Subjective or objective measures of street environment, which are more effective in explaining housing prices?

Abstract

Houses with better street design are found to relate to a price premium. Prior studies mainly present the street quality using objective indicators like tree counts and distance to parks with land use data, or most recently using the greenery view index extracted from street view imagery (SVI). We argue that objective indicators cannot completely describe people’s sense of a place, as perception is a highly subjective process. We hypothesize that subjective measures using visual surveys could capture more subtle human perceptions, thus providing stronger predictive power to housing prices. However, the role of subjectively measured street design qualities is less known due to the lack of large-scale perception data. To test our hypothesis, we first collected designers’ perceptions on five urban design qualities from pairwise SVIs rankings in Shanghai with an online visual survey. Unlike the mainstream of using generic image features, we followed urban design theory and used rule-based features, i.e., about thirty streetscape elements extracted from SVIs to train machine learning (ML) models to predict subjective perceptions. The predictive power of five qualities versus ten selected individual streetscapes on housing price were compared using the hedonic price model. Besides the standard ordinary least squares (OLS), spatial regression and geographical weighted regression (GWR) were also developed to account for the spatial dependence and heterogeneity effects. We found both subjectively measured design qualities and objective indicators outperformed housing structural attributes in explaining housing price. While the objective view indexes collectively explained more price variances, the five perceptions individually exhibited stronger strength. Third, less-studied perceptions like “human scale” showed stronger strength than commonly studied “safety” and “enclosure”. Fourth, less-studied view indexes like “person” and “fence” outperformed ubiquitous features like trees and buildings. Lastly, prior studies might have resulted in biased estimations due to ignoring the multicollinear issues between the sky, tree and building views. Our study addressed the effectiveness of incorporating subjective perceptions at a micro level to infer housing prices. Correlations between subjective perceptions were strong while that of objective indicators were negligible, therefore subjective perceptions can complement the objective indicators. The findings provide important reference to decision makers when selecting street quality indicators to infer urban design, city planning and community and housing development plans.