These functions use some conversion to and from the t distribution to provide the Cohen's d distribution. There are four versions that act similar to the standard distribution functions (the d., p., q., and r. functions, and their longer aliases .Cohensd), three convenience functions (pdExtreme, pdMild, and pdInterval), a function to compute the confidence interval for a Cohen's d estimate cohensdCI, and a function to compute the sample size required to obtain a confidence interval around a Cohen's d estimate with a specified accuracy (pwr.cohensdCI and its alias pwr.confIntd).

cohensdCI(d, n, conf.level = 0.95, plot = FALSE, silent = TRUE)

dCohensd(
x,
df = NULL,
populationD = 0,
n = NULL,
n1 = NULL,
n2 = NULL,
silent = FALSE
)

pCohensd(q, df, populationD = 0, lower.tail = TRUE)

qCohensd(p, df, populationD = 0, lower.tail = TRUE)

rCohensd(n, df, populationD = 0)

pdInterval(ds, n, populationD = 0)

pdExtreme(d, n, populationD = 0)

pdMild(d, n, populationD = 0)

pwr.cohensdCI(d, w = 0.1, conf.level = 0.95, extensive = FALSE, silent = TRUE)

## Arguments

n, n1, n2

Desired number of Cohen's d values for rCohensd and rd (n), and the number of participants/datapoints in total (n) or in each group (n1 and n2) for dd, dCohensd, pdExtreme, pdMild, pdInterval, and cohensdCI.

conf.level

The level of confidence of the confidence interval.

plot

Whether to show a plot of the sampling distribution of Cohen's d and the confidence interval. This can only be used if specifying one value for d, n, and conf.level.

silent

Whether to provide FALSE or suppress (TRUE) warnings. This is useful because function 'qt', which is used under the hood (see qt() for more information), warns that 'full precision may not have been achieved' when the density of the distribution is very close to zero. This is normally no cause for concern, because with sample sizes this big, small deviations have little impact.

x, q, d

Vector of quantiles, or, in other words, the value(s) of Cohen's d.

df

Degrees of freedom.

populationD

The value of Cohen's d in the population; this determines the center of the Cohen's d distribution. I suppose this is the noncentrality parameter.

lower.tail

logical; if TRUE (default), probabilities are the likelihood of finding a Cohen's d smaller than the specified value; otherwise, the likelihood of finding a Cohen's d larger than the specified value.

p

Vector of probabilites (p-values).

ds

A vector with two Cohen's d values.

w

The desired maximum 'half-width' or margin of error of the confidence interval.

extensive

Whether to only return the required sample size, or more extensive results.

## Value

dCohensd (or dd) gives the density, pCohensd

(or pd) gives the distribution function, qCohensd (or qd) gives the quantile function, and rCohensd (or rd) generates random deviates.

pdExtreme returns the probability (or probabilities) of finding a Cohen's d equal to or more extreme than the specified value(s).

pdMild returns the probability (or probabilities) of finding a Cohen's d equal to or less extreme than the specified value(s).

pdInterval returns the probability of finding a Cohen's d that lies in between the two specified values of Cohen's d.

cohensdCI provides the confidence interval(s) for a given Cohen's d value.

pwr.cohensdCI provides the sample size required to obtain a confidence interval for Cohen's d with a desired width.

## Details

The functions use convert.d.to.t() and convert.t.to.d() to provide the Cohen's d distribution.

The confidence interval functions, cohensdCI and pwr.cohensdCI, now use the same method as MBESS (a slightly adapted version of the MBESS function conf.limits.nct is used).

More details about cohensdCI and pwr.cohensdCI are provided in Peters & Crutzen (2017).

convert.d.to.t(), convert.t.to.d(), dt(), pt(), qt(), rt()

## Author

Gjalt-Jorn Peters (Open University of the Netherlands), with the exported MBESS function conf.limits.nct written by Ken Kelley (University of Notre Dame), and with an error noticed by Guy Prochilo (University of Melbourne).

Maintainer: Gjalt-Jorn Peters gjalt-jorn@userfriendlyscience.com

## Examples


### Confidence interval for Cohen's d of .5
### from a sample of 200 participants, also
### showing this visually: this clearly shows
### how wildly our Cohen's d value can vary
### from sample to sample.
cohensdCI(.5, n=200, plot=TRUE);
#>                     lo        hi
#> d=0.5, n=200 0.2178714 0.7809034
#> attr(,"plot") #> attr(,"class")
#>  "cohensdCI"

### How many participants would we need if we
### would want a more accurate estimate, say
### with a maximum confidence interval width
### of .2?
pwr.cohensdCI(.5, w=.1);
#>  1585

### Show that 'sampling distribution':
cohensdCI(.5,
n=pwr.cohensdCI(.5, w=.1),
plot=TRUE);
#>                      lo        hi
#> d=0.5, n=1585 0.3999342 0.5999124
#> attr(,"plot") #> attr(,"class")
#>  "cohensdCI"

### Generate 10 random Cohen's d values
rCohensd(10, 20, populationD = .5);
#>    1.26654753 -0.53642378  0.93488078  0.03580902  0.43708375  0.72018440
#>    0.52786352 -0.33929130  0.38615303  0.60876615

### Probability of findings a Cohen's d smaller than
### .5 if it's 0 in the population (i.e. under the
### null hypothesis)
pCohensd(.5, 64);
#>  0.9767939

### Probability of findings a Cohen's d larger than
### .5 if it's 0 in the population (i.e. under the
### null hypothesis)
1 - pCohensd(.5, 64);
#>  0.02320611

### Probability of findings a Cohen's d more extreme
### than .5 if it's 0 in the population (i.e. under
### the null hypothesis)
pdExtreme(.5, 64);
#>  0.04988584

### Probability of findings a Cohen's d more extreme
### than .5 if it's 0.2 in the population.
pdExtreme(.5, 64, populationD = .2);
#>  0.2406812