Figure 4.
Locating consistency of aging across different temporal segments of smartphone behavior
We used pairwise regressions relating the age-residuals across smartphone behavioral bins.
(A) In this example pair, the age-residuals of the two-dimensional bins are correlated to each other revealing a consistency—so individuals who show accelerated aging in the first two-dimensional bin (magenta) also show accelerated aging in the second bin (yellow).
(B) In this example pair, the age-residuals of the two marked bins are correlated to each other revealing an inconsistency. Here, in the first two-dimensional bin (magenta), individuals who had lower behavioral probability density than the fitted line were considered better performers (as this behavioral density increases with age, positive slope). In the second two-dimensional bin (yellow), individuals who had a higher behavioral probability density than the fitted line are considered better performers (as this behavioral density reduces with age, negative slope). The regression between the residuals shows a positive correlation—so individuals with accelerated aging in one bin show decelerated aging in another.
(C) All two-dimensional bins with significant age-related correlations were selected for the residual analysis. Here, we show the max R2 for the consistent pairs observed at a given two-dimensional bin. The corresponding line plots (each line depicting a significant correlation) show the location of the consistent correlational pattern linking the residuals, where the bins were separated by a bin distance of >5 bins.
(D) Same as “C” but for inconsistent pairs.