This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mhealth and uhealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.
Physical inactivity, now the fourth leading cause of death, is a primary element of noncommunicable diseases. Despite a great number of attempts, there is still a lack of effective approaches that can motivate sedentary populations to increase their levels of physical activity over a sustained period. Incentives for exercise can provide an immediate reward for increasing activity levels, but because of limited funding to provide rewards, previous programs using this approach have only shown short-term changes in behavior. Sweatcoin (Sweatco Ltd, UK) is an app-based platform that converts physical movement into virtual currency. The currency can be exchanged for goods and services on their marketplace, providing a continuous incentive to be active. This study investigates the physical activity behavior change observed in Sweatcoin users over a 6-month period of app usage.
The aim of this study was to investigate the change in physical activity (measured using daily step count) of a sample of Sweatcoin users, the longevity of the change, and whether this change can be predicted by demographic and other lifestyle variables.
Activity data from a sample of 5892 Sweatcoin users were used to analyze daily step count. Activity change was measured in terms of the percentage change in average daily step count for each month after registration, relative to that in the 3 months before using the app. Users were grouped according to having no or negative, moderate, or high activity change. A subset of users completed a questionnaire that allowed differences between groups in terms of activity and demographic status to be investigated using regression analyses.
Daily step count increased by 19% on average over the 6 months following registration (
The results highlight that an incentives-based app can induce significant physical activity behavior change, sustained over a 6-month period. Importantly, the results suggest that those typically lacking motivation to exercise (sedentary and high BMI) are most likely to be incentivized to increase their activity levels.
High levels of sedentary behavior increase the risk of cardiovascular disease, some types of cancer, and type 2 diabetes, particularly when combined with low levels of physical activity [
Adherence to physical activity programs has been found to be higher in individuals who exercise for enjoyment and social interaction rather than for fitness and appearance [
In this study, we investigated the behavior change exhibited by users of Sweatcoin and analyzed the change in physical activity following engagement with the app. In addition, we used a survey from a subsample of users to identify which populations were most likely to show the biggest activity change.
We hypothesized that users who will be motivated to increase their physical activity following engagement with the app can be predicted through a range of demographic and other self-reported lifestyle variables.
Sweatcoin is a free app available on iOS and Android platforms. The concept of Sweatcoin is to convert a user’s step count, as recorded by the sensors on a smartphone, into virtual currency [
Users accumulate Sweatcoins that can subsequently be spent on the marketplace, accessible through the app (
(a) Screenshots from the Sweatcoin app (as of December 2018). As a user’s step count (recorded by their smartphone) accumulates over the day, it is verified and converted into Sweatcoins. (b) The Sweatcoins are stored in the user’s wallet and can be subsequently used to purchase products, services, or subscriptions that are available on the marketplace.
A dataset containing daily step count for each user was used for the analysis of this study. The dataset contained no identifiable information, with the exception of a
On the basis of these inclusion criteria, 5952 Sweatcoin users were included in the daily step count dataset (
Flowchart of analysis stages with corresponding sample sizes.
Processing and analysis of the activity data were completed using R (version 3.4.1), a programming language specifically designed for statistical data analysis [
For each 30-day period within each user’s data, we calculated the mean, median, maximum, minimum, upper quartile, lower quartile, and SD of the daily step counts. However, for this study, we focused only on the mean values recorded. In addition to the mean daily step count recorded over the whole 30-day period, we further measured the mean value across all weekend days (Saturdays and Sundays only) occurring within the 30-day period along with the mean of the weekdays only (Monday through Friday) in that period.
The aim of the initial analysis was to quantify the change in physical activity (in the context of daily step count) after registering with the Sweatcoin app, relative to the 3-month period before registration. We did this by taking the mean across months −1 to −3 and subtracting this value from the mean daily step count of all 9 months. This resulted in a relative daily step count that was centered around zero for the period before registration and quantified any increase or decrease in the 6 months after registration. These relative step counts for the period after registration were subsequently converted to a percentage of the mean daily step count measured across months −1 to −3 to normalize against the large variation in activity levels across the sample.
We excluded users who had highly variable step counts in the 3-months preregistration period as the variability would impact the calculations of activity change in the postregistration period. This was achieved by ranking users according to the SD of their mean steps per day across months −1 to −3 and excluding those in the top percentile (equating to a SD >4020 steps/day). The final sample size for analysis was 5892 after removing these users (
Users were finally classified into groups according to their level of
On the basis of the grouping of users according to their change in activity following app registration, we wanted to determine if there were any demographic differences between these groups. To achieve this, we defined a set of demographic variables of interest to be tested (
Demographic and other variables included in the questionnaire, along with options or categories.
Variable | Options or categories |
Gender | Male, Female |
Age | <18 years [excluded], 18-24 years, and then 10-year increments up to 85 years or older |
Height | 1 m or less, increments of 10 cm up to 2 m. Equivalent feet and inches measurements also shown |
Weight | 40 kg or less, increments of 10 kg up to 120 kg, or >120 kg. Equivalent stones and pounds measurements also shown |
Education | High school (or equivalent), Grammar school, College degree, Bachelor’s degree, Master’s degree, Professional degree, Doctoral degree, or Other |
Employment | Student, Retired, Unemployed, Homemaker, Self-Employed, Private sector, or Public sector |
Income | <£10,000, then £10,000 increments up to £100,000, £100,000-£149,999, £150,000-£200,000, over £200,000, or would rather not say. (Equivalent US dollar amounts also shown.) |
Marital status | Living with another, Married/civil partnership, Separated, Divorced, Widowed, Would rather not say, or Other |
Dog owners | Yes, No |
Have children | Yes, No |
Regularly use a wearable fitness tracker | Yes, No |
Home location | Urban, Suburban, or Rural |
Commute type | Car, Bus, Train, Tram/Tube, Walk, Cycle, or Other/None |
Commute distance | No commute/no fixed place of work, <5 km, 5-10 km, 11-15 km, 16-20 km, 21-30 km, or >30 km |
Motivations to exercise | Respondents chose one option from: Increase my overall health, Lose weight, Gain strength, Improve my skills, Have fun, Spend time with friends, To look good, or Other |
Self-reported physical activity | On the basis of the General Practice Physical Activity Questionnaire [ |
Rigidity score | On the basis of the compulsive exercise test [ |
Walking pace | Slow (<3 mph), Steady, Brisk, or Fast (>4 mph) |
Other health/fitness apps used | Respondents chose from Never, Sometimes, or Regularly from the following apps: 7-Minute Workout, 8fit Planner, Calm Meditation, Calorie Counter, Fitbit, Headspace, MyFitnessPal, Nike, Strava, and Weight Watchers |
To recruit for the questionnaire survey, all users included in the original sample were shown an advert on the Sweatcoin marketplace inviting them to participate. Users who registered to participate were provided a link to the questionnaire (hosted on the Google Forms platform). Participant information and ethical approval information were provided on the first screen. Consent was registered by participants explicitly ticking a box to register their consent to take part and then continuing to the questionnaire. Those who completed the questionnaire were sent a £10 (United Kingdom) or US $10 retail voucher. The questionnaire typically took participants around 10 min to complete.
Over a period of 4 weeks, a total of 841 users completed the questionnaire (
The questionnaire data were coded and restructured as required to make each variable suitable for regression analysis. Dummy variables were created for multiple-choice responses. Self-reported activity and rigidity scores were calculated as per guidelines. However, it should be noted that rather than limiting the General Practice Physical Activity Questionnaire (GPPAQ) score to a maximum of 3, we used the total score from all activities. In addition to the questionnaire variables, the season of registration was added as a further set of dummy variables to ensure seasonality would be accounted for in the analysis.
