LoginSign Up
← Back to all sparks
O

ordbetareg

INFRA · APIS
Velocity0.0

Ordered Beta Regression Models with 'brms'

github.com/saudiwin/ordbetareg_pack ↗

A brms wrapper for bounded outcomes that keeps handing users more control over the model's internals.

r-packagebayesianregressionbrmsbounded-outcomes
◆Current state
ordbetareg fits ordered beta regression — outcomes bounded at both ends, with mass at the bounds — on top of brms. Each release has opened up a layer that was previously fixed: v0.5.0 added posterior predictive checks and user-specified bounds, and v0.8 added a full prior override that makes cutpoint modeling possible, plus component-wise prediction of the bottom, middle and top of the scale.
◆Where it's heading
The direction is from a convenience wrapper toward something users can take apart. The pattern across releases is the same move repeated at greater depth: what was hard-coded becomes an argument. Documentation is expanding in step, with the vignette now covering power analysis and cutpoint modeling, which suggests the audience is being taught to use the lower-level controls rather than the defaults.
◆Prediction
Given that priors, bounds and predicted components are now all exposed, the next release most plausibly extends the diagnostic and plotting surface rather than the model itself — though releases here are more than a year apart, so timing is unpredictable.

◆Recent moves

  1. 1y ago

    Manual priors unlock cutpoint modeling; new component predictions

    prior_manual lets users replace the package's priors outright, which is what makes cutpoint modeling reachable, while posterior_epred_ordbeta predicts the bottom, middle and top scale components separately and plot_heiss shows how those components move across factor levels. It is the fullest expression of the arc toward exposing the model's internals, with power analysis and cutpoint sections added to the vignette.

    View source ↗
  2. 3y ago

    ordbetareg release v0.7.0

    The release note defers entirely to the NEWS file, describing only "updates and extensive bug fixes", so the scope of this version cannot be read from the feed. It sits in a two-year gap between the capability-expanding releases on either side.

    View source ↗
  3. 3y ago

    Posterior predictive checks and user-set outcome bounds

    Bundling the 0.5.0 and 0.3.0 notes, this release adds pp_check_ordbetareg for outcomes that are part discrete and part continuous, lets users set their own normalisation bounds, and enables multiple imputation and multivariate responses through brms. The user-specified bounds are the first instance of the pattern v0.8 later repeats with priors.

    View source ↗