choice-metrics.com

Dear Ngene expert:
I’m more thankful than I can express for your previous reply. In order to reduce mistakes, I want to confirm my syntax with you finally.
This is my syntax:
Design
;alts=alt1*,alt2*,alt3
;rows=24
;block=4
;eff=(mnl,d)
;model:
U(alt1)=b1.effects[0.001]*A[1,0]+b2.effects[0.003|0.002|0.001]*B[3,2,1,0]+b3.effects[0.001|0.001|0.001]*C[3,2,1,0]+b4[0.001]*D[0,1,2,3]+b5[-0.001]*E[0,1,2,3]/
U(alt2)=b1*A+b2*B+b3*C+b4*D+b5*E/
U(alt3)=b0[0]
$

1 My variable D and variable E are continuous variables but I just need 0,1,2,3 to represent 100 yuan, 200 yuan, 300 yuan, 400 yuan instead of the number between 100 yuan and 400 yuan(such as 150 yuan,350 yuan etc.). Is the above syntax correct?

2 As shown above, the attribute level order of attribute A,B,C is 3,2,1,0, and that of D,E is 0,1,2,3. They have different orders, which seems a little strange, but is that right? Will this affect the efficiency of the design?

3 You told me before, when I use priors that are reasonable, the S error and S estimates will be meaningful. So if I get the reasonable priors, is the value of S estimates represents the sample size that I need to investigate? That is, assuming the value of S estimates is 1000, do I need to investigate 1000 people? If so, what if my S estimates is very very large? What does the value of S error stand for?

Thank you very much for your help. Best wishes for you

1. For efficient designs you need to put in the attribute levels that you are going to use in model estimation. Typically you code continuous variables with the actual levels, e.g. I would use 100 for 100 yuan. Since you are essentially using zero priors (only small values for the sign), the actual level does not matter, but once you use appropriate priors, the attribute levels matter.

2. Yes that is fine, no problem. The last level is the reference level and you can order them any way you want as long as the priors are consistent with that order.

3. The S-estimate will give you the minimum sample size you need for each parameter to be estimated at a 95% level of significance. Often you want to have a higher significance level so it is an estimate of the minimum. But since priors are not precise, the sample size estimate should also be treated as a rough guess only. An S-estimate of 1000 means indeed collecting data from 1000 people. If your design is blocked in 4 as in your case, you would need 4000 people since one respondent only sees 25% of the complete design. If your S-estimate is very large then either you need a very large sample size or your priors are not very accurate.

Michiel