Measuring Gentrification

1. INTRODUCTION

Background

A conceptual definition of gentrification by urban planner Benjamin Grant defines it as “a general term for the arrival of wealthier people in an existing urban district, a related increase in rents and property values and changes in the district’s character and culture” (Klibanoff, 2015). Any definition of gentrification would require acknowledgment of its context and special effects of it as a problem. Some of the noted significance of gentrification is that it pushes out longtime residents and businesses in communities through the form of forced displacement. Gentrification also reduces the amount of affordable housing and also results in the displacement of racial minorities and low-income residents (Richardson et al., 2019). All of which may foster discriminatory behavior by people in power (Chong, 2017).

We will be attempting to determine which tracts are at risk of displacement using gentrification predictor variables such as average household income, percentage of white-only residents, education demographics such as college attainment, and the share of renters against the city-wide average.

Research Question

Can we predict commonly used risk scores for gentrification based on Census data? And what features correlate with gentrification?

2. METHOD

Data Collection Procedures

In the article Mapping gentrification and displacement pressure: An exploration of four distinct methodologies, we followed Portland’s method: Census tracts were “determined to be vulnerable to housing displacement if three of four indicators – accounting for race, higher education, rent, and income, as indicated in Table 1 – are above the city-wide average” (Preis et al., 2020, p. 9). 

Portland’s method was the most appealing as it was categorical and focused on assessing the risk of gentrification for each tract for our year of study, 2019. Seattle and Los Angeles used a continuous risk score for each of their weighted variables. Philadelphia was categorical but based on the percent change from decennial census data. Ultimately, the results of this article were not as relevant for our study as it was focused on comparing the four cities' methodologies when applied to Boston.

While a major interest of this research was how different risk scores could be predicted and observed variable importance, we also chose to compare how these same risk predictors could be applied to various counties. By comparing the same risk predictor to another city with the same threshold, we can see the variation in how gentrified one county is versus the other. A limitation of this approach is that different counties will have varying historical and contextual factors. One county may have already faced rapid demographic change and increasing housing costs which would negatively impact how another county is predicted.

For the mapping, shapefiles were gathered at our level of interest from the National Historical Geographical Information System (NHGIS). Most examples are mapped at the County level for each State. We required down to the Census Blocks to map each City as it relates to the County. This allowed us to gather and compare the percentage of Races in the County. While our model had data at the census tract level for all of the variables mentioned, we were unable to gather shapefiles at the block level for Income and Educational Attainment. 

3. FINDINGS AND ANALYSIS

Findings 

Following the City of Portland’s gentrification and displacement risk assessment, we applied it to San Francisco’s 2019 census tracts as a training dataset. Once trained, it was applied to Multnomah county’s 2019 census tracts to test. The model correctly predicted that 4 of the tracts were at risk of gentrification and 148 of the tracts were not at risk. This is not far from our expectations as a majority of Portland is already gentrified and by San Francisco’s standards, they are hardly at any risk. The below graphic of Multnomah County shows that a high percentage of the population identifies as white and this backs why so few of the tracts are predicted as at risk.

An opportunity to run a new test on Pierce County, in Washington State, gave us the ability to compare our understanding of our predictor model against anecdotal accounts of gentrification that we know of in that region. ​​The model correctly predicted that 26 of the tracts were at risk of gentrification and 109 of the tracts were not at risk. With a lot more urban development in Pierce County, it is no surprise that the number of at risk of gentrification tracts exist compared to Multnomah and San Francisco county. The below graphic of Tacoma, WA, where a majority of the county’s population resides, shows that there is a much higher percentage of non-white.

Feature Importance

Unsurprisingly, the biggest indicators for gentrification with our model are the ones that are already established as traits of gentrification. The lowest indicators for gentrification according to our model are: % naturalized foreign-born from Oceania, % naturalized foreign born from Africa,  % of the population is of Hawaiian descent. These results are also not surprising since these ethnic groups represent an extremely small fraction of the population.

Something worth noting however is that the percentage of structures that were built after 2014 is a low indicator of gentrification. We believe that this could be due to higher construction costs in gentrified areas and historical building preservations that lead to developers remodeling and repurposing existing structures. A more plausible explanation is that our model is trying to predict gentrification, whereas the year structure was built indicates that gentrification has already happened a long time ago. 

Results

         Our models predicted commonly used risk scores for gentrification based on Census data with 83% accuracy. This was highly influenced by the number of true negatives that are a result of working with cities and counties that have already faced gentrification. There are difficulties with replicating our model in other cities, mainly stemming from contextual differences in gentrification among different cities. Out of the exploratory variables, it is not surprising that the percentage of newly built structures is a low indicator of being at risk for gentrification because, where new buildings are being erected, they have already experienced gentrification. While we were able to verify that the established variables for determining the risk of gentrification had the highest correlation with race, income, and educational attainment; renter share was surprisingly not as high. This may also be a result of the newly built structures that have taken over urban areas contributing to the high number of renters.

References

Chong, E. (2017, September 17). Examining the negative impacts of gentrification. Georgetown Law. Retrieved June 6, 2022, from https://www.law.georgetown.edu/poverty-journal/blog/examining-the-negative-impacts-of-gentrification/ 

Klibanoff, E. (2015, December 9). Explainer: What is gentrification? WHYY. Retrieved June 6, 

2022, from https://whyy.org/articles/explainer-what-is-gentrification/ 

Preis, B., Janakiraman, A., Bob, A., & Steil, J. (2020). Mapping gentrification and displacement pressure: An exploration of four distinct methodologies. Urban Studies, 58(2), 405–424. https://doi.org/10.1177/0042098020903011 

Richardson, J., Mitchell, B., & Franco, J. (2019, March 19). Shifting neighborhoods: Gentrification and cultural displacement in American cities. NCRC. Retrieved June 6, 2022, from https://ncrc.org/gentrification/ 

Stafford, A. (2020, September 19). Mapping Census Data. Medium. Retrieved June 6, 2022, from https://towardsdatascience.com/mapping-census-data-fbab6722def0