This review followed the initial guidelines delineated by Arksey and O’Malley [31] and the recent Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Extension for Scoping Reviews (PRISMA-ScR) [32]. A description of the protocol for this review is registered with the Open Science Framework.

This scoping review was guided by the following a priori research questions:

How effective social comparison features of apps are in changing PA behavior is also an important question. It is not included in the preceding list because we did not find any randomized controlled trials, narrative reviews, meta-analyses, or dismantling studies focused on social comparison app features or directly comparing the effects of different app features. We elaborate on this point in the Discussion section.

Inclusion and exclusion criteria were chosen by the first and last authors (DA and JS). Review articles were eligible for inclusion if they met the following criteria: (1) available in English; (2) published on or before May 31, 2019; (3) conducted a systematic or narrative review, or meta-analysis; (4) reviewed the features of commercially available smartphone apps or included formal intervention programs delivered via smartphone apps; and (5) used increasing PA or reducing sedentary time as a key behavioral outcome. An initial examination of the literature revealed that many reviews in the domain of mHealth combine PA with related weight control outcomes. Consequently, reviews that met the first 4 criteria and used weight loss or PA plus other behaviors (eg, diet and weight loss) as outcomes were included.

Reviews were excluded if they considered interventions that combined an app modality with other modalities (eg, websites, text messages, etc) because they might obscure conclusions specific to apps. Apps geared toward particular medical populations also were excluded because these interventions tend to promote multiple behavior changes and set illness-specific PA targets, rather than focusing on broad-based PA increases. Finally, apps involving gamified interventions were excluded because they typically engage a variety of social processes, in addition to social comparison, to create competition with other users or teams of users. Dissecting comparison features from features intended to induce other processes in this context can be very challenging.

We searched PubMed, PsycINFO, and Web of Science for publications related to the use of smartphone apps for increasing PA. Search terms were combinations of “physical activity” or “exercise” and “smartphone app(lication),” “mobile app(lication),” or “mHealth.” Resulting titles and abstracts were reviewed to determine relevance to our 6 research questions. Initial database and hand searches returned 3743 individual articles of which 2247 were duplicates, leaving 1496 unique articles. A PRISMA-ScR flowchart, shown in Figure 1, details the evaluation of each article for inclusion in this review. The majority of articles that were identified described empirical studies. Initial reviews were conducted by the first 3 authors (DA, MB, and KP) who were responsible for determining inclusion for an equal subset of identified articles. Final review and inclusion decisions were made by the first author (DA). The final set of 26 review articles were coded for the characteristics described in the following section.

The first and last authors (DA and JS) determined the types of data to be extracted from each article. The second and third authors (MB and KP, respectively) each independently read and extracted the following data from the 26 reviews: authors; year of publication; review of commercially available versus researcher-developed apps (or combination); number of apps reviewed; specific behavior change outcome targeted by the app (eg, overall PA, sedentary behavior, weight loss); percentage of apps that included social comparison features; the definition of social comparison; the specific features for inducing social comparison (eg, leaderboards); the social comparison dimension (eg, steps, physical fitness); and the presence (vs absence) and type or types of social comparison tailoring. Additional data extracted included the percentage of apps categorized as modeling/demonstrating a behavior, providing normative information about others' behavior, and social networking (eg, message boards). These features are associated with the opportunity to make comparisons, even if comparison is not considered the primary BCT induced.

For reviews that explicitly categorized features based on social comparison or other types of social influence (eg, modeling), the percentages attributed to social comparison processes were taken directly from the original published review. For reviews that did not use these terms, the percentages were calculated manually by reviewing the details available in the original published review, where possible (eg, references to social networking features or exposure to information about other users). As for all other data extraction, the second and third authors (MB and KP, respectively) independently determined the percentages of apps that categorized features as inducing social comparison or other social processes. The first author (DA) then calculated the interrater agreement (91%) and independently rated a subset of included reviews to verify the accuracy; the remaining discrepancies were resolved by consensus.

Among the 26 articles reviewed, the number of apps identified as promoting PA or weight control ranged from 12 [33] to more than 28,000 [34]. Of these 26 articles, 10 (38%) focused exclusively on apps intended to increase PA and 10 (38%) focused on weight loss, weight management, or obesity intervention (the largest subsets; see Table 1). The remaining reviews (6/26, 23%) considered a combination of diet, PA, and/or weight control/obesity prevention. The majority of articles reviewed only commercially available apps (19/26, 73%), primarily those available through the iTunes App Store (Apple operating system). Only 27% (7/26) of reviews appeared to include apps developed or empirically tested by researchers. The popularity of these reviews appeared to increase through 2014 (peaking in 2014-2016) and then decrease through 2019.

Social comparison is known to influence motivation and health behavior and is frequently manipulated in health behavior change interventions [9]. Comparison processes may be particularly useful for promoting PA with technology such as smartphone apps; objective measures of PA can be visualized and shared between app users, and users can see evidence of change in their relative standing by increasing their PA behavior over short time frames. Despite the interest in social comparison as a motivator of PA change and the exponential increase in publications about digital health interventions [59], no review to date has attempted to summarize existing literature on the social comparison features of PA apps. We undertook the present scoping review to address this gap and provide recommendations for future research in this area.

A modest proportion of the 26 available and eligible reviews of PA promotion apps categorized app features as eliciting social comparison (31%). Comparison fell behind modeling as a popular intervention process (54%) but was as common as social networking (38%; which also may facilitate comparison) and was more common than related processes such as norm referencing (12%). All the articles that included social comparison as a category used versions of the BCT taxonomy [9,13,54]. However, the versions differ in their definitions of social comparison. The original BCT taxonomy specifies that the potential target of comparison must be a nonexpert [9]; exposure to an expert is classified as modeling. Although modeling appeared more frequently in apps than did social comparison, the percentages of apps with features in each category differed modestly (ie, 35% vs 30%; see Table 3). Later iterations of the BCT taxonomy removed the requirement that only social comparisons with nonexperts would qualify [13,54]. Visual inspection of the percentage of apps classified as having social comparison features suggests that using the broader definition, ie, including experts, slightly increases the average proportions of apps that receive a social comparison designation (ie, 27% to 35%). The broader definition also is consistent with definitions of social comparison used in the mainstream comparison literature, where targets often include media figures or fashion models, in addition to peers [60].

Abraham and Michie’s [9] initial taxonomy also defined comparison as simply observation of another’s performance, which may occur in a variety of contexts (eg, group classes). Using this definition, PA app features such as social networking or message boards (where users can report on their performance) may count as social comparison [30,33]. In contrast, later versions explicitly state that attention must be drawn to the other’s performance and that contexts such as group classes do not necessarily induce comparison (vs other social processes) [13,54]. This definition implies that social networking and message boards would not count as social comparison, whereas leaderboards or competitions would [40,46].

The majority of reviews did not include any mention of specific dimensions of social comparison, and those that did made only vague references to dimensions (eg, comparison of behavior without specifying which behavior, eg, steps, etc). A recent meta-analysis suggests that comparison dimension provides information about the target’s relevance to the self; if relevance to the self is not clear, the individual might reflect on their target’s performance but not engage in comparative self-evaluation [61]. Owing to the many dimensions potentially relevant to PA promotion (eg, steps, calories burned, minutes of activity, and overall fitness) and the likelihood that these dimensions are not relevant for all app users [62], this review highlights the need for increased specificity in future work that describes social comparison features of apps.

As very few articles included descriptions of the specific features eliciting comparison, the exact degree of heterogeneity is unclear. What can be concluded is that existing reviews of PA apps show considerable variability in their approaches to defining and classifying social comparison. Specifically, comparison, modeling, and information sharing are not consistently differentiated. The heterogeneity associated with which features activate social comparison represents a challenge for future research to evaluate the unique effect of comparison as a mechanism of app-based behavior change, or its efficacy relative to other mechanisms [15]. Inconsistency in the definition of comparison also creates challenges for optimizing app-based interventions to address comparison preferences and needs between users, which may be either stable or dynamic. In this vein, PA app development has not yet integrated theoretical and empirical advances that the mainstream social comparison literature has made.

Interest in and responsiveness to social comparison information vary across individuals. This construct, called social comparison orientation (SCO) [63], has been positively associated with engagement in PA [64]. PA app users with strong SCOs may engage in comparison in response to a wide variety of social features in PA apps, including social networking and message boards, and they may find this information motivating. Here, comparison information is available, but the comparison process itself is not intentionally activated. In contrast, users with weaker SCOs may engage in comparison only when the comparison process is deliberately induced, such as by competitive challenges or leaderboards that display PA data ranked from most to least [65]. Social comparison features also may be ineffective for users with weaker SCOs. These hypotheses imply that PA app effectiveness might be improved by guiding users toward the types of social features that match their level of SCO or away from social comparison features at particularly low levels of SCO.

