How important is the type of working contract for job satisfaction of agency workers?

Abstract

Previous research finds that agency workers are less satisfied with their job than regular workers. This paper analyzes whether this difference can be explained by the duration of the working contract agency workers are employed on. The analysis leads to three results. First, agency workers’ contract type does not explain their lower job satisfaction. Second, agency workers on permanent contracts are significantly less satisfied with their job than regular workers on the same contract. Third, agency workers on fixed-term contracts do not differ in job satisfaction from regular workers on both fixed-term and permanent contracts. The difference in job satisfaction between permanently employed agency and regular workers can partly be explained by changes in the reference point. Overall, the results, however, lend support to the conclusion that agency workers on fixed-term contracts regard their employment as stepping stone while those on permanent contracts appear to be trapped in this type of employment.

1 Introduction

Between 1997 and 2017, the number of agency workers in Germany has increased rapidly from 181,000 to 1,043,000. Consequently, agency workers represented around 3% of the total German labor force in 2017 (BA 2018).¹ Typically, employees are directly employed with their employer on either a fixed-term or a permanent working contract.² Agency employment, in contrast, is a triangular employment relationship where workers are legally employed by their employment agency but “with a view to making them available to a third party, who may be a natural or legal person […] which assigns their tasks and supervises the execution of these tasks” (ILO 1997, No. 181, Article 1.1.b). Agency workers function as a useful supplement to regular employees as they allow firms to better balance business cycle fluctuations and to reduce costs (Matusik and Hill 1998). The increased use of agency workers is, however, critically observed by researchers and politicians, because agency contracts are typically considered to be less favorable than regular permanent contracts (De Cuyper et al. 2008).

On the one hand, researchers have indeed found that agency workers face poorer working conditions than regular employees, such as lower remuneration (Böheim and Cardoso 2009; Booth et al. 2002; Hamersma et al. 2014; Jahn 2010) and fewer training opportunities provided by the employer (Arulampalam and Booth 1998; Nienhüser and Matiaske 2006). Furthermore, workers on agency contracts are found to have a higher risk of social exclusion (D’Addio and Rosholm 2005) and of becoming unemployed (Antoni and Jahn 2009; Autor and Houseman 2010), which previous studies have identified to substantially reduce welfare (Clark and Oswald 1994; Kassenboehmer and Haisken-DeNew 2009; Winkelmann and Winkelmann 1998). More generally, Busk et al. (2017) show that relaxing regulations of the agency sector negatively affect agency workers’ well-being.

On the other hand, agency contracts may function as stepping stones to permanent employment (Addison and Surfield 2006; Green and Leeves 2004; Hopp et al. 2016) and provide a way for workers to better balance work and family life (Morris and Vekker 2001).³ Hence, there are several arguments for as well as against agency contracts, which makes it a priori difficult to judge whether such contracts are in sum less favorable than regular permanent contracts. Yet finding an answer to this issue is of interest to policymakers in all countries that have eased the use of agency workers in recent years or plan to do so in an attempt to make labor markets more flexible.

To assess whether agency contracts are less favorable than permanent working contracts, researchers compare the utility identical workers—who only differ in their working contract—derive from their job. Traditionally, the utility derived from a job has been approximated by the earned wage. Yet, over the past two decades, researchers have increasingly started to use workers’ reported job satisfaction as an alternative and more direct measure of the utility derived from a job (see, for example, Clark 2001; Hamermesh 2001). Using job satisfaction has the advantage that unobservable or unmeasurable job characteristics, such as the relationship with colleagues, are captured in addition to objective job characteristics such as wages.

In a literature review on the relationship between different work arrangements and job satisfaction, De Cuyper et al. (2008) find that workers who are employed on flexible contracts, such as fixed-term contracts, agency contracts, and seasonal contracts, as a group report lower levels of job satisfaction, on average, than workers who are employed on a permanent contract. However, in a meta-analysis, Wilkin (2013) stresses that workers on flexible contracts do not form a homogeneous group and that it is necessary to distinguish between those different types. Empirical evidence on the relationship between job satisfaction and fixed-term contracts is mixed. Chadi and Hetschko (2016), for example, provide evidence that relative to workers on a permanent contract, workers on a fixed-term contract are significantly less satisfied with their job. In contrast, Green and Heywood (2011) and D’Addio et al. (2007) find no such difference in reported job satisfaction.

