Testing individual regressors in logistic regression
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
*(Q) Tom Ryan's recent book chapter on logistic regression (1997; Modern Regression Methods) lists the Wald test as a 'poorly' performing test of individual predictors. Ryan goes on to describe 3 other tests: the likelihood ratio test, the scores test, and the exact conditional scores test. Apparently, SPSS 11.5 offers only the Wald test. Does any one have any input on this? *(A) Posted to SPSSX-L list on 2002/12/23 by Marta Garcia-Granero. To get the individual p values (either likelihood ratio test or exact conditional scores test), try this trick: LOGISTIC REGRESSION VAR="your dependent variable" /METHOD=BSTEP(LR) "list of IV" or /METHOD=BSTEP(COND) "list of IV" . (Rest of options come here) . /CRITERIA PIN(1) POUT(1). As you can see, you select backwards elimination of the IV, but forcing a PIN and POUT value of 1 (you don't eliminate anything from the model). This is the way of forcing SPSS give the individual p-values for the likelihood ratio test (BSTEP(LR)), or conditional test (BSTEP(COND)).