30 views

Skip to first unread message

Aug 29, 2022, 11:14:56 AMAug 29

to Biogeme

Dear all,

I am doing a discrete choice model to study how some variables affect how people choose air travel itineraries.

My model is like this: U = ASC + B_TF*TF + B_NS*NS + B_TC*TC + B_p*p + B_TDa*TDa + B_TDe*TDe.

To estimate the parameters, I designed a discrete choice experiment through a survey. In the survey I make 8 questions, in which I ask people to choose one of two options in each question. In each option I present a combination of the variables in the model.

To achieve this in BIOGEME, I was told that I should put 16 equations where the variables are varying, but the parameters B are the same, so I can estimate them. Regarding the ASC, I was told to put a different one for each of them, except for one that has to be zero.

My question is if in fact the ASC has to be different for every itinerary or only for each pair of options that come from the survey, because people are each time choosing between two alternatives. My second question is: if this is true, how can I calculate a unique ASC to put in the overall model? (I need a general model to calculate the utility of thousands of itineraries)

Apologize me if it is not clear. I tried to be a little bit direct while making the whole problem understandable.

Thank you in advance,

Tiago.

Sep 1, 2022, 3:09:31 AMSep 1

to tiago...@gmail.com, Bierlaire Michel, Biogeme

On 31 Aug 2022, at 20:07, tiago...@gmail.com wrote:

Thank you for the clarification.

I have another question. What can I do if some betas have an unexpected sign? From my results, B_TF and B_TC are positive and should be negative. Are there any methods to solve this?

First, look at your data to understand where it comes from. Then, improve your spefication. This is usually due to endogeneity issues, caused by the omission of important variables.

Thank you in advance. Best regards,Tiago.

Bierlaire Michel <michel.b...@epfl.ch> escreveu no dia terça, 30/08/2022 à(s) 12:29:

As the respondents had to choose between two alternatives, you need to estimate a binary logit model, with two alternatives, not 16.

Reparding the constant, it is the mean of the error term. It may capture the fact that respondents tend to respond more often to the first question.

--

You received this message because you are subscribed to the Google Groups "Biogeme" group.

To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/94d4c8a5-9d20-41cc-8fbc-07eb8a97cc4cn%40googlegroups.com.

<logit.ipynb>

Reply all

Reply to author

Forward

0 new messages

Search

Clear search

Close search

Google apps

Main menu