Constructing the Community Typology
Thirteen variables were used: five housing, six socioeconomic, and two household characteristics. The housing variables were percent of units built before 1940, percent of units built after 1995, percent vacant, percent detached single units, and percent owner occupied; the socioeconomic variables were percent African American, percent with less than a high school education, percent with a bachelor's degree or better, percent of persons less than 150 percent of the poverty line, percent working outside municipality of residence, and percent males unemployed; the household variables were percent of families with children under eighteen and percent of families which were female headed. Although we have spent considerable effort trying to improve the typology using measures of taxation, land use, and density, the alternatives proved statistically and intuitively inferior.
Communities were defined differently for the city and suburbs. To define communities in the city, we used the twelve planning analysis districts that the Philadelphia Planning Commission has historically used in its work; in the suburbs, the communities are the municipalities.
The validity of the typology was assessed three different ways: (1) its criterion validity was assessed using a variety of measures such as the percentage of households with incomes above $75,000; (2) the classifications were analyzed through a very different statistical methodology—a multiple discriminant analysis—using the same thirteen variables with and without random elimination of observations (the initial analysis correctly classified 91.8 percent of the cases; the random elimination analysis correctly classified 88.7); and (3) between the 2004 and 2005 reports, we sent the results of the 2004 cluster analysis to our Project Advisory Committee, county planning officials, and other regional specialists for comment. As a result, we reclassified 5 of the 364 communities for 2005, reflecting instances where group quarters distorted the classification, where there was a U.S. Census Bureau locational error, and where in a couple of instances census data did not reflect the situation “on the ground.”