贝叶斯潜变量倾向得分半联合模型研究与应用

A Bayesian Latent Variable Propensity Score Analysis Semi-Joint Model:Approach and Application

本文提出了一种贝叶斯潜变量倾向得分半联合模型(BS_LVM_PSA),探讨了如何将潜变量纳入倾向得分分析,同时引入先验信息,利用半联合贝叶斯方法进行参数估计。通过两个数值模拟来测算BS_LVM_PSA在特定环境的性能,并将BS_LVM_PSA应用于实例数据。模拟研究显示:第一,潜变量能够降低预处理协变量测量误差,提高处理效应估计精度;第二,不同匹配方法下,贝叶斯方法相对于频率学派的处理效应估计精度更高;第三,在小样本中,贝叶斯方法相比非贝叶斯方法预测精度和稳定性更高;第四,有先验信息的处理效应估计精度高于无信息先验,且在适度的先验精度下,处理效应估计更加可靠。实例分析中,利用本文提出的BS_LVM_PSA研究了社区扶贫政策的减贫效应。

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 information 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 elaborate 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 estimation;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 treatment 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.