Academic journal article
By Holbert, R. Lance; Stephenson, Michael T.
Journal of Broadcasting & Electronic Media , Vol. 47, No. 4
Empirical media effects research is a diverse field that continues to expand its theoretical, methodological, and analytical boundaries. As McLeod, Kosicki, and Pan (1996) have stated, "more complex models of media effects and more sophisticated statistical methods are being explored and used to connect previously isolated communication processes" (p. 241). One advanced multivariate statistical technique recently employed by media scholars is covariance-based structural equation modeling (SEM). We explore the underutilization of this technique by media effects scholars. In particular, we focus on the absence of indirect effects in media research and why the failure to move beyond the study of direct effects is inconsistent with some of the basic theoretical foundations of mass communication.
Once an adequate fit of the data has been obtained for a structural equation model, researchers are afforded the opportunity to study three types of influence: direct, indirect, and total effects (Bollen, 1987). A direct effect, the influence of one variable on another, is represented in a structural model by a single path. An indirect effect assesses the impact of one variable on another as that variable's influence works through one or more intervening variables (Hoyle & Kenny, 1999). In addition, researchers can disaggregate a total indirect effect that works through multiple intervening variables into specific indirect effects (Brown, 1997; Fox, 1980). Each specific indirect effect isolates and assesses the role of a single intervening variable in a given relationship. The total effect of one variable on another is the sum of its direct and indirect effects.
Indirect effects are generally overlooked in most empirical research (Alwin & Hauser, 1975). This state of affairs is worrisome given that, "if an indirect effect does not receive proper attention, the relationship between two variables of interest may not be fully considered" (Raykov & Marcoulides, 2000, p. 7). Even though researchers have understood for some time the importance of media's indirect effects (e.g., McGuire, 1986), a recent critical assessment reveals that the study of indirect effects via SEM is woefully inadequate across the communication sciences (Holbert & Stephenson, 2002). Holbert and Stephenson found that only 14.4% of communication studies using SEM from 1995 through 2000 analyze indirect effects in even the most cursory fashion.
This essay stresses that media effects research needs to systematically decompose its structural equation models into direct, indirect, specific indirect, and total effects. The conditional model of media influence states that media can (and most often do) act as mediators in a given process, or alternatively, media can work through one or more mediators before having an influence on a particular dependent variable (McLeod & Reeves, 1980). We begin by formally linking the study of mediation to media effects research, and then isolate the mass communication subfields of political and health communication to provide examples of how mediation is central to the discipline. Several different formulas for testing mediation have been offered by various researchers (MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002), and three classes of mediation formulas are summarized. We isolate how SEM software packages test for indirect effects and assess how well this procedure performs relative to other formulas. We isolate one mediation formula as being particularly promising, and provide an example of its use to reveal its simplicity and utility. A series of recommendations for the assessment of total and specific indirect effects is offered. Finally, we reemphasize that the systematic study of mediation in media effects research is a necessary but not sufficient condition for better understanding the role of media influence in a host of contexts.
Mediation in Media Effects Research
Conditional Effects Model
McLeod and Reeves (1980) explain that the conditional effects approach acknowledges that media do not have universal influence on all individuals and/or societies. …