A Methodological Approach for Causal Inference under Uncontrolled (and Possibly Latent) Pre-exposure to Treatment

Experiments have become one of the main tools for causal inference in political science, but scholars have raised concerns regarding the uncontrolled real-world pre-exposure of subjects to the information manipulated during the experiment. This pre-exposure can mislead conclusions about the treatment effects, and the problem is exacerbated because direct measurement of pre-exposure is not feasible in many studies due to the risk of confounding the experiment. This paper presents a method to estimate causal effects when pre-exposure is uncontrolled and directly unobservable. It formalizes the problem using the potential outcomes framework; decomposes the average treatment effect (ATE) into pre-exposure and exposure components; derives the bias that emerges when pre-exposure is ignored; establishes sufficient identification conditions; and introduces a bias-corrected estimator for the relevant causal parameters. The method is applied to analyze the impact of party cues on voters' policy attitudes.