Empirical research on the relationship between agency employment and job satisfaction is more clear cut: in general, agency workers are found to be significantly less satisfied with their job than regular workers on a permanent working contract.⁴ Using data from the first wave of the Household, Income and Labour Dynamics in Australia (HILDA) Survey, Wooden and Warren (2004) show that male agency (casual) workers report significantly lower levels of job satisfaction than regular male workers who hold a permanent contract. Applying panel estimation techniques to subsequent waves of HILDA, Buddelmeyer et al. (2015) find that the difference remains when controlling for unobserved heterogeneity. The negative relationship between agency contract and job satisfaction has further been established by De Graaf-Zijl (2012) and Jahn (2015) in the Netherlands and Germany, respectively.

A potential shortcoming of these studies is that they treat agency workers as a homogeneous group with regard to whether they are employed on a fixed-term or a permanent working contract. Not differentiating agency workers by their working contract may, however, result in biased estimates if agency workers on fixed-term contracts systematically differ from those on permanent contracts. For example, suppose that while agency workers on permanent contracts are as satisfied with their job as regular workers on permanent contracts, agency workers on fixed-term contracts are significantly less satisfied. In such a case, the difference in job satisfaction between agency workers as a group and permanently employed regular workers would be the result of the lower job satisfaction of agency workers on fixed-term contracts. Hence, we would mistakenly argue that agency employment is a bad policy instrument even though only employing agency workers on fixed-term contracts is undesirable but not employing them on permanent contracts.

The first contribution to the literature of this paper, therefore, is to account for this potential source of bias by considering the type of working contract in a regression of job satisfaction on agency status. A second contribution is that by comparing agency workers on a permanent contract with regular workers on the same contract, the present paper provides a more thorough analysis of the difference in job satisfaction that is solely attributed to the status of being an agency worker.

In this paper, I use data from the German Socio-Economic Panel, which allows me to control for a large set of personal and job characteristics as well as for individual fixed effects when regressing job satisfaction on agency status. Yet critically, there may still be unobserved factors that I cannot account for with the data at hand and, therefore, the correlations derived may not be interpreted as causal.

I find that, first, even though controlling for the contract type agency workers are employed on, they are on average still significantly less satisfied with their job than regular workers on the same contract. Second, the difference in job satisfaction is entirely driven by agency workers on permanent contracts who are significantly less satisfied with their job than regular workers on permanent contracts. In contrast, agency workers on fixed-term contracts do not differ in reported job satisfaction from regular workers on both fixed-term and permanent contracts. Hence, a permanent contract per se does not automatically imply high levels of job satisfaction.

The difference in job satisfaction between agency and regular workers on permanent contracts can at least partly be explained by changes in the reference point of agency workers who previously were regular employees, as the difference in job satisfaction becomes just insignificant when workers’ perception of just wages is controlled for. Since the difference in job satisfaction between agency and regular workers on fixed-term contracts is not affected by this modification, my results lend support to the conclusion that agency workers on fixed-term contracts regard their employment as a stepping stone into regular employment or as a mean to avoid unemployment, while agency workers on permanent contracts appear to be somehow trapped in this type of employment. In the course of this paper, I perform several robustness checks to validate my results. Yet they indicate that the estimates derived are not sensitive to the modifications made.

The remainder of this paper is organized as follows. Information on the institutional background of agency employment is provided in Sect. 2. Section 3 describes the dataset and provides summary statistics. Section 4 outlines the empirical strategy. Results, explanations and robustness checks are presented and discussed in Sect. 5. Section 6 concludes.

2 Institutional background

In Germany, the use of agency workers is settled in the so-called Labor Placement Act (Arbeitnehmerüberlassungsgesetz), which aims at providing social protection for agency workers, especially preventing them from exploitation. In addition, agency workers benefit from the same standard labor laws that also apply to regular workers, implying that they are entitled to health insurance, pension benefits, paid vacation, unemployment benefits, and, most importantly, employment protection (after a trail period of 6 months).

