NEWS

---

wbCorr 0.4.0 (2026-10-31)

This version returns to the simpler design of earlier wbCorr versions, while keeping the statistical fixes of 0.3.2.

Breaking changes

• Cluster means are again estimated from all available observations of each variable, as in psych::statsBy(). The centering_rows argument is removed. Results only change when data are missing.
• Removed the arguments between_inference (use inference = "none"), missing_data (use na.omit() for listwise deletion), and the deprecated bootstrap and weighted_between_statistics.
• Removed get_matrix_diagnostics() and get_matrix(numeric = TRUE). Unrounded coefficients are stored in the wbCorr object (e.g., object@within$correlations).
• Detailed tables no longer contain the diagnostic columns n_obs, n_clusters, n_boot_attempted, n_boot_valid, status, reason, inference_status, and inference_reason.

Restored

• Spearman correlations have p-values and confidence intervals again (Fisher z).
• method = "auto" uses Spearman for ordered factors again.
• Cluster-size weighted between-cluster correlations have analytic p-values and confidence intervals again.
• The cluster bootstrap reports p-values again.

Fixes

• Correlations based on two clusters are NA instead of a perfect correlation.
• wbCorr() no longer warns on every call that analytic p-values are approximations.
• The cluster bootstrap reports bias-corrected (BC) intervals, works with cluster-size weighting, and is much faster.

Documentation

• Rewrote the README with a worked example and its output. Removed the vignette, which repeated the README.

wbCorr 0.3.2 (2026-07-23)

Statistical correctness

• Reimplemented the one-way random-effects, single-measure ICC(1,1) with the ANOVA method-of-moments formula for unequal cluster sizes. Negative sample ICCs are retained, and non-estimable cases now return explicit diagnostics.
• Corrected Pearson analytic degrees of freedom and Fisher-z confidence intervals. Intervals are bounded by -1 and 1, and inference that is not supported by the available analysis units is reported as unavailable.
• Clarified that the Spearman option is a descriptive correlation of mean-centered scores within clusters and of cluster means between clusters. Unsupported analytic, row-jackknife, and cluster-size-weighted Spearman inference has been removed; whole-cluster bootstrap intervals remain available.

Inputs, diagnostics, and output

• Strengthened validation for data, cluster identifiers, method, confidence levels, bootstrap settings, weighting, and centering choices.
• Logical variables and declared two-level factors are supported with explicit 0/1 orientation. Other categorical inputs are rejected with guidance, and non-finite numeric values are treated as missing.
• Detailed tables now retain every requested variable pair, including pairs whose coefficient or inference cannot be estimated. Pair-specific observation counts, cluster counts, coefficient status, inference status, reasons, and bootstrap yield diagnostics are reported.
• Correlation-matrix diagonals are 1 only for variables with positive variance at the relevant level; otherwise they are NA.
• Added positive-semidefinite diagnostics for the unrounded within- and between-cluster matrices. Pairwise matrices that are not positive semidefinite now trigger a targeted warning, and missing_data = "listwise" provides an optional common-row analysis.
• get_matrix(..., numeric = TRUE) now returns unrounded numeric within, between, and merged matrices, including unrounded ICC diagonals.
• print() now provides concise, executable guidance for inspecting summaries and extracting full tables, numeric matrices, and ICCs.
• Plot panels now use the exact pair-specific rows, centering rule, method, and between-cluster weights used by the fitted object. Annotations come from the stored coefficient and p-value, and significance legends are omitted when p-values were not requested or are unavailable.
• to_excel() now accepts a data frame or matrix directly, supports mixed lists of tabular objects, errors clearly when there is nothing to write, and invisibly returns the supplied output path.

Development and release preparation

• Added reference tests against psych::statsBy() for within- and between-cluster correlations and ICC(1,1), and expanded automated regression-test coverage.
• Added a pkgdown website workflow and a vignette introducing the distinction between within- and between-cluster correlations.
• Removed obsolete self-generated test fixtures, a generated plot file, and tracked R session history. Added safeguards against including regenerated session or graphics artifacts in the repository and source package.
• Corrected the bundled-data documentation to report 5,000 observations from 100 participants measured over 50 days.
• Added repository citation metadata and a DOI badge that use the permanent Zenodo concept DOI, 10.5281/zenodo.20357592, rather than a release-specific DOI.