Keywords: Causal inference, real world evidence, observational study, propensity score, measurement error, unconfoundedness assumption
In observational studies, propensity score approaches under Rubin causal model framework are widely used to estimate average treatment effect (ATE). However, covariate measurement error is well-known challenge using observational data drawing real world evidence due to the violation of un-confoundedness assumption. Ignoring measurement error and using naïve propensity scores lead to biased ATE estimates. Only limited causal methods are found in this area of research, and there is no literature comparing their numerical performances. We identify four approaches to compare: Battistin and Chesher’s bias correction method, MaCaffrey and Lockwood’s inverse probability weighting method, and our latent propensity score methods estimated by EM algorithm and MCMC algorithm. Systematic simulation studies are conducted to compare their performances of ACE estimation under various rationales, small vs. large measurement error, Gaussian vs. binary outcome, continuous vs. discrete underlying true covariate, small vs. large treatment effect. The simulations show that, under Gaussian outcome, bias correction method and latent propensity score using EM algorithm perform best with small and large measurement error respectively; under binary outcome, the inverse probability weighting method and the latent propensity score method using MCMC algorithm perform best with small and large measurement error respectively. This is a joint work with Zhou Feng.