Theories of PA behaviour suggest that particular constructs are critical antecedents of behavioural engagement. These constructs are hypothesized as components of a causal chain, suggesting that if the mediators are changed, behaviour change should follow [6
]. Early reviews based on theoretical mediators of behaviour change, however, suggested that few formal tests of mediation had been conducted and limited evidence was available to support this proposition [3
]. Therefore, the purpose of this review was to provide an update of the literature on PA interventions that have included proposed mediators of behaviour, focusing specifically on primary prevention in adults since the time of these prior reviews.
The review yielded 29 studies from 27 independent samples to appraise our current understanding of PA mediators in interventions. Five studies were omitted from the analyses due to low quality but the other 22 trials showed moderate (n = 21) or high (n = 1) quality and thus relatively low risk of bias. Almost all studies did not meet the category for high quality because they failed to include a direct measure of physical activity behaviour and did not report on a pilot intervention to demonstrate that it could affect the mediators. Otherwise, the 22 trials generally showed many high quality features such as random assignment, a theoretical-base, reliable and valid measures of the mediators, and reliable measures of self-reported PA.
Overall, 11 studies showed that the intervention had an effect on PA behaviour change and all of these studies subsequently had an action theory link [59
]. That is, all 11 studies showed some evidence that the intervention also changed the proposed mediators. By contrast, a conceptual theory link [59
] was seldom reported (5/11 studies). Conceptual theory links demonstrate that changes in the mediators are related to the PA outcome. These are often the foundation for using a theory or mediator construct before initiation of the intervention [11
], but future work needs to test this link regularly in reported trials with mediators. Formal tests of mediation were also only conducted in six of the 11 cases where the procedure may have been appropriate (i.e., intervention effect on behaviour, evidence of action theory link, conceptual theory link or probable conceptual theory link). In terms of behavioural mediation by theory, TTM, SCT, TPB, PMT, and SDT all showed some evidence for action theory tests and all have shown evidence for conceptual theory tests in the past, but only the TTM employed tests of a mediated effect of its constructs. The results, when divided by theory, are too limited in number to make particular judgements at present.
A division at the construct level [6
], however, provides a larger sample for assessment. Self-regulation constructs (e.g., planning, behavioural processes) from trials where the intervention changed PA behaviour showed 75% (6 of 8 studies) support for action theory tests and all three of the conceptual theory tests conducted were significant. Mediated effect tests of the construct, however, were mixed with two showing support and two not providing evidence for mediation. Our appraisal of self-regulation is similar to the original comments made by Lewis et al. [9
]; the construct has the most support thus far but still demonstrates mixed findings. Still, it seems prudent to include a self-management and self-regulatory component to PA interventions.
Results of self-efficacy and outcome expectation-type constructs as mediators were weak or limited. Self-efficacy constructs among intervention studies that affected PA change showed relatively strong evidence for action theory (7 of 10 studies) and conceptual theory (3 of 4 studies reported) links, but a mediated effect was not supported in four of the five formal tests conducted. Outcome expectation constructs had similar results in terms of evidence for an action theory link (6/7 studies) but zero of the four tests for a mediated effect were significant. There was some notable differences between affective and instrumental outcome expectations (see [60
] for extended commentary) with positive changes in affective outcome expectations linked to positive changes in behaviour more than instrumental outcome expectations. Still, the relatively few studies on this topic and absence of any formal mediation tests render this point as speculative at present.
Social constructs were limited to only four studies where the intervention had produced significant changes in PA; however, three of these four studies showed an action theory link. Social support was also a mediator of behaviour change in one formal test of mediation, but was unable to show a mediated effect in the only other test with this construct. There was no evidence for the mediation capacity of subjective norm. Although limited literature precludes any definitive conclusions, social constructs, particularly social support, may have utility as mediators of change but findings are mixed at present.
A key finding of the review, however, was that half the interventions failed to change both behaviour and the proposed mediators through the action theory link. This does not challenge the internal structure of our leading theories and constructs at present as much as demonstrate that our interventions are generally ineffective. To evaluate the mediation capacity of a theory, the behavioural link and action link are important first steps in mediation [11
]. Pilot studies showing evidence that the intervention can change the proposed mediators are recommended in future research before large-scale trials are conducted.
The poor performance of PA interventions has been duly recognized [3
], and it is much easier to comment on this problem than to provide solutions. Nevertheless, it is important to provide some commentary on this issue. A most pragmatic possibility for these results may be attenuation from measurement error. For example, indirect (self-report) PA measures featured in these studies may lack the sensitivity to distinguish change between the groups and the psychological constructs may equally lack precision [3
]. Direct measures of PA are recommended in future trials. Still, this seems unlikely to be the sole reason for these null effects; many of the studies were able to demonstrate time effects (i.e., main effects), and the proposed mediators generally show moderate to large bivariate correlations with PA in prediction tests [e.g., [61
Clearly more innovation and higher fidelity interventions are needed. In the studies reviewed, there was a very similar genre of intervention. These typically focused on a persuasive educational component about the benefits of PA and hazards of inactivity followed by problem solving suggestions to regulate action and overcome barriers. Although this approach could be helpful to some, it was not helpful to change proposed mediators in over 50% of the cases reviewed and these null results were not readily identifiable as discrepant intervention styles from successful trials. The problem occurring may be that the approach is an insufficient band-aid to overcome the real-world obstacles and different values that some inactive participants experience. Social and environmental structures may be so grounded and geared to sedentary lifestyles that individual-level, inexpensive, patches may not resonate with the inactive populace [41
]. The limits of these "downstream" approaches have been recognized [62
]. Approaches at system-level social and environmental change may be needed to aid many people [63
]. This approach, of course, is costly and does not lend itself to the tight-budget three-year RCT; indeed, it is likely to conflict with other societal and industrial aims.
Interacting with these more systemic social and environmental issues may be systemic internal issues. Enacting a potentially fatiguing, boring, and time-consuming behaviour on a repeated basis in the face of other behavioural options and values is likely to pose an enormous daily challenge to many people. Some of this may arise from differences in genetic predisposition and other individual differences that are not easily intervened upon [64
], while some of these decisions may be the result of informed free will.
When considering these possibilities, it seems important for future interventions to become more innovative and target proposed mediators with a higher fidelity. Using the tenets of SCT as a guide[57
], the experiential qualities of the behaviour seem the most telling way to affect cognitions rather than passive approaches. Experiences of valued personal outcomes (e.g., enjoyment, pleasure, satisfaction) and behavioural control/self-efficacy, through shifts in behavioural, environmental and social experiences of PA may be the most effective intervention alongside increasing self-regulatory skills. At this time, we recommend that interventions focus on altering the behavioural experience in an attempt to improve fidelity and affect change in proposed mediators.
It is important to highlight the limitations of this review in order to provide a context for the results. First, the assessment is limited to published work and may be subject to publication bias. Given the high rate of null effects in these results, the bias may be minimal but no formal test of publication bias can be conducted. Second, the work contained in this review is limited to English written journals and thus the results cannot generalize to studies conducted and published in other languages. Finally, the review is limited to the search terms and data-bases contained in our methods section, which followed the precedent of Baranowski et al. [3
] and Lewis et al. [9
]. Studies that have not been abstracted with these key words will be missing from our review.