I have a distribution I want to predict from a categorical predictor
with 3 levels. The predicted variable are scores on the health of the
GI tract, one from each mouse. The distribution of scores has a good
number of mice that are normal and a good number that are abnormal. The
abnormal mice are variably so, with a positive skew. The additional
complication I have is that only extremes of the abnormal mice (high
and low tails of the abnormal distribution) were selected to gather
predictor data. So there are actually three groups of mice--normal,
variable low scoring, and variable high scoring. I can think of using
two models, one that compares normal versus abnormal, and another that
compares scores within abnormal mice, but what uses all the data and
the division in the abnormal mice? Would this be a case for ologit?
I've never had the opportunity to use ologit, yet.
