4.1: Facts First

To understand the need for a new theory of exercise and physical inactivity, it is first useful to critically appraise the theories of behavior change that have been used to study these behaviors in exercise psychology over the past three decades. First, we assess the results of intervention studies. Then we turn to the theories themselves (for more discussion on theories used to explain and change physical activity and exercise behavior, see Chapter 2, Rebar et al., 2021;and Chapter 5, Delli Paoli, 2021).

4.1.1. Intervention Studies

A few years ago, a meta-analysis¹ summarized and statistically re-evaluated the results of 82 randomized controlled trials² (RCTs) that tested the efficacy of theory-based interventions for increasing

Table 4.1. Theories of Behavior Change Used in PA Intervention Studies

Theories	Synopsis of the theory
Social cognitive theory(SCT; Bandura, 1986)	By observing others, how they behave within a social context and which experiences they create, the individual learns about the consequences of new behaviors. Outcome expectations, social-structural factors (facilitators and impediments), and the belief that one is capable of executing the behavioral steps necessary to achieve a goal (self-efficacy) can lead to behavioral goals that are then pursued by the individual.
Theory of planned behavior (TPB; Ajzen, 1991)	Behavioral beliefs form an attitude towards a behavior, normative beliefs result in subjective norms, and control beliefs determine one's perceived behavioral control over perceived barriers. These three interdependent elements determine whether an intention to change behavior is formed, which, if strong enough, is theorized to influence behavior.
Self-determination theory (SDT; Deci & Ryan, 1985)	Self-determination theory consists of several mini-theories. Motivation is theorized to be most effective when it is intrinsic (i.e., when a behavior is performed for its own sake; cognitive evaluation theory). When behavior is externally regulated, it can still be increasingly aligned with oneself (organismic integration theory). If behavior is perceived as supporting the basic psychological needs for autonomy, competence, and relatedness, then well-being and general functioning are optimized (basic psychological needs theory). In addition, interindividual differences are assumed in people's tendencies to appreciate autonomy, to focus on outcomes (such as rewards, gains, and approval), and to worry about their competence (causality orientations theory). The SDT posits that pursuing extrinsic goals leads to less wellness than pursuing intrinsic goals (goal contents theory),and that close personal relationships are essential for adjustment and well-being (relationships motivation theory).
Transtheoretical model (TTM; Prochaska & DiClemente, 1982)	The TTM integrates the functional elements of diverse previous theories. According to it, most characteristically, behavior change can be regarded as proceeding across discrete stages of change, namely precontemplation, contemplation, preparation, action, and maintenance (relapse is sometimes included as an additional stage). The model defines processes of change that are useful for facilitating stage progression (e.g.,consciousness raising from precontemplation to contemplation; stimulus control from action to maintenance). Stage progression is presumed to be accompanied by changes in subjective beliefs, e.g., the pros and cons of behavior change(decisional balance) and self-efficacy.

Note. According to Gourlan et al.(2016), 31 of the analyzed interventions referred to the TTM, 16 to SCT, 8 to TPB, 5 to SDT, only 1 to protection motivation theory(Rogers, 1983; not listed in the table), and 21 to combinations of the sefive theories.

The results of the meta-analysis (Gourlan et al., 2016) indicated a statistically significant difference in PA behavior (self-reported PA, device-measured PA, or a combination of both) in those who participated ina theory-based intervention (n= 10,574 individuals) compared to those who did not (n= 8,783 individuals). The effect size of this difference was quantified as falling in the range between d = 0.24 and d= 0.37 (95% confidence interval). This is typically characterized as a "small" to "medium" effect (0.2 < d < 0.5), according to convention (Cohen, 1988). Let's illustrate how small such an effect is.

Figure 4.1. Graphical Display of a Statistically Small (Part A) and a Large Effect Size (Part B)

Note. Standardized mean differences (d) are referred to as small (d = 0.20), medium (d = 0.50) and large (d = 0.80) effect sizes (Cohen, 1988). The d = 0.31 in Part A is taken from the results of the meta-analysis on the efficacy of theory-based PA interventions (Gourlan et al.,2016). For a more detailed explanation, see the body of the text.Source: https://rpsychologist.com/d3/cohend/(Magnusson, 2020).

