Panel Paper: Ova and Out: Using Twins to Estimate the Educational Returns to Attending a Selective College

Thursday, November 8, 2012 : 3:40 PM
Carroll (Sheraton Baltimore City Center Hotel)

*Names in bold indicate Presenter

Jonathan Smith, The College Board

Research shows that there is a strong positive relationship between college selectivity and graduation rates (Bowen and Bok 1998; Kane 1998; Alon and Tienda 2005; Horn and Carroll 2006; Long 2008; and Bowen et al. 2009).  Most papers condition on observable family characteristics, such as parents’ incomes and educations, to overcome selection bias.  However, there are still many unobservable family characteristics that can drive a student’s decision on where to enroll in college and the likelihood of graduating.  This paper overcomes the selection bias related to family background by using a new dataset of twins to identify how four-year college selectivity affects the probability of graduation.

Using College Board data, I identify approximately 17,000 sets of twins by matching students in the same high school, with the same last name, address, and date of birth.  This dwarfs previous studies that utilize twins, which typically have several hundred pairs.  The data is merged with National Student Clearinghouse data so we observe which colleges students enroll and whether they graduate.

I use a within-twin estimator to find the impact of college selectivity, as measured by SAT scores of enrollees, on graduation rates, without the bias of family background.  I also include person-specific controls that account for individual endowment (SAT) and motivation (high school GPA and number of Advanced Placement tests).  I find that every 100 average SAT points across a college’s matriculates increases the probability of a student graduating by 3.3 percentage points.  This estimate is 39% smaller than a regression that only controls for standard family background measures.  The effect is strongest between colleges whose average SAT scores are below 1100 but still differ by at least 100 points or between scores below 1100 and anywhere above 1100.  There is little effect between two colleges whose SATs are both above 1100 but still differ by at least 100 points.  I also find that the effect is stronger for male-male twins and male-female twins, rather than female-female twins.

Correctly estimating educational returns to college selectivity is important because it is well established that low-income students and under-represented minorities are less likely to apply and enroll in selective institutions (Manski and Wise 1983; Pallais and Turner 2006; Hill and Winston 2010).  Moreover, attending a more selective college and more years of schooling increases wages (James et al. 1989; Loury and Garman 1995; Ashenfelter and Krueger 1994; Behrman et al. 1996; Black and Smith 2001; Bowen and Bok 1998; Hoxby 1998; Kane 1998; Brewer et al. 1999; Monks 2000).  Therefore, education policy makers and students should fully consider the costs and benefits of selecting a college.

Full Paper: