Statistical Diagnosis Of Generalized Linear Modal Model With Missing Covariates

Today is big data's era,no matter in finance,economy,engineering technology or biomedicine,data statistics play an irreplaceable and increasingly important role.Statistical diagnosis is an important part of statistical analysis.It is mainly used to judge the rationality of fitting observed data in established models and is widely used in all kinds of statistical problems.However,in real life,the lack of data and the distribution of data often show spikes and thick tails,which seriously affect the accuracy of data analysis.Based on the continuous research of scholars in recent decades,modal regression is proposed to solve the problem of outliers or deviation of error distribution in the data.However,for the lack of data,especially the nonignorable missing of data,the parameter estimation and statistical diagnosis of the modal linear model have not been studied,not to mention the generalized linear modal model,which is more extensive than the linear model.Therefore,on the basis of the previous research,this paper studies the situation that cannot be ignored in the generalized linear modal model.Because the linear modal model is simple and easy to understand and is the most widely used classical model,at the same time,the linear modal model is also the simplest and most special form of the generalized linear modal model,so the research of this paper starts with the linear mode model.In this paper,we consider the parameter estimation and statistical diagnosis of linear mode model with nonignorable missing covariates,and then generalize the related conclusions to the generalized linear mode model.In this paper,we first use kernel regression to estimate the probability of loss of covariables.Then,based on the MEM algorithm,the model parameters are estimated by using the inverse probability weighted adjustment compound quantile method,and the specific expressions of the model parameters estimation are derived.At the same time,the local influence analysis of the model is carried out based on the likelihood function,and concrete expressions of the corresponding influence matrix are derived when the response variables and covariates are disturbed.Then,the theoretical results obtained in this paper are extended to the generalized linear mode model.Finally,the method proposed in this paper is proved to be effective and feasible by random simulation and concrete real sequence research.