Understanding Geographic Bias in Crowd Systems

Abstract

Crowd platforms are increasingly geographic, from the sharing economy to peer production systems like OpenStreetMap. Unfortunately, this means that existing geographic advantages or disadvantages (e.g. by income, urbanness, or race) may also impact these crowd systems. This thesis focuses on two primary themes: (1) how these geographic advantages and disadvantages interact with crowd platform services, and (2) how people’s geographic behavior within these platforms may lead to these biases being reflected. The first chapter in my thesis finds that sharing economy services fare less well in low-income, non-white, and more suburban areas. This chapter introduces the spatial Durbin model to the field of HCI, and shows that geographic factors like distance, socioeconomic status and demographics inform where sharing economy workers provide service. The second chapter in my thesis provides focuses on people in peer production communities contribute geographic content. By considering peer production as a spatial interaction process, this study finds that some kinds of content tend to be produced much more locally than others. Finally, my third contribution focuses on individual contributor behavior, and shows geographic “born, not made” trends. People tend to be consistent in the places, and kinds of places (urban, and non-high poverty counties) they contribute. The findings of this third study help identify mechanisms for how geographic biases may come about. Looking forward, my work helps inform an exciting agenda of future work, including building systems that provide individual crowd members sufficient geographic context to counteract these worrying geographic biases.