a question regarding the a-pripri opt-out probability.

[Anat Tchetchik]

Dear all NGENE users!
I have ran a design with 4 alternatives (one of which is a status-quo/opt-out) with prior means based on pilot study. What bothers me is that when looking atthe probabliltes created by the NGENE for tihs design -the opt-out alt. has always small probabilties and hence never chosen, while the other 3 alts. are being chosen evenly.
Is that a problem? shold I changed the prior means somehow so that the opt-out ootion will also be chosen sometine?

Below is the design syntax:

Design
;alts = alt1, alt2, alt3, alt4
;rows=72
;block=9
;eff=(mnl,d,fixed)
;bdraws = halton(400)
;model:
U(alt1) = b2[-0.7] * price[0.05,0.10,0.15,0.200,0.250,0.30] + b3.dummy[(u,0.2282,0.5928)|(u, 0.7542, 1.0376)]*bottom[0,1,2]+b4.dummy[(u,0.2601,0.5381)|(u,0.6157,0.8919)]*palagi[0,1,2]+ b5[(u,0.083,0.3478)] * comercialReg[0,1] + b6[(u,-0.4081,-0.1381)]*recreationalReg[0,1]+b7[(u,0.5242,0.764)]*facilities[0,1]/
U(alt2) = b2 * price + b3 * bottom+ b4*palagi+ b5 *comercialReg+ b6 * recreationalReg+ b7*facilities /
U(alt3) = b2 * price + b3 * bottom+ b4*palagi+ b5 *comercialReg+ b6 * recreationalReg+ b7*facilities
$

Best regared and many thanks in advance for any advice!
Anat

[Michiel Bliemer]

If the prior parameters are estimated from a pilot study (including the opt-out I suppose), you should not just manually change the parameters.
However, I think that maybe you did not estimate the model including the opt-out? Because the opt-out is a labelled alternative, and you should include a constant. So either include a constant in the opt-out alternative (so you specify U(alt4) = con[..] or something, or you should include a constant in the estimation for all other alternatives (so adding "con" to alt1, alt2, and alt3). This way, in the estimation process, it can account for the number of times the opt-out is chosen and the priors should correctly reflect the probabilities for each alternative.

If you did not have an opt-out in your pilot study, then you will have to guess this constant. So rather than changing the other priors (just leave them as they are), add the constant in the alt4 alternative and make a guess on how often the opt-out should be chosen within a certain design, and adjust the prior value of the constant manually.

Michiel

[Michiel Bliemer]

Another thing, you probably want to tell Ngene that alt1, alt2, and alt3 are unlabelled alternatives, such that Ngene can take out any strictly dominant alternatives. So I suggest changing the syntax to have:
;alts = alt1*, alt2*, alt3*, alt4

[Anat Tchetchik]

Dear Michiel!

Thank you very much for your answer! it actually solved the problem (we included the opt-out in the pilot so we had a constant for alt 4).
I really appreciate working with NGENE - it opens a whole lot of research opportunites for me.
Thanks you again.
Best
Anat