Dear Statalists,

I'm building a multinomial logit model to analyze ride-hailing behaviors. Each respondent was given 6 choice sets with 2 hypothetical ride-hailing trips characterized by 6 trip attributes and an opt-out option in each choice set. I first fit a conditional logit model using the -cmclogit- command in STATA 16, but since I wonder if there are random effects, I fit a panel data mixed logit model with the -cmxtmixlogit- command again. I'm now wondering how to compare the two models and decide which model to use. The standard deviations of the random effects are large in magnitude compared with the mean effects, is this enough to say that there is heterogeneity across individuals and the mixed logit model is better?

I tried to use the Hausman test and LR test, but I got some errors like this in both situations: 'when vce(cluster) exists, the test is invalid.' If I can use either test above, can anyone tell me how? I haven't found an answer so far. Thank you in advance!

Best,
Zhuoqun