Author Information

Raul E. Macchiavelli

Abstract

Since in longitudinal studies the covariance structure is often regarded as a nuisance parameter, the strategy has been to use a parsimonious covariance model that describes adequately the observed data and permits better inference on the parameters of interest. In this paper we present some diagnostic tools to choose an appropriate covariance structure and discuss some strategies for fitting it. The main diagnostic tool is the "residual", computed as the standardized difference between the elements of the fitted covariance (concentration or correlation) matrix and the corresponding unstructured matrix. SAS Proc Calis is a very efficient procedure that fits many covariance structures in models with no fixed effects. Based on this procedure, we discuss some strategies to choose initial values and improve convergence problems in certain commonly used structures.

Creative Commons License


This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Share

COinS
 
Apr 30th, 11:45 AM

SOME STRATEGIES FOR SELECTING AND FITTING COVARIANCE STRUCTURES FOR REPEATED MEASURES

Since in longitudinal studies the covariance structure is often regarded as a nuisance parameter, the strategy has been to use a parsimonious covariance model that describes adequately the observed data and permits better inference on the parameters of interest. In this paper we present some diagnostic tools to choose an appropriate covariance structure and discuss some strategies for fitting it. The main diagnostic tool is the "residual", computed as the standardized difference between the elements of the fitted covariance (concentration or correlation) matrix and the corresponding unstructured matrix. SAS Proc Calis is a very efficient procedure that fits many covariance structures in models with no fixed effects. Based on this procedure, we discuss some strategies to choose initial values and improve convergence problems in certain commonly used structures.