Dear Bill,
Thank you very much for your careful discussion of the issue.
It is not surprising that the deviance is the same whether you fit the
model using a factor response with weights or individual 0/1 responses.
I think this happens because the fitted probabilities in the saturated
models are
This is a very good question. You have spotted something that not many people
see and it is important.
The bland assertion that the "deviance can be used as a test of fit" can be
seriously misleading.
For this data the response is clearly binary, "Admitted" (success) or
"Rejected" (failure) a
2 matches
Mail list logo
