choice-metrics.com

Dear Michiel/ Andrew,

I am trying to generate the following design which also includes interaction effect. I got the priors from a pilot survey. At least three alternatives are repeated in different choice sets (not within choice sets). I was trying to figure our why this is happening but I am confused. Do you have any idea why this is the case in my design. I have only one attribute which is dummy but the rest are continuous (linear in utility parameter). I thank you very much for the assistance. Kind regards

Design
; alts = alt1*, alt2*, SQ
; rows = 12
; eff = (mnl,d)
; Model :
U(alt1) = amo[0.1]*amount[0,1,2]+fort.dummy[0.7]*forti[1,0]+cost[-0.02]*price[20,30,45,60,75,90]+amf[-0.12]*forti.dummy[1]*amount/
U(alt2) = amo*amount+fort*forti+cost*price+amf*forti.dummy[1]*amount/
U(SQ) = b[0]$

To understand your problem, can you please provide an design output that illustrates the problem?

Dear Michiel

Thank you for the swift response. As you can see alternative 1 in choice task is repeated in choice task 4 as alternative 2. Also alternative 2 in choice task 1 is repeated in choice task 12 as alternative 1. There are also other similar cases in the design.




MNL efficiency measures

D error 0.100469
A error 0.977896
B estimate 74.062959
S estimate 319.351926

Prior amo fort(d0) cost amf
Fixed prior value 0.1 0.7 -0.02 -0.12
Sp estimates 244.82875 16.282225 3.73975 319.351926
Sp t-ratios 0.125264 0.485735 1.013526 0.109678

Design
Choice situation alt1.amount alt1.forti alt1.price alt2.amount alt2.forti alt2.price alt1.forti.dummy[1]*alt1.amount alt2.forti.dummy[1]*alt2.amount
1 2 1 45 0 0 30 2 0
2 1 1 20 2 0 60 1 0
3 0 1 90 1 0 75 0 0
4 0 1 30 2 1 45 0 2
5 2 1 60 0 1 45 2 0
6 2 0 60 0 0 30 0 0
7 1 0 75 0 1 20 0 0
8 2 0 20 1 1 75 0 1
9 0 0 75 1 0 90 0 0
10 1 1 90 2 0 20 1 0
11 1 0 45 2 1 60 0 2
12 0 0 30 1 1 90 0 1

Why do you think this is a problem?
There are no repititions of choice tasks, repititions of alternatives is perfectly normal. It is even the way sequential orthogonal designs are created, in which each alternative is created twice. It is the differences between alternatives that is of interest in logit models, not the alternatives themselves.

Okay thank you Michiel.