Article Text

Financial incentives for physical activity in adults: systematic review and meta-analysis
1. Marc S Mitchell1,
2. Stephanie L Orstad2,
3. Aviroop Biswas3,
4. Paul I Oh4,
5. Melanie Jay5,
6. Maureen T Pakosh4,
7. Guy Faulkner6
1. 1 Faculty of Health Sciences, School of Kinesiology, Western University, London, Ontario, Canada
2. 2 Department of Medicine, Division of General Internal Medicine and Clinical Innovation, New York University School of Medicine, New York, New York, USA
3. 3 Institute for Work and Health, Toronto, Ontario, Canada
4. 4 Cardiovascular Prevention and Rehabilitation Program, University Health Network, Toronto, Ontario, Canada
5. 5 Departments of Medicine and Population Health, New York University School of Medicine, New York, New York, USA
6. 6 Faculty of Education, School of Kinesiology, University of British Columbia, Vancouver, British Columbia, Canada
1. Correspondence to Dr Marc S Mitchell, Faculty of Health Sciences, School of Kinesiology, Western University, London, ON N6A 5B9, Canada; marc.mitchell{at}uwo.ca

## Abstract

Objective The use of financial incentives to promote physical activity (PA) has grown in popularity due in part to technological advances that make it easier to track and reward PA. The purpose of this study was to update the evidence on the effects of incentives on PA in adults.

Data sources Medline, PubMed, Embase, PsychINFO, CCTR, CINAHL and COCH.

Eligibility criteria Randomised controlled trials (RCT) published between 2012 and May 2018 examining the impact of incentives on PA.

Design A simple count of studies with positive and null effects (‘vote counting’) was conducted. Random-effects meta-analyses were also undertaken for studies reporting steps per day for intervention and post-intervention periods.

Results 23 studies involving 6074 participants were included (64.42% female, mean age = 41.20 years). 20 out of 22 studies reported positive intervention effects and four out of 18 reported post-intervention (after incentives withdrawn) benefits. Among the 12 of 23 studies included in the meta-analysis, incentives were associated with increased mean daily step counts during the intervention period (pooled mean difference (MD), 607.1; 95% CI: 422.1 to 792.1). Among the nine of 12 studies with post-intervention daily step count data incentives were associated with increased mean daily step counts (pooled MD, 513.8; 95% CI:312.7 to 714.9).

Conclusion Demonstrating rising interest in financial incentives, 23 RCTs were identified. Modest incentives (1.40 US/day) increased PA for interventions of short and long durations and after incentives were removed, though post-intervention ‘vote counting’ and pooled results did not align. Nonetheless, and contrary to what has been previously reported, these findings suggest a short-term incentive ‘dose’ may promote sustained PA. • financial incentives • behavioural economics • physical activity • wearable devices • mhealth ## Statistics from Altmetric.com ## Request Permissions If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways. ## Introduction There is a clear dose–response relationship between physical activity (PA) and health with the greatest health benefits seen in physically inactive individuals who become more physically active.1–3 Routine PA, for example, contributes to the prevention of several chronic conditions such as type 2 diabetes2 and depression.4 It is widely recommended that for substantial health benefits adults participate in 150 min of moderate-intensity PA, or 75 min of vigorous-intensity activity, per week.5 Less strenuous light-intensity PA (not ‘huffing and puffing’) can also confer health benefits.6 Furthermore, light-intensity PA such as walking may be more attainable and more likely to be sustained on a population scale. Yet the average US adult accumulates only about half the recommended 10 000 daily steps.7 8 This ‘physical inactivity pandemic’ in the USA, and globally,9 carries a massive financial burden. Conservative estimates suggest physical inactivity costs the global economy53.8 billion US per year in direct healthcare expenses.10 Increasing population-level PA, therefore, is an important global public health priority.11

Behavioural economics, a branch of economics complimented with insights from psychology, has stimulated interest in using financial incentives to promote PA.12 Behavioural economics has shown that systematic errors in thinking, called ‘decision biases’, can lead to poor health outcomes.12 The ‘present bias’ is a relevant example when thinking about incentives. Sometimes referred to as ‘temporal discounting’, present bias describes how a person’s value of a reward (eg, better health) decreases the further away in time the reward is realised.13 Put another way, people tend to respond more to the immediate costs and benefits of their actions than to those experienced in the future.14 In the case of PA, the cost of the behaviour (eg, time, discomfort) is usually experienced in the present and thus overvalued while the benefits (eg, health, longevity) are often delayed and thus discounted, tipping daily decisional scales towards inactivity. According to behavioural economics, immediate incentives may be useful in emphasising a short-term PA benefit and motivate more people to be active. Maintaining fidelity to the ‘present bias’ in designing incentive interventions (ie, not delaying incentives) may increase intervention efficacy.15 16 Applying a broader range of behavioural economics concepts in the design of incentive programmes may boost intervention effects and reduce reward costs as well.16 According to a meta-analysis by Haff et al (2015) and a mapping review by McGill et al 17 (2018) contemporary incentive designs that leverage peoples’ ‘decision biases’ may improve the efficacy of incentive interventions.17 18

While behavioural economics has influenced the field of financial health incentives, recent technological advances have also made it easier to track and reward PA. For instance, in 2014, Apple introduced the iOS Health Kit app which translates smartphone accelerometer data into consumer-friendly health information (eg, steps per day).19 In the past, incentives in clinical trials were usually tied to gym attendance.20 Now incentives are often contingent on a wider array of PA outcomes (eg, steps per day, minutes of PA) measured by smartphone technology or wearable activity trackers.21 High rates of smartphone (8 in 10 people)22 and wearable device (1 in 10 people) use in the USA,23 for example, offer researchers and public health professionals unprecedented access to instantaneous PA data. These data can be used to set and re-set personalised PA goals, connect users with others, reward daily achievements and so on. This new ability to track and reward PA lends itself to population-level interventions where walking and other activities of daily living are a focus rather than targeting more structured, less accessible, and therefore more difficult to achieve exercise behaviours. The Carrot Rewards smartphone application (‘app’), developed in partnership with the Public Health Agency of Canada, is a recent example of incentives tied to smartphone-assessed step counts.24 Sweatcoin is another popular app that converts step counts into financial rewards.25

Despite the potential of incentives for promoting PA, many gaps in the literature remain. The best effect-estimate from the first known incentives-for-exercise meta-analysis in 2013 determined that incentives increased exercise session attendance, the most common PA outcome at the time, by 11.6%.20 Yet, few of the reviewed studies examined incentives over longer periods (≥6 months; n=1) or post-intervention (after incentives discontinued; n=3) to inform long-term or sustained effects. Only one study rewarded PA assessed by a wearable tracker. Subsequent reviews published between 2014 and 2017 have generally corroborated the 2013 review by suggesting that incentives stimulate but do not necessarily sustain PA.18 26–28 According to AMSTAR 2 criteria though,29 these reviews are not as rigorous.30 Notably, they may have omitted eligible studies26 27 and did not quantitatively pool data in a meta-analysis.18 28 31 Furthermore, no review to date has sought to disentangle the heterogeneity between studies through subgroup analyses.

We conducted this review to address several gaps in our understanding of the effect of incentives on PA in light of recent theoretical and technological advances. The primary objective of this review was to assess the randomised controlled trial (RCT) evidence examining the short-term (<6 months) and long-term (≥6 months) effects of incentives on daily step counts. Daily step counts was a priori selected as the primary outcome of interest given the growing use of smartphones/wearable trackers that monitor steps, the widely cited public health recommendations regarding steps (ie, 8000 to 10 000/day),32 and the ease with which studies reporting steps can be compared. Recent validation studies found that the iPhone step counting feature (version 6 or newer), as well as those for Android smartphones (eg, HTC, Motorola) and Fitbit trackers (eg, hip-worn Zip, wrist-worn Flex) were accurate in laboratory and field conditions.33–35 An important secondary objective of this review was to determine whether the effects of incentives on PA persist into the post-intervention period after incentives are withdrawn. Another secondary objective was to disentangle heterogeneity between studies through subgroup meta-analyses.

## Methods

### Electronic search

This study updates the authors’ previous systematic review and meta-analysis in which 11 RCTs were examined to determine the influence of incentives on exercise adherence in adults.20 We adapted the previous search strategy to capture articles not retrieved by the original one (online supplementary file 1). Seven electronic databases (CCTR, CINAHL, COCH, Embase, Medline, PsycINFO, PubMed) were searched for English-language, peer-reviewed studies using an RCT methodology published from January 2012 to May 2018 (the original review included articles published up to June 2012, before Apple introduced the Health Kit app). Reference lists of relevant studies were also hand searched for eligible papers.

### Eligibility criteria

RCT studies were included if they reported the effects of incentives on the PA of adults (aged ≥18 years). Studies rewarding multiple health behaviours were included (eg, weight loss, healthy eating, PA) if at least part of the incentive was allocated to a PA behaviour (eg, self-monitoring, gym attendance, daily step count) or outcome (eg, aerobic fitness). Incentives were defined as any cash or non-cash reward with a monetary value, not including gifts of negligible or symbolic value (eg, coffee mug).

### Study selection

Article records (titles, abstracts) were independently screened by two reviewers (SO and SK). A third reviewer (MM) was consulted where uncertainty existed. Full texts from eligible studies were retrieved and screened by one reviewer (SO). A second reviewer (MM) was consulted when a study’s eligibility was unclear. Reasons for study exclusion are presented in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines flowchart (figure 1).

Figure 1

Flowchart of included and excluded RCTs examining the impact of financial incentives on physical activity in adults. RCTs, randomised controlled trials.

### Data acquisition

Data from eligible studies were systematically abstracted using a protocol informed by the Task Force on Community and Preventive Service’s procedure for systematic reviews.36 Study-level (eg, intervention duration, PA outcomes, incentive design features) and participant-level information (eg, age, baseline PA) was extracted by one reviewer (SO; tables 1 and 2). Authors of included studies were contacted for missing data. A second reviewer (AB) audited all retrieved step count estimates.

Table 1

Summary of participant characteristics.

Table 2

Summary of study characteristics and head-to-head comparisons testing behavioural economics decision biases.

### Study quality

Two authors (SO and MM) independently assessed the methodological quality of included studies using the Effective Public Health Practice Project (EPHPP) Quality Assessment Tool for Quantitative Studies.37 This tool was developed to systematically appraise public health research studies. The EPHPP was chosen because it has demonstrated excellent inter-rater reliability for the final grade and more conservative bias estimates than the Cochrane Collaboration Risk of Bias Tool. With this tool, each of seven study components (selection bias, study design, confounders, blinding, data collection, dropouts) is assigned a weak (≥2 weak ratings), moderate (one weak rating) or strong (no weak ratings) score. Consensus on global quality assessment ratings was reached by two authors (SO and MM).

### Data analysis

A meta-analysis was performed on studies reporting changes in daily step counts. Effect sizes were calculated as the mean difference (MD) in daily step counts between study participants allocated to an incentive-based intervention arm and those allocated to a non-incentive control arm. We pooled study estimates that were statistically adjusted for baseline step counts as these were most consistently reported in the majority of studies. If baseline step count-adjusted effect estimates were unavailable in a study, we also used unadjusted estimates as these were more comparable than fully adjusted estimates. Nonetheless, findings were compared with studies reporting fully adjusted estimates in a sensitivity analysis (described in sensitivity analysis below). We conducted separate analyses of the effects of incentive-based interventions on step counts during (a) the intervention and (b) post-intervention periods. Pooled statistical effects were calculated using a random-effects model. Statistical heterogeneity was assessed using Cochran’s Q statistic and the I2 statistic of the proportion of total variation because of heterogeneity.38 The possibility of publication bias was examined by visually inspecting funnel plots for their skew and asymmetric shape and quantitatively by Egger’s test for asymmetry.39

Subgroup analyses were used to examine the robustness of the pooled estimates. Subgroups were compared by a number of participant and study characteristics selected a priori based on their potential to modify incentive effects.18 20 31 40–42 These include the following: (1) age (sampling only among older or younger participants vs non-specific), (2) sex, (3) sampling only overweight or obese participants vs non-specific, (4) income, (5) sampling only physically inactive or low active participants versus non-specific, (6) length of the intervention and follow-up periods, (7) monetary value of the incentive (above or below the median incentive value used in included studies), (8) PA measurement method and (9) the main behavioural economics (BE) concept informing the intervention (see table 3 for a list of these concepts). Where possible, sensitivity analyses were performed by comparing estimates based on study quality (weak to moderate vs strong quality), and estimates derived from studies reporting unadjusted or adjusted for baseline walking only versus adjusted estimates (statistically adjusted for participant characteristics as confounders, eg, age, income). The influence of individual studies on the pooled effect estimates was examined by removing one study at a time. All meta-analyses were conducted using Comprehensive Meta-Analysis V.3.43

Table 3

Behavioural economics decision biases used in financial incentive interventions.17 77 *

A narrative summary of all included studies is also provided using ‘vote counting’ (simple count of studies with positive and null effects) to explore short-term (<6 months), long-term (≥6 months) and post-intervention (after incentive removal) effects for studies with different outcomes (ie, gym visits, daily step count, self-monitoring, energy expenditure). Studies comparing two or more behavioural economics concepts (head-to-head) are also summarised.

## Results

### Study characteristics

From an initial return of 6038 studies, 202 full texts were assessed for eligibility (figure 1). In all, 23 studies involving 6074 participants were included in the review (64.42% female, mean age=41.20 years, mean body mass index=29.91 kg/m2; see table 1).15 21 44–64 Characteristics of the 23 included studies are outlined in table 2. In total, 19 of 23 studies were conducted in the USA.15 21 44–48 50–59 61 64 Sample sizes ranged from 19 to 1000 participants. Interventions lasted less than 12 weeks in four studies,46–48 64 12–23 weeks in 16 studies,15 44 45 49–52 54–59 61–63 and 24 or more weeks in three studies.21 53 60 No interventions extended past 26 weeks. Totally, 18 of 23 studies reported post-intervention PA21 44 45 47 48 50–52 54–61 63 64 with an average follow-up period of 17.5 weeks after incentive removal. One study received a weak quality rating,59 20 received moderate ratings,15 21 44–46 49–58 60–64 and two strong ratings (online supplementary file 2).47 48 Number of days a PA goal was met (eg, gym visits) was the most commonly reported outcome (n=17)21 44–48 50 51 54–58 60 61 63 65 66 though attendance expectations varied widely by study (eg, incentive for one gym visit in a week, nine visits in 6 weeks, etc). In all, 14 studies used wearable trackers or smartphone accelerometers to objectively-assess PA.15 21 45 49–51 53–58 62 64 Among these, 12 studies reported steps per day and were included in the meta-analysis.15 21 45 49–51 53–58 All 23 studies leveraged the ‘present bias’ (table 2). Despite delayed rewards in four studies (the other 19 out of 23 offered incentives within 7 days),15 45 47 60 this classification made sense given that instant PA data were available to all participants (eg, data from smartphone/wearable trackers, which participants knew was tied to a future incentive). All studies incorporated at least one other behavioural economics concept in addition to ‘present bias’. ‘Loss aversion’ was most commonly used (n=16),21 45 47 49–51 53 54 56–59 61–64 followed by ‘fresh start’ (n=13),21 44 48–51 53 58–63 ‘over-optimism‘ (n=9),45 49 51 54–58 ‘salience’ (n=12),15 44 46 48 52 54–57 60 61 64 ‘herd mentality’ (n=5)46 48 51 55 56 and ‘commitment’ (n=4).44 49 61 64 Different behavioural economics-informed designs with similar reward sizes were directly compared in 13 studies.15 21 46–48 50 51 54–57 60 64 See a full description of each study’s incentive design features including type, size and probability of rewards in online supplementary file 3.

### Meta-analyses

In total, 12 studies were included in the meta-analysis, including 2631 participants. Incentives increased mean daily step counts during the intervention period (pooled MD, 607.1; 95% CI: 422.1 to 792.1) and at follow-up assessment (pooled MD, 513.8; 95% CI: 312.7 to 714.9) (figure 2). As per Higgins and colleagues’ classification,38 heterogeneity was found to be high for studies during the intervention period (I2=80.8, p<0.0001; Q=114.5) and at follow-up (I2=85.1, p<0.0001; Q=120.8) which is expected when pooling data from multi-component behavioural interventions. However, heterogeneity was found to be moderate–low in the subgroup analyses suggesting differences may be due to study and participant characteristics (table 4). Studies that were minimally adjusted for baseline walking, or unadjusted, reported a higher pooled MD in daily steps at the intervention period (vs adjusted; 186 steps) and post-intervention (vs adjusted; 155 steps), although these differences were not found to be meaningfully different. Also, pooled effect estimates did not change substantially with the exclusion of any study. Publication bias was a possibility in both assessment periods as funnel plots were moderately asymmetrical (intervention period: Egger regression intercept, 2.23 (p=0.046) and at follow-up assessment: Egger regression intercept, 3.59 (p=0.020)) online supplementary file 4.

Figure 2

Pooled random-effects analysis of mean differences in daily step counts during the intervention period and at post-intervention follow-up.

Table 4

Pooled random-effects analysis of mean differences in daily step counts during the intervention period and at post-intervention follow-up, by subgroup variable.

## Discussion

Demonstrating the rising interest in financial incentives for PA promotion, 23 studies were identified over the last 6 years alone, compared with 11 in the initial review.20 Estimates from this meta-analysis suggest financial incentives increased daily step counts for short and long duration interventions by 607 steps, or approximately 10%–15%. This is consistent with our first review which also found modest but significant effects (ie, 11.6% increase in exercise session attendance over the short term). Notably, the median incentive size in this review was about $1.50 US per person per day, compared with$10.00 US in the original review. This efficiency may be due in part to recent technological advances that have made tracking and rewarding PA easier and more immediate, as well as a broader application of potent behavioural economics concepts like ‘loss aversion‘, 'fresh start', ‘over-optimism', and ‘salience’. Also, 18 out of the 23 included studies (78.3%) tracked PA in the post-intervention period, compared with only 3 out of 11 (27.3%) in the original review providing new insight into the quality of incentive-induced health behaviour change. Regarding sustained effects, vote counting indicated that only four out of 18 studies reported post-intervention benefits. On the other hand, when data were pooled in the meta-analysis, statistically significant daily step count differences were observed 3–6 months after incentives were removed, with an average difference of 514 steps post-intervention. These findings (vote counting vs meta-analysis) might be explained by the lower overall precision of individual studies with generally small sample sizes. Nonetheless, the positive post-intervention effect observed in our pooled analyses provides evidence to contradict the prevailing sentiment that extrinsic rewards undermine intrinsic motivation to engage in health behaviours and damage the potential for sustained improvements.67 68

The undermining effect of extrinsic awards has historically been based on the tenets of self-determination theory (SDT) and studies examining the impact of incentives on motivation to do enjoyable tasks such as completing puzzles. The assumption that these results can be extended to the use of incentives for health behaviour change has been challenged, however. Promberger & Marteau,68 for example, found no evidence that incentives undermine intrinsic motivation for health behaviours for which people begin with low levels of intrinsic motivation.69 In fact, for some, the opposite may be true. Cognitive evaluation theory, a subtheory of SDT, predicts that incentives might boost intrinsic motivation primarily through its action on self-efficacy (if incentives are contingent on realistic, confidence-promoting PA goals).70 Unfortunately, this review found no studies measuring self-efficacy or self-determined motivation over time to test this hypothesis. In a separate but related paper though Pope et al (2014) concluded that intrinsic motivation persisted among college students rewarded to attend the gym, even after incentives were discontinued.65 Crane et al 70 drew similar conclusions when they rewarded weight loss-related behaviours (eg, self-monitoring).71 In a feasibility RCT, Mitchell et al 72 found no preliminary evidence of undermining in a cardiac rehabilitation context.73 Moller et al (2012), on the other hand, found that incentives perceived as controlling, undermined intrinsic motivation to eat healthy and be active.74 It may be that given the modest nature of most incentives used in the reviewed studies, the rewards were not perceived as controlling by participants. While incongruences exist, this review challenges the assumption that incentives damage intrinsic motivation in all cases and that PA will not be sustained once the incentives are withdrawn. This assumption is consistent with the results of a meta-analysis by Mantzari et al (2015) in which smoking cessation was sustained for 3 months after incentive removal.31

