Solved – Individual specific variables in McFadden’s choice model

logitstata

From my understanding of McFadden's discrete choice model, individual specific variables such as income should not affect an individual's choice unless interacted with product characteristics.

But in Stata's command asclogit, which claims to estimate McFadden's model, allows individual variables without interacting with product characteristics.

I am finding this command bizarre.

Here is my answer to this question : ASCLOGIT DOES not interact the casevars with the product characteristics . It interact the casevars with ALTERNATIVE DUMMIES. This is done behind the scene. If you need interactions with product characteristics, you need to use clogit and manually generate interactions.

Best Answer

Stata's asclogit command formats and handles the interactions behind the scenes, but the interactions between the case-specific variables and the alternatives are performed as part of the analysis with asclogit.

The results of asclogit are the same as the results of Stata's clogit command when the case-specific variables are explicitly interacted with the alternatives.

The asclogit command is simply a convenient version of clogit and there is nothing "bizarre" about it.

Related Solutions

Solved – Problem with building mlogit model (with no alternative specific variables)

The <NA>s in var_5 are killing you. Student 30135 has a complete row of data, so this works:

> new.data <- student.data.m[student.data.m$student_id == 30135, ]
> predict(model.m, newdata = new.data)
state_1      state_2      state_3 
3.980402e-01 5.812384e-09 6.019598e-01

Consider using the mice package (on CRAN) for imputation of those missing values, or dropping var_5 if it's not that theoretically important.

Solved – how to check for robustness for categorical variables in Stata

In general, what econometricians refer to as a "robustness check" is a check on the change of some coefficients when we add or drop covariates. In linear regression models, this is pretty easy.

However, in a logit (or another non-linear probability model), it's actually quite hard because the coefficients change size with the total amount of variation explained in the model.

A solution for this was proposed by the sociologists Holm, Karlson & Breen in SMx 2012, SMR 2013. It is implemented in Stata via the khb command.

Best Answer

Related Solutions

Solved – Problem with building mlogit model (with no alternative specific variables)

Solved – how to check for robustness for categorical variables in Stata

Related Question