Descriptive Findings
Table 1 shows unweighted univariate descriptive statistics for our key micro- and marco-level variables averaged over all individuals (n = 18,596) and countries (N = 23). Regarding the key independent variable, 16% of the individuals are fixed-term employed, 79% are permanently employed and 5% are unemployed. Over all countries, the interviewees report on average 6.91 points on the life satisfaction scale ranging from 0 (“extremely dissatisfied”) to 10 (“extremely satisfied”). The standard deviation, however, is 2.26 units, indicating a high variation either across individuals and/or countries.
Table 1 Univariate descriptive statistics for both micro-level and macro-level variables. Concerning the macro-level characteristics, the index for social cohesion varies for the 23 countries of interest from − .97 for Bulgaria to 1.32 in Denmark with an average of zero and a standard deviation of .70 units. Regarding the ratio for the income inequality, the average is 4.53 with a standard deviation of .92 units. According to the index, Spain has the highest income inequality with a ratio of 6.5 and Norway the lowest, i.e. 3.2.
Hypothesis 1.1
Figure 1 shows the results regarding the first hypothesis for each country separately and ordered by effect sizes. To repeat, Hypothesis 1.1 states that fixed-term compared to permanent employment should reduce the perceived well-being of the individuals. The results are unambiguous concerning the effect direction. The coefficient plot shows that in all countries, the impact of fixed-term compared to permanent employment is estimated to be negative, holding constant the control variables. However, the effect sizes do seem to vary noticeably.
One example for the interpretation of the coefficient for France (FR) is: On average, being temporarily rather than permanently employed ceteris paribus decreases the life satisfaction by 1.19 units. The effect is statistically significant, as the confidence interval does not include zero. Moreover, it is the strongest negative impact size compared to the other countries. For Belgium (BE) we find the smallest negative effect. Having a temporary compared to a permanent job in Belgium on average decreases the well-being by only .17 units. The estimate is statistically insignificant. Moreover, for six additional countries, namely Ireland, Estonia, Norway, Switzerland, Slovenia, and Cyprus we do not find statistically significant effects. Nevertheless, the direction of the effects fits the expectations.
The range between the point estimates is 1.02 units, indicating a great variation in the effect sizes across countries. It is also noticeable that some of the countries with a rather liberal labor market (Great Britain, Netherlands) have a smaller effect, while the conservative ones (e.g. Spain, Italy, Germany, France) are among those countries with the greatest effect sizes.
Yet, regarding both effect directions and sizes in general for all the 23 countries, we find support for the upwards comparative effect. Moreover, the results suggest effect heterogeneity across countries.
Hypothesis 2.1
Figure 2 shows the results for the downwards perspective. The hypothesis states that fixed-term employees should have a greater subjective well-being compared to unemployed individuals. The findings suggest a more ambiguous picture compared to the upwards comparative effects. Holding the important confounders constant, for 18 of the 23 countries there are coefficients in the direction of expectation, whilst for five this is not the case.
For Norway (NO)—the largest impact size—being temporarily employed compared to being unemployed on average and statistically significantly increases the subjective well-being ceteris paribus by 1.25 units. The smallest negative effect size appears for Denmark (DK), where, on average and holding the confounders constant, jobs with a fixed duration compared to unemployment even decrease the life satisfaction by .27 units. This estimate is statistically insignificant as well as it is for Poland, Italy, Slovenia, and Portugal. Moreover, the range for the magnitude of the coefficients is 1.52 units, which again points to a great variation in the effect sizes across countries.
It is interesting to note that for some countries, which have a point estimate very close to zero, it does not seem to matter whether individuals are fixed-term employed or unemployed. This result points towards the argument from the literature, suggesting that this kind of job insecurity might feel as harmful as unemployment (Inanc 2018). However, for the large majority of the countries, to wit 18 countries, the effect is estimated to be positive.
When looking at the direction and significance of the effects, for five countries the data support the hypothesis. These findings indicate that temporary employees have a greater well-being compared to unemployed individuals. Nevertheless, for most of the countries, where the direction of the impact fits the expectation, we do not obtain a statistically significant estimate.
Hypothesis 1.2
To explain the effect of fixed-term employment on well-being, we suggest that some of the functions of the Latent Deprivation Model should matter. This assumption is tested by conducting mediation analyses, for which the results are shown in Table 2. Like previously explained, we consider three different estimates: the total effect (M1), the direct and the indirect effect (both included in M2–M5). Concerning Hypothesis 1.2, we assume that the effect size from M1 decreases in M2–M5 and that this decline (Δ) is statistically significant.
