Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Jul 20;547(7663):336-339.
doi: 10.1038/nature23018. Epub 2017 Jul 10.

Large-scale Physical Activity Data Reveal Worldwide Activity Inequality

Free PMC article

Large-scale Physical Activity Data Reveal Worldwide Activity Inequality

Tim Althoff et al. Nature. .
Free PMC article


To be able to curb the global pandemic of physical inactivity and the associated 5.3 million deaths per year, we need to understand the basic principles that govern physical activity. However, there is a lack of large-scale measurements of physical activity patterns across free-living populations worldwide. Here we leverage the wide usage of smartphones with built-in accelerometry to measure physical activity at the global scale. We study a dataset consisting of 68 million days of physical activity for 717,527 people, giving us a window into activity in 111 countries across the globe. We find inequality in how activity is distributed within countries and that this inequality is a better predictor of obesity prevalence in the population than average activity volume. Reduced activity in females contributes to a large portion of the observed activity inequality. Aspects of the built environment, such as the walkability of a city, are associated with a smaller gender gap in activity and lower activity inequality. In more walkable cities, activity is greater throughout the day and throughout the week, across age, gender, and body mass index (BMI) groups, with the greatest increases in activity found for females. Our findings have implications for global public health policy and urban planning and highlight the role of activity inequality and the built environment in improving physical activity and health.

Conflict of interest statement

The authors declare no competing financial interests.


