Controlling for confounding effects in spatial flow analysis with partial and full conditional FlowLISA methods

Abstract

Spatial flows refer to movement phenomena such as human migration and commodity exchanges between different regions. In recent years, various spatial analysis methods targeting flow data have been proposed. Nevertheless, the confounding effects are rarely addressed, which is likely to cause misinterpretations of the patterns that flows inherently exhibit. This study proposes a pair of flow-based Local Indicators of Spatial Association (LISA) methods, namely Partial and Full Conditional FlowLISA (PC-FlowLISA and FC-FlowLISA), to identify spatial flow patterns of network autocorrelation while controlling for multi-sourced confounding effects. We use publicly available taxi trip data in New York City to evaluate the effectiveness and demonstrate the usefulness of our methods. The results show that whether and how to control for the confounding effects can significantly influence the identification and interpretation of spatial flow patterns. The proposed methods help provide a crucial perspective for understanding the underlying mechanisms behind spatial interaction phenomena.