08-20-2015 04:52 AM
I have a question, when does degrees of freedom for a class variable (lets say with 5 levels) equal to zero in the logistic regressions Type III analysis of Effects?
In MLE class variable groups have very logical coefficients, but in Effects section, where you can find Wald etc., it says df = 0.
To have a better idea, output looks like this:
Type 3 Analysis of Effects
Effect DF Chi-Square Pr > ChiSq
VAR1 0 . .
Analysis of Maximum Likelihood Estimates
Parameter DF Estimate Error Chi-Square Pr > ChiSq Exp(Est)
VAR1 1 1 0.2912 0.0335 75.50 <.0001 1.338
VAR1 2 0 0 . . . . .
08-21-2015 10:58 AM
I usually see that when the level is fully equivalent to another level in the class, ie var1 level1 is fully equal to var1 level2 OR if it is the reference level.
08-24-2015 02:17 AM
This is not the case, level 1 and 2 completely differ with each other. It is in fact flag variable, with 0-s and 1-s (IV is equal to 0.05).
I suspect that this variable might be a linear combination of other variables, but how to check this fact?
08-24-2015 10:48 AM
See the last part of my answer - the reference level does not have an estimate. So for a variable that has two levels only, one will be missing.