Extended Data Figure 1
Extended Data Figure 1. Activity and obesity data gathered with smartphones exhibit well established trends
(a) Daily step counts across age and (b) BMI groups for all users. Error bars correspond to bootstrapped 95% confidence intervals. Observed trends in the dataset are consistent with previous findings; that is, activity decreases with increasing age, , , and BMI, , , and is lower in females than in males, , –.
Extended Data Figure 2
Extended Data Figure 2. Activity and obesity data gathered with smartphones are significantly correlated with previously reported estimates based on self-report
(a) WHO physical activity measure versus smartphone activity measure. The WHO measure corresponds to the percentage of the population meeting the WHO guidelines for moderate to vigorous physical activity based on self-report. The smartphone activity measure is based on accelerometer-defined average daily steps. We find a correlation of r=0.3194 between the two measures (p < 0.05). Note that this comparison is limited because there is no direct correspondence between the two measures—values of self-report and accelerometer-defined activity can differ, and the WHO confidence intervals are very large for many countries (Methods). (b) WHO obesity estimates, based on self-reports to survey conductors, versus obesity estimates in our dataset, based on height and weight reported to the activity-tracking app. We find a significant correlation of r=0.691 between the two estimates (p < 106). (c) Gender gap in activity estimated from smartphones is strongly correlated with previously reported estimates based on self-report. We find that the difference in average steps per day between females and males is strongly correlated to the difference in the fraction of each gender who report being sufficiently active according to the WHO (Pearson r=0.52, p < 103).
Extended Data Figure 3
Extended Data Figure 3. Activity inequality remains a strong predictor of obesity levels across countries when reweighting the sample based on officially reported gender distributions and when stratifying by gender or age
(a) Obesity versus activity inequality on country level where subjects are reweighted to accurately reflect the official gender distribution in each country (Methods). The gender-unbiased estimates are very similar to estimates using all data (r=0.953 for activity inequality and r=0.986 for obesity). (b) Obesity versus activity inequality on a country level for males and females. Activity inequality predicts obesity for both genders. (c) Obesity versus activity inequality on a country level across different age groups. We find associations between activity inequality and obesity persists within every single age groups. Older people are more likely to be obese (see y-axis ranging from 5% to 45% obesity for subjects older than 50 years) and more likely to get little activity (i.e., higher activity inequality on x-axis). These results indicate that our main result—activity inequality predicts obesity—is independent of any potential gender and age bias in our sample.
Extended Data Figure 4
Extended Data Figure 4. Relationship between activity inequality and obesity holds within countries of similar income
Out of the 46 countries included in our main result, we have 32 high income (green) and 14 middle income (orange) countries according to the current World Bank classification. We find that activity inequality is a strong predictor of obesity levels in both high income countries as well as middle income countries. While in middle income countries, iPhone users might belong to the wealthiest in the population, in high income countries iPhones are used by larger parts of the population. The fact that we find a strong relationship between activity inequality and obesity in both groups of countries suggests that our findings are robust to differences in wealth in our sample.
Extended Data Figure 5
Extended Data Figure 5. Graphical definition of activity inequality measure using the Gini coefficient
The Lorenz curve plots the share of total physical activity of the population on the y-axis that is cumulatively performed by the bottom x% of the population, ordered by physical activity level. The diagonal line at 45 degrees represents perfect equality of physical activity (i.e., everyone in the population is equally active). The Gini coefficient is defined as the ratio of the area that lies between the line of equality and the Lorenz curve (marked A in the diagram) over the total area under the line of equality (marked A and B in the diagram). The Gini coefficient for physical activity can range from 0 (complete equality) to 1 (complete inequality).
Extended Data Figure 6
Extended Data Figure 6. Activity inequality is a better predictor of obesity than the the average activity level
(a) Obesity is significantly correlated with the average number of daily steps in each country (LOESS fit; R2 = 0.47). (b) However, activity inequality is the better predictor of obesity (LOESS fit; R2 = 0.64). The difference is significant according to Steiger’s Z-Test (p < 0.01; Methods). This shows that there is value to measuring and modeling physical activity across countries beyond average activity levels. Activity inequality captures the variance of the distribution; that is, how many activity rich and activity poor people there are, allowing for better prediction of obesity levels. Figure repeated from Fig. 2a for comparison.
Extended Data Figure 7
Extended Data Figure 7. Female activity is reduced disproportionately in countries with high activity inequality
(a) Distribution of daily steps for females, males, and all users in representative countries of increasing activity inequality (Japan, United Kingdom, United States, and Saudi Arabia). While in countries with low activity inequality females and males get very similar amounts of activity (e.g., Japan), the distributions of female and male activity differ greatly for countries with high activity inequality (e.g., Saudi Arabia and United States). Activity distributions in these countries demonstrate that larger variances in activity (Fig. 1c) are due to a disproportionate reduction in the activity of females and not just an increase in variance overall. (b) Activity inequality increases with the relative activity gender gap on a country level (Methods). We find that the relative gender gap ranges between 0.041 (Sweden) and 0.380 (Qatar). The average daily steps for females is lower than for males in all 46 countries. The gender gap explains 43% of the observed variance in activity inequality (linear fit: R2 = 0.43). This suggests that activity inequality could be reduced significantly through increases in female activity alone.
Extended Data Figure 8
Extended Data Figure 8. Activity inequality-centric interventions could result in up to 4 times greater reductions in obesity prevalence than population-wide approaches
