (100–199 OR = 2.36; \(\geq\) 200 OR = 17.68) compared with the smallest reference group. See Section 1.3.4.2 and Section 3.4.3.2.

And finally, also time level effects seem to contribute to the risk. In univariable models, year and month fixed effects showed small but significant effects that could be confirmed in the multivariable model. Time-level determinants of an insurer-accepted workplace OA include the following.

Year. Over time, odds are generally lower than in the baseline year, with a marked reduction from 2020 onwards but showing a slightly increasing trend (2020 OR = 0.77 to 2023 OR = 0.84). See Section 1.3.6 and Section 3.4.5.
Month. Seasonal variation is present, with several months (April, May, July and December) below January levels and slightly elevated odds in June (OR = 1.09), September and October (OR = 1.08). See Section 1.3.6 and Section 3.4.5.

Overall, these findings indicate higher risk among men, younger to middle-aged workers, those on full-time contracts, and employees in very large firms, and lower risk among those with longer tenure, white-collar status, certain sectors, and in many months and recent years compared with their respective reference categories.

Wage, workforce segment, sector, biological sex, and insurer influence workplace occupational accidents patterns

The multivariable model shows good explanatory power for accepted workplace occupational accidents. Wage, workforce segment, and sector contribute the most, as indicated by their high marginal-to-conditional R² proportions in the corresponding univariable models. Biological sex and insurer also play a notable role. Both individual-level and company-level heterogeneity are key drivers in the observed patterns.

« previous

Figure 3.227 displays the estimated effect sizes for each category included in the model. The figure clearly highlights substantial effects for wage, workforce segment, and sector, as their estimates (represented by dots with whiskers) are positioned further from the OR = 1 reference line compared to other variables, indicating stronger associations with accepted workplace accidents.

3.5.5 A number of reflections on our commuting and workplace accident models

Our findings indicate that, based on the variables included in the models, it is challenging to formulate targeted prevention policies that would consistently reduce accident rates. This difficulty is primarily due to the substantial influence of random effects, reflecting unmeasured heterogeneity at both the individual and company levels. For workplace accidents (wplOA, see Section 3.5.4), the random effect at the company level is particularly pronounced, suggesting that organisational context plays a significant role in shaping safety outcomes. In contrast, for commuting accidents (comOA, see Section 3.5.3), the company-level random effect is negligible, indicating that such incidents are less influenced by organisational characteristics.

Both models include a range of fixed effects at the individual and organisational levels. In Section 3.5.4, several predictors show strong associations with accident likelihood, including age, sex, tenure, employment status, company size, and sector. Younger workers, males, full-time employees and those with shorter tenure are more likely to experience workplace OA. Organisational characteristics such as large company size (\(\geq\) 200 employees) and high-risk sectors, particularly construction, also contribute significantly. In comOA, while some individual-level predictors such as tenure and employment status remain relevant, the overall explanatory power of the model is lower, and company-level characteristics have limited predictive value.

Temporal variables provide additional insights. From 2020 onward, both models show an increased likelihood for OA, probably reflecting changes in work patterns, reporting behaviour, or external conditions such as the COVID-19 pandemic. Clear seasonal trends also emerge: commuting accidents are more likely to occur between November and February, while workplace accidents show higher likelihoods in March, June, September, and October. These patterns suggest potential levers for organisations, such as adjusting work schedules, operational planning, or communication campaigns to align with periods of elevated risk.

Although distance to work is included as a variable in the commuting model, it does not emerge as a strong predictor. This suggests that other factors -such as mode of transport and timing of travel- may be more relevant. These factors are potentially modifiable through company policies, including flexible start times, telework arrangements, and mobility programs. While the statistical contribution of company-level variance in comOA is minimal, large firms may still be well-positioned to implement and scale commuting-related interventions.

The current models do not include data on specific preventive measures, such as safety training, engineering controls, or behavioural interventions. Consequently, their effectiveness cannot be evaluated within the scope of this analysis. This limitation highlights the need for a follow-up phase in which the random effects are explored in greater detail. Incorporating additional contextual information -such as the presence of safety campaigns, onboarding programs, or infrastructure investments- could help reduce the unexplained variance and allow for a more precise assessment of targeted interventions. However, collecting such data over a sufficiently long period remains a practical challenge.

Sectoral differences in OA risk are to be expected, but they also raise questions about how organisations manage risk. This relates to the concept of safety culture, which reflects the extent to which risk awareness and management is embedded in organisational practices. Although safety culture is inherently difficult to measure, potential proxies, such as near-miss reporting rates, or the frequency of supervisor safety conversations, may offer valuable insights. Future research should explore how such indicators can be integrated into predictive models.

A further avenue for investigation involves evaluating the impact of preventive measures over time. This could include assessing whether interventions such as training or workplace modifications lead to measurable reductions in accident rates within companies, and whether effective practices are adopted across similar organisations or sectors. Such analyses could help clarify the sector-wide impact of prevention strategies and support evidence-based policy development.

From a prioritization perspective, workplace safety (or prevention of workplace accidents) represents the most substantial and controllable opportunity for prevention advisors, particularly in large firms and high-risk sectors such as construction. Commuting safety (or prevention of commuting accidents), while less influenced by organisational context, remains a relevant area for targeted, policy-driven interventions. These are most effective when strategically aligned with seasonal risk patterns, early stages of employment, and company-level decisions regarding mobility and scheduling.

Keep the focus on workplace occupational accidents

Since around 85% of accepted occupational accidents are workplace-related (wplOA) and only about 15% are commuting-related (comOA), the most impactful and controllable opportunities for company-facing prevention advisors lie in addressing workplace accidents. ESPP prevention advisors work directly with companies, and because the largest share of accidents occurs in the workplace, focusing on wplOA stays more important than on commuting-relating accidents.

« previous

The multilevel model for workplace accidents also shows a strong signal between companies, suggesting that prevention advisors can make a difference by tailoring safety interventions to specific company profiles. This requires a strategic prioritizing and integration of layering measures that fit the company’s context and workforce characteristics.

It’s important to recognize that some factors influencing OAs are structural, such as wage level, tenure, contract type, and company size. While these elements fall outside the direct influence of prevention advisors, understanding their role is essential for the development of effective strategies. Conversely, modifiable factors such as experience and awareness and can be influenced through training, onboarding, mentoring, and seasonal safety campaigns.

Future research could aim to integrate more contextual and cultural variables to better understand and reduce the random effects that currently limit the predictive precision of the models.

Recognize structural components like low-wage, full-time and young workers in different company size segments and look for modifiable opportunities

For commuting accidents, seasonal campaigns, flexible work arrangements and transportation mode policies could help mitigate risks. For workplace accidents, targeted training, onboarding, mentoring, improved safety protocols and all other prevention measures fostering a strong safety culture could become key strategies. Prevention advisors are well-positioned to support organisations in addressing these modifiable factors, while also supporting them in remaining aware of the structural components that shape risk profiles.

« previous

3.5.6 Frequence degree of workplace OA (freqOA, model 4)

In this multilevel multivariable regression model we examine the factors associated with the company’s frequency degree in a given year during the ten-year study period (freqOA).

The model includes 81,780 yearly observations (frequency degree for a given company in a certain year), across 17,670 unique mutual ESPP/PS companies for which information on all determinants included in the model is available.

Many companies do not experience workplace OAs, resulting in a frequency degree of zero. Hence, the median and 75ile frequency degree of our sample are both 0. The overall mean frequency degree across all companies, being strongly influenced by the larger values of companies with such countable OA, is 15.8.

The model includes both fixed effects (company level predictors) and random effects (group-level variation across companies in time). However, with a marginal R² of 0.058 and a conditional R² of 0.092, it accounts for only a small proportion of the variance, leaving 90.8% unexplained. This suggests that group-level factors do not substantially contribute to the outcome. The ICC was 0.04, suggesting that only 4% of the variance is attributable to differences between companies, the majority of variance being situated within companies over time. With an ICC smaller than 0.10, the use of a multilevel approach is not strictly required. But, its use facilitates the comparison across models. Additionally, these findings align with recent research of Hallowell et al. (2021), questioning the predictive validity of the Total Recordable Incident Rate (TRIR) -an international equivalent to the Belgian frequency degree, normalized per 200,000 instead of 1,000,000 worked hours- suggesting that this metric is largely random and lacks meaningful interpretability.

The R² ratio is relatively high (0.63), indicating that most of the explained variance comes from fixed effects. However, given the overall low explained variance, these fixed effects cannot be considered as strong determinants. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.228. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.229.

Determinants of the frequency degree include the following.

Insurance companies. The estimates -5 lowest (\(\beta\) = -2.3), 9 highest (\(\beta\) = +15.6)- suggest that differences between OA insurers in some way are correlated with the mean frequency degrees of their clients. This represents a new finding that needs further investigation in future studies. See Section 1.3.4.4 and Section 3.4.3.5.
Sector. Compared to agriculture, forestry an and fishing (level A), only construction (level F) and manufacturing (level C) show significant higher frequency degree estimates (\(\beta\) = +8.3 and \(\beta\) = +4.7, respectively). The lowest estimate is observed in accommodation and food service activities, (level I), with \(\beta\) = -10.2. The relative ranking of these results is in line with the frequency degrees reported in FEDRIS’ year report 2023 Fig 12.1c (Federal Agency for Occupational Risks, 2023). See Section 1.3.5 and Section 3.4.4.
Company size. Compared to the smallest companies with <5 workers, all other size categories appear to be show higher frequency degrees, following a U-shaped relationship. The highest estimates are observed in category between 5-9 employees (\(\beta\) = +8.6) and in companies with >200 employees (\(\beta\) = +11.0). A more moderate increase of frequency degree is seen in the 20-49 employee companies (\(\beta\) = +4.5). See Section 1.3.4.2 and Section 3.4.3.2.
Nature of the contract. A higher proportion of full-time workers significantly increases the frequency degree of a company (\(\beta\) = +4.7). See Section 1.3.3.3 and Section 3.4.2.3.
Workforce segment. A higher proportion of blue-collar workers significantly increases the frequency degree of a company (\(\beta\) = +13.9). See Section 1.3.3.1 and Section 3.4.2.1.
Salary level. Compared to having many workers in the highest earnings group (daily wage > € 250), having high proportions of workers in the segments <50 up to 125to149 workers significantly contributes to higher frequency degree estimates at company level (\(\beta\) = +6.9 to \(\beta\) = +8.5). See Section 1.3.2.5 and Section 3.4.1.5.

