Measurement Error and Bias in Causal Models in Accounting Research

“Measurement error biases against [finding results]” is an often-repeated phrase used to dismiss validity threats arising from measurement error. As a general rule, this phrase is false. We provide examples of commonly encountered circumstances where the variable of interest is exogenous––the gold standard for causal inference––but where measurement error in empirical proxies nonetheless bias in favor of rejecting a true null hypothesis. In addition, we show that the common practice of including high-dimensional fixed effects, specifically firm fixed effects, can exacerbate this bias and lead researchers to spuriously estimate a causal effect when none exists. Finally, we show that measurement error pervades the accounting literature, and illustrate the effect of measurement error on causal inferences in a popular quasi-natural experimental setting where we can observe the measurement error in the treatment variable. We encourage researchers to triangulate inferences across multiple empirical proxies and to report results from specifications with and without high-dimensional fixed effects