Rent Debt Methodology

This document describes our current methodology for estimating the number of renter households behind on rent and the total and per household rent debt for the United States as a whole, all states and counties, 562 cities and census-designated places, and 15 metropolitan regions, as presented in the Rent Debt Dashboard. It also describes methods used for incorporating Emergency Rental Assistance data from the US Treasury into the “Relief Map” on the dashboard. The dashboard was first released on April 21, 2021 and was last updated on November 2, 2021. The next update will be directly after the next data release from the Census Household Pulse Phase 3.3 survey, expected in late December. Our methodology used prior to the August 30 update can be found here.

Rent Debt Estimation Summary

Our rent debt estimates are based on two sources of data:

  1. Household rent and income data from the 5-year 2019 American Community Survey (ACS) summary file and microdata.

  2. Data on households behind on rent and number of months behind on rent from the U.S. Census Bureau’s Household Pulse Survey microdata for the United States, all 50 states, and the 15 largest metros. The Pulse survey is generally updated every two weeks. We rely upon the Pulse microdata, which is released two weeks after the tabular data is released. The most recent Pulse data is for September 29 - October 11, 2021.

We use the share of households behind on rent from the Census Household Pulse survey and the median contract rent paid by households from the American Community Survey, both broken down by income bracket, to determine the total amount of monthly rent owed by households behind on rent. We then multiply these monthly figures by the average number of months that households are in arrears based on the Pulse survey, to estimate total rent debt. This average is calculated via the weighted average of reported months behind for each behind household.

We believe these estimates are on the conservative side due to: 1) our use of rent data from the 2019 American Community Survey, since rents have risen, especially in high-cost cities and metros and 2) the likelihood that the Pulse survey is undercounting renters who are behind on rent and especially those who are many months behind on rent, given this is a vulnerable population.

Additionally, the fact that the Pulse survey provides a cross-sectional/point-in-time view, rather than a longitudinal one tracking the same renters over time, makes it impossible for us to understand how renters are moving in and out of debt.

Rent Debt Methodology Updates

This methodology was significantly revised beginning with the August 30 release in three ways:

  1. We now use Pulse data on the distribution of rent arrears to inform our rent debt estimates. A question on how many months the household is behind on rent was added in the Phase 3.2 survey which began July 21, 2021. Prior to this question being added, we used national estimates of the distribution of rent arrears derived from the University of Southern California’s Center for Economic and Social Research’s “Understanding Coronavirus in America” panel survey. The Pulse survey data is preferable for many reasons: it provides a much larger sample size, it allows for cross-tabulations with the other Pulse survey questions, and it will be regularly updated at the same interval as our other data inputs. This shift to using the Pulse data on the distribution of arrears has had a significant impact on our rent debt estimates due to the higher share of households that are less than 3 months behind in the Pulse survey compared with the USC survey. There was a 21 percent decrease in our total rent debt estimates across all geographies in the August 30 update, from $21.3 billion to $16.8 billion nationwide.

  2. We have added estimates for the 562 cities and census-designated places that have at least 10,000 renter households.

  3. We now use exponentially smoothed estimates across all available waves of survey data to report the number and share of households behind on rent, their characteristics, and the number of months they are behind on rent. Exponential smoothing assigns more weight to recent observations but also incorporates information from earlier survey waves, in order to increase the sample size and the robustness of the estimates, and to reduce the volatility we were seeing in the data which made it difficult to assess trends even in large states such as California. Previously, we used a two-wave moving average.

Rent Debt Estimation Full Methodology

The process for developing our estimates is as follows.

Calculate the number and share of households behind on rent by income bracket for the Pulse survey geographies (US, states, and 15 metros). We filter the Household Pulse Survey data to include only renting households paying a non-zero rent. We then assign these households a rent arrears status of “Behind” or “Not Behind” based on their response to the survey question: “Is this household currently caught up on rent payments?” We calculate the percentage of households in rent arrears as a share of renter households providing a “Yes” or “No” response to this question – the “behind rate” – by household income category and by geography for each survey wave. Households are grouped into three income categories: those with an annual income less than $50,000, those with an annual income between $50,000 and $100,000, and those with an annual income greater than $100,000. Households not reporting income are excluded from these calculations. An exponential smoothing model is then applied to these rates, using a value of 0.2 for the exponential smoothing parameter alpha. This weights the behind rate from the most recent survey wave most highly, but also incorporates information from each prior survey waves back to the wave ending August 31, 2021.

Estimate behind rates for counties. Behind rates are calculated separately for each of the 15 metropolitan areas and for the nonmetropolitan parts of each state; these rates are then assigned to counties based on whether they fall within one of the 15 metropolitan areas or elsewhere within a state. Within each geography, we vary the calculation of behind rates by income group depending on the number of unweighted observations:

  • If the number of unweighted counts of observations with a behind status and reported income are at least 50 across the past 8 survey waves for the three income groups, behind rates are calculated separately for each income group.

