Filling in the white space: Spatial interpolation with Gaussian processes and social media data

Area of Science:

• Computational Social Science
• Psychological Measurement
• Geospatial Analysis

Background:

• Survey data often lacks granular geographic coverage for psychological variables.
• Large-scale surveys like the Gallup-Healthways Well-being Index provide limited county-level estimates.
• High response thresholds for survey data yield biased and reduced population samples.

Purpose of the Study:

• To develop principled methods for interpolating spatial estimates of psychological variables.
• To assess the utility of geotagged social media data in improving interpolation accuracy.
• To extend the spatial coverage of life satisfaction estimates to more U.S. counties.

Main Methods:

• Utilized Gaussian Processes (GP), a Bayesian modeling approach, for spatial interpolation.
• Combined limited survey data with large-scale geotagged social media (Twitter) language data.
• Evaluated interpolation performance across geographic, socioeconomic, and language-based spaces.

Main Results:

• Gaussian Processes (GP) effectively interpolated life satisfaction estimates for a significantly larger number of counties.
• Geotagged Twitter language use improved interpolation accuracy compared to traditional socio-demographic and geographic measures.
• The enhanced method maintained convergent validity with external criteria, extending reliable estimates to 2,954 counties.

Conclusions:

• Bayesian techniques, particularly Gaussian Processes (GP), can reliably extend the spatial coverage of psychological variables.
• Geotagged social media language provides valuable insights into cultural similarity, enhancing traditional data.
• Open-sourced tools are available to facilitate the adoption of these advanced interpolation methods by researchers.