How to determine the sample size by Ngene?

[sukunta]

Dear Prof. Blimier,
I have tried to generate the choice set for a pilot study via efficient design. Thus, the prior was determined by the expected sign and value. First, I have determined

Code: Select all

Design
;alts=alt1*,alt2*,alt3
;rows=24
;block=2
;eff=(mnl,d)
; con
;model:
U(alt1)=b0[0.001]+b1.dummy [0.0004|0.0003|0.0002|0.0001|0.0001] * locate[5,4,3,2,1,0]+b2.dummy[0.0002|0.0001|0.0001] * pretest[3,2,1,0]+b3.dummy [0.0002|0.0001|0.0001] * posttest[3,2,1,0]+b4.dummy[-0.0002] *anonym [1,0]+ b5.dummy[0.0002] *method[1,0]+b6[-.0003]*price[0,100,300,700]/
U(alt2)=b0+b1*locate+b2* pretest+b3* posttest+b4*anonym+ b5.dummy *method+b6*price
$

. The d-error =0.209608, sp-estimate= 417885617.44.
Second, I have tried to increase all priors to 10 times. The d-error =0.209192, sp-estimate= 25276406.06.
Does the second better than the first? Can Ngene reduce the sample size in the other way? and How?
Sincerely yours,
Suukunta

[Michiel Bliemer]

For a pilot study where you do not have priors, I would use all near-zero priors. For example b6[-0.000001] * price[0,100,300,700] such that the utility contribution of each attribute is near-zero. With such noninformative priors, your sample size estimates are meaningless and should be ignored. There is no way to determine the sample size if you do not have reliable priors.

For the main study, you can copy the parameter estimates from the pilot study and use them as priors. Using such informative priors your sample size estimates will be meaningful.

Michiel

[sukunta]

Dear Prof Bliemer,
Thank you so much for your recommendation. Let me tell you that I have determined nearly zero prior for the new syntax. In addition, the priors were assumed the different values from a qualitative study as follows

Code: Select all

Design
;alts=alt1*,alt2*,alt3
;rows=24
;block=2
;eff=(mnl,d)
; con
;model:
U(alt1)=b0[0.00001]+b1.dummy [0.000004|0.000003|0.000002|0.000001|0.000001] * locate[5,4,3,2,1,0]+b2.dummy[0.000002|0.000001|0.000001] * pretest[3,2,1,0]+b3.dummy [0.000002|0.000001|0.000001] * posttest[3,2,1,0]+b4.dummy[-0.000002] *anonym [1,0]+ b5.dummy[0.000002] *method[1,0]+b6[-0.000004]*price[0,100,300,700]/
U(alt2)=b0+b1*locate+b2* pretest+b3* posttest+b4*anonym+ b5.dummy *method+b6*price
$

. I found that D-error=0.205269.
Sincerely yours,
Sukunta

[Michiel Bliemer]

That looks fine.