Maria Marchenko from the University of Manheim, Germany and the Higher School of Economics, Moscow yesterday presented a model to estimate the effect of an endogenous shock on future network performance. The presentation was held at HSE’s Center for Institutional Studies Research Seminar. The related paper “Endogenous Shocks in Social Networks: Effects of Students’ Exam Retakes on their Friends’ Future Performance” should be available at https://sites.google.com/site/mariavmarchenko/jmp.pdf.
The network consists of 1st and 2nd year undergrad students at the Nizhniy Novgorod branch of HSE. Students create links and ties to peers, while living in the same dorm and taking the same courses.
The endogenous shock is a retake of one of the network members, that is, that one of the students in such a network fails in an exam. HSE is highly selective, after three retakes a student will expulsed by default.
Now, the question is, to what extent, if at all, the retake of one of my friends affects my future performance, and the future performance of all my other friends?
Two major issues arise in such a setting. 1. The shock is highly endogenous. Proper instruments, IV, are required. Maria uses the individual characteristics of the friend of my friend as instruments.
- Estimation strategy. Maria uses a 2SLS approach. Probably oversimplifying her sophisticated model in a first step the dependent variable is the likelihood that I will fail in an exam, i.e. that I experience a retake. The residuals from this estimation are then taken for the second step. The depend variable is now the difference in my performance, in terms of average grading scores, between now (in the year the shock happens) and the next year. On the right hand side of the equation are an array of individual level characteristics, including tuition free place or not, higher education of parents, and high school exam and university entry scores; network characteristics, that is, and a term for correlated effects.
The effect of the shock on the network performance varies depending on the set of controls in the equation. But there is a negative effect; at maximum a retake of my friends will increase my future performance by .4 standard deviation, SD.
I very much like the basis idea of Maria’s work and the empirical approach. The crucial issue in studying peer effects is whether such an effect is physically, and logistically feasible, as Gigi Foster from UNSW’s Business School has highlighted in a presentation in the same seminar series roughly one year ago. In the case of students it is absolutely reasonable that there are potential spillover effects.
What I would like to see in a paper are some plots that demonstrate the predicted levels of the dependent variable “(change) in future performance” over the possible range of network characteristics, given an endogenous shock of retake; all other variables from the equation held at their mean value. This would also help to understand how robust the findings are.