A comparison of the approaches for gentrification identification

Abstract

Gentrification can be identified via a threshold-based method and/or a machine-learning approach. The former, which is simple and theoretically sound, is complementary to the latter, which is objective. In view of a lack of research on exploiting the strengths of both approaches, this study compares a threshold-based method to K -means clustering. Using the city of Auckland as a case study, we find that both approaches are in accord with each other. The maximum degrees of similarity (falling in the range 0–1) between the identification results of both approaches are 0.80 and 0.56 for binary and three-level identification, respectively. By comparison, it is evident that the threshold-based set of identification rules delineates gentrification more accurately. For example, a census tract with a confluence of housing reinvestment and at least one aspect of social upgrading is more likely to be identified as gentrified. Moreover, gentrification in Auckland assumes various appearances. Retaining a simple and universal conceptual and analytical framework for gentrification helps us focus on the essentials of this urban phenomenon: reinvestment and displacement.