NEWS.md

effectsize 0.8.7.xxx

• This release changes the licensing model of {effectsize} to an MIT license.

New features

• repeated_measures_d() to compute standardized mean differences (SMD) for repeated measures data.
  • Also supported in effectsize(<t.test(paired = TRUE)>)

Bug fixes

• nnt() now properly accepts the y argument.

effectsize 0.8.7

New features

• New function: interpret_fei()

effectsize 0.8.6

This is a minor update to bring effectsize in-line with the formula methods in t.test() and wilcox.test() in R>=4.4.0.

Breaking Changes

• cohens_d(), hedges_g(), p_superiority(), wmw_odds(), means_ratio() and rank_biserial() no longer support setting paired = TRUE when using the formula method.

Bug fixes

• eta_squared(<gam>) returns (approximate) effect sizes for smooths.

effectsize 0.8.5

New features

• interpret_cfi() gains a new rule option: "hu&bentler1999" ( #538 ).
• cohens_f() added option to return unbiased estimators (based on Omega- or Epsilon-squared).
• tschuprows_t() now returns an effect size corrected for small-sample bias. Set adjust = FALSE to preserve old behavior.
• w_to_v() and others for converting between effect sizes of Chi-square tests.
• arr() and nnt() for Absolute Risk Reduction or Number Needed to Treat.
• oddsratio_to_arr(), riskratio_to_arr(), nnt_to_arr() and their inverses.
• logoddsratio_to_*() and *_to_logoddsratio() have been added as convenient shortcuts for oddsratio_to_*(log = TRUE) and *_to_oddsratio(log = TRUE).
• Added all missing functions to convert between (log) OR, RR, ARR, and NNT.

Changes

• fei() gives a more informative error method for invalid table inputs (#566).
• convert_*() aliases are deprecated.

Breaking Changes

• *_to_riskratio() and riskratio_to_*() argument log not longer converts RR to/from log(RR).
• interpret_gfi() and friends: some previously named "default" rules have been re-labelled as "byrne1994".

Bug fixes

• riskratio() returns correct CIs (#584)
• d_to_r() correctly treats specifying only n1/n2 as equal group sizes (#571)

effectsize 0.8.3

Changes

• mahalanobis_d() now defaults to one-sided CIs.

New features

• means_ratio() for computing ratios of two means for ratio-scales outcomes (thanks to @arcaldwell49!)
• r_to_d() family of functions gain arguments for specifying group size ( #534 )
• r2_semipartial for semi-partial squared correlations of model terms / parameters.

Bug fixes

• ANOVA effect sizes for afex::mixed() now return effect sizes for the Intercept where applicable.
• Fixed error in cohens_w() for 2-by-X tables.
• Solved integer overflow errors in rank_biserial() ( #476 )
• Fixed issue in effectsize() for t-tests when input vectors has unequal amount of missing values.

effectsize 0.8.2

Breaking Changes

• omega_squared() and epsilon_squared() (and F_to_omega2() and F_to_epsilon2()) always return non-negative estimates (previously estimates were negative when the observed effect size is very small).
• rank_eta_squared() always returns a non-negative estimate (previously estimates were negative when the observed effect size is very small).

effectsize 0.8.1

Changes

• cohens_w() has an exact upper bound when used as an effect size for goodness-of-fit.

Bug fixes

• When using formula input to effect size function, na.action arguments are respected (#517)

effectsize 0.8.0

Breaking Changes

• {effectsize} now requires R >= 3.6
• fei(), cohens_w() and pearsons_c() always rescale the p input to sum-to-1.
• The order of some function arguments have been rearranged to be more consistent across functions: (phi(), cramers_v(), p_superiority(), cohens_u3(), p_overlap(), rank_biserial(), cohens_f/_squared(), chisq_to_phi(), chisq_to_cramers_v(), F/t_to_f/2(), .es_aov_*()).
• normalized_chi() has been renamed fei().
• cles, d_to_cles and rb_to_cles are deprecated in favor of their respective effect size functions.

Changes

• phi() and cramers_v() (and chisq_to_phi/cramers_v()) now apply the small-sample bias correction by default. To restore previous behavior, set adjust = FALSE.

New features

• Set options(es.use_symbols = TRUE) to print proper symbols instead of transliterated effect size names. (On Windows, requires R >= 4.2.0)
• effectsize() supports fisher.test().
• New datasets used in examples and vignettes - see data(package = "effectsize").
• tschuprows_t() and chisq_to_tschuprows_t() for computing Tschuprow's T - a relative of Cramer's V.
• mahalanobis_d() for multivariate standardized differences.
• Rank based effect sizes now accept ordered (ordered()) outcomes.
• rank_eta_squared() for one-way rank ANOVA.
• For Common Language Effect Sizes:
  • wmw_odds() and rb_to_wmw_odds for the Wilcoxon-Mann-Whitney odds (thanks @arcaldwell49! #479).
  • p_superiority() now supports paired and one-sample cases.
  • vd_a() and rb_to_vda() for Vargha and Delaney's A dominance effect size (aliases for p_superiority(parametric = FALSE) and rb_to_p_superiority()).
  • cohens_u1(), cohens_u2(), d_to_u1(), and d_to_u2() added for Cohen's U1 and U2.

