Research On Statistical Inference Of Non-probability Sampling Web Survey

With the development of information technology,the Internet has been widely used and popularized in people's lives and work.The sampling survey technology has also developed from traditional paper-pencil survey to Web survey.At present,Web survey is widely used in scientific research.However,Web survey based on Internet has its own limitations,which are mainly manifested in the following two aspects: one is sample selection can not fully guarantee the probability sampling;the other is that there are usually large coverage errors between the actual sampling frame and the target population.These limitations will lead to inability to directly use Web survey samples to make statistical inferences on the target population.In order to solve the above two problems in the Web survey,this thesis mainly discusses the following aspects:Firstly,in order to solve the problem of non-randomness of Web survey samples,we propose a randomization method for non-probability Web survey samples by using the method of the propensity score matching samples.When the auxiliary information of the target population unit is known,we can use the probability sampling method to sample a probability sample from the target population as an auxiliary sample,then,we match the non-probability sampling Web survey sample with the auxiliary probability sample based on the propensity score,which will produce a matching sample from the Web survey sample.Moreover,we demonstrate that the matching sample has the same randomness as the auxiliary probability sample.The results show that it is feasible to make statistical inference for target population based on matching samples.Secondly,based on matching samples from Web survey samples,this thesis discusses the construction of simple estimator and weighted estimator of the target population parameters,and demonstrates the unbiased and asymptotic of the estimators,and discussed the variance of estimators and their properties.The construction and properties of estimators are further studied when using matched samples to estimate population parameters by using the method of post-stratification estimation,and the sample selectivity bias of non-probability sampling Web survey is estimated.We explore an effective method for statistical inference of target population by using non-probability sampling Web survey samples.The results show that statistical inference of target population by using non-probability sampling Web survey samples is feasible.However,the accuracy of estimation is closely related to the result of sample matching in stratification estimation.Finally,in order to solve the problem of large coverage error between the target population and the Web survey population,we propose a method of estimating the unknown population size by using the capture-recapture model in bio-statistics based on the method of propensity matching samples.That is,we can estimate the Web population size of the actual sample of the Web survey by using the Logistic regression capture recapture model and the demographic characteristics of the sample unit of the Web survey.On this basis,in the process of estimating target population parameters by using propensity matching samples,we use the method of supplementary sampling domain estimation to estimate the target population parameters.That is to say,the supplementary probability sampling is carried out in the unmatched probability sample units,and the supplementary samples are investigated.Then,the target population parameters are inferred by combining the matched samples and supplementary samples.Furthermore,the self-selection bias and coverage error of Web survey samples are estimated.It solves the problem of statistical inference of target population by using Web survey samples when there are large coverage errors.