Relevance of education to occupation: a new empirical approach based on college courses

ABSTRACT

We introduce a new approach to measuring the match between education and occupation by using the number of college courses related to one's occupation. Previous studies have only considered the match between college ‘major’ and occupation. That approach ignores the content of education and the courses taken in college. We find that taking courses in college that are relevant to one's occupation is significantly associated with higher wages, which can be taken as evidence against the notion that returns to college are principally a matter of signaling. A student's wage increases, on average, by 1.5–2.1% for each matched course.

KEYWORDS:

• Returns to education
• skill match
• horizontal mismatch
• return to college courses

JEL CLASSIFICATION:

• J24
• I26
• J62

Acknowledgments

We would like to thank Judith Scott-Clayton, Wim Vijverberg, John Robst, Randy Reback, Basit Zafar, John Winters, Homa Zarghamee, Dan Thomas, Ben Zweig, Jasper Thomas, and participants at LERA and EEA for their insightful comments. The data and the code used in this paper may be found here: https://github.com/ahdvnd/Papers/tree/master/Mismatching.

Disclosure statement

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

Notes

1. In Sweden, higher vocational education (HVE) is an alternative to academic track of higher education and is post-secondary education that combines theoretical and practical studies in cooperation with employers and industry.

2. Respondents in NLSY97 were asked to take this test in 1997, and 78% of the respondents took the test. The NLSY79 respondents were administered the ASVAB for the purpose of constructing new national measures for the aptitudes of the American youth. The NLSY97 population was selected since it was a nationally representative sample of youth. The ASVAB consists of ten tests that measure knowledge and skill in areas such as general science, arithmetic reasoning, and word knowledge. Test scores in each section are used to build composite scores. ASVAB has been used in many social science studies but it has also been criticized for having large non-normal estimation errors (Schofield 2014). Another issue with the ASVAB score is that individuals in the sample took the test at different ages. Since test scores are likely to be associated with ages, they may not purely reflect individuals' ‘ability’.

3. We should note that of the individuals for whom transcript data was obtained, relevant data may be missing for some. During the PSTRAN study, multiple transcript requests were sent out if individuals had attended more than one undergraduate institution. In some cases, not all transcripts were obtained for reasons of non-response or refusal by the institution to release the transcripts.

4. For instance, in 2013, annual wages are reported for 2012 whereas occupations are reported for 2013.

5. See https://nces.ed.gov/ipeds/cipcode/resources.aspx?y=55 for the Guidelines for using the CIP to SOC Crosswalk.

6. All of the crosswalks that we use in this paper can be found here: https://github.com/ahdvnd/Papers/tree/master/Mismatching

7. We initially considered calculating the number of courses matched as a share of all courses required for an occupation as our matching score. However, this method would lead to the same matching score for a person who has taken 1 out of 2 required courses and a person who has taken 20 out of 40 required courses.

8. Robst (2007b) finds that women were somewhat more likely to report that they are matched to their occupation. 58 percent of women versus 53 percent of men in his sample reported that they are matched to their jobs. He argues, while men are more likely to be mismatched due to career-enhancing opportunities like promotions and changes in career interests, women are more likely to report working conditions and working constraints as their reasons for being mismatched.

9. There is no need for inclusion of a nonlinear term for experience in our model since there are only small variations in ages across our sample.

10. For years before 2008, the number of observations significantly drops since many individuals in the sample are still in school and not working and as a result do not have a CMQ. The last round of the PSTRAN study was done in 2013 and therefore we had to exclude NLSY data after 2013.

11. We thank one of the anonymous reviewers for this point.

12. We use mother's education due to the higher number of observations for this variable compared to father's education.

13. Another question is whether the effect of matching varies by major. In order to do this, we interacted major dummies with CMQ. The result is not shown here since we found no statistically significant interaction coefficient but are available upon request.

14. Note that in our cohort-based panel analysis, variable time is measured in years passed since graduation and not calendar years.

15. Furthermore, we investigate whether CMQ effect on wages changes over time, we included interaction of CMQ with dummy variables indicating years passed since graduation, and did not find any significant changes in the CMQ coefficient over the span of around 10 years that we observe. The results are not reported here but are available upon request.