Model estimates (dummy levels)

Experiments
1

I’ve noticed that the SurveyEngine modeller reports estimates for all levels of dummy coded categorical variables. I am used to needing to omit a level when including categorical variables so as to avoid the “dummy variable trap”. What kind of model specification is being used by the SurveyEngine here?

I can reproduce the same log-likelihood values reported by the SurveyEngine models in other software (e.g., Apollo in R), but I cannot do so without setting a reference category for my dummy variables. Any advice or explanations offered will be greatly appreciated!

2

This looks pretty old Thomas - the model used was a cox-regression using effects coded variables - and for novice user the redundant level utility was calculated as the remainder so all add up to zero (that’s effects coding).
We’ve moved to the apollo MNL modelling where everything is explicit now.