In his authoritative Statistical Power Analysis for the Behavioral Sciences, Cohen (1988) outlined a number of criteria for gauging small, medium, and large effect sizes estimated using different statistical procedures.Von Paul D. Ellis im Buch The Essential Guide to Effect Sizes (2010) im Text Interpreting effects
There is no wisdom whatsoever in attempting to associate regions of the effect size metric with descriptive adjectives such as “small,” “moderate,” “large,” and the like. Dissociated from a context
of decision and comparative value, there is little inherent value to an effect size of 3.5 or .2. Depending on what benefits can be achieved at what cost, an effect size of 2.0 might be “poor” and one of .1 might be “good.”Von Gene V. Glass, Barry McGaw, Mary Lee Smith im Buch Meta-analysis in social research (1981) auf Seite 104
Von Paul D. Ellis im Buch The Essential Guide to Effect Sizes (2010) im Text Interpreting effects auf Seite 41
Cohen’s effect size classes have two selling points. First, they are easy to grasp. You just compare your numbers with his thresholds to get a ready-made interpretation of your result. Second, although they are arbitrary, they are sufficiently grounded in logic for Cohen to hope that his cut-offs “will be found to be reasonable by reasonable people" (1988: 13). In deciding the boundaries for the three size classes, Cohen began by defining a medium effect as one that is “visible to the naked eye of the careful observer" (Cohen 1992: 156). To use his example, a medium effect is equivalent to the difference in height between fourteen- and eighteen-year-old girls, which is about one inch. He then defined a small effect as one that is less than a medium effect, but greater than a trivial effect. Small effects are equivalent to the height difference between fifteen- and sixteen-year-old girls, which is about half an inch. Finally, a large effect was defined as one that was as far above a medium effect as a small one was below it. In this case, a large effect is equivalent to the height difference between thirteen- and eighteen-year-old girls, which is just over an inch and a half.
Effect sizes seen in the social sciences are oftentimes very small
(Rosnow & Rosenthal, 2003). This has led to difficulties in their
interpretation. There is no agreement on what magnitude of effect
is necessary to establish practical significance. Cohen (1992) of fers the value of r .1, as a cut-off for “small” effects (which
would indicate only a 1% overlap in variance between two variables).
However, Cohen did not anchor his recommendations
across effect sizes; as such, his recommendations for r and d
ultimately differ in magnitude when translated from one to another.
For instance, Cohen suggests that r .3 and d .5 each
indicate a cut-off for moderate effects, yet r .3 is not the
equivalent of d .5. Other scholars suggest a minimum of r .2
(Franzblau, 1958; Lipsey, 1998) or .3 (Hinkle, Weirsma, & Jurs,
1988). In the current article, all effect size recommendations,
where possible, are anchored to a minimum of r .2, for practical
significance (Franzblau, 1958; Lipsey, 1998). These readily convert
from r to d for instance, without altering the interpretation.
Note that this is a suggested minimum not a guarantee that observed
effect sizes larger than r .2 are practically significant.
Such cut-offs are merely guidelines, and should not be applied
rigidly (Cohen, 1992; Snyder & Lawson, 1993; Thompson, 2002).Von Christopher J. Ferguson im Text An Effect Size Primer (2009)