Additional variability may exist with respect to users’ social comparison preferences and their affective and behavioral responses to comparisons. As noted, users may find comparisons to targets who are doing better with respect to PA (ie, upward comparisons) either inspiring or disheartening and may find comparisons to targets who are doing worse (ie, downward comparisons) either comforting or anxiety-provoking [18,23]. Which combinations lead to the greatest increases in PA (or lead to increases vs decreases) and for whom are significant empirical questions [25,66,67]. Basic research indicates that the opportunity to select a comparison target does not always lead to optimal affective or health-relevant outcomes, nor does it always fulfill comparers’ goals (eg, to feel better) [18,68,69]. Thus, providing information about only the targets that a PA app user wants may not lead to benefits. Providing only the targets that they do not want may create an aversive experience, however, and may lead users to discontinue engagement with the app [28].

The optimal combination of comparison target and affective response for increasing PA may differ between people. The best combination may also vary within the same person over time, as a function context (eg, precomparison mood), shift over the course of behavior change (eg, as users experience progress and setbacks) [56,70], and differ from users’ stated preferences, depending on whether users are just starting with the app or have been engaged for some time. The degree of within-person variability in social comparison preference and response (either affective or behavioral) remains unclear. The quantification of within-person variability and its responsiveness to social comparison interventions (eg, using N-of-1 designs) represent important next steps for PA app development and a broader understanding of social comparison processes [71].

Despite gaps in the social comparison literature, evidence suggests that the effects of social comparison and preferences for a comparison type differ between people and within people over time. This review, however, detected no reference to between- or within-person variability in comparison response/preference or to tailoring social comparison features of PA apps. In contrast, this review indicates that tailoring in PA apps is common with respect to goals and feedback, which suggests that technology for such tailoring is currently in use. Tailoring the PA app experience to match user characteristics such as SCO or user-relevant PA comparison dimensions might improve the app’s acceptability and engagement and, in turn, enhance PA outcomes [28]. Indeed, tailoring has been shown to outperform generic messaging in PA interventions across a range of modalities, including apps [48,72]. Tailoring also might discourage negative consequences of comparison (eg, giving up in response to a failure to match another user's achievements) by matching a user’s comparison preferences with the types of comparisons that optimize engagement in PA. Such tailoring will require nuanced assessment of the effect of factors such as SCO, dimensions of relevance, comparison preferences, affective response to comparison, and PA engagement. The adaptive capabilities of many existing apps and those under development may lend themselves to such tailoring [73].

Strengths of this scoping review include its use of preregistered methods, adherence to PRISMA-ScR guidelines, and a comprehensive search for relevant reviews to provide insights into how social comparison is currently applied in existing PA apps. A subset of pertinent articles may have been overlooked, but the extensive and systematic search increases confidence in the overall conclusions. Additional app comparison features (eg, specific dimensions and tailoring) may have not been described in the reviews or missed by our coders. As a check, we examined several primary sources of empirical data and failed to find these additional details. One exception, an empirical study by Mollee and Klein [28], demonstrated PA benefits of matching (tailoring) versus not matching comparison targets to user preferences. There is need for additional work of this kind to inform best practices for tailoring social comparison features of PA apps.

Although social comparison has been shown as effective for increasing PA in other types of interventions (eg, team-based competitions) [26], there are very few studies of the effectiveness of social comparison as a mechanism of change in PA apps (eg, randomized controlled trials, meta-analyses, and dismantling studies) to answer the question of whether, for whom, or under what circumstances social comparison features of apps produce positive changes in PA. Such research is critical to advance our basic understanding of comparison processes and their utility as BCTs, as is further information about within-person variability in comparison preferences and responses. This information would inform the necessary or sufficient social comparison features of PA apps needed for a successful intervention. To what extent our findings and conclusions apply beyond PA promotion (alone or in the context of weight control) to such health behaviors as smoking cessation or skin cancer prevention [74,75] remains to be addressed in future research.

This review documents that social comparison is frequently identified as a potential mechanism of behavior change in smartphone apps designed to promote PA, on par with mechanisms such as social networking (broadly defined). Behavioral modeling, which is considered in some reviews as a means of inducing social comparison, was the only comparison-related mechanism to appear in more reviews of PA apps than social comparison (as explicitly differentiated from other processes). Our findings highlight the need for careful consideration of social processes as behavior change mechanisms in app design and evaluation. Considerable gaps currently exist between theory and evidence relevant to social comparison and its implementation in PA apps. Greater attention to individual differences, dynamic responses, relevant PA dimensions, and comparison preferences and the potential to tailor apps on the basis of these characteristics may meaningfully improve the effectiveness of existing PA promotion apps.