Bug fixes

• Common-language effect sizes now respects mu argument for all effect sizes.
• mad_pooled() not returns correct value (previously was inflated by a factor of 1.4826).
• pearsons_c() and chisq_to_pearsons_c() lose the adjust argument which applied an irrelevant adjustment to the effect size.
• Effect sizes for goodness-of-fit now work when passing a p that is a table.

effectsize 0.7.0.5

Breaking Changes

effectsize now requires minimal R version of 3.5.

Bug fixes

• cohens_d() for paired / one sample now gives more accurate CIs (was off by a factor of (N - 1) / N; #457)
• kendalls_w() now deals correctly with singular ties (#448).

effectsize 0.7.0

Breaking Changes

• standardize_parameters(), standardize_posteriors(), & standardize_info() have been moved to the parameters package.
• standardize() (for models) has been moved to the datawizard package.
• phi() only works for 2x2 tables.
• cramers_v() only works for 2D tables.

New features

• normalized_chi() gives an adjusted Cohen's w for goodness of fit.
• cohens_w() is now a fully-fledged function for x-tables and goodness-of-fit effect size (not just an alias for phi()).
• Support for insight's display, print_md and print_html for all {effectsize} outputs.

Bug fixes

• kendalls_w() now deals with ties.
• eta_squared() works with car::Manova() that does not have an i-design.

effectsize 0.6.0.1

This is a patch release.

Bug fixes

• interpret.performance_lavaan() now works without attaching effectsize ( #410 ).
• eta_squared() now fully support multi-variate car ANOVAs (class Anova.mlm; #406 ).

effectsize 0.6.0

Breaking Changes

• pearsons_c() effect size column name changed to Pearsons_c for consistency.

New features

New API

See Support functions for model extensions vignette.

Other features

• eta_squared() family now supports afex::mixed() models.
• cles() for estimating common language effect sizes.
• rb_to_cles() for converting rank-biserial correlation to Probability of superiority.

Changes

• effectsize() for BayesFactor objects returns the same standardized output as for htest.

Bug fixes

• eta_squared() for MLM return effect sizes in the correct order of the responses.
• eta_squared() family no longer fails when CIs fail due to non-finite Fs / degrees of freedom.
• standardize() for multivariate models standardizes the (multivariate) response.
• standardize() for models with offsets standardizes offset variables according to include_response and two_sd ( #396 ).
• eta_squared(): fixed a bug that caused afex_aov models with more than 2 within-subject factors to return incorrect effect sizes for the lower level factors ( #389 ).

effectsize 0.5.0

Breaking Changes

• cramers_v() correctly does not work with 1-dimensional tables (for goodness-of-fit tests).
• interpret_d(), interpret_g(), and interpret_delta() are now interpret_cohens_d(), interpret_hedges_g(), and interpret_glass_delta().
• interpret_parameters() was removed. Use interpret_r() instead (with caution!).
• Phi, Cohen's w, Cramer's V, ANOVA effect sizes, rank Epsilon squared, Kendall's W - CIs default to 95% one-sided CIs (alternative = "greater"). (To restore previous behavior, set ci = .9, alternative = "two.sided".)
• adjust(), change_scale(), normalize(), ranktransform(), standardize() (data), and unstandardize() have moved to the new {datawizard} package!

New features

• pearsons_c() (and chisq_to_pearsons_c()) for estimating Pearson's contingency coefficient.
• interpret_vif() for interpretation of variance inflation factors.
• oddsratio_to_riskratio() can now convert OR coefficients to RR coefficients from a logistic GLM(M).
• All effect-size functions gain an alternative argument which can be used to make one- or two-sided CIs.
• interpret() now accepts as input the results from cohens_d(), eta_squared(), rank_biserial(), etc.
• interpret_pd() for the interpretation of the Probability of Direction.

Bug fixes

• kendalls_w() CIs now correctly bootstrap samples from the raw data (previously the rank-transformed data was sampled from).
• cohens_d(), sd_pooled() and rank_biserial() now properly respect when y is a grouping character vector.
• effectsize() for Chi-squared test of goodness-of-fit now correctly respects non-uniform expected probabilities ( #352 ).

Changes

• interpret_bf() now accepts log(BF) as input.

effectsize 0.4.5

New features

• eta_squared() family now indicate the type of sum-of-squares used.
• rank_biserial() estimates CIs using the normal approximation (previously used bootstrapping).
• hedges_g() now used exact bias correction (thanks to @mdelacre for the suggestion!)
• glass_delta() now estimates CIs using the NCP method based on Algina et al (2006).

