What’s sauce for the goose is sauce for the gander? Near-repeat victimization across different urban morphologies: Evidence from Almaty, Kazakhstan

Near repeat victimization (NRV) is a well-documented phenomenon, yet little is known about how it operates not across cities as a whole but within homogeneous urban morphologies. This study analyzes spatial and temporal patterns of property crime incidents from 2023 across urban morphology belts in Almaty, Kazakhstan. We use both the Knox test and interpretable machine learning (XGBoost with SHAP values) to predict victimization risk across three urban built environments (multi-level apartments, low-income detached houses, and high-income detached houses) based on prior incidents within spatio-temporal windows. The results show that although we observe partial evidence of the “Law of Crime Concentration”, overall NRV effects appear weak in Almaty. The machine learning approach further indicates that close-proximity crimes do not change recurrence risk, whereas more distant events in space and time are linked to higher odds of reoccurrence. Instead of a conventional “near repeat,” we observe a form of “distant repeat” victimization. This pattern suggests a specific type of crime displacement, where the most attractive target is not the site of a prior offence, but one that draws less attention from law enforcement. In addition, the morphology clusters do not differ substantially from one another — the key divide is not between grid-like and organic layouts per se, but between the high-class detached areas in the foothills and the rest of the city. These findings underscore the need to evaluate NRV in non-Western contexts using diverse methodological approaches, with explicit consideration of the urban morphology in which crimes occur. • Near repeat victimization (NRV) is not observed in the non-Western context of Almaty, Kazakhstan. • A distant repeat pattern emerges, driven by effects from outer time-space belts. • The main NRV divide lies between high-class foothill areas and the rest of the city, not grid vs. organic layouts. • Interpretable machine learning (XGBoost + SHAP) yields more nuanced results than the Knox test. • The Law of Crime Concentration holds: 50% of crimes occur within 5% of urban space.