Sending a police officer is not the only possible response to a mental-health or behavioural crisis. Across the United States, mobile crisis response programmes have increasingly placed clinicians, medics and crisis workers into calls that might otherwise reach police. The policy argument is compelling, but measuring whether these programmes actually reduce police workload has proved surprisingly difficult.
A new peer-reviewed study of CAHOOTS, the long-running Crisis Assistance Helping Out On The Streets programme in Eugene, Oregon, attempts to solve that measurement problem. Published in Science Advances, the research estimates that CAHOOTS directly substituted for police on about 18% of relevant responses. When the researchers also accounted for reductions in police calls consistent with a prevention effect, the estimated diversion rate rose to 23%. Among calls that could plausibly have been answered by either CAHOOTS or police, the crisis team handled about half.
Those numbers matter because previous claims about CAHOOTS have varied widely. Estimates of its diversion effect have ranged from roughly 3% to 20%, often because different analyses counted different things. The new study does not simply add another percentage to that debate. It proposes three separate measures designed to distinguish direct substitution from broader changes in demand for police response.
Why diversion is harder to measure than it sounds
CAHOOTS is the longest-running mobile crisis response programme in the United States. Its teams respond to situations such as mental-health crises, welfare concerns, substance-related problems and other calls that can sometimes be handled without a conventional law-enforcement response.
The simplest way to estimate diversion would be to count every CAHOOTS response and assume that each one replaced a police response. That approach is attractive but potentially misleading. Some calls may have been created specifically because a non-police service existed. Others might never have resulted in a police dispatch. In still other cases, a crisis team may resolve an earlier problem in a way that prevents a later call to police.
This means two programmes could handle the same number of incidents while having very different effects on policing. A programme could mostly absorb calls that police would otherwise have handled, or it could primarily serve demand that would not have reached police at all. It could also alter what happens after an initial intervention, changing future call volume.
Nathan Burton, Claire Herbert and Rori Rohlfs therefore treated diversion as a causal inference problem rather than a simple accounting exercise. Their analysis used computer-aided dispatch records obtained from the Eugene Police Department covering January 2016 through December 2021 and applied a systematic difference-in-differences design to estimate how police-handled call volume changed when CAHOOTS was available to respond.
Three measures answer three different questions
The researchers introduced three diversion estimators. The first, the Substitution Diversion Rate, asks how much of the crisis programme’s activity directly replaces police response. For CAHOOTS, this measure was estimated at 18%.
That estimate is important because it does not assume that every crisis-team call would otherwise have been handled by police. Instead, it focuses on the portion of CAHOOTS responses associated with a corresponding reduction in police-handled calls. Put differently, the 18% estimate is intended to capture substitution rather than simply programme activity.
The second measure expands the question. The Prevention-adjusted Diversion Rate allows for the possibility that a crisis response can reduce subsequent demand for police. Under this definition, the estimated diversion rate increased to 23%.
The difference between 18% and 23% is not proof that every missing police call was prevented by CAHOOTS. The study interprets the additional reduction as being consistent with a prevention effect. That distinction matters. Difference-in-differences methods can strengthen causal inference by comparing changes across appropriate groups and periods, but they still depend on assumptions about what would have happened in the absence of the programme.
The third estimator, the Overlapping-mandate Diversion Rate, approaches the issue from the opposite direction. Rather than asking what share of all CAHOOTS work replaced police, it asks what share of calls that either service could plausibly answer were handled by the crisis programme. The researchers estimated this at 50%.
Together, the three figures describe different layers of the same system. The 18% estimate represents direct substitution, the 23% estimate incorporates the broader reduction consistent with prevention, and the 50% estimate describes how work was divided within the subset of calls where the mandates of police and CAHOOTS overlapped.
The study used six years of dispatch records
The underlying administrative data span 1 January 2016 to 31 December 2021. The records identify incident classifications and responding units, including whether a CAHOOTS unit was dispatched. This gave the researchers a way to study actual emergency-response activity rather than relying on surveys of residents or responders.
The analytical strategy is a major part of the contribution. Difference-in-differences designs estimate an intervention’s effect by comparing changes in an exposed group with changes in a comparison group. In this setting, the researchers used patterns in call handling to estimate how police volume differed when CAHOOTS response was present, while attempting to separate the programme’s contribution from underlying differences in call demand.
This is a more demanding standard than dividing the number of CAHOOTS calls by the total number of calls received by a public-safety system. A raw share can describe workload, but it cannot by itself reveal the counterfactual: how many of those incidents would have required police if the alternative responder did not exist.
Why an 18% substitution rate can coexist with a 50% overlap rate
At first glance, the study’s headline estimates can look contradictory. If CAHOOTS directly diverted 18% of responses, how could it handle 50% of overlapping calls?
