---
title: "Standardized Differences"
output: 
  rmarkdown::html_vignette:
    toc: true
    fig_width: 10.08
    fig_height: 6
tags: [r, effect size, rules of thumb, guidelines, conversion]
vignette: >
  \usepackage[utf8]{inputenc}
  %\VignetteIndexEntry{Standardized Differences}
  %\VignetteEngine{knitr::rmarkdown}
editor_options: 
  chunk_output_type: console
bibliography: bibliography.bib
---

```{r setup, include=FALSE}
library(knitr)
options(knitr.kable.NA = "")
knitr::opts_chunk$set(comment = ">")
options(digits = 3)

set.seed(7)

.eval_if_requireNamespace <- function(...) {
  pkgs <- c(...)
  knitr::opts_chunk$get("eval") && all(sapply(pkgs, requireNamespace, quietly = TRUE))
}
```

This vignette provides a review of effect sizes for comparisons of groups, which
are typically achieved with the `t.test()` and `wilcox.test()` functions.

```{r}
library(effectsize)
options(es.use_symbols = TRUE) # get nice symbols when printing! (On Windows, requires R >= 4.2.0)
```

# Standardized Differences

For *t*-tests, it is common to report an effect size representing a standardized
difference between the two compared samples' means. These measures range from
$-\infty$ to $+\infty$, with negative values indicating the second group's mean
is larger (and vice versa).

## Two Independent Samples

For two independent samples, the difference between the means is standardized
based on the pooled standard deviation of both samples (assumed to be equal in
the population):

```{r}
t.test(mpg ~ am, data = mtcars, var.equal = TRUE)

cohens_d(mpg ~ am, data = mtcars)
```

Hedges' *g* provides a small-sample bias correction (for small sample sizes, $N < 20$).

```{r}
hedges_g(mpg ~ am, data = mtcars)
```

If variances cannot be assumed to be equal, it is possible to get estimates that
are not based on the pooled standard deviation:

```{r}
t.test(mpg ~ am, data = mtcars, var.equal = FALSE)

cohens_d(mpg ~ am, data = mtcars, pooled_sd = FALSE)

hedges_g(mpg ~ am, data = mtcars, pooled_sd = FALSE)
```

In cases where the differences between the variances are substantial, it is also
common to standardize the difference based only on the standard deviation of one
of the groups (usually the "control" group); this effect size is known as Glass'
$\Delta$ (delta) (Note that the standard deviation is taken from the *second* sample).

```{r}
glass_delta(mpg ~ am, data = mtcars)
```

For a one-sided hypothesis, it is also possible to construct one-sided confidence intervals:

```{r}
t.test(mpg ~ am, data = mtcars, var.equal = TRUE, alternative = "less")

cohens_d(mpg ~ am, data = mtcars, pooled_sd = TRUE, alternative = "less")
```

## One Sample

In the case of a one-sample test, the effect size represents the standardized
distance of the mean of the sample from the null value.

```{r}
t.test(mtcars$wt, mu = 2.7)

cohens_d(mtcars$wt, mu = 2.7)

hedges_g(mtcars$wt, mu = 2.7)
```

## Paired Samples

In a repeated-measures design, the same subjects are measured in multiple conditions or time points. Unlike the case of independent groups, there are multiple sources of variation that can be used to standardized the differences between the means of the conditions / times, and each provides a unique standardized mean difference.

The most basic option is compute from the paired samples difference scores and compute a one-sample effect size. This effect size, known as Cohen's $d_z$, represents the difference in terms of its homogeneity (a small but stable difference will have a large $d_z$).

```{r}
sleep_wide <- datawizard::data_to_wide(sleep,
  id_cols = "ID",
  values_from = "extra",
  names_from = "group",
  names_prefix = "extra_"
)

t.test(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]], paired = TRUE)

repeated_measures_d(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]], method = "z")

# same as:
hedges_g(sleep_wide[["extra_1"]] - sleep_wide[["extra_2"]])
```

Other options try to get close to the value that would have been reached if the samples were independant (see more info in the documentation of `repeated_measures_d()`):

```{r}
repeated_measures_d(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]])

repeated_measures_d(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]], method = "av")

repeated_measures_d(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]], method = "b")

repeated_measures_d(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]], method = "d")

# all closer to:
cohens_d(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]], ci = NULL)
```

