CoChat

A community response approach to mental health and substance abuse crises reduced crime

Open access
Dee, Thomas S., Pyne, Jaymes · 2022 · Science Advances

The DOI matches its Crossref record; 38 of the 38 references it cites were found in a registry or an index; 7 of the 36 sources its claims cite were checked for support.

Abstract

Police officers often serve as first responders to mental health and substance abuse crises. Concerns over the unintended consequences and high costs associated with this approach have motivated emergency response models that augment or completely remove police involvement. However, there is little causal evidence evaluating these programs. This preregistered study presents quasi-experimental evidence on the impact of an innovative “community response” pilot in Denver that directed targeted emergency calls to health care responders instead of the police. We find robust evidence that the program reduced reports of targeted, less serious crimes (e.g., trespassing, public disorder, and resisting arrest) by 34% and had no detectable effect on more serious crimes. The sharp reduction in targeted crimes reflects the fact that health-focused first responders are less likely to report individuals they serve as criminal offenders and the spillover benefits of the program (e.g., reducing crime during hours when the program was not in operation).

Overview

CoChat Summary

TL;DR

A preregistered quasi-experimental evaluation of Denver’s Support Team Assistance Response (STAR) pilot found that directing targeted 911 calls to a mental health clinician and paramedic instead of police reduced reports of targeted, less serious crimes by 34%, with no detectable effect on more serious crimes. The authors argue this reduction reflects both fewer individuals in crisis being recorded as criminal offenders and genuine crime reductions, including spillovers during hours when STAR was not operating.

Key findings

  • The difference-in-differences estimate indicates STAR reduced STAR-related offenses by 34% (b = −0.41, SE = 0.07, t = −6.09, P < 0.001); the difference-in-difference-in-differences estimate implies a 30% reduction (b = −0.36, SE = 0.05, t = −6.64, P < 0.001).
  • There was no detectable effect on STAR-unrelated, more serious crimes (b = −0.05, SE = 0.04, t = −1.18, P = 0.245).
  • The authors estimate the pilot prevented nearly 1400 criminal offenses (84.3 × 0.34 × 8 × 6 = 1376) across eight precincts over 6 months; STAR responded to 748 incidents, and baseline offenses per STAR-related incident imply just over 1000 fewer recorded offenses (748 × 1.4 = 1047), so the total reduction exceeds what service calls alone would predict.
  • The 6-month pilot cost $208,141, or $151 per offense reduced, compared with an average direct criminal justice cost of $646 per minor offense in 2021 dollars; police categorization also suggested a 14% reduction in overall crimes, concentrated in alcohol and drugs, disorderly conduct, and other crimes against persons.

Methods

  • Preregistered quasi-experimental DD and DDD designs using Denver National Incident-Based Reporting System/Open Data Catalog offense data; panel of 36 precincts from December 2019 to November 2020, with 432 precinct-month observations, 6 months before and after the STAR pilot, and 8 treated precincts.
  • STAR operated from 1 June 2020 for a 6-month pilot, Monday to Friday, 10 a.m. to 6 p.m.; outcomes were natural log counts of STAR-related versus STAR-unrelated offenses, with Eq. 1 including precinct and month fixed effects and precinct-clustered errors, and Eq. 2 stacking STAR and non-STAR offenses (n = 864) with precinct-month, month-offense, and precinct-offense fixed effects.
  • Robustness checks included event-study parallel-trends tests, prior-year placebo datasets, Poisson and negative binomial specifications, generalized synthetic control, comparative interrupted time series, randomization inference, and checks for spatial and temporal spillovers.

Limitations

  • The paper states the reduced-form analysis cannot exactly decompose STAR’s overall impact into reduced recording of existing offenses versus genuine crime reductions; it also notes no access to arrest and citation data and that offenses are not differentiated by whether they led to arrest or citation, which may confound results.
  • The paper states generalizability is uncertain, replication likely depends on implementation details such as dispatcher and field-staff recruitment/training and police coordination, and it is unsure whether STAR increased community trust and willingness to call 911, which would likely imply the estimates underestimate the true effect; a fuller cost-benefit reckoning would also include health care costs and benefits.
  • Evident from the text: the pilot was limited to 6 months and eight purposefully chosen central downtown Denver precincts whose residents were largely more affluent, educated, and white than the city as a whole, though all but one neighborhood were also designated displacement-vulnerable; the study period overlaps the COVID-19 pandemic, although the authors report robustness checks using only data after March 2020.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1126/sciadv.abm2106

