New York City dataset for CPTED-based visual unsafety and recorded crime-count mismatch analysis

This dataset supports the study “Where crime is low but visual unsafety is high: CPTED-informed street-view evidence of spatial mismatches across New York City.” It contains 2,639 polygon grid cells at a primary spatial resolution of 600 m × 600 m in EPSG:32618. Of these, 2,202 cells contain street-view observations and constitute the valid analysis sample. Attributes include a CPTED-based visual-unsafety score derived from daytime street-view imagery captured during 2022–2023, 2023 NYPD recorded property- and public-order-crime counts, 2022 ACS-derived socioeconomic controls, image-derived urban-decay indicators, and six built-environment measures. The visual-unsafety score is a multimodal large-language-model appraisal of visible CPTED-related streetscape cues and is not a direct measure of residents’ lived fear. Raw street-view images and participant-level survey responses are not included. The GeoPackage preserves all source values, expands truncated Shapefile field names, and includes an internal field dictionary and provenance table. The analysis sample can be reproduced by filtering image_observation_count > 0.