Revisiting the Empirical Foundations for Measuring the Capitalization of Access to Transit

摘要

In METRANS project number: 04-18, Transit Investment & the Capitalization of Access into Land Values, it was shown that the typical statistical tool-- standard heconic regression--used for assessing the capitalization of the transit access had potentially fatal flaws for use in informing policy making. Indeed, I showed that a set of commonly-applied hedonic analyses of house prices and transit capitalization produced highly unstable results in an urban context in which distinct housing submarkets existed in close proximity. This finding may explain the existing literature on capitalization of access to passenger rail; this large literature reveals no consistent estimate of access capitalization into surrounding land and property values. This lack of clear guidance is highly problematic for policy makers looking to make sound policy with regard to transit infrastructure investments. The specific problems have their origins in the standard tools of assessing capitalization in the presence more than one housing submarket in the data--a common occurance in urban areas. The solution I proposed in METRANS project 04-18--a more flexible, non-parametric approach--left several questions unanswered. With regards to the technical aspects of the non-parametric approach, the project provided little guidance as to formal hypothesis testing. In light of feedback the previous research generated, it appeared that this weakness could be addressed. The first part of this research project (METRANS 08-07) formalizes the statistical basis for non-parametric analysis of capitalization; the second would bring it to data. This first step is an essential element of gaining acceptance in a field dominated by the (demonstrably unstable) hedonic regression. The second step was initally proposed to revisit the Los Angeles data used in the previous research. In the end, a new data set from Chicago was used to provide broader empirical support for the findings.