Omitted because of collinearity.

[Nanna Brandt]

Hi. I have a dataset based on a survey including the 3 following variables:


1) Exposure to a wind turbine (no = 0, yes = 1)

2) Included in the planning proces (values from 0-6)

3) Voted for the incumbent politician (no = 0, yes = 1).


People were only asked if they were ‘included in the proces’ if there were actually exposed. When I’m running the following logistic regression both ‘Exposure’ and the interaction term is omitted because of collinearity.

logit Voted Exposure Included c.Exposure#c.Included


Is there solution to this problem?

Thank you very much.


// Nanna

[Carlo Lazzaro]

Nanna;
welcome to this forum.
I guess that -exposed- and -included- are perfectly collinear. Check your dataset and, if necessary, reconsider the right-hand side of your regression equation.

[Nanna Brandt]

Thank you very much for your reply, Carlo.

I'm examining if -included- is modifying the correlation between -exposure- and -voting-, wherefore I'm running a regression with an interaction term and cannot exclude this one.

The variables are not omitted if I omit -included- as main effect from the regression and run the following:

logit Voted Exposure c.Exposure#c.Included

Do you know if that's a 'legal' solution?

Best regards
Nanna

[Carlo Lazzaro]

Nanna:
no, it is not.
You cannoit pick up a conditional main effect and neglect the other one included in the interaction.