Evaluated ML and REML objectives by independent connected covariance blocks when the random and repeated grouping structures permit that decomposition. The fitted object continues to retain the complete marginal covariance for model methods.
Cached the optimized GLS state and the fixed-effect covariance Jacobian, removing repeated likelihood and Satterthwaite calculations.
Added simulate.lmm() with reproducible marginal Gaussian simulation and random-number-state preservation.
Added update.lmm() and extended predict.lmm() with standard errors, confidence intervals, prediction intervals, and partial re.form selection.
Added known relative precision weights, fixed-effect offset support, and custom contrasts to lmm().
Standardized ranef() to always return a named list of tibbles. This is an intentional pre-1.0 API change for single-term models.
Moved regular print() output to standard output, documented the residual scale returned by sigma(), and added a warning for the conservative multi-degree-of-freedom fallback.
Moved ggplot2 from Imports to Suggests; diagnostic plotting now checks for the optional package when called.
Added a comparison vignette for choosing among lmmix, lme4, lmerTest, nlme, and mmrm, and updated the documented scalability boundaries.
Replaced the maintainer’s GitHub no-reply address with a verifiable institutional address.
lmmix 0.3.0
Added fixed-band Toeplitz residual covariance through specifications such as structure = "toep(4)", while retaining full "toep" behavior.
Added an exact likelihood and covariance comparison with official PROC MIXED Example 79.6, including its line-source irrigation data and runnable SAS specification.
Extended confint() with "beta_" and "theta_" selectors. Covariance intervals use log and Fisher-z transformations to respect variance and correlation bounds.
Added reproducible parametric-bootstrap likelihood-ratio tests for nested models through anova(..., test = "parametric.bootstrap").
Added a vignette on covariance intervals, boundary-aware comparisons, and inspection of the simulated null reference distribution.
lmmix 0.2.1
Added reproducible comparisons with official SAS PROC MIXED examples 79.1, 79.2, and 79.5 from the SAS/STAT 14.3 documentation.
Added the split-plot, repeated-measures growth, and random-coefficients data used by those examples.
Added a runnable SAS validation program and machine-readable published targets for covariance parameters, likelihood criteria, fixed effects, BLUPs, and type III tests.
Added package and repository citation metadata for academic reuse.
Added CRAN submission preparation metadata and clarified the distinction between published SAS targets and a fresh SAS execution transcript.
lmmix 0.2.0
Added Kenward-Roger covariance adjustment, denominator degrees of freedom, and scaled multi-degree-of-freedom tests for REML fits.
Added independent crossed or nested random-effect terms through a backward-compatible list-of-formulas interface.
Added likelihood-ratio comparisons for nested models. Models that differ in fixed effects can be refitted automatically with ML, while covariance-model comparisons require ML and report the boundary-distribution caveat.
Added deterministic optimizer fallback attempts and retained complete attempt diagnostics in fitted objects.
Added na.action support for na.omit, na.exclude, and na.fail, including restoration of excluded rows in fitted values, residuals, and augmented data.
New grouping levels now require allow.new.levels = TRUE for conditional prediction instead of silently receiving a zero random contribution.
Added simultaneous Bonferroni confidence intervals for adjusted pairwise marginal-mean comparisons.
Added independent validation against lme4, lmerTest, mmrm, and emmeans for the new 0.2.0 capabilities.
lmmix 0.1.0
Added explicit profiled ML and REML estimation for Gaussian linear mixed models.
Added independent, compound-symmetric, AR(1), Toeplitz, and unstructured residual covariance matrices.
Added random intercepts and slopes with an unstructured covariance matrix.
Added Satterthwaite fixed-effect inference, type III tests, estimated marginal means, and pairwise contrasts.
Added standard model, broom, and emmeans methods.
Added plot.lmm() diagnostics that return ggplot2 objects for residual, normal Q-Q, and observed-versus-fitted displays.
Added human-readable table printing without dots in displayed headings while preserving standard broom column names for programmatic use.
Added the multilocation example from Milliken and Johnson (2009, Section 28.3) and validation tests against SAS PROC MIXED, nlme, lmerTest, and mmrm.
Added primary theoretical references for REML, Satterthwaite inference, multi-degree-of-freedom tests, and estimated marginal means.
Expanded the validation vignette with executable comparisons against nlme, mmrm, and lmerTest, plus every stored PROC MIXED target and its observed numerical difference.
Reworked the README, vignettes, package overview, and method documentation to describe the implemented API, validation evidence, and current limits.
Added combined-model convergence tests for every residual covariance structure.