:bar_chart: Computation and processing of models' parameters
flic and flac (logistf), mmrm (mmrm).model_parameters() now includes a Group column for stanreg or brmsfit
models with random effects.
The print() method for model_parameters() now uses the same pattern to
print random effect variances for Bayesian models as for frequentist models.
Fixed issue with the print() method for compare_parameters(), which
duplicated random effects parameters rows in some edge cases.
Fixed issue with the print() method for compare_parameters(), which
didn't work properly when ci=NULL.
The deprecated argument df_method in model_parameters() is now defunct
and throws an error when used.
The deprecated functions ci_robust(), p_robust() and standard_error_robust
have been removed. These were superseded by the vcov argument in ci(),
p_value(), and standard_error(), respectively.
The style argument in compare_parameters() was renamed into select.
p_function(), to print and plot p-values and compatibility (confidence)
intervals for statistical models, at different levels. This allows to see
which estimates are most compatible with the model at various compatibility
levels.
p_calibrate(), to compute calibrated p-values.
model_parameters() and compare_parameters() now use the unicode character
for the multiplication-sign as interaction mark (i.e. \u00d7). Use
options(parameters_interaction = <value>) or the argument interaction_mark
to use a different character as interaction mark.
The select argument in compare_parameters(), which is used to control the
table column elements, now supports an experimental glue-like syntax.
See this vignette Printing Model Parameters. Furthermore, the select
argument can also be used in the print() method for model_parameters().
print_html() gets a font_size and line_padding argument to tweak the
appearance of HTML tables. Furthermore, arguments select and column_labels
are new, to customize the column layout of tables. See examples in ?display.
Consolidation of vignettes on standardization of model parameters.
Minor speed improvements.
model_parameters().BFBayesFactor no longer drops the BF column if the
Bayes factor is NA.
The print() and display() methods for model_parameters() from Bayesian
models now pass the ... to insight::format_table(), allowing extra
arguments to be recognized.
Fixed footer message regarding the approximation method for CU and p-values for mixed models.
Fixed issues in the print() method for compare_parameters() with mixed
models, when some models contained within-between components (see
wb_component) and others did not.
Arguments that calculate effectsize in model_parameters() for htest,
Anova objects and objects of class BFBayesFactor were revised. Instead of
single arguments for the different effectsizes, there is now one argument,
effectsize_type. The reason behind this change is that meanwhile many
new type of effectsizes have been added to the effectsize package, and
the generic argument allows to make use of those effect sizes.
The attribute name in PCA / EFA has been changed from data_set to dataset.
The minimum needed R version has been bumped to 3.6.
Removed deprecated argument parameters from model_parameters().
standard_error_robust(), ci_robust() and p_value_robust() are now
deprecated and superseded by the vcov and vcov_args arguments in the
related methods standard_error(), ci() and p_value(), respectively.
Following functions were moved from package parameters to performance:
check_sphericity_bartlett(), check_kmo(), check_factorstructure() and
check_clusterstructure().
Added sparse option to principal_components() for sparse PCA.
The pretty_names argument from the print() method can now also be
"labels", which will then use variable and value labels (if data is
labelled) as pretty names. If no labels were found, default pretty names
are used.
bootstrap_model() for models of class glmmTMB and merMod gains a
cluster argument to specify optional clusters when the parallel
option is set to "snow".
P-value adjustment (argument p_adjust in model_parameters()) is now
performed after potential parameters were removed (using keep or drop),
so adjusted p-values is only applied to the parameters of interest.
Robust standard errors are now supported for fixest models with the vcov
argument.
print() for model_parameters() gains a footer argument, which can be
used to suppress the footer in the output. Further more, if footer = ""
or footer = FALSE in print_md(), no footer is printed.
simulate_model() and simulate_parameters() now pass ... to
insight::get_varcov(), to allow simulated draws to be based on
heteroscedasticity consistent variance covariance matrices.
The print() method for compare_parameters() was improved for models with
multiple components (e.g., mixed models with fixed and random effects, or
models with count- and zero-inflation parts). For these models,
compare_parameters(effects = "all", component = "all") prints more nicely.
dominance_analysis(), to compute dominance analysis
statistics and designations.ci_random in model_parameters() defaults to NULL. It uses a
heuristic to determine if random effects confidence intervals are likely to
take a long time to compute, and automatically includes or excludes those
confidence intervals. Set ci_random to TRUE or FALSE to explicitly
calculate or omit confidence intervals for random effects.Fix issues in pool_parameters() for certain models with special components
(like MASS::polr()), that failed when argument component was set to
"conditional" (the default).
Fix issues in model_parameters() for multiple imputation models from
package Hmisc.
Release for JOSS