:dragon: Compute and work with indices of effect size and standardized parameters
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).fei() gives a more informative error method for invalid table inputs (#566).convert_*() aliases are deprecated.*_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".riskratio() returns correct CIs (#584)d_to_r() correctly treats specifying only n1/n2 as equal group sizes (#571)mahalanobis_d() now defaults to one-sided CIs.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.cohens_w() for 2-by-X tables.rank_biserial() ( #476 )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} now requires R >= 3.6
fei(), cohens_w() and pearsons_c() always rescale the p input to sum-to-1.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.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.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().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.ordered()) outcomes.rank_eta_squared() for one-way rank ANOVA.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.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.p that is a table.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.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()).insight's display, print_md and print_html for all {effectsize} outputs.kendalls_w() now deals with ties.eta_squared() works with car::Manova() that does not have an i-design.pearsons_c() effect size column name changed to Pearsons_c for consistency.See Support functions for model extensions vignette.
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.effectsize() for BayesFactor objects returns the same standardized output as for htest.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 ).cramers_v() correctly does not work with 1-dimentional 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!).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!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).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.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 ).interpret_bf() now accepts log(BF) as input.