  • If unweighted counts of observations fall below 50 across the past 8 survey waves for either of the top two income categories (between $50,000 and $100,000 or over $100,000), but the two categories combined have more than 50 observations, a single rate is used for both of the categories.

  • If unweighted counts of observations fall below 50 across the past 8 survey waves for the top two income categories combined or for the lowest income category (less than $50,000), a single behind rate is used for all households in the state.

Determine median contract rent by household income bracket for states, the 15 metropolitan regions, large cities, and counties. Median monthly contract rents by income bracket for states, the 15 metropolitan regions, and 430 large counties are drawn directly from ACS microdata. We use median rent (rather than mean rent) based on the assumption that renters who are behind on rent are likely to have lower monthly rent than the average for each income bracket.

Estimate behind rates and median contract rent for cities. To estimate behind rates for the 562 cities and census-designated places recently added to the dashboard as well as the 2,790 counties not identified in the ACS microdata, we rely primarily on the ACS summary file with some inputs from the microdata. Specifically, we draw information from Table B25122 of the ACS summary file on the number of households by income bracket and gross rent bracket and use a Pareto interpolation procedure to estimate median monthly gross rent for each of the aforementioned income brackets in each geography. This procedure requires an upper bound for the top gross rent category ($2,000 or more), which is not provided in Table B25122. To adjust our estimate to reflect median contract rent (rather than median gross rent, which includes the cost of utilities), we also need an adjustment ratio to apply to our resulting Pareto estimates.

We estimate these data inputs for smaller city and county geographies using ACS microdata for the Public Use Microdata Area (PUMA) or PUMAs they intersect. This is accomplished using population-based crosswalks we developed between 2010 PUMAs and 2010 counties, and between 2010 PUMAs and 2010 census-defined places (which include all cities), by taking a population-weighted average of the PUMA-level measures for each smaller city and county geography. Following this approach, we estimate the maximum gross rent, median gross rent, and median contract rent for overall and for each income bracket. The estimated maximum gross rent is inputted into the Pareto interpolation procedure to estimate median gross rent by income bracket for each of the smaller city and county geographies. These initial estimates are then adjusted to reflect median contract rent by multiplying by the ratio of median contract to gross rent from the PUMA-based estimates. The approach seeks to utilize as much geographically-specific information from the ACS summary file as possible and substitutes in less geographically-specific information from the ACS microdata as necessary.

Estimate rent debt. To estimate rent debt at the county level, we apply the percentages of households behind on rent at the state or regional level (when available) to the county-level ACS data on median monthly contract rent by income bracket. We assume that differences between reported rents from the 2019 5-year ACS (which reflect a 2015-2019 average expressed in inflation-adjusted 2019 dollar values) and 2020 actual rents are negligible for households that have not moved in 2020, as those households were likely locked into pre-pandemic leases and/or month-by-month agreements with fixed/stable rents. The total amount of monthly rent owed by behind households is then calculated by multiplying estimated median monthly rent for each income category by the number of Pulse households in that income category and summing those values for each geography (city or county). Regional and statewide estimates are produced by summing estimates from their constituent county geographies.

These figures are converted to total rent debt by multiplying the estimated number of households behind on rent by a weighted average of the number of months behind on rent reported by respondents in the Pulse survey. For households that reported being behind but said they were 0 months behind, we assume that they are behind in the most current month and that they are also 1 month behind. Based on the two most recent waves of the Pulse survey, we estimate that households are 3 months behind on average, with 43.6% 1 month behind, 22.2% 2 months behind, 13.2% 3 months behind, 13.6% between 4 and 7 months behind, and 7.5% more than 7 months behind on rent. Because the Pulse dataset topcodes this value at 8 months but 17 months have passed between April 2020 (the first full month after initial stay-at-home orders) and August 2021 (the latest full month included in the survey), we evenly divide the 7.5% share of households more than 7 months behind on rent for each number of months between 8 and 17. To estimate rent debt for sub-national geographies, we apply this distribution of arrears to the number of behind households for each geography.

Emergency Rental Assistance Performance Data

The Relief Map on the dashboard was added with our August 30, 2021 update and presents the amount of Emergency Rental Assistance allocated to each state, county, and city on the dashboard along with the amount and share that has been distributed. This data is pulled directly from the monthly performance reports provided by the Treasury, which are generally released three weeks into the following month. State-level and city-level distribution and allocation figures are drawn directly from the Treasury reports, while county-level figures are calculated by summing the distribution and allocation amounts to both county governments and local governments within those counties. For the first phase of the Emergency Rental Assistance Program (ERA1), 23 counties and 9 local jurisdictions opted to reallocate their assistance funds to a higher level of government; funds allocated to those jurisdictions are not reported separately. For example, because Los Angeles County elected to reallocate its funds to the state of California, we do not report allocation and distribution data for Los Angeles County as a whole even though jurisdictions within Los Angeles County – including Los Angeles and Long Beach – have received separate allocations.