Mediators & Moderators Flashcards
Mediators: definition
Mediators refer to “an active organism [that] intervenes between stimulus and response” (Baron and Kenny, 1986)
Mediator analyses:
o Ask how and why interventions work (mediator = link between)
Analyzing mediators can illuminate how and why an intervention proves effective
Baron and Kenny (1986) enumerate four conditions that must be met in order to claim mediation:
o (1) the independent variable must affect the mediator in the first equation [i.e., the intervention/treatment predicts mediator A]
o (2) the independent variable must be shown to affect the dependent variable in the second equation [the intervention predicts outcome C]
o (3) the mediator must affect the dependent variable in the third equation [the mediator predicts outcome B]
o (4) If these conditions all hold in the predicted direction, then the effect of the independent variable on the dependent variable must be less in the third equation than in the second [essentially a fourth condition, dictating that the effect of the intervention on the outcome must be reduced upon introduction of the mediating variable] (Baron and Kenny, 1986, p. 1177)
Operating under the basic assumption that a causal relationship exists between an intervention and an observed outcome—an assumption made reasonable in the context of RCTs by the process of randomization—we can conceptually frame a mediator as an intervening variable on the causal pathway of an intervention, responsible for somehow shaping the relationship between stimulus and response
Moderators: definition
Moderators, such as socioeconomic class, gender, race, or level of incentive provided, alter “the direction and/or strength of the relation” between independent and dependent variables (Baron and Kenny, 1986)
Moderator analyses
o Ask for whom interventions work (moderator = alters, modifies effects)
o Are there differential effects for different subgroups?
- Important implications for ‘equity’ of effects; key topic in public health (Petticrew, 2011)
Moderator analyses yield insight into the individuals/subgroups for whom an intervention ‘works’
In particular, moderator analyses evaluate whether participant outcomes differ in accordance with baseline characteristics such as socioeconomic status, ethnicity, gender, and/or experienced severity of the problem under study
Differential subgroup effects can manifest in a variety of ways, with an intervention potentially suitable for certain subgroups, less effective for others, and actively harmful to yet another segment of the population
The field of public health in particular produces a significant body of research primarily concerned with examining the ‘equity’—or lack thereof—of intervention effects (Petticrew, 2011)
Value of conducting analysis of mediators in trials of social interventions
We should seek to supplement our confidence that an intervention “works” with an understanding of the underlying mechanisms that together culminate in the resulting effectiveness, crucially entailing the methodical analysis of mediators [and moderators]
Insofar as we wish our research to meaningfully inform practice, we must take care to unpack the ‘black box’ that often camouflages the causal processes and active ingredients driving an intervention
Mediator analyses elucidate how and why an intervention works, identifying the “active therapeutic components” of a treatment, and ultimately allowing us to refine treatments so as to maximize effectiveness and minimize cost
Even should the results of a robust randomized controlled trial indicate that an intervention works in achieving its desired outcome(s), conducting an analysis of potential mediators proves a worthwhile endeavor
Ultimately, to understand an intervention’s mediating variables is to understand how and why an intervention produces the effect(s) it does
o For example, in a study of 3-, 6-, and 9-month-old infants shown panels containing a single animated image interspersed among stationary but otherwise identical images, researchers found that accuracy in identifying the moving image—known as “visual search accuracy”—fully mediated the amount of time the infants spent viewing faces, the dependent variable under study (Frank, Amso, and Johnson, 2014)
By striving to fully appreciate “the mechanisms through which treatments operate,” we find ourselves better equipped to maximize treatment effectiveness, and simultaneously more able to reduce the monetary and human costs associated with a given treatment—“Active therapeutic components could be intensified and refined, whereas inactive or redundant elements could be discarded” (Kraemer, Wilson, Fairburn, and Agras, p. 878)
In addition to shedding light on the relative utility of an intervention’s constituent ingredients, mediator analyses provide valuable insight into the very nature of medical disorders and social phenomena, knowledge which is then utilized to develop more acutely targeted treatments
o Notably, when evidence emerged to suggest that cognitive behavioral therapy (CBT) works in treating panic disorders via the eradication of catastrophic thoughts related to bodily changes—a mediating mechanism—the cognitive theory of panic gained greater empirical substantiation (Clark, 1997)
When conducted in conjunction with one another, mediator and moderator analyses can maximize the equity of an intervention without necessarily compromising its effectiveness
Value of conducting analysis of moderators in trials of social interventions
We should seek to supplement our confidence that an intervention “works” with an understanding of the underlying mechanisms that together culminate in the resulting effectiveness, crucially entailing the methodical analysis of moderators [and mediators]
Moderator analyses ask for whom an intervention proves effective, seeking to uncover any differential subgroup effects, some of which may be contributing to the exacerbation of existing social disparities
