Sensors for Social Science Research

In the first study, we use sensors developed by ETH spinoff, Bonsai Systems, to study solar light adoption and use patterns in rural, Western Kenya. We find that study participants used the solar lights quite a lot - about four hours per day. Comparing sensor data to survey data about daily light use, we find that although survey data has considerable measurement errors at the individual level, self-reported use does not differ from sensor measurements on average. In addition, households that used the light a lot tended to underreport, while households that used it little tended to overreport - a pattern known as mean-reverting measurement error.

Graph1 — The graph shows the comparison between solar light usage in each hour of the day with self-reported usage by study participants.

In the second study, we use motion sensors developed by Sensen to study how nighttime activity in one informal settlement changed in response to the COVID-19 lockdown in South Africa. South Africa instated one of the strictest lockdowns in the world during the first pandemic wave, but many worried that lockdowns would not protect residents of informal settlements given the high density, prevalence of shared sanitation infrastructure, and low-income status of most residents in these neighborhoods. At the same time, the media sensationalized reports of people in informal settlements flouting rules.

Based on the sensor data, we find nighttime activity decreased a lot - 40% in paths and nearly 60% in shared courtyards, called compounds - and stayed low through the first month of lockdown.

Graph2 — The graph shows weekly average motion (measured as a count for each five-minute interval). Pink dots represent the five-minute average for paths and blue dots represent the five-minute average for compounds. Between May and June, we had some sensor attrition, so the darker pink and blue dots represent averages from a smaller number of sensors.

Reductions were primarily driven by less activity during commute hours and on Saturday and Sunday nights, suggesting people both commuted and socialized less. Though it is impossible to completely curtail outdoor activity in an informal settlement, we find residents did appear to stay indoors more at night, in line with lockdown rules.

Although sensors are not without their own limitations, both of these studies provide insight into how sensors can be used to check for or overcome measurement challenges in social science research focused on a wide range of human activities going forward.

Researchers: Yael Borofsky, Adina Rom, Isabel Günther

Publications

Rom, A., Günther, I. and Borofsky, Y., external page Using sensors to measure technology adoption in the social sciences. Development Engineering, 2020.