概率样本与非概率样本的整合估计方法OACHSSCD

In traditional probability sample collection,the increase in costs and the decline in response rates result in insuffi-cient valid samples,and the absence of target variables leads to biased estimations.In addition,the unknown sampling probability of non-probabilistic samples in online surveys and the method of estimating the sampling probability of non-probabilistic samples based on the Logistic regression model are sensitive to model specifications,which may lead to extreme probabilities and further cause the problem of high variation in the estimation results.To address the above issues,this paper proposes an integrated esti-mation method that combines probability samples and non-probability samples.Firstly,this method estimates the target variable of the probability samples through XGBoost to obtain the initial estimate,then further estimates the sampling probabilities of the probability samples and the non-probability samples,uses the kernel smoothing method to calculate the similarity of the sampling probabilities of the two types of samples,allocates the weights of the probability samples reasonably to the non-probability sam-ples according to the similarity,and performs weighted estimation of the known target variable based on the estimated weights,and finally,combines the two types of samples into one sample and adjusts the weights of the two types of samples by minimizing the MSE of the combined estimator,thereby achieving the estimation of the overall population.The results of both simulation and em-pirical studies indicate that,under various circumstances,the proposed method outperforms other methods in terms of bias and mean square error,demonstrating significant superiority.