mixedbiastest 1.1.0

• Add compact simulation printing and distinguish numerical validation failures from refit errors. Document plug-in conditioning, no-intercept autoscaling, and use an unbalanced simulation example.

• Correct the Karl-Zimmerman journal volume and page range and link the author manuscript and revisions.

• Clarify that Bias_Estimate is the internal random-effects alignment diagnostic, not an unbiased estimate of total bias or an automatic correction. Explain the additional residual-weight contribution when covariance estimates depend on the response, and the limits of fitted permutation-reference p-values.

• Add simulate_bias() for fixed-design conditional response-refit sensitivity analysis. The default holds the fitted random effects fixed, generates Gaussian residual errors, and re-estimates covariance parameters for every response. Custom generators can specify latent effects and errors for a stated scenario.

• Return replicate-level alignment, residual, total error and internal diagnostic, with means, Monte Carlo standard errors, convergence/failure records and singular fit counts. A centered-error control variate is available without discarding the raw errors; its summaries are withheld after any failed refit because selection can invalidate the mean-zero control-variate argument. This does not reproduce stochastic assignment automatically.

• Share fitted-model algebra between the existing diagnostic and the simulation; preserve the diagnostic API, retained Bias_Estimate column, permutation RNG ordering and two-sided +1 p-value convention.

mixedbiastest 1.0.3

• Correct term-specific covariance handling when several random-effects terms share a grouping factor. Earlier versions could use the first term’s variance for later terms.
• Correct permutations of multi-coefficient diagonal terms to preserve lme4’s group-major coefficient ordering. Handle diagonal covariance representations introduced in lme4 2.0, including the diag_special double-bar option.
• Transform diagnostics from automatically scaled fixed-effect designs back to the original coefficient scale, including intercepts and single predictors.
• Use lme4’s sparse fitted-model factors to compute the diagnostic without inverting the random-effects covariance or flooring zero variances.
• Enforce the documented exclusion of non-unit observation weights using the supported weights accessor, including fits with excluded observations.
• Validate permutation counts, contrasts, contrast names, and verbosity before calculation. List only retained fixed-effect coefficients for rank-deficient fits.
• Preserve full precision in returned bias estimates, permutation means and p-values; small nonzero permutation p-values are no longer rounded to zero. The two-sided test and +1 correction are unchanged.
• Run the regression suite during CRAN checks. Add independent dense-GLS and permutation-reference tests for supported inputs and recent lme4 changes. Gaussian unweighted fits with diagonal random-effects covariance remain the supported scope; correlated structures and GLMMs are not added.