Figure 4.1, part A shows two overlapping distributions³ with mean values differing by a "small" to "medium" effect size of d = 0.31 (i.e., the mean effect size for theory-based PA interventions, as identified by Gourlan et al., 2016).

The two distributions have 87.7% overlap, which gives a 58.7% chance that a study participant picked at random from the intervention group (the distribution presented in light blue) will have a higher score (e.g., PA level) than a person picked at random from the control group (i.e., "probability of superiority"). In other words, there isn't much more than a fifty-fifty chance (similar to a coin toss) that the intervention will change behavior at the level of a single individual.⁴ By comparison, Figure 4.1, part B illustrates two distributions that differ by a conventionally defined "large"effect size (i.e., standardized mean difference of d = 0.80). With a large effect, the probability of superiority increases to 71.4%.

In simpler terms, according to the results of the meta-analysis on the efficacy of theory-based PA interventions (Gourlan et al., 2016), if we randomly selected one of the 10,574 study participants who received a theory-based intervention, there is a less than 60% probability that this person would be more physically active than someone randomly selected from the 8,783 study participants who did not receive an intervention. Arguably, most exercise practitioners would consider this success rate to be discouragingly small for practical purposes.

Other interesting meta-analytic findings on the efficacy of theory-based PA interventions include the following.The success rates of interventions differed little, regardless of the theory that was used as their basis(Gourlan et al., 2016). Effect sizes are usually small across different meta-analyses(e.g., Compernolle et al., 2019; Grimmett et al., 2019; Ntoumanis et al., 2020; Romain et al. 2018). Moreover, there is evidence of publication bias⁵ in this research field (Gourlan et al., 2016). Studies of lower methodological quality tend to produce larger effect sizes than the (rarer)high-quality studies (Bernard et al., 2017). Ameta-analysis that included not only RCTs but also correlational studies reported slightly higher, but still small,effect sizes (McEwan et al., 2019);importantly, it also showed that theory-based interventions (148 studies, d = 0.48) and interventions not explicitly based on a theory (77 studies, d = 0.37) did not differ substantially in terms of their efficacy, as long as at least three behavior-change techniques were involved (e.g., social support, goals and planning, rewards and threats; Michie et al., 2013). Other meta-analyses have arrived at similar conclusions (e.g., Howlett et al., 2019; Rhodes et al., 2020).

4.1.1.1. In a Nutshell

Although authors of meta-analyses have concluded that PA interventions are effective (i.e., their efficacy, compared to control groups, is significantly different from zero), the practical meaningfulness of these results is questionable. A "small" effect is of limited practical relevance, especially when considering that effect sizes may be overestimated due to various methodological biases. Recent research has shown that basing an intervention on one of the theories discussed above does not meaningfully improve intervention effectiveness.

4.1.2. Explanation of Behavior (Theory Testing)

Numerous reviews on the theories of PA and exercise behavior change have been written, with almost identical conclusions(e.g., Buchan et al., 2012;Weinberg, 2018).One of the most recent review articles provided an historical and critical summary of individual theories and broad theoretical frameworks that have been used most frequently in exercise psychology over the past decades (Rhodes et al., 2019).

According to this review, SCT and TPB are important examples of the social cognitive framework (the TTM and PMT can also be classified within this broad theoretical framework). The reviewers characterized it as "the dominant framework for understanding physical activity" (Rhodes et al., 2019, p. 100). Theories within this framework are characterized by two common assumptions, namely (a) that influences from the social environment shape our views and how we may behave, and (b) that people use their expectations about the future outcomes of their actions and inactions as a basis for directing their behavior.The SDT was highlighted by the reviewers as a theory from the humanistic/organismic framework that has seen "a surge in research in the last decade"(p. 100). This theory emphasizes a view of human beings that is about realizing potential, personal growth,and development. Rhodes et al. (2019) further referred to dual-process theories as "the most recent and understudied framework" (p. 100). According to theories classified under this framework, behavior is influenced by deliberate thinking on the one hand,and more automatic mental phenomenaon the other (Evans & Stanovich, 2013).⁶ Theories from this framework have not been used as the basis of interventions in exercise psychology so far.

