Abstract

The inverse probability weighting (IPW) estimator is widely used for estimating the
average treatment effect (ATE) in causal inference. However, the IPW estimator is
prone to bias when the exposure variable is misclassified, even if the misclassification is
non-differential. In this paper, we propose a correction for the IPW estimator using the
method of moments (MoM) to account for misclassified exposure. We derive the corrected
IPW estimator, demonstrate its unbiasedness under misclassification, and evaluate
its performance through simulation studies. Furthermore, we discuss techniques for
estimating misclassification probabilities, including scenarios with and without validation
data or exposure replication. We also derive ranges of these probabilities for conducting
sensitivity analyses when no gold standard is available. Additionally, we apply the
proposed method to real-world data from the Bangladesh Multiple Indicator Cluster
Survey (MICS) 2019 to estimate the causal effect of the wealth index on ICT skills among
women aged 15-49, demonstrating the robustness of our method.