Research Methods

This research project uses traditional archival research in conjunction with geographic information systems (GIS) and spatial statistical methods to analyze mortgage lending patterns in Philadelphia between 1930 and 1960. In addition to determining how redlining affected Philadelphia's neighborhoods, this research also demonstrates methods that can be used to analyze current and historical redlining and other urban issues.

Scroll to:
Collecting Mortgage Data
Mapping Neighborhood Characteristics
Digitizing HOLC Maps
Analyzing Lending Patterns


Collecting Mortgage Data (return to top)

Finding detailed data on mortgages is one of the most challenging aspects of research on historical redlining. The Home Mortgage Disclosure Act (HMDA) currently requires that certain types of lenders provide data about where and to whom they make loans. However, no such mandate existed prior to HMDA's enactment in 1975. The decennial US Census has provided information about the age of housing, housing tenure (renter/owner), housing values, and duration of residence since 1940, but it does not include any information about lending. The Works Progress Administration Real Property Surveys of the 1930s, which provide similar data to the US Census, also does not include information about lending. So address-level mortgage data were collected from the Philadelphia Realty Directory and Service and the City of Philadelphia Archives.

The Realty Directory
The Philadelphia Realty Directory and Service, contains detailed information about all property transactions in the city of Philadelphia and is organized by date and address of the transaction. Random samples of all mortgages made between 1938 and 1960 were gathered from the Realty Directory for this project. Data include the address of the property, name of owner, address of owner (when owner lives off-site), name of lender, amount of mortgage, interest rate, and sales price ("consideration"). The Realty Directory also includes the date of the most recent transaction and appraised value for each property in the city for each year. The Realty Directory is available at the Free Library of Philadelphia for the years 1926-1959.

The Philadelphia City Archives
The mortgagee records at the The Philadelphia City Archives were also used because they index mortgages by the name of the lender, making it possible to collect a sample of loans made by a specific lending institution. Using these records is more time-consuming than the Realty Directory because it involves two stages. First, we used the mortgagee index to identify a particular loan. Using the clerk name, volume, and page number identified in the index, we were then able to request the actual mortgage instrument in order to obtain information on the address and loan terms. Using these records, we were able to collect random samples of loans made by the Home Owners' Loan Corporation, Berean Savings & Loan Association, Metropolitan Life Insurance Company, Colonial Mortgage Service Company, and South Philadelphia S&L.

Mapping Neighborhood Characteristics (return to top)

Data about the racial composition, income, and housing characteristics of neighborhoods were obtained from the 1934 Real Property Survey of Philadelphia (census-tract level) and the Bogue files of the 1940, 1950, and 1960 US Census of Philadelphia. The shapefile corresponding to census tracts for 1930 and 1940 was digitized for this project, but these shapefiles are now available from the Van Pelt Library, along with the Census attribute data tables.

Digitizing HOLC maps (return to top)

The HOLC maps were photographed at the National Archives II by a private vendor. The color photographs were then scanned and georectified in ArcView GIS 8.2, allowing them to be layered along with other maps. In order to create a more useful vector representation, each version of the HOLC map was also digitized. The areas defined by HOLC do not correspond to other political boundaries (such as wards) or administrative units (such as census tracts) from that time, but they did fall along streets. Rather than using online digitizing tools to draw HOLC boundaries, a 1990 census block shapefile was used and the boundaries among the census blocks that made up each HOLC area were dissolved. This ensured that the new HOLC shapefiles would correspond to current shapefiles for Philadelphia.

Analyzing Lending Patterns (return to top)

The mortgage lending data have been analyzed in several different ways for this project. A random sample of loans made by private lenders between 1938 and 1950 were mapped on top of the final (1937) HOLC map to determine if loans were made by private lenders in or near red areas. The resulting maps indicated that they were. See Hillier, Journal of Planning History [Hillier, Journal of Planning History]."

Spatial lag and spatial binary probit models were used to determine if the HOLC grade or distance to a red area on the HOLC maps affected the number of mortgages made on each property, the interest rate of the mortgage, or the type of lender for a random sample of loans made by private lenders between 1938 and 1950. These spatial models allow for observations (individual mortgages) to be spatially autocorrelated, unlike traditional OLS regression models, which assume independence. The number of mortgages and type of lender were not significantly related to the HOLC grade, but mortgages made in areas with higher HOLC grades did have higher interest rates. The spatial statistical programs used were written for Matlab by Tony E. Smith [Hillier, Journal of Urban History]. This same method was used to determine if the racial composition of an area affected the interest rate or loan amount for the same random sample of loans. Results varied [Hillier, Journal of Housing].

Ripley's local K functions were used to identify areas that received significantly fewer mortgages than would have been expected relative to the number of owner-occupied housing, based on the analysis of a random sample of loans made by private lenders between 1938 and 1950. The local K function program used was written for Matlab by Tony E. Smith [Hillier, Journal of Housing]. Ordinary Kriging was conducted with Geostatistical Analyst in ArcView 8.2 to interpolate a surface of interest rates, based on a random sample of loans made by private lenders between 1938 and 1950. Using this approach, contiguous areas where mortgages tended to carry higher interest rates were identified [Hillier, Journal of Housing].