Methodology

The Howard Center for Investigative Journalism sought to identify which areas of the United States might have the most difficulty protecting their homeless populations during the pandemic and analyzed four sets of data: homelessness rates per capita, poverty rates, ratio of primary care physicians to residents and number of additional shelter beds needed for social distancing.

The analysis yielded 43 counties of highest concern, representing 14 states from Hawaii to New York, and including 10 rural counties, 13 suburban counties, six urban counties, and 18 of the 50 largest cities in the country.

The four datasets were selected after consultation with sociologists, demographers and epidemiologists at the U.S. Centers for Disease Control and Prevention, the University of Pennsylvania, Boston University, the Models of Infection Disease Agent Study Coordination Center at the University of Pittsburgh, Carnegie Mellon University, and Northern Arizona University’s Center for Health Equity Research.

Each variable was classified to establish thresholds and identify counties that fell into the risk zones for all four variables. The homeless rate, poverty rate and additional shelter beds were all classified with the Jenks natural breaks algorithm, which created five classes based on natural groupings inherent in the data. The break between the first and second classes became the threshold for each variable. The doctors per 100,000 residents rate is measured inversely from the other three variables (i.e., higher is better), so quantile classification was used for that variable.

This classification enabled a vulnerability index for every county in the United States, ranging from 0 (not vulnerable for any of the four variables); 1 (vulnerable in any one variable); 2 (vulnerable for any two); 3 (vulnerable for any three) and 4 (vulnerable in all four). The counties fell largely along a normal distribution, with just 43 possessing a vulnerability index of 4.

The Howard Center then filed public records requests to governmental entities in those 43 counties and their major cities, including public health and human services departments and governing councils, to see how they prepared for and reacted to the threat of COVID-19. Public records requests also were made to those counties’ Continuums of Care, which are regional or local planning bodies that coordinate homeless assistance. The requests sought documents containing keywords, such as “homeless” and “COVID-19,” created from Jan. 1 to June 1, 2020.

Some governmental entities insisted that requests be narrowed to fewer keywords or a tighter time frame before searching for responsive records. Many records were provided free of charge, though some counties, such as Brevard in Florida, requested more than $5,500 for their records – a sum the Howard Center could not afford.

Of the 140 requests filed, 83 were partially or completely fulfilled, providing an inside look at how more than two dozen counties addressed the needs of their homeless populations in the pandemic.

To keep tabs on which of the 43 counties were experiencing spikes in COVID-19 cases and focus reporting efforts, the Howard Center turned to datasets updated daily by the CDC, Johns Hopkins University and Carnegie Mellon University. The CDC assembles cumulative case and death data from state and local jurisdictions with the assistance of its partner USAFacts, which the Howard Center used to produce a data visualization of how county infection rates changed over the summer.

Reporters also turned to datasets from academic researchers to deepen their understanding of COVID’s reach across the 43 counties. The CDC does not calculate a daily rate of new cases in each county, which could better indicate a spike or drop than cumulative numbers. However, researchers at Johns Hopkins produce this from CDC data by subtracting the previous day’s count from the latest count.

Carnegie Mellon researchers, noting inconsistencies in counting cases and deaths in various counties and states, devised their own “COVIDcast” mapping tool to create a tracking system they believe is more illustrative of where COVID-19 has the greatest impact nationwide. The researchers factor in the percentage of COVID-related outpatient visits with Facebook surveys and Google search trends to create an index of how affected a particular county is by COVID-19 each day. Higher numbers indicate more-affected counties.

The Howard Center also attempted to track deaths of homeless Americans across the country, identifying the Continuums of Care whose combined communities accounted for roughly half of the nation’s homeless population.

Reporters surveyed public dashboards and contacted county and city public health departments in those 26 Continuums of Care to see which were tracking COVID-19 infections or deaths among their homeless. Only 12 health departments provided death counts and 11 provided infection numbers, although most were noted as likely undercounts. Some death and infection data was collected from the Howard Center’s public records requests.

In an effort to track funding in the federal CARES Act targeted at helping community homeless populations, known as Emergency Solutions Grant Funding, the Howard Center analyzed and compared a government list of those who were to receive the funds with actual government spending as detailed on the federal website USAspending.gov.

Reporters obtained a list of jurisdictions that had completed grant agreements with the U.S. Department of Housing and Urban Development, a prerequisite to receiving the money, and then compared those against jurisdictions that had been given access to those funds.

Knowing that data can tell only part of the story, the Howard Center also reached out to homeless people across the country, distributing a dozen cheap smartphones with 30 days of credit to homeless service providers in California, New York, Florida, North Carolina and Arizona, who in turn shared the phones with homeless clients. The Howard Center also distributed a Google Voice phone number, which enabled other homeless people to leave recorded messages and connect with reporters directly.