Questions about using panel data

[Zahra Sanatgar]

Dear Professor Bierlaire,

 

I have a couple of basic question about developing a multinomial logit model using panel data in Python. I have followed the steps in 12panel.py. Here is what I have done:

1-    I define utility functions, associate utility functions with the numbering of alternatives, and associate the availability conditions with the alternatives

2-    Then I use the following few lines to estimate the model:

# The likelihood of one observation is given by the logit model

obsprob = models.logit(V, av, Choice)

# The likelihood of all observations for one individual (the trajectory) is the product of the likelihood of each observation.

logprob = PanelLikelihoodTrajectory(obsprob)

# Create the Biogeme object

biogeme = bio.BIOGEME(database, logprob)

biogeme.modelName = 'Panel'

 

# Estimate the parameters.

results = biogeme.estimate()

pandasResults = results.getEstimatedParameters()

print(pandasResults)

 

My specific questions are:

a-     Given that we are developing simple MNL, logprob is obtained by using PanelLikelihoodTrajectory. Essentially, there is no need for Monte-Carlo as no random parameters is involved. Do I understand it correctly?

b-    I get negative AIC and BIC. In addition, all variables have p-values that are either equal to 0 or 1. Any thoughts for potential reasons?

 

Thank you very much for your help.

 

 

 

[Bierlaire Michel]

You need to use PanelLikelihoodTrajectory only if you have an agent effect. If not, the fact that the data is panel is irrelevant, and you can treat them as cross-sectional data.

--
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/8dd24ae0-c6f5-4e60-811e-16f6df572e76n%40googlegroups.com.

[Zahra Sanatgar]

Dear Professor Bierlaire,

Thank you very much for your reply.

I do use PanelLikelihoodTrajectory. My question is that do I need to use Monte-Carlo while I do NOT have a mixture of logit models?

I also have a second question: I get negative AIC and BIC. In addition, all variables have p-values that are either equal to 0 or 1. Any thoughts for potential reasons?

All the best

ZS

[Bierlaire Michel]

On 13 Aug 2021, at 17:49, Zahra Sanatgar <sanat...@gmail.com> wrote:

Dear Professor Bierlaire,

Thank you very much for your reply.

I do use PanelLikelihoodTrajectory.

You should not. 

My question is that do I need to use Monte-Carlo while I do NOT have a mixture of logit models?

No. As I said, the fact that the data are panel is irrelevant for your model.

I also have a second question: I get negative AIC and BIC. In addition, all variables have p-values that are either equal to 0 or 1. Any thoughts for potential reasons?

Just estimate a simple logit model, without the trajectory.