California Estimated Displacement Risk Model

June 20, 2022    |    Authors: Tim Thomas, Karen Chapple, Julia Greenberg, Alex Ramiller, Emery Reifsnyder, Isaac Schmidt, Kate Ham

We would appreciate your feedback on this map. If you see a tract or area that does not seem right, please fill out this form to help us ground-truth the method and improve our model.

What does this map show?

UDP’s Estimated Displacement Risk (EDR) model for California identifies varying levels of displacement risk for low-income renter households in all census tracts in the state. Displacement risk means that in 2019—the most recent year with reliable census data—a census tract had characteristics which, according to our model, are strongly correlated with more low-income renter population loss than gain. In other words, the model estimates that more low-income households left these neighborhoods than moved in. The model uses 2015 – 2019 data, which means that correlations between tract characteristics and low-income renter population loss are based on this time period.

This map is a conservative estimate of low-income loss and should be considered a tool to help identify housing vulnerability. Displacement may occur because of either investment or disinvestment. Because this risk assessment does not identify the causes of displacement, we do not recommend that the tool be used to assess vulnerability to investment such as new housing construction or infrastructure improvements. To better understand the relationship between these developments and displacement, check out our studies on new housing construction and green infrastructure. We recommend combining this map with on-the-ground accounts of displacement to achieve a full understanding of the issue.

This model will be developed for the rest of the U.S. by 2023 and will continue to be improved as we gather new data sources.

How should I read this map?

The EDR provides three layers of displacement information. The “Overall Displacement” map layer shows the number of income groups experiencing any displacement risk. For example, in the dark red tracts (“2 income groups”), our models estimate displacement (Elevated, High, or Extreme) for both of the two income groups. In the light orange tracts categorized as “Probable”, one or all three income groups had to have been categorized as “Probable Displacement”.

The “50-80% AMI” layer on the map shows the level of displacement risk for low-income (LI) households specifically. Since we have reason to believe that our data may not accurately capture extremely low-income (ELI) households due to the difficulty in counting this population, we combined ELI and very low-income (VLI) household predictions into one group—the “0-50% AMI” layer on the map—by opting for the more “extreme” displacement scenario (e.g., if a tract was categorized as “Elevated” for VLI households but “Extreme” for ELI households, we assigned the tract to the “Extreme” category for the 0-50% layer). For these two layers, tracts are assigned to one of the following categories, with darker red colors representing higher displacement risk and lighter orange colors representing less risk:

  • Low Data Quality: the tract has less than 500 total households or the census margins of error were greater than 15% of the estimate (shaded gray).
  • Probable Displacement: the model estimates there is potential displacement of the given population in these tracts.
  • Elevated Displacement: the model estimates there is a moderate amount of displacement (e.g., 10%) of the given population.
  • High Displacement: the model estimates there is a relatively high amount of displacement (e.g., 20%) of the given population.
  • Extreme Displacement: the model estimates there is an extreme level of displacement (e.g., greater than 20%) of the given population.

Transparent tracts are not experiencing our definition of displacement according to the model. Some of these transparent tracts may be majority low-income experiencing small to significant growth in this population while in other cases they may be high-income and exclusive (and therefore have few low-income residents to begin with).

To help illustrate where low-income households are concentrated, we provide a layer showing the percent of households that are below 80% AMI. We also provide a segregation layer that shows which racial and ethnic groups make up more than 10% of a tract’s population. Finally, we provide an extra displacement overlay that can be turned on at the same time as the other map layers to show where displacement is occurring in relation to where these economic and demographic groups live.

How did we create this map?

The EDR is a first-of-its-kind model that uses machine learning and household level data to predict displacement. To create the EDR, we first joined household-level data from Data Axle (formerly Infogroup) with tract-level data from the 2014 & 2019 American Community Survey; Affirmatively Furthering Fair Housing (AFFH) data from various sources compiled by California Housing and Community Development; Longitudinal Employer-Household Dynamics (LEHD) Origin-Destination Employment Statistics (LODES) data; and the Environmental Protection Agency’s Smart Location Database.

We then used a machine learning model to determine which variables are most strongly related to displacement at the household level and to predict tract-level displacement risk statewide while controlling for region. We modeled displacement risk as the net migration rate of three separate renter households income categories: extremely low-income (ELI), very low-income (VLI), and low-income (LI). These households have incomes between 0-30% of the Area Median Income (AMI), 30-50% AMI, and 50-80% AMI, respectively. Tracts that have a predicted net loss within these groups are considered to experience displacement in three degrees: elevated, high, and extreme. We also include a “probable displacement” risk category in tracts that might be experiencing displacement.

What are the main limitations of this map?

  1. Because the map uses 2019 data, it does not reflect more recent trends. The pandemic, which started in 2020, has exacerbated income inequality and increased housing costs, meaning that our map likely underestimates current displacement risk throughout the state.
  2. The model examines displacement risk for renters only, and does not account for the fact that many homeowners are also facing housing and gentrification pressures. As a result, the map generally only highlights areas with relatively high renter populations, and neighborhoods with higher homeownership rates that are known to be experiencing gentrification and displacement are not as prominent as one might expect.
  3. The model does not incorporate data on new housing construction or infrastructure projects. The map therefore does not capture the potential impacts of these developments on displacement risk; it only accounts for other characteristics such as demographics and some features of the built environment. Two of our other studies—on new housing construction and green infrastructure—explore the relationships between these factors and displacement.