On March 26, 2020, version 1.0.1 of my R package shinybrms was released on CRAN (see here). This package provides a GUI (a Shiny app) for fitting Bayesian regression models using the R package brms which relies on Stan. Since it relies on Shiny, shinybrms may be run on a server and then accessed via a web browser (so that users don’t even need to install R and the necessary R packages). A few days ago, shinybrms version 1.1.0 (which also supports varying effects) has been released so that the current features are:
- For the (univariate) outcome: only a Gaussian, Bernoulli (with logit link), or negative binomial (with log link) distribution.
- For the predictors: nonvarying (a.k.a. population-level or “fixed”) and varying (a.k.a. group-level or “random”) effects as well as interactions.
- For the priors: All priors which are supported by brms.
- Most of brms's special features, like monotonic effects for ordinal predictors, nonlinear effects, or the possibility to specify standard errors for the outcome (for meta-analyses and meta-regressions) are not supported yet.
- For the inspection of the output, only a short summary (from
brms::summary.brmsfit()) and the possibility to launch the shinystan app is offered.
Of course, these features are still very limited. A lot of missing features are on my to-do list. Additionally, I plan to include (for example):
- a custom output inspection directly in shinybrms (additionally to the possibility to launch shinystan),
- model selection possibilities.
Further suggestions for improvements are very welcome.
Note that shinybrms does aim at replacing any of the great work of the JASP developers. Instead, it is meant as a supplement to JASP and all other GUIs for fitting Bayesian regression models. I am simply posting this announcement here as people who are already using a GUI for fitting Bayesian regression models might be interested in shinybrms.