R 4.3.{easystats}
packages.{easystats} package versions to avoid
user-facing warnings due to API changes upstream.Test and effect size details vignette is now available only on the package website (https://indrajeetpatil.github.io/statsExpressions/articles/).
Unused dataset has been removed:
movies_wide.
The cryptic but very useful parameter k has been
renamed to digits to improve its discoverability.
To be consistent, contingency_table() gains
alternative parameter, which defaults to two-sided
alternative hypothesis.
{afex} and
{PMCMRplus}.R 4.1
because a critical dependency ({pbkrtest}) requires this R
version.{statsExpressions} get a
new statsExpressions class and a print method for this
class.tidy_model_parameters() no longer removes columns
which contain only missing values.
Wilcox tests no longer provide exact p-values.
centrality_description() gets a new
conf.level parameter.
Updates tests and examples to insure against removal of suggested packages.
{effectsize} package
update.tidy_model_expressions() now returns a
NULL instead of an expression with empty strings.pairwise_comparisons() function to carry out
pairwise comparison tests for one-way designs.Expressions with statistical details were sometimes in a column
named label, while other times in expression
column. Now it will be consistently in the expression
column.
Additionally, glue expressions were stored parsed in some data frame outputs, while unparsed in others. Now it is consistently parsed.
The top.text parameter has been removed from all
functions. It was relevant only in the context of
{ggstatsplot} package. As that package no longer uses it,
it is no longer necessary to retain it.
{insight} package update.format_num() has been removed in favor of
insight::format_value().expr_template() has been renamed to more
informative add_expression_col() function and has a
different API. It returns a dataframe with the additional expression
column instead of just the expression.A number of effect size estimates and their confidence intervals
have changed due to respective changes made in {effectsize}
package version 0.5 release. For full details of these
changes, see: https://easystats.github.io/effectsize/news/index.html
For the same reason, the effect size for one-way contingency table has changed from Cramer’s V to Pearson’s C.
centrality_description() function added to describe
distribution for each level of a grouping variable and create an
expression describing a centrality measure.
Adds new experimental function
tidy_model_expressions() to create expressions for
dataframes containing tidied results from regression model
objects.
Removes the redundant bf_extractor function. The
tidy_model_parameters does the same thing.
Exports more utility functions
(long_to_wide_converter, format_num,
stats_type_switch) to get rid of reliance on
ipmisc package.
To be consistent with the expressions, the dataframe for Bayesian analysis now also contain log of Bayes Factor values.
The tidy_model_effectsize() function is no longer
exported as it is helpful only for the internal workings of the
package.
Given that these values can be really high, the statistic values for non-parametric tests were shown on a log scale, but this is a highly non-standard practice that has caused a lot of confusion among users. In light of this feedback, the functions no longer return these values on a log scale but in a scientific notation to keep the statistical expressions short.
VR_dilemma dataset, which lacked enough
variation to be a good dataset to use in examples or tests.There is a new JOSS paper about
{statsExpressions} package!! https://joss.theoj.org/papers/10.21105/joss.03236
The effect size for independent trimmed means two-sample test has been changed from explanatory measure of effect size to AKP’s delta, which is easier to understand and interpret since its a robust cousin of Cohen’s d.
one_sample_test and two_sample_test
gain alternative argument to specify alternative hypothesis
(#86).
Cohen’s d and Hedge’s g use non-pooled standard deviation (cf. https://psyarxiv.com/tu6mp/).
effectsize package).expr_* functions.effectsize.This is the first stable release of
{statsExpressions}!
There is good news and there is bad news that accompanies this milestone.
The bad news: The API for the
package has changed completely: All functions return a
dataframe, and not an expression, as a default. The
expression is contained in a list column in the dataframe itself. So, to
salvage your functions from breaking, you will have to add
$expression[[1]] to your function calls. For example, if
you were using the function expr_t_onesample(), you will
now have to specify expr_t_onesample()$expression[[1]], so
on and so forth. But, in general, the advice is to not
use any of the expr_* functions, which are vestigial names
for new avatars of these function and will be removed in future. The new
names are more intuitive, e.g., expr_t_onesample() is now
called one_sample_test(), etc.
The good news: There will not be any new changes to any of the current functions, except for any change necessary for maintenance or bug squashing. Well, to be more precise, this is true only for the functions that have “stable” badge.
BUG FIXES
grouped-tibble, the
function internally ungroups this (#79).afex has moved from
Imports to Suggests.tr = 0.1 to tr = 0.2
(which is what WRS2 defaults to).expr_template gains a new argument
bayesian, which can return an expression for Bayesian
analysis, which has a slightly different template. Additionally, it has
changed its conventions about the column names it expects.
Retires the additional caption-making functionality that was
unique to expr_meta_random when
type = "parametric". This was the only context in which
this feature was supported and was therefore inconsistent with the rest
of the package API.
Removes tidy_model_performance function, which is no
longer used internally.
Removes column containing log values of Bayes Factor
as they are relevant only for expressions.
All meta-analysis packages move from Imports to
Suggests to reduce the installation time for the
user.
