MNL estimation with two different Results

[Yashin Abdullah Ali]

 Dear Prof,

Estimation of the Unlabeled choice experiment with restricted design (generic parameter) with the attributes of the alternatives gives me the best answer. But if I wanted to include any explanatory variables, it has to be included by meaningful interaction with the attributes of the alternatives (according to the literature). 

I am getting two different stories from the estimation: 

Consider the following example: 

I have 4 Attributes > Charging speed, Price, reservation time & Charging station distance. Now I have 2 different stories while estimating the MNL model. 

Estimation procedure 1 (Model specification): 

If I include the beta_cost*price terminology and interact the socio-demographic variables with the price attribute, I get a good model fit but the socio-demographic variables are not statistically significant.

Estimation procedure 2 (Model specification): 

Whereas if I exclude b_cost * price terminology, the model fit is poor but the interacted socio-demographic variables are highly statistically significant.

It's a vice versa situation, I am not getting a good AIC value or P values with statistically significant results.

Are the utility functions for estimation procedure 2,  is the right approach to do?