I ran a Logistic Regression Model on SPSS v.20 and obtained Scores/Parameter estimates for a bunch of independent variables. The Classification Table in the Output claims a certain percentage of successful predictions based on its estimated probabilities, giving the numbers of correct predictions vis-a-vis observed cases. However, I am unable to reproduce the predicted probabilities that match those implied in the SPSS Output Classification Table, by manually computing e^(a+bx)/1+e^(a+bx) using the coefficients obtained in the output.
I tried using all the coefficients, as the SPSS manuals suggest as well as only the significant ones; yet my predicted probabilities do not come close to what the Classification Table shows.
Any idea why this discrepancy? I have all but 2 categorical variables as my Independent Variables. Any suggestion to help me reproduce those probabilities will be much appreciated. Thank you, in advance!