In practical terms, the model captures only a limited part of the observed variation in frequency degree, even after allowing every company being its own baseline via the random intercepts. This should not be interpreted as a failure of the modelling approach but rather aligns with expectations from the TRIR literature regarding short‑interval injury rates. Literature findings could be summarised as follows: OA are rare, subject to noise, and statistically unstable at the temporal resolution typically used in organisational reporting. While measures such as the frequency degree can provide directional insights and support risk stratification, they are not suitable for precise forecasting at the employer‑year level.

Variation in company-specific frequency degrees is highly random (>90%) and although factors such as insurer, sector, company size and other variables show statistical significance, their explanatory relevance remains limited

The multivariable model demonstrates limited explanatory power for frequency degrees. Occupational accident insurer companies, sector and size, and in employee fractions the nature of the contract, workforce segment and salary level have significant impacts on the model scores. However, if we consider the large portion unexplained variance (90.8%), the model is hardly useful in practice and confirms that caution is needed in using a TRIR-like safety measure such as the frequency degree at company level.

« previous

Figure 3.229 shows the different effect sizes of the categories included in the model. When compared with the previous model overviews, the figure shows wide confidence intervals indicating the high variability of frequency degrees.

Potential biasses to be aware of when using TRIR-like metrics such as a frequency degree

General underreporting. Our model’s nationality/region effects and negative coefficients for some groups might reflect reporting artifacts; peer reviewed studies document substantial undercounting and employer-level recordkeeping gaps (Azaroff et al., 2002; Rappin et al., 2016; Wuellner & Bonauto, 2014).
Distorted reporting. When TRIR is used as a performance target -for example, in contracts or bonus schemes- it can unintentionally distort reporting behaviour. Borderline cases may go unrecorded, undermining TRIR’s reliability as a safety metric. This issue is particularly pronounced across jurisdictions with varying reporting standards and compensation incentives. In Belgium, this distortion is less likely to affect frequency rate reporting, as it falls under federal rather than regional legislation. However, the risk remains significant when TRIR- like measures are imposed as a performance targets for contractors or third parties. In such cases, the pressure to meet targets can lead to strategic underreporting, ultimately compromising the integrity of safety data.
Confounding by workforce/industry mix, limiting cross-company benchmarking. Our large, significant effects for industry (e.g., Manufacturing “C”, Construction “F”), role mix (blue-collar, full-time) and establishment size mirror critiques that TRIR-like measures (such as a frequency degree) reflect many contextual factors beyond “safety performance” (Reiman & Pietikäinen, 2012)
Sensitivty to tenure composition. Our positive/significant coefficient for the 1–4 year tenure band is consistent with elevated early-tenure injury risk (“new worker” risk), implying TRIR will swing with hiring/turnover independent of controls (Bena et al., 2013).
Lagging indicators are of limited value for proactive control. The modest marginal R² and low ICC in our mixed model are consistent with the argument that lagging outcome metrics like TRIR capture multifactor results rather than system capability; leading/monitor indicators (e.g. safety training completion rates, near-miss reporting frequency, safety audits and inspections, PPE usage, safety engagement, hazard identification and risk reduction rate,…) are needed alongside measures as TRIR (Reiman & Pietikäinen, 2012).

« previous

3.5.7 Likelihood to experience a serious workplace OA (seriousOA, model 5)

This multilevel logistic multivariable regression model examines the factors associated with the likelihood an individual employed within a company experiences a serious workplace OA in a given month over the course of the ten-year study period (seriousOA).

The model includes 17,267 workplace OAs (0= no serious workplace OA, 1= serious workplace OA) for a given individual in a certain month, These observations span 14,198 unique individuals and 4,154 mutual ESPP/PS companies for which complete information on all model determinants is available.

Of the 17,267 workplace OAs, 1,136 are classified as serious, resulting in an overall individual chance of 6.6% for a workplace OA to be categorized as serious.

The random effect variance for individuals (insznr) is high (\(\tau_{00\,\text{insz}}\) = 552.81), indicating substantial unexplained heterogeneity at the individual level. Although the company-level variance is considerably lower (\(\tau_{00\,\text{crbnr}}\) = 4.14), suggesting that company affiliation does only contribute marginally to the classification of a workplace OA as serious, we retain the hierarchical clustering structure. This ensures accurate estimation of standard errors and facilitates meaningful comparison across models.

The model yields a conditional R² of 0.994, indicating that nearly all variation in classification the workplace OA as serious is explained when both fixed and random effects are considered. However, the marginal R² is only 0.005, showing that the fixed effects, such as part-time work, contribute virtually nothing in explaining this outcome. This suggests that the model’s explanatory power lies almost entirely in the between-individual variation, with the individual-level random effects accounting for nearly all of the explained variance. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.230. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.231.

Determinants of experiencing a serious OA include the following.

Company size. Larger companies (200 workers and more) were associated with lower odds for serious workplace OA (OR = 0.13). See Section 1.3.4.2 and Section 3.4.3.2.
Nature of the contract. Part-time workers had significantly lower odds for a serious workplace OA (OR = 0.38). See Section 1.3.3.3 and Section 3.4.2.3.
Temporal factors. The odds of a serious workplace accident was significantly higher in 2020 (OR = 3.59) and 2023 (OR = 3.55), relatively to the reference year 2014. See Section 1.3.6 and Section 3.4.5.

The multivariable model identifies significant employer level (company size), individual level (nature of the contract) and temporal level (year) determinants that influence the classification of a workplace OA as serious. However, without adjustment for multiple testing, these results may be misleading and may represent false positives. Nevertheless, the finding that the largest employer segment (200 workers and more) is associated with lower odds for a serious accident is consistent with previous research (Kines & Mikkelsen, 2003), lending credibility to this specific result.

As for model 1.1 (statusOA, see Section 3.5.2), a similar caveat applies here. Since this model primarily includes demographic and employment-related variables (e.g. age, wage, nationality, company size), it lacks information on the specific circumstances of the accident itself (e.g. the different administrative codes that are used to classify a workplace OA as serious). As a result, its capacity to explain why a particular accident is classified as serious is inherently limited. This limitation reflects the scope of the model rather than a flaw, and underscores the importance of accident-specific variables in accurately capturing the severity of workplace accidents.

Size of the company and nature of the contract seems to be associated with classification of a workplace occupational accident as serious, but individual level variation and accident specific factors might be far more important

The model demonstrates strong fit at the individual level but weak explanatory power from fixed effects. The dominant source of variation is individual-level heterogeneity, with company affiliation playing only a minor role. The model’s utility for identifying systematic patterns for classifying a workplace occupational accident as serious is limited by the absence of accident-specific variables.

« previous

Figure 3.231 shows the different effect sizes of the categories included in the model. When compared with the model overview figures of the individual models 1, 2, and 3 (see Section 3.5.1, Section 3.5.3, and Section 3.5.4), the present figure displays wide confidence intervals, indicating high variability in the classification of a workplace OA as serious. We also observe that only a few of the whiskers (fewer than five) do not cross the reference line OR = 1.

3.5.8 Days absent (validated) due to a workplace OA (nDaysTAO, model 6)

This multilevel multivariable regression model examines the factors associated with the IRR of the number of days absent due to a workplace OA over the ten-year study period (nDaysTAO).

The model includes 17,287 workplace OA across 4,168 mutual ESPP/PS companies for which complete information on all model determinants is available.

In the dataset used for this model, a substantial number of workplace OA (5,867 cases or 33.9%) result in zero days of absence from work. The median duration of absence is 5 days, while the mean is 24 days absence.

The model includes both fixed effects (company-level predictors) and random effects (group-level variation across companies over time). With a marginal R² of 0.200 and a conditional R² of 0.330, the model explains a substantial portion of the variance. This indicates that group-level factors substantially contribute to the outcome. The ICC was 0.16, suggesting that only 16% of the variance is attributable to differences between companies. Since the ICC is larger than 0.10, the use of a multilevel approach is justified (Section 3.2.2.2).

The model yields a relatively high R² ratio (0.61), indicating that most of the explained variance is attributable to fixed effects. Since the total explained variance is also relatively good, the identified fixed effects can be considered as valuable determinants. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.232. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.233.

Determinants of the FEDRIS-validated number of days absent due to a workplace OA include the following.

Age. The IRR gradually increases from 1.49 for workers in their twenties, to 5.30 for those aged 60 and above. This pattern confirms the commonly observed trend that older workers tend to have longer duration of work absence. See Section 1.3.3.3 and Section 3.4.2.3.
Sector. Compared to agriculture, forestry and fishing sector (level A), the lowest incidence rate ratios for days absent can be found in sectors M (Professional, scientific and technical activities, IRR = 0.41), L (Real estate activities, IRR = 0.49) and S (Other Service activities). The highest IRR was found in sector F (construction, IRR = 1.16) although this difference was not statistically significant compared to level A. See Section 1.3.5 and Section 3.4.4.
Company size. Compared to the reference category of companies with <5 than employees, the IRR for the number of days absent gradually decreases as company size increases, ranging from an IRR of 0.85 in the 20-49 employee segment to 0.72 in companies with 200 or more employees. This confirms previous literature indicating that smaller companies report more lost workdays following a workplace OA (Kines & Mikkelsen, 2003). See Section 1.3.4.2 and Section 3.4.3.2.
Workforce segment. Employees with a white-collar status showed a substantially lower IRR (IRR = 0.85) for the validated number of days absent. See Section 1.3.3.1 and Section 3.4.2.1.
Work experience. Longer work experience was associated with higher IRR for the validated number of days absent, with an increasing trend from 1.14 for employees with 1–4 years of experience to 1.31 for those with \(\geq\) 21 years (IRR = 1.31). See Section 1.3.2.3 and Section 3.4.1.3.
Insurance companies. IRR varied notably between OA insurers, ranging from 0.61 (Insurer 11) to 1.17 (Insurer 9). This suggests that differences between OA insurers in some way affect the number of days absent. This is a new finding that needs confirmation in future studies. See Section 1.3.4.4 and Section 3.4.3.5.
Salary level. Wage categories were associated with IRR of the number of days absent due to a workplace OA. Compared to the reference category of less than €50 daily wage, the highest IRR was observed in the €50–99 bracket (IRR = 1.31), while the lowest was found in the highest wage category of \(\geq\) €250 (IRR = 0.64). See Section 1.3.2.5 and Section 3.4.1.5.
Nature of the contract. Part-time workers had significantly lower IRR for validated days of absence (IRR = 0.75). See Section 1.3.3.3 and Section 3.4.2.3.
Temporal effects. The IRR for the number of days absent due to a workplace OA was significantly higher in 2016, 2019, 2021 and 2022, compared to 2014. Additionally, accidents occurring in July were associated with significantly lower IRR (IRR = 0.85) compared to those occurring in January. See Section 1.3.6 and Section 3.4.5.

