A Systematic Review and Meta-analysis of the Outcome Expectancy Construct in Physical Activity Research

Copyright © Society of Behavioral Medicine 2021. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.

This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Associated Data

Abstract

Background

Cognition-based theories dominate physical activity (PA) research, and many include a construct broadly defined as “beliefs about the consequences of behavior” (e.g., outcome expectancies, perceived benefits) hereafter referred to as perceived consequences.

Purpose

With the quantity of available research on this topic, it is important to examine whether the literature supports perceived consequences as a predictor of PA.

Methods

A meta-analysis examining longitudinal associations between perceived consequences and PA in adults was conducted. Studies were eligible if (a) perceived consequences were measured at a time point prior to PA, and (b) the target behavior was a form of PA. An omnibus meta-analysis estimating the mean effect of all included studies, and separate meta-analyses for perceived consequences content categories were conducted.

Results

This search yielded 6,979 articles, of these, 110 studies met inclusion criteria. Studies were published between 1989 and 2020, with sample sizes ranging from 16 to 2,824. All studies were evaluated as moderate to high quality. A small positive bivariate association was identified (r = 0.11; 95% CI [0.09, 0.13]) between perceived consequences and PA. Significant associations were identified for time, health, self-evaluative, psychological, and affective consequences. There was no association between perceived weight-related consequences and PA.

Conclusions

The findings emphasize the variability with which existing studies have examined perceived consequences in the PA literature. Future research might examine whether these are important distinctions for understanding PA. Overall, the results suggest utility in examining perceived consequences as a predictor of PA, but constructs with more robust associations may require priority.

Adults’ beliefs about the consequences of physical activity were positively associated with how much physical activity they did. However, adults’ beliefs about weight-related consequences of physical activity were not associated with their physical activity behavior

Introduction

Globally, over one-quarter of adults (27.5%) are insufficiently active and do not meet recommended physical activity (PA) guidelines [1]. Insufficient activity in adults is linked to increased risk of all-cause mortality, chronic disease, and reduced quality of life [2, 3].

Decades of health behavior change research has been dedicated to building a better understanding of PA behavior, and what can be done to increase PA. Much of this research has been framed from a cognitive perspective, with behavior posited to be a function of beliefs, including thoughts and evaluations of imagined end-states, and behavioral goals [4]. One construct common across cognitively orientated theories is perceived consequences, defined as the perceptions of what might occur as a result of a behavior, including the probability that engaging in a behavior might lead to various possible outcomes [5, 6] and has been labeled “outcome expectancies” [7], “behavioral beliefs” [8], “decisional balance” [9, 10], and “perceived benefits/barriers” [11] in the context of different health behavior theories. Although a commonly used term for this construct is “outcome expectancies,” in this study we have opted to use the umbrella term, “perceived consequences” to include outcome expectancies, as constructs from various theories of behavior change which share a general definition.

Some researchers have distinguished among subtypes of perceived consequences. For example, Bandura [12] proposed physical, social, and self-evaluative subtypes. This typology is frequently featured in the context of PA (e.g., [13]). Other subtypes of perceived consequences assessed in the PA literature include affective versus health-related [14], weight-related [15], or time-related (e.g., [16, 17]) consequences. In addition to (and independent of) content-related subtypes, perceived consequences are often characterized as either positive (beneficial) or negative (detrimental). For example, a positively valenced consequence about the health benefits of PA would be expected to predict higher levels of PA, whereas a negatively valenced consequence about the possibility of injury resulting from PA would be expected to predict lower levels of PA.

Understanding the role perceived consequences plays as a predictor of PA behavior is important for refining behavioral theories and improving the efficacy of PA interventions. However, to date, there have been few attempts to synthesize the abundance of research on perceived consequences and PA, as has been done for other cognitive variables, including behavioral intentions (e.g., [18]), self-efficacy (e.g., [19]), and attitudes and social norms (e.g., [20]).

One prior review of cross-sectional and longitudinal studies examining PA behavior from the lens of Social Cognitive Theory showed approximately one-third of studies including outcome expectancies reported a significant direct effect on PA [21]. A meta-analysis of predictors of all types of health behavior from a Health Action Process Approach perspective identified a small effect size (r = 0.15) for the zero-order correlation between outcome expectancies and health behavior [22]. Further, a meta-analysis evaluating mediators of PA interventions showed perceived consequences was a small, yet significant predictor of PA behavior change (r = 0.19 [23]). Finally, a narrative review of outcome expectancies in the PA literature [24], found mixed evidence for perceived consequences as a predictor of PA. There have, however, been no prior meta-analyses focused exclusively on the construct of perceived consequences as a predictor of PA.