The physical activity change class (
Classification of physical activity behavior change based on percentage change in daily step count relative to the 3 months preregistration period. The number of samples in each class is shown for both the daily step count data and the combined questionnaire responses with daily step count.
Class | Label | Range, % | Number of users in class, n (%) | |
Activity dataset (N=5892) | Questionnaire responses (N=728) | |||
0 | No or negative change in activity | <1 | 2172 (36.86) | 258 (35.4) |
1 | Moderate positive change in activity | 1-18.7 | 1542 (26.17) | 196 (26.9) |
2 | High positive change in activity | >18.7 | 2178 (36.96) | 274 (37.6) |
Ethical approval was received from the University of Warwick Biomedical and Scientific Research Ethics Committee (Approval Number: REGO-2017-2086). In addition, 2 members of the academic research team received honorary contracts with Sweatcoin to oversee the analysis of the anonymized activity data. Recruitment for the demographic questionnaire was further handled by employees of Sweatcoin to ensure participants’ activity data were not made identifiable outside of the company until they had provided consent to merge their activity data with the questionnaire responses.
Initially, mean daily step count was examined across three 3-month periods. In particular, we contrasted the mean daily step count for the 3 months before registration with the app and for two 3-month periods following app registration (months 1-3 and 4-6; see
For the remaining analyses, we no longer contrasted weekend and weekday activity and, hence, averaged the daily step count across continuous 30-day periods. The relative percentage activity change following app registration was calculated as described in the Methods section, revealing an overall mean increase of 18.7% (47.9%) in daily step count across the sample (
Mean daily step count across the sample (N=5892) was analyzed for the 3 months before app registration (Pre) and 6 months after registration (Post). (a) For analysis, these were grouped into 3-month periods; the results highlight the consistent increase in mean daily step count after registration. (b) The same measure is broken down over 30-day periods. Both plots show the data separated into weekday and weekend activity. Although weekend activity is lower, the overall pattern of increase after registration is similar. Error bars represent SE of the mean.
(a) Percentage change in daily step count for the period after registration, relative to the 3 months before registration (N=5892). The horizontal dashed line (gray) represents the overall average percentage increase of 18.7%. Error bars represent SE of the mean. (b) Individual data points (N=5892) of mean daily step count before app registration against the subsequent percentage change after registration.
From the overall mean activity change of 18.7% across 6 months, we defined user activity behavior change classifications as shown in
(a) Histogram showing distribution of user registrations in the sample data (N=5892). The main app launch can be identified in May 2016, whereas there is a further rapid acceleration around December 2016, which skews the data toward the winter and spring seasons. (b) User classifications of activity behavior change broken down across seasons; numeric values inside the circles represent the percentage of users in the associated class; sample sizes for each season are shown in italics under each chart. For winter and spring, there is a higher proportion of high activity change users than no or negative change users. Summer and autumn show an over representation of no or negative change users.
In
Descriptive statistics of key variables in the questionnaire (N=728).
