Efficient design with SQ alt. having dummy variables

[AnzeJap]

Thank you! I have tried with this restriction but it states that it could not find a design that meets them - even if I used relaxed ones. Is this a problem with a design? OR it simply cannot for some reason find one. I have also increased the number of rows to 36.

[Michiel Bliemer]

The syntax runs fine for me. I am using version 1.3.0.

You can try running it again since it may sometimes take a bit of time to find the initial design. If you increase the number of rows then of course you also need to adjust the restrictions. Relaxing the restrictions makes it easier to find a design, e.g., 4-8 is more restrictive than 2-10.

[AnzeJap]

Dear Michiel,

after running the code and reviewing the design there seem to be a few dominant alternatives (3 out of 36). The reason I guess are priors, which are to some extent (the sample was just over 50) unreliable in terms that they probably do not represent expected preferences as well as I hoped. For example the payment parameter (WTA) is not significant, negative and extremely small, but we expect it to significant and positive in the final sample. Thus, even looking at the choice probabilities there are no dominant alternatives, but after inspecting each of them some simply seem to have a very large probability of being chosen - large payments for forest owners in return of almost no additional forest activities. So my question is, do I need to leave the design as it is constructed by running the code or can I adjust some alternatives, like switching two payment levels between two alternatives in the same choice set? I did so almost every time in past when I was constructing orthogonal designs but I am not sure how acceptable this is for an efficient design as using priors is to avoid dominant alternatives in the first place.

Thank you!
Anže

[Michiel Bliemer]

The solution is simple: you need to tell Ngene that the alternatives are generic and that you want to check for dominant alternatives. You can do that by adding an asterisk to all alternatives of the same type as follows:

;alts = al1*, alt2*, sq*

Michiel

[AnzeJap]

Sure, but in case you detect dominant alternatives via post festum inspection because you expect priors are not entirely true, and checking dominance according to priors might not be completely reliable. Can you e.g. remove dominance 'by hand' - switching payment levels between two alternatives? Perhaps I am not making too much sense.

[Michiel Bliemer]

Yes it is fine to manually change the attribute levels in your design if you wish,as long as you do not introduce perfect correlations across attributes (which is unlikely). The design may be slightly less efficient, but that is generally not a major concern, so feel free to make slight changes to the design that you deem appropriate.

Michiel

[AnzeJap]

Dear Michiel,

thank you very much!

[AnzeJap]

Dear Michiel,

just to let you know that the design you advised me on worked OK, and we have done the survey and analyzed the results - mostly they make a lot of sense It took quite some time as we had to sample face-to-face in addition to the web survey. Thank you very much for your help!

Best regards,
Anže

[Michiel Bliemer]

Great to hear!