andrewheiss / diff-means-half-dozen-ways Goto Github PK

Hi. Thanks for all this work! I am trying to follow it, but cannot find tidyMCMC. It doesn't seem to be in the broom package anymore, and just using tidy gets the error "Warning: Error in : No tidy method for objects of class brmsfit". I can see https://github.com/bragqut/tidymcmc but that code is last modified 6 years ago, and doesn't seem to be right.

Thanks for any help - I have no idea about R, but am trying to fix that!

First of all, many thanks for this helpful thread!
It appears to me that there are slight mistakes in your calculations for the effect sizes, but please correct me, if I am wrong.
Your calculation is as follows:
mutate(cohen_d = diff_means / sqrt((b_sigma_genreAction + b_sigma_genreComedy)/2),
cles = dnorm(diff_means / sqrt((b_sigma_genreAction + b_sigma_genreComedy)), 0, 1))

but according to Cohen (1988) sigma for unequal variances should be calculated sigma' = sqrt((sigma1^2 + sigma^2)/2), and according to McGraw and Wong (1991), also this pooled standard deviation should be used. Additionally, I do not understand the approach with dnorm. Shouldn't you use pnorm? Accordingly the code should be in my opinion:

mutate(cohen_d = diff_means / sqrt((b_sigma_genreAction^2 + b_sigma_genreComedy^2)/2),
cles = pnorm(0, mean=(b_genreAction-b_genreComedy), sd = sqrt((b_sigma_genreAction^2 + b_sigma_genreComedy^2)), lower.tail = FALSE))

Looking for an answer!
Best

Hey,
sorry, I am too lazy for a pull request 🙈

Two minor issues:

• tidyMCMC is part of broom.mixed now, so should add this package as a dependency

• second one is about extracting the ci's:

infer_simulation <- tibble(estimate = diff_means$stat,
                           conf.low = boostrapped_confint$`2.5%`,
                           conf.high = boostrapped_confint$`97.5%`)

should be changed to:

infer_simulation <- tibble(estimate = diff_means$stat,
                           conf.low = boostrapped_confint$lower_ci,
                           conf.high = boostrapped_confint$upper_ci)

Thanks for the great tutorial!