Randomly Censored Partially Linear Model Statistical Diagnostics

This article mainly studied the partial linear model for random censor-ship data, lay in the statistical diagnosis method. This model explanatory variable has two kind of situations, one of the explanatory variables is the definite fixed design situation, the other is the stochastic design situation; Regarding the fixed design situation, after using synthesis data method processing data, we obtains the parameter and the non-parameter estimator using the smoothing spline method. In this foundation, use the data deletion model and the mean shift model, obtains outliers and the influencers diagnosis statistics SC_i statistics and generalized cook distance and so on. Afterwards, we also discuss the local influence analysis and get conscious influence matrix for case for weight perturbation and perturbation of response. Finally, examples are given to illuminate our results.Regarding the situation of stochastic design, we unify the least square estimator and the weighting core estimator to estimate parameter and non-parameter for transformed model of synthesizing data. Through the transformed original model, enable it to draw support the linear model basic diagnosis model - - the case deletion model and the mean shift model, carry on the diagnostic analysis to the original model, and has proven in this regarding the original model of two kind of basic diagnosis model approximate equivalence, thus we has given the test statistics of outliers, also has defined diagnosis statistics the original model in this foundation, like internal studentized residuals and the generalized cook distance and so on. Especially, aimed at the use the estimate method to give in particular in the model, we defined new generalized cook distance for the non-parameter. Finally, we confirmed this kind of diagnosis mentality validity through the example.