Variable | Frequency, % | |
Male | 64.7 | |
Femalea | 35.3 | |
18-24 | 39.7 | |
25-34 | 38.9 | |
35-44 | 17.3 | |
Over 45a | 4.1 | |
Underweight | 3.2 | |
Healthy weighta | 49.2 | |
Over weight | 21.6 | |
Obese | 26 | |
Yes | 67.7 | |
Noa | 32.3 | |
Not in employment | 6 | |
Student | 29 | |
Employed (private sector)a | 38.3 | |
Employed (public sector) | 19.4 | |
Self-employed | 7.3 | |
Less than £20k | 17.4 | |
£20k-£39k | 22.4 | |
£40k-£59ka | 33.7 | |
£60k-£79k | 12.8 | |
£80k or more | 13.7 | |
Married/civil partnership | 27.2 | |
Cohabiting | 15.7 | |
Divorced/separated | 4.3 | |
Singlea | 52.8 | |
Yes | 31 | |
Noa | 69 | |
Yes | 31.9 | |
Noa | 68.1 | |
Yes | 46.4 | |
Noa | 53.6 | |
Urbana | 38.2 | |
Suburban | 44.4 | |
Rural | 17.4 | |
Walk | 19.6 | |
Cycle | 4.3 | |
Bus | 9.5 | |
Car | 46.8 | |
Train | 6.6 | |
Tube/tram | 7.4 | |
Other | 5.8 | |
None/no fixed location | 10.7 | |
<5 km | 28.3 | |
5-10 km | 22 | |
11-15 km | 11.4 | |
16-20 km | 8.9 | |
21-30 km | 7.8 | |
>30 km | 10.9 | |
Increase overall healtha | 37.4 | |
Lose weight | 29.8 | |
Gain strength | 12.1 | |
Look good | 10.9 | |
Improve skills | 3.2 | |
Have fun | 4.3 | |
Spend time with friends | 2.3 | |
Inactive | 13.3 | |
Moderately inactive | 13.6 | |
Moderately active | 19.8 | |
Active | 53.3 | |
Slow (<3 mph) | 7.1 | |
Steady | 44.5 | |
Brisk | 38.6 | |
Fast (>4 mph) | 9.8 |
aResponse options were used as the reference response with which the other responses for that variable were compared in the regression model.
bGPPAQ: General Practice Physical Activity Questionnaire.
Frequency of usage of other fitness- and well-being–related apps (N=728).
App | Never use, % | Sometimes use, % | Regularly use, % |
7-Minute Workout | 76.1 | 19.2 | 4.7 |
8fit Planner | 88.7 | 8.1 | 3.2 |
Calm Meditation | 76.2 | 17.7 | 6.0 |
Calorie Counter | 81.6 | 12.9 | 5.5 |
Fitbit | 72.1 | 10.2 | 17.7 |
Headspace | 79.8 | 15.0 | 5.2 |
MyFitnessPal | 60.0 | 25.3 | 14.7 |
Nike | 74.0 | 18.1 | 7.8 |
Strava | 77.7 | 11.1 | 11.1 |
Weight Watchers | 88.6 | 7.3 | 4.1 |
On the basis of the classification of users according to their change in activity following app registration, we investigated the main predictors of this behavior change based on the demographic questionnaire variables. A multinomial logistic regression was run to compare users in the moderate and high activity change classes with respect to those in the no or negative activity change class. Independent variables included the demographic and other variables captured by the questionnaire (see
The results (N=728, χ2146=197.6
Multinomial logistic regression results for predictors of moderate activity behavior change classification versus no or negative activity change (N=728). Results show the odds ratios of the significant predictor variables (
Variable | Model coefficients (B) | SE | Wald | Odds ratio | |
Registered in winter | 2.45 | .67 | 13.36 | 11.54 | <.001 |
Registered in spring | 2.39 | .69 | 11.92 | 10.88 | .001 |
Registered in autumn | 1.43 | .71 | 4.05 | 4.17 | .04 |
Regular user of MyFitnessPal | −0.89 | .37 | 5.77 | 0.41 | .02 |
Multinomial logistic regression results for predictors of high activity behavior change classification versus no or negative activity change (N=728). Results show the odds ratios of the significant predictor variables (
Variable | Model coefficients (B) | SE | Wald | Odds ratio | |
Registered in winter | 1.54 | .46 | 11.13 | 4.67 | .001 |
Registered in spring | 1.62 | .49 | 11.08 | 5.05 | .001 |
Overweight | 0.61 | .27 | 5.18 | 1.83 | .02 |
Rigidity score | 0.07 | .03 | 5.72 | 1.07 | .02 |
GPPAQa score | −0.13 | .07 | 3.90 | 0.88 | .048 |
aGPPAQ: General Practice Physical Activity Questionnaire.
