In a recent City Journal article headlined "No, Seriously, Decriminalizing Drugs Kills People," the Manhattan Institute's Charles Fain Lehman claimed a recent National Bureau of Economic Research (NBER) paper provided conclusive evidence that "should lay to rest the debate over legalization." I published a rebuttal in Reason, arguing that the study did not close the debate. That provoked a response from Lehman on X.
I will respond to each of Lehman's substantive points and delve more deeply into the research. I communicated with the NBER study's co-authors, including University of Oregon economist Benjamin Hansen, who told me that the researchers "do not end any debate" with their report. I was also in touch with the co-author of a study on crime that Lehman cited on X. He offered a much more sober interpretation of his own research than the one Lehman presented in a Wall Street Journal op-ed.
The NBER paper focuses on Oregon and Washington's brief experiments with decriminalizing drugs, which began in February 2021 as a result of a ballot initiative and a court ruling, respectively. This topic is worthy of careful attention. States are policy laboratories. If other locales plan to try decriminalizing drugs, they should study closely what happened in Oregon and Washington.
The issue is complex because the decriminalization periods were marked by disruptions from COVID-19, fentanyl, and a season of social upheaval. Serious researchers using defensible methods are reaching different conclusions. I took issue with Lehman's claim that the issue was settled by the NBER study and his assertion that "it's hard to see decriminalization as anything but an abject failure."
The paper's authors don't see it that way. "Our research, if anything, restarts the debate," Hansen wrote via email, by raising new questions, such as why decriminalization was associated with more deaths and whether a future iteration of the same policy could avoid them.
The precise cause of the spike in fatalities during decriminalization is hard to tease out, according to David Hall, a doctoral student in economics at the University of Oregon and another co-author of the NBER paper. In an email, he said that decriminalization bundled together with many simultaneous changes caused the rise in overdose mortality. Identifying which of the factors drove the results, he said, "will be important and also really difficult."
Hansen emphasized that most of the rise in overdose deaths in both states came from the spread of fentanyl. The paper aims to estimate the extent to which decriminalization made the fentanyl crisis worse. Contrary to Lehman's claim, the broad debate on decriminalization of drugs has not been laid to rest.
Jumping to Conclusions
This isn't the first time advocates have claimed victory for their side. When a 2024 paper found no association between decriminalization and overdose mortality in Oregon, Scientific American quoted Todd Korthuis, the chief of addiction medicine at Oregon Health and Science University, who called it "the study that we were all waiting for." Korthuis said the study showed that fentanyl, not changes to the law, was "the whole story."
Emergency medicine physician Jeremy Faust offered a similar take in an Inside Medicine article. "Study debunks link between Oregon drug decriminalization and overdose deaths," the headline said. "Now we know that neither drug decriminalization nor drug recriminalization was responsible for increases in drug overdose deaths in Oregon and Washington," Faust wrote. "Fentanyl was (and is) the problem."
With one more year of data, the research points the other way. That's how social science works. But by declaring the issue settled for his side, Lehman committed the same error of leaping to an unjustified conclusion.
We do know that decriminalization in Oregon and Washington did not deliver the dramatic short-term benefits that advocates (including me) had hoped for. The NBER paper strengthens the case that it caused harm.
A Valuable Study
Is the NBER paper a good study that's worth taking seriously? I think so. Yet I noted that Lehman had selected "a single non-peer-reviewed working paper with equivocal findings from an even more equivocal literature" to slam the door on the debate. He replied:
A good rule of thumb is that anyone who uses "it's not peer reviewed!" as a rebuttal to a study probably doesn't know how to evaluate evidence. https://t.co/TI7Q97J8tx
— Charles Fain Lehman (@CharlesFLehman) July 15, 2026
Lehman's underlying point is sound. Plenty of good research hasn't been peer-reviewed. Lack of peer review is not a rebuttal, and I didn't offer it as one.
