Research On Matching Interpolation Model Inference Based On Dimension Reduction Of Propensity Score

In the context of big data,the proposal and application of network access fixed samples are receiving more and more attention.For propensity score matching,one of the processing methods,starting from the matching samples that are finally used for statistical inference,it is determined from the fixed samples.The matching samples that do not retain any target random sample information,and then study the matching units from the target sample with answering units and the units matched from the unanswered units from the fixed sample,together form an imputation sample,and propose Match the imputation model.The algorithm nodes of the model include:sample matching validity verification,target sample subset splitting,propensity score dimensionality reduction and compression,matching criterion to measure "approximate degree",and imputation sample formation and estimation,and finally after Monte Carlo simulation and empirical analysis.It is concluded that:1)Under the same conditions,the effect of the imputation estimator is more stable and closer to the true value than the matching estimator,and with the increase of the non-response rate of the target sample,the gap between the two will gradually narrow;2)The stability of the imputation estimator decreases with the increase of the non-response rate of the target sample,but increases with the increase of the critical ratio;3)The imputation estimator is not sensitive to the change of the critical ratio,that is,the non-response rate In the case of certainty,regardless of the ratio of the fixed sample to the target sample,the simple estimator can be used to estimate without weighting adjustment.In particular,if the non-response rate of the target sample is low([10%,50%]),the imputation estimator performs best when the critical ratio is[1.5,2],and as the critical ratio increases,the precision Instead,it declined.This means that when the imputation estimator is used for estimation,it may occur when the target sample has a low non-response rate,and it can achieve better results under the condition of a small number of fixed samples,which has strong practical value.