Hello,
We were asked to provide the statistical power of our ordinal model. Initially, we did not find evidence for any effects and because the sample size is small (n = 36), the model may have low power to reveal any effects. Despite finding some evidence for effects in an updated model with a phylogenetic covariance matrix and more appropriate priors, it could still be useful to check the statistical power by simulation-based power analysis, at least to please the editor :).
Our model has the following structure:
mod_phylo <- brm(Trend ~ treatA + treatB + treatC + treatD + treatE + (1|gr(Species, cov = phylo_cov)),
data = dt,
data2 = list(phylo_cov = phylo_cov),
family = cumulative(logit),
prior = priors)
The response variable, Trend, has three ordered levels (Decreasing, Stable, Increasing). Each predictor is a categorical variable with two levels (Yes, No). We hypothesized that the application of each treatment might have an effect on the Trend categories. It means we compare the probabilities between the Yes and No levels of a predictor for a given Trend category.
We would be very grateful for any advice regarding the implementation of phylogeny, as well as how to define the differences between the Yes and No levels of the treatments in the power analysis. The series of posts on Bayesian power analysis by @Solomon is very useful, though it did not help with complex models like ours.
Thanks in advance!
Jan