Chapter 28.

A century ago, psychology was a young field, greatly influenced by the physiological tradition of Hemholtz and Weber, who looked at psychological phenomena in an experimental fashion, and “only that which could be directly observed and studied under controlled laboratory conditions was deemed worthy of study” (Minton & Schneider, 1980, p. 7). Any individual variation in these experiments was just considered error. Galton, with his precocious nature and Darwinian influence, sought to apply “the principles of variation, selection, and adaptation to the study of human individuals” through his study of quantitative genetics, mental chronometry, and statistics (Anastasi & Foley Jr., 1949, p. 9). Likewise, Cattell (1890) sought to assess individual differences in his “mental tests,” which he administered to a generation of Freshman at Columbia College. It was the work of Binet and Henri (1896) and Stern (1900) focused psychologists on major concerns such as: (a) the nature and extent of differences in individuals and groups, (b) the factors that determined these differences, and (c) how the differences are manifested.
Paramount to this study of individual variation was the study of psychometrics and statistics, as “an intelligent interpretation of almost any study in differential psychology requires an understanding of certain fundamental statistical concepts” (Anastasi, 1958, p.9).  Yet few texts on individual differences include sections on appropriate statistical methodology to studying these issues, such as statistical moderation, mediation, and moderated mediation (but see Reyonds & Willson, 1985).  In fact, it was not until Barron & Kenny (1986) published their paper on testing these things systematically that the issue was brought to the fore.  In the twenty years since, the importance of individual differences has continued to grow, yet many researchers fail to use best practices in properly testing for these phenomena.  Consequently, this chapter will focus on the concepts of moderation, mediation, and moderated mediation within the framework of OLS multiple regression. Although more advanced modeling techniques are available [e.g., covariance structure analysis (Jaccard & Turrisi, 2003), multilevel models (Davison, Kwak, Seo, & Choi, 2002; Raudenbush & Bryk, 2002)], the ubiquitous use of multiple regression within the social sciences suggests that determining and delineating best practices are both necessary and advantageous. Moreover, alternative techniques can be overly complex and require a more refined statistical background for their interpretation and, currently, there is no consensus as to what non-regression procedure is “the best” for a given analysis. Third, and probably most important, the best practice concepts involved in mediation, moderation, and moderated mediation are not strictly dependent on a specific method per se. Thus, gaining an understanding of best practices within a multiple regression framework can help determine the same in more complex models.

Resources for walkthrough drescribed in the chapter:

Table 1
Description of variables CONTAINED IN THIS DATA SET