Seattle LEAD evaluation: arrest rates, reported odds and federal set-asides
Share of participants arrested before and after entering Seattle's Law Enforcement Assisted Diversion (LEAD) program or a system-as-usual control group, the odds estimates the University of Washington evaluators reported, how those odds translate into the chance of arrest, and the LEAD set-asides Congress enacted from 2018 to 2026. Includes the attributed statements on both sides of the 2026 dispute.
Coverage: October 2009 to July 2014 (evaluation window); federal budget years 2017 to 2026; statements 2015 to 2026
Built to check the widely repeated claim that LEAD participants were '58% less likely' to be arrested, and the Washington Policy Center's 2026 critique of it. Every number comes from the evaluators' own published report (March 2015) and peer-reviewed paper (2017), so the comparison uses the study's own figures rather than either side's summary.
Geography
Seattle, Washington, King County, Washington
Unit of observation
Evaluation group (LEAD or control) and period (before or after entry)
Update cadence
None planned; the evaluation is closed. Statements may be added.
Curator
FactHub Research Team
Curated data tables
These related tables form this versioned dataset release. Each
table has its own downloadable file and field definitions.
Share of participants arrested, before and after entry
Unadjusted percent of each group with at least one arrest (excluding the arrest at entry) in the period before and after entering the evaluation.
Unit of observation: Outcome x window x group x period
Unadjusted for the groups' differences in age, sex and time in the evaluation.
The LEAD 58% after entry is a share of people arrested; it is a coincidence that it equals the 58% in the odds result.
Robustness checks
Sensitivity analysis:Share arrested for non-warrant reasons, whole evaluation.
With arrests to serve warrants removed, LEAD fell from 76% to 52% and controls from 72% to 66%; the LEAD advantage shrinks and the model result was only marginally significant (p = .09).
Share arrested for non-warrant reasons, whole evaluation
Question: How do the arrest shares change when arrests to serve warrants are removed?
Result: Excluding warrant arrests, 76% of LEAD participants were arrested before entry and 52% after; controls went from 72% to 66%.
Operation: Select the whole-evaluation 'Non-warrant arrest' rows of the arrest-rates table.
Scope: All 318 participants (317 in the model), October 2009 to July 2014.
How much less often LEAD participants were arrested after entry
Question: After entry, how much lower was the share of LEAD participants arrested than the share of controls?
Result: After entry, 58% of LEAD participants were arrested versus 80% of controls: 22 points, or 27.5% lower. For non-warrant arrests the gap was 14 points, or 21.2% lower.
Operation: Gap in points = control after minus LEAD after. Percent lower = 1 minus (LEAD after / control after).
Raw after-entry comparison; it ignores that LEAD participants started with a higher arrest share (83% vs 77%), which would make the adjusted gap somewhat larger.
Odds estimates as the evaluators reported them
Question: What did the evaluators' models report for arrests and felony charges?
Result: The headline result is 58% lower odds of any arrest (p = .001). Removing warrant arrests gives 34% lower odds (p = .09); felony charges show 39% lower odds (p = .03).
Operation: Transcribe each reported interaction odds ratio, percent lower odds and p-value.
Scope: Collins et al. (2017) and the March 2015 report.
Odds ratios, not changes in the chance of arrest; see the odds-to-probability analysis.
Six-month results for non-warrant arrests and for charges were not significant.
What the reported lower odds mean for the chance of arrest
Question: If the control group's after-entry arrest share is the baseline, what arrest share do the reported lower odds imply for LEAD participants?
Result: With 80% of controls arrested, 58% lower odds means about 63% of LEAD participants arrested: about 22% less likely, not 58%. For non-warrant arrests, 34% lower odds means about 15% less likely.
Operation: Control odds = p / (1 - p). LEAD odds = control odds x (1 - reported lower odds). LEAD share = odds / (1 + odds). Percent lower = 1 - LEAD share / control share.
Scope: Illustration using the whole-evaluation after-entry control shares.
An illustration, not the evaluators' own conversion: the model ratio is adjusted and applies to change over time, while this applies it to raw after-entry shares.
LEAD set-asides in enacted federal budgets, 2018-2026
Question: How much federal money did enacted appropriations direct to LEAD, compared with WPC's list?
Result: Enacted budgets set aside at least $72.5 million for LEAD sites from 2018 to 2026, inside an existing opioid grant program of $180-200 million a year. WPC's own list sums to about $80 million; its headline says nearly $100 million.
