Repeating or Updating experimental design?

[Anat Tchetchik]

Dear all,

Three years ago, we created a Bayesian D-error efficient design (after piloting it for the priors) on consumers' preferences for E-waste recycling.
The data generated from this design was analysed via mixed-logit.
The Journal which research paper is under review with, have asked to repeat the experiment (as back then, the law enforcing E-waste recycling was new).
The prices have remained the same and so are the options (labelled)..
My question is should I use exactly the same design? Or is it better to update it based on the mixed-logit coefficients obtained, and create new experimental design wherein these coefficients serve as priors.
My main purpose is to have the best basis for comparison with the previous experiment..

Any help will appreciated,

Anat

[Michiel Bliemer]

I would advise to keep exactly the same design, since in comparisons this rules out differences due to a change in the survey.

The efficiency gain that you will get is likely to be relatively small and if sample size is not an issue and if the design worked fine last time, then I do not see an urgent need to make a change.

Michiel

[Anat Tchetchik]

Thank you very much Michiel! as always your advice is very valuable.

[Anat Tchetchik]

One last thing, we also have to repeat a Bayesian D-error efficient design on cell-phones' end-of-use option experiment (i.e. sell to recycling firm, trade-in or sell via second hand market). Yet here the prices have changed a bit (decreased) roughly at the same proportion across the different alts.
The question is whether changing (decreasing proportionally) the levels of the price for each alts. will make comparison with the previous experiment irrelevant?
Thanks
Anat

[Michiel Bliemer]

I think that you have two options:

(i) Keep the exact same attribute levels since these levels are hypothetical anyway (you are using a range) and do not need to exactly reflect the market prices.
(ii) Decrease the price attributes but keep the same experimental design, i.e. level 1 remains level 1, but the exact level is slightly lower.

Both options are fine since in both cases you keep as much similar as possible in order to enable statistical comparisons with high power.

Michiel

[Anat Tchetchik]

Thank you again Michiel!