The multivariable model highlights that individual-level factors (age, work experience, workforce segment, salary level, nature of the contract), employer-level factors conditions (sector, insurer, size), and temporal factors (year and month) are significant determinants of the validated number of days absent.

Although specific parameters of the OA itself (not included in the current model) may have a substantial impact on the number of days absent (Federal Public Service Health, Food Chain Safety and Environment, 2022), present findings indicate that general group-level predictions can be made. The determinants from the current model allow for the estimation of absence profiles at the company-level due to workplace OA.

Age, sector and size of the company are valuable determinants to estimate a number of (FEDRIS-validated) days absent from work, even without any specific information of the occupational accident itself

The multivariable model demonstrates high explanatory power for the (FEDRIS validated) number of days absent from work following a workplace occupational accident. Age, sector and company size have the highest marginal to conditional R² proportions in their corresponding univariable models. Company-level heterogeneity plays an important role and has to be taken into account.

« previous

Figure 3.233 shows the different effect sizes of the categories included in the model. The figure immediately indicates large effects for e.g. age and sector since their estimates (dots with whiskers) lie relatively further from the IRR = 1 reference line compared to the other variables included in the model.

3.5.9 Days absent (calculated) due to a workplace OA (lengtecor, model 6.1)

This multilevel multivariable regression model is completely analogous to the previous model. It examines the factors associated with the IRR of the number of days absent due to a workplace OA over the ten-year study period, but this time using absence duration calculated by Liantis (lengtecor).

This measure is derived directly from Liantis PS calendars, allowing for immediate estimation of absence duration without waiting for external validation by a governmental instance (i.e. FEDRIS, which kindly provided the validated absence data used in the previous model, see Section 2.6).

The model includes 9,822 workplace OA across 2,946 mutual ESPP/PS companies, for which complete information on all included determinants is available.

A notable proportion of cases (477, or 4.9%) in the dataset used for this model result in zero days of absence from work. The median calculated absence duration following a workplace OA is 10, and the mean is 33 days.

The model incorporates both fixed effects (company-level predictors) and random effects (group-level variation across companies over time). With a marginal R² of 0.111 and a conditional R² of 0.381, the model explains a moderate portion of the variance, indicating that group-level factors contribute substantially to the outcome. The ICC of 0.30, suggests that 30% of the variance is attributable to differences between companies. Since the ICC is much larger than 0.10, the use of a multilevel approach is justified (Section 3.2.2.2).

The relatively high conditional R² compared to the marginal R² (ratio of 0.29) indicates that company-level heterogeneity plays an important role. The fixed effects identified in the model can therefore be considered valuable determinants of calculated absence duration. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.234. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.235.

Determinants of the (Liantis-calculated) number of days absent due to a workplace OA include the following.

Age. The IRR progressively increases from 1.64 for workers in their twenties, to 3.93 for those aged 60 or more. This confirms the trend that older workers tend to be absent longer following an OA. See Section 1.3.3.3 and Section 3.4.2.3.
Sector. Compared to agriculture, forestry and fishing (level A), the lowest IRR values were found in sectors M (Professional, scientific and technical activities, IRR = 0.41) and L (Real estate activities, IRR = 0.41), while sector F (Construction) did not show a significant difference (IRR = 0.97, ns). See Section 1.3.5 and Section 3.4.4.
Work experience. Longer work experience was associated with higher IRR, with a clear increasing trend from 1–4 years (IRR = 1.13) to \(\geq\) 21 years (IRR = 1.43). See Section 1.3.2.3 and Section 3.4.1.3.
Company size. Compared to companies with <5 employees, the number of days absent decreases as company size increases. Significant reductions were observed in the 20–49 segment (IRR = 0.80), 50–99 (IRR = 0.75), and \(\geq\) 200 (IRR = 0.80). These findings support earlier research that smaller companies tend to report more lost workdays following an OA (Kines & Mikkelsen, 2003). However, the overall trend appears somewhat less clear than in the previous model. See Section 1.3.4.2 and Section 3.4.3.2.
Workforce segment. White-collar employees had significantly higher IRR (IRR = 1.17), which contrasts with the validated model where they had lower IRR. This difference should be further investigated. See Section 1.3.3.1 and Section 3.4.2.1.
Insurance companies. The variation observed between insurers, ranging from IRR = 0.77 (lowest, insurers 11 and 5) to IRR = 1.14 (highest, insurer 9), is consistent with the direction and magnitude in the former model. However, in the current model, this difference does not reach statistical significance. See Section 1.3.4.4 and Section 3.4.3.5.
Temporal effects. As in the previous model based on the validated absence data, the years 2019, 2021, and 2022 show the highest estimates for absence duration. However, this difference is not statistically significant compared to the reference year 2014. While the previous model identified July as the month with a notably lower IRR, the current model highlights August as significantly lower (IRR = 0.85). Together, both models suggest possible seasonal (summer) and/or holiday-related effects influencing absence duration. See Section 1.3.6 and Section 3.4.5.

The current multivariable model demonstrates that individual-level factors (age, work experience, workforce segment), employer-level conditions (sector, company size), and temporal factors (month) are significant determinants of the Liantis-calculated number of days absent. Although OA-specific parameters (e.g. injury type, severity) are not included, the model allows for group-level predictions of absence profiles based on structural characteristics.

Age, sector, and company size again emerge as strong predictors, confirming their utility in estimating group_level absence duration without needing OA specific information post occurrence of the accident. The model’s explanatory power, especially the high conditional R², underscores the importance of accounting for company-level heterogeneity in absence modeling.

Age, sector and size of the company are valuable determinants to estimate a number of (Liantis-calculated) days absent from work, even without any specific information of the occupational accident itself

The multivariable model demonstrates high explanatory power for the Liantis-calculated number of days absent from work following a workplace occupational accident. Age, sector and company size have the highest marginal to conditional R² proportions in their corresponding univariable models. Company-level heterogeneity plays an important role and has to be taken into account.

« previous

Figure 3.235 shows the different effect sizes of the categories included in the model. The figure immediately indicates large effects for e.g. age and sector since their estimates (dots with whiskers) lie relatively further from the IRR=1 reference line compared to the other variables included in the model.

3.5.10 Days absent modelling results compared

Liantis’ internally calculated number of days absent following a workplace OA -used in model 6.1 (Section 3.5.9)- appears to be a valid and useful proxy for modelling absence duration. However, this variable (lengtecor) is not a perfect substitute for the validated government variable (nDaysTAO) used in model 6 (Section 3.5.8).

Model 6.1 relies on Liantis’ internal data, which is more readily accessible and can be processed quickly. This makes it particularly valuable when validated data are not (yet) available. Although FEDRIS provided validated data for this study, there is currently no formal mechanism to access such data on a case-by-case basis outside this research context. As a result, the operational usefulness of model 6 remains limited.

Model 6, which uses validated absence data, explains more variance in absence duration (marginal R² = 0.200 vs 0.111), suggesting that it is more precise and captures a broader range of real-world variation. Additionally, the number of observations is nearly double (17,287 validated vs 9,822 calculated), reinforcing our earlier recommendation: it would be highly beneficial to structurally share the “final truth” concerning the number of days absent following an OA alongside the definitive status of closed OA files.

Only half of the validated cases could be processed directly via Liantis

Further research is needed to refine the algorithm that estimates absence duration using Liantis PS calendars since absence could only be estimated for 9,822 of the 17,287 validated cases.

« previous

Structurally sharing key outcomes with stakeholders would increase value for all parties

Critical outcomes -such as the final number of days absent, total cost of the accident, final dossier status, and seriousness of the OA- are currently not shared across stakeholders in a structured way. This limits the potential for evidence-based prevention and hinders efforts to support reintegration and return-to-work strategies.

« previous

Some predictors (such as age) exhibit stronger effects in the model based on validated data (Model 6), suggesting that this model captures more nuanced or extreme cases. In contrast, Model 6.1 tends to attenuate these effects, likely due to estimation errors or missing details in the internal algorithm used to calculate absence duration. Interestingly, Model 6.1 identifies a significant effect for biological sex, with men appearing to have significantly more calculated absence days. However, since this effect is not observed in Model 6, further research is needed to determine whether this finding reflects a true underlying difference or is merely an artifact of the internal calculation method. Interestingly, the higher ICC in model 6.1 (0.30 vs 0.16 in model 6) suggests stronger clustering by employer or sector. This may reflect systematic biases or limitations in how internal absence data is structured or recorded.

Use model 6.1 for operational monitoring and early warning; model 6 for research and validation

Given the easier accessibility of the source data and the reasonable predictive power, model 6.1 might be suited for operational use and early warning systems. If validated data were to become structurally available, hybrid models combining internal and external sources could be developed to enhance accuracy without sacrificing timeliness. Continued validation of both models is recommended, along with further investigation into discrepancies to identify potential biases and improve the internal absence calculation algorithm.

« previous

3.5.11 Direct wage cost due to a workplaceOA (costOA, model 7)

This multilevel multivariable regression model explores the determinants of the log-transformed cost (base 10) of OA over the ten-year study period. To avoid results approaching minus infinity, one euro was added to the cost variable, resulting in the transformation: \(\log_{10}(\text{costOA} + \text{€}1)\). This transformation was applied to reduce skewness and stabilize variance in the cost distribution.

The model includes 20,423 observations across 4,105 mutual ESPP/PS companies for which complete information on all included determinants is available.