"Working paper," though, is a phrase that covers everything from rough notes posted by a crank on his personal website to nearly publishable research by serious scholars. That's why I asked the authors about the status of their paper. Hall told me it "could still go through a bit of revision, but I think (hope?) less so than other working papers." Hansen added that while referees will demand new figures and robustness checks, "it's pretty rare that findings completely change."
Working papers circulated for comments can be valuable, and this one is. But completed, peer-reviewed work subjected to replication and objections by other researchers carries more weight. Provisional evidence deserves provisional language, even when it's good evidence.
Is It Statistically Significant?
How meaningful is it that Oregon and Washington were among the top 10 percent of states in overdose death increases during the decriminalization period? That finding, I wrote, was "on the border of statistical significance"—meaning suggestive but short of conclusive.
Lehman replied:
Among many other absurd things, this article appears to be saying that Oregon and Washington being in the top 10 percent of states for death increases is borderline statistically significant because a p-value of 0.1 is usually described that way. (H/t @besttrousers) https://t.co/hEbxCji6Gr pic.twitter.com/pbV0gCZzmK
— Charles Fain Lehman (@CharlesFLehman) July 15, 2026
Lehman assumed I was conflating these states' ranking in "the top 10" with a p-value of 0.1. That's not what I was saying, and I should have been clearer.
My point is that two decriminalization states landing in the top 10 percent of states for overdose death increases is, by itself, enough to go looking for a reason, but not so unlikely that we can rule out coincidence. The border of statistical significance is the line dividing results that can be plausibly dismissed from results that demand explanation.
If 10 geographically and culturally diverse states had decriminalized at different times and witnessed the 10 largest increases in overdose deaths immediately afterward, that would have clear statistical significance: The association between decriminalization and an increase in overdoses could not be plausibly ascribed to coincidence. If a single state decriminalized and saw an overdose increase, however large, the effect could have been caused by something else that changed around the same time.
In this case, with two similar states decriminalizing around the same time and landing in the top 10 percent of overdose increases, you have something in between. It takes painstaking analysis to determine whether the association is statistically significant at conventional levels. This painstaking analysis to rank decriminalization states against constructed controls is exactly what papers on both sides of the debate do.
Does Harm Reduction Work?
I argued that what kills drug users is not the opioid molecule but "not knowing whether the powder in front of them is a dose or a coffin." I cited Switzerland, which slashed drug fatalities by dispensing pharmaceutical-grade heroin, meaning users weren't poisoned by fentanyl. Here's Lehman's response:
also I just want to note that this clearly-AI-written paragraph is nonsense. Among other things, the idea that people only OD because they don't know what's in their drugs is belied by the surveys showing that when people get positive fent tests, they still use the drugs! pic.twitter.com/UVqSBrIM22
— Charles Fain Lehman (@CharlesFLehman) July 15, 2026
In the highlighted paragraph (which wasn't written by AI), I did not say that people overdose "only" because they don't know what's in their drugs. Overdoses increase when fentanyl in uncontrolled doses replaces more carefully dosed heroin. I stand by that point.
Lehman did not include a citation for his claim on X that "when people get positive fentanyl tests, they still use the drugs!" But he was likely referring to a 2018 study in the International Journal of Drug Policy, "Use of Rapid Fentanyl Test Strips Among Young Adults Who Use Drugs." Lehman linked to that study in a 2023 City Journal article to support his claim that "drug users, when informed that their drugs are adulterated with fentanyl, will often continue to use."
The authors of that study drew a different conclusion. They said their findings showed that "receipt of a positive test result was associated with positive risk reduction behavior change." Of the 31 users who got a positive result after testing their drugs or urine, they reported, "45% used smaller amounts, 42% went slower when using, 39% used with someone else, and 36% did a tester." They added that "few participants threw their drugs out (10%), sold them (10%), or gave them away (7%)."
