Fig 1.
Percentages of study time the app and badges were active.
Left panel shows the activity of the smartphone app. Right panel shows the activity of the sociometric badges. Only office hours (Mon-Fri 9am-5pm) during the 4-week period were considered.
Fig 2.
A) Percentage of scheduled Bluetooth scans that were made by each smartphone in the 4-week period. B) Scanning rates for each edge of the network. The scanning rate between node A and B is determined by the number of scans made by smartphone A and B combined, as the edges are undirected (symmetric). The horizontal solid line reflects a scanning rate of 1 scan every 15 min; the dashed line 1 scan every hour.
Table 1.
Contingency table across all office hours.
Table shows the number of times a particular edge of the network was detected (hit) or not (miss) by the smartphone app and the sociometric badges. Only office hours (Mon-Fri 9am-5pm) during the 4-week period were considered.
Table 2.
Contingency table when the app and badge are both active.
Table shows the number of times a particular edge of the network was detected (hit) or not (miss) by the smartphone app and the sociometric badges. Only time intervals when both the app and badge were active were considered.
Fig 3.
Weighted adjacency matrix of the social network mapped using the smartphone app and the sociometric badges.
Both axes reflect the 21 participants and each element reflects the percentage of time the two participants were in close proximity. A) Office hours, B) When both devices were active.
Fig 4.
Social networks mapped using survey data, smartphone app and sociometric badges.
To facilitate comparison the incoming and outgoing edges of from the survey data were combined to obtain an undirected network. In addition, the binary backbone was extracted from the weighted adjacency matrix of the smartphone app and the sociometric badges. A) Adjacency matrices for the survey data, smartphone app and sociometric badges, B) Topological representation of the corresponding networks. Node size represents its degree. Layout was rendered using the Kamada-Kawai algorithm.
Fig 5.
Correlation between resampled networks.
A) The weighted adjacency matrix of the app network was constructed using a fixed number of random samples for each participant to investigate potential biases resulting from unequal scanning rates. The number of required samples was varied from 10 to 500 samples. The Mantel test was again used to estimate the correlation with the networks constructed using badge and survey data. Colour patches show the 99% confidence interval estimated by resampling the network 1000 times. B) The size of the network that was compared decreased with increasing number of required samples, as participants with insufficient number of scans were excluded.