In the dataset used for this model, a substantial proportion of workplace OA (5,370 cases, or 26.3%) result in €0 direct wage cost. The median cost is €867 (corresponding to 2.94 on the log scale), while the mean cost, reflecting a highly skewed distribution, was €180.
The model includes both fixed effects (company-level predictors) and random effects (group-level variation across companies in time). With a marginal R² of 0.153 and a conditional R² of 0.378, it explains a substantial portion of the variance in OA-related costs. This indicates that group-level factors (differences between companies) contribute substantially to the outcome. The ICC of 0.27 suggests that 27% of the variance is attributable to differences between companies. Given that the ICC exceeds the commonly accepted threshold of 0.10, the use of a multilevel approach is justified (Section 3.2.2.2).

The R² ratio is relatively high (0.40), meaning that a moderate share of the explained variance in cost comes from fixed effects. Given the total explained variance is also relatively good, these fixed effects can be considered as valuable determinants of the direct wage cost due to a workplace OA. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.236. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.237.

Determinants of the (log-transformed) cost of workplace OAs include the following.

Workforce segment. White-collar workers show significantly lower costs (\(\beta\) = -0.30). See Section 1.3.2.2 and Section 3.4.1.2.
Sector. Several sectors are significantly associated with cost differences. For example, transport (level H, \(\beta\) = 0.45) and construction (level F, \(\beta\) = 0.39) show higher costs, while education (level P, \(\beta\) = -0.39) and information & communication (level J, \(\beta\) = -0.58) are associated with lower costs. These findings are consistent with previously described sector-specific risk profiles. For example, in the transport sector, road traffic accidents are often classified as workplace (rather than commuting) OAs and such accidents on the road tend to result in more severe consequences. See Section 1.3.5 and Section 3.4.4.
Nature of the contract. Part-time employment is associated with lower costs (\(\beta\) = -0.08), which may reflect differences in exposure duration or job type . See Section 1.3.3.3 and Section 3.4.2.3.
Age. Compared to the reference group (<20 years), direct wage costs increase with age. The effect becomes statistically significant from the 40–49 age group (\(\beta\) = 0.22) and remains so in older groups (e.g., 60+: \(\beta\) = 0.29). This aligns with previous findings that older workers tend to experience more severe injuries when workplace OA occur (although they show lower likelihoods to experience such workplace OA). Our study thus confirms both patterns described in the literature. See Section 1.3.2.1 and Section 3.4.1.1.
Salary level. A positive but diminishing gradient is observed: for instance, workers earning €50–99/day show an estimate of \(\beta\) = 1.72 in comparison with the lowest category.Lower but still substantial effects can be noticed in higher brackets e.g. \(\beta\) = 1.32 for workers earning \(\geq\) €250. See Section 1.3.2.5 and Section 3.4.1.5.
Company size. Employees in medium-sized firms (brackets 50-99, \(\beta\) = 0.22, to 200–499 employees, \(\beta\) = 0.29) tend to generate slightly higher direct wage costs, possibly due to more formalized reporting procedures or higher average wages. See Section 1.3.4.2 and Section 3.4.3.2.
Work experience. Longer work experience was associated with higher direct costs, with an increasing trend from 1–4 years (\(\beta\) = 0.09) to \(\geq\) 21 years (\(\beta\) = 0.18). See Section 1.3.2.3 and Section 3.4.1.3.
Insurance companies. Some significant effects can be noticed, e.g. insurer 9 (\(\beta\) = 0.66), 11 (\(\beta\) = 0.34) and 12 (\(\beta\) = 0.28) are associated with higher direct wage costs. This is a new finding that needs confirmation in future studies. See Section 1.3.4.4 and Section 3.4.3.5.
Temporal effects. In comparison to 2014, a kind of upward trend can be observed. Direct wage costs increase notably from 2016 onwards, with the highest costs occurring in 2021 (\(\beta\) = 0.30). Seasonal variation is also apparent: with May (\(\beta\) = -0.08) and December (\(\beta\) = -0.10) showing lower cost estimates compared to January. See Section 1.3.6 and Section 3.4.5.

The multivariable model highlights that individual-level factors (age, work experience, workforce segment, salary level, nature of the contract), employer-level factors (sector, size, insurer), and temporal factors (year and month) are significant determinants of the direct wage cost resulting from a workplace OA.

This means that, although the specific accident-related parameters (not included in the model) might have a very significant and relevant effect on the cost, also group-level estimates can be made using the determinants from the current model to predict group-level cost profiles due to workplace OA.

Sector, size, workforce, age and wage are valuable determinants to estimate the direct wage costs due to workplace occupational accident

The multivariable model demonstrates good explanatory power for the Liantis registered direct wage costs following workplace occupational accidents. Among the determinants, company-level factors, such as sector and company size, and individual-level determinants, such as workforce, age and wage have the highest marginal to conditional R² proportions in their corresponding univariable models. These findings highlight the importance of company-level heterogeneity, which must be accounted for in cost estimations.

« previous

Figure 3.237 shows the different effect sizes of the categories included in the model. The figure immediately indicates large effects for e.g. wage and sector since their estimates (dots with whiskers) lie relatively further from the grand mean = 0 difference reference line compared to the other variables included in the model.

3.5.12 Severity degree (validated absence) of workplace OA (sevOAval, model 8)

This multilevel multivariable regression model investigates the factors associated with the severity degree of a company in a given year over the ten-year study period (sevOAval).

The model includes 82,989 annual observations (representing the severity degree for a given company in a given year), across 17,735 unique mutual ESPP/PS companies for which complete information on all determinants included in the model is available.

Many companies do not experience workplace accidents that have to be included in the calculation of the severity degree. As a result, both the median and 75ile of severity degree are zero. The overall mean severity degree across all companies, being strongly influenced by the larger values of companies with such countable OAs, is 1.1.

The model includes both fixed effects (company-level predictors) and random effects (group-level variation across companies in time). However, with a marginal R² of 0.010 and a conditional R² of 0.016, it explains only a very small fraction of the variance, leaving 98.4% unexplained. This indicates that group-level factors do not substantially contribute to the outcome. The ICC was 0.01, suggesting that only 1% of the variance is attributable to differences between companies and that most variance is within companies over time rather than between companies. Although an ICC smaller than 0.10 implies that a multilevel approach is not strictly necessary, its use facilitates comparison with other models (Section 3.2.2.2).

The R² ratio is relatively high (0.010 / 0.016 = 0.63), indicating that most of the explained variance comes from fixed effects. However, given the total explained variance is very low, these fixed effects cannot be considered strong determinants of the outcome. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.238. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.239.

Determinants of the severity degree (validated absence) include the following.

Sector. Compared to agriculture, forestry and fishing (A), significantly lower severity degrees are observed in level I (accommodation & food service activities, \(\beta\) = -1.15) and level S (other service activities, \(\beta\) = -0.99). No statistically significant differences are found between level A and the remaining sectors. See Section 1.3.5 and Section 3.4.4.
Insurance companies. Relative to the reference insurer, several insurers are associated with higher average severity degrees such as insurer 9 (\(\beta\) = +1.03) and insurer 3 (\(\beta\) = +0.86). In contrast, insurer 5 shows a lower average severity degree (\(\beta\) = -0.28). These findings suggest that differences between OA insurers correlate with the mean employer-year severity degrees. This this is a novel finding that warrants confirmation in future studies. See Section 1.3.4.4 and Section 3.4.3.5.
Workforce segment. A higher fraction of bluecollar workers (arbeiders) is associated with a higher severity degree (\(\beta\) = +3.76). See Section 1.3.3.1 and Section 3.4.2.1.
Salary level. A higher fraction of low-wage workers (day wage <50) is significantly associated with a higher severity degree (\(\beta\) = +2.59). In contrast, mid-range and high-wage salary fractions do not significantly seem to contribute. See Section 1.3.2.5 and Section 3.4.1.5.
Company size. Compared to the smallest companies (<5 workers), significantly lower severity degrees are observed in larger company ize categories: 5–9 employees (\(\beta\) = -0.51), 10–19 employees (\(\beta\) = -0.71), 20–49 employees (\(\beta\) = -0.80) and 50–99 employees (\(\beta\) = -0.79). See Section 1.3.4.2 and Section 3.4.3.2.
Age. Only fractions in ages groups 40–49 years (\(\beta\) = +0.80) and 50–59 years(\(\beta\) = +1.16) seem to be associated with higher severity degrees. See Section 1.3.2.1 and Section 3.4.1.1.
Temporal factors. Severity degrees are significantly lower in 2020 (\(\beta\) = -0.84), 2021 (\(\beta\) = -0.60), 2022 (\(\beta\) = -0.55) and 2023 (\(\beta\) = -0.90) relative to the reference year. This pattern is consistent with a downward shift during and following the COVID-19 pandemic. See Section 1.3.6 and Section 3.4.5.

In practical terms, the model captures only a very small part of the observed variation in severity degrees, even after accounting for company-specific baselines through the inclusion of random intercepts. Similar remarks apply as the ones noted for the frequency degree model. Therefore, we do not recommend using severity degrees as point-precise measures at the employer-year level.

The variation in company-specific severity degrees appears largely random, with over 98% of the variance remaining unexplained. Although sector, insurer, workforce segment, salary level, company size, age and time (years) show statistically significant associations with the severity degree, their explanatory power is modest relative to the overall variability, limiting their practical utility as predictive factors.

The multivariable model demonstrates low explanatory power for severity degrees. Occupational accident insurer, sector and size, and in employee fractions the workforce segment, salary level and parts of the age distribution have significant impacts on the model scores. However, the model is hardly useful in practice and confirms that caution is needed when using such a measure for benchmarking at company level.

Variation in company-specific severity degrees (validated absence) is highly random (>98%) and although factors such as insurer, sector, company size and other variables show statistical significance, their explanatory relevance remains limited

The multivariable model demonstrates only very limited explanatory power for severity degrees. Occupational accident company insurers, sector and size, and in employee fractions the workforce segment, salary level and age have significant impacts on the model scores. However, if we consider the large portion unexplained variance (98.4%), the model is hardly useful in practice and confirms that caution is needed in using this safety measure at the company level.

« previous

Figure 3.239 shows the different effect sizes of the categories included in the model. When compared with Figure 3.221, Figure 3.225 and Figure 3.227, the figure shows wide confidence intervals indicating the high variability of severity degrees.

3.5.13 Severity degree (calculated absence) of workplace OA (sevOAcal, model 8.1)

In this final multilevel multivariable regression model, we examine the factors associated with severity degree of a company in a given year over the ten-year study period (sevOAcal). Unlike the previous model, this analysis uses Liantis-calculated absence days and not the FEDRIS validated absence days.

