pilot study for a MMNL model

[mires]

HI Manuel,

If your survey is Web one.. What I'm used to do, is to use firt observations, to get some good priors,
and plugging a new design of expêriments using those priors.
You won't lose the first observations but you will get higher precision over the following ones

Best regards
Naji

[mnarjes]

Dear Choicemetrics team,

Some months ago I asked you for advice regarding the design for my research. My data collection went very well and I am very thankful for having been able to count on your expertise!

Now I'm planning to do a new survey for a comparative study in another area, where the respondents share a common language and cultural background with those from my last survey (hence, it shouldn't be necessary to change the attributes nor the questionnaire format). Would you recommend using the same design I used for my last survey, or should I use the corresponding coefficient estimates as priors in a pilot study for the new survey?

Thank you in advance for your time and advice!

Best regards,

Manuel

[Michiel Bliemer]

Dear Manuel,

If the aim to do another separate study and not to compare the results with the earlier study? Then use the estimated coefficients as priors and create a new design.
Is your aim to compare the two studies? Then changing the experimental design makes the two studies less comparable (the design impacts upon the scale parameter in the logit model), and in that case I would opt for using the same design.

Michiel

[mnarjes]

Dear Michiel,

Thank you very much for your advice!

Best regards,

Manuel