I have a simple data set with a small number of experimental groups
with a balanced number of animals per group. Group differences are
the interest. The measures on each animal consist of 20 to 40 binary
repeated measures. I can analyze these data two ways. First, I can
use GEE with binomial family, logit link function, and exchangeable
correlation structure. Second, I can average the 0s and 1s for the
dependent variable to a summary statistic for each animal (proportion
or percentage) and use a simple oneway ANOVA. Each animal in this
data set has not been equally sampled. Averaging to a summary
statistic ignores this. Does GEE treat clusters with unequal sizes
differently?
