Job insecurity is one of the main disadvantages of temporary employment, but individuals in temporary jobs may in addition be exposed to higher levels of occupational health and safety (OHS) risks for a variety of reasons,1 2 especially as they may have shorter job tenures than their permanently employed counterparts due to frequent job changes.3 Starting a new job has been identified as a potential risk factor due to lack of knowledge of the physical environment, safety procedures and safety resources.4 5 Risks may be further heightened for temporary workers because organisations are often less committed to OHS training for such individuals.2 Temporary workers may also be given tasks that are less desirable and possibly more dangerous than those assigned to their permanently employed counterparts. Additionally, workers filling temporary positions may have little autonomy or legal recourse to protect themselves from unjust dismissal. Consequently, it may be difficult for them to refuse unsafe work or demand appropriate training.2 6

A number of studies have investigated the relationship between temporary employment and occupational health outcomes.7^–10 The evidence suggests that temporary workers have a higher risk of occupational injury.

Fewer studies have investigated the relationship between temporary employment and sickness absence.9 These studies have generally found that individuals in temporary employment have a lower rate of sickness absence relative to individuals in permanent employment,11 although some evidence suggests that absence varies depending on the nature of the temporary contract and the worker’s expectations.12 13 For example, in Belgium, temporary agency workers have a higher sickness absence rate than permanent workers, whereas fixed-term contract workers have a lower rate.12 In a study of the Netherlands, there was no difference in sickness absence between temporary and permanent workers, although temporary workers seeking full-time employment were less likely to be absent and had shorter absences than permanent workers.13

A few studies have focussed specifically on absence attributable to work-related exposures.10 14 15 Benavides et al investigated the relationship between various forms of temporary/fixed-term employment and absence due to “health problems caused by the main job” in a cross-sectional sample of 15 European Union countries.16 The study found that absence was generally higher for those in permanent employment, with the exception of fixed-term full-time employment. Kaminski investigated days away from work due to work injury in a sample of small to mid-sized manufacturing plants in mid-west United States.14 The study found no association between the proportion of temporary employees at a plant and the absence rate. Gimeno et al also investigated work-related sickness absence (WSA) rates for a cross-sectional sample of 15 European Union countries.15 The study found somewhat lower rates of absence amongst temporary workers compared to permanent workers.

Although the findings of a higher probability of injury/illness and lower probability of absence may appear paradoxical, the criteria for taking an absence can vary from individual to individual. Kristensen attributes this apparent paradox to different thresholds for taking sick leave.17 Temporary employees may have both positive and negative incentives that increase the injury/illness threshold required for an absence, such as the lack of wage-replacement benefits for absences, the desire to have a contract renewed or to secure a permanent job, the fear of dismissal, and avoiding negative impressions with coworkers and managers.9 A recent study by Virtanen et al finds that permanent employment protects some workers from termination despite high rates of work absence, suggesting that temporary employees should be wary of absence taking.18

What is missing from many of the above studies is a longitudinal framework and consideration of variables that change over time. In particular, job tenure changes over time and is important to consider since evidence suggests that individuals with shorter tenure have a higher level of work injury/illness claims than individuals with longer tenure.4 Tenure is associated with firm/job specific knowledge that can bear on risk exposure.5 19 It is also associated with seniority, which can bear on the ability to take an absence without negative reprisal from an employer. For example, in Canada individuals are often on probation for the first 3–6 months of a new job and can be dismissed without cause during this probationary period.20 21 Since individuals in temporary employment are less likely to have long tenure than their permanently-employed counterparts, not controlling for this variable may result in attributing outcomes associated with tenure to temporary employment. Prior health status should also be considered, since there may be some health selection effects into temporary employment (ie, less healthy individuals may be more likely to be in temporary employment). Most previous research in this area has not considered health selection effects.

In this study we investigate the role of temporary employment on WSA of 1 week or longer due to work exposures, with tenure and other contextual factors as confounders. Although the literature suggests that temporary workers may have a higher threshold for absence taking, we reason that this threshold is less likely to influence WSA associated with serious injury or illness (assuming that an absence of 1 week or longer is due to a serious health condition). In particular, work-related injuries and illnesses are covered by workers’ compensation systems, regardless of the type of job contract (ie, temporary or permanent), and hence it is more likely that a worker will file a claim and take an absence if the injury or illness is serious. In an effort to tease out the potential “threshold effects” associated with absence taking, in this study we focus specifically on work-related illnesses/injuries that result in prolonged work absence.

Social protection can result in differences in exposure to work-related hazards,22 so we include two measures to proxy for it, namely union membership and firm size.23 24 Union membership can confer a certain amount of protection from undue health risks, such as entitlement to OHS training and representation in cases where an employer fails to fulfil OHS obligations.25 Union membership might also be associated with higher short-term absence for minor injuries due to the protection provided by a union, but we assume that this is less likely to be a factor with serious injuries and illnesses. Firm size is also relevant because larger firms are thought to be safer as they are more likely to have standardised and well-developed human resource practices, formalised OHS and other training programs, and systematised/ongoing monitoring of safety performance.26^–28

As a consequence of factors related to exposure (ie, inadequate training and/or more dangerous task assignments), our working hypotheses for the present study are as follows: (1) after adjusting for job tenure and other confounders, individuals in temporary employment will have a higher risk of serious work-related injury/illness and a higher probability of WSA compared to permanent employees; (2) short tenure workers will have a higher rate of WSA than workers with longer tenure; (3) unionised workers will have a lower rate of WSA compared to non-union workers; and (4) workers in large firms will have a lower probability of WSA than workers in smaller firms.

METHODS

Data source and sample

The data source was the Canadian Survey of Labour and Income Dynamics (SLID), a nationally representative longitudinal survey of individuals aged 16 and older administered by Statistics Canada. The SLID consists of 6-year overlapping panels, with the first panel beginning in 1993, the second in 1996 and the third in 1999. At the end of each calendar year, individuals in a particular panel are questioned about their labour-market experiences in each month over the last year. This is repeated at the end of each year for that panel for a total of 6 years. Use of the SLID for this study was approved by the Research Ethics Board at the University of Toronto.

Our sample consisted of individuals aged 25–54 who were not full time students and began a job during the panel period. The sample was drawn from the third 6-year panel of the SLID, which spans the period 1999–2004. The time period used for the analysis was 2000–2003; the first year, 1999 was used to construct lagged explanatory variables and 2004 data were not available for analysis. The third panel of the survey was selected because it was the first to ask about the nature of the employment contract (ie, permanent versus temporary) of each job. The panel comprises approximately 15 000 households, with one individual selected from each household for more in-depth interviews. For these individuals, detailed information is collected on the characteristics of up to six jobs in a particular year, sociodemographic information, income sources and amounts, and all sickness absences of 1 week or more.

For our sample we selected individuals who began a new job within the time period of the panel and after 1999, were between the ages of 25 and 54 at the start of the job, were not full-time students and were not self-employed. The final sample consisted of 4777 individuals after removal of observations with missing data.

Outcome variable

The event of interest in this study is the first occurrence of a WSA after the start of a particular job. This is taken from a series of questions in the SLID that read as follows:

Not counting fully paid vacations, were you absent from the job for a period of one week or longer?” [If yes], what was the main reason for this absence [12 response categories provided with the first option being “own illness or disability”]? [If own illness or disability], was this due to a work related illness or injury [yes/no]?

This measure of WSA in the SLID has been compared to other measures of WSA employed by Statistics Canada and lost-time claim rates from a major insurer.29 The study found comparable trends between WSA rates from the SLID, WSA in another national longitudinal survey, and workers’ compensation rates for lost-time claims of 1 week or more. The findings from this study provide evidence of concurrent validity for the WSA measure.

Analysis was based on the person-job episode since job characteristics, including WSA, were identified on a monthly basis. An individual in a particular job was followed from the start of the job until the job ended or the individual exited to a WSA. If an individual changed jobs, a new person-job episode was started. For multiple concurrent jobs, a person-job episode was identified with each job. Person-job episodes in which an individual left the job without exiting to a WSA were right censored (ie, observations on the person-job episode ended without exit to a WSA). Person-job episodes that did not have complete responses for all covariates were deleted. Person-job episodes that started prior to the individual entering the panel were not included in the analysis due to difficulties in determining time on the job and whether the person had a WSA prior to the observation period.

