Table 4 reports the results of a series of linear random intercept models. The intra class coefficient of the empty model (not shown) indicates that about 19 percent of the variance in life satisfaction is observed at the country and country-round level. This suggests that well-being varies substantially across welfare states. In line with hypothesis 1 as well as previous studies (Wulfgramm 2014), we find that unemployed individuals have, on average, about a one scale point lower life satisfaction compared with workers in secure jobs (model 1). This estimate is already adjusted for all individual-level control variables as well as round fixed effects.Footnote 7 For workers in insecure jobs, the negative effect is smaller, but still practically and statistically significant. Workers who have a temporary contract or no contract at all, on average, have a 0.2 scale point lower well-being compared to those with permanent contracts (model 1). This confirms hypothesis 2 and is also consistent with previous findings (e.g., Scherer 2009). Overall, the effect of unemployment is five times larger than the effect of insecure jobs, suggesting that being employed is particularly important for individuals’ well-being.
Models 2–4 test our hypotheses about the moderating role of PLMP, ALMP, and EPL. The positive cross-level interaction of being unemployed with benefit generosity suggests that higher unemployment benefits buffer the negative effect of unemployment on well-being (model 2). A one standard deviation increase in benefit generosity is predicted to reduce the negative effect by about 0.16 scale points. In contrast, for insecure jobs the interaction coefficient is small and not statistically significant. Figure 1 illustrates these findings by plotting the respective average marginal effects with 90% confidence intervals (y-axis) against the standardized benefit generosity indicator (x-axis).
The flat line for insecure jobs shows that higher unemployment benefits do not alleviate the negative effect of having a temporary contract or no contract at all. In contrast, for the same person becoming unemployed in a country with a benefit generosity of one standard deviation below the mean, as compared to a country with one standard deviation above the mean, makes a difference of about 0.3 scale points in the negative effect on well-being. Given an average effect of about one scale point this equals a 30% reduction. Thus, while hypothesis 3 can be confirmed, hypothesis 4 has to be rejected.
For ALMP we find no moderating effect on the effect of having an insecure job on well-being (hypothesis 6). However, model 2 suggests that higher spending on ALMP is associated with a more negative effect of unemployment. Specifically, a one standard deviation increase in ALMP is predicted to increase the negative effect on well-being by about 0.17 scale points. Assuming that ALMP participants are mostly included among the unemployed, this finding contradicts hypothesis 5 and suggests that higher ALMP expenditure exacerbates the negative effect of unemployment. However, if one assumes that ALMP participants are mostly included with the employed, the same finding may be interpreted more positively. In this case, in countries or country-rounds with higher ALMP expenditure the unemployed are likely composed of individuals who have not been activated and who may be negatively selected in terms of their well-being and health.
With respect to EPL for regular employment (EPLR), we expected that a stricter protection increases the differences between labor market insiders and outsiders (hypothesis 7 and 8), resulting in more negative effects of unemployment and insecure jobs on well-being. The results of model 2, using the time-varying EPLR indicator, suggest that the extent of job security provisions does not moderate these effects. The respective interaction coefficients are close to zero and not statistically significant. However, to test our hypotheses about the effects of partial deregulation, we also generated a time-constant indicator distinguishing countries with low EPLR compared to those with high EPLR (High EPL regular in Table 4).Footnote 8 Repeating the analyses for model 2, using the time-constant EPLR indicator, yields somewhat different results (see model 3). In countries with high EPLR, the negative effect of unemployment on well-being is about −0.16 scale points larger than in countries with low EPLR. For insecure jobs, the results are consistent across model 2 and 3, providing evidence against hypothesis 8. For unemployment, the results, however, depend on the measurement of EPLR, with the time-constant indicator giving an estimate that is in line with hypothesis 7.
Model 4 tests our hypotheses about partial deregulation by including a three-way-interaction between employment status, the high EPLR indicator, and the time-varying indicator about restrictions on the use of temporary employment (EPLT). Our main interest is in the coefficients for the three-way-interactions reflecting the differences in the moderating effects of EPLT in countries with low and high EPLR.
To facilitate the interpretation of the results, Fig. 2 plots the average marginal effects of unemployment and insecure jobs on well-being against the standardized EPLT indicator. As can be seen from Fig. 2, these effects differ in countries with low and high EPLR.
In line with hypothesis 9, deregulating the restrictions on the use of temporary employment, that is moving in the negative direction on the standardized EPLT indicator, increases the negative effect of unemployment on well-being in countries with high EPLR, but not in those with low EPLR. In the latter, deregulation even has a buffering effect with respect to the negative well-being consequences of unemployment. The difference in the moderating effect of EPLT for countries with low and high protection of insiders is given by the coefficient of the three-way-interaction term in Table 4, model 4 (0.27).
For the effects of insecure jobs the pattern is similar. However, given that in countries with high EPLR the moderating effect of EPLT is almost zero, the finding is only partly in line with hypothesis 10. The positive interaction coefficient is mostly driven by the fact that in countries with low EPLR deregulating the restrictions on the use of temporary employment buffers the negative effect of insecure jobs on well-being.
Across models 2–4, we also find a negative moderating effect of the unemployment rate on the effect of unemployment, suggesting that individual unemployment has more negative consequences for well-being if aggregate unemployment is high. This result is at odds with the idea that unemployment hurts less if there is more of it around (e.g., Oesch and Lipps 2013).