The model includes 83,000 yearly observations (representing the severity degree for a given company in a given year), across 17,740 unique mutual ESPP/PS companies for which complete information on all determinants included in the model is available.

As in the previous model, many companies do not experience workplace OA that contribute to the calculation of the severity degree. Consequently, both the median and 75th percentile severity degree are zero. The overall mean severity degree across all companies remains approximately 0.9, largely driven by the higher values observed in companies with such countable OAs.

The model includes both fixed effects (company-level predictors) and random effects (group-level variation across companies in time). However, with a marginal R² of 0.007 and a conditional R² of 0.016, the model explains only a very small fraction of the variance, leaving 98.4% unexplained. The ICC is 0.01, indicating that only 1% of the variance is attributable to differences between companies, and most variance occurs within companies over time rather than between companies. As in the previous model, the use of a multilevel approach is not strictly required but facilitates comparison with other models (Section 3.2.2.2).

The ratio of marginal to conditional R² (0.007 / 0.016 = 0.44) suggests that most of the explained variance still comes from fixed effects, but since total explained variance is very low, these fixed effects cannot be considered strong determinants. Properties and random effects of the full multivariable model (right) in comparison with the corresponding null model (left) are presented in Figure 3.240. Fixed effects estimates of the multivariable full model in comparison the corresponding univariable models and existing literature are shown in Figure 3.241.

Determinants of the severity degree (calculated absence) include the following.

Sector. Compared to agriculture, forestry and fishing (level A), significantly lower severity degrees are observed in several sectors, including level I (accommodation & food service activities, \(\beta\) = -1.39), level S (other service activities, \(\beta\) = -1.35), and others such as the levels G, J, K, L, M, P, Q, R (all negative and significant). This pattern is more pronounced than in the previous model Section 3.5.12. See Section 1.3.5 and Section 3.4.4.
Insurance companies. Compared to the reference insurer, some insurers are associated with higher average severity degrees, such as insurer 9 (\(\beta\) = +1.21) and insurer 3 (\(\beta\) = +0.92), while others show either lower or non-significant effects. This is a novel finding that warrants confirmation in future studies. See Section 1.3.4.4 and Section 3.4.3.5.
Workforce segment. A higher fraction of blue-collar workers (arbeiders) is associated with higher severity degree (\(\beta\) = +3.71). See Section 1.3.3.1 and Section 3.4.2.1.
Salary level. A higher fraction of low daily wage earners (day wage <50) is associated with increased severity degree (\(\beta\) = +3.22). The effect is even larger than in the previous model. In contrast, midrange and high salary fractions do not significantly contribute. See Section 1.3.2.5 and Section 3.4.1.5.
Company size. Relative to the smallest companies (<5 workers), the severity degree is lower in company size classes with 5–9 (\(\beta\) = -0.58), 10–19 (\(\beta\) = -0.74), 20–49 (\(\beta\) = -0.82), and 50–99 employees (\(\beta\) = -0.85). See Section 1.3.4.2 and Section 3.4.3.2.
Age. Only the fraction of workers in age category 50–59 years shows a significant positive association (\(\beta\) = +0.79) with the severity degree. See Section 1.3.2.1 and Section 3.4.1.1.
Temporal factors. The severity degrees are significantly lower in the years 2018 (\(\beta\) = -0.58), 2020 (\(\beta\) = -0.90), 2021 (\(\beta\) = -0.74), 2022 (\(\beta\) = -0.65), and 2023 (\(\beta\) = -0.81) compared to the reference year, reflecting a downward trend during and following the COVID-19 period. See Section 1.3.6 and Section 3.4.5.

The variation in company-specific severity degrees appears largely random, with over 98% of the variance remaining unexplained. Although sector, insurer, workforce segment, salary level, company size, age and time (years) show statistically significant associations with the severity degree, their explanatory power is minimal relative to the overall variability, limiting their practical utility as predictive factors.

The multivariable model demonstrates low explanatory power for severity degrees. Although variables such as OA insurer, sector and company size, and employee-fractions of the workforce segment, salary level, and certain age group fractions have significant impact on the severity degree, the model is hardly useful in practice. These findings confirms that caution is needed when employing such measures for benchmarking at the company level.

Variation in company-specific severity degrees (calculated absence) is highly random (>98%) and although factors such as insurer, sector, company size and other variables show statistical significance, their explanatory relevance remains limited

« previous

Figure 3.241 shows the different effect sizes of the categories included in the model. When compared with Figure 3.221, Figure 3.225 and Figure 3.227, the figure shows wide confidence intervals indicating the high variability of severity degrees.

3.5.14 Summary overview of the discussed multivariable models

A summary overview of all multivariable models can be found in Table 3.146. In this table, the model quality of the of all discussed models is compared. For model 2 (Section 3.5.3), conditional R², R² ratio and ICC are presented between parentheses since they were estimated using data from the null model (see Figure 3.224).

Models 1 (Section 3.5.1), 1.1 (Section 3.5.2), 2 (Section 3.5.3), 3 (Section 3.5.4), and 5 (Section 3.5.7) are logistic regression models. A major distinction between these and many studies discussed in the literature review and earlier sections is the scope of the population analysed. In Models 1, 2, and 3 of the present study, we aimed to include all workers occupationally exposed to the risk of experiencing an OA, whereas many other studies focus solely on workers who have already experienced such accidents. This difference in population scope can distort our understanding of the relative importance of various determinants, as the distribution of characteristics in the general working population may differ substantially from that of the subpopulation affected by OAs.

Among Models 1, 2, and 3, Models 1 and 3 show relatively high marginal R² and R² ratios. This similarity is expected, given that approximately 85% of all OA are workplace-related. The high marginal R² values (in bold) suggest that these models explain a larger portion of the variance in the outcome variable, making them more informative and useful for understanding accident occurrence.

In contrast, the performance gap between Model 3 and Models 4 and 8 is substantial, particularly in terms of marginal R². Model 4, which examines the frequency degree of workplace OA at the company level per year, has a much lower marginal R², indicating limited explanatory power. The discrepancy is even more pronounced when considering conditional R² and the ICC: over 90% of the variance in frequency degrees remains unexplained by Model 4 and appears to lie within companies over time rather than between companies. This suggests that while workplace OA occurrence can be reasonably explained using individual- and company-level determinants (as in Model 3) on a monthly basis, this is not the case for company-level frequency degree patterns gathered at year basis.

Models 6 and 6.1 aim to explain variance in the duration of temporary absence following an occupational accident (OA), but differ in how the outcome variable is constructed. Model 6 (nDaysTAO) relies on absence data provided by the insurer, whereas Model 6.1 (lengtecor) calculates absence duration based on payroll records, which may offer a more accurate representation of actual absence. The marginal R² values (Model 6: 0.200; Model 6.1: 0.111) suggest that the fixed effects in Model 6 account for a larger portion of the variance in absence duration than those in Model 6.1. In contrast, the conditional R² values (Model 6: 0.330; Model 6.1: 0.381) indicate that Model 6.1 performs slightly better when random effects, such as company-level variation, are taken into account. This suggests that the payroll-based measure may be more sensitive to organizational context, potentially reflecting differences in how absence is recorded, managed, or influenced by internal company practices. The higher marginal R² in Model 6 likely reflects the standardized nature of insurer reporting, which aligns well with the fixed effects used. Meanwhile, the higher conditional R² in Model 6.1 implies that payroll-derived absence data may better capture contextual or structural influences, even if it is less predictable from fixed effects alone.

Model 7 aims to explain variance in the cost of occupational accidents (OA), using a transformed continuous outcome variable (\(\log_{10}(\text{costOA} + \text{€}1)\)). This transformation helps normalize the distribution of cost data. The model shows a marginal R² of 0.153, indicating that the fixed effects explain a moderate portion of the variance in OA-related costs. The conditional R² is higher at 0.378, suggesting that company-level random effects contribute meaningfully to the overall explanatory power. This pattern implies that while individual and accident-level predictors offer some insight into cost variation, a substantial part of the variance is structured at the organizational level. The ICC of 0.27 further supports this, pointing to clustering of cost outcomes within companies. These results highlight the importance of organizational context in shaping the financial impact of occupational accidents, and suggest that uncaptured company-specific factors (e.g. safety culture, wage policies,…), may play a significant role in cost outcomes.

Similar conclusions as for Model 4 apply to Model 8, which attempts to explain variance in yearly company severity degrees. These findings call for caution when using frequency or severity degree metrics to guide management decisions, shape company safety policies, or inform governmental and insurance frameworks for setting fees or fines. Like already clarified in the previous discussion section, overreliance on these measures may lead to misguided interventions if the underlying variability is not adequately captured.

Table 3.146: Summary overview model quality multivariable models

Parameter	mod1	mod1.1	mod2	mod3	mod4	mod5	mod6	mod6.1	mod7	mod8	mod8.1
outcome	hadOA	statusOA	comOA	wplOA	freqOA	seriousOA	nDaysTAO	lengtecor	costOA	sevOAval	sevOAcal
type	odds ratios	odds ratios	odds ratios	odds ratios	est.	odds ratios	incid. rate ratios	incid. rate ratios	est. (log10+1)	est.	est.
R² marg.	0.136	0.001	0.183	0.146	0.058	0.005	0.200	0.111	0.153	0.010	0.007
R² cond.	0.328	0.990	(0.951)	0.361	0.092	0.994	0.330	0.381	0.378	0.016	0.016
R² ratio	0.41	0.00	(0.19)	0.40	0.63	0.01	0.61	0.29	0.40	0.63	0.44
sigma2	3.29	3.29	3.29	3.29	2910.01	3.29	1.52	0.95	1.30	147.42	169.16
tau00 crbnr	0.19	0.00	0.00	0.20	111.38	4.14	0.30	0.41	0.47	0.88	1.51
tau00 insznr	0.76	329.80	52.45	0.91		552.81
ICC	0.22	0.99	(0.95)	0.25	0.04	0.99	0.16	0.30	0.27	0.01	0.01
N obs	7938537	31645	7938537	7938537	81780	17267	17287	9822	20423	82989	83000
N crbnr	16362	7218	16362	16362	17670	4154	4168	2946	4105	17735	17740
N insznr	291661	25200	291661	291661		14198
0	7906715	28848	7934355	7913871		16131
1	31822	2797	4182	24666		1136
rate	0.004	0.088	0.001	0.003		0.066

