I am aware of Cohen's and Sawilowsky's (2009) guidelines for interpretation of effect sizes. Are these applicable to r the effect size? If not, what is/are a reference(s) for those guidelines? TIA -Greg
Hello Gregory,
Cohen would have been the first to say that, intimate knowledge of context and the variables in question trumps his guidelines for interpreting magnitude of effect, no matter what the ES metric. That is, while his .1 = 'small', .3 = 'medium', and .5 = 'large' ES for correlations might make sense with some variable sets (and he does give examples in his text, Statistical power analysis for the behavioral sciences), with others these may be altogether too stringent or too relaxed.
Expert judgment (perhaps via the Delphi technique) might be one way to elicit some sense of the degree of import for an obtained correlation. A second way might be via review of existing literature on the variables/population in question (though that may be more a normative than a declarative method).
Good luck with your work!
In addition to what David Morse says, it also depends on the longevity/duration of whatever is correlated.
For things that are incidental (and hence, affect some Y only once), a correlation of 0.3 may still be very small. For things that keep occurring (and hence, influence some Y over and over again), a correlation of 0.1 can be very big.
Relying on rules of thumb is indeed not the best practice. Try in addition to compare the effect you find to other (well-established) effects in your field.
Gentlemen, both excellent points and I greatly appreciate the time you took to respond. -g
- Badges
- Science topic
- Similar topics
- Correction