Sample size calculation after the pilot

[a.tran@sydney.edu.au]

Dear Michiel,
I conducted the pilot, and Nlogit gave me. I updated the pilot design with the new beta to obtain the main design, and now I want to calculate the sample size for the main design.


Beta SE p-value
FP -0.08925 0.01534 0.0000
FN -0.26932 0.03519 0.0000
TS1 -0.58224 0.29286 0.0468
TS2 -0.42793 0.27977 0.1261
SE1 -0.75607 0.48958 0.1225
SE2 0.10498 0.45623 0.8180
Cost -0.000027628 0.000006469 0.0000

I referred to your paper published in 2013 with John, "Sample size requirements for stated choice experiments". Am I right to calculate N for each attribute (FP, FN, TS1, TS2, SE1, SE2, Cost), then take the maximum of the sample size needed? Why, for example, for FP, N >=((1.96*0.01534)/(-)0.08925) ^ 2 = 0.113487. Something must be wrong with my understanding (I am very new to DCE). Could you please guide me on where I can find info to calculate the sample size for DCE? Can Ngene do sample size calculations for us?


Below was what I ran in nlogit


create;if(TS=2)TS2=1$

create;if(SE=1)SE1=1$

create;if(SE=2)SE2=1$


nlogit

;lhs = choice,cset,altij

;choices = A, B

;model:

U(A) = ASC1 + FPb*FP + FNb*FN + TS1b*TS1 + TS2b*TS2 + SE1b*SE1 + SE2b*SE2+ Costb*cost/

U(B) = FPb*FP + FNb*FN + TS1b*TS1 + TS2b*TS2 $


Thank you very much!

[a.tran@sydney.edu.au]

Hi Michiel,
I got the answer from John, so that I will save questions for others.

Thank you and have a great weekend.

Anh

[Michiel Bliemer]

Ok great!

For others interested in the answer:

The sample size calculation in our paper is based on the standard error for a single respondent. The standard error from the pilot study therefore needs to be normalised by the number of respondents in the pilot study, so use se/sqrt(samplesize) instead of se. Ngene also does these calculations based on a generated experimental design (in the output they are listed as S-estimates).

Michiel