Quantitative Classification of Neighbourhoods: The Neighbourhoods of New Single-family Homes in the Portland Metropolitan Area

Abstract

This paper develops a quantitative method for classifying neighbourhood types and applies the characterization method to the neighbourhoods of new single-family homes in the Portland, Oregon, metropolitan area. The study first measures a set of 21 urban form attributes in neighbourhoods that contain newly constructed single-family homes. Factor analysis is then used to identify a small set of dimensions that capture essential differences in urban form. Finally, these factor scores are used as input to a cluster analysis to identify distinct neighbourhood types. The results demonstrate that most new single-family homes in Portland are built in new suburban neighbourhoods, but a substantial portion of single-family construction is occurring in traditional and neo-traditional urban neighbourhoods.

Notes

1. The literature on the relationship between urban form and travel behaviour and physical activity is too large to review in detail here. A few examples of how neighbourhood classification is used in these studies are offered here. See Handy (2004) for a comprehensive review.

2. To proceed with the calculations of variables such as Nbr_BlockSize, Nbr_LotSize, Commercial, Industrial, Public, OpenArea and Mfr, the first step is to identify those blocks, plots and commercial, industrial, public, open and multi-family land parcels with their centroids within the ¼-mile buffer area of the structures. Then the original values of acre (or square feet) of the blocks, plots and other land parcels identified as having the median values are retrieved. Therefore it is possible that the area values of these variables are larger than the area value of a ¼-mile buffer.

3. Varimax is used to maximize the variance of the squared loadings. Varimax is an orthogonal rotation method that simply rotates the axes of the first factor to a variable or group of variables and then rotates the subsequent factors to be at right angles (uncorrelated) with the first. By this way it removes the effects of variables that could be highly loaded on the first factor. Compared to unrotated factor solution, an orthogonal rotation minimizes the number of samples needed to account for the variation of distinct groups of variables.

4. K-means clustering begins with a grouping of observations into a predefined number of clusters. It evaluates each observation and moves it into its nearest cluster. The nearest cluster is the one that has smallest Euclidean distance between the observation and the centroid of the cluster. When a cluster changes by losing or gaining an observation, the cluster centroid is recalculates. At the end, all observations are in their nearest cluster.

5. Cluster analysis procedures are affected by the magnitude of the variables included, that is to say, variables with large numbers have a greater impact on the outcome of the analysis than variables with small magnitudes. To control for this imbalance, scaling is necessary to convert the original variable values to standard scores. Since the eight factor scores derived from the factor analysis are used here in the cluster analysis, the magnitude of the variables is not a concern here.

6. The ‘centroid’ of a cluster is the centre of gravity for the respective cluster and is the average point in the multidimensional space defined by the dimensions.