Skip to main content

Women, Men, and Job Satisfaction


We examine job satisfaction to determine whether gender differences found by previous researchers could be explained by constraints imposed by the specifications used. Applying those specifications to recent US data yields results similar to those previously found. However, clarification comes from applying specifications that allow for gender differences in sample selectivity and in the relative weights (β̂′s) of personal/job characteristics in evaluating satisfaction. We find that gender differences in the job satisfaction of married workers can largely be attributed to gender differences in β̂′s. However, more work is necessary to understand gender differences in job satisfaction among unmarried workers.

This is a preview of subscription content, access via your institution.


  1. As Budig and England [2001] have found, there is a wage disadvantage for married mothers, which lowers their opportunity cost of non-market work, so they may choose not to work.

  2. The traditional thinking among economists is that measurements of utility are subjective and interpersonal comparisons of such measurements are not useful. However, these researchers point to the numerous findings of strong relationships between job satisfaction and various aspects of worker behavior, as evidence that the study of interpersonal comparisons of self-reported measures of job satisfaction is indeed meaningful. For a survey of research on happiness in general, see Frey and Stutzer [2002].

  3. For seminal work on sample selectivity, see Heckman [1976].

  4. The specification for this portion of the analysis is fairly standard. For example, in estimating participation equations by gender, Christofides et al. [2003] use independent variables that reflect age, education, presence of children, whether disabled, whether married, whether a family head and whether an immigrant, along with controls for region of residence.

  5. While there is some overlap in 𝕏 and W, there are certain variables included in 𝕏 that are not included in W and vice versa. 𝕏 contains characteristics of one's job, while W does not. Household composition and unearned income (reflected here by number of children, presence of children under the age of 6, spouse's education level, and spouse's work status) are typically components of W but not of 𝕏. These patterns are true for the participation equations and the job satisfaction equations appearing in Clark [1997] and Sloane and Williams [2000], for example. One notable exception can be found in Donohue and Heywood [2004], where the presence of young children was an independent variable in one specification of a job satisfaction equation. The authors explained its inclusion by saying that they anticipated that the variable could serve as a control for sample selection issues associated with women; in fact, they found it to be related to higher job satisfaction for both white-collar female and blue-collar male workers. Here, we choose to control for sample selection issues by addressing job satisfaction and participation in separate equations.

  6. Details about this technique can be found in Heckman [1976].

  7. In studies of US workers, Donohue and Heywood [2004] used the 1988 National Longitudinal Survey of Youth, while Bender et al. [2005] used the 1997 National Study of the Changing Workforce. In studies of British workers, Clark [1997] used the 1991 British Household Panel Survey, while Sloane and Williams [2000] analyzed the 1986 Social and Economic Life Initiative Survey. In an international comparison of workers, Sousa-Poza and Sousa-Poza [2000] used data from the 1997 International Social Survey Program (ISSP); the US data of the ISSP have been collected as a module of the GSS since 1994.


  9. Historically, survey results were based on independently drawn samples of English-speaking persons 18 years of age or over, living in non-institutional arrangements within the United States. Starting in 2006 Spanish-speaking individuals were added to the target population.

  10. Since the beginning, the GSS employed a rotation design; permanent items appeared on every survey, but rotating items appeared on two out of every three surveys. The importance of the rotating scheme has increased over the years; more items have been shifted from permanent to rotating status, to make room for topical modules of special interest [Davis and Smith 2009, GSS codebook, Appendix Q].

  11. Sloane and Williams [2000] found that neither absolute pay nor objective (predicted) comparison pay were significant when entered jointly into a job-satisfaction equation. However, they did find that both absolute and subjective comparison pay were highly significant when entered jointly.

  12. For an income variable available in both the 2002 and 2006 versions of the GSS, the categories for annual earnings are: less than $1,000; between 1,000 and 3,000; between 3,000 and 4,000; between 4,000 and 5,000; between 5,000 and 6,000; between 6,000 and 7,000; between 7,000 and 8,000; between 8,000 and 10,000; between 10,000 and 15,000; between 15,000 and 20,000; between 20,000 and 25,000; and 25,000 and above.

  13. The combined total number of respondents to the 2002 and 2006 GSS was 7,275. We excluded 845 who were over the age of 70. Both health and education of spouse were “rotating items” in the GSS in these years. As a result, a total of 1,900 respondents were never asked about health and 760 married individuals were never asked about spouse's education. The losses of observations largely explain the reduction to the 4,012 observations used in the first stage of estimation.