The Labor Placement Act was introduced in 1972 but has been modified on several occasions since then.⁵ The last big reform took place in 2003 alongside the so-called “Hartz-reforms”, aimed at making the German labor market more flexible. The key changes of the reform in 2003 were the elimination of the so-called reemployment ban, the synchronization ban, and the maximum period of assignment. Until 2003, the reemployment ban permitted that agencies hire workers only for a specific job, fire them, and rehire them again. The synchronization ban permitted that agencies hire workers only for the period in which they would be lent out to the user firm. The maximum period of assignment of an agency worker to a user firm was limited to 24 months prior to 2003. After this assignment period, the agency was required to replace the worker with another one. The rationale behind this provision was to prevent user firms to substitute agency workers for regular employees.⁶

Besides the elimination of these regulations, the principle of “equal pay and equal treatment” was introduced in 2003. Equal treatment requires that from the first day of assignment agency workers should be subject to the same working conditions than regular employees at the user firm. Equal pay, on the other hand, requires that they earn the same wage like regular workers who perform the same tasks. While equal working conditions are fixed by law, equal pay could be circumvented by agencies if they sign a collective agreement. Almost all agencies took advantage of this possibility since approximately 97% of all agencies had signed such an agreement by the end of 2003 (Jahn 2010), implying that agencies workers, on average, still earn less than regular workers at the user firm for doing the same job. In an evaluation of this reform, Busk et al. (2017) provide evidence that equal treatment of agency workers does not outweigh the cost of abolishing reemployment ban, synchronization ban, maximum period of assignment, and equal pay, as they find that agency workers’ job satisfaction decreased significantly following the reform in 2003.

Taken together, the changes in 2003, on the one hand, improved working conditions of agency workers. On the other hand, agency workers still earn less than regular workers and the reform notably increased flexibility in the agency sector. Hence, we would expect that permanently employed agency workers are more satisfied than temporarily employed agency workers since they have more stable working relationships, which are typically associated with less uncertainty.

3 Data and summary statistics

The data used in the present study are drawn from the German Socio-Economic Panel (SOEP), a representative household panel that interviews approximately 20,000 individuals on a yearly basis.⁷ The SOEP contains detailed information on respondents’ personal and job characteristics. The question of whether a respondent is an agency worker has been asked in 2001 for the first time. However, to ensure the same legal conditions over the whole investigation period, I restrict the analysis to waves 2003–2016. In addition, I remove the self-employed and those currently in (occupational) education or retraining. Furthermore, I follow Busk et al. (2017) and only keep male workers belonging to the typical German working-age population (20–65 years) for the same reasons as mentioned there. First, previous research has shown that men and women on flexible employment contracts have different job satisfaction regimes (Clark 1997; Green and Heywood 2011; Jahn 2015; Souza-Poza and Sousa-Poza 2003), which is likely due to women’s stronger preferences for flexible work arrangements and for balancing family and working life.⁸ Second, the German agency work sector is highly segmented by gender. While female agency workers are mainly employed in service and clerical occupations, male agency workers dominate in blue-collar occupations (Jahn 2010). Third, the vast majority of agency workers in Germany are male (about 70% according to the Federal Employment Agency (BA 2018)). Last, any analysis on flexible employment contracts is likely confronted with selection issues regarding female labor force participation. Hence, the subsequent analysis is based on male workers only.⁹ The various reduction steps result in an unbalanced panel of 52,520 observations from 12,338 respondents.

Following previous research, individuals’ reported job satisfaction is used as a proxy for individual utility (for example, Clark 2001; Green and Heywood 2011; Jahn 2015). Job satisfaction is assessed by the question “How satisfied are you with your job (if employed)?” with responses ranging from 0 (“completely dissatisfied”) to 10 (“completely satisfied”). Agency workers are identified by the question “Is this work through a temporary employment agency?” Independent of whether they are employed at a temporary employment agency, in the next question respondents are asked “Do you have a fixed-term or permanent employment contract?”

Job satisfaction depends on workers’ personal and job characteristics that may also drive selection into agency work. I capture this issue by including control variables for observable personal and job characteristics. Using a fixed effects model allows me to additionally account for time-invariant characteristics. Personal characteristics include years of unemployment experience, unemployment in the previous year (binary), education (ISCED-1997, six categories), age (four categories), marital status (four categories), children below 14 in the household (binary), and subjective health status (three categories).¹⁰

With regard to observable job characteristics, I control for net labor income in the last month (in logs and without extra payments), extra wage in the last year (binary; equals one in case of, for example, holiday or Christmas pay), actual working hours per week, work hour mismatch (binary; equals one if desired work hours do not correspond to actual hours), hours of overtime in the last month, degree of work autonomy (higher values imply greater autonomy), job position (binary; equals one if in manager or supervisor position), and years of tenure¹¹ (linear and squared). Moreover, I account for job changes (binary; equals one if respondents have a new job since the last interview) because recent work by Chadi and Hetschko (2016) shows that the tenure–job satisfaction relationship is of non-linear nature.¹²

Table 1 provides summary statistics distinguished by contract type. Regular workers on permanent contracts are on average less satisfied with their job than regular workers on fixed-term contracts. In general, agency workers are less satisfied with their job than regular workers. Within the group of agency workers, those on fixed-term contracts (6.873) report higher levels of job satisfaction than those on permanent contracts (6.244).

