Effect Sizes and Groups
People have been studying social behavior and personality for a long, long time. Although many suggest that Triplett’s 1898 study marks the start of the scientific investigation of interpersonal processes, in all likelihood the field’s roots reach even further back in time (Stroebe, 2012). In any case, in the last 100 years researchers have conducted thousands of studies of social behavior, which–taken individually–may shed only a pinpoint of light on an intriguing social psychological questions, but when synthesized support more general, sweeping conclusions.
Reviews of previous work generally come in two flavors: narrative and quantitative. When writing a narrative review, the researcher examines previous research carefully and draws general conclusions about the strength of the relationships among the variables that have been investigated. When writing a meta-analytic review, in contrast, the researcher combines the results of previous research statistically to determine, quantitatively, the strength of the relationships under study. Although the size of the samples and the number of studies analyzed influence these estimates of relationship, those that fall between .1 and .2 are considered small, from .2 to .5 moderate, and those above .5 large.
When Richard, Bond, and Stokes-Zoota (2003) examined 100s of prior meta-analytic studies of various social psychological processes, they discovered that the average effect size in those studies was .21, a low to moderately strong effect. But, when they looked more closely across topics, they discovered that some relationships were particularly paltry, whereas others were more robust. Studies the relationship between personality and behavior, for example, are often considered relatively unsubstantial by social psychologists, yet they were consistently stronger (r = .22) than the relationships documented in studies of influence (r = .12), attribution (r = .14), and expectancies (r = .16). And what one area of study had yielded the strongest support for predicted relationships between the variables specified in its theories? Leading the way, across all 18 topics identified by Richard, Bond, and Stokes-Zoota (2003), was the scientific study of groups and their dynamics, with mean r of .32.