Given a fixed activity budget (100 daily steps per individual) to distribute across the population, we compare an inequality-centric strategy which equally distributes this budget to minimize activity inequality (100/X% daily steps increase for the activity-poorest X% where X minimizes the country’s resulting activity inequality; Methods) and a population-wide strategy which equally distributes the budget across the entire population (100 daily steps per individual; Methods). Based on our simulations, we find that the inequality-centric strategy would lead to predicted reductions in obesity prevalence of up to 8.3% (median 4.0%), whereas the population-wide approach would lead to predicted reductions of up to 2.3% (median 1.0%).
Extended Data Figure 9
Extended Data Figure 9. Relationship between walkability and activity inequality holds within US cities of similar income
Walkable environments are associated with lower levels of activity inequality within socioeconomically similar groups of cities. We group the 69 cities into quartiles based on median household income (data from the 2015 American Community Survey). We find that walkable environments are associated with lower levels of activity inequality for all four groups. The effect appears attenuated for cities in the lowest median household income quartile. These results suggest that our main result—activity inequality predicts obesity and is mediated by factors of the physical environment—is independent of any potential socioeconomic bias in our sample.
Extended Data Figure 10
Extended Data Figure 10. Differences in country level daily steps are not explained by differences in estimated wear time
Users have an average span of 14.0 hours between the first and last recorded step, our proxy for daily wear time (Methods). While on an individual level, longer estimated wear time is associated with more daily steps (r=0.427, p < 1010), on a country level, there is no significant association between wear time and daily steps (r=−0.086, p = 0.57). Line shows linear fit using the 46 countries with at least 1000 users. This suggests that differences in recorded steps between countries are due to actual differences in physical activity behavior and are not explained by differences in wear time.
Figure 1
Figure 1. Smartphone data from over 68 million days of activity by 717,527 individuals reveal variability in physical activity across the world
(a) World map showing variation in activity (mean daily steps) measured through smartphone data from 111 countries with at least 100 users. Cool colors correspond to high activity (e.g., Japan in blue) and warm colors indicate low levels of activity (e.g., Saudi Arabia in orange). (b) Typical activity levels differ between countries. Curves show distribution of steps across the population in four representative countries as a normalized probability density (high to low activity: Japan, United Kingdom, United States, Saudi Arabia). Vertical dashed lines indicate the mode of activity for Japan (blue) and Saudia Arabia (orange). (c) The variance of activity around the population mode differs between countries. Curves show distribution of steps across the population relative to the population mode. In Japan, the activity of 76% of the population falls within 50% of the mode (i.e., between light gray dashed lines), whereas in Saudi Arabia this fraction is only 62%. The United Kingdom and United States lie between these two extremes for average activity level and variance. This map is based on CIA World Data Bank II data publicly available through the R package “mapdata”.
Figure 2
Figure 2. Activity inequality is associated with obesity and increasing gender gaps in activity
(a) Activity inequality predicts obesity (LOESS fit; R2 = 0.64). Individuals in the five countries with highest activity inequality are 196% more likely to be obese than individuals from the 5 countries with lowest activity inequality. (b) Activity inequality is associated with reduced activity, particularly in females. The figure shows the 25th, 50th, and 75th percentiles of daily steps within each country along with 95% confidence intervals (shaded) as a linear function of activity inequality. As activity inequality increases, median activity (50th percentile) decreases by 39% for males (blue) and by 58% for females (red). (c) Obesity-activity relationship differs between males and females and between high and low activity individuals. The plot shows the prevalence of obesity as a function of daily number of steps across all subjects in all countries (with 95% confidence intervals). For both males (blue) and females (red), a larger number of steps recorded is associated with lower obesity, but for females, the prevalence of obesity increases more rapidly as step volume decreases (232% obesity increase for females vs. 67% increase for males; comparing lowest vs. highest activity).
Figure 3
Figure 3. Aspects of the built environment, such as walkability, may mitigate gender differences in activity and overall activity inequality
(a) Higher walkability scores are associated with lower activity inequality, based on data from 69 United States cities (LOESS fit; R2 = 0.61). (b,c) Walkability is linked to increased activity levels. Curves show average steps recorded throughout the day in United States cities with the top 10 walkability scores (green) and bottom 10 walkability scores (blue). (b) On weekdays, walkable cities exhibit a spike in activity during morning commute (9:00), evening commute (18:00) and lunch times (12:00), while activity is relatively constant and lower overall in less walkable cities. (c) On weekend days, people in more walkable cities take more steps throughout the middle of the day, thus walkability is associated with higher activity levels even when most people do not work or commute. (d) Higher walkability is associated with more daily steps across age, gender, and BMI groups. Bars show the steps gained per day for each point increase in walkability score for 24 United States cities, including 95% confidence intervals (assuming linear model; Methods). Positive values across all bars reveal that, with increasing walkability, more steps are taken by every subgroup. The effect is significantly larger for females overall (left), with the greatest increases for women under 50 years (middle) and individuals with a BMI less than 30 (right).

Similar articles

See all similar articles

Cited by 98 articles

See all "Cited by" articles


    1. Hallal PC, et al. Global physical activity levels: surveillance progress, pitfalls, and prospects. Lancet. 2012;380:247–257. - PubMed
    1. Lee IM, et al. Effect of physical inactivity on major non-communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet. 2012;380:219–229. - PMC - PubMed
    1. UN Secretary General. Prevention and control of non-communicable diseases. 2011 Accessed April 21, 2016.
    1. WHO. Global recommendations on physical activity for health. World Health Organization; Geneva: 2010.
    1. Kohl HW, et al. The pandemic of physical inactivity: global action for public health. Lancet. 2012;380:294–305. - PubMed

Publication types

MeSH terms