Table 1 Summary statistics by contract type.

Around 43% of agency workers on fixed-term contracts are between 20 and 29 years old, compared to 20% of agency workers on permanent contracts. This could explain why agency workers on fixed-term contracts are less likely to be married and more likely to report being in better health than agency workers on permanent contracts.

More than half of the fixed-term regular and agency workers report having a new job compared to around 10% of regular workers on permanent contracts and 37% of permanently employed agency workers. Regular permanent employees report the longest tenure (13 years) followed by agency workers on a permanent contract (6 years). Regular fixed-term employees report on average 3 years of tenure and agency workers on a fixed-term contract on average 2 years.

With regard to remuneration, Table 1 shows that regular workers on fixed-term contracts and agency workers earn significantly less than regular workers on a permanent contract. Approximately, every second agency worker on a permanent or fixed-term contract and every second regular worker on a fixed-term contract are entitled to bonus payments. In contrast, three out of four regular workers on permanent contracts received an extra wage in the previous year. Regular workers on a permanent contract have the lowest amount of experienced years of unemployment, followed by regular workers on a fixed-term contract. Concerning agency workers, those with a permanent contract spent fewer years in unemployment than those with a fixed-term contract. A similar pattern is found with respect to whether workers were unemployed in the previous period: Almost 20% of agency workers on a fixed-term contract were unemployed in the year before getting a fixed-term agency contract, compared to 13% of regular fixed-term workers and 13% of permanently employed agency workers.

4 Empirical methodology

The point of departure for the empirical analysis is the following regression specification typically used in the literature on the relationship between type of working contract and job satisfaction (for example, Green and Heywood 2011; De Graaf-Zijl 2012; Jahn 2015):

$$jsat_{it} = \alpha + \beta_{1} F_{it} + \beta_{2} A_{it} + \gamma 'X_{it} + \delta 'Z_{it} + \lambda_{t} + \zeta_{i} + \varepsilon_{it} ,$$

(1)

where $jsat_{it}$ corresponds to reported job satisfaction for worker i in year t. $F_{it}$ is a binary variable which equals one if worker i is employed on a fixed-term contract in year t and zero otherwise. Similarly, $A_{it}$ is a binary measure which equals one if worker i is an agency worker in year t and zero otherwise. $X_{it}$ and $Z_{it}$ are vectors of observed personal and job characteristics, including the variables discussed in the previous section. $\lambda_{t}$ and $\zeta_{i}$ are time and individual fixed effects, respectively. $\varepsilon_{it}$ represents the individual stochastic error term with mean zero and constant variance.

Contrary to previous studies, $F_{it}$ and $A_{it}$ are defined as not being mutually exclusive in Eq. (1), implying that the estimated agency coefficient, $\beta_{2}$, reflects the average difference in job satisfaction between agency workers and regular workers with the same contract and other identical characteristics.¹³ If the difference in job satisfaction between agency workers and regular workers can be explained by the duration of the working contract they are employed on, we would expect that the difference vanishes once the contract type is controlled for: $\beta_{2} = 0$ (ceteris paribus).

The descriptive statistics in Table 1 highlight that agency workers on a permanent contract report substantially lower levels of job satisfaction than regular workers on the same contract, while the difference between agency and regular workers on a fixed-term contract appears to be a little bit smaller. To examine whether these differences remain in a regression analysis, an interaction term between contract type, $F_{it}$, and agency status, $A_{it}$, is added to Eq. (1). As a result, the difference in job satisfaction between agency and regular workers on permanent contracts is simply reflected by the agency main effect. Whether agency workers and regular workers on a fixed-term contract differ in reported job satisfaction can be tested from the sum of agency main effect and interaction effect. The interaction coefficient itself measures whether a fixed-term contract has a different impact on job satisfaction for regular than for agency workers.

