Enriching Survey Data with Big Geodata: A Computational Approach to Intergroup Dynamics and Neighborhood Contexts

This dissertation examines how neighborhood context shapes interethnic group relations. The current empirical evidence on the intergroup relations-neighborhood nexus reveals two central research gaps that guide this work. First, there is a need to expand the range of available neighborhood context indicators. Second, there is a need to move beyond coarse administrative geographies as approximations of neighborhood context toward smaller-scale, more individualized approximations of neighborhood context that may better capture local dynamics. To address these gaps, this dissertation employs an innovative methodological approach. It augments georeferenced survey data from Belgian National Election Study 2020 with big geodata sources and applies computational social science methods (i.e., supervised machine learning and GIS algorithms). This dissertation draws on several big geodata sources. The primary dataset is OpenStreetMap, which forms the central focus of the dissertation. Additional datasets include georeferenced survey data, geotagged Twitter/X data, and Google Maps data. To address the first gap, the dissertation focuses on novel, big geodata-based built environment neighborhood indicators of neighborhood context, such as spaces of encounter (e.g., parks, schools, cafés) and Islamic points of interest, together with more established administrative indicators, such as ethnic diversity and socioeconomic status. To address the second gap, the dissertation employs both small-scale administrative geographical units and custom-made GIS-based geographical units, such as walking- and driving-time neighborhood contexts as approximations of immediate neighborhood context. First, the findings show while novel built environment indicators enrich understanding of neighborhood contexts for intergroup relations, they complement rather than replace established contextual indicators, such as socioeconomic and ethnic diversity. Likewise, by augmenting georeferenced survey data from Belgian National Election Study 2020 with GIS-based walking- and driving-time neighborhood approximations, the results demonstrate that the relationship between neighborhood indicators and intergroup outcomes depends on how neighborhood context is approximated and measured. Second, the dissertation also demonstrates how big geodata from OpenStreetMap, as a central data source, can enrich survey data contextually and enable the inclusion of geocontextual variables that were previously inaccessible or complicated to capture. It provides a methodological guide on how to use OpenStreetMap for survey data augmentation. Overall, the dissertation highlights that neighborhoods are multidimensional contexts and should be treated as such. Similarly, it demonstrates the methodological value of augmenting survey data with big geodata and the use of computational methods. It offers concrete use-cases for both neighborhood context indicator operationalization and neighborhood context approximation.