Comparing the Performance of Spatial Centrality Measures Through Predicting Improved-Property Values: Evidence from Porto Alegre, Brazil

This is the authors' abstract. We don't add key points for this paper.

Several spatial centrality measures have been proposed to capture the differentiation of urban space. Accessibility measures appraise ease of access to other locations; intermediation measures describe a location’s control over network movement. Commonly used measures overlook attractors and the decay of interaction with distance, and preferential centrality and Freeman–Krafta centrality were introduced to incorporate, respectively, the agglomeration of activities and distance decay. This article compares five measures – closeness, betweenness, Freeman–Krafta centrality, and preferential centrality with and without its agglomeration term – in Porto Alegre, Brazil. The dependent variable is improved-property value per square metre of land, built from 2022 property transactions and the municipal unit registry. The comparison is made for 7,775 address records and, after aggregation, for 812 hexagonal cells of 500 m. Performance is defined once, as the coefficient on standardized predictors and the trend-only prediction error under spatially blocked cross-validation. In the headline specifications the three accessibility measures are more closely associated with the outcome than the two intermediation measures at both levels, and preferential centrality with agglomeration has the largest standardized coefficient and the lowest prediction error at both, although its advantage over the version without agglomeration is small and does not hold under every specification. Betweenness and Freeman–Krafta centrality are indistinguishable from zero at the address level and positive and significant at the 500 m cells. Built-area controls substantially attenuate the accessibility associations, highlighting the importance of building intensity and property composition.