5 Results

5.1 Empirical results

Table 2 displays the estimation results of Eq. (1). The dependent variable in each column is workers’ reported job satisfaction. Workers’ personal characteristics as well as federal state, year, and 1-digit industry classification dummies (constructed by the SOEP group based on the NACE classification) are included in each regression. Workers’ job characteristics are additionally controlled for in the even columns of Table 2 to check whether agency workers may be compensated for undesirable characteristics associated with such a contract by so-called equalizing differences in terms of hedonic labor market theory (for example, Rosen 1974), such as (extra) wages or a higher degree of work autonomy.¹⁴

Table 2 Job satisfaction effects of different contract types.

Column (1) presents estimation results when Eq. (1) is estimated using pooled OLS.¹⁵ On average, agency workers are 0.519 points less satisfied with their job than regular workers on the same contract, holding everything else constant. The difference is statistically significant at the 1% significance level. Regarding fixed-term contracts, the coefficient is positive but statistically indistinguishable from zero at least at the 10% level. If job characteristics capture positive benefits of agency contracts, the agency coefficient is expected to turn more negative when these observable factors are controlled for. Yet column (2) shows that the opposite is the case: the agency coefficient becomes more positive, implying that differences in job characteristics partially explain why agency workers are less satisfied with their job than regular workers on the same contract.¹⁶

Column (3) displays estimation results when identification is based on workers’ within variation across years by means of the fixed effects model.¹⁷ The estimates suggest that a notable part of the job satisfaction penalty of agency workers can be explained by unobserved heterogeneity. The penalty drops to 0.243 points but remains significant at the 1% level. Similar to the pooled OLS estimates, the agency coefficient becomes more positive when job characteristics are controlled for.

Finally, column (5) of Table 2 shows that agency workers on a permanent contract are on average 0.305 points less satisfied with their job than regular workers on a permanent contract and other comparable characteristics. This difference is even larger than the combined average difference in the previous two columns, implying that the gap in job satisfaction between agency and regular workers on a fixed-term contract needs to be smaller. Indeed, the difference in job satisfaction between agency and regular workers on fixed-term contracts is − 0.152 points as calculated by the coefficient sum of agency indicator and interaction term; the difference is indistinguishable from zero at conventional levels. In addition, the null hypothesis that the coefficient sum of fixed-term, agency measure and interaction term is different from zero cannot be rejected using the F statistic, implying that agency workers on a fixed-term contract do not differ in reported job satisfaction from regular workers on permanent contracts. Lastly, regular workers on fixed-term contracts are significantly more satisfied with their job than agency workers on permanent contracts.¹⁸ The positive coefficient of the interaction term indicates that a fixed-term contract might have a more positive impact on job satisfaction for agency workers than for regular workers. The coefficient is, however, insignificant at conventional levels. Column (6) again indicates that the coefficients do not alter when job characteristics are controlled for. Yet, since a larger fraction of the (within) variation in each specification can be explained when differences in job characteristics are accounted for, they are from now on included in each of the subsequent regressions.

Table 2 shows that the difference in job satisfaction between agency and regular workers on permanent contracts is highly statistically significant. To illustrate the economic significance of the effects, I calculate the relative increase in income that is needed to compensate the worker for the lower level of job satisfaction resulting from being an agency worker.¹⁹ The estimated coefficient of log net labor income in the last month is 0.668, indicating that a 1% increase in income increases job satisfaction on average by 0.0067 points. Compared to the fact that agency workers on permanent contracts are on average 0.302 points less satisfied with their job than regular workers on the same contract, this suggests that monthly net wages of permanently employed agency workers need to be raised by around 45% to make up for lost job satisfaction resulting from being an agency worker.²⁰

5.2 Explanations

These results certainly come as a surprise. Typically, permanent contracts are associated with higher (formal) job security than fixed-term contracts and, therefore, we would expect that agency workers on permanent contracts report higher levels of job satisfaction than agency workers on fixed-term contracts, everything else equal.

One explanation for this result might be that agency workers on fixed-term contracts regard their agency employment as a stepping stone into regular employment or as a mean to avoid unemployment. This hypothesis is supported by the finding that agency workers on fixed-term contracts are younger and more likely to be unemployed in the period before they start working for an employment agency (see Table 1). Moreover, since a high share of older agency workers is employed on permanent contracts, their lower job satisfaction may be traced back to the fact that they are somehow trapped in agency employment; and this persistent situation cannot be alleviated by being employed on a permanent contract either. This point is supported by the number of permanent agency workers in the sample who dominate fixed-term agency workers and by the high average tenure of permanently employed agency workers (see Table 1).

