Hello, I have two small questions for people experienced in discrete choice experiments and use of mixed logit models.
I am currently conducting a DCE with forced choices with 3 alternatives presented (no opt out). I have 4 attributes with three levels and 1 with two levels. This means 162 unique choicecards.
1) To calculate the number of unique choicetasks for two alternatives in a full factorial design, the formula is (162*161)/2 . How do I compute this when there are 3 alternatives presented.
2) How do I code the alternative specific constant , since I have 3 alternatives presented. If it were a choicetask with two alternatives I would code this as alternative 1 ==>0. and alternative 2= 1
Is it okay to code alternative 1=0 alternative 2=0 en alternative 3=1. ?
Thanks
I'm not sure to completely understand your question, but I would say that your problem is similar to what is solved by a multinomial logit. In transport modal choice models for instance, you can have several alternatives that can be identified by a numerical ID (car = 1, bus = 2, train= 3...).
Example (free) package softwares :
- The MLogit R package (https://cran.r-project.org/web/packages/mlogit/index.html), which documentation is very well written with numerous examples
- The MNLogit R package (similar to MLogit, but faster)
- Biogeme, a Python package (https://biogeme.epfl.ch). Powerful and flexible... and written by a Belgian :-)
Good luck !
I'll say a little about alternative specific constants ..... as you have it set up everyone is choosing from three alternatives. Suppose you think option 3 is special in some way --- so a constant (1) defined for that alternative (and zero for the other options) will allow you to experiment. Does option three end up being chosen (the "chosen alternative") a lot? If so, your alternative specific constant will have a positive sign.
You are able to do this because everyone has option 3 in their choice set.
You can also have an alternative specific variable --- say you code the income of the respondent for the third alternative -- so the data would be [0, 0, y] ..... and if income is unusually associated with that choice your revealed preference data should show it.
We had an interesting variant --- suppose everyone has slightly different choice sets (supermarkets) --- so we cannot have a normal alternative specific constant as there is no guarantee of a consistent appearance of the choice in the options --- however, everyone has a "nearest" alternative so we could code a "nearest alternative specific constant" (or indeed a nearest alliterative specific variable).
The software options from Prof. Bart Jourquin should be helpful!
That is why I said "as you have set it up." The experimental set up here appears to have everyone choosing from the same three options.
Extra info: My experiment is unlabeled.
Every participant will need to answer 5 choicetasks with each one presenting 3 options .(option A , option B, Option C) All three options differ per choice task.
Momently I am working with one block, so every participant receive the same questions.
Do I need to include a constant when coding the data in stata?
Thanks to all for clarifying .... clearly, I was not looking at this correctly.
What I thought was going on was three identical discrete choices with multiple levels of potential scores on 5 independent variables or measures: 4 of the measures (with 3 possible levels each) and 1 added measure (2 possible levels).
Obviously a very different problem from yours. My apologies.
Good wishes.