12 Instrumental Variables Regression
As discussed in Chapter 9, regression models may suffer from problems like omitted variables, measurement errors and simultaneous causality. If so, the error term is correlated with the regressor of interest and so that the corresponding coefficient is estimated inconsistently. So far we have assumed that we can add the omitted variables to the regression to mitigate the risk of biased estimation of the causal effect of interest. However, if omitted factors cannot be measured or are not available for other reasons, multiple regression cannot solve the problem. The same issue arises if there is simultaneous causality. When causality runs from \(X\) to \(Y\) and vice versa, there will be an estimation bias that cannot be corrected for by multiple regression.
A general technique for obtaining a consistent estimator of the coefficient of interest is instrumental variables (IV) regression. In this chapter we focus on the IV regression tool called two-stage least squares (TSLS). The first sections briefly recap the general mechanics and assumptions of IV regression and show how to perform TSLS estimation using R. Next, IV regression is used for estimating the elasticity of the demand for cigarettes — a classical example where multiple regression fails to do the job because of simultaneous causality.
Just like for the previous chapter, the packages AER (Christian Kleiber & Zeileis, 2017) and stargazer (Hlavac, 2018) are required for reproducing the code presented in this chapter. Check whether the code chunk below executes without any error messages.
library(AER)
library(stargazer)References
Kleiber, C., & Zeileis, A. (2017). AER: Applied Econometrics with R (Version 1.2-5). Retrieved from https://CRAN.R-project.org/package=AER
Hlavac, M. (2018). stargazer: Well-Formatted Regression and Summary Statistics Tables (Version 5.2.2). Retrieved from https://CRAN.R-project.org/package=stargazer