Power analysis for conjoint survey experiments. (forced choice)

[Tom Ford]

Hello,

I am planning a conjoint survey experiment and would like to perform some power calculations.

Is there anyone that can provide some guidance on how to do this in Stata?

The objective is to understand teacher opinions of school reforms. The plan is to use a two-stage cluster sampling strategy.
I will first select a random sample of schools (~40) and then interview randomly sampled respondents in each school (~30).

How can I do power calculations that account for the fact that responses will be clustered at the school level?

Note that this article shows that answers are clustered at the individual level is unlikely to have a large effect on calculations https://osf.io/preprints/socarxiv/9yuhp/.

Few additional features of the design:

It will be a forced choice model.
Respondents have to decide between two school reform profiles.
Each respondent will have ~4 tasks.
The number of attributes is 8.
The maximum number of levels is 5.
Each respondent is forced to choose between two reform profiles.
The total number of tasks is 4 tasks.
My hunch is that my effect size will be between 0.02 and 0.10

Note that my N= 40*30*2*4=9600. In other words, the number of respondents times the number of profiles times the number of tasks.

Is there a way I can use the

power

command to make power calculations for similar designs?

I want to know if I am sufficiently powered (.8?) or whether the number of clusters (40) dramatically impacts my ability to find the significant coefficients in the population.

Thanks a lot in advance for your help

[Oskar Lindgren]

Hi Tom,

Did you find any way to do this? Having some similar issues with a study of mine.

Best,
Oskar

[Tom Ford]

You can try to use this shiny app https://mblukac.github.io/posts/2020...poweranalysis/