| Based on Athey’s causal tree,this article considers how to reduce the influence of confounding bias on the estimation of causal effects when the data is the observational data.This paper combines the causal tree method with the propensity score stratification method,which uses the propensity score stratification method to stratify the data to achieve a higher homogeneity of the data within the strata and the distribution of various confounding factors at each level tends to be consistent,in order to reduce the influence of the confounding bias in the causal tree estimation.In this paper,based on the propensity score stratification method to get the data stratified,and the t statistic is used to measure the similarity of propensity scores within the strata.If the t value is greater than 1,the stratification will continue according to the median within the strata until t<=1,and finally divide the data into 5 to 10 levels.Based on the results of propensity score stratification,this article proposes two methods:the first method is to establish a causal tree at each level based on the results of propensity score stratification;the second method is to use the result of propensity score stratification as an explanatory variable added into the estimation of causal tree.The simulation results show that two methods proposed in this paper performs better than the causal tree method in both randomized experimental data and observational data,which have higher accuracy and lower MSE.For data with confounding bias,two methods proposed in this paper still have good estimation results,so it can be known from the simulation results that the propensity score stratification combined with the causal tree method reduces the influence of confounding bias on the estimation.This paper applies the method of establishing a causal tree within stratification to the estimation of impact of government subsidies on corporate profits.A total of 2,710 companies were selected,of which 849 companies receiving government subsidies,accounting for 31.3%,of which 1150 companies with a positive effect,accounting for 42.4%.Anathor modified method was used to study the impact of government subsidies on the number of employees in enterprises.A total of 578 companies were selected,of which 286 companies had a positive effect on the number of employees,accounting for 49.5%.This analysis is an application of causal inference in policy evaluation. |
