Recommendataion for models (Random Effects Logit, Conditional logit, etc)?

[olivia kim]

In my dataset, I have total of 50,000 observations. They are dyads of products(100) and students(500).
With this, I'm basically looking at what products students are willing to buy.

Variables are following.

• Product level chacteristics : color size speed popularity price
• Student level characteristics : gender grade family-income
• Student-product level characteristics : use experience, disability-support
• Student dummy
• Student class dummy
• Student enrolled year dummy

Dependent variable : Buy (coded 1 if the student is willing to buy the product, student can choose up to 5 products)


With students buying multiple products and NO ranking of their preference, I am not sure what model to run.
Discrete choice model assumes once choice per student, so I don't think this is the model to go.
Random-effects logit might be a good choice but I lack confidence.

ANy suggestions?

[Erik Ruzek]

Are you interested in trying to discern which features of products and students are related to the buying decision? If so, then you would need to utilize something like a random or mixed effects model, which allows you to account for both correlations in buying behavior expected because of repeatedly measured students and products. It also allows you to include predictor variables at the student and product level. I'd recommend reading the Stata mixed manual on logistic models carefully as well as this really helpful article by Sommet and Morselli (https://rips-irsp.com/articles/10.5334/irsp.90) that has extensive appendices with code for Stata, R, etc. The Bristol LEMMA on multilevel models is also quite useful (https://www.cmm.bris.ac.uk/lemma/).

[olivia kim]

Erik Ruzek Yes, exactly! I am trying to look into what features of products are related to students' buying decisions. Thanks.