Another explanation for the lower job satisfaction of agency workers on permanent contracts might be that being employed for a very long time with the same agency and user firm changes the reference point for one’s own job satisfaction. While at the beginning agency workers still value not being unemployed, as time goes by, they may start comparing themselves with regular employees of the user firm who probably enjoy more benefits. However, the principle of equal treatment requires that agency workers are subject to the same working conditions than regular employees at the user firm.²¹ Additionally, the principle of equal pay requires that agency workers earn the same as regular user firm workers who do the same job. Yet as discussed above, 97% of agencies circumvented equal pay by signing a collective agreement. Hence, a likely candidate that might shift the reference point of agency workers is wages.

In addition to controlling for wages and extra payments, I consider this argument by including a binary indicator in Eq. (1) that equals one if workers evaluate their current wage as just and zero otherwise. Doing this also allows me to at least partly consider the argument that fixed-term agency workers simply have more attractive work environments compared to permanently employed agency workers. The question whether workers perceive their current income just has been asked only in waves 2005, 2007, 2009, 2011, 2013, and 2015. The following analysis is, therefore, restricted to these six waves, implying that the number of observations used in the analysis drops from 52,520 to 22,340 and the number of individuals from 12,338 to 9742.

Column (1) of Table 3 shows the fixed effect results when Eq. (1) is estimated for these six waves. The results are similar to those reported in Table 2, indicating that the restriction to these six waves has no impact on the results derived so far. Next, column (2) presents estimates when the binary indicator for just wages is included in Eq. (1). First of all, workers who evaluate their current wage just are on average 0.323 points more satisfied with their job than workers who do not. The agency coefficient, however, only slightly decreases when just wages are controlled for, indicating that they can only partly account for the difference in job satisfaction between agency workers and regular workers on the same contract. Similar to Table 2, column (3) shows that agency workers on permanent contracts are still significantly less satisfied with their job than regular workers on the same contract. However, regular workers on fixed-term contracts are no longer more satisfied with their job than agency workers on permanent contracts, as the null hypothesis that the sum of fixed-term and agency main effect is different from zero can no longer be rejected at conventional levels. Like in the first two columns, controlling for just wages slightly decreases the agency coefficient but it eventually turns just insignificant (p value 0.103). This finding thus indicates that a change in the reference point might explain why agency workers on permanent contracts are less satisfied with their job than regular workers on such contracts.

Table 3 Fixed effect estimation of job satisfaction effects controlling for workers’ perception of just wages.

Critically, agency workers on a permanent contract might be a very exceptional group of people that hardly compare to other workers included in the analysis. For example, some change in (working) life may have caused lowering job satisfaction and becoming an agency worker employed on a permanent contract. Such a likely change could be the practice of employers to dismiss costly regular workers and rehire them as agency workers. This practice could also explain the high average tenure of agency workers on permanent contracts (see Table 1), as they could indicate years of employment with the user firm; that is, agency workers may not be willing to consider themselves as outsiders in their current job after having worked several years for the same employer as a regular employee. This practice is at least partly restricted by §§ 9 and 10 of the Labor Placement Act, yet only for a period of 6 months since the firing, implying that after 6 months it is indeed possible for employers to rehire fired regular employees as agency workers.

I seek to account for this argument in two ways. First, I follow Schäfer (2012) and exclude those workers from the sample who did not report a job change, but were not in agency work in the previous period; this applies to 398 observations. Second, I use propensity score matching to control for a potential compositional bias of agency and regular workers, which, however, comes at the expense of not being able to control for time-invariant unobserved heterogeneity.

Table 4 shows that the previous results do not alter when the 398 observations who did not report a job change but were not in agency work in the previous period are excluded. If we assume that those 398 observations are indeed those who had been rehired as agency workers, then this finding indicates that the above results are not biased by workers who are not willing to consider themselves as outsiders in their current firm after having worked several years as a regular employee for the same employer.

Table 4 Job satisfaction effects of different contract types after Schäfer’s (2012) sample restriction.

Using propensity score matching with 20 neighbors and replacements on a probit model to the full sample does also not lead to differing results, as can be seen in Table 5.²² Restricting the matching procedure to the subsample of permanent and fixed-term workers [columns (2) and (3)] neither changes the results derived so far. Additionally, column (4) shows that in the subsample of agency workers, those on fixed-term contracts still report significantly higher levels of job satisfaction than those on permanent working contracts.