Each person-job episode was broken into monthly observations, since monthly data on each job were recorded in the survey. Hence, the unit of analysis was the person-job-month. For person-job episodes included in the sample, all months of observation after the first absence were excluded in order to streamline the modelling and avoid bias. If there was a second absence within a job, it could be related to the first absence (ie, it might be a reoccurrence or worsening of an existing condition) and therefore not be a distinct and separate occurrence. A similar modelling approach has been used in a study of youth and WSA.30

Explanatory variables

The key explanatory variable was a variable indicating whether the job was temporary or permanent. The SLID question that this variable is developed from reads as follows:

Is your job permanent, or is there some way that it is not permanent (eg, seasonal, temporary, term, casual, etc)?

A set of variables were used to identify tenure or number of months from start of employment (1–3 months or 4–6 months compared to more than 6 months) based on monthly information provided on each job. The specific clusters used to define tenure were chosen based on the fact that jobs generally have a 3–6 months’ probationary period in Canada, during which one can be dismissed without cause.20 21 Interaction terms of the temporary job variable with each duration category were tested to determine if individuals in temporary jobs had higher probabilities of absence in each category. Other tests for model specificity included: a model with only the temporary variable and an intercept; the full model without control for prior health; the model without controls for tenure run with all the data and then with person-job-month data for only 1–3 months and 1–6 months (ie, all person-job episodes were censored at 3 months for one model and at 6 months for another). Results from these tests are available upon request.

Indicators for union membership and firm size category of 19 people or less were included as proxies for social protection. The cut-point of 20 was chosen because this is the minimum firm size for which a joint health and safety committee is required in Canada.31

Other explanatory variables used to control for risk exposure were job type (manual, mixed or non-manual), the number of hours worked in the month, an indicator of multiple concurrent job holding, age (in years), sex and educational attainment (less than high school, high school or post-secondary education). The job type variables were developed from the Standard Occupation Classification Codes using the Institut de recherche Robert Sauvé en santé et en sécurité du travail occupation coding system.32 An indicator of self-reported health status that was fair or poor (compared to good, very good or excellent) prior to commencement of the job was included to control for health selection effects.

Statistical model

We used the complementary log-log (clog-log) function for continuous time processes33 to model the time from the start of a job to exit to the first occurrence of WSA.34 35 This approach treats the within-period hazard as constant (exponential) with an across-period baseline that takes the form of a step function with no pre-defined shape.36 Like the Cox regression model, the clog-log transformation invokes a proportionality assumption in hazards not odds; hence, it provides a discrete time analog for the continuous time hazard model. The clog-log is best for analysing interval censored data in which events unfold in continuous time, although information about event occurrence is limited to discrete-time intervals like the monthly job information available in SLID.37 38 This statistical approach is also better at handling situations where there are a lot of ties than the standard Cox model. Three periods were defined based on months on the job (1–3 months, 4–6 months and more than 6 months). Covariates were integrated into the clog-log functional form using the proportional hazard approach.

Some within-job time varying covariates varied on a monthly basis, others on a yearly basis. The former includes job tenure, the multiple job holding identifier, hours worked and age. The latter includes contract type (the temporary job identifier), union membership, firm size, the job type identifier, education and prior health status. The only within-job constant covariate was sex.

Due to the self-reported health status question being asked only once a year (at year end), job episodes starting in 1999 (the first year of the panel) could not be used. Additionally, the last year of the panel (2004) was not available for analysis. Since the SLID is administered through a stratified cluster sampling, we used scaled weights for observations derived from population weights provided by Statistics Canada. The weights take into account sampling probabilities and non-response.

Models were run in SAS v 9.1 using PROC GENMOD. Correlation of errors for within person-job units over time was controlled for by specifying an autoregressive error structure. This autoregressive structure corrects for correlation across months within a person-job episode but not across different job episodes held by a person.

RESULTS

Descriptive statistics

Table 1 provides detailed descriptive information for the sample. The final sample consisted of 4777 individuals who experienced 7953 distinct job episodes and 167 WSA. The episodes contributed 115 488 person-job-month observational units. Overall, 2.1% of the job episodes ended in a WSA. Approximately 34% of all job episodes were temporary jobs. Temporary jobs were less likely to be of longer duration, with just 23% of job episodes longer than 6 months classified as temporary. Of all the person-job-months from temporary jobs, 49% were for durations of greater than 6 months.

