Statistical Diagnosis Of Generalized Nonlinear Regression Models Based On Empirical Likelihood Under Missing Data

This paper studies the statistical diagnosis problem of generalized nonlinear model with missing data based on empirical likelihood,developed the nonlinear model,using the empirical likelihood method to estimate the parameters,and construct the asymptotic confidence interval of the parameters.When the responses are missing at random,take the probability of missing ?(x)=0.5,0.8 respectively,and the sample size n=20,50,100 respectively.repeat each case simulation 2000 times,through the simulation,draw the conclusion:the coverage rate of the empirical likelihood method and the general method are relatively large,compared to the lack of probability;on fixed ?,as the sample size n increases,the average interval length of the empirical likelihood method and general method are shorter,the coverage rate increased.When the sample size n is fixed,the greater the ? is,the empirical likelihood method and general method of the coverage rate increased,the average interval length were shorter,but the empirical likelihood method is more obvious than the general method on improving the coverage.Then the statistical diagnosis is carried out on the model,introduceing how to detect the deviation between the actual data and the model established,then analyse the case-deletion measur e and the local influence in the model,and put forward the empirical likelihood distance,Cook distance diagnosis statistics as well as the standardized residual character.Finally analyze with practical examples,select the appropriate model to find out the influence in the data,verify the validity of diagnostic statistics.