The heterogeneous impacts of business cycles on educational attainment

ABSTRACT

This study examines the impact of fluctuations in the unemployment rate before high school graduation on educational attainment measured 30 years later. I find evidence that important heterogeneity is masked by estimating average effects across the ability distribution. Using data from the 1979 National Longitudinal Survey of Youth, this analysis identifies individuals who are on the boundary between pursuing and not pursuing additional education. Exposure to a higher unemployment rate at age 17 is associated with higher educational attainment for men in the 60–80th quintile of the ability distribution. There is no evidence of an effect beyond this quintile.

KEYWORDS:

• Education
• business cycles
• unemployment rate
• countercyclical
• AFQT
• NLSY79
• opportunity cost

JEL CLASSIFICATION:

• I20
• I22
• J10
• J18
• J24

Acknowledgments

I thank Hani Mansour, Benjamin Hansen, Daniel Rees, Peter Kuhn, Kelly Bedard, Susan Averett, and the seminar participants at UC Santa Barbara for their helpful comments and support. National Longitudinal Survey of Youth 1979 data and Geocode information are provided by the Bureau of Labor Statistics. All errors and omissions are my own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 For more evidence of countercyclicality see Corman (1983), Kane (1994), Berger and Kostal (2002), and Barr and Turner (2013). Other studies find that the relationship can differ by gender, undergraduate major, degree level, race, and country of origin (Bedard and Herman 2008; Black and Sufi 2002; Card and Lemieux 2000; Sakellaris and Spilimbergo 2000).

2 Using sibling pairs, Altonji and Dunn (1996) explore heterogenous returns to education by parental educational attainment.

3 These findings are relevant to the literature investigating the impact of unfavorable economic conditions at job market entry on long-term earnings (Brunner and Kuhn 2014; Genda, Kondo, and Ohta 2010; Kahn 2010; Kwon, Milgrom, and Hwang 2010; Oreopoulos, von Wachter, and Heisz 2012; Oyer 2006, 2008; Stevens 2008) and to the literature on the health impacts of recessions (Ruhm 2000, 2015).

4 For more information about the NLSY79 and the NLSY79 state identifiers, see appendix sections 1.1 and 1.2 (see online supplemental file at 10.1080/09645292.2017.1336511).

5 Appendix Table A lists the range of the unemployment rate in each state.

6 Degree attainment is an alternative measure (Belman and Heywood 1991; Hungerford and Solon 1987). Unfortunately, the NLSY79 is ill-suited to investigate this outcome. If I construct and use a measure of ‘highest degree ever attained’ looking across all years of data, samples sizes shrink by 18–40%. Nonetheless, using this alternative measure of educational attainment gives results consistent with those based on completed years of education and are discussed later on.

7 Eight years of data is made available by the Inter-University Consortium for Political and Social Research. The final year is available from Cornell University's Institute for Social and Economic Research.

8 Minimum wage data is from the U.S. Department of Labor. It is not clear whether 17 year olds observe real or nominal rates, but this is not a concern in specifications with cohort fixed-effects. For details on how the average manufacturing wage is calculated in the Current Population Survey, see Appendix section 1.4.

9 More details are provided in Appendix section 1.5 (see online supplemental file at 10.1080/09645292.2017.1336511).

10 Average state Pell grant award is not available in 1974. Results are robust to the exclusion of average state Pell grant award.

11 Using a similar fixed-effects empirical strategy, Card and Lemieux (2000) find that a 1 point increase in the overall unemployment rate at age 17 leads to a .008 increase in years of completed education. This is very close to my estimate of .01.

12 Including a region-specific time trend, the coefficient is still .103, but is no longer statistically significant. Coefficients in the first three quintiles remain small and insignificant. The top quintile coefficient becomes more negative and significant.

13 Using AFQT deciles corroborates the results in Table 3. Appendix Table B indicates that the coefficient estimated in the 60–80th quintile is coming from 70-80th decile.