Study On The Generalized Ridge & Principal Correlation Estimate And Admissibility

The least squares estimate of the regression parameters (shortened form LS estimate) has many good nature, for example, when error is normally distribution in regression model, it has the smallest variance in all unbiased estimates class. But, when assumes X shows strange state, by now although its variance is the smallest in the linear unbiased estimate class, but its mean squares error is very actually big.Therefore many scholars devote to the improvement estimate, proposed many new estimates .a very important kind of estimates is unbiased estimates. And in numerous unbiased estimates, what affects the biggest is the ridge estimated, generalized ridge estimate, principal components estimate, principal correlation estimate and Stein estimate and so on.In this paper, we propose a new biased estimate of the regression parameters—the generalized ridge and principal correlation estimate in the third chapter, We give its some properties and prove that it is superior to LSE (least squares estimate), principal correlation estimate, ridge and principal correlation estimate under some certain conditions when we select MSE(mean squares error) and PMC(pitman closeness) criterion respectively. In the fourth chapter, promotes the generalized ridge and principal in growth curve model, similarly had proven its optimality, and its optimality has be confirmed through the actual data.The admissibility is the most minimum request to an estimate. In the fifth chapter, this paper proved the principal correlation estimate is admissible, and considering the admissibility of Linear estimates in growth curve model under matrix loss with respect to an Incomplete Ellipsoidal Restriction ,we use the fact that the admissibility of the Linear estimate in growth curve model with an Restriction under matrix loss is identical with that of the Linear estimate in the necessary and sufficient single Linear model under matrix loss .From that we obtain the necessary and sufficient conditions for DYF+C to be admissible estimates of estimable function KBL in the non- homogenous linear class.