01-31-2012 05:01 PM
Hi, I ran a proc logistic on a binary response with a few covariates, some continuous and some categorical. When I checked the TYPE III analysis of effect, there is one line for a categorical covariate with DF = 0 and missing estimate. What is wrong here?
TYPE III Analysis of Effect
Effect DF Chi-square Pr>Chisqr
Age 1 58999.5037 <.0001
Indicator 0 . .
gender 1 3266.6988 <.0001
01-31-2012 05:27 PM
What does proc freq on the variable (indicator) do, with similar modelling restrictions (ie any where clause you used)?
02-01-2012 10:20 AM
Logistic Regression has a linear model component. If any variable can be expressed as a linear combination of others, then it will show up with 0 degress of freedom and no estimate.
For instance, if you had 0-1 binary variables called Female and Male, then <Female = 1 - Male> and one of them is redundant. Usually, it is not as bovious as my example and you might need to do some two way frequencies or correlations to tease out the dependency.