
Undercounting Homelessness: Limits of the Point in Time Count
January 31, 2025On Tuesday, January 28, volunteers and city workers across Maricopa County municipalities fanned out across their respective communities to try and spot people who might be homeless in what is called the annual “Point in Time” Homelessness count. Once found, a person is asked where they slept that night and a series of other questions, including if there were others in their group at night. The surveys are required by the federal Department of Housing and Urban Development (HUD). Typically the numbers become the official count of the number of people experiencing homelessness.
In Maricopa County on January 22, 2024, 9,435 people were counted as being homeless, which included 5,359 in shelters or transitional housing and 4,076 who were on the streets or in vehicles or in structures not designed to be housing, such as abandoned buildings.
That figure is an undercount, probably a significant undercount. The shelter count should be fairly accurate, only limited to the degree some housing options are excluded. Excluded options might be a shelter that does not get federal money and does not participate or someone normally unsheltered who happens to have gotten a cheap motel room that night.
Counting people living on the streets or in vehicles is a significant challenge. Logically, most people would realize that trying to find all the people living on the streets in a morning is virtually impossible, especially with the relatively small numbers of people doing the count. Yet, when these homeless census figures are cited, they are often not contextualized as being an undercount.
To get a sense of the undercount, the research director of the Grand Canyon Institute and a volunteer who works with the homeless went to the Tuesday event hosted by the Aris Foundation at Broadway and McClintock in Tempe. The Aris Foundation gatherings are attended by a wide range of people, from those who are unhoused, to those in shelters as well as housed people with low incomes. Aris provides a range of services, such as clothes, medical checks, and free basic cell phones in addition to a free meal.
The event happened on the same day as the count. While it starts at 5 p.m., people begin lining up for food and clothes early. GCI focused on the food line and interviewed 33 people before the event began. Of the 33, 15 were people who had slept on the streets the previous night in Tempe. Only four of the 15 people, however, had been surveyed by the Point in Time count.
What’s the likelihood that the Point in Time Count actually reached HALF the unsheltered population in Tempe?
GCI conducted a binomial test (counted v. non counted) to test the probability that if the true likelihood of being surveyed was 50%, what was the likelihood if you found 15 unsheltered people that only 4 of them would be surveyed. The result was about 4% of the time. That was assuming people on the street only had a 50% chance of being surveyed. A similar test would be to think of a coin flip and how often heads would come up no more than 4 times out of 15. You would find the probability is 6%.
What about a 67% (two-thirds) chance of being surveyed? The likelihood that only 4 of 15 would be surveyed when two-thirds of the folks on the street are surveyed is 0.2%, essentially impossible.
Even though one should be careful about generalizing about a sample of 15 people, this sample suggests quite clearly that fewer than half of Tempe’s unsheltered population were included in its Point in Time survey.
The Point in Time count only needs to reach one person in a group that sleeps together to count everyone, since how many people in your “household” is one of the questions. While some unsheltered people sleep with others, relatively few do. One of the volunteers in Tempe’s Point in Time Count indicated in their assigned area only one person surveyed was with at least one other person that night. Six of the 15 surveyed at Aris indicated being part of a group of at least two, and they represented two of the four who were counted.
Making that adjustment did little to improve the likelihood of being counted. The 15 people surveyed at Aris represented 24 unsheltered people, including others in their group, of whom only 6 were counted. However, GCI estimates that up to 8 of them might have been counted based on people who were not surveyed but had group members not present at Aris. A binomial test found that only modestly improved the likelihood that half of Tempe’s unsheltered population was counted to 7.5 percent.
Tempe is illustrative of how unsheltered people are systematically undercounted by the Point in Time survey.
The Point in Time survey does tell us a great deal about the people who were surveyed as many important questions are asked, we just should not expect that the “census” is accurate, especially as it relates to people who are “unsheltered.”
For more information, contact: Dave Wells, Research Director, Grand Canyon Institute, at DWells@azgci.org or at (602) 595-1025, Ext. 2.
The Grand Canyon Institute (GCI) is a nonpartisan, nonprofit organization dedicated to informing and improving public policy in Arizona through evidence-based, independent, objective, nonpartisan research. GCI makes a good faith effort to ensure that findings are reliable, accurate, and based on reputable sources. While publications reflect the view of the Institute, they may not reflect the view of individual members of the Board.