Jessica Finlay, David Rigby, Amber DeJohn, Yue Sun, Brigette A. Davis, Arwa Aldulaimy, Desiree Alvarez‐McNelis, Ainsley Bowie, Karis Hawkins, Zhe Lin, Weining Kan, Taylor S. Ketterhagen, Xinyu Lin, Stephen B. Liwur, Mallory Sagehorn, Laine Sullivan, Lucy Vaughan, Shangrui Zhu, Ania Berry, Margaret T. Hicken, Michael Esposito
INTRODUCTION: Neighborhood "third places" are increasingly studied as contextual determinants of cognitive health, yet the reliability of geospatial datasets is poorly understood. METHODS: We evaluated Advan Research, Data Axle, FourSquare, and the National Establishment Time Series (NETS) across five categories: cafes/coffee shops, civic/social organizations, libraries, performing arts/museums, and recreation centers/gyms. Bayesian multilevel logistic regression models estimated locational accuracy and categorical validity for 13,168 coder ratings of 4876 unique businesses. Qualitative content analysis examined reflections from 18 coders. RESULTS: Advan showed high locational accuracy (≥95%), with Data Axle and FourSquare performing well for most categories and NETS the lowest. Inaccuracies stemmed from outdated or incorrect addresses and non-fixed locations. Advan, Data Axle, and FourSquare performed moderately well for categorization (70% to 97%) and NETS the worst. Misclassification reflected ambiguous purposes, misleading names, and uncertainty around third places. DISCUSSION: Study-specific dataset selection, triangulation, cleaning, error calibration, and clearer third place conceptualization are critical to strengthen neighborhood-based dementia research.