panel data - Pandas

[Shelly ben zvi]

Dear group members

I estimated a mixed multinominal model for panel data to predict mode choice using Pandas.

For now I have a very basic model with only ACCs, error components, time and cost attributes.

With Pandas, I got a positive insignificant travel time coefficient which doesn't make sense.

When I run the same model with the same data in Bison biogeme , I get negative and significant travel time coefficient which make a lot of sense. 

I tried scaling my time and cost in various ways as I believed that could be the issue but nothing seems to fix the positive insignificant time. 

When I ran logit model (no panel) with Pandas, I got negative time+cost similarly to Biosn. So the problem is only with the panel data. 

I attach the codes and outputs for Bison and Pandas.

Any help will be highly appreciated. 

Shelly