- Remove ‘Zelig’ from suggested packages, as it was removed from CRAN.

`model_parameters()`

now also transforms standard errors when`exponentiate = TRUE`

.`model_parameters()`

for`anova()`

from mixed models can now also compute effect sizes like eta squared.`model_parameters()`

for`aov()`

gains a`type`

-argument to compute type-1, type-2 or type-3 sums of squares.`model_parameters()`

for Bayesian models gains a`standardize`

argument, to return standardized parameters from the posterior distribution.- Improved
`print()`

method for`model_parameters()`

for nested`aov()`

(repeated measurements). - You can now control whether
`demean()`

should add attributes to indicate within- and between-effects. This is only relevant for the`print()`

-method of`model_parameters()`

.

- Fixed
`model_parameters()`

for`anova()`

from*lmerTest*models.

- Alias
`model_bootstrap()`

was removed, please use`bootstrap_model()`

. - Alias
`parameters_bootstrap()`

was removed, please use`bootstrap_parameters()`

. - Alias
`model_simulate()`

was removed, please use`simulate_model()`

. - Alias
`parameters_simulate()`

was removed, please use`simulate_parameters()`

. - Alias
`parameters_selection()`

was removed, please use`select_parameters()`

. - Alias
`parameters_reduction()`

was removed, please use`reduce_parameters()`

. - Functions
`DDR()`

,`ICA()`

and`cmds()`

are no longer exported, as these were intended to be used internally by`reduce_parameters()`

only. `skewness()`

and`kurtosis()`

always return a data frame.

- Added support for
`arima`

(*stats*),`bife`

(*bife*),`bcplm`

and`zcpglm`

(*cplm*)

- Improved print-method for
`model_parameters.brmsfit()`

. - Improved print-method for
`model_parameters.merMod()`

when fitting REWB-Models (see`demean()`

). - Improved efficiency for
`model_parameters()`

(for linear mixed models) when`df_method = "kenward"`

. `model_parameters()`

gets a`p_adjust`

-argument, to adjust p-values for multiple comparisons.- Minor improvements for
`cluster_analysis()`

when`method = "kmeans"`

and`force = TRUE`

(factors now also work for kmeans-clustering).

`p_value_kenward()`

,`se_kenward()`

etc. now give a warning when model was not fitted by REML.- Added
`ci()`

,`standard_error()`

and`p_value()`

for*lavaan*and*blavaan*objects. - Added
`standard_error()`

for*brmsfit*and*stanreg*objects.

- Run certain tests only locally, to reduce duration of CRAN checks.
`skewness()`

,`kurtosis()`

and`smoothness()`

get an`iteration`

argument, to set the numbers of bootstrap replicates for computing standard errors.- Improved print-method for
`factor_analysis()`

. `demean()`

now additionally converts factors with more than 2 levels to dummy-variables (binary), to mimic*panelr*-behaviour.

- Fixed minor issue with the
`print()`

-method for`model_parameters.befa()`

. - Fixed issues in
`model_parameters()`

(for linear mixed models) with wrong order of degrees of freedom when`df_method`

was different from default. - Fixed issues in
`model_parameters()`

(for linear mixed models) with accuracy of p-values when`df_method = "kenward`

. - Fixed issues in
`model_parameters()`

with wrong test statistic for*lmerModLmerTest*models. - Fixed issue in
`format_parameters()`

(which is used to format output of`model_parameters()`

) for factors, when variable name was also part of factor levels. - Fixed issue in
`degrees_of_freedem()`

for*logistf*-models, which unintentionally printed the complete model summary. - Fixed issue in
`model_parameters()`

for*mlm*models. - Fixed issue in
`random_parameters()`

for uncorrelated random effects.

`skewness()`

now uses a different method to calculate the skewness by default. Different methods can be selected using the`type`

-argument.`kurtosis()`

now uses a different method to calculate the skewness by default. Different methods can be selected using the`type`

-argument.

- Added support for
`cglm`

(*cglm*),`DirichletRegModel`

(*DirichletReg*)

- Added new vignettes on ‘Standardized Model Parameters’ and ‘Robust Estimation of Standard Errors’, and vignettes are now also published on CRAN.
- Improved handling of robust statistics in
`model_parameters()`

. This should now work for more models than before. - Improved accuracy of
`ci.merMod()`

for`method = "satterthwaite"`

and`method = "kenward"`

. `select_parameters()`

for*stanreg*models, which was temporarily removed due to the CRAN removal of package**projpred**, is now re-implemented.

`dof_betwithin()`

to compute degrees of freedom based on a between-within approximation method (and related to that,`p_value_*()`

and`se_*()`

for this method were added as well).`random_parameters()`

that returns information about the random effects such as variances, R2 or ICC.`closest_component()`

as a small helper that returns the component index for each variable in a data frame that was used in`principal_components()`

.`get_scores()`

as a small helper to extract scales and calculate sum scores from a principal component analysis (PCA,`principal_components()`

).

