CI for Effect Sizes d* and g*

Objective

We describe how to estimate confidence intervals for Cohen’s d* and Hedges’ g*, effect sizes for two independent samples where the variances are not assumed to be equal. These effect sizes are described in Two Sample t-test with Unequal Variances. We start by reviewing these effect sizes.

Cohen’s d* and Hedges’ g*

Cohen’s d* is defined by

Cohen's d*

where

s* for Cohen's d*

Hedges’ g* is a less biased version of d*, and is defined by

Hedges' g*

where m = df*/2 anddf* for g*

Confidence Interval

We estimate confidence intervals for d* and g* using the same approach as for d and g (see CI Functions for Effect Sizes d and g). In particular, first we estimate the noncentrality parameter (ncp*) by

ncp*

Next, we find the confidence interval for ncp* by using the Real Statistics NT_NCP function. The end points of this CI are =NT_NCP(α/2, df*, ncp*) and =NT_NCP(1-α/2, df*, ncp*).

We then use this confidence interval to find a CI for d*. This is done by noting that

Relationship: ncp* and d*

Thus, once we have a CI for ncp* we can multiply the end points of this CI by the factor on the right side of the above expression to obtain a CI for d*. Finally, we multiply the end points of the d* CI by the usual factor (involving the gamma function) to obtain a CI for g*.

Example

Example 1: Find the 95% confidence interval of the effect size for Example 2 of Two-Sample t-Test with Unequal Variances.

We show the results of the t-test on the left side of Figure 1 to obtain the needed descriptive statistics. The results on the right side of the figure shows how to estimate the desired confidence intervals.

Confidence intervals

Figure 1 – 95% confidence interval for d* and g* effect sizes

Worksheet Functions

Real Statistics Functions: The Real Statistics Resource Pack provides the following array functions.

T_EFFECT3(m1, m2, s1s2, n1, n2, lab, alpha, iter) = column array with the values Cohen’s d*, Hedges’ g*, and the lower and upper confidence interval limits for d* and g* based on a two independent sample t-test with unequal variances for sample 1 with mean m1, standard deviation s1  and sample size n1, and sample 2 with mean m2, standard deviation s2 and sample size n2.

TT_EFFECT3(R1, R2, lab, alpha, iter, iter0, prec) = T_EFFECT3(m1, m2, s1s2, n1, n2, lab, alpha, iter, iter0, prec) where m1 = AVERAGE(R1), s1 = STEV.S(R1), n1 = COUNT(R1), m2 = AVERAGE(R2), s2 = STEV.S(R2) and n2 = COUNT(R2).

alpha is the significance level (default .05). If lab = TRUE (default FALSE) then an extra column of labels is appended to the output.

The last three arguments are as for the NT_NCP function (see Noncentral t Distribution), except that if iter = 0 then the Hedges and Olkin estimate of the confidence interval is employed (iter0 and prec are not used), while if iter > 0 (default 1000) then the estimate of the confidence interval described above based on the noncentrality parameter is used. In this case, iter, iter0, and prec are as for the NT_NCP function (see Noncentral t Distribution).

Example

Referring to the data on the left side of Figure 1, we see that the array formula

=T_EFFECT3(B4,C4,SQRT(B5),SQRT(C5),B6,C6,TRUE)

produces the output shown in Figure 2, which agrees with the results shown on the right side of Figure 1.

T_EFFECT3 function example

Figure 2 – T_EFFECT3 output

Examples Workbook

Click here to download the Excel workbook with the examples described on this webpage.

References

Delacre, M., Lakens, D., Ley, C., Liu, L., Leys, C. (2021) Why Hedges’ g*s based on the non-pooled standard deviation should be reported with Welch’s t-test
https://psyarxiv.com/tu6mp/download

Howell (2010) Confidence intervals on effect size
https://www.uvm.edu/~statdhtx/methods8/Supplements/MISC/Confidence%20Intervals%20on%20Effect%20Size.pdf

Lecoutre B., (2007) Another look at the confidence intervals for the noncentral t distribution
https://digitalcommons.wayne.edu/jmasm/vol6/iss1/11/

Steiger, J. H., Fouladi, R. T. (1997) Noncentrality interval estimation and the evaluation of statistical models
https://statpower.net/Steiger%20Biblio/Steiger&Fouladi97.PDF

Hedges, L. V. and Olkin, I. (1985) Statistical methods for meta-analysis. Academic Press
https://www.researchgate.net/publication/216811655_Statistical_Methods_in_Meta-Analysis

Leave a Comment