Insights from the multivariable models overview

Broader population scope for likelihood modelling is an added value. Models 1, 2, and 3 include all workers exposed to OA risk, unlike many studies that focus only on those already affected. These likelihood models yield a more realistic view of the full range of determinants.
Strong explanatory power for individual-level OA occurrence models. Models 1 and 3 show high marginal R² and R² ratios, indicating they effectively explain notifications (Model 1) and workplace-related accepted OA (Model 3). Their similar performance is unsurprising, given that 85% of accepted OA are workplace-related.
Limited insights from company-year level models call for caution regarding frequency and severity degrees. Model 4 (OA frequency degree per company per year) and Models 8/8.1 (OA severity degree per company per year) show low marginal and conditional R², suggesting most variance lies within companies over time, not between them. The poor explanatory power indicates these metrics may be unreliable for policy or management decisions.
Data source impacts model performance for absence duration modelling. Model 6 (FEDRIS-validated data) explains more variance via fixed effects, while Model 6.1 (payroll data) captures more company-level variation. Differences in model quality show how data origin affects explanatory power and sensitivity to organizational context.
Cost of OA is company-dependent. Model 7 shows moderate fixed-effect explanatory power but strong company-level clustering, implying that organizational factors heavily influence OA-related costs.
Low explanatory power models still yield important information. Models assessing whether an accepted workplace OA is classified as serious (Model 5) or whether a notified OA is refused (Model 1.1) show very high conditional R² but extremely low marginal R². This indicates strong clustering effects (e.g., company-level variation) but weak predictability from the fixed effects included in the models. Importantly, this does not imply flawed analysis. The study deliberately focuses on the role of determinants (antecedents), excluding OA-specific circumstances. These models still offer meaningful conclusions: the included determinants appear to play little to no role in influencing refusal or seriousness classification decisions. In other words, they are not sources of discrimination, a valuable finding in itself.
Random effects matter. Several models (e.g., 6/6.1, 7) show that company-level random effects significantly contribute to outcome variance, underscoring the importance of organizational context. Mixed-effects modelling is not just a tool for statisticians, it is a crucial approach for capturing real-world complexities in OA analyses.

« previous

3.6 Strengths and limitations of the study

A broad and large study. The present study is broad and large. We investigated a large number of OA (>200k), across multiple NACE-BEL-2008 sectors (>20), happening in many companies (48k – 69k), in a large population of exposed workers (>600k), in a substantial timespan (ten years). In this ways, the study is able to fill significant gaps (see Section 1.4).

Mixed model approach. The application of a mixed-effects model approach to predict the individual likelihoods of reporting an OA or having a commuting or workplace OA represents a significant methodological advancement over existing research, which predominantly relies on aggregated descriptive statistics of OA that already took place. By accounting for repeated observations within individuals and clustering within companies, this approach corrects for the non-independence of observations, thereby reducing bias in parameter estimates and standard errors. Furthermore, our models enable the decomposition of variance across multiple hierarchical levels, offering insights into whether reporting an OA is primarily driven by individual characteristics or organizational factors. This multilevel perspective not only enhances explanatory power but also provides actionable intelligence for targeted prevention strategies, a dimension largely absent from current national reporting systems.

Correction for a representative sample. An extra variable was included in all models to correct for the non-representative distribution of Liantis mutual ESPP and PS customers across Belgian provinces, sectors and company size. This correction enhances the generalizability of our findings to the broader Belgian working population, addressing a common limitation in OA research where samples may not accurately reflect the diversity of the workforce.

Including insurers as a determinant. To our knowledge, this is the first study in which the OA insurer is included as a determinant for a broad series of relevant outcomes. In this way, our study is highly relevant in filling a significant gap in the literature.

External benchmarking. By comparing our study sample with official statistics from NSSO and FEDRIS, we were able to benchmark our findings against established national data sources. This external validation strengthens the credibility of our results and places them within the broader context of OA research in Belgium. However, this approach also introduces a limitation. To align with the categorisations used in these official datasets, we had to simplify or recode certain variables, resulting in a loss of granularity. Variables such as age, wage, and company size, for instance, could have been treated as continuous variables as well, potentially leading to more nuanced and statistically powerful models.

No analysis at the accident mechanism level. This study does not include an analysis of specific accident mechanisms (e.g., falls from height, machinery-related incidents). Although such details are recorded once an OA has occurred, we deliberately chose to exclude them from our analyses. The primary reason is that this information is only available for individuals who have already experienced an OA. Including these variables would restrict the analysis to a subset of the workforce, potentially introducing selection bias and limiting the generalizability of our findings. Our focus was on identifying determinants of OA occurrence and outcomes across the entire exposed workforce, rather than examining the specific circumstances of accidents post-occurrence. While future research could explore accident mechanisms in greater detail, this study prioritizes a broader perspective on OA.

Waterfall loss. For likelihood modelling, information was also needed from employees not experiencing any OA. The availability of this necessary detailed information reduced the potential of 69,157 Liantis ESPP only employers to 47,820 Liantis ESPP and PS mutual customers. As a consequence, the original 293,938 ESPP only notifications were reduced to 90,619 notifications from 52,240 unique employees within mutual Liantis ESPP and PS customers. From this point on, further reduction to achieve complete cases for all determinants included in the models could be necessary (e.g. mod1.1 25,200 unique individuals left).

Underreporting bias. A recent systematic review investigated underreporting of OA (Kyung et al., 2023). While across the world several government based (regulatory compliance) and insurer/compensation based (financial benefits) reporting systems co-exist, underreporting risks are overall very high. Multiple studies across diverse industries and countries found rates ranging from 20% to over 90%, depending on the sector, injury severity, and reporting system. Belgium relies in first instance on a compensation based system. In such a system, underreporting occurs when workers fear retaliation, lack awareness, or perceive the claims process as burdensome. Some studies show that not all workplace injuries are reported to insurers, especially less severe cases, due to these barriers (Azaroff et al., 2002). Figure 2.1 reminds us of this the tip of the iceberg problem: incidents and light accidents are de facto not included in the study. For all other accidents, the choices of multiple actors during the notification process can influence whether the OA will actually be notified or not. It is crucial to keep this underreporting bias throughout the whole study in mind when interpreting the results. However, since the underreporting problem of OA is present in many published studies, we expect our results to be comparable and complementary to existing scientific literature.

Including insurers as a determinant. Insurers associated with companies that did not report OA may be underrepresented in the dataset, as insurer information is only reliably available for cases where the company was a client of Liantis Risk Solutions. In other cases, the insurer was inferred retrospectively via the FEDRIS Fonds voor Arbeidsongevallen (old Dutch name of FEDRIS before the fusion with FBZ, FAT in French) (FAO) number following the occurrence and notification of an OA. And although including the insurer as a determinant can be regarded as a strength of our study leading to new insights, it can also introduce bias and certainly reduced the number of due to the necessity of complete cases for modelling.

Excluding certain sectors. Although we aimed to include all NACE-BEL-2008 sectors, some sectors had to be excluded due to insufficient data availability. For example, the public sector is largely absent from the dataset since Liantis primarily serves private sector employers. Also mixed private/public sectors proved to be difficult to include. Our study thus does not provide insights in the level 1 sectors B, D, E, O, U and T.

Collapsed categories. Unlike the NSSO and FEDRIS statistics, which cover the entire Belgian population, this study is limited to the mutual clients of Liantis ESPP and PS. As with sector classifications, certain categories of key variables used in official statistics may be underrepresented or overrepresented in this dataset. When these variables are cross-tabulated with other important determinants, small or empty cells can occur, which may prevent statistical models from converging. A clear example is company size: while the univariable analysis (see Section 3.4.3.2) allowed for detailed categories such as 200–499, 500–999, and 1000+ employees, the multivariable models (see Section 3.5) failed to converge with this level of granularity. As a result, all categories with 200 or more employees had to be collapsed into a single group.

Wage category interpretation. One important limitation in this study concerns the interpretation of wage categories, particularly the lowest category (e.g., less than €50/day). This group differs significantly from the other wage categories and should be interpreted with caution. Due to the nature of the available raw payroll data, it was not possible for every record to verify to what extent the reported wage figures reflect actual gross salary levels. Wage calculations are inherently complex and influenced by a variety of contextual factors. For instance, low reported wages may not necessarily reflect low compensation, but rather periods of long-term absence, such as extended sick leave or disability. In such cases, the employee remains formally employed but is not actively working, and therefore not exposed to occupational risks. Additionally, during these absence periods, wage costs may no longer be borne directly by the employer, as compensation may be covered by social security mechanisms. Furthermore, a significant data quality issue must be acknowledged regarding the lowest wage category. A gross wage of less than €50 per day is not realistic in today’s context and likely results from anomalies in the dataset. These anomalies include records with extremely low or zero wages, which may be linked to specific employment statutes that could not be traced during the analysis. Examples include certain learning contracts or company directors/entrepreneurs with reported wages of €0. While such cases were present in the database, they could not be reliably identified or excluded. Therefore, this lowest category should not be considered a true representation of low-wage workers, but rather a heterogeneous group with mixed characteristics. Another layer of complexity arises from the temporal scope of the dataset, which spans multiple years. Wages from 2014 are not directly comparable to those from 2023, as indexation and inflation are not taken into account. This limits the interpretability of absolute wage values over time. Therefore, it is essential to focus on the relative positioning of wage categories within the model, rather than on their absolute numeric values or coefficients. Interpreting wage categories in relative terms provides a more robust and meaningful understanding of their role in the analysis.

Excluding OA specific variables without counterpart. A final limitation of this study is the exclusion of certain informative, accident-specific variables from the multivariable models. Factors such as shift work, long working hours, usual professional activity, specific task at the time of the accident, and type of workstation are known to significantly influence occupational accident outcomes. However, these variables could not be included due to the lack of structurally available counterparts for workers in the Liantis mutual ESPP and PS customer base who did not experience an OA. Without comparable data for the non-accident group, including these variables would have introduced bias and undermined the validity of the models.

