![Sample our Education journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5932/TOC-Subject-Sample-2021-Education.jpg"})
ABSTRACT
This paper estimates the monetary return to quality in US graduate education, controlling for cognitive ability and self-selection across award level, program quality, and field-of-study. In most program types, I cannot reject the hypothesis of no returns to either degree completion or program quality. Important exceptions include master's programs in health science, where completion substantially increases earnings, and in MBA and professional degree programs, where program quality has a positive influence on earnings. I explore the job characteristics that predict greater earnings among students with tertiary education, and I estimate the returns to quality in terms of non-monetary job benefits.
KEYWORDS:
Disclosure statement
No potential conflict of interest was reported by the author.
Notes
1. For example, it is not a priori clear how a high-ranked master's program in Public Administration compares to a low-ranked JD program.
2. Because this is a survey of UG students in American schools, it excludes almost all students who come to the US exclusively for graduate education. The graduate student data I use is representative only of students who earn a US bachelor's degree and then enroll in a US school for further education. This is population is largely, but not entirely, made up of US citizens.
3. Programs are typically left unranked because the program is so small or unknown at a national level that the survey cannot gather sufficient data to evaluate its quality. Many small master's programs in the field of education that focus on training teachers for local districts, for example, are unranked.
4. There are only 35 such observations, in the fields of Philosophy, Design, and Communications Technology.
5. That is, it describes the top 25% of students, ranked by their graduate program's quality, and so on.
6. In the bottom half of the quality distribution, mean ability is below zero, the average score of all students who take the SAT or ACT exams, whether or not they attend or complete UG education. This difference is not, however, statistically significant.
7. We cannot, however, reject equivalence of ability between terminal bachelor's students and students in unranked master's of humanities and social science programs, unranked doctoral programs, and a few cells in the health sciences.
8. Notice that at the median, the field of education is comparable to the humanities and social sciences. Presumably teacher salary schedules put a cap on income dispersion at the high end of the income distribution in this field.
9. The one other program group whose quality data has low coverage is the humanities. There are many programs in the arts (performing, studio, or literary) that are not ranked.
10. A growing literature models the choice of field as a matter of learning. This learning can take place in terms of updating a student's expected returns to a major (Arcidiacono, Hotz, and Kang Citation2012) or in terms of updating a student's information about their aptitude in one field or another (Stinebrickner and Stinebrickner Citation2014). The B&B sample does not have the frequency, nor the expectations-elicitation questions, required to model choice at this level.
11. In sensitivity analysis, I will also model choice j as across degree and quality (but not field). In all cases, the comparison group, for whom all Equation (Equation2(2) ) parameters are normalized to zero, is individuals who do not earn a graduate degree.
12. Included regressors were student expectations over income, average net tuition, program length, and selectivity. Preferences over these choice-level characteristics were allowed to vary systematically with the observable individual-level characteristics of ability and gender. The step-2 income equation results are not substantially changed when using probabilities predicted by this method rather than the multinomial logit.
13. This expectation is calculated by regressing 2002 income on ability and broad field of UG study, allowing for full flexibility in parameters according to race and gender. This regression is run including only those students who never attended graduate education. Predicted income is then calculated for all students.
14. The increased likelihood of professional degree completion is consistent with Schapiro, O'Malley, and Litten (Citation1991), but inconsistent with other findings (Millett Citation2003). The literature on the effect of UG debt on the likelihood of continuation to graduate education is generally inconclusive.
15. Because of the typical timing of educational enrollment and the very low rates of unemployment in this highly skilled cohort, the experience variable goes a long way towards controlling for the effects of entering the job market in a recession (see, e.g. Oreopoulos, von Wachter, and Heisz Citation2012). Introducing additional indicators for the year of labor market entry does not change any results.
16. It is also possible that the differences are due to changes over time in the value of each degree, or to cohort effects in the returns to each degree, since the B&B cohort of doctors is too young to appear in the SESTAT data.
17. In Dale and Krueger (Citation2002), the results are more mixed. Correcting for selection sometimes does not change their estimates of the returns to quality relative to the OLS benchmark, as when we measure UG quality with mean student SAT scores, or Barron's rankings. In many other specifications (which are the main results of their paper), controlling for selection causes the estimated return to quality to disappear.
18. The difference in ‘sticker prices’ between top and bottom ranked master's programs, ignoring financial aid, is around $7000 annually.
19. Wage is calculated as annual labor market income, divided by annual hours. Annual hours is calculated from the B&B's ‘typical weekly hours worked’ variable, under the assumption of a 50-week work year.
20. The B&B only allows us to investigate the presence of non-wage benefits. It contains no data with respect to the generosity of these benefits.