The Survey That Decides Who Gets Housed Was Never Validated. Phoenix Retired It Last Year.

By Common Ladder · August 13, 2026 · 8 min read

If you have ever asked for housing help in an American city, someone sat you down with a long list of questions. Whether you use drugs. Whether you have been beaten up recently. Whether you talked to the police after witnessing a crime. How many times you rode in an ambulance in the last six months. Whether you have "planned activities, other than just surviving, that make you feel happy and fulfilled."

You were probably not told the answers were being added up — in Los Angeles, the packet case managers worked from carried an instruction in capital letters: "DO NOT VOLUNTEER THE SCORE OR THE SCORING PROCESS."⁴ The number that came out the other end shaped whether you got an apartment, a time-limited subsidy, or a place on a list. It is worth knowing how well that number works.

What the score actually decides

HUD's 2017 coordinated entry notice requires every Continuum of Care to run a standardized assessment and prioritize people consistently across the region. It tells communities to "develop or select standardized tools" gathering only what is needed to establish severity of need and eligibility. It does not name a tool.¹ Into that blank space, in 2013, stepped the Vulnerability Index – Service Prioritization Decision Assistance Tool: fast, free, and available when hundreds of communities suddenly needed something.⁴

What it produced was one number on a 0–17 scale. Above a threshold, usually 8, you were a candidate for permanent supportive housing. Between 4 and 7, rapid rehousing. Below that, neither. Matchers worked the list from the top down — as one Los Angeles matcher put it, "We filter our 17s, our 16s, and 15s — that's where we go first."⁴ The question is whether that number measures what it claims to.

What the validity evidence says

The first independent examination came in 2018, using HMIS records for 1,495 people assessed between 2014 and 2016. Several items turned out to be weakly related, or related in the wrong direction, to the vulnerability they were meant to capture; the socialization and daily-functions domain and the health items performed particularly poorly. The headline result: "VI-SPDAT total scores did not significantly predict risk of return to homeless services, while type of housing was a significant predictor."³

Read that again, because the order matters. What happened to you afterward — whether you got permanent supportive housing or a rent subsidy that ran out — predicted whether you came back. Your score did not.

Dennis Culhane of the University of Pennsylvania, whose group tested whether the tool could identify who would die, be incarcerated or be hospitalized without help, put it bluntly: "The VI-SPDAT was like flipping a coin in terms of the predictability."⁴ A 2025 systematic review landed in the same place: very limited evidence supports the psychometric properties of the instruments communities are actually using.⁵

The score predicts you'll come back — not that you'll succeed

The largest study is published by HUD's own research office. Petry and colleagues followed 11,036 unsheltered single adults aged 25 and over, assessed between 2015 and 2018 across 16 Continuums of Care in four regions of the country, and asked one narrow question: after exiting into housing, did they return to the homeless system needing housing within 365 days?²

Just under 30% did. And each additional VI-SPDAT point was associated with a 19% increase in the odds of returning (AOR 1.19, 95% CI 1.16–1.22).²

That association is real, but it is not validation. It says the tool picks up something about difficulty. It does not say the tool sorts people into the right interventions — and the paper's own numbers suggest the intervention matters far more than the score.

Look at people who scored exactly 8, right at the threshold. Of those placed in permanent supportive housing, 15.8% returned to homelessness. Of those with the same score placed in rapid rehousing, 48.8% returned. At a score of 9: 22.5% versus 65.3%.² Identical people on paper. Three times the failure rate, depending on which door was open that week.

And the models explained about 4% of the variance (pseudo R² = 0.04).² Ninety-six percent of what determines whether someone stays housed is not in the score, and mostly is not in the data at all. The authors say so directly: "a multitude of factors are not captured by Homeless Management Information System data that are likely critical to preventing returns to homelessness."²

One further finding deserves its own sentence: Native American and Alaska Native people had 70% higher odds of returning to homelessness than White people overall (AOR 1.70, CI 1.16–2.50), and 83% higher after rapid rehousing specifically.² That is a disparity in what happens after placement, and no assessment tool touches it.

The racial gap in scoring

In 2019, C4 Innovations analyzed coordinated entry data across four counties in three states and found people of color received statistically significantly lower scores than White people, with race predicting score.¹⁰ Four years later The Markup obtained more than 130,000 assessments from Los Angeles going back to 2016 and found the same pattern holding year after year. Among young adults in 2021: 67% of White respondents reached the highest priority group, versus 56% of Latino and 46% of Black respondents. Among adults 25 and older: 39%, 35% and 33%.⁴ In the same period Black residents were about 9% of Los Angeles County's population and 30% of its homeless population.⁴

The most plausible explanation is not a coding error. The questionnaire asks people to disclose drug use, mental illness, arrests and police contact to a stranger who controls their access to housing — and people with good historical reasons to distrust that transaction disclose less.⁴ A tool that rewards candor will systematically under-score anyone who has learned that candor is expensive.

By December 2020 the developer, OrgCode, announced it would stop investing in the VI-SPDAT, citing persistent misuse in the field and the need to accelerate work on racial equity; support ended entirely in 2022.⁶ The instrument that ranked hundreds of thousands of Americans for housing has been formally unsupported for four years. Los Angeles has since built a replacement with USC and UCLA researchers — the LA Housing Assessment Tool, 19 questions instead of 35, phased in alongside the old survey through spring 2026.¹⁴

What happened in Phoenix

Maricopa County had its own version of the evidence. A 2023 evaluation of the regional coordinated entry system — five focus groups with 45 people who had actually tried to get housing here, plus a provider survey and an HMIS analysis — found that "Black people have lower median scores than White people across assessment types," and that while referral rates were broadly similar across racial groups, White clients with referrals waited noticeably less time than Black and Native American clients.¹²

The county did not wait to be told twice. A 2021 regional racial equity report named redesigning the assessment as one of four goals; a workgroup built a local instrument, the Maricopa Assessment and Prioritization tool, or MAP, with fewer questions, no predetermined score buckets, and trauma-informed framing. It was piloted from January 2024 alongside the VI-SPDAT for roughly eighteen months, and the workgroup then recommended formal retirement of the VI-SPDAT effective June 30, 2025, with MAP-only assessment from July 1.⁷

Then the Arizona Housing Analytics Collaborative looked at the results, and this is where the story earns its place in an evidence brief rather than a press release.

Under the VI-SPDAT, White individuals and families had been overrepresented in the high-priority group, Black and Latin American individuals and families underrepresented. Under MAP, prioritization for individuals came out racially equitable. For families it stayed slightly inequitable — small differences, but consistent over a year and additive over time, driven largely by a chronic-homelessness item worth almost a third of all available points.⁸

The evaluators also found a plain implementation defect: individuals were "erroneously receiving additional points for items intended for families only," and some items carried the wrong point values. The errors "appeared to benefit White individuals and families disproportionately, and to artificially de-prioritize or push Black individuals and families further down the priority list." The first recommendation was to get the vendor to fix it.⁸

Phoenix's transition is also less complete than the headline suggests. The retirement covers single adults only: as of July 2025 the Family Housing Hub was still using the family version of the VI-SPDAT, and the CoC's publicly posted coordinated entry policies — last board-updated in February 2025 — still describe VI-SPDAT acuity tiers.¹¹

None of that is a reason for cynicism about the change. Maricopa built a tool, measured it, published the flaws and named the fix — which is exactly what the field has been asking communities to do. It is also a reminder that replacing an instrument is the beginning of the work, not the end.

Being precise about what this establishes

This is not an argument against assessment. Rationing happens whether or not anyone writes it down. The alternative was first-come-first-served, provider discretion, and whoever could get to the right office on the right morning. Those systems had disparities too; they just produced no data anyone could audit.

A better tool does not create a housing unit. Eric Rice, the USC researcher who spent years building the Los Angeles replacement, is candid about it: "these vulnerability tools that I'm working on trying to improve are a function of a failed system where we don't have enough resources dedicated to solving the problem."⁴ The score is the interface to scarcity. Fixing the interface does not end the scarcity.

And a fair score is not a fair outcome. Allegheny County, Pennsylvania replaced the VI-SPDAT in 2020 with a predictive instrument. Scores under the new tool were distributed similarly across racial groups and decisions aligned better with those scores — yet researchers found no corresponding decrease in disparities in the rate at which people were actually served.⁹ Everything downstream of the number — who is on the by-name list, which buildings accept which histories, who can be found when a unit opens, who has documents — carries its own inequities, and no assessment tool touches them.

And none of the studies here randomized anything. The Petry paper says so in its own limitations: no control group, no random assignment, and real selection bias, because providers exercise discretion over who gets which resource.² The 15.8%-versus-48.8% gap at a score of 8 is consistent with rapid rehousing being a poor match for high-acuity people. It is also consistent with providers steering whoever they judged most likely to succeed into permanent supportive housing. Both are probably true.

What follows from this

For anyone running or governing a coordinated entry system, the evidence points at a short list with little to do with picking a better questionnaire.

Audit your own scores by race, and publish it. Maricopa's evaluator found a live scoring bug because someone was looking. Most communities have never run the analysis; it takes an afternoon with HMIS data.

Measure the assignment, not just the assessment. The largest finding in this literature is that where someone lands matters more than what they scored. A system that assesses beautifully and then places high-acuity people into time-limited subsidies because that is what is in stock will generate returns to homelessness regardless of the instrument.

Stop asking questions you cannot justify. Any item inviting someone to disclose criminalized behavior to a stranger gets answered differently depending on how much that person has to lose. If an item predicts nothing — and several demonstrably do not — it is costing trust for free.

Tell people they are being scored. The Los Angeles instruction not to volunteer the scoring process was defended as sparing people from feeling reduced to a number. Its practical effect was that people could not tell they had been ranked low and could not ask to be reassessed. Reassessment was available the whole time; Chantel Jones got housed after a second survey with a case manager she trusted more.⁴ Nobody had told her that was an option. That fix is free, and it hands the most power to the person answering the questions — which may be why it keeps not happening.

Phone numbers and access points on this page were verified against the Maricopa Regional CoC's published guidance in August 2026. Programs change — call before you go, or call 211. Have a correction or an updated figure? Email corrections.commonladder@gmail.com.