Can consumer big data reveal often-overlooked urban poverty? Evidence from Guangzhou, China

Abstract

In the evolving landscape of poverty research, especially in China, the focus has shifted from eliminating absolute poverty to relieving relative poverty. Although much of the existing studies have begun to use built environment big data, such as remote sensing and street view imagery, to measure poverty, peoples' consumption, an essential indicator of poverty receives less attention. This study delves into the relationship and spatial disparity between poverty measured by consumer big data and multidimensional poverty measured based on the census data. We investigated 1731 communities in Guangzhou as case study regions and combined their residents' mobile phone metadata and spatial cost of living data as the input consumer big data. Then, we constructed Index of Multiple Deprivation (IMD) levels based on the census data and built random forest classification model based on our consumer big data to predict IMD level at community level. The result shows that the predicted poverty of 81.11% communities were generally consistent with the IMD level, indicating that the consumer big data poverty mapping provided a viable poverty measurement from consumer behavior perspective. The SHapley Additive exPlanations' values result shows that Pinduoduo (a low-cost online shopping mobile application) contributes the most to predicted poverty from consumer behavior. For spatial disparities, poverty mapping based on consumer big data is more sensitive to the poverty in suburban developing neighborhoods and affordable housing communities compared with the IMD. The urban poverty mapping based on consumer big data offers a timely portray of communities' socio-economic challenges and consumption-related poverty, and provides support and evidence for accurate urban poverty alleviation strategies.