                        Changes in version 1.0-2                        

Changes in results

  - hoa() now computes the modified likelihood root correctly for
    contrasts with more than one nonzero coefficient, such as averages,
    trends or Tukey-type comparisons in models with correlated
    coefficients. Previously the nuisance information was taken from a
    sub-block of the covariance matrix, which is only correct for
    contrasts that select a single coefficient. Adjusted statistics,
    intervals and p-values change for such contrasts, in some cases by
    as much as the whole higher-order correction. Pairwise contrasts in
    cell-means models (for example the Dunnett example in the vignette)
    were nearly unaffected.

  - hoa() also uses the working weights at the final constrained fit,
    which changes its results slightly (up to about 1e-4) for all
    contrasts.

  - hoa() no longer discards every profile point for larger data sets.

Bug fixes

  - plot() failed for intervals from confint(..., alternative = "less").

  - expit() is now an exported generic, so expit(confint(...))
    transforms intervals to the probability scale as documented.

  - mcprofile() failed when all grid points lay on one side of the
    estimate.

  - mcprofile() for lm objects failed when the data, weights or formula
    were defined inside a function, or when the data had the same name
    as an argument of mcprofile() (for example CM).

  - confint() now checks adjust and alternative. A misspelled
    alternative previously gave a confusing error, or could silently
    return an unrelated object called ci from the workspace as the
    intervals. Unambiguous abbreviations such as alternative = "g" are
    now accepted.

  - mcprofile() now gives a clear error for models with aliased (NA)
    coefficients, and for a contrast matrix with the wrong number of
    columns.

Other changes

  - Profiling is faster: each constrained fit solves the quadratic
    program once per iteration, and the starting values on each side of
    the estimate are chosen better.

  - Requires ggplot2 3.4.0 or later. plot() for intervals no longer uses
    the deprecated geom_errorbarh().

  - The citation uses bibentry().

  - Added a test suite.
