The perils of OLS: Using robust estimators to address least squares' sensitivity to outlying points.

摘要

Least squares regression models have been one of the most time-honored tools researchers employ in the study of political science. Much like physical tools least squares models have their own limitations. In the presence of points that exert undue influence, OLS produces results that contain model fit bias. Therefore I advocate for the use of MM-estimators to, at a minimum, act as a robustness check against OLS. MM-estimators are not only resilient to data points that induce least squares' fit bias, but are also inherently resilient against heteroskedasticity thus making them an ideal tool for political science research.