Measuring "third places": Comparing neighborhood data for cognitive health research

Area of Science:

• Urban Health
• Geospatial Data Science
• Cognitive Epidemiology

Background:

• Neighborhood "third places" are increasingly recognized as crucial for cognitive health.
• However, the reliability of geospatial datasets used to identify these places is not well understood.
• This gap limits the accuracy of neighborhood-based health research.

Purpose of the Study:

• To evaluate the locational accuracy and categorical validity of four major geospatial datasets.
• To compare the performance of Advan Research, Data Axle, FourSquare, and the National Establishment Time Series (NETS).
• To identify sources of error in geospatial data for neighborhood third place research.

Main Methods:

• Assessed four datasets (Advan, Data Axle, FourSquare, NETS) across five business categories.
• Employed Bayesian multilevel logistic regression for locational accuracy and categorical validity.
• Utilized 13,168 coder ratings of 4,876 unique businesses and qualitative coder reflections.

Main Results:

• Advan Research demonstrated high locational accuracy (≥95%).
• Data Axle and FourSquare showed moderate accuracy and categorization, while NETS performed poorly.
• Common errors included outdated addresses, non-fixed locations, and ambiguous business classifications.

Conclusions:

• Dataset selection, data triangulation, and cleaning are vital for reliable neighborhood dementia research.
• Clearer conceptualization of "third places" is needed to improve data accuracy.
• Addressing geospatial data limitations is critical for advancing cognitive health research.