I am assessing a dataset from a demonstration project where a continuous
outcome variable was measured repeatedly on 7 demonstration farms. I have a
total of 372 observations on the 7 farms. I used xtreg to assess the
association of the outcome with a single continuous independent variable
and have found several problems in the data. First, it fails the Hausman
test. Second, the residuals are heterocschedastic, decreasing with fit; and
are skewed as well.
My question: Would it be reasonable to evaluate the statistics from this
model by bootstrapping with the cluster(farm) option?
Thanks for any insights.
Buzz Burhans
Department of Animal Science
Cornell University
