Heterogeneous returns to U.S. college selectivity and the value of graduate degree attainment
Publisher: Bank of Canada Ottawa
Subject: C30 | Labour markets | I21 | returns to education, heterogeneous treatment effect, selection, data combination
jel: jel:I21 | jel:C30
Existing studies on the returns to college selectivity have mixed results, mainly due to the difficulty of controlling for selection into more-selective colleges based on unobserved ability. Moreover, researchers have not considered graduate degree attainment in the ana... View more
1The New York Times, for example, has published articles targeted at parents and students with titles such as "Does it Matter Where You Go to College?" (Nov. 29, 2010), "Is Going to an Elite College Worth the Cost?" (Dec. 17, 2010), and "Do Elite Colleges Produce the Best-Paid Graduates?" (Jul. 20, 2009).
2Throughout this paper, I use the word "college" to mean a four-year undergraduate program in the United States, which is equivalent to a university undergraduate program in Canada.
3Oreopoulos and Salvanes (2011) review the non-pecuniary benefits of an additional year of schooling. A similar outcome may hold for the dimension of college selectivity. See also Haveman and Wolfe (1984) and Ge (2011) for non-market returns of education.
4Enrolment in graduate programs rose about 67 percent between 1985 and 2007, while undergraduate enrolment increased by about 47 percent. Total autumn enrolment in undergraduate programs increased from 10.6 million to 15.6 million from 1985 and 2007, while that in graduate programs increased from 1.4 million to 2.3 million (NCES (2009)).
5See Buchinsky (1994) and Brand and Xie (2010) for a discussion of the heterogeneous returns to schooling.
6In this paper, I consider mainly cognitive abilities, due to data limitations. See Heckman, Humphries, Urzua, and Veramendi (2011) for the effects of non-cognitive abilities on labour market outcomes.
7Heckman and Navarro (2007) discuss the semi-parametric identification of dynamic discrete choice and dynamic treatment effects using the factor structure model. Empirical applications of factor structure model of Carneiro, Hansen, and Heckman (2003) can be found in Cunha and Heckman (2008), Heckman, Stixrud, and Urzua (2006), Cooley, Navarro, and Takahashi (2009), Cunha, Heckman, and Schennach (2010), and Heckman, Humphries, Urzua, and Veramendi (2011).
8This is essentially mapping the information from multiple test scores into fewer dimensions. I call these unobserved abilities. The specification allows measurement errors to enter into the test score equation, so the estimates are robust to measurement errors by construction.
9See Ridder and Moffitt (2007) and Ichimura and Martinez-Sanchis (2005) for details about data combination.
10See Figure 5 of Choy and Cataldi. (2011) for the years until the enrollment for graduate and first-professional students (after Bachelor's degree; measured in 2007-2008), by degree program. Also, Krantz, Natale, and Krolik (2004) summarize the labour market conditions for the United States in 2003.