Job satisfaction across Europe: An analysis of the heterogeneous temporary workforce in 27 countries

The consequences of temporary jobs for job satisfaction are not clear. This article examines the effect of two crucial moderators in the association between temporary contracts and job satisfaction: the reason for being a temporary worker and the duration of temporary contracts. Using the ad-hoc module of the 2017 EU Labour Force Survey (EU-LFS), this study examines 27 European countries separately. Results show that involuntary temporary workers (those who wanted a permanent contract but could not find one) tend to be less satisfied than permanent employees. However, voluntary temporary workers (those who prefer temporary over permanent jobs) and temporary workers in apprenticeships or probation periods are generally as satisfied as permanent employees. Shorter contracts frequently exert negative effects on job satisfaction, but only among involuntary temporary workers. Results differ between countries: the differences between temporary and permanent workers are insignificant in Scandinavian countries but large in the post-Socialist states.

) and different kinds of temporary workers for the overall sample -Full models corresponding to Figure 1. 5 Table A2: Linear regression estimates. Difference in job satisfaction between permanent (ref.
) and different kinds of temporary workers, by country -Full models corresponding to Figure 2. 7 Table A3: Linear regression estimates. Difference in job satisfaction between permanent (ref.
) and different kinds of temporary workers with different contract durations for the overall sample -Full models corresponding to Figure 3. 10 Table A4: Linear regression estimates. Difference in job satisfaction between permanent (ref.) and involuntary temporary workers with different contract durations, by country -Full models corresponding to Figure 4. 12 Table A5: Linear regression estimates. Difference in job satisfaction between permanent (ref.) and instrumental temporary workers with different contract durations, by country -Full models corresponding to Figure 5. 16 Table A6: Linear regression estimates. Difference in job satisfaction between permanent (ref.) and voluntary temporary workers with different contract durations, by country -Full models corresponding to Figure 6. Table B1: Average marginal effects from multinomial logistic regression models. Difference in job satisfaction between permanent (ref.

III -Robustness tests
) and different kinds of temporary workers for the overall sample -Replications of models in Figure 1 and Table A1. 20 Table B2: Average marginal effects from multinomial logistic regression models. Difference in job satisfaction between permanent (ref.
) and different kinds of temporary workers, by country -Replications of models in Figure 2 and Table A2. 21 Table B3: Average marginal effects from multinomial logistic regression models. Difference in job satisfaction between permanent (ref. ) and different kinds of temporary workers with different contract durations for the overall sample -Replications of models in Figure 3 and Table A3.
23 Table B4: Average marginal effects from multinomial logistic regression models. Difference in job satisfaction between permanent (ref.) and involuntary temporary workers with different contract durations, by country -Replications of models in Figure 4 and Table  A4 24 Table B5: Average marginal effects from multinomial logistic regression models. Difference in job satisfaction between permanent (ref.) and instrumental temporary workers with different contract durations, by country. -Replications of models in Figure 5 26 and Table A5. Table B6: Average marginal effects from multinomial logistic regression models. Difference in job satisfaction between permanent (ref.) and voluntary temporary workers with different contract durations, by country. -Replications of models in Figure 6 and Table  27 A6. Table B7: Linear regression estimates. Determinants of job satisfaction -Income and household composition included as control variables, compared to reference models ( Figure 2 and Table A2). 28 Table C1: Descriptive statistics of the sample -Correspondence with samples from Figure  31 1, Table A1 and Table B1.  Table A2 and Table B2. Table C3: Descriptive statistics of the sample -Correspondence with samples from Figure  35 3, Table A3 and Table B3.  Table A4 and Table B4. Table C5: Descriptive statistics of the sample -Correspondence with samples from Figure 39 5, Table A5 and Table B5.

