Coding data of an opt out alternative in a choice experiment

[Will Wright]

Hi,

I am working on putting together the survey questionnaire for a choice experiment. This survey will be administered by a third party and I need to specify how I would like each variable coded. I plan on using effects coding for the variables in the actual choice experiment. Each choice set will contain two alternatives and an opt out option where the respondent can choose neither option. My question is how to code the opt out option for the choice experiment variables for running either asclogit or mixlogit? Should these variables be coded as missing?

Thank you,

Will Wright

[Joseph Coveney]

I don't know much about choice experiments, so pardon my naïveté, but isn't opt out its own choice option? And wouldn't opt out be conceptually different from neither? That is, I can opt in (choose to participate in the survey and respond to the item) and yet tell you that I don't care for either of your alternatives presented for the item. And, on the other hand, I might very well have a preference for one of the alternatives presented, but I wish to op out, that is, to decline to tell you which--this is why it seems that it's its own choice option. So, again it's simple-minded, but I see four options that the participant could be thinking of: first alternative, second alternative, neither alternative presented, and op out.

[Will Wright]

I didn't do a very good job of explaining my question. What I was meaning by opt out was what you have called the neither option. Opt out implying that the respondent prefers neither if the two offers made in the choice experiment. My problem is more regarding how Stata will read the data. Each attribute in the choice experiment is be coded as L-1 variables, L being the number if levels that attribute may take. In each choice set - two choice options plus the neither option - each possible choice is included in the dataset as a row or observation with a dummy variable for choice taking the value of 1 if the option was chosen or 0 if not. The 'neither' option does not take a specific level for any of the attributes so therefore the value of each of the variables describing then is undefined. For this reason it seems like the values of these variables should be coded as missing?

[Arne Risa Hole]

Will

If you code the variables as missing for the opt-out alternative you will effectively remove this alternative from the choice set. Another option is to code the variables as zeroes and include an alternative-specific constant for the opt-out alternative.

Arne

[Cordula Kiel]

Dear Arne (or all others able to comment on this question),

one further question related to this topic, as I also have an opt-out alternative in my choice design and I am not very sure how to implement it.
In my case, the opt-out alternative is also coded with zeros for each variable of the alternative.
But the opt-out alternative was very frequently chosen in the experiment, in more than 50 percent of the cases (with varying attribute levels for the other alternative) it was preferred over the other two available options, so opt-out is quite important.
My supervisor told me that it would therefore be important to include an ASC for the opt-out option.
I have seen different examples for such an ASC. Either the other two alternatives are coded as 1 and the opt-out alternative is coded as zero, or the other way round: the opt-out option gets a 1 assigned and the other two alternatives get zeros.

I am not sure which alternative makes more sense or how this decision affects the results and the interpretation of the ASC.
Could you please explain the difference?

Thanks a lot in advance!
Cordula

[Irene Bodega]

Dear users,

I followed the instructions given above about coding the optout alternative levels as zeroes. At the moment of estimating the clogit, I created a variable taking the value of 1 when the optout alternative is chosen and 0 otherwise. What is the correct procedure for the inclusion of such a variable into the model? Is there a specific way to include it as an ASC?

Thank you in advance,

Irene

[Charlotte Hammer]

Hey everybody,

I am also struggling with including an opt-out coding for my choice experiment in asclogit and mixlogit.
I coded the opt out variables as 0 in my data (as recommended by Arne Hole), but how do I include an alternative-specific constant for the opt-out alternative? What is exactly meant by that - do I have to define my data or the do-file?

Thank you
Charlotte

[Maya Gheroufella]

Drear Charlotte Hammer,

it is also my case, did you find any answers ?

Maya.

[Charlotte Hammer]

No, unfortunately I didn't.

[Maya Gheroufella]

Dear Charlotte,
I have found this post " Several questions: asclogit - clogit - mixlogit | dummy coding vs. effects coding | interaction effects | Hausman test " and it says: " we should created an alternative-specific constant for the opt-out option (1: opt-out, 0: alt1/alt2), as suggested by Arne Risa Hole in another thread".

I don't know if it helps !

Maya

[Hanna Ihli]

Dear Maya and Charlotte,

did you find a solution? I face the same challenge.

Thanks.
Hanna

[Maya Gheroufella]

Dear Hanna,
I have found this paper, it gives more details about how to model the opt-out alternative.
(https://www.researchgate.net/publica...es_to_Consider)

l' am still working on it, so if you have any other reference I would be glad to be aware of it.