Educational mismatch and retirement

ABSTRACT

Using a panel data set of scientists in the US, we examine the hypothesis that workers in jobs poorly matched to their education are more likely to retire. In pooled estimates, we confirm that the mismatched are more likely to retire and that among retirees, the mismatched retire at younger ages. Hazard function estimates also support the hypothesis. Workers with longer accumulated periods of mismatch are significantly more likely to retire in discrete-time duration models that account for both reasonable controls and worker heterogeneity. These findings suggest that educational mismatch and its consequences are concentrated among late-career employees.

KEYWORDS:

• Educational mismatch
• retirement
• scientists
• duration models

JEL CLASSIFICATION:

• J24
• J26

Acknowledgements

The authors thank the US NSF for access to the restricted data. However, the use of the NSF data does not imply NSF endorsement of the research methods or conclusions contained in this paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. See McGuinness and Wooden (2009) for evidence that job changes typically do not improve match quality.

2. In examining Estonia, Lamo and Messina (2010) also present evidence that both the incidence and wage penalty of mismatch increase with age.

3. As 1993 is the first year of our data, the workers we study do not face mandatory retirement provisions. Although such provisions were prevalent in the USA prior to 1988, they have since been typically outlawed as age discrimination. Thus, although public or private pensions may generate incentives to retire, mandatory retirement provisions in contracts do not drive our results. While the matched and mismatched would, in theory, face the same set of incentives to retire from any pension provisions, the SDR does not contain any information on pension characteristics or even provision and so we cannot test for such pension incentive effects in the analysis below.

4. Although there is a public-use version of the data available, we employ a restricted data version of the survey, since information such as salary, detailed discipline of education and job, and race are masked in the public-use version. More information about both versions of the survey can be found at the Scientists and Engineers Statistical Data System (SESTAT) website, or http://sestat.nsf.gov.

5. The methodology is implemented in this paper by using the ‘ pgmhaz8’ subroutine available in STATA (Jenkins 2004).

6. We note that the controls behave roughly as anticipated with age, female gender, divorce, disability, hard science and engineering degrees (relative to social science degrees), years since degree, employment in business or government (relative to academia), and main activity being management or teaching (compared to research) emerging as significant positive determinants and black or Hispanic (relative to white) race/ethnicity, never married, non-native citizenship, and supervising others emerging as significant negative determinants. The full estimations are available upon request.

7. As the anonymous referees pointed out, such a response may also fit with a ‘linked lives’ perspective in which younger female partners of potentially retired men have lower tolerance for disadvantageous job situations.

8. Conflicting income and substitution effects make the influence of earnings on retirement ambiguous. For the retirement probability regressions, the substitution effect seems dominant, and those with higher earnings (and thus a higher opportunity cost to leisure) delay retirement. But among those who have retired, the income effect dominates, since leisure in retirement is likely a normal good, and so retirement happens at earlier ages for those with higher income. This income effect will also be supplemented by the higher pension wealth that is surely accumulated by those with higher income levels, giving further incentive to retire at earlier ages. 

9. We thank a reviewer for this suggestion.

10. Including log salary in the regressions does not qualitatively affect the results in any significant way. These results are available from the authors.

11. An alternative specification could use just the most recent periods. Indeed, workers experiencing severe mismatch anytime in the previous two waves have statistically significant higher odds of retiring compared to those who had no mismatch over the previous two waves. Full results of these estimations are available upon request.

12. While earnings and job tenure have both been shown to be influenced by mismatch, as a robustness check we included these variables in the hazard estimates with no qualitative change in the results. These are available upon request.