Operation: For each budget year, read the enacted joint explanatory statement and any committee report it makes binding; record the 'no less than' amount named for the Law Enforcement Assisted Diversion (LEAD) model. Sum the verified amounts.
Scope: Department of Justice opioid grant program (called COAP, then COSSAP, then COSSUP), budget years 2017 to 2026.
Set-asides are floors, not amounts actually awarded; award data that names LEAD is too incomplete to total.
Does not count state, local or private LEAD funding.
Statements on record
Exact quotes, unaltered, with what surrounded them, the primary
source, and an archived copy of it.
“Over the entire course of the evaluation to date, people in LEAD were 58% less likely than people in the control group to be arrested”
Susan E. Collins, Seema L. Clifasefi and Heather S. Lonczak, University of Washington LEAD evaluation team
One-page evaluation summary attached to the April 8, 2015 LEAD press release ·
·
Assertion
Context: The summary's second bullet, under 'This evaluation found that LEAD was associated with significantly reduced recidivism.' The full report it summarized said '58% lower odds.' In August 2026 the evaluators said they had used 'likelihood' where they should have used 'odds' and were correcting it.
“People in LEAD were 58% less likely than people in the control group to be arrested.”
Seattle Police Department, City police department and LEAD partner
SPD Blotter news post ·
·
Assertion
Context: The post announcing the evaluation repeats the coalition press release; the line follows a sentence saying the evaluation showed LEAD significantly reduced recidivism. The post was still online unchanged on November 3, 2026.
“LEAD participants 58% less likely to be arrested for any crime in any jurisdiction compared to a similarly situated control group.”
LEAD Seattle-King County, Program website, run by Purpose Dignity Action
Evaluation page on leadkingcounty.org (Internet Archive copy) ·
·
Assertion
Context: Summary line on the page for the 2017 housing and employment paper, as archived April 29, 2026. By November 2026 the page instead said participants 'had 58% lower odds of re-arrest for any crime in any jurisdiction.'
“Independent UW researchers funded by Laura & John Arnold Foundation found LEAD reduced recidivism for all crimes by 58%, reduced felony charges by 39%”
Purpose Dignity Action (LEAD), LEAD's operator, presenting to the Seattle City Council Public Safety Committee
Slide 'Evaluation & research' in the committee's agenda materials ·
·
Assertion
Context: Shown in an update on LEAD use since Seattle's 2023 drug ordinance. The same slide continues that LEAD 'resulted in 41 fewer days in jail per person annually, and lowered the odds of entering prison by 88%.'
“It is a 58% reduced likelihood of being arrested compared to a comparison group.”
Lisa Daugaard, LEAD co-creator, Purpose Dignity Action
Interview with The Center Square ·
·
Assertion
Context: Responding to WPC, she first said 'It's not a 58% reduction in the absolute number of arrests for the lead participants.' She added, 'But WPC is not wrong,' and said the figure 'should not be sort of flattened or further simplified as a 58% reduction in the number of arrests.'
“No clear difference in arrests rates can be reliably claimed, but also cannot be altogether ruled-out.”
Eric Zimmerman, Policy Director, Public Safety Solutions, Washington Policy Center
National Retraction Brief, Part 1, version 1.3, section III.B ·
·
Assertion
Context: Under the heading 'Arrests (Unclear differences),' after noting the 22-point unadjusted difference and arguing that warrant arrests, recruitment, policing and geography differed between groups in ways likely favorable to LEAD.
“For example, compared to the small number estimated as gaining jobs during the study, three times as many LEAD participants died.”
Eric Zimmerman, Policy Director, Public Safety Solutions, Washington Policy Center
National Retraction Brief, Part 1, version 1.3, section I ·
·
Assertion
Context: Offered as an example of claims that 'treat null or negative findings as successes by removing the context.' The brief later says the cohort death rate was 'three times higher than its estimated employment rate.'
“This is equivalent to a 13% success rate in securing permanent housing for the LEAD participants who lacked such housing upon enrollment.”
Eric Zimmerman, Policy Director, Public Safety Solutions, Washington Policy Center
National Retraction Brief, Part 1, version 1.3, section III.E ·
·
Assertion
Context: Derived from 'an estimated average of 19' of 146 unhoused participants appearing housed in a given month of the 18-month follow-up. The brief adds that with shelter included LEAD 'had just a 35% success rate.'
“Let's say 163 or 164 people didn't have jobs when they came into LEAD, and how many additional people in that group gained jobs? [...] The answer is three”
Eric Zimmerman, Washington Policy Center
The John Curley Show, KIRO Newsradio (as reported by MyNorthwest) ·
·
Assertion
Context: Discussing the evaluators' finding of 46% higher odds of being on an 'employment continuum,' which he called 'code for jobs.' He continued that the number could not be connected 'to 46% anything.'