  14. Only 2,732 of the 4,012 observations used in the first stage of estimation were part- or full-time workers. Only 1,477 of these are used in the second stage of estimation. The reduction in numbers is largely due to the fact that about 1,100 workers were never asked about quality of working life in 2006 and in both years combined over 500 respondents with income did not provide information about that income. Additionally we excluded 163 part- or full-time workers with data because they were self-employed. While it would be interesting to explore the job satisfaction of self-employed workers separately, the numbers of self-employed workers were too small to support such an investigation here.

  15. In previous studies of job satisfaction, researchers have used either health status as represented by reported limitation on activity [Donohue and Heywood 2004; Bender et al. 2005] or self-reported evaluations of health [Clark 1997]. In the 2002 and 2006 GSS data, only the latter is available for use here.

  16. When asked about the fairness of earnings in comparison to others doing the same type of work, respondents were asked to indicate if earnings were much less than deserved, somewhat less than deserved, about what was deserved, somewhat more than deserved or much more than deserved. For four variables, a statement about working environment was made (e.g. “My fringe benefits are good.”) and respondents were asked whether the statement was “very true,” “somewhat true,” “not too true,” or “not at all true.” In regards to amount of employee input on the job, respondents were asked whether they “strongly agree,” “agree,” “disagree,” or “disagree strongly” with the statement “I have a lot of say about what happens on my job.” In terms of ability to take time off, respondents were asked to indicate whether taking time off from work for personal or family matters is “not hard,” “not too hard,” “somewhat hard,” or “very hard.” As for the frequency with which the job interferes with family life and the worker can make changes in starting or quitting times on a daily basis, respondents were asked to indicate “often,” “sometimes,” “rarely,” or “never.” In answering a question about whether the respondent finds work stressful, respondents were asked to indicate “always, ” “often, ” “sometimes, ” “rarely,” or “never. ” Table 1 indicates how responses were collapsed into fewer categories for 11 related variables.

  17. This finding could very well reflect the larger wage penalty associated with married mothers, found by Budig and England [2001].

  18. These findings with respect to MALE and MARRIED are robust across different specifications (with and without sample weights, with and without HEALTH, with and without AGE squared, with and without D06) and across different samples (70 years old and younger, 60 years old and younger).

  19. More specifically, Appendix Table A3 shows the results of first-stage probit estimation for the likelihood of being at work full-time or part-time (as opposed to being with a job but not at work, unemployed, laid off, in search of a job, retired, or in school, keeping house or doing something else), by gender and marital status. Appendix Table A1 defines additional independent variables used in the analysis; Appendix Table A2 provides associated descriptive statistics for the sample used in this first-stage of estimation.

  20. See Heckman [1976, p. 479].

  21. Interestingly, a significant negative coefficient for lambda is also frequently found in earnings equations for women [ Dolton and Makepeace 1987; Wright and Ermisch 1991; Sloane and Williams 2000].

  22. These results are based on first-stage samples that include workers who were not asked questions about job satisfaction and who were subsequently excluded in the samples of workers used in the second-stage estimation, because of missing data. The reduction in the number of workers from the first to the second stage of estimation is unfortunate. To avoid the loss of valuable information about the decision to work, however, we retained the full set of workers for the first-stage estimation, rather than omitting them throughout the analysis.

  23. The estimation is repeated with different specifications (without HEALTH; with AGE squared; with TENURE squared) and a different sample (those under the age of 66). When the sample is limited to those under the age of 66, the results are similar except that the P-value for the lambda of married women slips to 0.12. When HEALTH is omitted from the ordered-probit stage, the lambdas of both married men and married women are significant at conventional levels of significance. When TENURE squared is added, it is associated with a negative but insignificant effect; with its addition, the P-value of the lambda for married women slips to 0.13 while the lambdas for all other groups remain insignificant. When AGE squared is added, a significant U-shaped relationship between age and job satisfaction is found for married men only and only when the model includes a correction for sample selection; as a result of its inclusion, the lambda for married women becomes rather insignificant (P-value=0.22) while the lambda for married men becomes highly significant (P-value=0.03). Given the oddness of the latter results and the nature of the variable itself, we feel it more relevant for the first-stage probit and less relevant than TENURE squared for the second-stage ordered probit.

