:dragon: Compute and work with indices of effect size and standardized parameters
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).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.mad_pooled() (the robust version of sd_pooled()) now correctly pools the the two samples.standardize_parameters() + eta_sqaured() 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 ).kendalls_w() now actually returns correct effect size. Previous estimates were incorrect, and based on transposing the groups and blocks.effectsize now supports R >= 3.4.
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 ).standardize() can now deal with dates ( #300 ).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.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():
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.adjust() properly works when multilevel = TRUE.cohens_d() family / sd_pooled() now properly fails when given a missing column name.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.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.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.eta_squared() family of functions returns non-partial effect size for one-way between subjects design (#180).hedges_g() correctly implements the available bias correction methods ( #222 ).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).eta_squared() / F_to_eta2 families of function now has the Eta2 format, where previously was Eta_Sq.cramers_v is now Cramers_v
effectsize() added support for BayesFactor objects (Cohen's d, Cramer's v, and r).cohens_g() effect size for paired contingency tables.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.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 ).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.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 ).standardize_parameters() (reducing co-dependency with parameters) - argument parameters has been dropped.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.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.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).r_to_odds() family is now deprecated in favor of r_to_oddsratio().interpret_odds() is now deprecated in favor of interpret_oddsratio()
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 )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(...).ranktransform() ( #87 ).standardize() for standard objects with non-standard class-attributes (like vectors of class haven_labelled or vctrs_vctr).effectsize() for one sample t.test(...) ( #95 ; thanks to pull request by @mutlusun )