A Sum of Places: Exploring Street Segment Composition as a Risk Factor of Crime Hot Spots

Zach Drake

Advisor: Charlotte Gill, PhD, Department of Criminology, Law and Society

Committee Members: David Weisburd, Beidi Dong, Martin Andresen

Online Location, #online
October 07, 2026, 10:00 AM to 12:00 PM

Abstract:

Crime concentrates tightly at micro-places. Most place-based criminology has treated street segments as homogeneous units or examined individual places in isolation. Research using street segments as the unit of analysis has treated street segments as opaque containers of routine activities. Other research using individual places as the unit of analysis has mostly ignored how co-located places might affect each other in regards to criminal opportunities. Routine activities theory, crime pattern theory, and even social disorganization theory all imply to various degrees that places near each other impact each other.

This dissertation examines whether a street segment can be better understood as the sum of its places, an aggregation of the routine activities of the individual places along it. The composition of the street segment in terms of the types of places that sit along it are tested as predictors of whether the street segment is a crime hot spot. This is explored using simple counts of place types, interaction effects of frequently co-located place types, and using latent combinations of all possible place types. Hot spots are computed and labeled across four crime categories independently, violence, larceny, burglary, and drugs. Whether a street is a hot spot is computed using three different classification methods including traditional 25\% and 50\% cumulative proportion thresholds as well as Jenks natural breaks classification. Both arrest and incident hot spots were predicted independently to assess how place composition affects both crime and crime policy. Binomial logistic regression, augmented by co-occurrence analysis for interaction effects and principal component analysis for identifying latent combinations, is used to assess the streets composition as a predictor for each crime category and hot spot classification method combination. False discovery rate correction and spatial autocorrelation diagnostics were applied to each model to counteract the effects of the dense dataset and multiple comparisons.

The findings suggest that place composition has an additive effect on whether a street is a crime hot spot, but the interactions of place types and the latent classifications do not have much of an effect. Several types of places, including shopping, food and drink, and lodging establishments, consistently predicted hot spots, with the effect strongest for property crimes and weak for violence and drugs. Combinations of place types did not amplify risk beyond the individual effects. Principal component analysis revealed weak latent structures, indicating that commercially derived place categories may not capture the dimensions by which street segments vary in meaningful ways. The Jenks natural breaks classification consistently produced stronger model fit than arbitrary thresholds, though at the cost of statistical power for identifying individual predictors. Arrests showed weaker or non-existent relationships, with many models failing to converge, implying that any relationship between street segment composition and crime is at best only applicable to crime incidents themselves.

This study suggests thinking of street segments as the sum of their places is a viable line of inquiry. The additive effects of the count models, the absence of strong combination effects, and the weak latent structure in place categories together indicate that this approach has potential, but the effect is likely nuanced and limited. Further development of this research depends on several advances the field will need to pursue. These include place data collected and aligned with crime over time rather than as a single snapshot, place taxonomies grounded in routine activities and opportunity structures rather than in commercial search, and extension to cities with more varied land use patterns where place composition may exhibit stronger latent structures. This dissertation details this way of thinking, assesses what it can and cannot yet explain, and outlines the data and conceptual infrastructure that would allow place composition analysis of crime hot spots to mature.