- The minimum needed R version is now bumped to
R 4.3
. - Fixes failing tests due to changes in
{easystats}
packages.
- Upgrade
{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 package licence has been updated to an MIT license.
- Updates code to fix warnings coming via updates to easystats packages.
-
The cryptic but very useful parameter
k
has been renamed todigits
to improve its discoverability. -
To be consistent,
contingency_table()
gainsalternative
parameter, which defaults to two-sided alternative hypothesis.
- Internal changes only.
- Fixes failures on R-devel due to changes in the formula interface for a paired test.
- Following dependencies move from being weak to strong dependencies given their
importance in key functions:
{afex}
and{PMCMRplus}
.
- The minimum needed R version is now bumped to
R 4.1
because a critical dependency ({pbkrtest}
) requires this R version.
- All returned data frames from
{statsExpressions}
get a newstatsExpressions
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.
- Maintenance release.
-
centrality_description()
gets a newconf.level
parameter. -
Updates tests and examples to insure against removal of suggested packages.
- Fixes breakages due to
{effectsize}
package update.
- When any of the necessary numeric column values are missing, for these rows
tidy_model_expressions()
now returns aNULL
instead of an expression with empty strings.
- The
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 inexpression
column. Now it will be consistently in theexpression
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.
- Fixes breakages due to
{insight}
package update.
- The
format_num()
has been removed in favor ofinsight::format_value()
.
- The
expr_template()
has been renamed to more informativeadd_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 version0.5
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. Thetidy_model_parameters
does the same thing. -
Exports more utility functions (
long_to_wide_converter
,format_num
,stats_type_switch
) to get rid of reliance onipmisc
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.
- Removes
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
andtwo_sample_test
gainalternative
argument to specify alternative hypothesis (#86). -
Cohen's d and Hedge's g use non-pooled standard deviation (cf. https://psyarxiv.com/tu6mp/).
- The output dataframes now contain columns with additional information about
how confidence intervals are computed (thanks to
effectsize
package).
- Retires all vestigial
expr_*
functions.
- Adapts failing tests due to changes in
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$expression[[1]]
to your function calls. For example, if you were using the functionexpr_t_onesample()
, you will now have to specifyexpr_t_onesample()$expression[[1]]
, so on and so forth. But, in general, the advice is to not use any of theexpr_*
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 calledone_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
- If the entered dataframe is
grouped
-tibble, the function internally ungroups this (#79).
- To reduce dependency load,
afex
has moved fromImports
toSuggests
.
- To avoid confusion among users, the trimming level for all functions is now
changed from
tr = 0.1
totr = 0.2
(which is whatWRS2
defaults to).
-
expr_template
gains a new argumentbayesian
, 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
whentype = "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
toSuggests
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 bytr
argument. This should result only in minor changes in correlation coefficient estimates.
-
To be consistent with
ggstatsplot
's overall syntax philosophy thetype
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 functiontwo_sample_test
. -
expr_anova_parametric
,expr_anova_nonparametric
,expr_anova_robust
,expr_anova_bayes
are now removed in favor of a single functiononeway_anova
.
-
-
{statsExpressions}
no longer internally relies ontidyBF
. 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 fromtidyBF
have also been removed.
BUG FIXES
contingency_table
ignoredratio
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 notrcompanion
. 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 ofez
for within-subjects ANOVA. -
expr_template
gainspaired
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
andtidy_model_performance
to toggle betweeneasystats
andtidymodels
naming conventions. -
Drops
broomExtra
from dependencies in favor ofparameters
+performance
. -
Removes the unused and vestigial
Titanic_full
dataset.
-
Removes the alias
expr_onesample_proptest
. -
The
expr_template
function retireseffsize.df
argument. Now all details need to be entered only indata
. -
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 ofrcompanion
. 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 returningNULL
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 returningNULL
. -
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 forggstatsplot
. -
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 ofrcompanion
. -
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
- Major bug introduced in
0.4.0
release forexpr_anova_parametric
: changingconf.level
doesn't work and function defaults to0.90
CIs (#32).
- Removes extra space contained in subtitles for Bayes Factor results (#31).
-
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 fromeffectsize
, which is now added as a dependency. Similarly,{statsExpressions}
now relies oneffectsize
to compute effect sizes for ANOVA designs, instead ofsjstats
. -
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.
- All expressions now contain name of the statistical test carried out.
- Adds a new function
corr_objects
to reduce dependency load ofggstatsplot
. This is an experimental function and should be avoided until it stabilizes.
- New function to carry out meta-analysis:
expr_meta_bayes
.
- New function to carry out meta-analysis:
expr_meta_parametric
,expr_meta_robust
,bf_meta
.
expr_template
function now expects two dataframes:data
andeffsize.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 likebroom
.
-
Minor tweaks to how widehat is displayed in some of the expressions.
-
Cramer's V is bias-corrected by default.
-
Removes
MCMCpack
fromDepends
. -
All effect size texts now contain
^
on top to signify that these are estimates.
- Maintenance release to fix additional check issues on
CRAN
.
-
Fixing tests for the new release of
rcompanion
dependency. -
Minor code refactoring.
- First release of the package.