The unappreciated relevance of auxiliary assumptions for evaluating theory-based interventions in health psychology

Applicability to interventions

We have highlighted the role that auxiliary assumptions play in theory prediction. But the application of auxiliary assumptions is not only relevant for theory prediction; it is important that interventions developed using theory also consider auxiliary assumptions (St Quinton et al., 2021). To see why, consider that interventions involve something observable, whether it is a marketing campaign, policy change, or other kind of treatment designed to induce behavior change. Therefore, just as it is necessary to conjunct theory with auxiliary assumptions to traverse the gap between nonobservational terms in theories and observational terms in empirical hypotheses, it is similarly necessary to conjunct theory with auxiliary assumptions to traverse the gap between nonobservational terms in theories and observables pertaining to interventions.

In turn, if an intervention fails, there are two plausible places to put the blame. First, the theory might be wrong, thereby causing the intervention to fail. Second, at least one auxiliary assumption may be wrong, thereby causing the intervention to fail through no fault of the theory. Therefore, effective theory-based interventions depend not only on the quality of the theory, but on the quality of the auxiliary assumptions too. In the subsequent section, we consider the TPB as an illustration.

Theory of planned behavior intervention auxiliary assumptions

A significant issue in assessing the usefulness of the TPB in intervention is how the theory was used. Interventions must target a relevant determinant (or combination of determinants) specified in the theory. That is because the importance of the underlying determinants need not be equal in all behaviors, persons, or populations (Trafimow & Finlay, 1996). For example, intention to engage in one behavior may be influenced by attitude whereas subjective norm could be the main driver of another behavior. Fishbein and Ajzen (2010) provide detailed guidelines on the formative work required to understand the contribution of determinants. Following this work, an intervention designer would understand which determinant(s) should be the focus of intervention.

Despite the importance of this formative work, it is surprising how infrequently this is undertaken, especially when compared to the number of studies that have adopted the theory. This may lead to interventions targeting psychological determinants irrelevant to the target behavior or population. For example, change is unlikely if an intervention addresses normative factors, but intentions are influenced more strongly by attitude. To illustrate this, Stasson and Fishbein (1990) had shown seat belt use in safe driving situations to be influenced by attitude and seat belt use in risky driving situations to be influenced by normative factors. To confirm these predictions, Trafimow and Fishbein (1994) manipulated attitude towards seat belt use in both situations and, in accordance with the initial predictions, found these manipulations had a greater impact on intentions to wear a seat belt in safe than in risky driving situations. Undertaking formative work to understand what to change ensures the intervention provides relevant psychological treatment. An intervention not applying the theory correctly has clear implications for theory effectiveness. It is therefore important to consider how theory was used in the intervention.

Despite this importance, one cannot establish effectiveness if a relevant determinant is not targeted or if the theory is not applied correctly. Yet, despite such importance, these issues are not part of the theoretical propositions outlined by the TPB; they constitute auxiliary assumptions. To be clear, the TPB outlines the causal influence of attitude, subjective norm, and perceived behavioral control to behavior through intention. Whether the theory is applied correctly at the empirical level, although needed to test these hypotheses, is nevertheless auxiliary to the theory itself. Failure to implement these is no fault of the theory and cannot provide a valid assessment of its utility. It is therefore important to consider such auxiliary assumptions to appraise theory effectiveness. We will now consider some auxiliary assumptions attached to intervention delivery.

A significant challenge for intervention designers in promoting health behavior change is modifying the theorized determinants. This is important if the theoretical propositions are correct, that is, that changing specific psychological determinants will lead to behavior change. This is also key if the theory is going to be put to the test. Clearly, it is not possible to test whether intention results in behavior change without first changing intention. And to change intention, we need to modify at least one of the underlying determinants. The designer must therefore select an appropriate strategy (or strategies) to include in the intervention. The taxonomy of behavior change techniques developed by Michie et al. (2013) provides a toolbox of strategies. The decision as to which of the 93 strategies to select could be facilitated by drawing on recent work attempting to map techniques onto psychological determinants (Carey et al., 2018; Connell et al., 2018). The assumption of this work is that an intervention designed to modify a specific determinant would fare better if one of the recommended strategies is used. For example, Carey et al. (2018) identified eight techniques that could effectively modify a person’s attitude, such as providing health consequences or incentivizing behavioral participation. Selecting an appropriate strategy (or strategies) is especially important given its influence on intervention effectiveness (McEwan et al., 2019). An intervention targeting a theoretical determinant, such as attitude, but adopting a poor strategy, is unlikely to demonstrate positive outcomes. But this failure is not a theoretical issue. To be clear, the TPB posits change will occur if specific determinants are altered. Now, failure to change these determinants is not a fault of the theory but rather indicts a poor auxiliary assumption attached to the theory in relation to intervention (Trafimow, 2015). Failure to change attitude, for example, might not say anything about the effectiveness of the TPB in behavior change; it might say more about the intervention applied to engender change. That the treatment successfully influences the crucial determinant is another auxiliary assumption needed to traverse the distance from the nonobservational theoretical terms to the observational terms in intervention hypotheses.

