University Campus Bicycle Theft: A Curated, Subject Matter Expert–Validated Dataset with Event-Level Variables Extracted from University of Arizona Police Report Narratives, 2024-2025

This event-level dataset contains bicycle theft–related incidents derived from police reports provided by the University of Arizona Police Department (UAPD). The dataset was constructed via our information-extraction pipeline that generates structured variables from unstructured natural-language report narratives, producing a machine-readable resource for quantitative analysis of campus bicycle theft. The dataset was approved for public release by the University of Arizona Police Department. All extracted variables were manually validated by subject matter review. The current release contains 472 incidents from 2024–2025. This temporal scope was selected to reduce major confounding changes: (i) post-2023 shifts in campus course modality, (ii) a transition in records management systems at UAPD, and (iii) updates to narrative writing protocols. Selected variable definitions: Report Subtype: Incident classifications include larceny of bicycles and bicycle parts, and larceny of electric bicycles and scooters. When narratives explicitly indicated probable cause for an attempted bicycle theft, related reports classified as suspicious activity, burglary, vandalism, or narcotics were included. Cases labeled suspected larceny were excluded when probable cause was not supported in the narrative. Attempted: Incidents with strong probable cause for theft in which the bicycle was not ultimately taken. No Lock: No mention of any locking or securing mechanism in the report narrative. Other Lock: Any security method not categorized as a chain, U-lock, or cable lock, including bicycle lockers/enclosures, unspecified lock mentions, or thefts from locked dorm rooms. Event Location: Reported location of the incident. USD Total Value: Reported U.S. Dollar amount of the bicycle, lock, or associated damages. Approximately one-third of incidents lack value estimates; missing values were not inferred because victims frequently did not know the cost. Setting these values at "$0.00" maintains integrity of police report narratives. An updated release including 2026 incidents is anticipated in early 2027. The associated Python extraction pipeline will be released as open-source software following manuscript review. Files are under embargo until the associated preprint is publicly available, after which the dataset will be released openly.