Measurement Error in Causal Inference

Abstract

Valid causal inference, the pursuit of explanation of phenomena and evaluation of treatment effects relies on several testable and untestable assumptions. Among those, one that is often disregarded is that all the variables involved in the analyses are correctly measured. Conversely, it is often ignored that employing a causal inference framework in the study of measurement error can allow us to better characterize the source, the impact, and the remedies to this ever-present issue in data science. The literature at the intersection of measurement error and causal inference is very rich, albeit relatively recent. The purpose of this chapter is to provide a series of points of reflection to bridge these two vast fields of research. It describes uses of the causal inference framework to communicate assumptions about the measurement error mechanism and the impact of measurement error in the estimation of some of the most popular causal contrasts. The chapter also describes causal inference principles that should guide measurement error correction to improve upon causal inferences.