choice-metrics.com

I have run the following design and the output looks good, except I am getting some duplicates. Specifically, all attribute levels for alternatives 2 and 3 (or 4 and 5) are identical in some rows. I do not think this should be happening. Can someone please advise? (I only show the code for the first segment because the others are almost identical and my problem does not have to do with pivoting). Many thanks!


Design

;alts(Luxury) = car1, car2, car3, car4, car5
;alts(Trucks) = car1, car2, car3, car4, car5
;alts(CarsMid) = car1, car2, car3, car4, car5
;alts(CarsEcon) = car1, car2, car3, car4, car5

;rows = 120
;eff = fish(mnl,d,mean)
;bdraws = halton(500)
;fisher(fish) = Module7(Luxury[.08], Trucks[.445], CarsMid[.152], CarsEcon[.323])

;model(Luxury):
U(car1)= a1*range_gas[300] + b1[(n,-.000142,.0000137)]*price.ref[78219]+ b3[(n,-.174,.0154)]*refuel_gas.ref[.234] + b33[(u,0,0)]*refuel_electric.ref[.234] /

U(car2)=b1*price.piv[5%,15%,25%,50%]+ b2ev[(n,.0263,.002287)]*range_ev[50,75,100,200]+ b3*refuel_electric.piv[-30%,-40%,-50%,-60%]+ b5*hov[0,1]+b6*range_ev*refuel_electric + b7[.001]*range_ev*refuel_gas + b8*range_ev*price/

U(car3)=b1*price.piv[5%,15%,25%,50%]+ b2ev[(n,.0263,.002287)]*range_ev[50,75,100,200]+b3*refuel_electric.piv[-30%,-40%,-50%,-60%]+ b5*hov[0,1]+b6*range_ev*refuel_electric + b7[.001]*range_ev*refuel_gas + b8*range_ev*price/

U(car4)= a1*range_gas[300] + b1*price.piv[5%,15%,25%,50%] + b2phev[(n,.0263,.002287)]*range_phev[10,20,40,60] + b3a*refuel_electric.piv[-30%,-40%,-50%,-60%] + b3b*refuel_gas.piv[0%,-10%,-20%,-30%] + b5*hov[0,1] + b6*range_phev*refuel_electric + b7[.001]*range_phev*refuel_gas + b8*range_phev*price/

U(car5)= a1*range_gas[300] + b1*price.piv[5%,15%,25%,50%]+ b2phev[(n,.0263,.002287)]*range_phev[10,20,40,60]+b3a*refuel_electric.piv[-30%,-40%,-50%,-60%] + b3b*refuel_gas.piv[0%,-10%,-20%,-30%] + b5*hov[0,1] + b6*range_phev*refuel_electric + b7[.001]*range_phev*refuel_gas + b8*range_phev*price

I ran your syntax, and Ngene does what you ask it to. You have specified your alternatives as labelled, therefore having identical attribute levels across alternatives would not be a problem. However, I am assuming you actually require some of the alternatives to be unlabelled, i.e. alternatives 2 & 3, and alternatives 4 & 5. You then have add an * to the alternatives, i.e. car2*, car3* in the ;alts property. When you do that, you will notice that Ngene has a hard time finding a design of 120 rows in which there are no repeating attribute levels, as Ngene uses randomisation algorithms. I can only guess that that is the problem, as I am not sure what is exactly the problem is due to the highly complex syntax (pivoting, Bayesian, interactions, ...).

You probably would like to indicate that alternatives 2 and 3 are unlabelled and belong together, and that alternatives 4 and 5 are unlabelled and belong together. The command ;alts = car1, car2*,car3*, car4*, car5* will not work, as car2 and car4 are different and cannot be unlabelled. There is at this stage no option in Ngene to select two or more groups of unlabelled alternatives. We have not encountered the need for this before, but your syntax clearly indicates this could be important, so we will add it to our wish list.

Michiel,

Many thanks for your reply. You properly diagnosed the problem. I'll rethink my design to see if I can avoid having two different pairs of unlabeled alternatives.

Thanks again.