symmetric matrix component correlations

02:27 Jan 30, 2016 

This has nothing to do with correlations between the components of a matrix. That makes no sense statistically speaking.

In GEE, there are within-subject correlations that have to accounted for by making some assumption about the correlation structure for the measurements repeated at different time points (i.e. pairwise correlation between repeated measurements on a single subject at different time points). The one probably being used here is known as the "exchangeable" or "compound symmetry" model. This has Cor (Y_ij, Y_ik) = rho and so it's symmetric.

Best thing is probably to ask the authors do they mean a "matriz de correlação" with "simetria composta".

17:25 Jan 29, 2016 

https://onlinecourses.science.psu.edu/stat504/node/180

===========================================

Generalized Estimating Equations (GEE)

Assumptions:

(...)
Covariance specification.
(...)
Here are four correlation structures:
Independence - **correlation between time points is independent**

===========================================

Se o contexto for o mesmo que o do link acima (e parece ser) a correlação é entre os time points, e não entre os componentes. Aliás, a estrutura gramatical da frase também sugere isso.

Me parece que o intuito da análise é estabelecer correlações entre dados coletados (no caso, os time points), e isso é feito agrupando-os em componentes em uma matriz (no caso, uma matriz simétrica). Então, as correlações são entre os dados, e não entre os componentes.