Partial choice set design in a labelled DCE

[pakhi]

Thank you very much for your response, Prof. Bliemer. I could only get the design running with a 10-30 restriction, and like you mentioned, it did not satisfy all constraints due to the restrictions in my design. If I decide to go ahead with the pilot test using the design I currently have, how important it is to have level balance and it is alright to manually change some levels?

Kind regards,
Pakhi

[Michiel Bliemer]

Attribute level balance is not crucially important and yes you can manually make changes (as long as you do not create perfectly correlated attributes or omit certain levels of dummy coded coefficients). After making manual changes, you could always import and evaluate the design using ;alg = eval(...) to make sure that it is fine.

Michiel

[pakhi]

Thanks a lot for your guidance Prof. Bliemer!

[pakhi]

Hi Prof. Bliemer,

I wanted to follow-up on this thread regarding my PCSD DCE. To reiterate, I am using five labelled alternatives (video, telephone, app, online, f2f). These are my attributes and levels: admin (0, 1), tech (0, 1, 2), duration (30, 60, 90, 120), and feedback (0, 1). Please note, ‘admin’ for alternatives app and online is always 0; and ‘tech’ for f2f is always 0 so I have removed them from the codes and candidate set. I am treating all attributes as alternative-specific parameters except for ‘feedback,’ which will be a generic parameter.

I have completed a pilot survey and am using Bayesian priors from the estimates obtained from the survey for my main DCE design. However, I keep getting an undefined d-error. I have seen many posts on this forum regarding undefined d-errors but cannot determine if the issue lies with my code or how I have defined my candidate set, and hence multicollinearity.

The codes are pasted below. For the candidate set, am I correct in indicating ‘feedback’ as generic in the candidate set spreadsheet? For example, I have named the column for the feedback attribute as ‘feedback’ in all alternatives instead of video_feedback, telephone_feedback, etc.

I have tried using different types of draws, but that didn’t work either. I also tried reducing the number of Bayesian priors from 11 to 9, but that also gave me an undefined d-error.

Ngene codes:

Code: Select all

Design
;alts = video, telephone, app, online, f2f
;rows = 50
;block = 5
;eff = (mnl, d, mean)
;alg = mfederov(candidates = pcsd_pilot.csv)
;bdraws = Gauss (2)

;model:
U(video) = video[-1.42913]
+ b1_video.dummy[(n, -0.03656, 0.24705)]  * admin_video[0, 1]
+ b2_video.dummy[0.39052]                 * tech_video[1, 2]
+ b3_video[(n, 0.38072, 0.33673)]         * duration_video[30, 60, 999]
+ b4.dummy[(n, 0.12529, 0.10148)]         * feedback[0, 1]
/
U(telephone) = telephone[-1.66766]
+ b1_telephone.dummy[(n, -0.12181, 0.25296)]  * admin_telephone[0, 1]
+ b2_telephone.dummy[(n, 0.24292, 0.25143)]   * tech_telephone[1, 2]
+ b3_telephone[(n, 0.42640, 0.31660)]         * duration_telephone[30, 60, 999]
+ b4                                          * feedback
/
U(app) = app[-1.42343]
+ b2_app.dummy[(n, 0.09786, 0.28830)]    * tech_app[1, 2]
+ b3_app[(n, 0.39204, 0.36730)]          * duration_app[30, 60, 999]
+ b4                                     * feedback
/
U(online) = online[-0.85626]
+ b2_online.dummy[(n, -0.16117, 0.27827)]  * tech_online[1, 2]
+ b3_online[(n, 0.05708, 0.35007)]         * duration_online[30, 60, 999]
+ b4                                       * feedback
/
U(f2f) =
  b1_f2f.dummy[-0.24854]              * admin_f2f[0, 1]
+ b3_f2f[(n, 0.19027, 0.13373)]       * duration_f2f[90, 120]
+ b4                                  * feedback
$


Thank you in advance for your help.

Kind regards,
Pakhi

[Michiel Bliemer]

There are several issues.

1. If your prior is positive, your unavailability level needs to be -999 since the utility needs to go to a large negative number to disappear from the choice set.

2. Some of your priors do not seem correct, it is not possible to have a prior value of 0.38 for an attribute with levels 30 and 60 as this would lead to a far too large utility. This seems to be an issue with all your duration attributes. Please use the same levels/units in model estimation and experimental design.

Michiel

[pakhi]

Thank you very much Prof. Bliemer. I fixed the errors I made and the code worked now.

Kind regards,
Pakhi