ApolloChoiceModelling forum

Hi Apollo Community!

Congratulations on the fantastic package !!

I am very new to discrete choice modeling, but I have to create a model for my thesis, and my supervisor suggested apollo.
Apologies for the very basic questions! (Still learning)
I have written the code and managed to create the five-level nesting structure that I need.
However, since I have too many alternatives (27) and a lot of independent variables (17 per alternative), the model ends up being very complicated.
I have created a test dataset ( to make sure that it runs properly and I have NA in all: s.e., t.rat.(0) , Rob.s.e. & Rob.t.rat.(0))
I do not know if the problem comes from the test data set (do I need more individuals (currently have 188))? Is there a problem with my model nesting structure?
I have read on the forum about bootstrap and tried it but had no luck!
I have also read somewhere in the forum that it is advised to start working on simpler models and then progressively making it complicated, but I have no idea on how this procedure is done? Do we start with less variables in the model? with a simpler nesting structure?

I would appreciate any help or advice!

Thank you very much!
Best Regards,
Vivian

Hi Vivian,

Welcome to the community. Glad you like the page

If you get all NAs in the s.e. and t-ratios, it is likely your model is misspecified. It could be that some parameters are not identified, for example because you put the same explanatory variable with the same coefficient in all alternatives, or because you have a perfect explanator among your explanatory variables (e.g. respondents always choose alternative 1 when dummy variable x1=1, and never choose it when x1=0). It is not possible to say what is the issue without looking at your code and data.

In general, it's always advisable to start with a simple MNL model and build from that. So I would recommend you start by estimating an MNL model (i.e. no nesting structure) on your data. If you still have issues, you know that the problem is in the definition of the utilities, and not in the nesting structure. So you can focus on solving that. Only after successfully estimating an MNL I would recommend moving to a nested logit.

Try estimating some MNL models with your data. If you keep having issues, you can write again attaching your data and script.

Best of luck!
David

Hello,

I will try this and get back to you as soon as possible!

Thank you very much!
Best regards,
Vivian