A1235
Title: On prediction in mixed linear models when random factors and errors are correlated
Authors: Tatjana von Rosen - Stockholm University (Sweden) [presenting]
Azadeh Chizarifard - Karolinska Institutet (Sweden)
Abstract: The inference about mixed linear models is often based on the assumption that the vector of random effects and the vector of random errors are independent. The inference in these models usually concerns the estimation of the fixed effects and prediction of random effects. The aim is to consider a more general mixed linear model allowing for the correlation between random effects and random errors. In particular, the best linear unbiased prediction is of interest. The necessary and sufficient conditions are derived for the equality of the best linear unbiased predictors (BLUPs) under two mixed linear models with different covariance matrices using a new approach, which is based on solving a system of matrix equations. The proposed approach can be also used to study the equality of the best linear unbiased estimators of the fixed effects. The relationship with some existing conditions for the equality of the BLUPs is demonstrated.
