Conventionally, in longitudinal studies, the mean structure has been thought to be more important than the covariance structure between the repeated measures on the same individual. Often, it has been argued that, with respect to the mean, the covariance was merely a ‘nuisance parameter’ and, consequently, was not of ‘scientific interest’. Today, however, one can see that from a formal statistical standpoint, the inferential problem is entirely symmetric in both parameters. In recent years there has been a steady stream of new results and we pause to review some key advances in the expanding field of covariance modelling, In particular, developments since the seminal work by Pourahmadi (1999, 2000) are traced. While the main focus is on longitudinal data with continuous responses, emerging approaches to joint mean-covariance modelling in the GEE, and GLMM arenas are also considered briefly.
[1]
M. Pourahmadi.
Joint mean-covariance models with applications to longitudinal data: Unconstrained parameterisation
,
1999
.
[2]
Gilbert MacKenzie,et al.
On modelling mean‐covariance structures in longitudinal studies
,
2003
.
[3]
Calyampudi R. Rao,et al.
The theory of least squares when the parameters are stochastic and its application to the analysis of growth curves.
,
1965,
Biometrika.
[4]
M. Pourahmadi.
Maximum likelihood estimation of generalised linear models for multivariate normal covariance matrix
,
2000
.
[5]
Modelling marginal covariance structures in linear mixed models
,
2001
.
[6]
Michael G. Kenward,et al.
A Method for Comparing Profiles of Repeated Measurements
,
1987
.