Figure 7

Out-of-sample${\mathit{R}}^{\mathbf{2}}$for models of predicting #bostonstrong. (a) Overall, the geo-social model performs significantly better than all other models. (b) The predictor ‘visit’ adds more predictability than ‘geo’ in predicting the #bostongstrong occurrences in farther cities (top 50% in terms of geo-distance). Error bars show the bootstrapped 95% confidence intervals of the out-of-sample $R^2$.