Hi all,

I have a few questions about asclogit. I am currently working with some ride-hailing data. The ultimate goal of the project is to infer some measure for value of time and I am trying to pin down the baseline model. Just to give you some background, each passenger has 3 options (quit, A, and B). A and B are two types of services that the platform provides and B is more expensive but entails shorter wait time. I observe alternative-specific prices and wait times, but price and wait time are both 0 for quit. The choice set is the same for every passenger and one alternative is chosen out of the three. I also observe some passenger-specific characteristics including distance, trip duration, time of request, and such, but there is a problem. If the passenger chooses either A or B, trip distance and trip duration are recorded, but if the passenger quits, then trip distance and duration are 0. For example,

id alternative chosen price wait time distance trip duration weather time ....
1 quit 1 0 0 0 0
1 A 0 5 10 0 0
1 B 0 9 3 0 0
2 quit 0 0 0 3 8
2 A 1 10 8 3 8
2 B 0 19 2 3 8
3 quit 0 0 0 10 20
3 A 0 30 5 10 20
3 B 1 40 1 10 20
...

When I run clogit only with price and wait time, I get insignificant estimates. When I add distance and estimate alternative-specific coefficients, Stata reports no convergence with and without alternative-specific intercepts for A and B. There is no problem with adding only alternative-specific intercepts or weather/time characteristics without controlling for distance, however.

As a sidenote, I tried running similar specifications on Matlab and R as well. They generate reasonable estimates for price and wait time coefficients with only alternative-coefficients for distance, but when I add alternative-specific intercepts, price and wait time coefficients become highly insignificant. Can someone please help with this?