Recent studies have found a large earnings premium to attending a more selective college, but the mechanisms underlying this premium have received little attention and remain unclear. In order to shed light on this question, I develop a multidimensional signaling model relying on college grades and selectivity that rationalizes students' choices of effort and firms' wage-setting behavior. The model is then used to produce predictions of how the interaction of the signals should be related to

more »

... es, namely that the return on college GPA should fall the more selective the institution attended. Using five data sets that span the early 1960s through the late 2000s, I show that the data support the predictions of the signaling model, with support growing stronger over time as college sorting by ability has increased. The findings imply that return to college selectivity depends on GPA, something previously not recognized in the literature, and they can rationalize why employers learn more quickly about college graduates' productivity than less educated workers'. JEL Classification Codes: I20, I21, J31 for data on college selectivity. All errors are my own. This paper makes two substantive contributions toward understanding the college selectivity premium. First, it develops a novel, multidimensional signaling model of ability between college graduate workers and prospective employers. In equilibrium, the utility-maximizing behavior of these agents leads to a specific-and empirically testable-relationship between the two dimensions of the signal, college selectivity and grade point average (GPA), and starting wages. While the full model is elaborate, the crux is intuitive. Students sort into different colleges by ability, which means that college selectivity is a valuable signal of ability to employers. If graduating from a more selective school sends a more precise signal of ability than graduating from a less selective school, the marginal informational benefit of an additional signal, such as GPA, is reduced. When it comes to wage setting, we would expect the relative weight that firms place on the GPA signal to be lower at more selective colleges. Consequently, the change in log wages with respect to a change in GPA should be smaller when selectivity is higher. Furthermore, the ability sorting across college types also implies that the selectivity premium should fall as GPA rises. The intuition here is that high-GPA students benefit less from attending a selective school because they have demonstrated their ability through their GPA; but for a lower-GPA student at a selective college, firms will discount the noisier signal and place more weight on the college type. Second, the paper empirically tests the implications of the model. Employing five nationally representative data sets that span five decades, I consistently find strong support for the predictions of the signaling model. The return on GPA is lower at selective colleges and falls as the threshold of selectivity rises. The selectivity premium is highest for those with lower GPAs and declines as GPA rises. Moreover, both of these phenomena have become more pronounced over time as ability sorting across colleges has increased. The paper proceeds as follows. In the next section, I review some of the recent literature on the returns to college selectivity and employers learning about workers. Sections 3 and 4 develop, characterize solutions, and derive predictions for a multidimensional signaling model in the context of college graduate workers whose productivity firms cannot perfectly observe. Section 5 describes the data sets and empirical methodology that are used to explore and test the implications of the model, while Section 6 presents the results of these tests. Section 7 concludes.