A standard statistical correction used by city planners for decades actually makes their research less accurate.
Urban planning studies that try to control for personal preferences to remove self-selection bias are mathematically flawed. Researchers typically assume that removing these biases is necessary to understand how city design affects travel behavior. Mathematical proofs demonstrate that the common method of conditioning on these preferences creates a collider that amplifies the very errors it tries to fix. This flaw means that many of the assumptions about how bike lanes or transit-oriented developments change habits might be based on bad math. Correcting this mistake will force a total rethink of how we measure the impact of new infrastructure.
Fallacies of residential self-selection bias in travel behavior research: a reconceptualization
SocArXiv · hku4d_v2
Residential self-selection (RSS) is widely recognized as a source of bias in studies of how the built environment affects travel behavior. The standard remedy is to control for individual travel attitudes, residential preferences, and sociodemographic characteristics. We show, however, that this approach misidentifies the nature of the bias and can worsen it. Using Directed Acyclic Graphs (DAGs) to represent the causal structure of the travel behavior data-generating process, we demonstrate that