performance 0.15.2

Bug fixes

• Fixed failing CRAN checks, related to the latest rstanarm update.

performance 0.15.1

Changes

• display() now supports the tinytable format, when format = "tt".

• Better handling of non-converged lavaan-models in model_performance().

performance 0.15.0

New functions

• item_omega(), to calculate the McDonald's Omega reliability coefficient.

• item_totalcor() calculates the total correlation of an item with the
  sum of all other items in a scale. If corrected = TRUE, the total
  correlation is corrected for the number of items in the scale (which is
  equivalent to item_discrimination()).

• Column names of item_reliability() were changed to be in line with the
  easystats naming convention and to be consistent with the output of other
  related functions.

Changes

• check_itemscale() now work with factor analysis results, from
  parameters::factor_analysis().

• item_reliability() now includes the item-total correlation, and information
  about Cronbach's alpha and mean inter-item correlation in the printed output.

• cronbachs_alpha() now work with factor analysis results, from
  parameters::factor_analysis().

• Formatting of p-values in test_likelihoodratio() is now consistent with
  formatted p-values from other functions.

• Added following methods for psych::fa(), psych::principal(), item_omega(),
  psych::omega(), and parameters::factor_analysis(): check_normality(), check_residuals(), check_outliers(), and model_performance().

• item_alpha() was added as an alias for cronbachs_alpha().

• Further functions get a display(), print_md() and print_html() method.

Bug fixes

• Fixed issue in check_predictions() for binomial models with a response
  defined as proportion or matrix of successes and trials.

• print_md() for objects returned by check_itemscale() now include the footer
  with information about Cronbach's alpha and mean inter-item correlation.

performance 0.14.0

Breaking Changes

• The "Increased SE" column in the output of check_collinearity() was renamed
  into "adj. VIF" (=adjusted VIF). Furthermore, the computation of the adjusted
  VIF now correctly accounts for the numbers of levels (i.e. degrees of freedom)
  for factors.

New functions

• New function check_group_variation() to check within-/between-group
  variability (this function will replace check_heterogeneity_bias() in
  future releases.)

• New functions performance_reliability() and performance_dvour(). These
  functions provide information about the reliability of group-level estimates
  (i.e., random effects) in mixed models.

Changes

• Singularity checks with check_singularity() are now more efficient and also
  include the random effects for the dispersion component (from package
  glmmTMB). Furthermore, a check argument allows to check for general
  singularity (for the full model), or can return singularity checks for each
  random effects term separately.

Bug fixes

• Fixed issue with wrong computation of pseudo-R2 for some models where the
  base-model (null model) was updated using the original data, which could
  include missing values. Now the model frame is used, ensuring the correct
  number of observations in the returned base-model, thus calculating the
  correct log-likelihood and returning the correct pseudo-R2.

• Fixed examples in check_outliers().

performance 0.13.0

Breaking changes

• check_outliers() with method = "optics" now returns a further refined
  cluster selection, by passing the optics_xi argument to dbscan::extractXi().

• Deprecated arguments and alias-function-names have been removed.

• Argument names in check_model() that refer to plot-aesthetics (like
  dot_size) are now harmonized across easystats packages, meaning that
  these have been renamed. They now follow the pattern aesthetic_type, e.g.
  size_dot (instead of dot_size).

Changes

• Increased accuracy for check_convergence() for glmmTMB models.

• r2() and r2_mcfadden() now support beta-binomial (non-mixed) models from
  package glmmTMB.

• An as.numeric() resp. as.double() method for objects of class
  performance_roc was added.

• Improved documentation for performance_roc().

Bug fixes

• check_outliers() did not warn that no numeric variables were found when only
  the response variable was numeric, but all relevant predictors were not.

• check_collinearity() did not work for glmmTMB models when zero-inflation
  component was set to ~0.

performance 0.12.4

Changes

• check_dag() now also checks for colliders, and suggests removing it in the
  printed output.

• Minor revisions to the printed output of check_dag().

Bug fixes

• Fixed failing tests that broke due to changes in latest glmmTMB update.

performance 0.12.3

New functions

• check_dag(), to check DAGs for correct adjustment sets.

Changes

• check_heterogeneity_bias() gets a nested argument. Furthermore, by can
  specify more than one variable, meaning that nested or cross-classified
  model designs can also be tested for heterogeneity bias.

performance 0.12.2

Patch release, to ensure that performance runs with older version of
datawizard on Mac OSX with R (old-release).

performance 0.12.1

General

• icc() and r2_nakagawa() get a null_model argument. This can be useful
  when computing R2 or ICC for mixed models, where the internal computation of
  the null model fails, or when you already have fit the null model and want
  to save time.

• icc() and r2_nakagawa() get a approximation argument indicating the
  approximation method for the distribution-specific (residual) variance. See
  Nakagawa et al. 2017 for details.

• icc() and r2_nakagawa() get a model_component argument indicating the
  component for zero-inflation or hurdle models.

• performance_rmse() (resp. rmse()) can now compute analytical and
  bootstrapped confidence intervals. The function gains following new arguments:
  ci, ci_method and iterations.

• New function r2_ferrari() to compute Ferrari & Cribari-Neto's R2 for
  generalized linear models, in particular beta-regression.

• Improved documentation of some functions.

Bug fixes

• Fixed issue in check_model() when model contained a transformed response
  variable that was named like a valid R function name (e.g., lm(log(lapply) ~ x),
  when data contained a variable named lapply).

• Fixed issue in check_predictions() for linear models when response was
  transformed as ratio (e.g. lm(succes/trials ~ x)).

• Fixed issue in r2_bayes() for mixed models from rstanarm.

performance 0.12.0

Breaking

• Aliases posterior_predictive_check() and check_posterior_predictions() for
  check_predictions() are deprecated.

• Arguments named group or group_by will be deprecated in a future release.
  Please use by instead. This affects check_heterogeneity_bias() in
  performance.

General

• Improved documentation and new vignettes added.

• check_model() gets a base_size argument, to set the base font size for plots.

• check_predictions() for stanreg and brmsfit models now returns plots in
  the usual style as for other models and no longer returns plots from
  bayesplot::pp_check().

• Updated the trained model that is used to prediction distributions in
  check_distribution().

Bug fixes

• check_model() now falls back on normal Q-Q plots when a model is not supported
  by the DHARMa package and simulated residuals cannot be calculated.