“According to an initial study, it successfully reduces recidivism by as much as 60 percent.”
Suzan DelBene, U.S. Representative (D-Washington)
House floor debate on the Comprehensive Opioid Abuse Reduction Act, Congressional Record p. H2307 ·
·
Assertion
Context: Offering an amendment to let opioid grants fund community-based diversion programs sponsored by law enforcement. The sentence before introduced LEAD as a 2011 Seattle pilot that offers help rather than jail time. The amendment passed by voice vote; the final law does not name LEAD.
“U.S. Congress misled by false LEAD crime-reduction claims, leading to nearly $100 million in federal funding”
Washington Policy Center (David Boze), Free-market think tank
Headline of the Part 2 web summary ·
·
Assertion
Context: Headline of a summary of Part 2 of Eric Zimmerman's brief. The linked PDF lists federal set-asides by year but gives no total; elsewhere it says hundreds of millions of public dollars went to LEAD programs, counting all levels of government.
Retrieved November 3, 2026. Copyright Elsevier; figures quoted for commentary. Values quoted with attribution; the paper itself is not redistributed.
Reliability:
Peer-reviewed study.
Has a stake in this finding. Written by the University of Washington team that evaluated LEAD, funded by the Laura and John Arnold Foundation. The authors declared no conflicts; LEAD program staff sat on the evaluation advisory committee.
Retrieved November 3, 2026. Public report; figures quoted for commentary. Values quoted with attribution.
Reliability:
Academic.
Has a stake in this finding. The same UW evaluators' report to LEAD's policy group, before peer review. Copy hosted by IssueLab (Candid).
Retrieved November 3, 2026. Public press release. Quoted with attribution.
Reliability:
Party to the dispute.
Has a stake in this finding. Issued by the program's backers to announce the evaluation; hosted on the UW evaluators' lab website.
Retrieved November 3, 2026. Public web page (Internet Archive copy). Quoted with attribution.
Reliability:
Party to the dispute.
Has a stake in this finding. Website of the LEAD program run by Purpose Dignity Action (formerly the Public Defender Association).
Retrieved November 3, 2026. Public record. Quoted with attribution.
Reliability:
Party to the dispute.
Has a stake in this finding. Presented by LEAD's operator, Purpose Dignity Action, while the council considered LEAD funding and expansion.
Retrieved November 3, 2026. Public document. Quoted with attribution.
Reliability:
Advocacy group.
Has a stake in this finding. Free-market think tank; the brief recommends pausing LEAD expansion and restoring arrest and jail for drug crimes.
Retrieved November 3, 2026. News article; quoted for commentary. Quoted with attribution.
Reliability:
News report.
Talk-radio station article reporting Zimmerman's on-air remarks.
The Center Square, 'Creator defends drug diversion program amid conflicting success claims' (August 12, 2026), as republished by News-ShieldInterview with LEAD co-creator Lisa Daugaard.
Retrieved November 3, 2026. News article; quoted for commentary. Quoted with attribution.
Reliability:
News report.
Wire story by Carleen Johnson of The Center Square, a nonprofit news service, republished by a member paper.
Retrieved November 3, 2026. Public web page. Quoted with attribution.
Reliability:
Advocacy group.
Has a stake in this finding. Summary by David Boze of Part 2 of the WPC brief; the linked PDF lists set-asides by year but shows no total.
Retrieved November 3, 2026. Public domain (U.S. government work). Public domain.
Reliability:
Government.
Enacted public law, published by the Government Publishing Office.
Limitations and reproducibility
Reproducible from public sources: Arrest percentages are read from Figures 1-4 and Table 4 of the March 2015 UW report and Figure 2 of Collins et al. (2017); label order was checked against the PDF's text stream. The post-entry gap and odds-to-probability conversions are arithmetic on those published values, shown in each analysis's operation.
The evaluation was not randomized. Groups were balanced with propensity score weights, but unmeasured differences between LEAD and control participants may remain.
Arrest percentages are unadjusted raw values read from published figures, rounded to whole percents by the evaluators.
One evaluation of one program in one city (318 people, 2009-2014); results may not carry over to LEAD programs elsewhere or today.
Control-group deaths were not systematically recorded, and the study did not measure crimes that did not lead to arrest.
lead-arrest-rates.csv
text/csv · 1.4 KB (1,455 bytes)