Exported reproducible R code now runs as written. Centering statements refer to columns of dat, the script selects the analysis sample before centering, and refitting reproduces the model estimated by mlm_fit().
Level-by-level equations now place predictors at the level where they vary. Higher-level predictors appear in the intercept equation and cross-level interactions appear in the equation for the slope they moderate, instead of every term being written as a Level-1 predictor.
Equations are built from the fitted coefficients, so each dummy code of a categorical predictor has its own coefficient and random-effect indices in the Tau matrices match the equations. Previously the two could disagree when a factor with more than two categories preceded a random slope.
Generalized models are written for the linear predictor with the link function shown, without a Level-1 residual term.
Centering now uses the analysis sample. Grand and cluster means are computed from rows that are complete on all model variables, so centered predictors have mean zero among the observations the model uses. Results are unchanged for complete data.
Methods text, the reproducibility checklist, and exported code now read the estimator from the fitted model and the optimizer settings recorded by mlm_fit(), so they cannot report REML for a model estimated with ML or a different maxfun than the one used.
The installed Shiny app was running an older copy of the modeling code than the package. The copies are synchronized again, dev/sync-app.R refreshes them, and a test fails if they drift.
A random-effect block with neither an intercept nor slopes no longer produces the invalid term (0 | group).
Improvements
mlm_fit() returns an object of class "mlm_result" with a print() method and the estimation settings; "mlm_spec" objects also print a summary.
mlm_fit() explains specifications it cannot fit: centering a categorical predictor, an unknown centering method, a random slope for a variable that is not a fixed predictor, a missing data frame, or no random effects. The same problems appear as stop items in the app’s model-readiness table.
Three-level nested models are written with pi, beta, and gamma coefficients. Crossed designs are written as a single mixed-model equation.
mlm_latex_equations() also returns a coefficient key linking each symbol to its estimate, the notation used, and an explanatory note. The app shows the key on the Equations tab and reports include it.
mlm_apa_tables() gains exponentiate = TRUE for odds ratios (logit link) and incidence rate ratios or exp(b) (log link) with confidence intervals. The app shows these columns for generalized models.
Cluster-mean centering can target a specific grouping factor with list(center = "CWC", group = "districtid").
Variable names that are not syntactic R names are quoted in formulas and handled in generated code.
The readiness table checks whether grouping IDs form the declared structure and warns when lower-level IDs are reused across higher-level units, which makes lme4 treat a nested design as crossed.
LaTeX tables escape #, $, {, }, ^, ~, and > in addition to the characters handled before.
Added a testthat suite covering specification, centering, fitting, equations, reporting, readiness checks, and app synchronization.
CRAN preparation
Retitled the package “A Guided Interface for Mixed-Effects and Multilevel Models” and added method references to the Description.
example_hsb() no longer changes the caller’s random-number state, and accepts seed = NULL to draw from the current stream.
Examples run without \donttest{} wrappers.
The app no longer errors in the variance-covariance preview when no grouping factor is selected, and its missing-package message no longer points to a script that is not installed.
Added an automated test that fits the example model in the bundled app and checks that every output renders.
Public beta preparation
Added HLM-style feature parity tracking with mlm_hlm8_parity(), plus a dedicated vignette describing parity and validation evidence.
Restored rendered equation displays in the Shiny app with local HTML/CSS rendering for Greek symbols, subscripts, multi-line equations, and Tau matrices while keeping raw LaTeX available for export.
Made the getting-started vignette execute a lightweight package-scope example during vignette builds while keeping install and app-launch chunks unevaluated.
Added software-article scaffolding, citation metadata, and machine-readable software metadata outside the CRAN source package.
Removed external MathJax/CDN loading from the Shiny app so the installed app does not depend on a remote JavaScript service at runtime.
Added a publication-readiness checklist for the next CRAN and beta-testing milestones.
Strengthened validation for example-data generation and model specification when selected variables are missing from the data.
Expanded backend tests for manuscript-facing outputs, including APA tables, raw LaTeX, wrapped equations, Quarto report text, software reporting, and app/backend synchronization.
Expanded the GitHub bug-report template so testers can report model structure, centering, interactions, and export-specific issues more clearly.
Added a local publication-readiness rehearsal script that checks app/backend copy synchronization, public documentation for local paths, release-file presence, and source tarball contents.
Added GitHub issue-template links to the documentation and beta testing guide.
Published GitHub prerelease v0.1.0-alpha.2 for structured beta testing.
Added a structured beta testing guide for early users and linked it from the documentation site.
Expanded the GitHub feedback template so testers report model structure, exported-code checks, and app section details.
Exported core backend helpers for model specification, formula generation, fitting, equations, APA tables, and software/package reporting.
Added mlm_supported_models() and a “Supported Models and Production Scope” vignette so users can see supported, experimental, planned, and out-of-scope features before using the app in real analyses.
Added APA-style software and R package reporting for reproducible manuscripts, including Shiny app display, Quarto report output, and LaTeX export.
Added optional papaja citation code generation for users who want APA-style R/package citations and BibTeX references in R Markdown manuscripts.
Added package citation metadata and expanded multi-platform check preparation.
Polished README installation, demo, documentation, and feedback guidance.
Added package-level documentation and runnable interactive example for run_mlmr().
Added CRAN submission comments and release checklist scaffolding.
Added a GitHub Actions workflow for building and publishing the pkgdown documentation site.
Kept root development app files out of the built package; the installed Shiny app is shipped through inst/app/.
Development version
Added the initial Shiny app for mixed-effects and multilevel model building.
Added the HSB-style example data workflow.
Added model specification, formula generation, centering, fitting, diagnostics, APA-style tables, model equations, Tau matrix display, and reproducible code export.
Added Quarto, R script, LaTeX, HTML, and Word-compatible export scaffolds.
Added package metadata, GitHub URLs, roadmap, and initial backend tests.
Added model-readiness checks before fitting custom models.
Added direct data upload support for CSV, TSV/TXT, Excel, SPSS, SAS, and Stata files.
Added loading indicators for slower diagnostics plots.