When the incongruous effects of a treatment materialize so as to either disproportionately benefit an already advantaged subgroup or disproportionately harm a traditionally underprivileged group, then the treatment in question may in fact serve to further exacerbate existing social disparities, engendering criticism on a social justice front
The field of public health in particular produces a significant body of research primarily concerned with examining the ‘equity’—or lack thereof—of intervention effects (Petticrew, 2011)
o Famously, a preponderance of evidence strongly suggests that media campaigns targeted at reducing cigarette use prove most effective at achieving their aim among the socioeconomically well-off, thereby widening already established inequalities (Lorenc et al., 2013)
To the extent that we care about the broader societal implications of our interventions, rather than simply whether or not they ‘work’ in the most rudimentary sense, moderator analyses become an invaluable tool
Upon identifying inequitable effects of an intervention through moderator analyses, we can work to better understand why certain subgroups respond more positively than others to a treatment by investigating the potential mediating mechanisms at play
When conducted in conjunction with one another, moderator and mediator analyses can maximize the equity of an intervention without necessarily compromising its effectiveness
Moderator analyses also present a means by which to capitalize on the increasing interest among policymakers in promoting more targeted, highly tailored interventions, with the initial investment incurred in developing a well-founded knowledge base of what works for whom seen as more than justified by the resulting ability to maximize efficiency and minimize risk
The considerable influence wielded by intervention research pertaining to the topic of subgroup analysis extends to “policy decisions around programmatic aims (e.g., Upward Bound), funding decisions (e.g., Even Start), and new initiatives targeting funding towards evidence-based programs (e.g., teen pregnancy and home visitation)” (Supplee et al., 2013, p. 107)
Research should not exist in a vacuum, and with the present demand for subgroup-specific interventions unlikely to abate in the foreseeable future, moderator analyses represent an especially promising avenue for bridging the gap between research findings and clinical practice
Insofar as we wish our research to meaningfully inform practice, we must devote the time, energy, and resources necessary to establish the existence of any differential subgroup effects prior to the widespread implementation of an ‘effective’ intervention, so as to minimize the risk of unknowingly inflicting harm on a large scale, and ultimately maximize efficiency
Challenges in conducting mediator analyses
The continued prevalence of outcome reporting bias in primary (i.e., main effect) analyses of trials, as highlighted by the work of Dr. Kerry Dwan and colleagues (2008), raises concerns that such ‘cherry picking’ of favorable results may abound to an even greater extent in secondary analyses of mediators and moderators
Challenges in conducting moderator analyses
The continued prevalence of outcome reporting bias in primary (i.e., main effect) analyses of trials, as highlighted by the work of Dr. Kerry Dwan and colleagues (2008), raises concerns that such ‘cherry picking’ of favorable results may abound to an even greater extent in secondary analyses of moderators and mediators
Reflecting his deep misgivings about the validity of conclusions drawn from moderator analyses, statistician and epidemiologist Sir Richard Peto even goes so far as to claim that “only one thing is worse than doing subgroup analyses—believing the results”
Critics note with apprehension the prohibitively low statistical power of a great deal of secondary analyses, finding that “[m]any [major clinical trial] reports put too much emphasis on subgroup analyses that commonly lacked statistical power” (Assmann, Pocock, Enos, and Kasten, 2000)
Mark W. Lipsey (2003) delineates the difficulties that can arise from confounding among moderator variables, a relatively likely turn of events in which the moderators under observation are found to be related to each other as well as to effect sizes
Issues of replicability have also been identified in relation to the secondary analyses of intervention data, with various moderator analyses yielding divergent—and sometimes contradictory—results (Baydar, Reid, and Webster-Stratton, 2003; Beauchaine, Webster-Stratton, and Reid, 2005; Gardner, Hutchings, Bywater, and Whitaker, 2010)
Resolving criticism of mediator and moderator analyses
Though each of the above concerns reflects a potential limitation of secondary analyses, many of these obstacles can be satisfactorily addressed, and none are sufficient to warrant a complete abandonment of systematic inquiry into mediator and moderator variables
Heightened awareness of the potential limitations of secondary analyses grants us a valuable opportunity to institute measures designed to diminish the impact of foreseeable stumbling blocks, such as through the systematic pooling of individual-level data from a relatively large number of trials, which serves to not only increase power for subgroup analyses, but greatly enhances transparency and generalizability as well
The risk of ‘cherry picking’ only those results that suit a given narrative can be minimized by greater transparency, achievable in part through the explicit pre-specification of hypotheses, including whether a hypothesis is exploratory or confirmatory in nature, as well as its underlying rationale (Rothwell, 2005)
Pooling data from a number of trials can address the lack of statistical power common to secondary analyses, while also providing for greater generalizability across contexts (Brown et al., 2013; Gardner et al., 2017)
Fundamentally, much of what constitutes ‘good practice’ in conducting secondary analyses mirrors or closely tracks that which is expected in primary analyses, with transparency—and pre-registration—being of paramount importance no matter the level of analysis