The present study is a meta-analysis of perceived consequences as a predictor of PA behavior. Only longitudinal studies were included to allow for stronger inference (relative to cross-sectional studies) regarding the directionality of the association between perceived consequences and physical activity. We did not seek to study the effects of any intervention on perceived consequences, but instead sought to examine the association between (and thus potential temporal associations of) perceived consequences and PA behavior. An omnibus meta-analysis was conducted along with putative moderators, including study design (intervention vs. observational), population (clinical vs. non-clinical), underlying theoretical framework (e.g., Social Cognitive Theory), inclusion or exclusion of perceived outcome value, method of PA measurement (self-report vs. objective), time between assessment of perceived consequences and PA, and study quality. Additional separate meta-analyses were conducted for different categories of perceived consequences, including health, affective, physical, social, self-evaluative, psychological, weight, and time, and also all scales which had a negative valence.

Methods

Study Design

A meta-analysis adhering to the PRISMA standard [25, 26] examined the effect of perceived consequences of PA on PA behavior in longitudinal reports of both intervention and observational studies (Supplementary File 1).

Eligibility Criteria

To include a study: (a) the target behavior must be a form of PA; (b) perceived consequences must be measured, and the form of PA behavior referenced within the scale item(s) must be the same behavior measured in the study (e.g., if the scale measured beliefs about the consequences of a non-PA behavior such as weight loss, or dieting, it would be ineligible); (c) participants must be 18 years or older; and (d) perceived consequences must be measured at a time point before PA.

Studies were excluded if they (a) were not reported in peer-reviewed publications (i.e., book chapters, dissertations); (b) were not written in English; and (c) used a case-study or single-participant trial design (i.e., n of 1 studies). Additionally, studies were excluded if (a) instead of PA behavior a proxy for PA was measured, such as physical fitness (e.g., VO₂ Max) or PA intentions; (b) PA was represented within a mixed measure of other behaviors, such as attendance at intervention events targeting both diet and PA behavior; or a measure focused on physical therapy or physical rehabilitation attendance; (c) measures which conflated PA behavior with PA intentions (i.e., stages of change for PA); or (d) assessments of perceived consequences featured outcomes related to PA, but not PA behavior itself (i.e., beliefs about the consequences of weight loss).

Search Strategy

Literature searches were completed in May 2020 in seven databases: (a) PubMed, (b) PsycInfo, (c) CINAHL, (d) EMBASE, (e) ISI Web of Knowledge, (f) PsycArticles, and (g) Social SciSearch. The search strategy was developed by all study authors and executed by one author (LCB). The search was not restricted by any parameters. The search terms included construct names from behavior change theories which were categorized in a prior line of research as being the “same as” or “indistinguishably similar to” the construct “beliefs about consequences” [5, 27] as defined above as “the perceptions of what might occur as a result of a behavior, including the probability that engaging in a behavior might lead to a set outcome,” and terms associated with PA similar to previous reviews in the PA literature [19, 23]. The full search string is in Supplementary File 2.

Screening

Citations from the search were screened for duplicates in EndNote and uploaded into Covidence [28]. One reviewer (LCB) reviewed titles and abstracts using pre-defined inclusion criteria and screened for duplicates not identified by EndNote. To reduce the likelihood of excluding a relevant study before full-text review, during title and abstract screening, studies were removed according to more restrictive criteria: (a) study design irrelevant (e.g., cross-sectional design, or case study); (b) sample irrelevant (e.g., non-human or pediatric populations); (c) not in English; (d) review paper (e.g., systematic review); (e) non-peer reviewed (e.g., a dissertation). Two reviewers (LCB and JAE) performed full-text screening to determine if a study met all inclusion criteria. Interrater reliability was moderately good, with 87.5% agreement, κ = 0.73. Unresolved were reviewed by DMW and RER, such that consensus was reached for 100% of articles screened.

