yes, thank you!

Oh sorry! It looks like it might be the issue mentioned in the post. Don't know why it didn't work on previous debugging attempts though. Must have done something wrong. If anyone else reads this, r-shinylive still needs to be the latest dev version.

Relevant app just doesn't load (shinylive in a Quarto doc). Isolated it to just loading ggplot2. And found this post, which kind of looks like @transport-talk.bsky.social : github.com/r-wasm/webr/... I may have to do more debugging. Thought I had the latest r-shinylive. munsell trick does not work

Thank you! I'll take a look. This might do the trick

And I also typed this, but don't see it in my reply now. Weird: So, the part about "skewed distributions of predictors and outcomes" may not be correct. Depending on the setup and nonnormality, Bayesian estimation is sometimes robust, but not in general.

didn't need the "sometimes" in there, but whatever, it's late

So, the part about "skewed distributions of predictors and outcomes" may not be correct. Depending on the setup and nonnormality, Bayesian estimation is sometimes robust, but not in general.

Bayesian estimation sometimes makes distributional assumptions about the data that bootstrap does not. We put blavaan into this paper (on nonnormality) in part to illustrate that it is not inherently robust to distributional assumption violations: osf.io/preprints/ps...

Agree that once a joint distribution is found, just multiply a*b and then use that to find intervals. Bootstrap, Bayes, MC, are similar in this regard. But, while it doesn't matter if a or b have normal sampling distributions...

Add robustbase to the list of #rstats packages. If it were really that the estimates were ok but standard errors not trustworthy, I wonder if there is an appropriate sandwich covariance matrix (sandwich package worth a look). I can’t say what’s best.

Like others say, go for it. Most times I've done something similar it's helped in other contexts. e.g., knowing one sampler has helped with custom programming simulated annealing, Metropolis-Hastings Robbins-Monro, etc., where there was no apparent software implementation for my purpose. 👍

So, I take it from this discussion that (a) such a thing does not exactly exist publicly, at least in a great/flexible format, and that (b) one/both of you are going to create something?

Computer adaptive test, as in use of IRT to shorten a test, reduce the number of items needed to assess certain constructs