Nonidentifiability in stochastic models of illness and death.

The interpretation of animal survival experiments in which disease incidence is determined at death or following sacrifice is shown to involve certain ambiguities. In particular, quantities of interest such as the expected duration of life for an animal contracting a specific disease at a specific age are found to be nonidentifiable. An example is constructed in which two populations of animals will appear similar to the experimenter but in which animals contracting a particular disease in one population may have double the life expectancy of similarly afflicted animals in the other population.