Data Extraction

Data from included studies were extracted using a pre-specified data extraction form and coding guidelines (Supplementary File 2). The following study characteristics were extracted (where available) from all articles: authors and publication year, sample size, effect size (unadjusted correlation coefficient, or t-value), and author label for perceived consequences. Additional characteristics of the assessment of perceived consequences were also extracted for each study: measure name (often different from the construct label), whether the measure was previously used (yes or modified from previous version, no), item stem (e.g., “If I exercise regularly, then…”), an example item, the response scale, the number of items on the scale, valence of the scale or subscales (positive vs. negative), and content category of the scale or subscales (e.g., physical, social, self-evaluative). To determine the content category of the perceived consequences, primary emphasis was placed on the label(s) provided by study authors, followed by an examination of the items. To be included in the final analysis, a content category needed to appear in more than five study samples; these included content categories where the beliefs about the consequences of PA were either: affective (e.g., improve my mood), health (e.g., reduce my chances of heart disease), physical (e.g., make me more toned), psychological (e.g., improve my mental well-being), social (e.g., improve my social life), self-evaluative (e.g., improve my self-esteem), time (e.g., take up too much time), and weight (e.g., help me control my weight). Content categories which appeared in five or fewer study samples were not included in the content category analyses. Also not included in the content category analyses were measures of perceived consequences that assessed an array of content categories within a single scale and thus labeled as undifferentiated [29]. All of the scales were included in the omnibus test, even if they were not included in a separate content category analysis.

We also extracted the following a priori putative moderators: study design (intervention vs. observational), population (clinical vs. non-clinical), underlying theoretical framework, inclusion of perceived outcome value (expectancy only, expectancy × value, expectancy−value combination), PA method of measurement (self-report, objective, other), time between perceived consequences and PA (number of weeks), and study quality. Study quality was assessed with a 10-item measure based on similar measures of study quality and risk of bias in previously published systematic reviews and meta-analyses of theoretical predictors of PA behavior [30–32]. All questions were answered on a yes-no basis, with “yes” answers indicating good study quality. Each “yes” counted as “1” and each “no” counted as a “0.” A study was considered low quality if it scored below a 3 on the scale, scores between 3 and 7 were considered moderate, and scores 8–10 were consider high quality, with the intention to exclude any studies from the review with low quality [31, 32]. A detailed description of the study quality criteria is listed in Supplementary File 3.

Data Analysis

Data were analyzed with Comprehensive Meta Analysis software using random effects models for correlation values. The set of studies investigated does not meet the assumptions for a fixed effects model [33]. We conducted multiple meta-analyses including: (a) an omnibus meta-analysis of all studies, with negatively valenced scales/subscales reverse-scored, and examination of previously specified putative moderators; (b) separate meta-analyses for each perceived consequences content category; and (c) a meta-analysis including all negatively valenced scales. Intervention and observational studies were included together in the omnibus meta-analysis in order to obtain an overall estimate of the association between perceived consequences and PA regardless of study design and assessment characteristics. We then conducted more focused moderator analyses or separate meta-analyses to examine differences in the perceived consequences−PA relationship across various design and assessment factors. This approach allowed us to obtain both general and more focused estimates of the association between perceived consequences and PA.

Intervention and observational studies were included together in the analyses as the goal of this study was to understand associations between perceived consequences and PA, rather than the effect of any intervention to improve PA which included perceived consequences as a baseline measure. To optimize integration with the observational studies, unadjusted correlation coefficients were the primary target for data extraction. If a study provided a correlation coefficient or t-value which included all participants (control and intervention), this was included instead of separate analyses by intervention condition. More information is listed in Supplementary File 2. All analyses were conducted at the level of the study sample rather than the study, as several studies included separate analyses for independent samples of participants (e.g., intervention vs. control groups). If the same study sample contributed multiple scales, and, therefore, multiple effect sizes to an analysis, the mean of the effects within sample was computed and the combined score was used in the omnibus meta-analysis [34]. However, because many of the study samples contributed multiple perceived consequences content categories, therefore violating necessary assumptions of independence for a moderator analysis, we conducted separate analyses for each content category. Due to small numbers of studies for each of the content categories, moderator analyses were not conducted for the meta-analyses of the perceived consequences content categories. For all studies, the baseline value of perceived consequences and the most temporally distal PA time point (i.e., final follow-up) were extracted. We opted to include the most distal PA time point in order to draw strength from the methodological rigor of the longitudinal designs used in the studies reviewed. In cases where both an objective and self-report measure of PA were reported, the analysis includes the objective measure.

