LoginSign Up
← Back to all sparks
B

bayestestR

ANALYTICS
Velocity0.0

Describe and test the parameters of Bayesian models in R

github.com/easystats/bayestestR ↗

Bayesian diagnostics get stricter defaults while the Stan backend list widens

bayesiandiagnosticsstaneasystatsr-stats
◆Current state
bayestestR is the diagnostics and hypothesis-testing layer of the easystats stack, and its recent releases have concentrated on two things: reporting the right uncertainty numbers by default, and accepting posterior draws from more sources. The 0.18.x line added CmdStanFit support alongside the existing rstanarm/brms paths and switched effective-sample-size reporting to tail-ESS. Bug-fix releases in between are mostly CRAN-check maintenance.
◆Where it's heading
The package is converging on a single posture: work with raw MCMC draws from anywhere, and report the diagnostic that actually governs the interval being shown. Successive releases have swapped defaults rather than added surface area, and the efficiency work in 0.16.x aimed squarely at large brms and rstanarm fits. Output formatting is drifting toward the shared easystats display() and tinytable path.
◆Prediction
Expect continued backend coverage on the Stan side and further alignment of print/display behavior with insight and the rest of easystats; the entries do not show a push into new inference methods.

◆Recent moves

  1. 2mo ago

    mcse() gains a centrality argument

    A narrow addition in the same direction as the 0.18.0 diagnostics work: mcse() can now return the Monte Carlo standard error matched to the centrality measure being reported. The rest of the release clears CRAN check failures.

    View source ↗
  2. 2mo ago

    CmdStanFit support and tail-ESS as the default diagnostic

    The most substantive release in this window: cmdstanr fits become first-class inputs, diagnostic_posterior() accepts raw MCMC samples, and the ESS column now reports tail-ESS instead of the older n_eff. That default swap changes what users read off every diagnostic table.

    View source ↗
  3. 11mo ago

    rope() gains complement probabilities; display() methods added

    Adds the complementary above/below-ROPE probabilities and a display() path with tinytable output, continuing the shift of formatting into shared easystats infrastructure. The long-deprecated rnorm_perfect() is removed.

    View source ↗
  4. 1y ago

    describe_posterior() efficiency and multinomial handling

    Incremental performance work on describe_posterior() plus handling fixes for multinomial responses and brms mixture models. Consistent with the large-model efficiency push that started in 0.16.0.

    View source ↗
  5. 1y ago

    effects argument changes behavior for large brms/rstanarm fits

    A rebase on the current insight release that makes many-parameter models cheaper to summarize, and redefines effects = "all" to exclude group-level estimates. Users who relied on the old meaning need effects = "full".

    View source ↗
  6. 1y ago

    Tail ESS returned from effective_sample() and its callers

    The first step toward the tail-ESS default that lands in 0.18.0, alongside fixes for response-level columns on marginaleffects objects and a MAP estimate edge case.

    View source ↗