The geography of road-crash co-location: space-time association of events in Warsaw and Chicago

Abstract

Road crashes of different kinds are not distributed independently of one another: some concentrate in the same places and within a short time of each other, while others clearly avoid one another. The tool for measuring such association is the co-location quotient (CLQ) in its space-time form (GTWCLQ), which compares the neighbourhood of one kind of event with what would be expected under a random arrangement. This paper is applied and didactic: it shows what the measure says about urban space, how to read its results and how not to over-interpret them. The material consists of all road events recorded in Warsaw between 2016 and 2024 (271,706 events, SEWIK register) and in Chicago in 2023 (109,545 events, municipal open-data portal). Three groups of results emerge. First, the scale of association is a property of the kind of event rather than of the city: potholes cluster at the scale of tens of metres and hours, pedestrian crashes along arterial corridors, and sideswipe collisions hardly at all. Second, the spatial pattern of pedestrian crashes is the same in both cities: the dense core yields values at or below the random level, while pronounced clusters lie on peripheral arterial roads. Third, the relation between two kinds of events can be one-sided: environmental categories such as animal and tree collisions occupy a subset of urban space, so urban collision types appear in their neighbourhood in ordinary proportion while the converse does not hold. This one-sidedness can be measured and tested, and the test also rejects one pair that looked one-sided on the raw values alone. The computation, including the improvements introduced here – the full scale surface instead of a single bandwidth, a reference distribution obtained by permuting type labels within blocks of similar terrain, and an explicit test of directional difference – is set out in plain terms in Appendix A, together with a twelve-item list of what to report when applying the measure to one’s own data.