Open alexpkeil1 opened 5 years ago
Note that a GEE-like approach to clustered data (e.g. longitudinal measurements on individuals) can be done using qgcomp.boot functions with the id
parameter set to the cluster identifier. Like a GEE, the mean function (regression parameters) will be the the same as a GLM (in expectation for GEE, but exactly in the case of qgcomp). However, the bootstrap variance estimates will be clustering appropriate.
Note this approach has now been implemented by Welch et al in https://doi.org/10.1016/j.envint.2021.106787
There is also a gist with a brief example of this: https://gist.github.com/alexpkeil1/1887e2f87d92e682c517dd8fe74415b4
Hello, I was wondering if using qgcomp.glm.boot with the "id" specification with a paired time-varying exposure and paired time-varying outcome for each cluster was appropriate? Does this estimate the change in outcome driven by "unit change" in mixture quartiles between timepoints? Or is it better to use a difference measure for qgcomp that is not time-varying and use that to assess the impact of per quartile increase in mixture concentration change on outcome between timepoints?
Would be nice to have a longitudinal qgcomp with some of the following possibilities: