Nicholas Johnson, Bartosz Bonczak, Arpit Gupta, and Constantine E. Kontokosta have submitted a working paper to the SSRN repository, "Working 9 to 5? Measuring Hyperlocal Worker Productivity with Public WiFi Network Data."
The accurate estimation of workday length is essential to estimate total labor supply, and has a significant bearing on the assessment of labor productivity and worker well-being. Using probe request data from a 53 access-point, publicly-accessible Wi-Fi network in the Lower Manhattan district of New York City, we develop a method to measure localized worker activity patterns. Our Wi-Fi network data consist of over 10,000,000 probe requests per day, accounting for approximately 9.5 million unique devices over the study period from April 2017 to September 2017. We describe worker activity at various spatial and temporal aggregations in order to define baseline workday patterns and compute the workday length. We find a substantial population with characteristic workday lengths (e.g. 9am-5pm) during the workdays, as well as diurnal activity patterns that are consistent with expected worker behavior. These temporal patterns provide sufficient evidence to reinforce our assumptions about the ability to identify worker populations from Wi-Fi data. Finally, we compute the workday length for each identified worker and aggregate these workday lengths to estimate collective workday patterns to understand the uniformity of worker behavior. We find workday lengths of 7 hours and 40 minutes on average, which shorten substantially on Fridays and days surrounding holidays. We also find considerable seasonal variation in total workday hours supplied in our study area. This dynamic pattern of hours-worked suggests that our methodology is able to accurately assess workday lengths at high spatial resolution and temporal frequency. The ability to quantify hyperlocal worker activity patterns has a broad range of applications, including estimates of localized economic output and changes in labor supply.