Influence Analysis For Linear Mixed Models

Diagnostic methods are useful for assessing the adequacy of assumptions underlying a modeling process and for identifying unexpected characteristics of the data that may seriously influence conclusions or require special attention. So, we hope to provide an analyst a complete picture of points which excessively influence the fitted regression equation. Mixed linear models (Linear models with multiple sources of error) are frequently used in biology, medicine, economics, determining sampling designs and quality control procedures, and so on. Standard estimation methods (quadratic likelihood estimation method) for linear mixed models are sensitive to bizarre observations. Such influence can completely distort an analysis and lead to inappropriate actions and conclusions. So, We are interested in developing case梔eletion diagnostics for linear mixed models. Any observations that have large inflience for the variance components will affect the entire covariate analysis. Therefore, it is appropriate to examine the variance components befire examing the covariates. So, we will examine case梔eletion results for estimates of the variance components. The discussion will be fcused on the Restricted Maximum Likelihood (REML) method and the Minimum Variance Quadratic Unbiaced Estimation (MINQUE). In essence, th REML method deals with linear combinations of the observed values whuse expectations are zero. Genetic Algorithms (GA) is a highly effect numeical optimization method. In this paper, GA is used to calculate REML estimator. An example is presented.