Skip to contents

Changelog

Source: NEWS.md

MAIHDA 0.1.9

Bug Fixes

  • Clarified the Shiny dashboard PVC/HUD interpretation so negative PVC values are shown as variance unmasking rather than as unexplained interaction variance.
  • Fixed the coverage workflow failure-artifact upload configuration.

MAIHDA 0.1.8

CRAN release: 2026-05-16

General Updates & New Features

  • Added plot_prediction_deviation_panels() function for visualizing predicted values and identifying deviant cases.
  • Added plot_risk_vs_effect() function to create a quadrant scatterplot comparing overall marginal predicted risk against pure intersectional effects.
  • Added plot_effect_decomposition() function to visually decompose the total deviation from the overall mean into additive and intersectional components.
  • Replaced the redundant “caterpillar” plot with the “predicted” plot in plot() and the interactive dashboard.
  • Added automatic tertile binning (via an autobin parameter) for numeric grouping variables with more than 10 unique values in make_strata().
  • Updated the interactive Shiny Dashboard (run_maihda_app()) to include the new visualizations and a toggle for auto-binning continuous strata variables.
  • Added detection for binomial data. fit_maihda() will now automatically detect binomial outcomes and switch to the appropriate family.

Bug Fixes

  • VPC/ICC Calculation Fix: Corrected the residual variance estimation for binomial and ordinal models. The package now accurately applies the theoretical level-1 variance (π2/3\pi^2 / 3 for "logit" links and 11 for "probit" links) internally when summarizing models or bootstrapping the variance partition coefficient, avoiding deflated VPC/ICC metrics.

MAIHDA 0.1.7

CRAN release: 2026-04-05

General Updates & New Features

  • Added stepwise_pcv() function to sequentially estimate proportional change in variance (PCV) by adding predictors one-by-one.
  • Added a fully-featured interactive Shiny Dashboard (via run_maihda_app()) for visual data exploration, model fitting, and performance visualization.
  • Improved bootstrap methods for more efficient confidence interval estimation.
  • Added missing documentation block for the maihda_sim_data dataset to resolve R CMD check warnings.
  • Updated test suite setup: tests/testthat.R was modified to correctly use test_check("MAIHDA") instead of shinytest2.
  • Added importFrom(stats, as.formula) for the stepwise_pcv function to prevent undefined warnings.
  • Updated introduction.Rmd vignette: added standard CRAN installation instructions, and improved text clarity.

MAIHDA 0.1.0

CRAN release: 2026-04-03

Initial Release

  • Initial CRAN submission
  • Added make_strata() function for creating intersectional strata
  • Added fit_maihda() function for fitting multilevel models with lme4 (default) or brms engines
  • Added summary() function for variance partition and stratum estimates
  • Added predict_maihda() function for individual and stratum-level predictions
  • Added plot() function with three plot types:
    • Caterpillar plots of stratum random effects
    • Variance partition coefficient visualization
    • Observed vs. shrunken estimates comparison
  • Added compare_maihda() function for comparing models with bootstrap confidence intervals
  • Added comprehensive documentation and vignettes
  • Added unit tests for core functionality

Bug Fixes and Improvements

  • Enhanced make_strata() to properly handle missing values (NA) in input variables:
    • Observations with missing values in any stratum variable are now assigned NA stratum
    • Missing values are no longer included as valid stratum categories
    • Added comprehensive tests for missing value handling