6
$\begingroup$

I wanted to use poisson glm model which would take serial correlation (AR-1) and overdispersion into account. I was directed to use gee, but gee (function of gee package in R) requires parameter id which I don't have an idea what that means. The documentation says it identifies the clusters but I don't have an idea what the cluster concept is about. Can you please explain the cluster concept of gee? Is it somehow related to the autocorrelation thing?

$\endgroup$

1 Answer 1

6
$\begingroup$

I never used this kind of models but a quick Google search reveals that

Generalized Estimating Equations (GEE) (Liang and Zeger 1986) are a general method for analyzing data collected in clusters where 1) observations within a cluster may be correlated, 2) observations in separate clusters are independent, 3) a monotone transformation of the expectation is linearly related to the explanatory variables and 4) the variance is a function of the expectation. It is essential to note that the expectation and the variance referred to in points 3) and 4) are conditional given cluster-level or individual-level covariates.

(source: Halekoh and Højsgaard, 2006 in JSS paper on geepack library)

So this kind of models seem to be designed especially for clustered data and if your data is not clustered then this does not seem to be a right model for you. If you need a model that accounts for autocorrelated errors you may try GLS.

As about what is clustered data - we say that the data is clustered if there is some grouped structure, e.g. students are grouped in schools, patients in hospitals etc. If you want to account for group effects, then you use models that let you define such structure (e.g. linear mixed models). The structures can be hierarchical: students grouped in classes, classes in schools, schools in districts etc. or even crossed: students grouped in schools and at the same time grouped by the neighborhood they live in (that is possibly different than school localization).

$\endgroup$
2
  • $\begingroup$ thanks Tim! Points 1 & 2 of your citation answer my question perfectly. $\endgroup$
    – Tomas
    Apr 14, 2015 at 11:00
  • 1
    $\begingroup$ Additionally, a cluster can represent an individual in a longitudinal study where repeated measurements are taken. The id, in this case, would be a unique identifier representing each individual in the study. Each individual has many repeated measurements, which are almost always correlated. This is why GEE models are attractive when analysing data obtained from a longitudinal study. $\endgroup$
    – Edward
    May 5, 2020 at 12:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.