choice-metrics.com

Dear all,

I'm trying to generate an unlabelled design with 2 alternatives, each with 5 (non-linear) attributes of 5 levels + the status-quo(SQ) alternative. As you can see in my syntax below (all priors are set to 0 because this is a pilot), I have specified the utility of the SQ to be given by a constant and have included this constant in the AVC matrix for the computation of the efficiency measures. So, here is my question: Shall I count the SQ alternative as an independent choice observation-which implies that I need 9 rows in order to estimate my 17 parameters- or not, which in turn means I need 17 rows for my 17 parameters??

My intuition is to include them, since If we can calculate the choice probability of, let's say the first alternative and the status-quo, we can determine the probability for the second alternative. However, I'm not sure since SQ provides no information about the tradeoffs between the attributes.

Code: Select all

Design
;alts = alt1, alt2, SQ
;rows=????
;eff = (mnl,d)
;con
;model:
U(alt1) =b2.dummy[0|0|0|0] * X1[1,2,3,4,5] + b3.dummy[0|0|0|0] *X2[1,2,3,4,5] +
b4.dummy[0|0|0|0]*X3[1,2,3,4,5]+b5.dummy[0|0|0|0]*X4[1,2,3,4,5] /
U(alt2) =b2.dummy * X1 + b3.dummy *X2 +b4.dummy*X3+b5.dummy*X4 /
U(SQ)=b1[0] $

Best Regards,
Achilleas

Your intuition is correct, you count the SQ as an extra alternative, as it does provide extra information. So 9 rows are sufficient. Ngene also automatically checks for the correct number of rows, if you would try 8 rows, Ngene would tell you that this is not sufficient.

Probably it is best to go for 10 rows, as this would lead to an attribute level balanced design.

Michiel

Thanks a lot Michiel!