Neighbourhood avoidance, residential turnover and measured segregation in an agent-based urban growth model

Residential preferences and aggregate segregation need not vary in the same direction when urban growth involves continuing relocation and settlement consolidation. This study extends an agent-based peripherisation model motivated by Latin American urban dynamics. Three income groups compete for residential locations near high-income settlements, which act as proxies for access to urban amenities. The extension introduces neighbourhood-dependent consolidation of low-income settlements, avoidance of low-income neighbours during residential search, and neighbourhood-triggered relocation of middle- and high-income residents. Reported sensitivity experiments use a toroidal grid of 51 × 51 cells and ensembles of 100 runs, assessed through spatial dissimilarity, spatial information theory, and exposure/isolation indices. Lower neighbourhood-tolerance thresholds substantially increase the reported mean number of abandonment events, from 44.00 at a threshold of eight to 2369.80 at four. Nevertheless, the strongest avoidance scenario has lower global dissimilarity than intermediate thresholds, particularly when consolidated settlements are present initially. Reported trajectories and exposure profiles suggest that repeated relocation opens settlement opportunities and sustains spatial contact between groups. Longer post-target search also reduces measured segregation. These findings concern a stylised, uncalibrated simulation rather than a forecast for a particular city. They show why spatial segregation measures should be interpreted alongside residential turnover, group composition and consolidation, rather than as direct indicators of residential tolerance or equitable urban outcomes. Research preprint, not peer reviewed. Derived from a 2022 master’s dissertation. No new simulation experiments were performed during article preparation. Generative AI assistance is disclosed in the manuscript. The deposit of manuscript files does not provide the original simulator or run-level research data.