For data containing repetition in each condition/subject, another effect size (residual _d_) is also available:

```{r}
data("rouder2016")

head(rouder2016)

repeated_measures_d(rt ~ cond | id, data = rouder2016, method = "r")
```

## For a Bayesian *t*-test

A Bayesian estimate of Cohen's *d* can also be provided based on `BayesFactor`'s
version of a *t*-test via the `effectsize()` function:

```{r, eval = .eval_if_requireNamespace("BayesFactor"), message=FALSE}
library(BayesFactor)
BFt <- ttestBF(formula = mpg ~ am, data = mtcars)

effectsize(BFt, type = "d")
```

## (Multivariate) Standardized Distances

When examining multivariate differences (e.g., with Hotelling's $T^2$ test), Mahalanobis' *D* can be used as the multivariate equivalent for Cohen's *d*. Unlike Cohen's *d* which is a measure of standardized *differences*, Mahalanobis' *D* is a measure of standardized *distances*. As such, it cannot be negative, and ranges from 0 (no distance between the multivariate distributions) to $+\infty$.

```{r}
mahalanobis_d(mpg + hp + cyl ~ am, data = mtcars)
```

## Means Ratio

Instead of the *difference* between means, we can also look at the *ratio* between means. This effect size is only applicable to **ratio scale** outcomes (variables with an absolute zero).

Lucky for us, miles-per-gallon is on a ratio scale!

```{r}
means_ratio(mpg ~ am, data = mtcars)
```

Values range between $0$ and $\infty$, with values smaller than 1 indicating that the second mean is larger than the first, values larger than 1 indicating that the second mean is smaller than the first, and values of 1 indicating that the means are equal.

# Dominance Effect Sizes

The rank-biserial correlation ($r_{rb}$) is a measure of dominance: larger
values indicate that more of *X* is larger than more of *Y*, with a value of
$(-1)$ indicates that *all* observations in the second group are larger than the
first, and a value of $(+1)$ indicates that *all* observations in the first
group are larger than the second.

These effect sizes should be reported with the Wilcoxon (Mann-Whitney) test or
the signed-rank test (both available in `wilcox.test()`).

## Two Independent Samples

```{r, warning=FALSE}
A <- c(48, 48, 77, 86, 85, 85)
B <- c(14, 34, 34, 77)

wilcox.test(A, B, exact = FALSE) # aka Mann–Whitney U test

rank_biserial(A, B)
```

## One Sample

For one sample, $r_{rb}$ measures the symmetry around $\mu$ (mu; the null
value), with 0 indicating perfect symmetry, $(-1)$ indicates that all
observations fall below $\mu$, and $(+1)$ indicates that all observations fall
above $\mu$.

```{r}
x <- c(1.15, 0.88, 0.90, 0.74, 1.21, 1.36, 0.89)

wilcox.test(x, mu = 1) # aka Signed-Rank test

rank_biserial(x, mu = 1)
```

## Paired Samples

For paired samples, $r_{rb}$ measures the symmetry of the (paired) *differences*
around $\mu$ as for the one sample case.

```{r}
x <- c(1.83, 0.50, 1.62, 2.48, 1.68, 1.88, 1.55, 3.06, 1.30)
y <- c(0.88, 0.65, 0.60, 2.05, 1.06, 1.29, 1.06, 3.14, 1.29)

wilcox.test(x, y, paired = TRUE) # aka Signed-Rank test

rank_biserial(x, y, paired = TRUE)
```


# Common Language Effect Sizes

Related effect sizes are the *common language effect sizes* which present information about group differences in terms of probability.

## Two Independent Samples

### Measures of (Non)Overlap

These measures indicate the degree two independent distributions overlap: Cohen's $U_1$ is the proportion of the total of both distributions that does not overlap, while *Overlap (OVL)* is the proportional overlap between the distributions.

```{r}
cohens_u1(mpg ~ am, data = mtcars)

p_overlap(mpg ~ am, data = mtcars)
```

Note the by default, these functions return the parametric versions of these effect sizes: these assume equal normal variance in both populations. When these assumptions are not met, the values produced will be biased in unknown ways. In such cases, we should use the non-parametric versions ($U_1$ is not defined):

```{r}
p_overlap(mpg ~ am, data = mtcars, parametric = FALSE)
```

### Probabilistic Measures

*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.  

```{r}
p_superiority(mpg ~ am, data = mtcars)
```

Here, this indicates that if we were to randomly draw a sample from `am==0` and from `am==1`, 15% of the time, the first will have a larger `mpg` values than the second.