Cites

38 of 38 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1. Changes in criminal offenses before and after the STAR pilot implementation. Thick black lines surround the police precincts where the STAR program was active.
Fig. 1. Changes in criminal offenses before and after the STAR pilot implementation. Thick black lines surround the police precincts where the STAR program was active.CC BY · PubMed Central
Fig. 2. Estimated effects of the STAR program on criminal offenses. The DD estimates are based on 432 precinct-month observations and condition on precinct fixed effects and month fixed effects. The DDD estimates are based on the stacked precinct-month data for STAR and non-STAR offenses (n = 864). The DDD estimates condition on fixed effects unique to each category of the following two-way interactions: precinct-by-month offense, precinct-by-STAR offense, and month-by-STAR offense. The outcome variables are the natural log of the offense counts. Dots are the coefficients; bars are 95% confidence intervals. See table S4 for numerical results.
Fig. 2. Estimated effects of the STAR program on criminal offenses. The DD estimates are based on 432 precinct-month observations and condition on precinct fixed effects and month fixed effects. The DDD estimates are based on the stacked precinct-month data for STAR and non-STAR offenses (n = 864). The DDD estimates condition on fixed effects unique to each category of the following two-way interactions: precinct-by-month offense, precinct-by-STAR offense, and month-by-STAR offense. The outcome variables are the natural log of the offense counts. Dots are the coefficients; bars are 95% confidence intervals. See table S4 for numerical results.CC BY · PubMed Central
Fig. 3. Event-study model. The DD event-study estimates are based on 432 precinct-month observations and condition on precinct fixed effects and month fixed effects. The outcome variables are the natural log of the STAR-related and STAR-unrelated offense counts. The event-study estimates identify for each outcome the regression-adjusted treatment comparison differences by month relative to the first time period (i.e., December 2019). The vertical line separates pretreatment months from the months after the STAR pilot program that began in June 2020. See table S5 for numerical results.
Fig. 3. Event-study model. The DD event-study estimates are based on 432 precinct-month observations and condition on precinct fixed effects and month fixed effects. The outcome variables are the natural log of the STAR-related and STAR-unrelated offense counts. The event-study estimates identify for each outcome the regression-adjusted treatment comparison differences by month relative to the first time period (i.e., December 2019). The vertical line separates pretreatment months from the months after the STAR pilot program that began in June 2020. See table S5 for numerical results.CC BY · PubMed Central

Verification

Claims the paper makes

6 Valid · 0 Not valid · 15 Not checked · 60 Own results and interpretations

55 more claims

Create a free account to see all 81 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 11
  • Probably not a substantial claim. 12
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 38 references

Research works printed but not linked: 4

  1. National Academies of Sciences Medicine and Engineering Proactive Policing: Effects on Crime and Communities (National Academies Press 2018).
  2. G. L. Kelling, J. Q. Wilson, Broken windows. Atl. Mon. 249, 29–38 (1982).
  3. D. Mitchell The Right to the City: Social Justice and the Fight for Public Space (Guilford Press 2003).
  4. “DJP helps launch alternative public health emergency response pilot in Denver” press release (Denver Justice Project 2020).

Cited by

See all 87

Metadata

Authors
Dee, T. S., & Pyne, J.
Affiliations
  • National Bureau of Economic Research (US)
  • Stanford University (US)
Corresponding author
  • Dee, Thomas S.
  • Pyne, Jaymes
Venue
Science Advances
Published
2022-06-10
Volume
8
Issue
23
Pages
eabm2106-eabm2106
ISSN
2375-2548 (ISSN-L)
OpenAlex
W4282942270
Cited by
87
References
34
Citations by year
2026: 36 · 2025: 17 · 2024: 12 · 2023: 16 · 2022: 6
FWCI
41.62
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Science Advances (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • OSF Preprints (OSF Preprints) (submittedVersion, cc-by)
  • OSF Preprints (OSF Preprints) (submittedVersion)
Topics
  • Social Sciences › Political Science and International Relations › Policing Practices and Perceptions
  • Social Sciences › Sociology and Political Science › Crime Patterns and Interventions
  • Social Sciences › Law › Criminal Justice and Penology
Keywords
community response, mental health crisis, emergency calls, crime reduction, trespass, public disorder, resisting arrest, spillover effects, de-escalation, Denver

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.