Other auxiliary issues concerning intervention delivery also exist. There are many ways to communicate an intervention, and each delivery mode has potential pros and cons. Face-to-face delivery requires participants to be present during the intervention, but personalized treatment can be provided using this modality. Online modalities are convenient and have widespread reach, but this delivery mode is reliant on technological availability and competence (Erbe et al., 2017). Other factors requiring consideration are intervention dose and frequency. Modifying attitude towards one health behavior may take a single session over a couple of weeks whereas changing attitude towards another health behavior could require many sessions over several months. These issues are, of course, important in influencing intervention effectiveness, though the theory says nothing about them. Interventions delivered using a website could prove unsuccessful if the sample have limited access to the internet. Similarly, attempting to modify erroneous deep-seated attitudes may fail to cause change if delivered on only a single occasion. But hopefully it is clear that despite influencing effectiveness, these are auxiliary assumptions attached to intervention. Instead of providing a test of the theoretical assumptions, such as whether perceived behavioral control changes intention or whether positive intentions translate into behavior, intervention effectiveness may be strongly influenced by auxiliary assumption quality. Appraising the effectiveness of the TPB within intervention therefore requires consideration of these auxiliary assumptions. It is unsound to draw conclusions about the theory without proper consideration of auxiliary assumption quality.

An intervention designer may carefully consider all of the above, yet behavior change is still lacking. The degree to which an intervention is delivered as intended by the designers is another important issue (Dusenbury et al., 2003). The deliverer may not implement the intervention correctly in practice and therefore experience challenges to fidelity. It is also important that participants respond to and engage with intervention material (Hasson et al., 2012). The intervention may be delivered accordingly, but the information may not be acknowledged, or relevant activities part of the strategy may not be undertaken. Consider the strategy of self-monitoring, which requires participants to monitor how often they engage in the behavior. Now, this strategy can alter a person’s perception of control; successful behavioral accomplishments increase the amount of control one perceives to have. But participants may not actually record their behavior, or the intervention deliverer may not correctly instruct participants on the self-monitoring requirements. On these occasions, intervention effectiveness can be influenced, but this is not necessarily a consequence of the underlying theory. Instead, fault lies with the way the intervention was delivered. This therefore represents auxiliary assumptions attached to intervention delivery when testing the theory. The designers need to make auxiliary assumptions that traverse the gap from what they intend to theoretically manipulate (e.g., perceived behavioral control) to whether the implemented delivery actually accomplishes that. Of course, it is important to also consider the auxiliary assumptions previously outlined: that perceived behavioral control is a relevant determinant underlying the behavior; that self-monitoring is an appropriate strategy; and that a relevant mode and frequency is adopted. Only when such auxiliary assumptions are considered can assumptions at the theoretical level be tested.

We have suggested auxiliary assumptions need to be considered when testing theory in intervention, and that one of the main issues is the inability to change psychological determinants. To demonstrate this point, Trafimow (2015) provides an example relating to increasing broccoli consumption through attitude change. He suggests that if we can somehow manage to make everyone like eating broccoli through making attitude towards consuming it more positive, then behavior change is likely to occur (assuming relevant control and normative influences). This may prove to be a difficult undertaking given the (potential) unpleasant taste of broccoli. But this problem is not with the theory; it is with knowing how to make people like eating broccoli. It is much easier to speculate that broccoli eating would increase if people’s attitudes towards eating broccoli could be made positive than it is to figure out how to make people like eating broccoli. But we can only test the theoretical assumption by modifying attitude in intervention. The problem, therefore, is not with the theory, but the auxiliary assumptions attached to the theory in intervention.

We can see how the TPB’s utility can be misrepresented by examining an intervention conducted by Sniehotta (2009). The intervention was developed to change university students’ attendance at sports facilities. The intervention successfully managed to change intention, but behavior change was not seen in recorded attendance. It was therefore concluded that the TPB did not suffice in guiding behavior change interventions. However, as Ajzen (2015) pointed out, closer inspection of the formative data suggested that, despite being motivated to attend the facilities, perceptions of control were weak. Focus should therefore have been on ensuring intention translation was possible by attending to issues of control. Additionally, the information provided in the intervention to modify cognitions was relatively poor (Ajzen, 2015). Intervention failure was therefore more a consequence of poor auxiliary assumptions than of poor theory.