All robust tests in this package were based on trimmed means,
except for correlation test. This has been changed: the robust
correlation measure is now Winsorized correlation, which is based on
trimming. Therefore, the beta argument has been replaced by
tr argument. This should result only in minor changes in
correlation coefficient estimates.
To be consistent with ggstatsplot’s overall syntax
philosophy the type argument can be used to specify which
type of statistical approach is to be used for all functions.
t_parametric, t_nonparametric,
t_robust, t_bayes are now removed in favor of
a single function two_sample_test.
expr_anova_parametric,
expr_anova_nonparametric, expr_anova_robust,
expr_anova_bayes are now removed in favor of a single
function oneway_anova.
{statsExpressions} no longer internally relies on
tidyBF. All Bayesian analysis is carried out in this
package itself. This was done to make the maintenance of this package
easier and helps with some major internal code refactoring. As such, all
re-exported functions from tidyBF have also been
removed.
BUG FIXES
contingency_table ignored ratio argument
while computing Cramer’s V for one-sample test. This is
fixed.All non-parametric functions now use effectsize
package to compute effect sizes and not rcompanion. This
would lead to some changes in effect sizes and their confidence
intervals reported by the respective functions.
Robust one-sample test is changed from one-sample percentile bootstrap to bootstrap-t method for one-sample test, which uses trimmed mean like the rest of the robust functions in this package.
Package internally relies on afex instead of
ez for within-subjects ANOVA.
expr_template gains paired
argument.
Internal refactoring to catch up with changes made to
effectsize. Tests are adapted to these changes as
well.
Sample size information in expressions is pretty-formatted.
Adds two new helper functions: tidy_model_parameters
and tidy_model_performance to toggle between
easystats and tidymodels naming
conventions.
Drops broomExtra from dependencies in favor of
parameters + performance.
Removes the unused and vestigial Titanic_full
dataset.
Removes the alias expr_onesample_proptest.
The expr_template function retires
effsize.df argument. Now all details need to be entered
only in data.
All meta-analyses are now carried out using
expr_meta_random and the individual functions have been
removed.
All effect sizes for contingency tabs are now calculated via
effectsize instead of rcompanion. This would
lead to slight differences in effect sizes and their CIs but the
computations will be faster. Additionally, the lower bound will never be
negative and will be restricted to [0,1].
contingency_table function has been made less
robust. It now fails instead of returning NULL when it is
not supposed to work. This is done to be consistent with the other
functions in the package which also fail instead of returning
NULL.
expr_anova_parametric always applies sphericity
correction for p-values for repeated measures ANOVA.
expr_anova_parametric retires non-partial variants
of effect sizes (eta-squared and omega-squared, i.e.) for parametric
analyses.
The t-test and ANOVA tests get subject.id
argument relevant for repeated measures design.
Retires the vestigial stat.title argument. It was
originally intended to give more info on the tests, but now the
expressions themselves contain these details.
For paired ANOVA designs, partial = TRUE is
recognized by effect sizes.
Retires bias.correct argument for contingency table
analysis. It is rarely justifiable to use the biased version of Cramer’s
V.
Adapts tests to changes made in the correlation
package.
Subtitles for correlation tests make clear the type of statistic.
Small p-values (< 0.001) are now shown in scientific format.
Adapts to changes made in tidyBF package.
Re-exports correlation::correlation needed for
ggstatsplot.
The t_nonparametric subtitle now clarifies whether
it’s a Wilcoxon test or a Mann-Whitney test.
Thanks to Sarah, the package has a hexsticker. :)
Confidence intervals for Spearman’s rho are computed using
correlation instead of rcompanion.
All relevant functions get rid of messages argument
as the functions no longer print a message when bootstrapped CIs are
used.
The effect size measure for paired robust t-test is now changed to robust (trimmed-Winsorized) standardized difference similar to Cohen’s d.
BUG FIXES
0.4.0 release for
expr_anova_parametric: changing conf.level
doesn’t work and function defaults to 0.90 CIs (#32).Removes the experimental corr_objects
function.
All Bayes Factor related functions have now moved to the new
tidyBF package and are re-exported from there.
Minimum R version bumped to R 3.6.0.
Retires the internal effsize_t_parametric helper
function in favor of relying functions from effectsize,
which is now added as a dependency. Similarly,
{statsExpressions} now relies on effectsize to
compute effect sizes for ANOVA designs, instead of
sjstats.
For parametric t-tests and ANOVAs, confidence intervals for effect sizes are estimated using the noncentrality parameter method. Centrality-based methods are deprecated.
Correlation analysis is carried out using
correlation package, which is now added as a
dependency.
corr_objects to reduce dependency
load of ggstatsplot. This is an experimental function and
should be avoided until it stabilizes.expr_meta_bayes.expr_meta_parametric, expr_meta_robust,
bf_meta.expr_template function now expects two dataframes:
data and effsize.df that contain the details
needed for creating expressions instead of providing each individual
values. This makes the function more friendly work with using modeling
packages like broom.Minor tweaks to how widehat is displayed in some of the expressions.
Cramer’s V is bias-corrected by default.
Removes MCMCpack from Depends.
All effect size texts now contain ^ on top to
signify that these are estimates.
CRAN.Fixing tests for the new release of rcompanion
dependency.
Minor code refactoring.