Table 2 Pooled linear regression models with mediation analyses for the effect of fixed-term employment compared to permanent employment. On average, we find that fixed-term employees are—holding the confounding variables constant—.833 units less satisfied with their lives compared to permanent employees. This is also what we expected in the basic Hypothesis 1.1 and thus also holds across countries. It is statistically significant as t = − 11.75.
Adding the mediating variable for social contacts, the impact only slightly decreases to − .817 units. The point estimate is still statistically significant as t = − 7.49. Thus, the coefficient from M1 only slightly decreases in M2 and this decline Δ = − .016 is not statistically significant (z = − 1.46). This leads us to reject the hypothesis on social contacts being a mechanism for the comparison of fixed-term to permanent employees.
Another latent function, which we suggest mediating the impact is the status. Adding this mediator in M3, the total effect declines in both empirical size and statistical relevance to a direct effect of − .508 units (t = − 6.45). The indirect effect of − .325 units—about 39% in relative terms—is statistically significant (z = − 9.49). Thus, we find support for status to explain some part of the effect of fixed-term employment on well-being.
The greatest decline, however, can be observed for adding the manifest function of the income (M4): the total effect reduces to the direct effect of − .335 units, that is to say a 60% (Δ = − .498, z = − 10.83) decline in the size of the point estimate. Thus, the data support the importance of worries about income to be one explanation for the negative impact.
Adding all mechanisms simultaneously (M5) decreases the total effect of − .833 units to a much smaller point estimate of − .255 units (t = − 3.80), which is still statistically significant. The decline of .579 units proves statistical significance (z = − 13.04) and indicates in relative terms a 70% weakening.
Whereas the results show that adding the mediating variables notably decreases the main correlates, still there remains a − .255 unit effect size, which we cannot explain. This finding might refer to the fact that we were restricted in the choice of mediators by the data source. Nevertheless, the results give first hints for utilizing the Latent Deprivation Model when explaining the main effects. It is a reasonable theoretical approach to explain the association between fixed-term employment compared to permanent employment and well-being.
Hypothesis 2.2
With the results of the previous analyses in mind, we now return to the downwards comparative effects. Table 3 reveals the results of the mediation analyses. Even though we could not find support for every single country for a positive connection to exist, on average (M1) it turns out to be statistically significant and in the direction of expectation (β = .581, t = 5.75). Thus, ceteris paribus individuals who are temporarily employed rather than not having a job, report on average a .581 unit higher life satisfaction. Within the framework of the Latent Deprivation Model, we suggest social contacts, status, income, activity, time structure and supra-individual aims to explain this positive correlate.
Table 3 Pooled linear regression models with mediation analyses for the effect of fixed-term employment compared to unemployment. However, when it comes to social contacts, adding the variable to the model (M2) again does not notably decrease the positive impact: the point estimate decreases to .560 units (t = 4.48), which implies a statistically insignificant net difference of Δ = .021 (z = 1.09). Thus, we do not find support for social contacts to mediate the downwards comparative effects.
Regarding the status, the analyses (M3) reveal its important role as a mediator: the indirect effect of Δ = .134 is statistically significant with z = 3.18.
Similar to the results for the upwards comparison, the perceived financial situation is the most important explanation for the total effect. Incorporating it (M4) decreases the point estimate by 85% and the t-value to .75. Thus, the effect even becomes statistically insignificant when accounting for the financial resources.
Different from the previous models, we now also include activity (M5). We find that this does neither cause a change in the initial effect size nor in its statistical significance (Δ = − .001, z = − .43). Owing to this, we need to reject the hypothesis of activity being a mediator.
When we add all mechanisms simultaneously (M5), the total effect of θ = .581 (t = 5.75) can be decomposed into a direct effect of Θ = .129 (t = 1.18) and an indirect effect of Δ = .452 units (z = 7.76). In relative terms, this means that 78% of the initial total effect can be explained by the added mediating variables. Vis-à-vis 22% remain unexplained. This unexplained share of the effect might be due to lacking the possibility to test for the mediating impact time structure and supra-individual aims. However, the findings pinpoint to the mediating role of both the manifest as well as the latent functions.
Comparing the results of Hypothesis 1.2 and 2.2, it is interesting that the explanatory power of the mediating variables seems to be very similar in magnitude. This finding also strengthens the interest in the results of the following analyses regarding social cohesion as a country-level moderator, which we assume to substitute the micro-level mechanisms. Table 4 includes the estimates for the cross-level interaction effects.