For each meta-analysis, a 95% confidence interval was calculated. Two statistics were computed to assess dispersion across studies: the Q-statistic and the I 2 statistic. A significant Q-statistic falsifies the null hypothesis that the perceived consequences−PA correlation is the same across all studies. The I 2 statistic provides an index of the proportion of true variance in the observed effect sizes across studies, rather than variance due to sampling error, with higher values reflecting more true variance. Publication bias was assessed using Rosenthal’s classic fail-safe N [35] and Duval and Tweedie’s trim and fill procedures [36].

Results

The literature search yielded 6,979 potentially relevant records. After removal of duplicates, and title and abstract screening, the full text of 1,035 records were assessed. A total of 90 studies (k) with 96 independent samples (k′) met the inclusion criteria, and were included in the meta-analysis. An additional 114 studies met inclusion criteria but did not contain all relevant data. Of these, we successfully obtained additional data from the authors of 20 studies (with 27 independent samples), thus yielding a total of 110 studies, and 123 independent samples in the meta-analysis. Further detail about study selection can be found in Fig. 1 . See Supplementary File 4 for all extracted data, and Supplementary File 5 for a reference list of all included studies.

PRISMA chart detailing the study inclusion processes.

Study Characteristics

Included studies were published between 1989 and 2020. Sample sizes ranged from n = 16 to n = 2,824. Study authors used 21 different terms to describe perceived consequences, with “outcome expectations” the most frequently used term, followed by “outcome expectancies,” “decisional balance,” “response efficacy,” and “behavioral beliefs” (Supplementary File 6). Of the 110 included studies, 60 were intervention studies and 50 were observational studies. The majority of studies assessed non-clinical populations (k = 72). Most studies referred to a theory (k = 104), with Social Cognitive Theory referred to most often (k = 51), followed by the Health Action Process Approach (k =12). Most studies did not include a measure of perceived outcome value (k = 92). PA behavior was most frequently measured using self-report methods (k = 79). The length of time between the measure of perceived consequences and PA behavior ranged from two hours to four years, with a median of 12 weeks and a mean of 20.2 weeks. On average study quality was high (mean = 8.02), with no studies considered low quality (range = 4–10). There were 32 studies identified in the search which were published between 1970 and 1988, however none of these studies met the inclusion criteria. For additional details about study characteristics, see Table 1 .

Table 1.

Note. k = number of studies, For “Perceived consequences survey measure,” measures “designed by study team” are those for which the authors created a new measure for the study, or did not report that the measure was based on an existing measure of perceived consequences. For all other measures, authors might have modified the existing measure reported. PA physical activity.

Characteristics of the Perceived Consequences Measure

Across the 110 included studies, 75 used a previously established or modified version of a previously established measure. The most often used pre-existing measures were: the Outcome Expectations for Exercise Scale (k = 11 [37]), Decisional Balance (k = 10 [38]), the Multidimensional Outcome Expectations for Exercise Scale (k = 10 [13]), the Health Beliefs Survey (k = 4 [39]), and the Expected Outcomes and Barriers for Habitual Physical Activity scale (k = 4 [40]). The item stem used in each study varied greatly, with 51 unique item stems extracted. The most frequently identified item stem was “Exercise will…” (k = 13). The number of items used to measure perceived consequences on either the scale or sub-scale level varied (mean = 5.92), with some sub-scales within studies using a single item to measure perceived consequences (k = 11), and one scale using as many as 43 items [42]. Nearly all studies (k = 109) assessed positively valenced perceived consequences, including a majority of studies which only assessed positively valenced consequences (k = 86). There were 23 studies which included at least one positively valenced, and also one negatively valenced scale. One study only assessed negatively valenced consequences [43].

Omnibus Test

There was a small significant effect of perceived consequences on subsequent PA behavior (r = 0.11; 95% CI [0.09, 0.13]), with a range of r = 0.00 to r = 0.50. Details about individual study correlations, and sample sizes are in Fig. 2 . There was significant heterogeneity across studies (Q(109)=309.2, p < .001), with 65% of the variance observed across studies explained by true systematic differences in effect sizes rather than sampling error (I 2 = 64.8). Rosenthal’s (1979) classic fail-safe N was 8,182, suggesting a much larger number of studies with null findings would need to be identified to potentially nullify the findings. The estimate of the unbiased effect size, using Duval and Tweedie’s [36] trim and fill, suggests six studies are potentially “missing” from the analysis, and that the trim and fill imputed effect size is smaller than the observed effect size (r = 0.10, 95% CI [0.09, 0.11]). A funnel plot depicting these findings is in Supplementary File 7.

Forest plot for the omnibus meta-analysis test. In the content category column, “Combined” refers to studies which had multiple scales to measure perceived consequences, and in the omnibus meta-analysis, the mean effect across these scales was used in the calculation of the point estimate. The label “+ Undifferentiated” refers to perceived consequences scales which were undifferentiated in their content (i.e., referring to an array of different categories of consequences), and positively valenced. Some studies appear as more than one row, these are studies which contributed more than one independent sample to the analysis (e.g., an intervention vs. control group). Each square in the figure represents a study’s individual contributing correlation coefficient to the meta-analysis, with the bars depicting the confidence interval for that study. The correlation coefficients are plotted on a −1.0 to +1.0 horizontal axis. The mean effect is indicated by a diamond at the bottom of the plot.

Moderator Analyses

Moderator analyses were conducted in an attempt to understand the heterogeneity in effects across studies with respect to study design, population, underlying theoretical framework, inclusion of perceived outcome value, PA method of measurement, time between the assessment of perceived consequences and PA, and study quality. To be considered as a putative moderator, a minimum of five studies were required in each moderator variable category [34]. No moderator variables were excluded based on this criterion; however, for the moderator “underlying theoretical framework,” some theories appeared infrequently (k < 5), and were collapsed into a level labeled “other.” The only variable that significantly moderated the main effect of perceived consequences on PA behavior was whether the study design was an intervention or observational study (p < .01), with a slightly larger effect found for observational (r = 0.14) compared with intervention studies (r = 0.08). The latter findings were unchanged regardless of whether control group samples within the intervention studies were treated as “observational” study samples or “intervention” study samples. The moderator variable for the time between the assessment of perceived consequences and PA was assessed as both a categorical and a continuous variable. The results were non-significant for both analyses, and therefore we presented the results of the categorical variable analysis to be consistent with the presentation of the other moderator analyses. All other putative moderators tested were non-significant ( Table 2 ).

Table 2.

Moderator analyses of the omnibus test of perceived consequences and physical activity

Note. k′, number of independent samples.

*Significant moderating variable.

Perceived Consequences Content Category

There were more than 10 different perceived consequences content categories, including health-related, physical, social, self-evaluative, affective, psychological, weight, and time-related. The largest pooled mean effect size was for time, for which all scale items were negatively valenced (e.g., “exercise would interfere with my other commitments”; k′ = 7; r = −0.22; 95% CI [−0.39, −0.03]). Small positive effect sizes were identified for psychological consequences (k′ = 11; r = 0.14; 95% CI [0.04, 0.23]), affective consequences (k′ = 14; r = 0.13; 95% CI [0.06, 0.19]), health consequences (k′ = 27; r = 0.11; 95% CI [0.07, 0.15]), and self-evaluative consequences (k′ = 16; r = 0.07; 95% CI [0.01, 0.12]). Nine samples assessed consequences of PA such as energy, safety, injury, and pride, which were not assessed frequently enough (k′ 5) to examine their independent effect. Three categories of consequences had non-significant associations with subsequent PA: physical consequences (k′ = 20; r = 0.03; 95% CI [−0.04, 0.10]), social consequences, (k′ = 16; r = 0.09; 95% CI [−0.003, 0.16]), and weight and/or weight loss consequences (k′ = 8; r = −0.03; 95% CI [−0.11, 0.04]). Despite the negative association, all scale items for the weight content category were positively valenced, for example, “physical activity.. will help me to lose weight.” The largest number of perceived consequences scales were “undifferentiated” (i.e., there was a broad range of content within a single scale; k′ = 74) and were not analyzed as a separate content category.

Negatively Valenced Perceived Consequences Scales

Orthogonal to the perceived consequences content categories, negatively valenced scales (k′ = 26; as reported above, 7 of these were from the time content category) showed a small negative association with future PA (r = −0.12, 95% CI [−0.05, −0.18]).

Discussion

This meta-analysis examined longitudinal associations between perceived consequences and PA in adults, across observational and intervention studies. In the omnibus meta-analysis, a small positive association was identified across 123 included samples from 110 studies. When looking to understand or change PA behavior, these findings suggest a focus on perceived consequences of PA may have reliable but modest utility. Findings are comparable to other reviews of the association between perceived consequences and PA [21, 23, 24].