`n_clusters()`

gets the option`"M3C"`

for the`package`

-argument, so you can try to determine the number of cluster by using the`M3C::M3C()`

function.- The
`print()`

-method for`model_parameters()`

gets a`select`

-argument, to print only selected columns of the parameters table. `model_parameters()`

for meta-analysis models has an improved`print()`

-method for subgroups (see examples in`?model_parameters.rma`

).`model_parameters()`

for mixed models gets a`details`

-argument to additionally print information about the random effects.`model_parameters()`

now accepts the`df_method`

-argument for more (mixed) models.- The Intercept-parameter in
`model_parameters()`

for meta-analysis models was renamed to`"Overall"`

. `skewness()`

gets a`type`

-argument, to compute different types of skewness.`kurtosis()`

gets a`type`

-argument, to compute different types of skewness.`describe_distribution()`

now also works on data frames and gets a nicer print-method.

- Fixed issue in
`model_parameters()`

when`robust = TRUE`

, which could sometimes mess up order of the statistic column. - Fixed issues in
`model_parameters()`

with wrong`df`

for`lme`

-models. - Fixed issues in
`model_parameters.merMod()`

when`df_method`

was not set to default. - Fixed issues in
`model_parameters.merMod()`

and`model_parameters.gee()`

when`robust = TRUE`

. - Fixed issues with
*coxph*models with only one parameter. - Fixed issue in
`format_p()`

when argument`digits`

was`"apa"`

. - Fixed issues in
`model_parameters()`

for`zeroinfl`

-models.

- Fix CRAN check issues, caused by removal of package ‘projpred’.

- The column for degrees of freedom in
`model_parameters()`

was renamed from`df_residuals`

to`df_error`

for regression model objects, because these degrees of freedom actually were not always referring to*residuals*- we consider`df_error`

as a more generic name. `model_parameters()`

for standardized parameters (i.e.`standardize`

is not`NULL`

) only returns standardized coefficients, CI and standard errors (and not both, unstandardized and standardized values).`format_ci()`

was removed and re-implemented in the**insight**package.

`model_bootstrap()`

was renamed to`bootstrap_model()`

.`model_bootstrap()`

will remain as alias.`parameters_bootstrap()`

was renamed to`bootstrap_parameters()`

.`parameters_bootstrap()`

will remain as alias.`model_simulate()`

was renamed to`simulate_model()`

.`model_simulate()`

will remain as alias.`parameters_simulate()`

was renamed to`simulate_parameters()`

.`parameters_simulate()`

will remain as alias.`parameters_selection()`

was renamed to`select_parameters()`

.`parameters_selection()`

will remain as alias.`parameters_reduction()`

was renamed to`reduce_parameters()`

.`parameters_reduction()`

will remain as alias.

- Added support for
`vgam`

(*VGAM*),`cgam`

,`cgamm`

(*cgam*),`complmrob`

(*complmrob*),`cpglm`

,`cpglmm`

(*cplm*),`fixest`

(*fixest*),`feglm`

(*alpaca*),`glmx`

(*glmx*),`glmmadmb`

(*glmmADMB*),`mcmc`

(*coda*),`mixor`

(*mixor*). `model_parameters()`

now supports`blavaan`

models (*blavaan*).

- Better handling of
`clm2`

,`clmm2`

and`stanmvreg`

models. - Better handling of
`psych::omega`

models.

`dof_satterthwaite()`

and`dof_ml1()`

to compute degrees of freedom based on different approximation methods (and related to that,`p_value_*()`

and`se_*()`

for these methods were added as well).`rescale_weights()`

to rescale design (probability or sampling) weights for use in multilevel-models without survey-design.

- Robust estimation (like
`standard_error_robust()`

or`ci_robust()`

) can now also compute cluster-robust variance-covariance matrices, using the*clubSandwich*package. `model_parameters()`

gets a`robust`

-argument, to compute robust standard errors, and confidence intervals and p-values based on robust standard errors.- Arguments
`p_method`

and`ci_method`

in`model_parameters.merMod()`

were replaced by a single argument`df_method`

. `model_parameters.principal()`

includes a`MSA`

column for objects from`principal_components()`

.

- Fixed issue in
`model_parameters()`

with non-typical ordering of coefficients for mixed models. - Fixed issues with models of class
`rlmerMod`

. - Fixed minor issues
`model_parameters.BFBayesFactor()`

.

