Error: Aborting run. After approx 10 min, no design found

[aguerrero]

Dear Choice-Metrics team,

Thank very much for developing this software.
I am trying ngene for a pilot study. My basic specification is below which I expect to estimate through a mixed logit. In order to obtain the priors for a Bayesian efficient design, I am first estimating also a Bayesian eff. design based on a pilot with 20 to 30 respondents (2-4 choice sets per respondent).

By the exception of orthogonal design, others design seems to work (fractional, MNL-D-efficient with code effects), however, the syntax bellow is aborted after 10 minutes.

This a shorter syntax, since the initial specification had an status-quo alternative and some constrains to require more realistic choice tasks.
Please I would really appreciate any help.

Best regards,

Andrea
-------------------------------------------------
Design
;alts = alt1*, alt2*, alt3*
;rows = 35
;eff = (mnl,d,mean)
;rseed=12345
;block = 3
;model:
U(alt1) = b1[0.659] + b2[(u,-2,-0.8)]* A[0,1,2,3,4,5] + b3[(u,-0.06,-0.02)]*B[0,1,2,3,4,5] + b4[(u,-1.3,-0.2)]*C[0,1,2]+
b5[(u,-0.8,-0.3)]* D[0,1,2,3,4,5,6,7,8] + b6 * E[1,2,3] + b7 *F[1,2,3,4,5,6] + b8 * G[1,2] + b9*H[1,2,3,4,5] /
U(alt2) = b1+ b2 * A + b3 * B + b4 * C + b5 * D + b6 * E + b7 * F + b8 * G + b9 * H /
U(alt3) = b2+ b2 * A + b3 * B + b4 * C + b5 * D + b6 * E + b7 * F + b8 * G + b9 * H
$

[Michiel Bliemer]

The warning indicates that there are many cases of dominancy. By using the * behind the alternatives, Ngene checks for dominance. You are asking for 35 choice tasks (which is quite a lot) without any dominant alternative, however the default algorithm of Ngene cannot find such a design that is so heavily constrained.

Instead, you can use the modified Federov algorithm by adding for example:

;alg = mfederov(candidates = 2000)

It takes some time for Ngene to find designs, but give it some time. Note that the modified Federov does not guarantee attribute level balance, so you may wish to impose some lower and upper bounds on the number of times each attribute appears, i.e. + b4[(u,-1.3,-0.2)]*C[0,1,2](5-15,5-15,5-15), which requires that each attribute level appears between 5 and 15 times.

[aguerrero]

Thanks Michiel.