Cohen's \(U_1\), \(U_2\), and \(U_3\), probability of superiority,
proportion of overlap, Wilcoxon-Mann-Whitney odds, and Vargha and Delaney's
A are CLESs. These are effect sizes that represent differences between two
(independent) distributions in probabilistic terms (See details). Pair with
any reported stats::t.test() or stats::wilcox.test().
p_superiority(
x,
y = NULL,
data = NULL,
mu = 0,
paired = FALSE,
parametric = TRUE,
ci = 0.95,
alternative = "two.sided",
verbose = TRUE,
...
)cohens_u1(
x,
y = NULL,
data = NULL,
mu = 0,
parametric = TRUE,
ci = 0.95,
alternative = "two.sided",
iterations = 200,
verbose = TRUE,
...
)
cohens_u2(
x,
y = NULL,
data = NULL,
mu = 0,
parametric = TRUE,
ci = 0.95,
alternative = "two.sided",
iterations = 200,
verbose = TRUE,
...
)
cohens_u3(
x,
y = NULL,
data = NULL,
mu = 0,
parametric = TRUE,
ci = 0.95,
alternative = "two.sided",
iterations = 200,
verbose = TRUE,
...
)
p_overlap(
x,
y = NULL,
data = NULL,
mu = 0,
parametric = TRUE,
ci = 0.95,
alternative = "two.sided",
iterations = 200,
verbose = TRUE,
...
)
vd_a(
x,
y = NULL,
data = NULL,
mu = 0,
ci = 0.95,
alternative = "two.sided",
verbose = TRUE,
...
)
wmw_odds(
x,
y = NULL,
data = NULL,
mu = 0,
paired = FALSE,
ci = 0.95,
alternative = "two.sided",
verbose = TRUE,
...
)
A data frame containing the common language effect sizes (and optionally their CIs).
A numeric vector, or a character name of one in data.
Any missing values (NAs) are dropped from the resulting vector.
x can also be a formula (see stats::t.test()), in which case y is
ignored.
An optional data frame containing the variables.
a number indicating the true value of the mean (or difference in means if you are performing a two sample test).
If TRUE, the values of x and y are considered as paired.
This produces an effect size that is equivalent to the one-sample effect
size on x - y.
Use parametric estimation (see cohens_d()) or
non-parametric estimation (see rank_biserial()). See details.
Confidence Interval (CI) level
a character string specifying the alternative hypothesis;
Controls the type of CI returned: "two.sided" (default, two-sided CI),
"greater" or "less" (one-sided CI). Partial matching is allowed (e.g.,
"g", "l", "two"...). See One-Sided CIs in effectsize_CIs.
Toggle warnings and messages on or off.
Arguments passed to or from other methods. When x is a formula,
these can be subset and na.action.
The number of bootstrap replicates for computing confidence
intervals. Only applies when ci is not NULL and parametric = FALSE.
For parametric CLES, the CIs are transformed CIs for Cohen's d (see
d_to_u3()). For non-parametric (parametric = FALSE) CLES, the CI of
Pr(superiority) is a transformed CI of the rank-biserial correlation
(rb_to_p_superiority()), while for all others, confidence intervals are
estimated using the bootstrap method (using the {boot} package).
"Confidence intervals on measures of effect size convey all the information
in a hypothesis test, and more." (Steiger, 2004). Confidence (compatibility)
intervals and p values are complementary summaries of parameter uncertainty
given the observed data. A dichotomous hypothesis test could be performed
with either a CI or a p value. The 100 (1 - \(\alpha\))% confidence
interval contains all of the parameter values for which p > \(\alpha\)
for the current data and model. For example, a 95% confidence interval
contains all of the values for which p > .05.