It is also possible that Lehman was referring to another multi-state observational study of fentanyl test strips published last year in the same journal. That study also found that few users discarded drugs after a positive result, but it reported "a four-fold reduction in the amount of drugs used following a positive fentanyl test strip result." It also found that a positive result was associated with "participants taking turns" and "having naloxone nearby."
Both of these studies are evidence of harm reduction working at the margin, which is all it ever claimed to do. Furthermore, the fact that a person facing withdrawal will take chances on a dose of unknown potency does not refute the claim that deaths would plummet if a pharmaceutical-quality alternative were widely available, or the conclusion that overdoses are mostly a product of the drug war.
The Cost of the Drug War
I wrote that Lehman "ignored the other side of the ledger." I noted that Washington's drug possession arrests fell by 91 percent, avoiding around 20,000 arrests, and that hundreds of thousands of prior convictions became eligible to be vacated after State v. Blake, the 2021 Washington Supreme Court decision that temporarily decriminalized simple drug possession. I added that those criminal justice burdens fell most heavily on black and Native American residents, who were also dying at the highest rates. Here is Lehman's reply:
Yeah man I don't think it was worth an extra 1,000 drug OD deaths to prevent 8,000 possession arrests, especially given that those possession arrests basically never resulted in prison time. pic.twitter.com/d00v986MNT
— Charles Fain Lehman (@CharlesFLehman) July 15, 2026
Where did Lehman get these figures? He seems to be comparing the study's estimate of additional overdose deaths during the entire post-decriminalization period of 43 and 28 months in two states against roughly a year's worth of avoided arrests in one state. My claim, in any case, was that Lehman ignored one side of the ledger. I didn't offer my opinion on which side was heavier. Readers can weigh deaths against arrests themselves, but they need both numbers stated on the same basis.
Roughly 260,000 felony drug possession convictions became eligible to be vacated, along with refunds of the fines and fees paid. All that was thrown out by the court decision that forced decriminalization. Tack on what a felony drug record does to employment, housing, personal relationships, and child custody. Arrests can ruin lives.
Lehman's point is also logically inconsistent. If possession arrests were inconsequential to users, as he suggests, how could the removal of that threat have caused 1,000 deaths? A toothless sanction does not change behavior very much.
Was There a Surge in Crime?
I wrote that Lehman folded in a "surge in crime," even though the mortality study says nothing about crime. I added that "the arrest study he implicitly relies on found no statistically significant increase in violent or property offenses attributable to Measure 110." He replied:
This is literally false? He didn't even bother to follow the links to the study (https://t.co/urWRKLdWOP) I was pointing to? What the fuck? pic.twitter.com/K3lXxSEjLM
— Charles Fain Lehman (@CharlesFLehman) July 15, 2026
I did follow his link, which went to a Wall Street Journal op-ed Lehman wrote on the topic. That piece discusses the paper linked in Lehman's X post: a 2025 working paper titled "The Unintended Consequences of State-Level Drug Decriminalization Laws: Implications for Arrest and Crime Rates." When I wrote that the study "he implicitly relies on found no statistically significant increase in violent or property offenses attributable to Measure 110" (the Oregon ballot initiative), I had mistakenly looked at a different paper linked in the same op-ed. Reason has added a correction to my original article.
Lehman's Wall Street Journal op-ed nevertheless overstates the 2025 paper's findings, calling them "startling." I read the study and contacted its authors to get the full story. Like the mortality paper, this study has not yet been reviewed, which is why it should be discussed with provisional language. It finds statistically significant increases in serious violent and property crime in Oregon and Washington relative to other states, concentrated in Portland.
In the Journal, Lehman wrote, "A new study finds that Oregon's and Washington state's experiments with decriminalizing drug possession caused a surge in serious violent and property crime, especially in Portland and Seattle." In fact, the Seattle results are mixed, with robbery and drug offenses up but burglary and murder down relative to other large Washington cities.