Parts of the **parameter** package are restructured and functions focussing on anything related to effect sizes are now re-implemented in a new package, **effectsize**. In details, following breaking changes have been made:

- Functions for computing effect sizes (
`cohens_f()`

,`eta_squared()`

etc.) have been removed and are now re-implemented in the**effectsize**-package. - Functions for converting effect sizes (
`d_to_odds()`

etc.) have been removed and are now re-implemented in the**effectsize**-package. `standardize()`

and`normalize()`

(and hence, also`parameters_standardize()`

) have been removed ;-( and are now re-implemented in the**effectsize**-package.

- Added support for
`aareg`

(*survival*),`bracl`

,`brmultinom`

(*brglm2*),`rma`

(*metafor*) and`multinom`

(*nnet*) to various functions. `model_parameters()`

for`kmeans`

.`p_value()`

,`ci()`

,`standard_error()`

and`model_parameters()`

now support*flexsurvreg*models (from package**flexsurv**).

`degrees_of_freedom()`

to get DoFs.`p_value_robust()`

,`ci_robust()`

and`standard_error_robust()`

to compute robust standard errors, and p-values or confidence intervals based on robust standard errors.`demean()`

to calculate de-meaned and group-meaned variables (centering within groups, for panel-data regression).`n_parameters()`

to get number of parameters.`n_clusters()`

to determine the number of clusters to extract.`cluster_analysis()`

to return group indices based on cluster analysis.`cluster_discrimination()`

to determine the goodness of classification of cluster groups.`check_clusterstructure()`

to check the suitability of data for clustering.`check_multimodal()`

to check if a distribution is unimodal or multimodal.- Add
`plot()`

-methods for`principal_components()`

.

- Added indices of model fit to
`n_factors()`

(Finch, 2019) `standard_error()`

for mixed models gets an`effects`

argument, to return standard errors for random effects.- The
`method`

-argument for`ci()`

gets a new option,`"robust"`

, to compute confidence intervals based on robust standard errors. Furthermore,`ci_wald()`

gets a`robust`

-argument to do the same. `format_p()`

gets a`digits`

-argument to set the amount of digits for p-values.`model_parameters()`

now accepts (non-documented) arguments`digits`

,`ci_digits`

and`p_digits`

to change the amount and style of formatting values. See examples in`model_parameters.lm()`

.- Improved
`print()`

method for`model_parameters()`

when used with Bayesian models. - Added further method (gap-statistic) to
`n_clusters()`

.

- Interaction terms in
`model_parameters()`

were denoted as nested interaction when one of the interaction terms was surrounded by a function, e.g.`as.factor()`

,`log()`

or`I()`

. - Fixed bug in
`parameters_type()`

when a parameter occured multiple times in a model. - Fixed bug with
*multinom*-support. - Fixed bug in
`model_parameters()`

for non-estimable GLMs. - Fixed bug in
`p_value()`

for*MASS::rlm*models. - Fixed bug in
`reshape_loadings()`

when converting loadings from wide to long and back again.

`format_value()`

and`format_table()`

have been removed and are now re-implemented in the**insight**package.

`parameters()`

is an alias for`model_parameters()`

.`p_value()`

,`ci()`

,`standard_error()`

,`standardize()`

and`model_parameters()`

now support many more model objects, including mixed models from packages*nlme*,*glmmTMB*or*GLMMadaptive*, zero-inflated models from package*pscl*or other modelling packages. Along with these changes, functions for specific model objects with zero-inflated component get a`component`

-argument to return the requested values for the complete model, the conditional (count) component or the zero-inflation component from the model only.

`parameters_simulate()`

and`model_simulate()`

, as computational faster alternatives to`parameters_bootstrap()`

and`model_bootstrap()`

.`data_partition()`

to partition data into a test and a training set.`standardize_names()`

to standardize column names from data frames, in particular objects returned from`model_parameters()`

.`se_kenward()`

to calculate approximated standard errors for model parameters, based on the Kenward-Roger (1997) approach.

`format_value()`

and`format_ci()`

get a`width`

-argument to set the minimum length of the returned formatted string.`format_ci()`

gets a`bracket`

-argument include or remove brackets around the ci-values.`eta_squared()`

,`omega_squared()`

,`epsilon_squared()`

and`cohens_f()`

now support more model objects.- The
`print()`

-method for`model_parameters()`

now better aligns confidence intervals and p-values. `normalize()`

gets a`include_bounds`

-argument, to compress normalized variables so they do not contain zeros or ones.- The
`method`

-argument for`ci.merMod()`

can now also be`"kenward"`

to compute confidence intervals with degrees of freedom based on the Kenward-Roger (1997) approach.

- Fixed issue with wrong computation of wald-approximated confidence intervals.
- Fixed issue with wrong computation of degrees of freedom for
`p_value_kenward()`

. `paramerers_standardize()`

resp.`standardize()`

for model objects now no longer standardizes`log()`

terms, count or ratio response variables, or variables of class`Surv`

and`AsIs`

.

- Added a
`NEWS.md`

file to track changes to the package