Cohens d effect sizes

Question

I'm using Cohens d effect sizes as a means of providing a magnitude based inference in a study I'm conducting.

I just wanted to clarify if I'm correct here: Anything between .0 and .2 is small. Anything between .2 and .8 is moderate Anything above .8 is large

If anybody could let me know if this is correct that would be much appreciated.

score 4 · Answer 1 · answered Jul 31 '15 at 18:58

4

I am probably in the minority (or maybe the only one) who thinks that giving qualifiers such as "small", "medium", and "large" without any consideration for the subject matter is a bit silly. (That 0.5 should mean the same thing for wildlife biology, hydrology, and quantum physics just doesn't make sense to me.)

If for your subject matter such qualifiers have a strong consensus, then go for it.

answered Jul 31 '15 at 18:58

JimB

2,043
8
14

Actually, Cohen himself would strongly agree with you. See: http://stats.stackexchange.com/a/87968/1934 – Wolfgang Jul 31 '15 at 20:58
That's good to hear. If we only had to use the above guidelines and adhere to 0.05 statistical significance, then we wouldn't need subject matter experts or even need to think. Life would be much simpler. – JimB Jul 31 '15 at 21:22

score 3 · Answer 2 · answered Jul 31 '15 at 18:59

Those values are largely arbitrary. I believe Cohen surveyed some published studies in his field and found that the mean $d$ was $.5$ with a standard deviation of $.3$. However, the means and SDs will vary by field. Moreover, they aren't necessarily the same thing as importance for anything. That is, people will naturally assume that a large effect is important and a small effect is not, but you will get a very large effect by comparing the heights of adults and 3 year olds, without having made an important discovery. In general, I think the only person who can say what is "large" in your research is you.

score 1 · Answer 3 · answered Jul 31 '15 at 20:22

In the lines of the answers by @gung and @Jim Baldwin, the paper Translating the Statistical Representation of the Effects of Education Interventions into More Readily Interpretable Forms (full text available) states that effect sizes of 0.2 and 0.3 are very significant and important in Education.

Cohens d effect sizes

3 Answers3