Bug fixes

• eta_squared() family returns correctly returns the type 2/3 effect sizes for mixed ANOVAs fit with afex.
• cohens_d() family now correctly deals with missing factor levels ( #318 )
• cohens_d() / hedges_g() minor fix for CI with unequal variances.

Changes

• mad_pooled() (the robust version of sd_pooled()) now correctly pools the the two samples.

effectsize 0.4.4-1

New features

• standardize_parameters() + eta_squared() support tidymodels (when that the underlying model is supported; #311 ).
• cohens_d() family now supports Pairs() objects as input.
• standardize_parameters() gains the include_response argument (default to TRUE) ( #309 ).

Bug fixes

• kendalls_w() now actually returns correct effect size. Previous estimates were incorrect, and based on transposing the groups and blocks.

effectsize 0.4.4

effectsize now supports R >= 3.4.

New features

• standardize_parameters() now supports bootstrapped estimates (from parameters::bootstrap_model() and parameters::bootstrap_parameters()).
• unstandardize() which will reverse the effects of standardize().
• interpret_kendalls_w() to interpret Kendall's coefficient of concordance.
• eta_squared() family of functions can now also return effect sizes for the intercept by setting include_intercept = TRUE ( #156 ).

Bug fixes

• standardize() can now deal with dates ( #300 ).

effectsize 0.4.3

Breaking Changes

• oddsratio() and riskratio() - order of groups has been changed (the first groups is now the treatment group, and the second group is the control group), so that effect sizes are given as treatment over control (treatment / control) (previously was reversed). This is done to be consistent with other functions in R and in effectsize.

New features

• cohens_h() effect size for comparing two independent proportions.

• rank_biserial(), cliffs_delta(), rank_epsilon_squared() and kendalls_w() functions for effect sizes for rank-based tests.

• adjust() gains keep_intercept argument to keep the intercept.

• eta_squared() family of functions supports Anova.mlm objects (from the car package).

• effectsize():

  • supports Cohen's g for McNemar's test.

  • Extracts OR from Fisher's Exact Test in the 2x2 case.

• eta2_to_f2() / f2_to_eta2() to convert between two types of effect sizes for ANOVA ( #240 ).

• cohens_d() family of functions gain mu argument.

Bug fixes

• adjust() properly works when multilevel = TRUE.

• cohens_d() family / sd_pooled() now properly fails when given a missing column name.

Changes

• effectsize() for htest objects now tries first to extract the data used for testing, and computed the effect size directly on that data.

• cohens_d() family / sd_pooled() now respect any transformations (e.g. I(log(x) - 3) ~ factor(y)) in a passed formula.

• eta_squared() family of functions gains a verbose argument.

• verbose argument more strictly respected.

• glass_delta() returns CIs based on the bootstrap.

effectsize 0.4.1

Breaking Changes

• cohens_d() and glass_delta(): The correction argument has been deprecated, in favor of it being correctly implemented in hedges_g() ( #222 ).

• eta_squared_posterior() no longer uses car::Anova() by default.

New features

• effectsize() gains type = argument for specifying which effect size to return.

• eta_squared_posterior() can return a generalized Eta squared.

• oddsratio() and riskratio() functions for 2-by-2 contingency tables.

• standardize() gains support for mediation::mediate() models.

• eta_squared() family available for manova objects.

Changes

• eta_squared() family of functions returns non-partial effect size for one-way between subjects design (#180).

Bug fixes

• hedges_g() correctly implements the available bias correction methods ( #222 ).

• Fixed width of CI for Cohen's d and Hedges' g when using non-pooled SD.

effectsize 0.4.0

Breaking Changes

• standardize_parameters() for multi-component models (such as zero-inflated) now returns the unstandardized parameters in some cases where standardization is not possible (previously returned NAs).

• Column name changes:

  • eta_squared() / F_to_eta2 families of function now has the Eta2 format, where previously was Eta_Sq.

  • cramers_v is now Cramers_v

New features

• effectsize() added support for BayesFactor objects (Cohen's d, Cramer's v, and r).

• cohens_g() effect size for paired contingency tables.

• Generalized Eta Squared now available via eta_squared(generalized = ...).

• eta_squared(), omega_squared() and epsilon_squared() fully support aovlist, afex_aov and mlm (or maov) objects.

• standardize_parameters() can now return Odds ratios / IRRs (or any exponentiated parameter) by setting exponentiate = TRUE.

• Added cohens_f_squared() and F_to_f2() for Cohen's f-squared.

• cohens_f() / cohens_f_squared()can be used to estimate Cohen's f for the R-squared change between two models.

• standardize() and standardize_info() work with weighted models / data ( #82 ).

