Heaven knows I'm miserable now: overeducation and reduced life satisfaction

Abstract

Recently the supply of young graduates entering the UK labour market has undergone a sharp increase. A possible consequence of this is an increase in the number of individuals who are overeducated for the jobs that they do subsequent to participating in higher education. Using British panel data and dynamic panel analysis, I demonstrate that overeducation amongst the young has increased, and that the overeducated are less satisfied with life than their peers who are not overeducated. This result appears to fade over time, with more recently overeducated individuals being both more common and less dissatisfied with life, consistent with the notion that relative effects and comparisons are important for life satisfaction.

Keywords:

• life satisfaction
• happiness
• overeducation
• education
• dynamic panel analysis
• BHPS

Acknowledgements

For comments on earlier drafts, helpful discussions and advice I am grateful to Nick Adnett, Andrew Clark, and Geoff Pugh. I am grateful to seminar participants at Staffordshire University and Universität Flensburg, and participants at the 2012 IWAEE event in Catanzaro, Italy, for valuable suggestions. I am also grateful to the editor of the Journal, Colin Green, and two anonymous referees who provided very useful comments and suggestions. The usual disclaimer applies.

Notes

1. This raises the additional possibility of promotion being a pathway out of overeducation. Due to the construction of the relative overeducation variable it is unlikely to be captured in this study or with the data used in this study, but it is a potential source of bias in capturing the happiness impact of overeducation. This is returned to in the methodological discussion. I thank Geoff Pugh for this suggestion and valuable discussions. This notion of a potential ‘settling in’ period is also discussed by Dolton and Vignoles (1997).

2. A further dummy variable was created which measures relative overeducation by both occupation and gender combined: on the basis that males may compare themselves primarily with other males and females with other females. In practice, the results from this addition are qualitatively the same as those for the dummies mentioned above, and as such are not discussed further.

3. The actual amount of years is quite varied dependent upon the reference group.

4. Overeducation is more prevalent than unemployment (not shown), affecting between 1.5 and 3 times as many individuals in this sample, depending upon how overeducation is measured.

5. This is particularly the case with this study due to its restricted age range focus.

6. This is often untested in the happiness economics literature, and suggests that some of the static estimates in this area are potentially misspecified.

7. Calculating long-run coefficients is straightforward and undertaken in the results section. A recent paper that uses this dynamic panel method in an education context is by Pugh, Mangan, and Gray (2011), a paper which was formally commended at the 2012 British Educational Research Association.

8. For the initial conditions test, the key diagnostic is the difference-in-Hansen test, and no evidence of an initial conditions problem is found. This is to be expected somewhat with happiness data, because one year's values have only a limited influence in the values of other years. The initial conditions – the relationship between the unobserved fixed effects and the observables at the time of the start of the panel subset employed – may almost never be particularly important for well-being data (assuming the choice of exogeneity or endogeneity of the regressors has been carefully considered).

9. The two fewer waves that can be employed with this created reference group to measure overeducation for data availability reasons explain the lower number of observations used for the estimations when compared to the number used in Table 1.

10. Tests for the joint significance of overeducation and the interaction term demonstrated that they are jointly significant when they are individually significant in each of the estimates here.

11. I thank Ilona Ebbers for this suggestion.