Hopefully someone out there can answer a question regarding an equivalent
F-test for logistic regression.
I have results of S-Plus that I am trying to interpret. The residual deviance
from S-Plus - is this the same as the -2 log likelihood? I am trying to assess
a parameter similar to an F-test for linear regression. Following is my data:
Binomial model
Response: binary
Terms added sequentially (first to last)
Df Deviance Resid. Df Resid. Dev
NULL 3102 2896.336
COMPCOEF 1 7.77848 3101 2888.558
elev 1 42.34300 3100 2846.215
slope 1 5.76162 3099 2840.453
strmslop 1 3.59616 3098 2836.857
hydgrp.num 1 62.00777 3097 2774.849
lutypes.sqrt 1 7.69644 3096 2767.153
So, the residual deviance presented above - it this the same as the -2 log
likelihood? Would an equivalent function to an F test then be (2896.336 -
2767.153) ? Can I then calculate p <- 1-pchisq((2896.336-2767.153),3096) ?
Thanks for any help you can provide.
David Slayter
|