How many sample sizes do I need for a pilot study?

[sukunta]

Dear Prof. Michiel Bliemer,
I design syntax for pharmacists' counseling services. My syntax is

Code: Select all

design
;alts=alt1*,alt2*,alt3
;rows=12
;block=2
;eff=(mnl,d)
;con
;alg=mfederov
;model:
U(alt1)=b0[0.000001]+b1.dummy[0.000002]*private[1,0]+b2.dummy[0.000002]*open[1,0]+b3[0.000002]*time[10,20,30]+b4.dummy[0.000002|0.000001]*channel[2,1,0]/
U(alt2)=b1*private+b2*open+b3*time+b4*channel/
U(alt3)=b0[0.000001]
$

D-error is 0.153 and Sp estimates are large values. How many sample sizes do I need for a pilot study?
Sincerely yours,
Sukunta

[Michiel Bliemer]

Hi,

The S-estimates produced by Ngene are only useful if you have specified informative priors from the literature or a pilot study. In this case you have used (near) zero values, therefore your S-estimates will be extremely large and should be ignored since they have no meaning.

For a pilot study you typically use 10% of your total sample. This sample may not always be sufficient to obtain reliable priors, there is no magic number for how many respondents you need in a pilot study.

Michiel

[sukunta]

Dear Prof. Michiel Bliemer,
Thank you so much for your reply. The sample size for the main survey is 300, my pilot study is 30, thus the sample size for block=2 is 60. If I have a limitation of the sample size, can I conduct block 2 design in 30? or I use the design with no block and 6 rows but the D-error of the model is 0.33432.

Sincerely yours,
Sukunta

[Michiel Bliemer]

The D-error reported by Ngene assumes a single responent facing all 12 choice tasks, i.e. 1 design replication. If you have 300 respondents and you block the design in two, then the number of design replications is 150, which would be the "sample size" assumed in Ngene. So it depends whether you compute sample size in terms of respodents or in design replications. If you have 300 respondents (not design replications) then you would typically collect data from 30 respondents in your pilot, meaning each block is given to 15 respondents such that you will have data from 15 design replications. Of course there is nothing stopping you from taking 20% of your sample for your pilot to collect data from 60 respondents, giving each block to 30 respondents.

Michiel

[sukunta]

Dear Prof. Michiel Bliemer,
Thank you so much for your reply.

Sincerely yours,
Sukunta