In this study, we have investigated physical activity behavior change in users engaging with an app that converts step count into virtual currency (Sweatcoins). Using an initial sample of 5892 users who had been registered with the app for 6 months or longer, we used anonymized daily step count data to analyze activity levels for the 6-month period following registration and compared it with that recorded for the 3-month period before registration. Importantly, our results found a significant and consistent increase in step count following app registration, which averaged 18.7% across the period and across all users. In an earlier pilot analysis, we previously reported a figure of 19.5% because of a slight variation in inclusion criteria [
The respondents to the subsequent questionnaire were mainly from the young adult population, which is likely to be representative of Sweatcoin’s overall user demographic. Therefore, our results do not necessarily generalize across all age groups. However, the captured demographic is an important generation to target, with many physical activity programs focusing specifically on children or older adults but few targeting young adults and capitalizing on their engagement with smartphone technologies [
The classification of users into levels of activity behavior change and subsequent demographic survey data allowed us to examine in detail which users are most likely to increase activity following registration with the app. Despite the wide range of variables captured, the equivalent variance accounted for by the logistic regression model was relatively low at 26.8%, highlighting that there are likely additional lifestyle events not captured that could contribute to increased activity behavior. In particular, we did not capture any medical conditions the respondents may have had at the time that may impact the levels of physical activity. Weather-based effects [
Taking seasonality into account, we found additional variables that distinguished moderate and high activity change groups from those showing no or negative change. Regular users of MyFitnessPal were significantly less likely to be in the moderate change group compared with the no or negative change. MyFitnessPal [
For the high change group, we found that a number of variables were significant predictors of this group relative to no or negative change. We found a small positive relationship to a user’s rigidity score [
The overall consistent increase in physical activity over the 6-month period following registration highlights a differentiation from other incentive-based and goal-oriented programs reported, which typically show only short-term behavior change effects [
Monetary and other tangible incentives have been effective in increasing physical activity behavior in sedentary populations [
There are some limitations to this study. In particular, we were unable to establish a control group as a direct comparator to the intervention group we analyzed. This would have required the recruitment of a similar sample (ie, N>5000) of non-Sweatcoin app users who would share their daily step count data over matched 9-month periods. However, the dataset we used included both pre- and postregistration data, such that the preregistration data provided an objective baseline measure of activity. Furthermore, during the period that the data were generated, users were unaware that it would be used for this kind of analysis, and hence, there was no risk of experimental bias in the data.
There was also risk of bias being introduced because of sampling of users. We only used data from iPhone users because of the limitation of the app only being able to access daily step count data via Apple HealthKit on these devices. This could have potentially skewed the demographic profile of the sample recruited, although it has recently been reported that there are few demographic and personality differences between iOS and Android users [
In conclusion, the Sweatcoin concept allows users to continuously be incentivized to be physically active through generation of virtual currency from steps. Through analysis of a sample of Sweatcoin users, we have observed a sustained increase in physical activity (measured by daily step count) over a 6-month period, with users identified as overweight and less active most likely to show the highest increases in activity after registering with the app.
Full regression results.
body mass index
General Practice Physical Activity Questionnaire
global positioning system
odds ratio
This study was supported by an Innovate UK grant (Ref: 80753-504440) awarded to Sweatco Ltd and author ME.
All authors contributed to the design of the study. FE and EK led data processing and analysis of step count data. ME, FE, and EK worked on data processing and analysis of survey data. ME wrote the manuscript. All the authors commented on and edited the manuscript.
ME received a grant in partnership with Sweatco Ltd from Innovate UK. He was provided with an honorary unpaid research position within Sweatco Ltd for the period of the project. FE was provided with an honorary unpaid research position within Sweatco Ltd for the period of the project. AD and OF are directors and shareholders of Sweatco Ltd. Sweatco Ltd received a grant in partnership with ME (University of Warwick) from Innovate UK to conduct the study. EK is an employee and shareholder of Sweatco Ltd.