LoginSign Up
← Back to all sparks
P

performance

ANALYTICS
Velocity0.0

Assessment of regression model quality

github.com/easystats/performance ↗

performance keeps adding ways to check a model you have already fitted.

r-languagemodel-diagnosticsbayesianeasystatsregression
◆Current state
performance is at 0.17.1, which added check_priors() for prior predictive checks on Bayesian models and gave check_overdispersion(), check_model() and check_predictions() arguments to control residual type and plot range. The releases before it are a similar mix: a -2LL criterion column in test_likelihoodratio(), Bayesian predictive checks routed through modelbased, and in 0.16.0 a set of breaking renames including RMSA to the correct RMSR.
◆Where it's heading
Two consistent habits. Diagnostics keep gaining arguments to narrow what is examined — ppc_range, x_limits, maximum_dots, show_ci — which reads as a package being used on models large and awkward enough that the defaults stopped working. And simulated residuals via DHARMa keep displacing standard ones as the basis for the checks themselves.
◆Prediction
With check_priors() newly added and Bayesian predictive checks now routed through modelbased, the next release most likely extends the Bayesian diagnostic set rather than reworking the frequentist checks.

◆Recent moves

  1. 1mo ago

    check_priors() added; overdispersion plots use simulated residuals

    Adds check_priors() for prior predictive checks on Bayesian models, plus ppc_range and x_limits arguments for zooming into plots on wide-range responses. The overdispersion plot switches to DHARMa simulated residuals for glmmTMB and mixed models, fixing distorted output.

    View source ↗
  2. 2mo ago

    -2LL criterion column and unified Bayesian predictive checks

    test_likelihoodratio() gains a Criterion column carrying -2LL per model, and Bayesian check_predictions() moves onto modelbased::estimate_prediction() so its output matches every other supported model. Collinearity fixes for fixest and ordinal models come with it.

    View source ↗
  3. 6mo ago

    Breaking renames plus point-count and CI controls in check_model()

    Breaking renames: the psych FA metric is corrected from RMSA to RMSR, and the first argument of check_model(), check_predictions() and check_convergence() becomes model. check_model() gains maximum_dots to cap plotted points and show_ci to control confidence intervals.

    View source ↗
  4. 8mo ago

    check_autocorrelation() methods for DHARMa objects

    Adds check_autocorrelation() methods for DHARMa and simulate_residuals() objects, plus documentation work. A small extension of the simulated-residual thread.

    View source ↗
  5. 10mo ago

    Fixes CRAN checks after an rstanarm update

    A release that exists to clear CRAN checks after an rstanarm update. Nothing user-facing.

    View source ↗
  6. 11mo ago

    tinytable output format in display()

    display() picks up the tinytable format and non-converged lavaan models are handled more gracefully in model_performance(). Two small additions.

    View source ↗