Changes in version 1.2.0                        

User visible changes

  - NEW FEATURE: ?? change in =effects= ??
  - NEW FEATURE: a delta method via the =estimate= function can now be
    performed on the output when combining mixed model via =rbind= and
    =mlmm=.
  - NEW FEATURE: pooling has been added as a possible method, next to
    adjustment for multiple testing for the output of =anova=, =rbind=,
    and =mlmm=.
  - NEW FEATURE: a formula can now be input to =summarizeNA= in order to
    evaluate the missing data patterns per group.
  - NEW FEATURE: an argument =force= can passed to =anova= to do a
    likelihood ratio test (LRT) without checking whether the models are
    nested nor whether they are fitted via REML.
  - FIX: bug when having many variables in the formula of =summarize=.
  - BREAK: =summarize= no longer evaluate correlations. This has been
    moved to a new function called =correlate=. #+BEGIN_SRC R :exports
    both :results output :session R :cache no
    LMMstar.options(param.optimizer = c("init.cor" = "overall"))
    #+END_SRC

Internal change

  - re-organize what the output of =anova= contain and corresponding
    methods (=coef=, =model.tables=, =vcov=, =iid=) to better align
    methods applied to the output of =anova=, =rbind=, and =mlmm=.
  - BREAK: the optimization method of lmm is now checking whether the
    score w.r.t. to all parameters (instead of only the mean and
    variance parameters) is close to 0. This could impact the solution
    found by lmm.
  - BREAK: initialization of the correlation parameters is now performed
    by averaging time-specific residual correlations instead of
    computing an overall correlation. This could impact the solution
    found by lmm. This behavior can be reverted using:
  - NEW FEATURE: update method for the output of =rbind= and =mlmm=.

                 Changes in version 1.1.0 (2024-05-12)                  

User visible changes

  - NEW FEATURE effects.lmm to evaluate average counterfactuals.
      - FIX: reparametrisation when using sigma() inside estimate
      - BREAK: update the calculation of the influence function with
        REML which may impact multiple testing adjustments. Use
        LMMstar.options(REML2ML=FALSE) to revert to old version.
      - CHANGE: Rename argument data into newdata in extractors such as
        logLik, information, model.frame, model.matrix, residuals,
        score, vcov
      - CHANGE: Rename argument keep.newdata into keep.data
      - CHANGE: Argument at for partial residual calculation (instead of
        using an attribute on argument type).
      - REMOVE: dependency on emmeans and sandwich.

Internal change

  - partial derivative of the residuals w.r.t. model parameters can be
    exported as an attribute grad.
      - standard deviation for random effect computed via delta method
        instead of numerical differentiation

                 Changes in version 1.0.0 (2023-11-08)                  

User visible changes

  - NEW FEATURE: support lme4 syntax for random intercepts.

Internal change

  - substantial re-wrote of the functions for handling covariance
    patterns.