• Added hardlyworking (simulated) dataset, for use in examples.

• interpret_* ( #131 ):

  • interpret_omega_squared() added "cohen1992" rule.

  • interpret_p() added Redefine statistical significance rules.

• oddsratio_to_riskratio() for converting OR to RR.

Changes

• CIs for Omega-/Epsilon-squared and Adjusted Phi/Cramer's V return 0s instead of negative values.

• standardize() for data frames gains the remove_na argument for dealing with NAs ( #147 ).

• standardize() and standardize_info() now (and by extension, standardize_parameters()) respect the weights in weighted models when standardizing ( #82 ).

• Internal changes to standardize_parameters() (reducing co-dependency with parameters) - argument parameters has been dropped.

Bug fixes

• ranktransform(sign = TURE) correctly (doesn't) deal with zeros.

• effectsize() for htest works with Spearman and Kendall correlations ( #165 ).

• cramers_v() and phi() now work with goodness-of-fit data ( #158 )

• standardize_parameters() for post-hoc correctly standardizes transformed outcome.

• Setting two_sd = TRUE in standardize() and standardize_parameters() (correctly) on uses 2-SDs of the predictors (and not the response).

• standardize_info() / standardize_parameters(method = "posthoc") work for zero-inflated models ( #135 )

• standardize_info(include_pseudo = TRUE) / standardize_parameters(method = "pseudo") are less sensitive in detecting between-group variation of within-group variables.

• interpret_oddsratio() correctly treats extremely small odds the same as treats extremely large ones.

effectsize 0.3.3

New features

• standardize_parameters(method = "pseudo") returns pseudo-standardized coefficients for (G)LMM models.

• d_to_common_language() for common language measures of standardized differences (a-la Cohen's d).

Changes

• r_to_odds() family is now deprecated in favor of r_to_oddsratio().

• interpret_odds() is now deprecated in favor of interpret_oddsratio()

Bug fixes

• phi() and cramers_v() did not respect the CI argument ( #111 ).

• standardize() / standardize_parameters() properly deal with transformed data in the model formula ( #113 ).

• odds_to_probs() was mis-treating impossible odds (NEVER TELL ME THE ODDS! #123 )

effectsize 0.3.2

New features

• eta_squared_posterior() for estimating Eta Squared for Bayesian models.

• eta_squared(), omega_squared() and epsilon_squared() now works with

  • ols / rms models.

• effectsize() for class htest supports oneway.test(...).

Bug fixes

• Fix minor miss-calculation of Chi-squared for 2*2 table with small samples ( #102 ).

• Fixed miss-calculation of signed rank in ranktransform() ( #87 ).

• Fixed bug in standardize() for standard objects with non-standard class-attributes (like vectors of class haven_labelled or vctrs_vctr).

• Fix effectsize() for one sample t.test(...) ( #95 ; thanks to pull request by @mutlusun )

effectsize 0.3.1

New features

• standardize_parameters() now returns CIs ( #72 )

• eta_squared(), omega_squared() and epsilon_squared() now works with

  • gam models.

  • afex models.

  • lme and anova.lme objects.

• New function equivalence_test() for effect sizes.

• New plotting methods in the see package.

effectsize 0.3.0

New features

• New general purpose effectsize() function.

• Effectsize for differences have CI methods, and return a data frame.

• Effectsize for ANOVA all have CI methods, and none are based on bootstrapping.

• New effect sizes for contingency tables (phi() and cramers_v()).

• chisq_to_phi() / cramers_v() functions now support CIs (via the ncp method), and return a data frame.

• F_to_eta2() family of functions now support CIs (via the ncp method), and return a data frame.

• t_to_d() and t_to_r() now support CIs (via the ncp method), and return a data frame.

• standardize() for model-objects has a default-method, which usually accepts all models. Exception for model-objects that do not work will be added if missing.

• standardize.data.frame() gets append and suffix arguments, to add (instead of replace) standardized variables to the returned data frame.

• eta_squared(), omega_squared() and epsilon_squared() now works

  • output from parameters::model_parameters().

  • mlm models.

Bug fixes

• Fix cohens_d()'s dealing with formula input (#44).

• sd_pooled() now returns the... pooled sd (#44).

Changes

• In t_to_d(), argument pooled is now paired.

effectsize 0.2.0

Bug fixes

• standardize.data.frame() did not work when variables had missing values.

• Fixed wrong computation in standardize() when two_sd = TRUE.

• Fixed bug with missing column names in standardize_parameters() for models with different components (like count and zero-inflation).

effectsize 0.1.1

Changes

• News are hidden in an air of mystery...

effectsize 0.1.0

New features

• standardize_parameters() and standardize() now support models from packages brglm, brglm2, mixor, fixest, cgam, cplm, cglm, glmmadmb and complmrob.

Bug fixes

• Fix CRAN check issues.