Sample Selection Model And Its Application In Study Of The Influence Of Medical Expenses

Sample selection model are multiple regression equation systems which is made up of selection equation and result equation.It allows residual from two equations isn't independent. And dependent variable from result equation is censored as missing value by above or below a specified value.Then this special character has good applicability prospect in statistics modeling. At the same time,making use of the correlation information with two equations residual,the parameter estimators of this model are more efficient,at some general situation,than the traditional single regression equation in which dependent variable is missing.In some epidemiology research and clinical trial,we collect a random sample of people,hospital,and so on,but such unit sometimes makes some behavior which will destroy random sampling as a result.On account of sample selection bias with unit behavior,it is disappointed that we only require excellent sampling design.Although whether respondents take part in investigation is due to their own judgments,this decision isn't occurred by completely random mode such as throwing coins.It is influenced by other factors which are also noticed by investigators.Due to the common effect of some factors,there are correlation between selection equation and result equation.Hence,sample selection model have important practical value in the field of medical research.In this paper,accompany with the data from medical expenditure,we have attempted to discuss systematically the medical modeling and typical estimation method of sample selection model,and then extend the context to the non-parametric estimation method further.At last,we pay our attentions to the statistical modeling and estimation methods for the effects of Residents' Medical Expenditure.In the first chapter,we introduce the basic structure of sample selection model.And then accompany with missing data of chronic medical expenditure,we explain mechanism of statistical modeling.Based on the partial sample,we analyze the reason that ordinary linear square estimators are biased.Chapter 2,we focus our attentions to likelihood estimator and two-step estimator of sample selection model.Along with years,we introduce the development and merits of parametric and semi-parametric methods,and their application in example.At the same time,we emphasize semi-parametric two-step estimation method which is the key in our paper,and introduce the selection methods of kernel density function and optimal bandwidth in kernel regression.Finally,we apply Nadaraya-Watson kernel regression with Gaussian density kernel and optimal bandwidth which is obtained by solve-the-equation plug-in method.Chapter 3,we work as simulation experiments and application example analyze.The relative performance of the regression estimators is studied in relation to the joint distribution of the error terms,the degree of censoring,and the degree of correlations between the errors terms.When censoring is high and correlation is strong,likelihood estimator and two-step estimator are efficient.When the distributional assumption of joint normality is violated,censoring is high, and correlation is weak,the efficient of two-stage local regression estimator is better.In experiment analysis,when the distributional assumption isn't satisfied and correlation exists,we select two-stage local regression estimator in application.And compare with the study of Residents' Medical Expenditure,we make reasonably,explain about results.Chapter 4 is topic summary.In this chapter,we compare with estimate methods,and discuss Collinearity and Heteroskedasticity problem further.In the end,we have made the bold forecast to software application and theory development about sample selection model.We use SAS 9.0 and STATA 9.0 statistics software as platform to handle the analyses of our simulation experiments and application example.