Duration modelling results

Table 2 presents the hazard ratio for each variable, which is a comparison of the hazard of an individual with the characteristic of interest compared to an individual without that characteristic, all else being equal.34

The hazard ratio for the temporary job variable was not different from one based on the 95% confidence interval (CI) (0.29 to 1.43). The variable indicating tenure of 1–3 months was also not different from one (95% CI 0.39 to 1.43). In contrast, the variable indicating tenure of 4–6 months was significantly different from one (95% CI 0.18 to 0.71). Individuals with 4–6 months of tenure were 64% less likely to exit to a WSA than individuals with tenure of more than 6 months. We also tested interaction terms between the temporary job variable and the duration dummies (models not shown), but these also had 95% CI that straddled one. Models run without controls for health and without controls for job tenure had similar findings. Testing with data censored at 3 months and at 6 months also results in an insignificant coefficient for the temporary variable.

Union membership had a 95% CI of 1.00 to 2.49, indicating a 58% higher probability of WSA than for individuals without union membership. In contrast, individuals in small firms were equally as likely to have a WSA as individuals in large firms.

Two other job characteristics were associated with health risk exposure, namely physical demands and concurrent job holding, whereas the sociodemographic variables were not. Individuals in manual jobs were 151% more likely to have a WSA compared to those in non-manual jobs (95% CI 1.48 to 4.22). The indicator of multiple concurrent job holding had a 95% CI of 0.09 to 0.59, suggesting that individuals holding multiple jobs were 77% less likely to exit to a WSA.

Individuals reporting fair or poor health prior to the job episode were substantially more likely to have a WSA (95% CI 1.76 to 6.07). These individuals had a 227% higher probability of WSA than individuals in good, very good or excellent health.

DISCUSSION

The main finding of our study is that the association between temporary employment and WSA of 1 week or longer was not significant after controlling for tenure, job characteristics, sociodemographics and prior health status. One of the categories of tenure was significant at the conventional level of 5% or less, but its sign was in the opposite direction to that hypothesised. Union membership, a variable used to proxy for social protection, was significant, but its sign was also opposite to that hypothesised.

Impact of temporary employment and tenure on sickness absence

In contrast with previous studies15 16 we found that individuals in temporary jobs were as likely to have a WSA as their permanently employed counterparts. Our study was longitudinal, considered absences over several years and controlled for prior health status, so this might explain why our results differed from these studies. Furthermore, we only considered absences of 1 week or longer, whereas the previous studies considered total absences in a year including those of less than 1 week.

Our model controlled for job tenure based on the premise that health risk exposures and workers’ compensation claims are higher when one begins a job, as suggested in previous studies.3^–5 19 Short tenure also connotes less job security39 and hence the possibility of a higher threshold for absence taking. The two forces work in opposite directions, but we thought that the latter would be less relevant for an absence of 1 week or longer. Our findings provide modest support that the disincentive for absence taking due to short tenure dominates, although we cannot completely rule out the possibility that the results are driven by lower risk exposure for tenure of 4–6 months. This finding is consistent with studies that show that workers with longer tenure are more likely to have a sickness absence.39 40

The role of union and firm size

We hypothesised that unions may be vigilant about health and safety issues, and hence, that unionised workers would be less likely to have a WSA. But unions also provide the security for workers to be able to take absence leave without fear of reprisal from an employer, although we thought this would be less of an issue with serious work injuries and illnesses. The two forces work in opposite directions, and our results suggest the latter dominated. Other studies have reported similar findings.41 42

Also based on the same social protection argument, we postulated that employees of larger firms are less likely to have a WSA due to more standardised and well-developed human resource practices, more formalised OHS training programs, and ongoing monitoring of safety performance. Other studies have found that absenteeism for small employers is relatively low and sickness absence increases with company size.15 16 This latter finding could be explained by larger firms being more likely to facilitate other employees covering the workload for an absent worker, making it less onerous to negotiate time off work, particularly for short absences. We did not find this to be the case for longer absences, although the hypothesis of social protection based on firm size was also not supported in our study.