These findings are comparable to meta-analyses of other cognitively orientated constructs in the PA literature. Previous meta-analyses on self-efficacy [19] and intention [18] identified small, significant associations with PA. Point estimates from a meta-analysis of Theory of Planned Behavior constructs (i.e., intentions, perceived behavioral control, attitudes, subjective norms) were either comparable, or larger (r = 0.18−0.45) than those in the present study [44]. The present findings are smaller in magnitude than the effects identified in meta-analyses of constructs associated with PA which are not commonly found in cognitively orientated theories such as habit (r = 0.46 [45]) and identity (r = 0.44 [46]).

Only one significant moderator of the omnibus findings was identified. The association between perceived consequences and PA was found to be significantly weaker in the intervention studies than the observational studies. The weaker association in intervention studies could be a result of unaccounted for intervention effects on perceived consequences and their potential effects on PA behavior. Indeed, research evaluating perceived consequences as a mediator of PA behavior change interventions supports the overall findings of this meta-analysis, that the association between perceived consequences and PA is significant, but small [23]. In addition, previous findings suggest that those who participate in PA interventions often have high levels of motivation to be physically active at baseline (even if PA levels are low), as opposed to a wider range in motivation among those who enroll in observational studies. Thus, there may be restricted variability in perceived consequences in intervention studies relative to observational studies thus attenuating prediction of future PA behavior [47].

We included both intervention and observational studies in our omnibus meta-analysis since the association of interest was between baseline perceived consequences and future PA behavior—an association that could be examined in the context of either intervention or observational studies. While PA behavior may have been more likely to change across time in intervention studies, we felt it was best to include both types of studies in the omnibus test and then examine study design as a moderator.

The null results for all other a priori moderators suggest the association between perceived consequences and PA behavior is similar across populations, underlying theoretical frameworks, inclusion of perceived outcome value, PA method of measurement, time between perceived consequences and PA, and study quality. Overall, perceived consequences are extremely robust to the study method effects evaluated in this meta-analysis.

There were no significant differences in the association between perceived consequences and PA when examined in clinical (r =0.14), compared with non-clinical (r = 0.10) populations. In the present meta-analysis, the clinical populations identified included cancer-patient populations, those with diabetes-related conditions, arthritis, multiple sclerosis, cardiovascular disease, and pregnancy. Some of the included clinical study samples assessed perceived consequences which were specific to the clinical population, such as, “reduce my chances of getting cancer,” “reduce my chances of getting another stroke,” and “exercise will improve my bones,” in an osteoarthritis sample. However, the assessment of perceived consequences was not always aligned with the clinical condition being studied. For example, the Multidimensional Outcome Expectations for Exercise Scale, [13], which includes items relevant to certain clinical populations (i.e., “exercise will improve the functioning of my cardiovascular system”), and items not necessarily of clinical relevance (i.e., “exercise will improve my social standing”), was used in a study of prostate cancer survivors [48], and two studies of patients with multiple sclerosis [49, 50]. It is unclear in this analysis the extent to which the salience of a clinical condition is associated with clinically relevant consequences of PA. Previous research on disease-specific PA consequences supports the present findings, and suggests a person’s clinical status does not influence a person’s perceptions of the consequences of PA, but rather perceived consequences of PA are consistent across different clinical populations [51].

The lack of variability in findings across different theories is not surprising given that each of the theories defines the perceived consequences construct similarly [5]. Moreover, the reliable but small effect of perceived consequences on PA is comparable to reviews of perceived consequences and PA conducted from a specific theoretical lens, such as Social Cognitive Theory [21], and Health Action Process Approach [22].

The structure of some perceived consequences measures inherently prompt a person to consider both the likelihood and the perceived importance or value of the outcome (e.g., “A major benefit of PA for me is…” [52]). Other theoretical conceptualizations [53, 54] distinguish and assess outcome likelihood and outcome value as separate constructs, theorizing additive or multiplicative effects. The present findings did not suggest a predictive advantage to incorporation of outcome value in assessment of perceived consequences, which is consistent with findings from a prior narrative review [24], and direct tests of including an outcome value component [55, 56].

