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