| Regression analysis is arguably the most prevalent statistical method,regression models like mean regression and median regression has been widely used in real life.In the last decades,with the emergence of the research of missing data,the regression models with missing data has been thoroughly studied and the research of missing value has been expanded to other parametric and nonparametric models.Statistical diagnosis of models were also developed during this period.However,researchers gradually noticed that when applied to some real datasets,mean regression and median regression would generate large bias,which inevitably leads to bad estimations,especially when these classical models are used to fit the datasets with nonsymmetric error distributions.To overcome these problems,the modal linear model has been developed,which has the merits of classical regression models,as well the good fit to nonsymmetric distributed error.However,there is no relevant research about the parameter estimation and statistical diagnosis problems with regard to the modal linear model with missing data.This paper thus considers the parameter estimation and statistical diagnosis of modal linear model with missing values in independent and dependent variables.We first employ the mode imputation and based on kernel nonparametric imputation methods to deal with missing values in dependent variable;besides,we use single interpolation,multiple imputation and MEM algorithm to deal with missing values in independent variables.We develop the parameter estimation methods of the relative model and study the statistical diagnosis of case-deletion model and the perturbation model.The effectiveness of the used methods is finally demonstrated by simulations of synthetic data and real data. |
