FWIW, there's also a fast and simple alternative in R called greta (https://greta-stats.org)

Looks “great-ah”!

What’s the advantage of using Stan when now there is numpyro and pymc. The newer frameworks seem much more flexible and performant than Stan.

For things that do not require more expressiveness than what Stan offers, I have found the speed and quality of sampling is outstanding. Furthermore, thanks to static typing, complex models are easy to write. To be fair, Pyro is also great and is my go-to for massive models where MCMC is unfeasible or for those requiring more expressiveness and/or deep components. I have less experience with PyMC.

brms and rstanarm make using it very easy. Blavaan likewise. The transition from Conventional model syntax is great scaffolding.

I feel the technical barrier of adoption of bayesian methods is still high enough to deter potential users.

It is much less challenging with Bambi[1] and brms[2].

  [1] https://bambinos.github.io/bambi/
  [2] https://paul-buerkner.github.io/brms/

If you put the links in plaintext they're not clickable.

I completely agree with you and this is why we are developing 4d-modeller:

https://4dmodeller.github.io/fdmr/

It's still in early stages but the concept is that these methods are mostly opaque even to highly technical users. So we start with shiny apps that help you build a model and that will work up to (not yet implemented, but we will do a sprint in a couple weeks) wrapper functions, which helps you get things going until you start wanting to get more and more complex. We just ran a Hackathon and participants with no R or Bayesian experience were able to make models.

It is still the Linux of statistical software. Rstanarm is getting quite good though for R users.

For someone too lazy to read all the docs, is most/all of the lavaan model syntax implemented?