statsExpressions

MAJOR CHANGES

• The package licence has been updated to an MIT license.

MINOR CHANGES

• Updates code to fix warnings coming via updates to easystats packages.

MAJOR CHANGES

• 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.

• Internal changes only.

• Fixes failures on R-devel due to changes in the formula interface for a paired test.

MAJOR CHANGES

• Following dependencies move from being weak to strong dependencies given their
  importance in key functions: {afex} and {PMCMRplus}.

BREAKING CHANGES

• The minimum needed R version is now bumped to R 4.1 because a critical
  dependency ({pbkrtest}) requires this R version.

NEW FEATURES

• All returned data frames from {statsExpressions} get a new
  statsExpressions class and a print method for this class.

MINOR CHANGES

• 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 new conf.level parameter.

• Updates tests and examples to insure against removal of suggested packages.

• Fixes breakages due to {effectsize} package update.

MINOR CHANGES

• When any of the necessary numeric column values are missing, for these rows
  tidy_model_expressions() now returns a NULL instead of an expression with
  empty strings.

NEW FUNCTIONS

• The pairwise_comparisons() function to carry out pairwise comparison tests
  for one-way designs.

MAJOR CHANGES

• 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.

• Fix tests for insight 0.17.0

BREAKING CHANGES

• The format_num() has been removed in favor of insight::format_value().

MINOR CHANGES

• The 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.

BREAKING CHANGES

• 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.

NEW FUNCTIONS

• 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.

MAJOR CHANGES

• 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.

MINOR CHANGES

• Removes VR_dilemma dataset, which lacked enough variation to be a good
  dataset to use in examples or tests.

MAJOR CHANGES

• 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/).

MINOR CHANGES

• The output dataframes now contain columns with additional information about
  how confidence intervals are computed (thanks to effectsize package).

BREAKING CHANGES

• Retires all vestigial expr_* functions.

MINOR CHANGES

• 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
  $expression[[1]] to your function calls. For example, if you were using
  the function expr_t_onesample(), you will now have to specify
  expr_t_onesample()$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

• If the entered dataframe is grouped-tibble, the function internally
  ungroups this (#79).

MINOR CHANGES

• To reduce dependency load, afex has moved from Imports to Suggests.

BREAKING CHANGES

• To avoid confusion among users, the trimming level for all functions is now
  changed from tr = 0.1 to tr = 0.2 (which is what WRS2 defaults to).

MAJOR CHANGES

• 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.

BREAKING CHANGES

• 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.

MAJOR CHANGES

• 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.

MINOR CHANGES

• Package internally relies on afex instead of ez for within-subjects
  ANOVA.

• expr_template gains paired argument.