Dear Allstaters:
Following a matched case-control logistic regression one interprets the
probability of a case (1) relative to the matched control (0). Referring
to Hosmer and Lemeshow, the estimated probability for the matched data is
the probability that a subject with x=1 is the case compared to a control
with x=0.
I see how predicted probabilities are generated and interpreted for sample
data, but how does one apply that model to a series of out-of-sample
observations that aren't matched? Can I justify exp(B)/1+(expB) or should
I present predictions in terms of odds ratios exp(B)?
Any and all suggestions are appreciated.
Chris
|