Hi!

I'm currently analyzing survey data (916 respondents in total) using the -asclogit- or -cmclogit- command, and I am trying to put individual-specific variables (socio-demographics) in the model. I'm interested in the value of travel time, which is coef(trip_time)/coef(trip_cost), with trip_cost and trip_time be the alternative-specific variables. Since most of the socio-demographics in my data set are categorical with more than two categories, I did the pairwise chi-squared tests between them. It seems some of the variables are correlated (the chi-square test has a very small p-value). I have three plans on how to cope with this issue:

1. put the socio-demographics sequentially and keep only the significant ones. Since the socio-demographics are just control variables, the multicollinearity issue between them should not affect the coefficients of trip_cost and trip_time so much, and indeed, the coefficients of trip_cost and trip_time are pretty stable as I put more socio-demographics in the model.

2. instead of including the socio-demographics in the model, use them as subgroups. Fit the model with only alternative-specific variables separately for each subgroup. For example, what is the value of travel time for the low income group, medium income group and high income group, etc. In this way, the value of travel time are more comparable, but the model may be not so accurate.

3. combine 1 & 2, fit the model with alternative-specific variables and significant sociodemographics in subgroups of the insignificant control variables.

Which plan is the best? If neither of them is good, please give some advice on what I should do. Thanks!!

By the way, the McFadden’s R-squared is very small for my models... only 0.09-0.1, while it seems a model with McFadden’s pseudo R-squared between 0.2-0.4 can be considered as a good fit. Is there something I can do to improve the goodness of fit of my model? Or is it ok if I don't report the goodness-of-fit in the paper? Thanks!

Best,
Zhuoqun