Note that a confidence interval including 0 does not indicate that the null
(no effect) is true. Rather, it suggests that the observed data together with
the model and its assumptions combined do not provided clear evidence against
a parameter value of 0 (same as with any other value in the interval), with
the level of this evidence defined by the chosen \(\alpha\) level (Rafi &
Greenland, 2020; Schweder & Hjort, 2016; Xie & Singh, 2013). To infer no
effect, additional judgments about what parameter values are "close enough"
to 0 to be negligible are needed ("equivalence testing"; Bauer & Kiesser,
1996).
Some effect sizes are directionless--they do have a minimum value that would
be interpreted as "no effect", but they cannot cross it. For example, a null
value of Kendall's W is 0, indicating no difference between
groups, but it can never have a negative value. Same goes for
U2 and Overlap: the null value of \(U_2\) is
0.5, but it can never be smaller than 0.5; am Overlap of 1 means "full
overlap" (no difference), but it cannot be larger than 1.
When bootstrapping CIs for such effect sizes, the bounds of the CIs will
never cross (and often will never cover) the null. Therefore, these CIs
should not be used for statistical inference.
These measures of effect size present group differences in probabilistic terms:
Probability of superiority is the probability that, when sampling an observation from each of the groups at random, that the observation from the second group will be larger than the sample from the first group. For the one-sample (or paired) case, it is the probability that the sample (or difference) is larger than mu. (Vargha and Delaney's A is an alias for the non-parametric probability of superiority.)
Cohen's \(U_1\) is the proportion of the total of both distributions that does not overlap.
Cohen's \(U_2\) is the proportion of one of the groups that exceeds the same proportion in the other group.
Cohen's \(U_3\) is the proportion of the second group that is smaller than the median of the first group.
Overlap (OVL) is the proportional overlap between the distributions.
(When parametric = FALSE, bayestestR::overlap() is used.)
Wilcoxon-Mann-Whitney odds are the odds of
non-parametric superiority (via probs_to_odds()), that is the odds that,
when sampling an observation from each of the groups at random, that the
observation from the second group will be larger than the sample from the
first group.
Where \(U_1\), \(U_2\), and Overlap are agnostic to the direction of the difference between the groups, \(U_3\) and probability of superiority are not.
The parametric version of these effects assumes normality of both populations and homoscedasticity. If those are not met, the non parametric versions should be used.
Cohen, J. (1977). Statistical power analysis for the behavioral sciences. New York: Routledge.
Reiser, B., & Faraggi, D. (1999). Confidence intervals for the overlapping coefficient: the normal equal variance case. Journal of the Royal Statistical Society, 48(3), 413-418.
Ruscio, J. (2008). A probability-based measure of effect size: robustness to base rates and other factors. Psychological methods, 13(1), 19–30.
Vargha, A., & Delaney, H. D. (2000). A critique and improvement of the CL common language effect size statistics of McGraw and Wong. Journal of Educational and Behavioral Statistics, 25(2), 101-132.
O’Brien, R. G., & Castelloe, J. (2006, March). Exploiting the link between the Wilcoxon-Mann-Whitney test and a simple odds statistic. In Proceedings of the Thirty-first Annual SAS Users Group International Conference (pp. 209-31). Cary, NC: SAS Institute.
Agresti, A. (1980). Generalized odds ratios for ordinal data. Biometrics, 59-67.
sd_pooled()
Other standardized differences:
cohens_d(),
mahalanobis_d(),
means_ratio(),
rank_biserial()
Other rank-based effect sizes:
rank_biserial(),
rank_epsilon_squared()
cohens_u2(mpg ~ am, data = mtcars)
p_superiority(mpg ~ am, data = mtcars, parametric = FALSE)
wmw_odds(mpg ~ am, data = mtcars)
x <- c(1.83, 0.5, 1.62, 2.48, 1.68, 1.88, 1.55, 3.06, 1.3)
y <- c(0.878, 0.647, 0.598, 2.05, 1.06, 1.29, 1.06, 3.14, 1.29)
p_overlap(x, y)
p_overlap(y, x) # direction of effect does not matter
cohens_u3(x, y)
cohens_u3(y, x) # direction of effect does matter
Run the code above in your browser using DataLab