I -Methodological aspects A. Selection of observations
Starting with the full original sample from the Ad-Hoc module of the EU-LFS of 2017, the observations that had missing values for the following variables were discarded: -Job satisfaction.
-Type of work contract (permanent or temporary).
-Reason for being a temporary worker (among those with temporary contracts). -Occupation. -Education. -Nationality.
-Working time (see the characteristics of working time below).
-Number of hours worked in the second job (among those with a second job). -Tenure.
-Duration of temporary contract (observations with missing values for this variable were only discarded when contract duration was used as independent variable).
In addition, workers with the following characteristics were discarded: -Workers whose questionnaire was answered by a third person (proxy interviews).
-Self-employed workers and family workers.
-Workers aged over 64 years.
-Army and military workers.
-Workers who live in a different country than the country where they work.
-Workers who devote more than 10 hours per week to a second job.
Finally, other countries were not analysed because of specific issues: -Slovenia was not analysed because "reason for being a temporary worker" was missing for the whole sample.
-Iceland and Latvia were not included because they contained too few temporary workers to perform a reliable analysis.
-Croatia was not included because the data for Croatia was not provided by Eurostat.

B. Details about specific variables
Education is introduced as a continuous variable, following the ISCED 2011 scale. This classification avoided collinearity issues with Occupation, that was introduced mostly as a categorical variable.
Working time is introduced a category with three items: "Full-time work", "Part-time work" and "Marginalwork". In most of the cases, this classification depended on the number of hours worked per week "as usual" in the main job. Hence, "full-time work" refers to more than 30 hours of work per week, "part-time work" refers to between 15 and 30 hours of work per week, and "marginal work" refers to less than 15 hours of work per week "as usual". In those cases where the number of hours worked per week "as usual" was missing, the number of hours of work during the week of reference was used instead, following the same criteria. If this variable was also missing, the self-classification provided by the worker was used instead. However, workers could only classify themselves as "parttime" or "full-time" workers.Therefore, the analysis might underestimate, to some extent, the number of real "marginal workers". Insummary, the classification gives priority to the usual number of hours per week, then to the number ofhours of work during the reference week, and finally to the selfclassification provided by the worker. Those observations where all these variables were missing were discarded.

C. Robustness tests: Multinomial models
The dependent variable was ordinal (4-point Likert scale), but it was recoded as if it was a continuous one, allowing for the performance of linear regression models. Nonetheless, this violated some regression assumptions. When the dependent variable is ordinal, ordinal logistic regression models aremore appropriate. However, this technique was not fully adequate as it entailed the violation of the assumption of parallel lines. For this reason, multinomial logistic regression models were performed instead. To allow for the comparison of coefficients, average marginal effects are provided instead of odds or odds ratio. As in some countries there were some categories of the dependent variables that contained very few observations, two categories were collapsed: "Not satisfied at all" and "Satisfied to a small extent" were recoded in the same category. Table B1-B6 report the average marginal effects of the independent variables of these multinomial logistic regression models. These models included the same control variables as the models they replicate (these are indicated in the table).

17301
Adjusted R-squared 0.185 0.090 Note: * p < 0.05, ** p < 0.01, *** p < 0.001. (i) Indicates that the categories "Part-time" and "Marginal work" are included under the same category. (j) Indicates that the categories "EU/EFTA" and "Non EU/EFTA" were included under the same category. (k) Indicates that Occupation was included as a continuous (instead of categorical) variable. (!) Indicates that the coefficient is unreliable because of few observations, according to Eurostat guidelines. na Refers to coefficients that are not shown because of the low number of observations.   ) and different kinds of temporary workers for the overall sample -Replications of models in Figure 1 and Table A1. Coef.

Mean / Percentage
(SD) Note: (i) Indicates that the categories "Part-time" and "Marginal work" are included under the same category. (j) Indicates that the categories "EU/EFTA" and "Non-EU/EFTA" were included under the same category. (k) Indicates that Occupation was included as a continuous (instead of categorical) variable. (!) Indicates that the coefficient is unreliable because of few observations, according to Eurostat guidelines. na Refers to frequencies that are not shown because of the low number of observations. Note: (i) Indicates that the categories "Part-time" and "Marginal work" are included under the same category. (j) Indicates that the categories "EU/EFTA" and "Non-EU/EFTA" were included under the same category. (k) Indicates that Occupation was included as a continuous (instead of categorical) variable. (!) Indicates that the coefficient is unreliable because of few observations, according to Eurostat guidelines. na Refers to frequencies that are not shown because of the low number of observations.