Exercise contagion in a global social network

Abstract

We leveraged exogenous variation in weather patterns across geographies to identify social contagion in exercise behaviours across a global social network. We estimated these contagion effects by combining daily global weather data, which creates exogenous variation in running among friends, with data on the network ties and daily exercise patterns of ∼1.1M individuals who ran over 350M km in a global social network over 5 years. Here we show that exercise is socially contagious and that its contagiousness varies with the relative activity of and gender relationships between friends. Less active runners influence more active runners, but not the reverse. Both men and women influence men, while only women influence other women. While the Embeddedness and Structural Diversity theories of social contagion explain the influence effects we observe, the Complex Contagion theory does not. These results suggest interventions that account for social contagion will spread behaviour change more effectively.

Figure 1. Peer effects in global running behaviours.

The panels display social influence coefficients from second-stage regressions in the two-stage least squares specification for friends' behaviour at time t influencing ego at time t, t+1 and t+2 for (a) distance ran in kilometres (km), (b) pace in km per minute, (c) running duration in minutes and (d) calories burned. Bars are 95% confidence intervals. (e) The table at the bottom of the figure compares social influence coefficients and s.e. from the IV models to those from the OLS models and provides the OLS overestimates of social influence as a percentage of the IV estimates.

Figure 2. Heterogeneity in social influence effects across relationships.

The panels display social influence coefficients across dyadic relationships in which ego is (a,b) a more or less active runner than their friends, (c) a more or less consistent runner than their friends and (d) either the same or a different gender than their friends. Bars are 95% confidence intervals.

Figure 3. Testing structural theories of networked contagion.

The panels describe the structural correlates of social influence in the distance run (in km). Panel (a) estimates the social influence effects of the number of distinct friends that run and the number of distinct components of friends that run independently, in separate regressions (separated by the dotted line). Panel (b) directly compares, in the same regression, the number of distinct friends that run (supporting Complex Contagion theory) and the number of distinct network components of friends that run (supporting Structural Diversity theory) as structural moderators of social influence effects. The positive estimate for the number of distinct network components of friends that run and the negative estimate for the number of distinct friends that run, when both are analysed together in (b), supports the Structural Diversity theory. Panel (c) tests whether embedded dyadic relationships with mutual friends transmit influence more effectively than relationships with no mutual friends (supporting Embeddedness theory). The social influence coefficient estimated for embedded relationships (Regression 2) is statistically significantly greater than the social influence coefficient estimated for non-embedded relationships (Regression 1) (t-statistic=2.45, N=10.7M). Bars are 95% confidence intervals.

Figure 4. The strength and exogeneity of weather patterns as instrumental variables for running behaviours.

Panel (a) displays daily correlations between precipitation in New York and Chicago at time t and precipitation in the rest of the United States at time t, t+1, t+2 and t+3. White colouring indicates no correlation, while progressively darker green colouring indicates proportionally stronger correlations. Panel (b) displays daily correlations between running activity per capita and binary indicators of rainfall above the annual average and temperature <35° or >85° in New York and Chicago, respectively. At nearly each spike in rainfall or extreme temperature, running declines markedly, visually demonstrating the strength of the instruments at the daily level. Panel (c) displays running activity per capita on the y axis and precipitation in millimetres on the x axis, while (d) displays running activity per capita on the y axis and temperature in degrees Fahrenheit on the x axis for the top 15 running cities in the United States during the 5-year period. These two panels show strong correlations between the weather and running behaviour and demonstrate that the correspondence of precipitation and temperature to running display different functional forms, necessitating differential approaches to constructing the precipitation and temperature instruments. The inset panel displays the aggregated run traces of the same number of randomly chosen global positioning system-enabled runners on a dry and mild day, a rainy and mild day, a dry and cold day and a rainy and cold day in New York and Chicago, with fewer traces indicating fewer runs by those runners.