Dear all,
I have one question concerning the mlogit command in Stata and/or R. In the model I’m working on the individuals don’t face the same choice set each other; for example I have 10 alternatives but for some customers the choice set is reduced to 8. Is it possible to take this into account when I use mlogit?
Thank you very much for your attention
Haven't heard about it in Stata (maybe, someone else has). As a workaround, I'd consider merging some alternatives so that the number of these "larger" alternatives to which every respondent is exposed is the same for each respondent, if that's makes sense, of course.
It depends on the case scenario and what does it mean exactly to have a customer faced with more choices versus less. Is it a good thing or a bad thing?
You could for example create a new variable for this specific exposure and examine it as a continuous variable (mediator or confounder in your model) . You can also turn it to a categorical variable by creating a cut-off point or dividing the sample into quartiles or quintiles- (of course you should be able to explain why did you choose this specific cut-off point in your paper) and also plug this new categorical variable as a mediator/confounder in the logistic equation.
I hope this answers your question.
Hi Daniel Guhl , the data set(yogurt) you showed on your R script have same choice set (yoplait, dannon, hiland, weight) for all the cases. Andrea Baldin probably asked if it's possible to use mlogit for the data set where choice set varies from case to case. I am currently stuck with similar type of problem in doing a transit route choice estimation.
Say, I have 100 observed transit data with their path attributes (ie. travel-time, cost, waiting time) and socio-demographic attributes (ie. age, income) of passengers. Each observed trip have multiple transit path options, thus choice set varying between 5~15 for all 100 cases.
Is traditional MNL model applicable for such scenario without any reference or baseline alternative? Can I do it using mlogit, or is there any other R package available for it?
Thank you.
Dear Daniel Guhl,
Thank you very much for this detailed explanation. I went through various literature which were saying traditional MNL is only applicable for same size choice set, and thought "mlogit" package is built upon that assumption.
Can I ask one more question? Is it possible to estimate MNL model without any alternative specific constants (ASC)? Suppose, the alternatives or choice set are given in a way that there's no reference or baseline alternative, so only wants to estimate the parameters of the attributes.
Thanks in advance.
-Asad
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R