Predictors of Student Success

The world of education is filled with policies which aim to improve student outcomes. While there are many factors which may potentially contribute to the academic success of a student, there are similarly a wide variety of measures which can be used to determine student success. This dissertation examines some specific policies and environments to which students are exposed and aims to determine their effects on those students. In doing so, the following chapters identify and explain particular predictors of student success.The first chapter, "Course Closed: The Short- and Long-Run Impacts of Course Shutouts on University Students", examines how students are affected when they are unable to register for a university course. For a variety of reasons, demand for seats in some college courses can exceed the available supply. At the same time, many universities employ policies which allow some students to register for classes before others, creating a situation in which some students may find classes to be full when it is their time to register. In this chapter, I study how these "course shutouts" affect university students. Utilizing data from a university in which students are assigned to registration times in a quasi-random order, I use these registration times as an instrument to determine the causal effect of course shutouts on a variety of student outcomes. I show that, within a given term, students who are forced to register later experience more course shutouts and that these shutouts cause students to both attempt and earn fewer units. Students who get shutout of classes are more likely to end up in classes that start before 10 AM and are taught by instructors with historically low pass rates. All of these effects are particularly strong for students in their first two years at the university. Surprisingly, I show that students' cumulative number of shutouts across their academic career is not predictive of time to graduation, major changes, or dropout rates. However, students who accumulate more shutouts in their first few years at the university do exhibit higher rates of summer school enrollment, suggesting that they may respond to the adverse effects of shutouts by making up units in the summers. I also find this effect to be strongest for those students who show up to the university without any degree applicable units from community college courses or advanced placement exams. All together these results suggest that policies which routinely place particular students at the end of the registration period – specifically first year students and those without incoming, degree applicable units – may increase the cost of graduation for those students by increasing their likelihood of attending summer school.The second chapter, "Non-Linear and Heterogeneous Effects of Peer Gender Composition on Academic Performance", explores the ways in which students' peer groups can influence academic performance. A long-standing literature has shown the gender composition of a student's peer group to be a relevant predictor of his or her performance in school. Most specifically, female students perform best when more of their peer group is female, with mixed results for males. Yet while some work has been done to establish a linear relationship between peer group gender composition and academic performance, little is known about the relevant non-linearities and heterogeneity of this relationship. Moreover, as students often non-randomly sort into schools and classrooms, plausible exogenous variation in peer gender composition is rare. To address these two points I utilize data from a randomized control trial in Duflo, et al. (2011) to explore the potential non-linearities and heterogeneities of the effects of peer gender composition alluded to in previous work. Results from this chapter show that students perform best when they are placed in a classroom with a fairly balanced gender composition, as opposed to one where students are predominantly of one gender. As part of the experiment, classrooms were randomly assigned one of two types of teachers: an existing, civil-service teacher or a new teacher on a one year contract who are shown to more consistently be in the classroom and teaching than their civil-service counterparts. I find that the effect of being in a more gender balanced classroom is strongest when students are taught by a contract teacher. Overall, these findings suggest that the effect of peer gender composition may not only be non-linear, but dependent on classroom environment.The third chapter, ``The Effects of Changing the Registration Policy at a Large Public University'', examines the effect of a change in the policy that determines students' assignment to registration times at a large public university. Prior to the policy change, the university would divide the student body into 12 registration groups based on the first three letters of their last name. Each term these 12 name groups would each be assigned to one of 12 registration periods, creating an ordering of students that was unrelated to their progress towards graduation. Following the policy change, the university began determining students' academic progress, a number from 0 to 100 based on the proportion of degree applicable units completed. Under this policy, students are placed in registration times in descending order of academic progress, with the students with the most progress registering first. My estimates suggest that this reordering of students led to fewer waitlists and shutouts – instances when students were unable to get into a class from a waitlist – for third and fourth year students. However, the subsequent increase in waitlists and shutouts for underclassmen outweighed the decrease for upperclassmen, leading to an aggregate increase in these measures. Ultimately, seniors saw no increase in their cumulative number of units attempted and earned while that of freshmen, sophomores, and behind-schedule juniors decreased relative to the previous policy