Power Analysis & Sample Size estimation

[Gokay Kirtil]

Hi all !

For my research, before discussing the methodology section of the study, I need to add a section regarding the minimum number of sample needed. In social sciences, it's generally discussed under the power analysis topic. So for a Bayesian SEM to be conducted, I think I need a different methodology than frequentist power analyses. Is this bayes factors ?

I found this on the internet : https://www.andywills.info/rminr/power-bayesian.html 

Can you tell me that if this one is enough for me or do you we have any other alternatives ?

Regards, 

Gokay

[Mauricio Garnier-Villarreal]

Hi Gokay

There is no agreement on what "power analysis" means in Bayesian methods.

I have seen 2 main recommendations

- The Bayes Factor approach as you shared: very similar to the frequentist power anlsysi, and mainly using the BF instead of the p-value

- Stability of the posterior distribution: this one looks at the sample size when the SD of the posterior distribution doesnt shrink anymore. So, the minimal sample size for the most stable variability of the posterior.

Either way, I havent seen it applied in SEM. You could take from examples of the package simsem, that can do power analysis in SEM with lavaan (see example 18 https://github.com/simsem/simsem/wiki/Vignette ). But would need to adjust the code to use blavaan. And this will also increase the computation time

In general, if you use the BF approach, I think that should be enough for the proposal. But that will always depend on the review board, and how familiar they are with these methods

take care