The association between perceived consequences and PA was not significantly different if PA was measured by self-report (r = 0.12) versus objective measures (r = 0.10) such as accelerometry. Although some studies identify differences in physical activity behavior rates by measurement type [57, 58], the results suggest the association between perceived consequences and PA does not vary systematically with how PA is measured. Therefore, it does not appear that any possible inflation of self-reported PA is associated with how people self-report their beliefs about the consequences of PA. The association between perceived consequences and PA tracks in the same general direction and scale, even if the overall magnitude of PA captured by objective measures might be different from self-reported PA. This is similar to other reviews on various PA constructs that have shown differences in the associations with different types of PA measurement [23].

In the present review, there was substantial heterogeneity in the amount of time between the measures of perceived consequences and PA behavior, ranging from a matter of hours, to four years. The amount of time between assessments did not moderate the omnibus perceived consequences−PA effect. Only two studies reviewed here used ecological momentary assessment (EMA) methods [59, 60], which allow for more frequent testing of the perceived consequences−PA association. Thus, the present analyses do not permit an evaluation of whether the perceived consequences-PA association is different in a within-subjects context. The two studies which used EMA included between-subjects effects of perceived consequences and PA, which were included in the present meta-analyses. We chose to include the between-subjects effects to be consistent with the other included effects in the meta-analyses. There were not enough within-subjects effects identified in the search to conduct a sub-analysis of these findings.

The association between perceived consequences and PA was not moderated by study quality, perhaps because nearly all of the included studies were of at least mid-range quality (i.e., greater than five on a 10-point scale), with average study quality on the higher end of the study quality scale (mean = 8.02; see Supplementary File 8 for study quality data).

In addition to the omnibus meta-analysis and examination of putative moderators, we conducted separate meta-analyses for specific content categories of perceived consequences of PA. The largest association between a specific content category and PA behavior was for perceived consequences about time (r = −0.22). All of the time scales were negatively valenced perceived consequences. For example, many studies asked participants to rate the extent to which being physically active would “take up too much time.” Time-related consequences are immediate and thus an often-valued consequence relative to other possible outcomes of PA, such as health (r = 0.11, e.g., “reduce chances of disease”), or self-evaluative consequences (r = 0.07, e.g., “improve my self-esteem”), for which the associations with PA behavior were notably smaller. Indeed, many of the health-related benefits of PA are not realized until far into the future, and improvements to one’s self-esteem may not feel as perceptible or as apparent as interfering with time commitments.

Time is also one of the most commonly cited barriers to PA [61]. The distinction between time as a barrier to present PA behavior, versus time as a consequence of future PA has potential implications for interventions. When time is conceptualized as a barrier to current PA, this can be attributed to issues of goal conflict, motivation, or excuses [62], which makes selecting targeted intervention technique to increase PA difficult. However, considering time as perceived consequence, an evaluation of beliefs about hypothetical future behavior, extricates motivation and excuses, and permits a shift towards thinking about goal priorities which are also future-focused [56, 63]. Previous research identified perceived time-related beliefs as a more direct predictor of walking behavior than beliefs about time as a barrier to PA [56]. The present findings underscore the importance of future interventions addressing perceived consequences about time in order to increase PA. Future research could consider how targeting goal priorities, which are shown to positively predict health behavior change over and above interventions to address goal conflict [63], might help address time-related perceived consequences, and thus increase PA.

We found small positive associations between affective (r = 0.13) and psychological (r = 0.14) consequences and PA. These findings may also reflect the more immediate temporal proximity of affective outcomes to PA behavior [14], relative to other more distal outcomes (e.g., lose weight, and get fit). For these more distal outcomes, there was almost no association with whether or not a person believes that PA will result in either physical (r = 0.03) changes (e.g., “exercise will strengthen my bones” [13]), or weight-related (r = −0.03) changes (e.g., “exercise will help me lose weight” [17]) and PA behavior. From an applied perspective, the null findings for physical and weight-related consequences are important to consider. Popular media tends to over-accentuate the importance of exercise for weight loss and other physical appearance-focused consequences [64]. Exercise-based weight loss programs also continue to be a top trend in the fitness industry [65]. Not only does this conflict with research on the efficacy of PA as a stand-alone mechanism for weight loss [66], but the lack of association between weight-related beliefs about PA and subsequent PA behavior further diminishes the value of emphasizing the weight loss benefits of PA. Highlighting other consequences of PA, such as the affective or psychological benefits, might be more relevant for promoting PA participation.