### Limitations and future directions

Our results should be interpreted with caution in light of some key limitations. First, the meta-analyses were limited to studies reporting mean changes in steps per day. Analyses of other PA variables may have yielded different results. A second limitation is the small sample size in the majority of studies. Examining studies with larger sample sizes is expected to lower heterogeneity estimates and better elucidate findings. Additionally, research examining potential mechanisms (eg, self-efficacy, self-determined motivation) through which incentives influence behaviour would be beneficial. Fourth, we acknowledge publication bias and the possibility that selective reporting may undermine the generalisability of our findings. Furthermore, our subgroup analysis yielded overlapping CIs and therefore we cannot be certain as to how the incentives-for-PA effects may differ by participant and study characteristics. More studies are needed that sample under-represented groups (eg, lower-income adults). Fifth, the clinical benefits of PA are usually reserved for those who meet recommended levels of moderate-intensity to vigorous-intensity PA and sustain PA behaviour for longer periods,32 and so more longitudinal research examining increases in moderate–vigorous PA is also needed. No intervention or follow-up period lasted longer than 6 months which limits the applicability of these findings. Last, the external validity of the results is limited. A review of quasi-experimental studies evaluating incentives in ‘real-world’ settings (eg, incentive-based workplace wellness programme, government initiatives, apps) would provide valuable insight into the effectiveness of different incentive designs in practice.

## Conclusion

This systematic review and meta-analysis found that incentives increased PA for interventions of short and long durations and after incentives were removed, though the count of studies with positive post-intervention effects was modest (see figure 3 for Infographic summary). Nonetheless, and contrary to what has been suggested for years, a short-term incentive ‘dose’ may promote sustained PA post-intervention. Improvements, therefore, can be expected when technology-enabled, behavioural economics-informed incentives are added to multi-component PA interventions. More work is needed, however, to replicate these findings in light of some of this review’s limitations.

Figure 3

Infographic summary.

### What is already known?

• Two systematic reviews and one meta-analysis have determined that financial incentives increase physical activity (PA) in the short term (3 months or less) and while in place.

• The evidence regarding long-term (6 months or more) and sustained (after financial incentives are withdrawn) PA increases is mixed.

### What are the new findings?

• In all, 23 randomised controlled trials were identified over the last 6 years demonstrating the rising interest in financial incentives.

• Modest incentives ($1.40 US/day on average and as small as$0.10 US/day) increased PA for interventions of short and long durations and after incentives were removed.

• Subgroup meta-analyses and ‘vote counting’ provide insight for incentive programme optimisation. More immediate (within 7 days) incentives for individualised daily step goal achievement (roughly 10%–15% above baseline) offered for longer periods (24 or more weeks) to lower active adults (<7–8000 daily steps), for example, are recommended.

• In total, 13 studies compared different behavioural economics-informed incentive designs suggesting these can be harnessed to boost incentive effectiveness as well.

## Acknowledgments

We acknowledge Subhan Kangatharan (SK) for screening titles and abstracts.

## Footnotes

• Contributors All co-authors meet the ICMJE authorship criteria.

• Funding Marc Mitchell received post-doctoral fellowship funding from the Canadian Institutes of Health Research to support this project. The other authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

• Competing interests From 2015 to 2017 Marc Mitchell was the sole proprietor of a program evaluation company called Incentive Avenue Inc. Currently, he is the Principal Behavioural Insights Advisor for an incentive-based mHealth application company called Carrot Insights Inc. He reports consulting income from Carrot Insights Inc. in the past 36 months and company stock options as well. Furthermore, Stephanie Orstad worked as an independent contractor for Incentive Avenue Inc. in 2016. The other authors declare that no competing interests exist.

• Patient consent for publication Not required.

• Provenance and peer review Not commissioned; externally peer reviewed.