The study window runs from 2019 through 2022, so it includes one year before COVID-19, the George Floyd protests, and a policing collapse that hit Portland and Seattle harder and longer than almost anywhere in the country. Decriminalization in Oregon and Washington both started in early 2021, which is also the year the FBI changed its crime reporting methods, causing well-documented gaps in agency coverage.
The paper's authors are appropriately careful about all of this: They describe factors that "may have" contributed to crime trends and do not claim to isolate one. Meanwhile, a federally funded study covering 2008 through 2024 attributes the crime increases in Oregon to trends predating decriminalization, and an analysis of Portland and Eugene 911 calls found no change relative to comparison cities.
I sent my own characterization of the 2025 study Lehman highlighted to its authors. University of South Carolina criminologist Hunter Boehme, a co-author of the paper, replied for the group, confirming that I had accurately represented their within-city findings and thanking me for noting the care with which the preprint states them.
Lehman took one working paper with nuanced findings from a contested literature, cherry-picked carefully hedged findings, and presented it as settled fact.
A Conclusive Result?
I wrote that the mortality study's result "hinges on rejecting one modeling choice" from a prior paper on the same topic by political scientist Michael Zoorob and colleagues. Lehman replied:
Just to spell it out: this is completely false. Not only do the authors of the new paper look at a variety of alternative specifications; they also retest the referred to "modeling choice" with more data, showing that even under that assumption, the fent effect goes away. pic.twitter.com/W2HCNMB9mP
— Charles Fain Lehman (@CharlesFLehman) July 15, 2026
I went over this issue at length with two of the study's co-authors, Hansen and Hall, and I stand by my characterization of the study as hinging on "rejecting one modeling choice."
The authors reject Zoorob's fentanyl control. That control is the share of police drug seizures containing fentanyl. Decriminalization cut non-fentanyl seizures, shrinking the denominator. Since fentanyl was policed as the greater menace, fentanyl seizures likely shrank less. If Hansen and Hall are correct, Zoorob's control exaggerated the increase in fentanyl.
The authors built two alternatives: fentanyl reports per capita, which is a measure Zoorob's own team used in earlier work, and fentanyl's share of opioid reports, which puts fentanyl in both the numerator and denominator. Under those measures, they report, the estimated effect of decriminalization is positive and significant in most specifications.
Hansen and Hall say the prior debate about overdose trends and decriminalization's effect on overdose deaths was premature. The earlier dueling studies ran through 2021 or 2022. If you extend them through 2023 using each study's own methods, Hansen and Hall say, they all now point in the same direction: unusually large overdose increases in Oregon and Washington. Their timing tests also cut against the simplest rival story: that fentanyl merely reached the Pacific Northwest late and explains everything by itself.
The authors also question whether it is appropriate to control for fentanyl at all. Hansen offered an analogy from labor economics: You can explain most of the gender wage gap by controlling for occupation, but that would absorb the effect it was meant to isolate if the factor you are studying influences occupational choice. Similarly, if decriminalization aided the spread of fentanyl, controlling for fentanyl blames the drug for a policy error.
Overdose deaths in Oregon and Washington broke away from their statistical twins after decriminalization. Why they broke away is a hard question, and Hansen does not pretend otherwise.
Hansen offered his own list of rival explanations, including hospital staffing, homelessness policy, and hidden shifts in drug smuggling networks, adding that such things are "incredibly challenging to get good data on." The size of the estimate also depends on modeling choices. Even with the extra year of data used in the new study, the choice of fentanyl measurement changes the drug's estimated effect by a factor of seven, and the four-figure death increase cited by Lehman requires assuming no fentanyl effect at all.
Hansen et al.'s NBER paper has advanced the debate, but it has not laid the issue to rest. The study's authors don't say that it has. Lehman is the one making that claim.
The post Did Decriminalizing Drugs Lead to More Deaths in Washington and Oregon? appeared first on Reason.com.
