LAUSR

Abstract Although an increasing number of articles examining statistical control variables exist, the use of control variables receives less scrutiny than other methodological topics, and research has yet to show how adding or dropping a control variable can damage causal inferences. By showing the unexpected consequences of including and excluding a control variable, this article challenges the view that additional control variables always strengthens causal conclusions. This study first illustrates the purification role of statistical control variables, and then quantifies the omitted variable bias. Using the Directed Acyclic Graphs (DAG), this essay differentiates confounding bias from overcontrol and endogenous selection bias. While using control variables may at times rule out alternative explanations, it is equally possible that adding control variables introduces overcontrol and endogenous selection biases, creating alternative interpretations rather than ruling them out. This paper concludes by discussing how and whether scholars should include certain control variables.