The Use of Instrumental Variables in Peer Effects Models (original) (raw)
Related papers
Literature Review: Empirical Identification of Peer Effects
This paper is a literature review of peer effects in economics. It surveys peer effects in theory, field experiments, their econometric identification, and empirical results. The primary objective of this paper is to determine how to identify peer effects in empirical settings. I find that the foremost problem of identification is the reflection problem, which consists of distinguishing between exogenous and endogenous variables. I discuss econometric techniques that can be used to resolve the reflection problem and analyze a variety of contexts in which these techniques are applied. I examine a number of circumstances in which peer effects have been measured and isolate variables that affect their magnitude and behavior. I find that considering particular contexts and the types of interactions are important in understanding what techniques to apply in future research.
Inference on Peer Effects with Missing Peer Data: Evidence from Project STAR
SSRN Electronic Journal, 2000
This paper studies peer effects on student achievement among first graders randomly assigned to classrooms in Tennessee's Project STAR. The analysis uses previously unexploited pre-assignment achievement measures available for 60 percent of students. Data are not missing at random, making identification challenging. The paper develops new ways, given random assignment of individuals to classes, to identify peer effects without imposing other missing-data assumptions. Estimates suggest positive effects of mean peer lagged achievement on average. Allowing heterogeneous effects, evidence suggests lowerachieving students benefit more than higher-achieving students do from increases in peer mean. Further, the bias in a widely used, poorly understood peer-effects estimator is analyzed, implying that caution is warranted in interpreting many peer-effects estimates extant in the literature.
Endogenous peer effects: fact or fiction?
The Journal of Educational Research, 2015
ABSTRACT This study examines endogenous peer effects, which occur when a student’s behavior or outcome is a function of the behavior or outcome of his or her peer group. Endogenous peer effects have important implications for educational policies such as busing, school choice and tracking. In this study, we quantitatively review the literature on endogenous peer effects through the use of meta-analytic methods. We find a significant and positive endogenous peer effect. It appears to be a genuine empirical effect but is dependent on the measure of educational outcomes, the peer group, publication status and publication year.
Estimating peer effects in longitudinal dyadic data using instrumental variables
Biometrics, 2014
The identification of causal peer effects (also known as social contagion or induction) from observational data in social networks is challenged by two distinct sources of bias: latent homophily and unobserved confounding. In this paper, we investigate how causal peer effects of traits and behaviors can be identified using genes (or other structurally isomorphic variables) as instrumental variables (IV) in a large set of data generating models with homophily and confounding. We use directed acyclic graphs to represent these models and employ multiple IV strategies and report three main identification results. First, using a single fixed gene (or allele) as an IV will generally fail to identify peer effects if the gene affects past values of the treatment. Second, multiple fixed genes/alleles, or, more promisingly, time-varying gene expression, can identify peer effects if we instrument exclusion violations as well as the focal treatment. Third, we show that IV identification of peer...
Clean Evidence on Peer Effects
Journal of Labor Economics, 2006
While confounding factors typically jeopardize the possibility of using observational data to measure peer effects, field experiments offer the potential for obtaining clean evidence. In this paper we measure the output of subjects who were asked to stuff letters into envelopes, with a remuneration completely independent of output. We study two treatments. In the "pair" treatment two subjects work at the same time in the same room. Peer effects are possible in this situation and imply that outputs within pairs should be similar. In the "single" treatment, which serves as a control, subjects work alone in a room and peer effects are ruled out by design. Our main results are as follows: First, we find clear and unambiguous evidence for the existence of peer effects in the pair treatment. The standard deviations of output are significantly smaller within pairs than between pairs. Second, average output in the pair treatment largely exceeds output in the single treatment, i.e., peer effects raise productivity. Third, low productivity workers are significantly more sensitive to the behavior of peers than are high productivity workers. Our findings yield important implications for the design of the workplace. JEL Classification: D2, J2, K4.
Identification of peer effects using group size variation
Econometrics Journal, 2009
This paper considers the semiparametric identification of endogenous and exogenous peer effects based on group size variation. We show that Lee (2006)'s linear-in-means model is generically identified, even when all members of the group are not observed. While unnecessary in general, homoskedasticity may be required in special cases to recover all parameters. Extensions to asymmetric responses to peers and binary outcomes are also considered. Once more, most parameters are semiparametrically identified under weak conditions. However, recovering all of them requires more stringent assumptions. Eventually, we bring theoretical evidence that the model is more adapted to small groups. JEL Classification: C14, C21, C25
Labour Economics, 2014
This research was partially funded by the Institute for Education Sciences. Gaston Illanes and Gabriel Kreindler provided expert research assistance. Seminar participants at EALE, Maryland, Warwick, and Queens provided helpful comments. Special thanks go to Bruce Sacerdote, who patiently walked me through his earlier analyses and graciously supplied new results, and to Steve Pischke, for extensive discussions and feedback repeatedly along the way. Thanks also go to many of my other peers for helpful discussions and comments, especially
Accounting for Peer Effects in Treatment Response
SSRN Electronic Journal, 2014
Accounting for Peer Effects in Treatment Response * When one's treatment status affects the outcomes of others, experimental data are not sufficient to identify a treatment causal impact. In order to account for peer effects in program response, we use a social network model. We estimate and validate the model on experimental data collected for the evaluation of a scholarship program in Colombia. By design, randomization is at the student-level. Friendship data reveals that treated and untreated students interact together. Besides providing evidence of peer effects in schooling, we find that ignoring peer effects would have led us to overstate the program actual impact.
Right Peer, Right Now? Endogenous Peer Effects and Achievement in Victorian Primary Schools
Social Science Research Network, 2013
This paper presents estimates of endogenous peer effects in pupils' school achievement using data on national test scores, across multiple subjects and cohorts, for the population of primary school pupils in Years 3 and 5 (aged 7/8 and 9/10 years) in the Australian state of Victoria. Identification is achieved via school-grade fixed effects and instrumental variables (IV), exploiting plausibly random differences in the age distribution of peers and their gender mix across cohorts. The results provide strong evidence for the existence of endogenous peer effects across all subjects, with the IV estimates close in magnitude to the corresponding fixed-effects estimates, although less precisely estimated. In reading, for example, a one point increase in peers' average test scores leads to between a .14 and .39 point increase in own test score, with similar ranges across other subjects.