Impact of job characteristics and sociodemographics

As might be expected, individuals in manual jobs were more likely to have a WSA. By contrast, holding multiple concurrent jobs was associated with reduced probabilities of WSA. This finding is notable given that this variable was included in the model to control for increased exposure due to increased work time across multiple jobs. It may be that individuals with multiple jobs have a higher absence taking threshold, perhaps due to increased work engagement or more resilience to adverse health exposures.

None of the sociodemographic characteristics (specifically age, sex and educational attainment) were significant, a finding aligned with other research that focuses on absence related to work exposures rather than sickness absence in general. Factors such as age, sex and educational attainment are more likely to be associated with general health and sickness absence unrelated to work exposures.41 43^–45 In our study, we controlled for prior health status, hence mitigating the need to proxy for general health with sociodemographic characteristics.

Strength and limitations of the study

The key strengths of this study are the longitudinal nature of analysis and the control for job tenure, health selection effects, within-person-job correlation of observations, and a number of time-varying individual and job characteristics. This improves on previous studies, many of which have relied on cross-sectional data or less detailed longitudinal data. Most previous studies have not controlled for tenure or considered time-varying characteristics or health selection. Furthermore, our use of self-reported, population-based survey data has advantages over workers’ compensation claims data because temporary workers may be less inclined to make a claim even if an injury or illness is compensable,1 6 and some industries in some jurisdictions in Canada are not covered by workers’ compensation.

There are a few limitations to our study. First, it relied on a self-report survey administered at year end; consequently, recall bias within a reporting period might occur. A respondent may remember a WSA but not the exact date on which it occurred. However, recall bias is likely to be minimal due to the focus on absences of 1 week or greater. Absences of this length would involve reasonably serious health issues that would be difficult to forget. The use of absences of 1 week or greater has other shortcomings. First, such absences are less frequent and therefore there are fewer of these absences, there being only 167 in a population of 4777 individuals. The number of absences is even small when subsamples are tested, such as that with only the first 3 months or the first 6 months. Second, the variable differs from that used in previous studies which focus on all absences regardless of duration.

Another issue is that respondents may have differed in their interpretation of what constitutes sickness attributable to work exposures. Some may have focused only on physical health conditions, whereas others may have defined it more broadly to include mental health issues such as stress, anxiety and depression. Furthermore, the aetiology of long latency conditions may be difficult for respondents to associate with particular exposures.

A third issue is that the variable indicating temporary employment captured a broad range of jobs. Some contract work may be quite insecure (eg, seasonal, temporary agency, short-term contract work), whereas other contract work may be almost as secure as permanent jobs (eg, multi-year contract work). This issue is addressed by De Cuyper et al10 who emphasise the heterogeneity of temporary employment.

Some assumptions made in the analysis may also introduce bias. For individuals with multiple concurrent jobs we attributed the absence to the main job, whereas WSA might have arisen from an exposure in a concurrent secondary job. To test the robustness of our assumption, we ran models in which an absence was assigned to secondary jobs for individuals with multiple concurrent jobs. We found that the results were similar to the primary model in terms of magnitudes, signs and significance levels of coefficients (results from these models are available upon request).

Conclusions and future research

Our findings suggest that temporary employment is not associated with an increased probability of WSA of 1 week or longer after controlling for tenure and several other characteristics. We cannot completely rule out the possibility that both exposure and incentives for temporary workers are important but act in opposite directions and net out to zero. Other factors such as job tenure and union membership bear on either exposures and/or the ability to take time off work to recover from even a serious work injury or illness, although the latter appears to dominate. Other factors are also critical, such as working in a manual job, having multiple concurrent jobs, and prior health status.

Our major contribution to this literature is the modelling of the longitudinal aspect of employment experiences, and the inclusion of job tenure, which previous studies suggest is an important aspect of adverse health exposures at work and associated sickness absence.3^–5 19 Prior health status and indicators for multiple concurrent job holding have not been used in studies of temporary employment and yet appear to be important variables.

Continued investigation of the contributions of differential exposure and absence taking behaviour between temporary and permanent employment in a longitudinal framework is important in future research on the topic. Furthermore, the impact of different types of temporary work arrangements (eg, seasonal, temporary agency, long-term contract work) on absence taking behaviour warrants exploration, as do more direct measures of social protection.