The Person and the Situation

by Lee Ross

A much more sensible convention for defining effect size was suggested by Cohen (1965, 1977), who suggested that the magnitude of experimental effects should be judged relative to the variability of the measure in question. Thus, by Cohen’s criterion, a difference between two means that corresponded to a quarter of a standard deviation in the distribution of the relevant measure would be deemed small, a difference corresponding to half a standard deviation would be deemed moderate, and a difference corresponding to a whole standard deviation would be deemed large. This statistical definition, and other related ones, assess effect size relative to all nonspecified, “random” determinants of variability or, in other words, relative to “noise.” The definition effectively finesses, in fact ignores, all considerations of the nature of the variable under consideration and the units of measurement involved. Therein lies both its major virtue and, as our discussion of the other two criteria will make clear, also its chief drawbacks.