I want to compare model $A$ with 5 variables and model $B$ with 3 variables. I want to use the likelihood ratio test to do this. Note that this is logistic regression. How do I do this in SAS?

  • 3
    $\begingroup$ Is the three variable model a sub-model of the five variable model? If so, you can take the $-2\cdot{\rm loglike}$ from the outputs and do it manually, if nothing else. $\endgroup$ – Macro Nov 16 '11 at 1:23
  • $\begingroup$ @Macro: And you use the intercept and covariates values of $-2 \log L$ for each model and divide by 2? $\endgroup$ – Terry Nov 16 '11 at 1:38
  • $\begingroup$ No. Under the null, the difference between the $-2 \log L$ values follows a $\chi^{2}_{d}$ distribution (as $n \rightarrow \infty$) where $d$ is the number of parameters deleted to arrive at the submodel. You can use the CDF of that distribution to get the $p$-value. $\endgroup$ – Macro Nov 16 '11 at 22:59

See this example from the documentation which shows how to construct and compare different ROC curves for the models: http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_logistic_sect066.htm

Notice in the output that each model includes the -2 log L. Also notice that the final table of this output includes the chi-square values and p-values for the pairwise comparisons.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.