Clogit with interaction terms

[Liam Lee]

Dear,

I have a question regarding clogit. I conducted a choice experiment about home batteries, where respondents had to make 6 hypothetical choices. They could choose from battery A, battery B, or opt-out (not purchasing any battery). Each proposed battery had different characteristics (price, capacity, power, charge cycles, etc.).

My data file was transformed so that each row represented a choice. 1 respondent = 18 rows (6 questions with 3 choices). I ran a conditional logit (clogit) model for the main effects. (with ASC alternative specific constant)

clogit Choice ASC1 ASC2 Price Capacity Power CycleEfficiency ChargeCycles Expandable, group(Obs) cluster(Resp)

Now I should run another clogit model but with interaction terms (which remain constant with a respondent) e.g. age, electricity meter, income. Unfortunately, I can't find how to correctly add these interaction terms to the model. The outcome is often "omitted".

Can someone help me? thank you.

[Andrew Musau]

clogit is conditional fixed effects logit, and you need within-variation (i.e., variation within your grouping variable) to estimate the effect of a variable. Below, your group variable is "Obs" which I assume means observation, and I have no idea what this achieves as observations are singletons.

clogit Choice ASC1 ASC2 Price Capacity Power CycleEfficiency ChargeCycles Expandable, group(Obs) cluster(Resp)

For experimental data where the group variable is the participant, you are likely not to reject the random effects assumption as participants are randomly sampled from the population. Unless this is not true for your design, you can estimate a random effects model which will allow you to identify the effects of variables that do not vary within your grouping variable.