Cohen's $U_2$ is the proportion of one of the groups that exceeds the same proportion in the other group, and Cohen's $U_3$ is the proportion of the second group that is smaller than the median of the first group.

```{r}
cohens_u2(mpg ~ am, data = mtcars)

cohens_u3(mpg ~ am, data = mtcars)
```

<!-- ```{r, eval = requireNamespace("ggplot2"), echo=FALSE} -->
<!-- mu <- tapply(mtcars$mpg, mtcars$am, mean) -->
<!-- sigma <- sd_pooled(mtcars$mpg, factor(mtcars$am)) -->

<!-- xlim <- sort(mu) + c(-4, 4) * sigma -->

<!-- seg_u2 <- data.frame(x = mean(mu), -->
<!--                      xend = xlim, -->
<!--                      y = -exp(-5), -->
<!--                      yend = -exp(-5), -->
<!--                      color = c("0", "1")) -->

<!-- ggplot2::ggplot() +  -->
<!--   ggplot2::stat_function(ggplot2::aes(fill = "0"), size = 1, geom = "area", alpha = 0.6, -->
<!--                 fun = dnorm, args = list(mean = mu[1], sd = sigma), xlim = xlim) +  -->
<!--   ggplot2::stat_function(ggplot2::aes(fill = "1"), size = 1, geom = "area", alpha = 0.6, -->
<!--                 fun = dnorm, args = list(mean = mu[2], sd = sigma), xlim = xlim) + -->
<!--   # U3 -->
<!--   ggplot2::stat_function(ggplot2::aes(fill = "1"), size = 1, geom = "area",  -->
<!--                          color = "black", outline.type = "full",  -->
<!--                 fun = dnorm, args = list(mean = mu[2], sd = sigma), xlim = c(xlim[1], mu[1])) + -->
<!--   ggplot2::geom_vline(ggplot2::aes(xintercept = mu[1]), linetype = "dashed") +  -->
<!--   ggplot2::annotate("text", x = mu[1]-2, y = -seg_u2$y, label = "U3") +  -->
<!--   # U2 -->
<!--   ggplot2::geom_vline(ggplot2::aes(xintercept = mean(mu))) +  -->
<!--   ggplot2::geom_segment(ggplot2::aes(x = x, xend = xend, y = y, yend = yend, color = color), data = seg_u2,  -->
<!--                         arrow = ggplot2::arrow(), -->
<!--                         size = 0.4) +  -->
<!--   see::scale_color_material(aesthetics = c("color", "fill"), name = "am") +  -->
<!--   ggplot2::annotate("text", x = mu[2], y = 1.4*seg_u2$y, label = "U2 for am=1", hjust = "left") +  -->
<!--   ggplot2::annotate("text", x = mu[1], y = 1.4*seg_u2$y, label = "U2 for am=0", hjust = "right") +  -->
<!--   ggplot2::theme_bw() -->

<!-- ``` -->

Here too we have a non-parametric versions when the assumptions of equal variance of normal populations:

```{r}
p_superiority(mpg ~ am, data = mtcars, parametric = FALSE)

cohens_u2(mpg ~ am, data = mtcars, parametric = FALSE)

cohens_u3(mpg ~ am, data = mtcars, parametric = FALSE)
```

## One Sample and Paired Samples

For one sample, *probability of superiority* is the probability that, when sampling an observation at random, it will be larger than $\mu$.

```{r}
p_superiority(mtcars$wt, mu = 2.75)

p_superiority(mtcars$wt, mu = 2.75, parametric = FALSE)
```

For paired samples, *probability of superiority* is the probability that, when sampling an observation at random, its *difference* will be larger than $\mu$.

```{r}
p_superiority(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]],
  paired = TRUE, mu = -1
)

p_superiority(sleep_wide[["extra_1"]], sleep_wide[["extra_2"]],
  paired = TRUE, mu = -1,
  parametric = FALSE
)
```

## For a Bayesian *t*-test

A Bayesian estimate of (the parametric version of) these effect sizes can also
be provided based on `BayesFactor`'s version of a *t*-test via the
`effectsize()` function:

```{r, eval = .eval_if_requireNamespace("BayesFactor")}
effectsize(BFt, type = "p_superiority")

effectsize(BFt, type = "u1")

effectsize(BFt, type = "u2")

effectsize(BFt, type = "u3")

effectsize(BFt, type = "overlap")
```

# References