Data collection and model specifications

The data are extracted from the ESS, which is an academically driven multi-country survey that has developed a series of social indicators, including attitudinal indicators. Ten ESS surveys have been conducted since 2002, and we use the latest (tenth) round survey in 2020 with 18,060 valid respondents. The life satisfaction question reads: All things considered, how satisfied are you with your life as a whole nowadays? We use this as the dependent variable in our analysis. The independent variables are the paid and unpaid working hours per week. Detailed survey questions and descriptions of the indicators are listed in Table ​Table2.2. In addition to the life satisfaction and working time variables, we also include personal characteristic indicators, including health, social inclusion, social trust, feelings of safety, digitalization, income, marital status, gender, age, religion, and education. The mediating role of health is tested in this study and we further explore the worktime–satisfaction nexus in the three income levels (low-, mid- and high-income) and six job categories (central or local government, other public sector (such as education and health), state-owned enterprise, private firm, self-employed, and other). We divide income level into three equal groups with low-, mid- and high-income.

Survey questions and descriptions of the variables from the ESS dataset

We recode Education as follows: we set 520 and below (including 520 and 000) to 1, between 610 and 620 to 2, between 710 and 720 to 3, and 800 and above to 4. Marital status is recoded, where 1 denotes married, otherwise it is 0. The Health variable includes both physical and mental health

Source: European Social Survey [33]

Robustness checks are applied by replacing life satisfaction with happiness such that the happiness question reads: Taking all things together, how happy would you say you are? Strictly speaking, life satisfaction and happiness have different connotations: the former reflects an individual’s cognitive judgment about the compatibility of living circumstances based on their own work and life experiences [48, 49], while the latter is a hedonic/emotional evaluation of their current state of mind [50]. For example, Lara et al. [51] regarded life satisfaction as the cognitive indicator of well-being and examined its association with current happiness. However, Schyns [52] found a close association between life satisfaction and happiness and suggests an interchangeable use of these two indexes. Mainstream literature follows this course and employs these two indexes to explain the individual’s subjective well-being [53–56]. Caner [57] estimates and compares the regression results using life satisfaction and happiness as outcome variables respectively. This study replaces life satisfaction with happiness to check its robustness. The reliability of our analysis is further verified if the outcomes after variable substitution are similar. Pairwise correlations for the dependent variables and the explanatory variables are reported in Table ​Table3.3. The results illustrate two facts: first, most variables are significantly correlated at the 10% level, and second, working time is negatively correlated with life satisfaction as well as other explanatory variables except for gender and age. The observations highlight the importance of careful multivariate econometric analysis.

Correlation analysis

^adenotes that the correlation is significant at the 5% significance level (2-tailed). Job category is not included because it is not an ordinal variable and we use it to divide different groups

The ordered probit model was proposed by McElvey and Zavoina [58] for the analysis of categorical, non-quantitative choices, outcomes, and responses. To tackle the single crossing property problem inherent in standard logit/probit models (i.e., that the signs of the marginal effects can only change once when moving from the smallest to the largest categories), Boes and Winkelmann [59] propose four alternative models: the generalized threshold, random coefficients, finite mixture, and sequential models. The ordered probit model is suitable for this study [60] considering that the dependent variables—life satisfaction and happiness—are ordinal data that range from 0 to 10. More importantly, the ordered probit model takes into account unobserved heterogeneity and ordinarily in life satisfaction scales while using full information contained in the data [1]. As both the ordered probit and logit models are commonly employed to analyze such ordinal data, we choose the former since it is widely used in the related literature [18, 61, 62]. The basic equation of the ordered probit model is:

where yi represents the dependent variable and yi∗ the latent variable, denoting 11 levels of life satisfaction. Xi is a vector of explanatory variables that assesses the attribution of life satisfaction, and βi is the coefficient of Xi, a vector of estimated parameters to be projected, which represents the impact magnitude of the independent on the dependent variables. Finally, εi is unobserved white-noise disturbance, where Eεi.

Moreover, since the coefficients of the ordered probit model cannot be directly explained while the estimators are very similar to the ordinary least squares (OLS) model, we also construct the following alternative econometric specification following Ronning and Kukuk [63]:

where Satisfactioni is the life satisfaction level reported by individual i, Worktimei is the reported working hours per week reported by individual i, ΣIndividuali is the vector of the respondent’s individual characteristics, and μi is an error term. It is worth noting that the results are presented in forest plots to be visually friendly, referring to Becker and Kennedy [64], Lechner and Okasa [65] and Kostka et al. [66].