That there are, in fact, only a handful of frequently used theories in exercise psychology can also be gleaned from a quick search in the Web of Science database.⁷ Figure 4.2 illustrates how often the frameworks and theories have been cited in research publications over the past three decades (from 1990 to 2020). The large recent increase of interest in the SDT is obvious, and so is the even more recent interest in dual-process theories.

Figure 4.2. Growth in the Popularity of Theories of Behavior Change in Research Related to PA and Exercise

Let us now examine the empirical support for these theories within exercise psychology by focusing on meta-analyses of (cross-sectional and prospective) correlational studies (rather than interventions). With regard to the TPB, it can be stated that attitude and perceived behavioral control account for roughly 30% of the variance in behavioral intention(with a modest and typically nonsignificant contribution of subjective social norms); intention is a significant predictor of behavior and has been estimated to account for approximately 20% of the variance in self-reported PA (Downs & Hausenblas, 2005). With the variables defined in the SCT (especially self-efficacy and goals), the portion of accounted variance in behavior reaches 31% (Young et al., 2014). From meta-analyses on SDT, it can be concluded that effect sizes are slightly higher. Results suggest that the correlation between self-determined motivation and PA behavior can range between r = .30 and r = .45 (Hagger & Chatzisarantis, 2009; Vasconcellos et al., 2020, for a more recent study on adaptive and maladaptive outcomes in physical education). Comparable correlations have been reported from the synthesis of study results based on predictions of the TTM. For example, later stages of change seem to be associated with higher levels of PA, changes in self-efficacy are moderately consistent with the TTM's predictions, stage progression from contemplation on is associated with small increases in perceived pros and large decreases in perceived cons, and behavioral processes of change peak from precontemplation to contemplation and from preparation to action(Marshall & Biddle, 2001). Meta-analyses on dual-process theories do not yet exist.⁸

Notice that, while in the earlier section that referred to meta-analyses quantifying the effect size associated with theory-based interventions compared to control, the effect size was Cohen's d,which represents the standardized mean difference in the outcome variable (e.g., PA) between the intervention and control groups. In contrast, in meta-analyses that examine relations between variables, the effect sizes used are the "percentage of accounted variance" (r²) and the correlation coefficient (r). While keeping in mind that randomized controlled trials of interventions can be used to draw inferences about causation whereas correlational studies cannot, the different indices of effect size (d, percentage of accounted variance,and the correlation coefficient) can be converted to each other for comparison purposes.A conventionally large effect size of d= 0.80 can be considered analogous to 26% of accounted variance; correlation coefficients of r = .30 are characterized as medium, while those of r = .50 are described as large (Cohen, 1988). Thus,it seems that correlational empirical evidence obtained so far is consistent with the postulates of the various theories.

That said, there are again good reasons to look beyond statistics when trying to understand why these few theories have been preferred over others. First, it should be noted that a common thread among all of the above-mentioned meta-analyses is the generally poor methodologies of most of the included original studies; again, higher methodological quality is typically associated with lower effect sizes (e.g., Hagger & Chatzisarantis, 2009). Secondly, specifically regarding the meta-analyses summarizing the results of correlational studies, it is important to place these results in the context of a crucial methodological pitfall. When two variables are measured from the same source (e.g., only the respondents themselves), via the same response channel, such as questionnaires (e.g., attitude and intention, or autonomous motivation and behavior), at the same physical location, and at the same time (e.g., back-to-back, as part of the same battery of measures), their intercorrelations tend to be inflated by so-called "common method bias"(Armitage & Conner, 2001; Podsak off et al., 2003). This means that part of the apparent intercorrelation between variables reflects not overlap between the constructs per se but rather the commonality of the measurement methods. Remarkably, despite this known source of bias, most exercise psychology studies measure not only psychological variables, but also PA and exercise behavior using questionnaires. Thirdly, recent analysis shows that the medium and large effect sizes reported in many psychological studies must be evaluated with utmost caution when the studies in question were not preregistered (i.e., if the researchers did not detail their intended research aims, hypotheses, and methods prior to conducting their study, by entering this information in a published "protocol"paper, a public registry, or both). A recent analysis revealed that in preregistered psychological studies, effect sizes often appear greatly reduced or even halved(Schäfer & Schwarz, 2019). In exercise psychology,preregistration has been extremely rare until recently.