List Of Acronyms

CBE: Crossroads Bank for Enterprises (KBO in Dutch)
COVID-19: Coronavirus Disease 2019
ESAW: European Statistics on Accidents at Work
ESPP: External Service for Prevention and Protection at work (EDPB in Dutch)
FAO: Fonds voor Arbeidsongevallen (old Dutch name of FEDRIS before the fusion with FBZ, FAT in French)
FEDRIS: Federaal agentschap voor beroepsrisico’s
GMQ: General Medical Questionnaire (AMV in Dutch)
ICC: Intraclass Correlation Coefficient
INSZ: Identificatienummer Sociale Zekerheid (rijksregisternummer of BIS-registernummer)
IRR: Incidence Rate Ratio
ISCO: International Standard Classification of Occupations
NACE-BEL: Belgian version of the the Europese activiteitennomenclatuur (NACE)
NIS: Nationaal Instituut voor de Statistiek
NSSO: National Social Security Office
OA: Occupational Accident
OR: Odds Ratio
PC: Paritair Comité (joint collective agreement or sectoral committee)
PPE: Personal Protective Equipment
PS: Payroll Services
RS: Risk Solutions
RSZ: Rijksdienst voor Sociale Zekerheid
SEATS: Signal Extraction in Arima Time Series
TRIR: Total Recordable Incident Rate
crbnr: enterprise identification number within CBE
insznr: personal identification number within NSSO (INSZ number)

Abatemarco, D. J., Delnevo, C. D., Rosen, M., Weidner, B. L., & Gotsch, A. R. (1995). Medical Surveillance Practices of Blue Collar and White Collar Hazardous Waste Workers. Journal of Occupational and Environmental Medicine, 37(5), 578–582. https://doi.org/10.1097/00043764-199505000-00005

Ahonen, E. Q. (2006). Risk of fatal and non-fatal occupational injury in foreign workers in Spain. Journal of Epidemiology & Community Health, 60(5), 424–426. https://doi.org/10.1136/jech.2005.044099

Ahonen, E. Q., & Benavides, F. G. (2016). Injury at work and migrant workers: a priority for a global agenda in occupational health. Occupational and Environmental Medicine, 74(4), 231–232. https://doi.org/10.1136/oemed-2016-103964

Alali, H., Abdel Wahab, M., Van Hecke, T., & Braeckman, L. (2016). Work accident victims: a comparison between non-standard and standard workers in Belgium. International Journal of Occupational and Environmental Health, 22(2), 99–106. https://doi.org/10.1080/10773525.2016.1168588

Alamgir, H., Yu, S., Chavoshi, N., & Ngan, K. (2008). Occupational injury among full-time, part-time and casual health care workers. Occupational Medicine, 58(5), 348–354. https://doi.org/10.1093/occmed/kqn026

Assuralia. (2025a). Steeds meer arbeidswegongevallen met zachte mobiliteit. https://press.assuralia.be/copy-steeds-meer-arbeidswegongevallen-met-zachte-mobiliteit

Assuralia. (2025b). Studie over de evolutie van de arbeidswegongevallen met zachte mobiliteit in de periode 2019-2024. https://cms.assuralia.be/sites/default/files/2025-09/Mobilite-douce-en-AT-Rapport-2024-NL.pdf

Azaroff, L. S., Levenstein, C., & Wegman, D. H. (2002). Occupational Injury and Illness Surveillance: Conceptual Filters Explain Underreporting. American Journal of Public Health, 92(9), 1421–1429. https://doi.org/10.2105/ajph.92.9.1421

Bena, A., Giraudo, M., Leombruni, R., & Costa, G. (2013). Job tenure and work injuries: a multivariate analysis of the relation with previous experience and differences by age. BMC Public Health, 13(1). https://doi.org/10.1186/1471-2458-13-869

Breslin, F. C. (2006). Trial by fire: a multivariate examination of the relation between job tenure and work injuries. Occupational and Environmental Medicine, 63(1), 27–32. https://doi.org/10.1136/oem.2005.021006

Brooks, M. E., Kristensen, K., van Benthem, K. J., Magnusson, A., Berg, C. W., Nielsen, A., Skaug, H. J., Maechler, M., & Bolker, B. M. (2017). glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. The R Journal, 9(2), 378–400. https://doi.org/10.32614/RJ-2017-066

Cabrera-Flores, J. (2023). The relationship between work accident rates and economic activity: Evidence from peru (2016-2021). Tec Empresarial, 17(3), 20–32. https://doi.org/10.18845/te.v17i3.6847

Craig, M. J., Liu, C., Zhang, F., & Enriquez, J. (2023). Sex-based differences in odds of motor vehicle crash injury outcomes. Accident; Analysis and Prevention, 195, 107100. https://doi.org/10.1016/j.aap.2023.107100

Delgado-Fernández, V., Rey-Merchán, M., López-Arquillos, A., & Choi, S. D. (2022). Occupational traffic accidents among teachers in spain. International Journal of Environmental Research and Public Health, 19. https://doi.org/10.3390/ijerph19095175

Fabiano, B., Currò, F., & Pastorino, R. (2004). A study of the relationship between occupational injuries and firm size and type in the Italian industry. Safety Science, 42(7), 587–600. https://doi.org/10.1016/j.ssci.2003.09.003

Fadzil, S. S., & Bulgiba, A. (2023). Commuting Accidents and Its Associated Factors Among Public Hospital Nurses in Selangor, Malaysia. Asia Pacific Journal of Public Health, 36(1), 119–122. https://doi.org/10.1177/10105395231207234

Federal Agency for Occupational Risks. (2023). Statistical Year Report 2023 - Occupational Accidents in the Private and Public Sector. https://www.fedris.be/sites/default/files/assets/NL/Jaarverslagen/BCAO-2024-12-13%20-%20bijlage%201%20-%20%20Jaarverslag%202023%20voor%20de%20%20private%20en%20publieke%20sector%20-%20NL.pdf

Federal Public Service Health, Food Chain Safety and Environment. (2022). Arbeidsongevallen Publieke Sector 2016-2020. https://www.health.belgium.be/nl/studie-arbeidsongevallen

Feijó, F., Menegale, C., Cardoso, F. M., Santos, L., & Lemos, M. (2018). Gender inequalities in work accidents and occupational musculoskeletal disorders: Findings from the 2013 brazilian national health survey. Occupational and Environmental Medicine, 75, A131–A132. https://doi.org/10.1136/oemed-2018-ICOHabstracts.374

Finnish Institute of Occupational Health. (2025). Commuting by bicycle enhances work ability and outweighs the increased risk of commuting accidents. https://www.ttl.fi/en/topical/press-release/study-commuting-by-bicycle-enhances-work-ability-and-outweighs-the-increased-risk-of-commuting

Fontaneda, I., Camino López, M. A., González Alcántara, O. J., & Ritzel, D. O. (2019). Gender differences in lost work days due to occupational accidents. Safety Science, 114, 23–29. https://doi.org/10.1016/j.ssci.2018.12.027

Gholamizadeh, K., Zarei, E., Yazdi, M., Rodrigues, M. A., shirmohammadi-Khorram, N., & Mohammadfam, I. (2023). An integration of intelligent approaches and economic criteria for predictive analytics of occupational accidents. Decision Analytics Journal, 9, 100357. https://doi.org/10.1016/j.dajour.2023.100357

Gómez, V., & Maravall, A. (1997). Programs TRAMO and SEATS: Instructions for the user. Mimeo, Banco de España.

Guerreschi, E. M., Francini, C., Giomarelli, A., & Nante, N. (2024). Injury risk in foreign workers: a proposal for a prevention plan in South Tuscany. European Journal of Public Health, 34(Supplement_3). https://doi.org/10.1093/eurpub/ckae144.2282

Hägg, S. A., Torén, K., & Lindberg, E. (2015). Role of sleep disturbances in occupational accidents among women. Scandinavian Journal of Work, Environment & Health, 41(4), 368–376. https://doi.org/10.5271/sjweh.3495

Hale, T., Angrist, N., Goldszmidt, R., Kira, B., Petherick, A., Phillips, T., Webster, S., Cameron-Blake, E., Hallas, L., Majumdar, S., & Tatlow, H. (2021). A global panel database of pandemic policies (Oxford COVID-19 Government Response Tracker). Nature Human Behaviour, 5(4), 529–538. https://doi.org/10.1038/s41562-021-01079-8

Hallowell, M., Quashne, M., Salas, R., Jones, M., MacLean, B., & Quinn, E. (2021). The statistical invalidity of TRIR as a measure of safety performance. Professional Safety, 66(4), 28–34. https://aeasseincludes.assp.org/professionalsafety/pastissues/066/04/F2Hallowelletal_0421.pdf

Hargreaves, S., Rustage, K., Nellums, L. B., McAlpine, A., Pocock, N., Devakumar, D., Aldridge, R. W., Abubakar, I., Kristensen, K. L., Himmels, J. W., Friedland, J. S., & Zimmerman, C. (2019). Occupational health outcomes among international migrant workers: a systematic review and meta-analysis. The Lancet Global Health, 7(7), e872–e882. https://doi.org/10.1016/s2214-109x(19)30204-9

Hendricks, K. J., Hendricks, S. A., & Marsh, S. (2024). Workplace injury and death: A national overview of changing trends by sex, united states 1998-2022. American Journal of Industrial Medicine. https://doi.org/10.1002/ajim.23687

Jeong, B. Y. (2021). Prevalence of occupational accidents and factors associated with deaths and disabilities in the shipbuilding industry: Comparisons of novice and skilled workers. Work, 69(3), 997–1005. https://doi.org/10.3233/wor-213530

Kaur, H., Wurzelbacher, S. J., Bushnell, P. T., Bertke, S., Meyers, A. R., Grosch, J. W., Naber, S. J., & Lampl, M. (2023). Occupational injuries among construction workers by age and related economic loss: Findings from ohio workers’ compensation, USA: 2007–2017. Safety and Health at Work, 14(4), 406–414. https://doi.org/https://doi.org/10.1016/j.shaw.2023.10.003

Kines, P., & Mikkelsen, K. L. (2003). Effects of Firm Size on Risks and Reporting of Elevation Fall Injury in Construction Trades. Journal of Occupational and Environmental Medicine, 45(10), 1074–1078. https://doi.org/10.1097/01.jom.0000085887.16564.3a

