Instrumental Variables Estimation in Political Science: A Readers’ Guide

Abstract

The use of instrumental variables regression in political science has evolved from an obscure technique to a staple of the political science tool kit. Yet the surge of interest in the instrumental variables method has led to implementation of uneven quality. After providing a brief overview of the method and the assumptions on which it rests, we chart the ways in which these assumptions are invoked in practice in political science. We review more than 100 articles published in the American Journal of Political Science, the American Political Science Review, and World Politics over a 24-year span. We discuss in detail two noteworthy applications of instrumental variables regression, calling attention to the statistical assumptions that each invokes. The concluding section proposes reporting standards and provides a checklist for readers to consider as they evaluate applications of this method. P olitical scientists frequently seek to gauge the effects of independent variables that are measured with error or are systematically related to unobserveddeterminantsofthedependentvariable.Recognizingthatordinaryleastsquaresregressionperformspoorly in these situations, an increasing number of political scientists since the 1970s have turned to instrumental variables (IV) regression. IV regression in effect replaces the problematic independent variable with a proxy variable that is uncontaminated by error or unobserved factors that affect the outcome. Instrumental variables regression is designed to relax some of the rigid assumptions of OLS regression, but IV introduces assumptions of its own.WhetherIVisinfactanimprovementoverOLSdepends on the tenability of those assumptions in specific applications (Bartels 1991). In order to help readers judge the tenability of IV assumptions, researchers must provide pertinent evidence and argumentation. Readers must have access to certain basic statistics that shed light on the susceptibility of the IV estimator to bias. Readers also need a description of thecausalparametertobeestimatedandanargumentexplaining why the proposed instrumental variable satisfies