To recap, many important considerations are required in the development and implementation of interventions adopting the TPB. However, such considerations do not provide tests of effectiveness at the theoretical level. These are intervention auxiliary assumptions required to bridge the gap between nonobservational theoretical terms and hypotheses about interventions. The TPB provides a causal role of psychological determinants to behavior; however, because intervention failure can be blamed on theory or on auxiliary assumptions, it is unsound to jump to the conclusion that theory is at fault. It would take substantial empirical investigation of the quality of the auxiliary assumptions, with strong data upholding that quality, before it would be sensible to blame theory for intervention failures. Though we have adopted the TPB to demonstrate our points, it is important to note the suggestions made are applicable to the many other psychological theories used to develop intervention, such as social cognitive theory (Bandura, 1986), health belief model (Rosenstock, 1974), COM-B model (Michie et al., 2014), and self-determination theory (Deci & Ryan, 1985). To do so, one would obviously need to consider the specific psychological content outlined in the theory. But given the limited support for theoretically informed interventions (Dalgetty et al., 2019; Prestwich et al., 2014; Rhodes et al., 2017), and the subsequent questions regarding the utility of behavioral theories (Hagger & Weed, 2019), we recommend increased appreciation of the role of auxiliary assumptions in traversing the distance between psychological theory and theory-based interventions.

It is also important to note we are not suggesting these issues are not considered by researchers. Indeed, studies often report, for example, the strategies used to change psychological determinants, and it is often acknowledged that intervention effectiveness depends on such strategies. However, what has not been made explicit is how these represent auxiliary assumptions and how their careful consideration is crucial for drawing sound conclusions about the worth of the theory. Again, an ineffective intervention could mean that the theory is at fault, or it could mean that there is one or more faulty auxiliary assumptions. Given that the failures of theory-based interventions can be due to poor theory or to poor auxiliary assumptions, it is worthwhile to consider how to distinguish the two possibilities. This is a complex issue that deserves its own article, but we can recommend a few items to consider. First, there are psychometric issues, such as the reliability and validity of the measures. Many researchers seem to think that obtaining a nice factor structure establishes construct validity, but more is needed. Construct validity involves a matching of empirical with theoretical relations (Cronbach & Meehl, 1955), and if this has not been accomplished, and it rarely has been in our opinion, then construct validity has not been accomplished either (Trafimow & Osman, 2022). In that case, intervention failure can plausibly, though not definitively, be attributed to a failure in the auxiliary assumption of construct-valid measures. Another item, to which we alluded earlier, is whether the intervention influenced the theoretical constructs it was supposed to influence. One way to determine this is to measure the theoretical constructs of interest. For example, if an intervention is designed to increase exercise behavior through influencing underlying attitudes, the researchers could measure attitudes in the experimental and control conditions. If there is no difference in attitudes in the two conditions, then the intervention may not have influenced them, and the failure to increase exercise behavior might reasonably be blamed on the failure to influence attitudes rather than on the theory. That said, however, even if an effect on attitudes is obtained, the intervention may have unintentionally influenced another relevant construct in the opposite direction. If the intervention increased attitudes, but at the cost, unbeknownst to the researcher, of decreasing perceived behavioral control, the intervention can scarcely be expected to succeed even if the theory is correct. In general, when there is a lack of measures of relevant theoretical constructs, that may shift the weight of suspicion in the direction of auxiliary assumptions as opposed to theory. A third item is whether the researchers have made important auxiliary assumptions explicit (Trafimow, 2012). If not, that would be a reason to suspect that the researchers have not carefully considered these auxiliary assumptions, and that lack of consideration might be a reason to similarly suspect that an intervention failure should be blamed on auxiliary assumptions. Finally, the track record of the theory’s degree of corroboration in previous tests is an important factor to consider. When a theory has survived a set of severe tests but an intervention based on it nevertheless fails, the theory may still be wrong (Popper, 1963), but more likely, the problem would lie with the auxiliary assumptions. In this case, the initial search for the reason for failure might best focus on auxiliary assumptions.

Although we have provided examples of auxiliary assumptions, many more require attention when testing theory in intervention. For example, when participants provide psychological assessments, it is not explicitly assumed that participants can read questionnaires, that participants understand what is asked of them, or that participants are honest in their responses. Yet theory effectiveness depends on these assumptions. We acknowledge there are many auxiliary assumptions to consider, and some are more difficult to accommodate than others. However, behavior change is, after all, difficult, and successfully influencing health behavior is crucial for health and wellbeing. We hope that bringing attention to the cruciality of auxiliary assumptions aids researchers in more accurately evaluating the worth of theories. This would provide a significant contribution to the field through improving the soundness of theory evaluation in the field of health behaviors.