The
selection was via a sequence of likelihood ratio tests on several
nested models. A power model in which the correlation between
observations from the same carcass decays as the time delay
between the observations increases was selected as the most
appropriate.