  24. A number of studies have found that dissatisfied workers are more likely to quit their jobs. (See Böckerman and Ilmakunnas [2009] as a recent example that provides a listing of others.) However, it is not clear that any of these have investigated whether the responsiveness of quit rates to job dissatisfaction is any greater among married women than married men.

  25. These findings about significant gender differences in coefficients could be misleading, if there is multicollinearity among the X variables. That is, significant gender differences in other coefficients could be masked by multicollinearity among X variables. Calculations of variance inflationary factors (VIFs) for these variables do not seem to indicate severe multicollinearity, however. For married women, the highest VIFs for the X variables in the job satisfaction equation are under 3.5 (trade 2.5, service 3.4, ln hours 2.5 and full-time status 2.4). For married men, the highest VIFs are a bit higher (white collar 5.5, blue collar 5.6, lambda 3.5 and age 2.8, with the latter two occurring in the model with sample selection correction only). For unmarried workers, primarily the VIFs for types of occupations or industries exceed 2.5 (men white collar 3.2, men blue collar 3.6, women trade 4.0, women service 5.1). Additionally, for unmarried women in the model with sample selection correction, the VIF for lambda is 2.9.

  26. These findings could also be misleading, if there is misspecification in the model. Misspecification of the model would be problematic in this regard, in that the estimated coefficients of identified variables (e.g. FT) could be biased because of some omitted relevant variable(s) correlated with them. If any such omitted variables differ in importance to the job satisfaction of men and women, then the extent of the bias would vary by gender and the differing effects of these omitted variables, rather than the identified variables, could in fact be the source of gender differences in job satisfaction. This comment applies to the discussion that follows as well.

  27. Notably, within the unmarried group there are significant gender differences in household composition: number of children (averaging 0.63 for women but only 0.24 for men) and presence of children under the age of six years (for 13 percent of women but only for 6 percent of men). The direction of these differences is exactly the opposite in the case of married workers; in comparison to married men who work, married women who work have fewer children on average (0.75 as opposed to 0.9) and are less likely to have children under the age of six years (16 percent as opposed to 25 percent). As a result of different life circumstances, unmarried women who work may feel less satisfaction from their jobs than unmarried men who work, in a manner unrelated to gender differences in the X's being used here or in the β̂′s associated with those X's. The finding of a significant positive coefficient for MALE is not altered by simply including measurements of household composition (number of children and presence of children under the age of six) among the X's in the job satisfaction equation for unmarried workers, however. The results of such an effort are not shown in Table 6, but are available upon request.


  • Akerlof, George A., Andrew K. Rose, and Janet L. Yellen . 1988. Job Switching and Job Satisfaction in the US Labor Market. Brookings Papers on Economic Activity, 2 (2): 495–582.

    Article  Google Scholar 

  • Bartel, Ann P. 1981. Race Differences in Job Satisfaction: A Reappraisal. Journal of Human Resources, 16 (2): 294–303.

    Article  Google Scholar 

  • Bender, Keith A., Susan M. Donohue, and John S. Heywood . 2005. Job Satisfaction and Gender Segregation. Oxford Economic Papers, 57 (3): 479–496.

    Article  Google Scholar 

  • Berg, Nathan, and Donald Lien . 2002. Measuring the Effect of Sexual Orientation on Income: Evidence of Discrimination? Contemporary Economic Policy, 20 (4): 394–414.

    Article  Google Scholar 

  • Böckerman, Petri, and Pekka Ilmakunnas . 2009. Job Disamenities, Job Satisfaction, Quit Intentions, and Actual Separations: Putting the Pieces Together. Industrial Relations, 48 (1): 73–96.

    Google Scholar 

  • Borjas, George J. 1979. Job Satisfaction, Wages and Unions. Journal of Human Resources, 14 (1): 21–40.

    Article  Google Scholar 

  • Budig, Michelle J., and Paula England . 2001. Wage Penalty for Motherhood. American Sociological Review, 66 (2): 204–225.

    Article  Google Scholar 

  • Christofides, Louis, Qi Li, Zhenjuan Liu, and Insik Min . 2003. Recent Two-Stage Sample Selection Procedures with an Application to the Gender Wage Gap. Journal of Business & Economic Statistics, 21 (3): 396–405.

    Article  Google Scholar 

  • Clark, Andrew E. 1997. Job Satisfaction and Gender: Why Are Women So Happy at Work? Labour Economics, 4 (4): 341–372.

