Evaluating Demographic Clusters as a Relevant Unit of Aggregation for Neighborhood Effects

Jonathan Tannen, Princeton University

Neighborhood studies have traditionally relied on fixed boundaries--usually the census tract--as the unit of aggregation, with acknowledged imprecision. Despite this, the literature has long understood neighborhoods--explicitly and implicitly--as clusters of racially similar households. This paper proposes a new unit of aggregation for measuring neighborhood effects by directly estimating those clusters, both as a potential solution to the well-known Modifiable Areal Unit Problem and for larger insights into the spatial boundaries of neighborhood effects. I develop a method for identifying fine-resolution racial clusters in US Census data using a Bayesian clustering model. I then examine the properties of these clusters compared to tracts and block groups, and test their explanatory power in predicting crime.