Originally Posted by MontanaMan
Originally Posted by Jordan Smith
[
Where is anyone here saying you need single-digit ES? Single-digit SD, maybe, but SD is very different from ES. In a Gaussian distribution, SD is approximately 1/6th of ES.

Actually, the estimate of SD is ES/4. And that will be "rough".

Then to get a realistic estimate of the real world variance, you should use at least +/- 3 SD, or to be even more certain, +/- 6 SD.

MM
That entirely depends on your confidence requirement. Assuming a meaningful sample size, ES is well-described by a range of 6*SD (+/- 3 from the mean) that captures approximately 99.7% of the population. Using 4*SD captures 95% of the population, which is still a fairly good estimate of ES.