The answer lies in the denominator. The Substitution Diversion Rate concerns the relationship between crisis-team responses and police responses that were actually displaced. The Overlapping-mandate Diversion Rate narrows attention to calls that fall within the operational territory of both services. Within that smaller pool, CAHOOTS handled about half.
This distinction has practical consequences for cities trying to forecast staffing. A crisis programme’s total call count should not automatically be translated into the same number of police calls saved. Yet looking only at direct substitution can also miss a programme’s role within the subset of incidents for which alternative response is genuinely feasible.
The prevention result broadens the policy question
The increase from an 18% direct substitution estimate to a 23% prevention-adjusted estimate is arguably the study’s most conceptually interesting result. Emergency-response policy is usually evaluated at the moment a call is dispatched. But a successful crisis intervention could influence what happens hours, days or weeks later.
A non-police team may connect someone with services, de-escalate a recurring problem or resolve a situation without creating additional legal or interpersonal consequences. If those mechanisms reduce later calls, an evaluation focused only on immediate dispatch substitution would understate the programme’s effect.
At the same time, the study’s language around prevention deserves attention. The analysis identifies reductions in police-handled call volume that are consistent with prevention. Administrative dispatch data cannot observe every pathway through which a future incident does or does not occur. The estimate therefore supports a broader effect beyond direct substitution without turning the underlying mechanism into a certainty.
What the findings mean for public governance
For local governments, the research offers a more disciplined way to ask whether civilian crisis response changes the demand placed on police. That question is relevant not only to debates about policing, but also to budgeting, emergency-service design, mental-health provision and the allocation of specialised public-sector skills.
The findings suggest that mobile crisis programmes should not be judged solely by the proportion of all emergency calls they attend. Their effect depends on how much of their work truly substitutes for police, whether they change later demand, and how much of the call landscape is genuinely shared between services.
The three-estimator framework could therefore be useful beyond Eugene. A city considering an alternative-response programme could separately estimate direct substitution, prevention-adjusted diversion and overlapping jurisdiction. That would produce a more informative picture than a single headline diversion rate.
It could also help prevent exaggerated claims in either direction. A programme handling thousands of calls is not necessarily replacing thousands of police responses. Conversely, a relatively modest direct substitution percentage does not mean the programme has little operational importance if it handles a substantial share of calls within its appropriate mandate or reduces future demand.
Important limits remain
The study focuses on one programme in one US city. CAHOOTS has operated for decades and is embedded in Eugene’s particular dispatch system, service network and local institutions. Newer programmes in larger cities, rural areas or jurisdictions with different call classifications may produce different diversion patterns.
The analysis also depends on administrative dispatch records. These are valuable because they capture real service activity, but they were created for operational rather than research purposes. Call classifications, dispatch practices and recording conventions can affect what appears in the data.
Difference-in-differences designs additionally rely on assumptions about the counterfactual trend. The method is designed to estimate causal effects more rigorously than a raw before-and-after comparison, but it cannot make every unobserved difference disappear. The prevention component is particularly important to interpret cautiously because lower later police volume can be consistent with prevention without directly documenting each prevented incident.
Finally, diversion is only one outcome. The study addresses police call volume rather than providing a complete evaluation of patient health, safety, equity, responder workload, public satisfaction or programme cost. A system could divert calls effectively while performing differently on any of those other dimensions. Those outcomes require separate evidence.
A clearer way to count alternative response
The central lesson is methodological as much as political. Asking how many calls a crisis team answered is not the same as asking how many police responses it replaced.
For CAHOOTS, the researchers estimate that direct substitution amounted to 18%, that the figure reached 23% when reductions consistent with prevention were included, and that the programme handled 50% of calls within the overlapping mandate it shared with police. Those measures do not collapse into a single definitive score. They describe different parts of how an alternative emergency-response system changes public-service demand.
As more governments experiment with sending health and crisis professionals instead of police to selected calls, that distinction may become increasingly important. The question is no longer simply whether an alternative responder showed up. It is what would have happened if that responder had not been there.
Source Information
Study: Quantifying CAHOOTS: Measuring a mobile crisis response program’s impact on police response through call diversion and prevention
Authors: Nathan Burton, Claire Herbert and Rori Rohlfs
Journal: Science Advances, Volume 12, Issue 39
Publication date: 25 September 2026, electronically available 23 September 2026
DOI: 10.1126/sciadv.aeb8064
Study design: Difference-in-differences analysis of computer-aided dispatch data from Eugene, Oregon, covering 2016 to 2021
Data availability: Study data documentation and reproducibility materials are archived through Dryad and associated code repositories.
