Someone puts Cohen's d in a nutshell, please

[gianmarco]

Hi,

I came across Cohen's d statistic and I would very appreciate if anyone could provide me with a brief explanation of it and with an example rooted in a context of use.

I understand that it is equal to the difference between two sample means divided by the pooled SD. But, I am wondering what kind of information is Cohen's d actually conveying. Beside, I understand that it can be provided along with t-test. Am I right?

Any help is appreciated.

Gm

[Lazar]

Hi,

I came across Cohen's d statistic and I would very appreciate if anyone could provide me with a brief explanation of it and with an example rooted in a context of use.

I understand that it is equal to the difference between two sample means divided by the pooled SD. But, I am wondering what kind of information is Cohen's d actually conveying. Beside, I understand that it can be provided along with t-test. Am I right?

Any help is appreciated.

Gm

Cohen's d is a measure of effect size. Simply put it indicates the amount of different between two groups on a construct of interest in standard deviation units. It is given for two reasons:
1. It is used as a counter-point to significance tests were it gives an indication of how big or small a significant difference is. This difference can then be compared to Cohen's estimates of what is typical of a small, medium, or large effect.
2. To provide a common metric on which to compare effects for meta-analysis or what not when outcome variables may be measured on different scales.

As an example suppose we found a significant difference in self-esteem in a sample tested before an intervention and then again afterwards. A Cohen's d of .50 would suggest that the intervention program was associated with a half of one standard deviation increase in self-esteem.

and yes it can be used with t-tests.

[gianmarco]

Hi Lazar,
thanks for your reply and for your explanation.

Can Cohen'd be used for un-paired t-test, or it only works with paired t-test?

Beside, I guess it cannot be put at work in case of un-equal variance (i.e., Welch't-test) ? right?

Thanks
Gm

[Lazar]

Sorry I should have mentioned that for paired samples one should use a correction for the correlation between the variables over time (see Morris and DeShon - from memory it is equation 6 or 8).

For more info on how to use it in a range of circumstances see the paper above or Cohen's book.

While I am at it this presentation on multilevel meta-analysis might be of some use for calculating d for a range of different things (see the powerpoint slides for detail).

[gianmarco]

Hi Lazar,
thanks again for your detailed reply and for the documentation provided.

I will take a look at all that material.


Thank you.
Regards,
Gm