Higher Education Peer Effects and Contextual Admission Policy

点击翻译按钮获取中文摘要

This builds on my LEO HE research on the causal effect of "courses" on student financial outcomes. This figure shows our recent estimates of the (average) earnings differentials of graduates (relative to the reference subject - History) at age 30, controlling for pre-HE educational attainment. There is wide variation across subjects in these, and the earnings of the lowest subjects are very close to non-graduates in the data. The weakness of the research is that, while it controls for the prior achievement of the student, it does not control for the prior achievement of fellow students - i.e. it ignores peer effects. The estimation of peer effects suffers from "the reflection problem" whereby the average performance of a group influences the performance of the individuals that comprise that group (Manski, REStuds 1993). The solution to this problem is to have an idea of how students' reference groups form - operationalised in the "peers of peers" method (Angrist Labour Economics, 2014). I used this method to model peer effects in English secondary schools (using the NPD component of LEO) in Mendolia et al (OxEconPapers 2018) - it relied on knowing which secondary school students had attended which primary schools. We can use this idea in this HE context since LEO tells us who went to which HEI to do what, and how well they did in terms of subsequent earnings. And we know, who each graduate's HE peers were and which school they attended at 16-18, and how well they did. This would be the first ever application of the peers of peers method to HE students a would form the student's "job-market" paper. Thanks to the Office for Students, we know HE institutional "access and participation plans". Over 25% of HEIs practice contextual admission (CA). GRADE provides UCAS information, important for modelling the counterfactual, on which student went where to do what, it also shows where they also applied to go, but didn't. CA affects who goes where and changes the peers one gets, which changes one's outcomes. This second paper would be a significant (co-authored) extension that would tell us how much labour market "value" a course added, independently of the quality of the mix of peers. Armed with this we could simulate the effects of any amount of CA allowing for its complex spillover effects. This is challenging problem and "set identification" (ie estimating bounds on causal effects) is the state-of-the-art literature to tackle it. This final paper would be a tool that would power impact.