Table 5 Job satisfaction effects of different contract types using propensity score matching.

In a similar spirit, I test whether agency workers may differ from other individuals in the likelihood of leaving the panel, because such a behavior may also interact with both contract type and job satisfaction. To test for attrition, I employ a fixed effects logit model regressing the likelihood of leaving the panel in the next wave on agency status, type of working contract and all other covariates.²³ The coefficient of having a fixed-term contract is negative and significant at the 1% level (− 0.210, p value 0.007). The coefficient of the agency indicator is − 0.194 with a corresponding p value of 0.107. If an interaction term between agency status and fixed-term contract is included, the fixed-term coefficient remains negative and significant (− 0.174, p value 0.013). The agency coefficient becomes smaller and decreases even further in significance (− 0.018, p value 0.894), indicating that agency workers on permanent contracts and regular workers on permanent contracts do not differ in the likelihood of leaving the panel.

Besides these empirical checks and the descriptive evidence that relative to agency workers on fixed-term contracts those on permanent contracts are on average older and less likely to be in unemployment in the year before they become an agency worker (see Table 1), I did not find any serious evidence that individuals who become agency workers on a permanent contract form an exceptional group of people. Another argument against this claim might be that agencies have an incentive to employ only those agency workers on a permanent contract who have a high likelihood of being lent out to a user firm, because it is the agency that has to pay the agency worker’s wage even in case when the agency worker is currently not lend out.

5.3 Robustness checks

Besides testing explanations for the findings derived, I perform several robustness checks to validate my results. A first robustness check picks up recent findings that workers’ perceived job security rather than the formal security provided by the contract type matters for job satisfaction (Origo and Pagani 2009; Jahn 2015).²⁴ But column (1) of Table 6 displays that the results are not sensitive to controlling for workers’ perceived job insecurity.

Table 6 Different robustness checks.

The next check tests whether economic conditions or shocks have an influence on the use of agency work and reported job satisfaction, since the observational period 2003–2016 includes the years of the financial crisis (2007–2009), which had an impact on agency employment as the number of agency workers decreased by around 18% from 760,500 to 625,667 between 2008 and 2009 (BA 2018).²⁵ To check this argument, I estimate Eq. (1) excluding the years 2007–2009, but column (2) shows that the results prove robust to this restriction.²⁶

In a last check, I test whether the results alter when ordinality rather than cardinality of the dependent variable is assumed by estimating Eq. (1) using the “Blow-up and Cluster” fixed effects ordered logit estimator introduced in Baetschmann et al. (2015). Yet, column (3) provides evidence that the results are also not sensitive to the use of a different estimator.

6 Conclusion

In this paper, I analyze the relationship between agency status and job satisfaction on basis of the German Socio-Economic Panel. Contrary to previous research, I do not treat agency workers as a homogeneous group with regard to their type of working contract but rather consider their contract type in a regression of job satisfaction on agency status. I find that agency workers are on average significantly less satisfied with their job than regular workers on the same contract. In a more detailed analysis, I provide evidence that this difference is entirely driven by agency workers on a permanent contract, who are significantly less satisfied with their job than regular workers on the same contract. On the other hand, agency workers on a fixed-term contract do not differ in reported job satisfaction from both regular workers on the same contract and those on a permanent contract.

Tests within my sample indicate that the difference in job satisfaction between agency and regular workers on permanent contracts can at least partly be explained by changes in the reference point of agency workers who previously were regular employees, as the difference becomes just insignificant when workers’ perception of just wages is controlled for. Yet, since this finding is derived from a smaller number of observations and the difference in job satisfaction between agency and regular workers on fixed-term contracts is not affected by this modification, the results rather lend support to the conclusion that agency workers on fixed-term contracts regard their employment as a stepping stone to regular employment or as a means to avoid unemployment, while agency workers on permanent contracts appear to be somehow trapped in this type of employment.

My results are derived by controlling for a wide variety of workers’ personal and job characteristics and, in addition, time-invariant unobserved heterogeneity. Moreover, they prove robust to several modifications. Yet critically, there may as well be unobserved factors that vary across both individuals and time and thus, are not eliminated when applying panel techniques, implying that the correlations derived may not be interpreted as causal. Ideally, we would like to have an exogenous source of variation, determining whether one is an agency worker on a permanent contract or on a fixed-term contract. Detecting a causal relationship between job satisfaction and agency status, therefore, provides a fruitful avenue for further research.