Abstract
Propensity score (PS) methods reduce bias in observational studies under certain assumptions. These methods are often employed in large databases, including administrative claims data and drug registries, where misclassification of the exposure is likely. We conducted a simulation study to compare the bias and variance obtained from three common PS implementations (adjustment, matching, and stratification) for varying degrees of nondifferential exposure misclassification. Our simulation is based on realistic parameters from the Norwegian Mother and Child Cohort Study, and mimics an example where the true effect of medication exposure causes a substantial reduction in birth weight.