A Bayesian latent variable propensity score semiunion model(BS_LVM_PSA)is introduced,that discusses how to include latent variable into propensity score analysis,and incorporates prior infor-mation into latent variable equation,propensity score equation and outcome equation,then the semi-joint Bayesian method is used to estimate the parameters.Two Monte Carlo simulations are presented to elab-orate the actual performance of BS_LVM_PSA in a specific environment,and the BS_LVM_PSA is applied to study the actual data.Results of the simulation studies show that:First,latent variables can reduce the measurement error of preprocessing covariates and improve the accuracy of propensity score esti-mation;Second,under different matching methods,the better the treatment effect estimation effect of Bayesian method relative to frequentist approach;Third,in small sample,Bayesian approach has higher prediction accuracy and stability than Non-Bayesian approach;Fourth,the prediction accuracy of treat-ment effect with information prior is higher than that without information prior,and with moderate prior accuracy,the estimation of treatment effect is more reliable.In the case study,the BS_LVM_PSA proposed in this paper is used to study the poverty reduction effect of community poverty alleviation policies.
维普
万方数据