Page 1 of 1

Could I improve my design more?

PostPosted: Sat Apr 08, 2023 11:51 pm
by sukunta
Dear Michiel Blimer,
I have run the syntax for around 8 hours by
Code: Select all
Design
;alts=alt1*,alt2*,alt3
;rows=24
;block=2
;eff=(mnl,d,mean)
; con
;model:
U(alt1)=b0[(n,1.007,0.563)]+b1.dummy [ (n,0.161,0.263) | (n,0.032,0.299)|(n,-0.002,0.317)|(n,0.184,0.228)|(n,-0.206,0.292)] * locate[5,4,3,2,1,0]+b2.dummy[(n,0.334,0.189)|(n,0.103,0.214)|(n,0.187,0.256)] * pretest[3,2,1,0]+b3.dummy [(n,-0.098,0.228)|(n,-0.012,0.172)|(n,0.017,0.1779)] * posttest[3,2,1,0]+b4.dummy[(n,0.068,0.14)] *anonym [1,0]+ b5.dummy[(n,0.107,0.174)] *method[1,0]+b6[(n,-0.001,0.0003467)]*price[0,100,300,700]/
U(alt2)=b0+b1*locate+b2* pretest+b3* posttest+b4*anonym+ b5.dummy *method+b6*price
$
and the prior with s.e. from a pilot study. The d-error was 0.2089, however, the sp-estimate was a very large number. Could I improve my design more?
Sincerely yours,
Sukunta Muadthong

Re: Could I improve my design more?

PostPosted: Tue Apr 11, 2023 4:34 pm
by Michiel Bliemer
Based on the means of your priors, the sample size estimates indicate that:
- ASC should be easy to estimate
- Pretest dummy coefficients may require a few hundred respondent for most of them, but some categories have very small coefficients and may not be statistically significant if the priors turn out to be true
- Posttest dummy coefficients may require a few hundreds respondents and do not seem a worry
- Anonym coefficients seem to indicate that this attribute is not very important, so this attribute may not come out as statistically significant, which could also be a result, not everything needs to be important in decision making
- Method dummy coefficients may require a few hundreds respondents and do not seem a worry
- Price coefficient should be easy to estimate

Given that there is a large amount of uncertainty about the priors, you should treat some of the sample size estimates as only a best guess but could be very different. I do not see any major issue, I think that the sample size estimates look what they typically look like. There is nothing else you can do to improve them.

Michiel

Re: Could I improve my design more?

PostPosted: Fri Apr 14, 2023 1:50 am
by sukunta
Dear Prof.Bliemer,
Thank you so much for your recommendation.
Sincerely yours,
Sukunta