Table 4 Random slope linear regression models for the cross-level interaction effects of social cohesion. Hypothesis 3
First, we again return to the well-being effects comparing fixed-term to permanent employment. Now we are interested in whether an increasing level of social cohesion on the macro-level reduces the negative impact. More specifically, this assumption refers to the difference between temporarily and permanently employed individuals in their well-being on the micro-level.
The coefficient on intra-class correlation (ICC) of the empty model (not shown) indicates that about 17% of the variance in life satisfaction of fixed-term and permanent employed individuals is observed at the country-level. This indicates a substantial variation across welfare states. The likelihood ratio test on the random slope indicates that it is reasonable to include a random slope for the micro-level independent variable of fixed-term employment (\( {\text{X}}^{2} = 28.20, p < 0.01 \)).
M1 includes the random slope specification. The model reveals that, controlling for all important micro- and macro-level confounding variables, on average fixed-term employment decreases the subjective well-being of individuals by .69 units, when comparing them to permanent employees. Therefore, we again find support for Hypothesis 1.1.
The interaction term is added in M2 and equals .19 units. This indicates that a one unit increase in social cohesion ceteris paribus decreases the negative well-being effect of fixed-term compared to permanent employment by .19 units. Hence, the greater the social cohesion in a country is, the smaller the well-being difference between fixed-term and permanent employees. The direction of the moderating term is in line with our theoretical assumptions and the test statistics (t = 2.22) backs up the importance of social cohesion as macro-level moderator.
Figure 3 visualizes the cross-level interaction. It includes the linear predictions for both fixed-term and permanent employees regarding their well-being for specific values of social cohesion. For instance, for countries with a social cohesion of − 1 unit, the prediction for the subjective well-being of fixed-term employees is 5.29 units. For permanent employees living in those countries, the models predict a 6.17 unit well-being. The difference equals the impact, i.e. fixed-term employees have a .88 unit lower subjective well-being compared to permanent employees in countries with a very low social cohesion. For countries in which the social cohesion is high, i.e. 1 unit, the effect is − .60 (7.47–8.07). Thus, the greater the social cohesion is, the smaller the well-being difference between fixed-term and permanent employees.
Eventually, we find support for social cohesion on the macro-level to diminish the negative impact of fixed-term employment compared to permanent employment on subjective well-being.
Hypothesis 4
For the well-being effects regarding fixed-term employed compared to unemployed individuals, the theory also assumes social cohesion to diminish the impact by substituting the role of social contacts. More precisely, Hypothesis 4 suggests a greater social cohesion to balance out the differences between fixed-term employed and unemployed individuals, such that they should turn into a zero correlate.
The coefficient on intra class correlation (ICC) of the empty model (not shown) indicates that about 16% of the variance in life satisfaction of fixed-term employed and unemployed individuals is observed at the country-level. Again, this suggests a high heterogeneity across welfare states. The likelihood ratio test on the random slope indicates that there is no statistically significant improvement when including a random effect for the micro-level independent variable of fixed-term employment compared to a model without it (\( {\text{X}}^{2} = 0.69, p = 0.70 \)). However, since we theoretically assume the downwardly comparative effect to vary across countries, which the results of Hypothesis 2.1 also indicated, we include a random slope anyway.
The simultaneous estimation procedure reveals that, ceteris paribus, fixed-term employees have a .47 greater subjective well-being compared to unemployed individuals (M1). This effect is statistically significant (t = 4.98). Thus, the results also support Hypothesis 1.2. Specifically, the effect of the upwards comparison seems to be stronger compared to the downwards comparatively effects. This is in line with the findings for Hypothesis 1.1 and 1.2, where we fitted regression models for each country separately.
Adding the cross-level interaction effect in M2, results in an interaction term of .02, which is statistically insignificant (t = .13). In greater detail, this means that one unit increase in social cohesion increases the positive effect by .02 units. Specifically, the direction is in opposite to what we expected, but is also very close to zero. Thus, there is no support for the moderating effect of social cohesion on the well-being effect of fixed-term employment compared to unemployment.
Figure 4 shows that the linear prediction lines for the subjective well-being for fixed-term employees and unemployed individuals dependent on social cohesion almost run parallel. When there is very low social cohesion, i.e. − 1 unit, the effect of fixed-term employment compared to unemployment is .45 units (5.26–4.81). When there is a high level of social cohesion, i.e. 1 unit, the impact is .49 units (7.41–6.92). Moreover, the 95% confidence intervals overlap, which means that the differences are not statistically significant.
Therefore, we cannot find support for the hypothesis that social cohesion moderates the effect of fixed-term employment compared to unemployment on subjective well-being.