choice-metrics.com

Dear all,

I am still learning about CE and I would like to have your insights into a problem I am facing.

I am trying to make a D-efficient (before making a Bayesian efficient one) for a CE presenting 3 varieties alternatives to farmers: 2 hypothetical varieties (alt1 and alt2) and 1 reference that is commonly used by farmers (alt3).

I have conducted a pilot (N=40) to get priors to be included in the following design. I have only included the priors that have a significant value and put a "0" value for the others.

The problem is I can't seem to obtain a valid design after 1000 evaluations. NGene sends back the following message:
"There may be a problem with the specification of the design. A common problem is that the choice probabilities are too extreme (close to 1 and 0), perhaps because some or all of the prior values are too large".

Would you be able to help and point me towards the right direction/paper to solve the issue?

The design is the following:

Design

;alts = alt1, alt2, alt3
;rows = 18
;block = 3
;eff = (mnl,d)
; model:

U(alt1) = b1.dummy[1.100|0]*yld[2,1,0] + b2.dummy[0|0]*mat[2,1,0] + b3.dummy[0|0]*dor[2,1,0] + b4.dummy[1.565|1.648]*lbr[2,1,0] + b5.dummy[0.798|1.051]*tbs[2,1,0] + b6[-0.001]*pri[1800,2300,2700,3500,3800,4200]/
U(alt2) = b1*yld + b2*mat + b3*dor + b4*lbr + b5*tbs + b6*pri/
U(alt3) = b0[0] + b7[1.100]*yldref[1] + b8[0]*matref[1] + b9[0]*dorref[0] + b10[0]*lbrref[0] + b11[1.051]*tbsref[1] + b6*priref[3250] $


Best wishes,

Carla

Hi all,

As a follow up from my previous message, It looks like the problem comes from specifying the utility of the reference alternative (all attribute levels known). If I don't include a utility function for alt3 but instead include the corresponding constant in alt1 and alt2, then NGene produces a design. Would that be a valid alternative?

Best wishes,

Carla

I believe that the syntax below does what you are looking for, I have made several changes. The reason why your syntax did not work is because you used different parameters for alt3 and since these levels do not vary these parameters cannot be estimated. But you can estimate generic parameters across the three alternatives as I have done below.

Code: Select all

Design
;alts = alt1*, alt2*, alt3
;rows = 18
;block = 3
;eff = (mnl,d)
;alg = mfederov(candidates = 2000)
;require:
alt3.yld = 1,
alt3.mat = 1,
alt3.dor = 0,
alt3.lbr = 0,
alt3.tbs = 1
; model:
U(alt1) = b1.dummy[1.100|0]*yld[2,1,0]
        + b2.dummy[0|0]*mat[2,1,0]
        + b3.dummy[0|0]*dor[2,1,0]
        + b4.dummy[1.565|1.648]*lbr[2,1,0]
        + b5.dummy[0.798|1.051]*tbs[2,1,0]
        + b6[-0.001]*pri[1800,2300,2700,3500,3800,4200](2-4,2-4,2-4,2-4,2-4,2-4)
        /
U(alt2) = b1*yld + b2*mat + b3*dor + b4*lbr + b5*tbs + b6*pri
        /
U(alt3) = b0[0] + b1*yld + b2*mat + b3*dor + b4*lbr + b5*tbs + b6*pri_ref[3250]
$

Michiel

Dear Michiel,

Thank you very much for your answer. Your syntax works indeed. That's really helpful.

Have a nice day,

Carla