To test the robustness of these findings, Table 5 reports the results of the analyses using pooled linear regression models including country and round fixed effects (models 5–8). Note that these models are based on 23/109 instead of 26/112 countries/country-rounds, because the three countries that have not been observed repeatedly (see Table 3) do not contribute to the estimates of the moderating effects of labor market policies.
Overall, the point estimates are very similar to those reported for the linear random intercept models. Therefore, we only shortly summarize the main findings. Higher unemployment benefit generosity buffers the negative effects of unemployment, but is of little importance with respect to the well-being consequences of insecure jobs. We also find a negative moderating effect for ALMP (model 6). As for the multi-level analyses, the effect of EPL for regular employment (EPLR) depends on the measurement. Only the time-constant indicator (model 7) gives a point estimate in line with hypothesis 7, stating that a stricter protection of labor market insiders increases the negative effect of unemployment on well-being. For insecure jobs, the effects are once more small and not statistically significant.Footnote 9
With respect to the partial deregulation hypotheses, the analyses including country fixed effects confirm the findings of the linear random intercept models. For unemployment, we find that deregulating the restrictions on the use of temporary employment (EPLT) increases its negative effects in countries with high EPLR, but alleviates the well-being consequences in countries with low protection for insiders. For insecure jobs, the positive interaction coefficient is rather due to the buffering effect of deregulation in countries with low EPLR.
Table 6 presents the results for the linear random intercept models. The intra class coefficient of the empty model (not shown) indicates that about 6 percent of the variance in health is observed at the country and country-round level. Compared with well-being, this suggests that cross-national differences in self-rated health are smaller than those in life satisfaction, but still practically and statistically significant.
To test hypothesis 1 and 2, model 1 includes the employment status variable as well as the individual-level control variables and round fixed effects. Compared with workers in secure jobs, unemployed individuals, on average, have an about 7.3 percentage point lower probability to be in very good or good health. In contrast, workers in insecure jobs have an only about 1.3 percentage point lower probability of reporting very good or good health. This is in line with hypotheses 1 and 2 as well as confirms the findings for well-being, showing that unemployment is associated with more negative consequences compared with insecure jobs.
Model 2 includes the macro-level variables and the respective cross-level interactions between employment status and labor market policies. In contrast to our findings for well-being, the results reveal no positive moderating effect of benefit generosity for the effect of unemployment on health. Hypothesis 3 can, therefore, only be confirmed for life satisfaction. One interpretation may be that unemployment benefits are more conducive to the psychological than the less malleable, physical aspects of health and that well-being is more closely related to the former.
In line with the analyses on well-being, higher expenditures on ALMP increase the negative effect of unemployment on health. A one standard deviation increase in spending per unemployed increases the negative effect on health by about 2.4 percentage points. However, as for well-being this result may be interpreted differently (see above and conclusions section).
Considering EPLR, the results for both the time-varying (model 2) and time-constant indicator (model 3) are in line with hypothesis 7, stating that a stricter protection for labor market insiders is associated with more negative effects of unemployment on health. A one standard deviation increase in EPLR is predicted to increase the negative effect of unemployment on health by about 1 percentage point (model 2) and in countries with a high EPLR the negative effect of unemployed is increased by about 2.9 percentage points (model 3). For the effect of insecure jobs, the EPLR again seems to be of little importance. For the restrictions concerning the use of temporary employment (EPLT), the moderating effect differs between countries with low and high protection for insiders. Specifically, in high EPLR countries, a standard deviation decrease (deregulation) in EPLT is predicted to increase the negative effect of unemployment on health by about 1.7 percentage points (= −0.027 + 0.044) while in countries with low protection for insiders the negative effect is expected to be alleviated by about 2.7 percentage points (model 4). In contrast to the findings for well-being, the coefficient of the three-way-interaction term for insecure jobs is, however, not statistically significant. For the control variable, unemployment rate, the results are opposite to the findings concerning well-being meaning that higher aggregate unemployment reduces the negative effect of individual unemployment on being in very good or good health.
Looking at the point estimates of the pooled linear regression models including country and round fixed effects, we find very similar results. Table 7 presents the respective models. Similar to the linear random intercept models, the only practically significant moderating effects are found for unemployment.
Based on within country changes in labor market policies, model 6 suggest no buffering effects of benefit generosity, but a negative moderating effect of expenditures on ALMP. Also for the analyses concerning the EPL for regular employment (EPLR) and the partial deregulation, the results of the linear random intercept models are mostly confirmed.
Before concluding, we briefly report on some additional robustness checks. First, we tested whether our results are affected by influential country-rounds and countries by re-estimating a specific model many times and deleting each country-round or country once.Footnote 10 Concerning the analyses excluding each country-round once, we find that our results are quite robust and that the estimates for the cross-level interactions vary only slightly. However, for the analyses excluding each country once (all country-rounds of the respective country are excluded), we find that the positive buffering effect of unemployment benefit generosity depends on the inclusion of Portugal. Similarly, the negative moderating effect of high EPLR depends on Germany being included.Footnote 11 Second, we tested whether the results for benefit generosity are sensitive to measurement. Using the expenditure on PLMP per unemployed as a percentage of GDP growth per capita as an alternative indicator (see Table 1 for definitions), we do not find a positive moderating effect of higher benefit generosity on the effect of unemployment on well-being. Moreover, we find an alleviating effect of higher benefit generosity for the negative effect of insecure jobs on health. The implications of these robustness checks will be discussed in the next section.