Measuring Gentrification with Mortgage Application Data
Richard W. Martin (University of Georgia) & James Stevens (Wofford College)
Ruth Glass coined the term gentrification in 1964, and it has been widely used since to describe significant neighborhood change. Although many opinions exist around the precise definition, the origin of the term refers to higher-income households moving into lower-income neighborhoods. When the phenomenon is occurring, or has occurred, it is usually obvious to residents of a given neighborhood and city. However obvious this might be at the ground-level, when researchers study the topic across geographics and time, the identification task becomes more difficult. It is for this reason that we introduce several new measurement tools that can help reveal the presence and timing of the forces involved in the gentrification process.
Researchers who study gentrification in the United States face several serious constraints regarding data, especially when it comes to trying to pinpoint the onset of gentrification. Census tracts are typically used as a proxy for neighborhoods and comprehensive data on tracts from the U.S. Census Bureau comes from two sources. The first is the decennial Census. While the decennial Census provides a wealth of data on every census tract in the country, the downside of relying on it to study gentrification is that, by definition, the data is only provided every ten years and only in years ending with zero. This leaves researchers no choice but to identify “gentrifiable” tracts (those that are candidates to gentrify) based on their characteristics at the beginning of a decade and to identify the tracts that gentrify based on changes that occur over the next ten or twenty years. This means that gentrification is limited to being a process that begins in years ending in zero and occurs over a period that is at least ten years long. Additionally, the tract-level results from the decennial census are typically not available until up to three years after the census year. Because of this, the decennial census does not provide a timely picture of what is occurring in a tract and is more useful for looking back and describing what happened. This, then, makes it difficult to identify tracks that are experiencing changes consistent with gentrification until well after the changes have begun.
The second source of comprehensive data on census tracts is the American Community Survey (ACS). The ACS is an annual nationwide survey that is intended to provide census-style data in the years between the release of decennial census data. The ACS provides timely and thorough data for the nation, states, and counties each year with each sample year’s data typically available only a few months into the following year. However, the ACS is less useful for census tracts. Because of the small sample sizes at the tract level, useful annual estimates are not possible on an annual basis. Instead, census tract data for the ACS are released in rolling five-year samples in which the samples from the previous five years are combined. This yields a large enough sample to provide accurate data at the census tract level. However, the combined sample masks annual changes and does not allow one to observe a reliable picture as to what a tract is experiencing during a particular year. Thus, like the decennial census, ACS data is not particularly useful in identifying tracts that may be starting to gentrify.
Our study presents a new approach to measuring U.S. gentrification at the census tract level. We utilize a popular public data set – the Home Mortgage Disclosure Act (HMDA) Loan Application Register – and leverage it in an unconventional way. The premise of our study is that there is valuable information about household income levels contained within this mortgage banking dataset. More specifically, the mortgage application data can provide a window into the income levels of households who are seeking to move into a neighborhood. We develop several new measures that benchmark the income of mortgage applicants against existing homeowners in a neighborhood. In our study we show statistically significant relationships between these measures and more traditional gentrification over the 2010 to 2017 period. Furthermore, instead of using a binary measure of gentrification, our tool allows one to gauge the breadth and intensity of the gentrification forces occurring from homebuyer demand in a given tract.
Most importantly, the free and publicly available dataset provides researchers new flexibility in considering the beginning and ending of the gentrification process, of which much is still unknown. The new methodology allows researchers to observe tracts experiencing gentrification pressures on an annual basis with only a short lag between the collection year and the release of the data. We are hopeful that these measurement tools might be useful in advancing our knowledge of the gentrification phenomenon in the U.S. To that end, we have provided an annual tract-level dataset of our gentrification pressure measures to Sage so the data can be available and used by readers of Urban Affairs Review.
Richard W. Martin is an associate professor of real estate in the Terry College of Business, at the University of Georgia.
James A. Stevens is an assistant professor of finance at Wofford College, in Spartanburg, South Carolina.