These findings emphasize the importance of studying specific content categories of perceived consequences. In part, studying specific content categories would facilitate disentangling the motivating features of a perceived consequence which may be conflated with temporal proximity of the outcome. For example, while losing weight or improving one’s health may be highly valuable, the near-term realization that a person does not have enough time to be physically active today, may take precedence in making decisions about PA behavior. Further, while some of the present findings for specific content categories of perceived consequences are small, or null, these findings all report on between-person associations of perceived consequences and PA. An open question remains as to whether there are within-person associations such that an individual’s increases in, for example, perceived physical consequences, leads to increases in PA over time.

The largest number of studies in this meta-analysis were classified as having undifferentiated perceived consequences content and included items from multiple other content categories, such as health, affect, or time-related consequences. This is similar to a prior review, which identified 71 unique representations of outcome expectancies, many of which were labeled as “various” by that study’s authors [21]. While inclusion of perceived consequences items from different content categories within a single measure is not a methodological flaw per se, it serves to obscure our understanding of the association between specific content categories of perceived consequences and PA behavior. In a review of elicitation studies on behavioral beliefs about PA, several types of beliefs emerged as themes which were frequently reported as perceived advantages to PA. The most frequently reported advantages of PA closely mirror content categories identified in the present study (e.g., physical, psychological, health, weight, social, affective [67]). Researchers might consider using these classifications in subsequent research, which were frequently identified in the literature in this meta-analysis, and synthesized from several studies of open-ended responses from participants describing the advantages of PA.

Limitations

A limitation for meta-analyses in general is that study findings are often not reported in a way that permits data synthesis, and/or studies may not report findings on certain associations due to non-significant findings. For this review, a large number of studies (k = 114) report measuring perceived consequences and PA, but data on the association between perceived consequences and PA could not be extracted, and authors (k = 94) did not respond to our attempts to retrieve the data. Given the results of the publication bias analyses, we do not anticipate that the addition of these studies would nullify the effects, although they could alter the magnitude of the point estimate. Another potential limitation is the string of search terms used. It is possible that we have not identified all possible phrases for describing perceived consequences, or that studies for which the primary focus was on other theoretical constructs may not have included perceived consequences as a keyword, so may not have been discovered in the search. The former limitation should be mitigated by the inclusion of all frequently used labels for perceived consequences previously systematically identified in prior research [5, 27]. Further, in this study we only included the published literature on perceived consequences and PA. Despite limited evidence of publication bias, it’s possible that the grey literature might yield different findings. Lastly, a study protocol for this review was not publicly pre-registered; however, the study procedures were agreed upon by all study authors prior to the screening stage of the study. All study procedures have been described in detail such that the search could be replicated by another research team.

Future Directions

The present findings do not negate the importance of evaluating the role of perceived consequences when seeking to understand PA behavior in future research. Rather, these findings highlight the substantial heterogeneity with which perceived consequences have been studied and operationalized in the existing literature. Additional research might seek to systematically evaluate some of the nuances associated with a person’s perceptions of the consequences of future PA behavior, and whether these subtleties are important for understanding and predicting PA. For example, it may be more beneficial to focus on specific beliefs, rather than an array of undifferentiated beliefs when seeking to understand PA in specific populations. Future research might also consider whether PA interventions seeking to change perceived consequences use the same or similar behavior change techniques (e.g., Michie et al. 2013), and whether these behavior change techniques produce changes in perceived consequences which are associated with PA behavior. When comparing these findings to meta-analyses of similar, and non-similar theoretical constructs, it seems it may be important to consider how the cumulative evidence for different theoretical perspectives, and theoretical constructs should inform the direction of future theory-based physical activity research.

Conclusions

The findings from this study point to a small association between perceived consequences (i.e., beliefs about consequences) and subsequent PA behavior across 110 studies conducted among adults. Study design (observational vs. intervention) significantly moderated the effects, with a smaller average effect identified in intervention studies. The findings report wide-ranging variability in how perceived consequences were studied in the PA literature identified, including a variety of construct names, and measures used. Small associations were identified for time, affect, psychological, and self-evaluative consequences. Notably, no association was identified with either physical or weight-related consequences and PA. The findings underscore the importance of enhanced reporting quality in the literature in order to ensure both transparency and replicability of study findings. Likewise, using shared language, and shared operationalizations of theoretical constructs, serves to advance behavior change theory for future theory evaluation and refinement.