Nonlinear Regression Models With Missing Data Based On The Statistical Likelihood Of Diagnostic Experience

This paper studies statistical diagnosis problems of the nonlinear regression models with missing data. First, we assume that the responses are missing at random; we impute missing data and get complete sample, the parameters are estimated by the empirical likelihood, then we develop adjusted empirical log-likelihood ratio and obtain the asymptotic confidence interval; we consider the following two missing mechanisms under the MAR assumption and get the following conclusion:when missing mechanisms p(x) is more larger under the same sample size n, the coverage probabilities is more higher and the average length is more shorter; when the sample size n is more larger under the same missing mechanisms p(x), the coverage probabilities is more higher and the average length is more shorter. Compared with the normal approximation method, the empirical likelihood method can improve the coverage rate. Second, we can get the approximate first step for parameters based on the deletion model case, we can develop empirical likelihood distance, Cook distance and standard pseudo-residuals statistics. Finally, two examples are given to illustrate the validity of diagnostic measures.