I happened to read this paper over the past few days:
Learning about academic ability and the college drop-out decisionTodd StinebricknerThe University of Western OntarioRalph StinebricknerBerea CollegeAbstract: We use unique data to examine how college students from low income families form expectations about grade performance/academic ability and to examine the role that learning about these factors plays in the college drop-out decision. With respect to beliefs, at college entrance students are considerably overoptimistic. Subsequent updates depend on both initial beliefs and new information arriving in the form of grades, with individual heterogeneity in the weights assigned to initial beliefs and grades depending on individual-specific views about the underlying reasons for grade performance that are suggested to be of importance by a Bayesian model. While students with the worst grade performance update dramatically, they tend to remain overoptimistic because they understate the importance of permanent factors in determining their performance. With respect to drop-out, we find that learning about grade performance/ability plays a very prominent role; our predictions suggest that drop-out would be reduced by 41% if no learning occurred about these factors. Finally, the paper contributes directly to a recent literature examining gender differences in educational attainment and gender differences in other behavior. We find that the substantial gender difference in drop-out in our sample is predicted almost entirely by academic differences (1st year grades and beliefs about future grades), and we find direct evidence of gender differences in effort and gender differences in the non-pecuniary cost of effort. As to why poorly performing males decided to enter college, we find that, at entrance, males are substantially more overoptimistic than females.
The paper relies on the unique data from the Berea Panel Survey that the two Stinebrickner's designed and continue to field. This work, and a string of other papers using the same data, really illustrate the value of thoughtful data collection that is well-informed by the recent literature on measurement - particularly expectations measurement in this case, by the relevant theoretical literature, and by specific questions of academic and policy interest.
The results of the study suggest that many students enter college with remarkably misguided ideas about their own ability and about the workload required to succeed in college. It seems to me that high schools ought to be able to provide this information relatively cheaply and that doing so would be of great social benefit. While some learning at college is both inevitable and optimal, four year colleges in particular are very expensive for society to provide and having large numbers of students attend for only one or two years in order to learn that they are poorly suited for college-level work is a serious waste of resources.
Full disclosure: Todd is a friend and former colleague.