plot.resid() now picks its default Q-Q reference
distribution from fit when supplied
(distribution defaults to NULL).plot() method for glm objects,
which replaced the standard stats::plot.lm() diagnostics
once sure was loaded; use plot(resids(fit), fit = fit)
instead.glm fits with a
cloglog link, which used the wrong Gumbel orientation for
the latent distribution (#45).NULL when calling
plot() on a fitted model without a covariate.CITATION, DESCRIPTION, and references.resids.multinom and
resids.matrix for multinomial and general ML
classifiers.prior.weights) for Gaussian
and Gamma GLMs.sim_data() to dynamically
simulate datasets representing different model misspecifications.codecov.yml,
and a curated _pkgdown.yml reference index.ggplot2/gridExtra to tinyplot via
standard S3 plot() methods.plot.resid() arguments to leverage standard
base R graphical parameters (col, pch,
cex, etc.).utils.R
file into separate generic-specific files.run_bootstrap_reps().3.1 to
4.0.0 in DESCRIPTION.devel is now the
default development branch.generate_surrogate()/generate_residuals()
(previously ~90% duplicated) into a single code path.get_bounds.orm/ncat.orm/get_mean_response.orm
(now aliased to their lrm equivalents) and collapsed
get_distribution_function()/get_quantile_function()
from 12 S3 methods into two functions over shared lookup tables."_PACKAGE"
sentinel.DESCRIPTION paragraph describing the
removed ggplot2-based autoplot().ggplot2-based autoplot()
methods in favor of standard plot() S3 methods.ggplot2 and gridExtra package
dependencies from Imports.testthat testing framework and rebuilt the
entire test suite using tinytest.slowtests/ directory as its test
cases are fully covered by the package vignette and the unit test
suite.df1 through
df5 and their documentation, replacing them completely with
sim_data().mgcv Suggests entry, the stale
LazyData field, the pr-commands.yaml workflow,
and a stray tracked vignettes/sure.R build artifact.gof() validation checking due to
operator precedence.nsim > 1 bootstrap crashes on model classes
that do not support nobs().vglm models
containing "link" suffixes.clm and
polr models with collinear (aliased) predictors.tinytest::at_home() gating:
test_package()’s own default silently skipped roughly 63%
of the test suite (104 of 164 expectations) in every environment,
including CI, regardless of any environment variable.truncdist package.stats @importFrom entries
(pgamma, pnbinom, ppois) flagged
by R CMD check.Incorporated pkgdown website.
Added paper URLs and ORCIDs to DESCRIPTION file (#30).
Added sure vignette
incorporated autoplot methods for glm, lrm, orm, polr, vglm
autoplot() can now return multiple plots (#16).
Specifying method = "latent" now works for binomial
GLMs (#27).
Specifying method = "jittering" now issues a warning
(at least until it has been fully tested).
New function surrogate for returning the surrogate
response values used in calculating the surrogate-based residuals. The
surrogate response values can be useful for checking the proportionality
assumption of fitted cumulative link models, among other
things.
Jittering (on both the probability scale and the response scale)
is now available for fitted cumulative link models based on packages
MASS, ordinal, rms, and
VGAM (#18).
Added support for vector generalized additive models from the
VGAM package (i.e., objects of class
"vgam").
New data sets df4 and df5 for
illustrating various uses of the surrogate residual for diagnostics an
ordinal regression models.