Statistical
Two-sample KS testing was performed with alpha = 0.05 for each. First, distributions of "first seen" times were compared on both weekdays versus weekends as well as between weekdays to determine whether the data were from the same source distribution (null hypothesis). The KS statistics and p-values for both, respectively, were (0.001, p=0.743) and (0.141, p=0.0), indicating that the null hypothesis cannot be rejected for similarity of weekday arrival times, but that a statistically significant difference exists between weekdays and weekends.
KS testing was also performed under the same conditions for duration in the network. The KS and p-values were (0.022, p=1.359e-111) and (0.154, p=0.0), indicating that duration in network does vary both across weekdays and weekends as well as individual weekdays.
Future work:
Future work could easily include increasing temporal granularity of analysis. For example, when a larger range of data are collected that include a full year of data with ranges of temperatures and weather types, it would be interesting to analyze short-term effects of weather on the arrival, departure, and duration of clients in the network. More generally, quantitative trends in the significant changes in duration, arrival, and departure in the network could be observed. Integrating these observations with external data for the region, including construction, air quality, and indicators such as holidays or events such as financial crises or storms, would provide further insight into the pulse of Lower Manhattan and its patterns of behavior.
Links:
Bibliography:
Campbell, A. (02 August 2006). People-centric urban sensing. WICON '06 Proceedings of the 2nd annual international workshop on Wireless internet. Retrieved from
dl.acm.org