Introduction to instrumental variables and their application to large-scale assessment data

Abstract

In the social sciences, estimating causal effects is particularly difficult. Gold standards are set by randomized experiments in many cases expensive, unenforceable for ethical and practical reasons. Recent research has drawn attention to techniques that under some conditions, could estimate causal effects on non-experimental observable data. One technique is the instrumental-variables (IVs) approach. This approach is used to determine variation that is exogenous in treatment and to estimate causal inferences. This paper begins by explaining the logic of IVs and then reviews the literature on the use of the IVs approach in the educational context. The most common types of IVs and the guidelines for selecting appropriate variables are explained. The statistical background of IVs estimation is described, which is followed by a discussion of the assumptions that underlie statistical procedures. Finally, empirical examples that use data from the Polish extension of the Programme for International Student Assessment are presented to estimate the effects on student learning outcomes of having at least one neighborhood friend in the classroom.