Job satisfaction: towards internalizing the feeling of inequality between men and women

ABSTRACT

The more pronounced job satisfaction among women is generally observed despite their less favorable work situation compared to men. However, regression analysis alone in a sample of non-comparable men and women may be subject to model misspecification. Our work uses an innovative matching procedure, Coarsened Exact Matching (CEM), to address this issue and analyze the reasons for the differential in job satisfaction between men and women with the same characteristics. Data from the Sixth European Working Conditions Survey are considered including five measures of satisfaction with career development prospects taken as a new measure. The results show that women are more satisfied with job security, while they seem less satisfied with their career development prospects. A similar level of satisfaction is observed between men and women with regard to social relations, overall satisfaction and salary. Exceptionally, the youngest women, or those with higher education, or employed at a higher hierarchical level, or working in male-dominated sectors, expressed levels of satisfaction that were the opposite of the other women. This is likely due to the fact that these women align their job expectations with those of their male counterparts.

KEYWORDS:

• Job satisfaction
• gender
• Coarsened Exact Matching
• Europe

JEL CLASSIFICATION:

• J16
• J28
• C14
• O33

Acknowledgments

The author thanks the associate editor and the two anonymous reviewers for their helpful and constructive comments He also thanks Jean-pascal GUIRONNET

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Individuals can be close on their propensity scores but far apart on their characteristics (Fowler 2012).

² We excluded people over the age of 65 to conform to International Labor Organization standards on the normal age range for employment. The self-employed are excluded because of the inapplicability of some of the questions used in the study, such as those referring to the size of the establishment or the number of hours normally worked per week, and differences in the determinants of job utility such as income and other working conditions (Blanchflower and Oswald 1992; Clark 1997).

³ The statements were: 1-I feel I am paid well for the effort I put in and the work I do. 2-My job offers good prospects for career growth. 3-I am not in danger of losing my job in the next six months. 4-I generally get along well with my coworkers.

⁴ We have corrected the wage by Purchasing Power Parity to take into account the differences between countries.

⁵ We implement the procedure using the command: $cem$ on stata (Blackwell et al. 2009).

⁶ Studies suggest considering variables that simultaneously influence the decision to participate in treatment and the outcome variable (Rosenbaum and Rubin 1983). Since we do not perform a treatment, we chose the variables based on those that influence job satisfaction.

⁷ We entered several variables assumed to influence job satisfaction as the second predictor in a regression where the gender variable was the first predictor (considering both the quadratic terms of the continuous variables and their interaction terms with the categorical variables as well as the interaction terms between categorical variables) in a simple logit regression. According to this method, we chose the variables that influence all the variables to be explained.

⁸ We have eleven age intervals ([15-19], [20-24], [25-29], [30-34], [35-39], [40-44], [45-49], [50-54], [55-59], [60-64] and 65, which means, for example, that individuals aged between 15 and 19 will be put in the same group and likewise for the other intervals. We introduce the variable into the algorithm by specifying in parentheses the cutoff points according to the thresholds of the intervals.

⁹ We tried more coarsening to get more matches (e.g. opting for 10-year age-gap intervals, etc.), but ended up with almost the same number of observations. As we do not want to delete any variables, since all of them are considered important, our coarsening strategy was therefore limited to this level, and we preferred to include the industry and occupation variables independently, to avoid matching individuals in different types of jobs.

¹⁰ Approximately, $34\mathrm{\%}$ of employees are excluded from the sample. Of these individuals, $29\mathrm{\%}$ are women and $39\mathrm{\%}$ are men. Among the unmatched men and women, most do not match on wage ($20\mathrm{\%}$ of men and $12\mathrm{\%}$ of women), followed by type of occupation ($11\mathrm{\%}$ of men and $13\mathrm{\%}$ of women), followed by sector of activity ($5\mathrm{\%}$ of men and $2\mathrm{\%}$ of women), while for the other variables used in the matching the percentage does not exceed $0.5\mathrm{\%}$.

¹¹ The measure of multivariate imbalance is a relative quantity that depends on the data set and the variables used in the matching. As the two distributions overlap, this measure decreases and tends to 0. It provides both a measure of imbalance for the overall sample, and on a variable by variable basis (see, Iacus, King, and Porro 2012 for more details).

¹² The propensity score is estimated through a logistic regression with the following specification: $Pr$(women dummy = 1)=$\mathrm{F}$(age, age-squared/100, education dummies, health dummy, relationship dummy, number of children in the household dummy, etablishment size dummies, occupational catégory dummies, industry dummies, log income, full time job dummy, permanent worker dummy, partner’s weekly hours of work dummies, household income dummies, country dummies).

¹³ The common support is defined as an interval whose values are probabilities resulting from the estimation of the propensity scores, which makes it possible to restrict the sample to individuals likely to be matched. In this case, individuals who have a propensity score outside this range are excluded from the sample. This conforms to the CEM algorithm, which restricts the data through the formation of strata.

¹⁴ Matching is performed with the commands “psmatch2” (Leuven and Sianesi 2003).

¹⁵ Following the literature, we included a set of variables related to personal and household characteristics: age, age squared out of 100, education level, in a couple, number of hours worked by partner, number of children in the household (under 2, 2–6, and 7–14), number of elderly in the household, and household size (Clark 1997; De Galdeano 2002; Perugini and Vladisavljević 2019). Each of the variables is then interacted with the dichotomous variable for gender (Clark 1997; De Galdeano 2002).

¹⁶ Clark (1997) considers that women who are younger in the workforce, or who have completed higher levels of education, or who are exposed to good jobs or who work in a male-dominated work environment have the same expectations of work as men. They therefore express a lower level of satisfaction with their work.

¹⁷ According to the OECD, people between the ages of 15 and 24 are those who enter the labor market after completing their education.

¹⁸ The robustness of all results was checked at two levels: first, by performing the country-by-country estimates. Second, by considering the satisfaction variables as cardinal variables and applying linear techniques, producing OLS regressions where the selection effect is accounted for through the introduction of the inverse of the Mills ratio (Heckman 1979) from the estimation of the selection equation specified in section III. In general, the results are robust.