Koranyi, I., Jonsson, J., Rönnblad, T., Stockfelt, L., & Bodin, T. (2018). Precarious employment and occupational accidents and injuries a systematic review. Scandinavian Journal of Work, Environment & Health, 44(4), 341–350. https://doi.org/10.5271/sjweh.3720

Kreshpaj, B., Bodin, T., Wegman, D. H., Matilla-Santander, N., Burstrom, B., Kjellberg, K., Davis, L., Hemmingsson, T., Jonsson, J., Håkansta, C., & Orellana, C. (2021). Under-reporting of non-fatal occupational injuries among precarious and non-precarious workers in Sweden. Occupational and Environmental Medicine, 79(1), 3–9. https://doi.org/10.1136/oemed-2021-107856

Kyung, M., Lee, S.-J., Dancu, C., & Hong, O. (2023). Underreporting of workers’ injuries or illnesses and contributing factors: a systematic review. BMC Public Health, 23(1). https://doi.org/10.1186/s12889-023-15487-0

Lalive, Rafael and Ruf, Oliver and Zweimuller, Josef. (2006). Wages and Risks at the Workplace: Evidence from Linked Firm-Worker Data. https://conference.iza.org/conference_files/TAM2006/253.pdf

López, M. A. C., Alcántara, Ó. J. G., & Fontaneda, I. (2017). Gender differences in commuting injuries in spain and their impact on injury prevention. BioMed Research International, 2017. https://doi.org/10.1155/2017/3834827

Margolis, K. A. (2010). Underground coal mining injury: A look at how age and experience relate to days lost from work following an injury. Safety Science, 48(4), 417–421. https://doi.org/https://doi.org/10.1016/j.ssci.2009.12.015

McVittie, D., Banikin, H., & Brocklebank, W. (1997). The effects of firm size on injury frequency in construction. Safety Science, 27(1), 19–23. https://doi.org/10.1016/s0925-7535(97)00048-9

Monteiro Ferreira, M. J., Silveira Correa, F. G., Lacerda, E. M., Hajat, S., & de Araújo, L. F. (2020). Analysis of Risk Factors in Occupational Accidents in Brazil. Journal of Occupational & Environmental Medicine, 62(2), e46–e51. https://doi.org/10.1097/jom.0000000000001793

Morassaei, S., Breslin, F. C., Shen, M., & Smith, P. M. (2012). Examining job tenure and lost-time claim rates in Ontario, Canada, over a 10-year period, 19992008. Occupational and Environmental Medicine, 70(3), 171–178. https://doi.org/10.1136/oemed-2012-100743

Mudenha, W. F., Naicker, N., & Singh, T. (2022). Impact of the macro-environment on the reporting of occupational injuries and illnesses by low-income workers compared to middle-income workers in South Africa: a mixed-methods study protocol. BMJ Open, 12(8), e063384. https://doi.org/10.1136/bmjopen-2022-063384

Nakagawa, S., Johnson, P. C. D., & Schielzeth, H. (2017). The coefficient of determinationR²and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. Journal of The Royal Society Interface, 14(134), 20170213. https://doi.org/10.1098/rsif.2017.0213

Nenonen, N. (2011). Occupational accidents in the finnish local government sector: Utilisation of national statistics. International Journal of Injury Control and Safety Promotion, 18, 321–329. https://doi.org/10.1080/17457300.2011.589005

Neves, T. V., & Fonseca, B. M. C. (2023). Are men more likely to have serious occupational accidents? Findings of reported cases in palmas, northern brazil. Revista Brasileira de Medicina Do Trabalho, 21. https://doi.org/10.47626/1679-4435-2023-909

O’Hara-Wild, M., Hyndman, R., & Wang, E. (2024). Feasts: Feature extraction and statistics for time series. https://CRAN.R-project.org/package=feasts

Picchio, M., & Ours, J. C. van. (2016). Temporary Jobs and the Severity of Workplace Accidents. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2817291

Picchio, M., & Ours, J. C. van. (2017). Temporary jobs and the severity of workplace accidents. Journal of Safety Research, 61, 41–51. https://doi.org/10.1016/j.jsr.2017.02.004

Ponsin, A., Fort, E., Hours, M., Charbotel, B., & Denis, M.-A. (2020). Commuting Accidents among Non-Physician Staff of a Large University Hospital Center from 2012 to 2016: A Case-Control Study. International Journal of Environmental Research and Public Health, 17(9), 2982. https://doi.org/10.3390/ijerph17092982

Rappin, C. L., Wuellner, S. E., & Bonauto, D. K. (2016). Employer reasons for failing to report eligible workers’ compensation claims in the BLS survey of occupational injuries and illnesses. American Journal of Industrial Medicine, 59(5), 343–356. https://doi.org/10.1002/ajim.22582

Reiman, T., & Pietikäinen, E. (2012). Leading indicators of system safety Monitoring and driving the organizational safety potential. Safety Science, 50(10), 1993–2000. https://doi.org/10.1016/j.ssci.2011.07.015

Salerno, S., & Giliberti, C. (2019). Non-vehicle commuting in italy: Need for ergonomic action for women’s lower limbs? Applied Ergonomics, 83, 102982. https://doi.org/10.1016/j.apergo.2019.102982

Salerno, S., & Giliberti, C. (2021). Commuting accidents while walking: A women occupational risk. Epidemiologia e Prevenzione, 45 5, 368–377. https://doi.org/10.19191/EP21.5.P368.103

Salminen, S. (2000). Traffic accidents during work and work commuting. International Journal of Industrial Ergonomics, 26, 75–85. https://doi.org/10.1016/S0169-8141(00)00003-2

Salminen, S. (2004). Have young workers more injuries than older ones? An international literature review. Journal of Safety Research, 35(5), 513–521. https://doi.org/https://doi.org/10.1016/j.jsr.2004.08.005

Salminen, S. (2011). Are Immigrants at Increased Risk of Occupational Injury? A Literature Review. The Ergonomics Open Journal, 4(1), 125–130. https://doi.org/10.2174/1875934301104010125

Takahashi, A., & Miura, T. (2016). At what age is the occupational accident risk high? Analysis of the occurrence rate of occupational accidents by age. In P. Arezes (Ed.), Advances in safety management and human factors (pp. 3–10). Springer International Publishing. https://doi.org/10.1007/978-3-319-41929-9_1

VIAS Institute. (2018). Vrouwen achter het stuur: minder boetes, minder ernstige ongevallen. https://www.vias.be/nl/newsroom/femmes-au-volant-moins-daccidents-moins-de-p-v/

Wuellner, S. E., & Bonauto, D. K. (2014). Exploring the relationship between employer recordkeeping and underreporting in the BLS Survey of Occupational Injuries and Illnesses. American Journal of Industrial Medicine, 57(10), 1133–1143. https://doi.org/10.1002/ajim.22350

Yedla, A., Kakhki, F. D., & Jannesari, A. (2020). Predictive Modeling for Occupational Safety Outcomes and Days Away from Work Analysis in Mining Operations. International Journal of Environmental Research and Public Health, 17(19), 7054. https://doi.org/10.3390/ijerph17197054

Yong Jeong, B. (1999). Comparisons of variables between fatal and nonfatal accidents in manufacturing industry. International Journal of Industrial Ergonomics, 23(5), 565–572. https://doi.org/https://doi.org/10.1016/S0169-8141(98)00020-1

3.1 Representativeness of the employer data

3.2 Statistical methodology

3.2.1 Descriptives

3.2.2 Modelling

3.2.2.1 Datasets

3.2.2.2 Occurence

3.2.2.3 Severity

3.3 Outcomes

3.4 Descriptive and univariable analysis of potential determinants

3.4.1 Individual factors

3.4.1.1 Age

3.4.1.1.1 Descriptives

3.4.1.1.2 Models

3.4.1.2 Biological sex

3.4.1.2.1 Descriptives

3.4.1.2.2 Models

3.4.1.3 Work experience (seniority with the employer)

3.4.1.3.1 Descriptives

3.4.1.3.2 Models

3.4.1.4 Nationality (foreign workers)

3.4.1.4.1 Descriptives

3.4.1.4.2 Models

3.4.1.5 Salary level

3.4.1.5.1 Descriptives

3.4.1.5.2 Models

3.4.1.6 Residential location (province)

3.4.1.6.1 Descriptives

3.4.1.6.2 Models

3.4.1.7 Commuting distance

3.4.1.7.1 Descriptives

3.4.1.7.2 Models

3.4.1.8 Use of Personal Protective Equipment (PPE)

3.4.1.9 Health problems: quality of hearing

3.4.1.9.1 Models

3.4.1.10 Sleeping disorders

3.4.1.11 Substance abuse

3.4.1.12 Personality

3.4.1.13 Number of years of schooling

3.4.2 Job related factors

3.4.2.1 Occupation and workforce segment

3.4.2.1.1 Descriptives

3.4.2.1.2 Models

3.4.2.2 Contract type (permanent/indefinite/unlimited versus temporary/definite/limited) (precarious employment)

3.4.2.2.1 Descriptives

3.4.2.2.2 Models

3.4.2.3 Nature of the contract (full-time versus part-time employment)

3.4.2.3.1 Descriptives

3.4.2.3.2 Models

3.4.2.4 Working in shifts and long working hours

3.4.2.5 Hazards in work system

3.4.2.5.1 Usual professional work activity

3.4.2.5.2 Specific activity

3.4.2.5.3 Type of workstation

3.4.2.5.4 Risks (ESPP)

3.4.3 Organisation related factors

3.4.3.1 Work location

3.4.3.1.1 Descriptives

3.4.3.1.2 Models

3.4.3.2 Company size

3.4.3.2.1 Descriptives

3.4.3.2.2 Models

3.4.3.3 Precarious employment

3.4.3.3.1 Subcontracting

3.4.3.4 Psychosocial risk factors

3.4.3.4.1 Work stress

3.4.3.4.2 Job satisfaction

3.4.3.4.3 Work relationships

3.4.3.4.4 Safety climate and safety culture

3.4.3.5 Insurance companies

3.4.3.5.1 Descriptives

3.4.3.5.2 Models

3.4.3.6 Investments in prevention

3.4.4 Sectoral differences

3.4.4.1 Section NACE-BEL 2008

3.4.4.1.1 Descriptives

3.4.4.1.2 Models

3.4.4.2 Joint labour committees

3.4.4.2.1 Descriptives

3.4.4.2.2 Models

3.4.5 Temporal patterns