On the other hand, it allows you to explore boundaries of your conclusions. If say, we did find strong evidence for the alternative for some JNDs, under which JND assumptions would this hold.

yeah, maybe we should make this clearer. It looks like as if even for the most liberal ROPE, evidence for the alternative hypothesis is rather weak. So our results indicate: we don't know either way.

oh that is a great idea! (and a nice argument for using ranges.

I found your vignette already super clear and straightforward, but great to hear its an ongoing process. We discovered your package during writing this paper and where so happy that you guys had already implemented it (particularly the logspline estimation). Thanks for this amazing package!

You might want to start with our 2019 tutorial then before doing the latest one :)

Im pretty sure @meanwhileina.bsky.social throws around DAGs left and right. @casillas.bsky.social and @scoretta.bsky.social might also have used them.

We hope its useful. Let us know what you think and whether it all makes sense to you. It will likely remain a preprint and can be changed if things are unclear. end/🧵

osf.io/preprints/ps...

...and its sequel we wrote 6 years later: "Bayesian regression modelling (for factorial designs): A tutorial"

osf.io/preprints/ps...

Now it has been 6 years again and we thought it is time to drop Part 3 sticking to the same phenomenon and data set as the Original. 7/🧵

We simply felt that this simple and intuitive method should be communicated in our own way.

This tutorial is basically a long awaited 2nd sequel (!?) of
@bodowinter.bsky.social's original tutorial on linear mixed effects models: bodo-winter.net/tutorials.html ... 6/🧵

It also forces us to carefully think about what we mean with meaningful differences. A feature of this method, not a bug.

None of this is new of course. Not only that, but it has already been baked into a wonderful #rstats package: bayesTestR

easystats.github.io/bayestestR/

5/🧵

3. Relating these probabilities allows us to calculate the Bayes Factor for and against a meaningful difference between conditions.

Advantage of this combination of methods: This is computationally very slim (as compared to e.g. bridge sampling) and it's intuitive. 4/🧵

2. quantify the probability of posterior samples being inside the ROPE (i.e. practically equivalent to 0 = no meaningful effect) relative to being outside the ROPE (evidence for the alternative hypothesis). And we do that for both the posteriors before seeing the data and after seeing the data. 3/🧵

Using an example from speech production research, we propose to

1. define a region of practical equivalence (ROPE) through what we consider the smallest effect size of interest (SESOI)

(that's tough, and here we try that with reference to just-noticeable differences from perception research) 2/🧵

“Simine Vazire talks the talk and walks the walk,” writes @briannosek.bsky.social @cos.io.
🏆 The €150,000 award allows her to continue prioritizing research rigor and honors the mentors, colleagues, and students who made her work possible.
Short interview: www.youtube.com/watch?v=0H3g... (3/4)

Unfortunately, I don't have much time for additional meetings currently. One major point of user friction IMO:

Your bread crumbs are uninformative (and unlinked). Given that the sidebar on the left is also generic, users do not know where they are and have no easy way to navigate repositories.

… so glad I ran 20 experiments this time!

If your p-value remains stubbornly above 0.05, there are some creative ways to describe that as well, see this blog post: mchankins.wordpress.com/2013/04/21/s...