4.1.2.1. In a Nutshell

SDT and SCT are the most commonly used theories in studies that have attempted to predict or change PA and exercise behavior. The number of publications from studies conducted on the basis of these theories is still growing steadily. Results from these studies have been generally congruent for decades:the psychological variables postulated in these theories are correlated with PA and exercise with medium effect sizes(albeit with the caveat that these effect sizes may be inflated).

4.1.3. So, Where is the Problem?

The problem is that researchers in exercise psychology continue to focus on only a handful of theories, producing consistently similar results, although it is known from meta-analyses that these theories are of very limited practical value as the basis of behavior-change interventions.We believe that, half-a-century after the advent of the field of exercise psychology, it is counterproductive to gloss over this fact or to reframe PA interventions as being "effective"solely because they are significantly different from zero,and insist that they have the potential for meaningful long-term population-level changes in PA and health on a global scale(sentences to this effect can be found in all of the meta-analyses summarized here). This approach is counterproductive because, while accepting the current situation as the unavoidable status quo, the number of people who live with reduced quality of life, disability,and shortened life spans due to physical inactivity continues to increase (Guthold et al., 2018).

The alternative approach we advocate is to fundamentally question the established theoretical foundations of exercise psychology. While providing part of the answer, the few popular theories clearly do not suffice to adequately explain the reasons for physical inactivity and behavior change. They offer only a partial answer regarding what needs to be done in practice to produce meaningful and sustained behavior change.

---

References

¹ A meta-analysis is a quantitative synthesis of past studies, integrating the results from (ideally, all) empirical studies on a given subject. One of the criticisms against meta-analyses is that they may aggregate results from both well conducted and poorly conducted studies(e.g., Borenstein et al. 2009).

² In RCTs, study participants are randomly assigned to at least one treatment group and a comparison or control group. An outcome variable under investigation(e.g.,PA)is assessed before and after a treatment period (e.g., an activity-promotion program in the intervention group vs. a comparison treatment in the control group),to determine whether it changed to the same or a different extent across the different groups. If the study was conducted well and values of the outcome variable in the two groups are found to differ significantly, it is reasonable to conclude that any differential change was due to the treatment.Researchers consider RCTs to be the most powerful research design for investigating intervention effects (Bickman & Reich, 2015).

³ Statisticians like to think of measured values (e.g., exercise minutes per week) in samples of study participants as being distributed in such a way that extreme values are rare and mid-range values are much more frequent(as in a Gaussian "normal distribution").

⁴ We refer here to the so called "common language effect size" (McGraw & Wong, 1992) and its two basic assumptions: (a) that the distributions of the intervention and control groups will be normal, and (b) that the two distributions will have equal variances.

⁵ Publication bias occurs when studies with statistically significant results are more likely to appear in the published literature,whereas studies with undesirable or nonsignificant findings tend to remain unpublished. This tendency leads to an overestimation of the apparent efficacy of interventions, including an overestimation of the efficacy of theory-based PA interventions in this case (Schäfer & Schwarz, 2019).

⁶ In addition, there is the socioecological framework(Rhodes et al., 2019). Due to emphasizing mostly environmental and policy factors, this framework falls largely outside the psychological perspective discussed here.

⁷ The Web of Science is a database whose scope is limited only to approximately 12,000 highest-quality scientific journals, according to the judgment of a selection committee.

⁸ However, the association between automatic evaluations towards PA and PA behavior (i.e.,the contribution of one of the two classes of processes postulated within dual-process theories) has been analyzed.Researchershave found a "small"effect size (Chevanceet al., 2019).