    Article  Google Scholar 

  • Clark, Andrew E. 2001. What Really Matters in a Job? Hedonic Measurement Using Quit Data. Labour Economics, 8 (2): 223–242.

    Article  Google Scholar 

  • Clark, Andrew E., and Andrew J. Oswald . 1996. Satisfaction and Comparison Income. Journal of Public Economics, 61 (3): 359–381.

    Article  Google Scholar 

  • Clegg, Chris W. 1983. Psychology of Employee Lateness, Absence, and Turnover: A Methodological Critique and an Empirical Study. Journal of Applied Psychology, 68 (1): 88–101.

    Article  Google Scholar 

  • Davis, James A., and Tom W. Smith . 2009. General Social Surveys, 1972–2008, [machine-readable data file] /Principal Investigator, James A. Davis; Director and Co-Principal Investigator, Tom W. Smith; Co-Principal Investigator, Peter V. Marsden; Sponsored by National Science Foundation. — NORC ed. — Chicago: National Opinion Research Center [producer]; Storrs, CT: The Roper Center for Public Opinion Research, University of Connecticut [distributor], 1 data file (53,043 logical records)+1 codebook (2,656 pp.). — (National Data Program for the Social Sciences, no. 18).

  • Dolton, Peter J., and Gerald H. Makepeace . 1987. Marital Status, Child Rearing and Earnings Differentials in the Graduate Labor Market. Economic Journal, 97 (388): 897–922.

    Article  Google Scholar 

  • Donohue, Susan M., and John S. Heywood . 2004. Job Satisfaction and Gender: An Expanded Specification from the NLSY. International Journal of Manpower, 25 (2): 211–234.

    Article  Google Scholar 

  • Flanagan, Robert J., George Strauss, and Lloyd Ulman . 1974. Worker Discontent and Work Place Behavior. Industrial Relations, 13 (2): 101–123.

    Google Scholar 

  • Freeman, Richard B. 1978. Job Satisfaction as an Economic Variable. American Economic Review, 68 (2): 135–141.

    Google Scholar 

  • Frey, Bruno S., and Alois Stutzer . 2002. What Can Economists Learn from Happiness Research? Journal of Economic Literature, 40 (2): 402–435.

    Article  Google Scholar 

  • Hamermesh, Daniel S. 2001. The Changing Distribution of Job Satisfaction. Journal of Human Resources, 36 (1): 1–30.

    Article  Google Scholar 

  • Heckman, James 1976. The Common Structure of Statistical Models of Truncation, Sample Selection and Limited Dependent Variables and a Simple Estimator for Such Models. Annals of Economic and Social Measurement, 5 (4): 475–492.

    Google Scholar 

  • Kristensen, Nicolai, and Edvard Johansson . 2008. New Evidence on Cross-Country Differences in Job Satisfaction Using Anchoring Vignettes. Labour Economics, 15 (1): 96–117.

    Article  Google Scholar 

  • Mangione, Thomas W., and Robert P. Quinn . 1975. Job Satisfaction, Counterproductive Behavior, and Drug Use at Work. Journal of Applied Psychology, 60 (1): 114–116.

    Article  Google Scholar 

  • Sloane, Peter J., and Hector Williams . 2000. Job Satisfaction, Comparison Earnings, and Gender. Labour, 14 (3): 473–502.

    Article  Google Scholar 

  • Sousa-Poza, Alfonso, and Andrés A. Sousa-Poza . 2000. Taking Another Look at the Gender/Job-Satisfaction Paradox. KYKLOS, 53 (2): 135–152.

    Article  Google Scholar 

  • Vroom, Victor H. 1964. Work and Motivation. New York: Wiley.

    Google Scholar 

  • Wright, Robert, and John F. Ermisch . 1991. Gender Discrimination in the British Labor Market: A Reassessment. Economic Journal, 101 (406): 508–522.

    Article  Google Scholar 

Download references

Author information

Authors and Affiliations




Tables A1, A2 and A3

Table A1 Definitions of additional variables
Table A2 Summary statistics of variables for first-stage probit, by marital status and gender
Table A3 Probit results: Parameters of index function for being at work

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Carleton, C., Clain, S. Women, Men, and Job Satisfaction. Eastern Econ J 38, 331–355 (2012).

Download citation

  • Published:

  • Issue Date:

  • DOI:


  • job satisfaction
  • gender differences

JEL Classifications

  • J16
  • J28