Effect size is an easy way of quantifying the difference between two groups which is more advantageous than only using tests of statistical significance. It seeks to reiterate the size of difference rather than confounding this with sample size. Researchers cannot but report on the effect sizes in their results for quantitative studies. However, my readings on the subject with reference to Cohen’s (1988) classifications provide contradictory materials. I need help to satisfy curiosity
For r from Pearson correlation, Cohen (1988) gives the following interpretation:
small, 0.10 – < 0.30
medium, 0.30 – < 0.50
large, ≥ 0.50
But it can't be stressed enough that the interpretation varies by the discipline and expectations of the experimenter. You can't blindly follow Cohen. There is no substitute for what you subjectively determine is "practically important".
_______________________
Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd Edition. Routledge
"Correlation and Linear Regression" in Mangiafico, S.S. 2016. Summary and Analysis of Extension Program Evaluation in R, version 1.13.6. http://rcompanion.org/handbook/I_10.html .
Cohen discusses several different effect size statistics, which are used in different situations. Each one has a different interpretation. It's helpful to think of these statistics as related to statistical tests. For example, Cohen's d is an effect size statistic used with two-sample t-tests. r is used with correlation. phi is used with 2 x 2 chi-square test of association.
Which are you thinking about?
As Salvatore said, there are different measures of effect size depending on what statistical test you are using.
Since you mention difference between 2 groups, my guess would be that you are talking about Cohen’s d.
According to Cohen (1988), 0.2 is considered small effect, 0.5 medium and 0.8 large. Reference is from Cohen’s book, Statistical Power Analysis for the Behavioral Sciences (2nd ed).
However as with any rule of thumb caution should be exercised. What is considered large effect size in Biomedicine could be considered small effect in Sociology, so actual cut-off values may differ from field to field.
Hope this helps.
Take care,
Ivan
In fact Cohen based his guidelines on visual differences in the height of growing girls, so I agree with the above caution!
@Adrian --- just to be clear for readers of this thread --- Cohen does use that as an example in one case, but, no, that's not how he came up with interpretations for his effect size statistics.
I thank you all your insightful submissions especially Salvatore. I am interested in the r for correlation.
For r from Pearson correlation, Cohen (1988) gives the following interpretation:
small, 0.10 – < 0.30
medium, 0.30 – < 0.50
large, ≥ 0.50
But it can't be stressed enough that the interpretation varies by the discipline and expectations of the experimenter. You can't blindly follow Cohen. There is no substitute for what you subjectively determine is "practically important".
_______________________
Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd Edition. Routledge
"Correlation and Linear Regression" in Mangiafico, S.S. 2016. Summary and Analysis of Extension Program Evaluation in R, version 1.13.6. http://rcompanion.org/handbook/I_10.html .
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size