Analysis for partial choice set design

[pakhi]

Dear team,

We are conducting a labelled partial choice set design with the following labels: video, telephone, app, online, and face-to-face (f2f). In our candidate set design, f2f appears consistently, while any two alternatives from the remaining four vary, resulting in three alternatives per person.

We have collected pilot data and are now seeking help with the analysis. We believe that this could be managed by setting up the data for NLOGIT, for example, by using cset = 5, alti = 1,2,3,4,5, and -999 for the attribute levels of the specific alternative that were not shown. Could you assist if this is the correct approach?

NLOGIT codes are pasted below for reference.

Thank you for your help in advance.

Kind regards,
Pakhi

Code: Select all

Nlogit

;lhs = choice, cset, alti
;choices = video, telephone, app, online, f2f
;checkdata

;model:
U(video) = video + v_admin0*admin0 + v_tech0*tech0 + v_tech1*tech1 + v_dur*dur + v_feed0*feed0/

U(telephone) = telephone + t_admin0*admin0 + t_tech0*tech0 + t_tech1*tech1 + t_dur*dur + t_feed0*feed0/

U(app) = app + a_admin0*admin0 + a_tech0*tech0 + a_tech1*tech1 + a_dur*dur + a_feed0*feed0/

U(online) = online + o_admin0*admin0 + o_tech0*tech0 + o_tech1*tech1 + o_dur*dur + o_feed0*feed0/

U(f2f) = f_admin0*admin0 + f_tech0*tech0 + f_tech1*tech1 + f_dur*dur + f_feed0*feed0 $


[johnr]

Hi Pakhi

First, you should be aware that Nlogit has its own specific forum for answering questions, and questions about other software really should be posted on the software's website, not here. That said, with respect to your question, you can control variable choice set sizes via the data in a different manner than you propose. In most software, including Nlogit, the software will ignore the entire observation if you use a missing value code (-999 in Nlogit case) - this is called listwise deletion. That is, it won’t ignore the alternative, it will throw the whole observation out entirely. In your case, as every choice task will have at least two missing alternatives, the model shouldn’t estimate at all, as all observations will be ignored.

The way you need to handle the data is similar to what you suggest, with a few minor changes.

Data for each choice observation should include only data on the three alternatives shown to respondents. Alternatives not present should not appear in your data at all. In your case, that means that the cset variable should be 3 for all observations, not 5.

You control the alternatives via the alti variable you have. If for a choice observation it sees alti = 1, 3 and 5, it firstly knows there are only three alternatives in this observation (via the cset variable) and those three are video, app, f2f.
If the next observation has alti = 1,2,4, then it again knows there are three alternatives and these are video, telephone, online.

As such, you control this information via the alti variable in your dataset.

Hope this helps

John

[pakhi]

Thank you for your response John, this is very helpful. And apologies for not posting this on Nlogit website.

Kind regards,
Pakhi