CoChat
Sign Up
PreprintCited by 2Policing Practices and PerceptionsCrime Patterns and InterventionsCriminal Justice and Penology

What is the Best Response? Examining the Impact of Police and Their Alternatives

Open access
Ba, Bocar, Chen, Patton, Cheng, Tony, Eies, Martha, Holz, Justin·2025·National Bureau of Economic Research
DOI 10.3386/w34344
Open this paper in CoChat — free account

Abstract

This paper evaluates Durham, North Carolina's HEART program, which diverts nonviolent 911 calls from police. Using a difference-in-differences design, we find that HEART reduces crime reports, arrests, and response times—primarily through civilian phone and in-person responses. The program increases future 911 calls, fostering public trust, and is fiscally self-sustainable.

Integrity signals

Registry

Crossref · 10.3386/w34344

Cites

0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

10 of 25 cited sources checked · 3 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Crisis calls, if mishandled, can generate significant externalities, from community distrust to legal risks.

    “Crisis calls, if mishandled, can generate significant externalities, from community distrust to legal risks (Bor et al., 2018;Ang, 2020;Ang et al., 2024).”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites Bor et al., 2018

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Ang, 2020 source
    • cites Ang et al., 2024 source

      The Abstract reports that high-profile police violence may erode community engagement and that crime victims became less likely to report incidents because of mistrust of police. The Conclusion reports a substantial decline in 911 calls per gunshot. The paper does not address mishandling of crisis calls or resulting legal risks; no page numbers are supplied in our copy. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
  • Premisecitation

    One limitation of contingent valuation (CV) studies is that stated willingness to pay (WTP) may overstate true WTP, as some respondents may feel social pressure to indicate support for policies, even if they would not do so in an actual decision-making context.

    “One limitation of contingent valuation (CV) studies is that stated willingness to pay (WTP) may overstate true WTP, as some respondents may feel social pressure to indicate support for policies, even if they would not do so in an actual decision-making context.(Diamond and Hausman, 1994;List, 2025;Bursztyn et al., 2025).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Diamond and Hausman, 1994 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites List, 2025 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites Bursztyn et al., 2025 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
  • Study design and conductown data

    Across all outcomes, the bounds from the Rambachan and Roth (2023) approach confirm the robustness of the DiD estimates.

    “Across all outcomes-crime reports, arrests, response time, and backup requests-the bounds confirm the robustness of our DiD estimates.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Rambachan and Roth (2023) source
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.
  • Findingown data

    The reductions are primarily through civilian phone and in-person responses, rather than police-civilian co-responses.

    “primarily through civilian phone and in-person responses, rather than policecivilian co-responses.”
    Quote in the paperSubstantial claim
  • Findingown data

    Violent crime reports and arrests also declined more for calls handled by HEART than for those handled by police.

    “although violent crime reports and arrests also declined more for calls handled by HEART than for those handled by police.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 110
  • Classified as probably not a substantial claim. 11
  • The quote only partly states this claim. 1

  • May combine two claims.

    Interpretationcitation

    The demographic composition of the reference group confirms that Black individuals and men remain most affected, which reinforces how alternative crisis response models can reduce law enforcement contact among historically heavily policed populations.

    “The demographic composition of the reference group, individuals processed by police, confirms that Black individuals and men remain most affected, which reinforces how alternative crisis response models can reduce law enforcement contact among historically heavily policed populations (Ba et al., 2021;Cox et al., 2024).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Ba et al., 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Cox et al., 2024 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    To mitigate concerns about hypothetical bias, budget neglect, and sample validity, we follow best practices from both the CV and survey experiment literatures.

    “To mitigate concerns about hypothetical bias, budget neglect, and sample validity, we follow best practices from both the CV (Bishop et al., 2017;Domínguez and Scartascini, 2024) and survey experiment literatures (Stantcheva, 2023).”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Bishop et al., 2017 source
    • cites Domínguez and Scartascini, 2024 source
    • cites Stantcheva, 2023 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    To enhance consequentiality, the survey told participants that their responses would be shared with Durham mayor Leonardo Williams, policymakers and local law enforcement to inform future decisions.

    “To enhance consequentiality, the survey told participants that their responses would be shared with Durham mayor Leonardo Williams, policymakers and local law enforcement to inform future decisions (Bishop et al., 2017).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Bishop et al., 2017 source

      The ECONOMIC VALUE, TESTS OF CONSISTENCY section (page unavailable) explains that respondents see a survey as consequential when they believe their choices may influence a policy decision. It also says the authors emphasized their study’s importance for policymaking, but it does not say responses were shared with Durham’s mayor, policymakers, or law enforcement. (LLM, conf=0.83)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Similar trade-offs in cost, accountability, and service quality characterize the outsourcing of other criminal justice services, such as prison management.

    “Similar trade-offs in cost, accountability, and service quality characterize the outsourcing of other criminal justice services, such as prison management(Mukherjee, 2021;Mukherjee and Sanders, 2021;Yegen, 2021).”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites Mukherjee, 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Mukherjee and Sanders, 2021 source

      The Introduction discusses the costs of prison temperature control, and the Data section describes Mississippi correctional facilities. Neither addresses outsourcing prison management or trade-offs in cost, accountability, and service quality. Searches for private and contracted prison management found no passage addressing that claim. (LLM, conf=0.74)

      Experimental — accuracy not yet measured
    • cites Yegen, 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Findingown data

    HEART reduces crime reports, arrests, and response times.

    “HEART reduces crime reports, arrests, and response times”
    • Also stated in: § 1 Introduction · plain
      “HEART's rollout led to significant reductions in crime reporting, arrests, and response times.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Conditional parallel trends holds.
    • The authors state this as an assumption: No anticipation holds.
    • The authors state this as an assumption: No spillovers holds.
    • The authors state this as an assumption: Call classification is consistent.
    • The authors state this as an assumption: The comparison group is exogenous.

  • May combine two claims.

    Findingown data

    Co-responses have minimal impact on crime or arrests, with patterns in outcomes resembling those of calls receiving traditional police handling.

    “In contrast, coresponses have minimal impact on crime or arrests, with patterns in outcomes resembling those of calls receiving traditional police handling.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The effects are smaller for mental health crises and disturbances and are minimal for calls related to violent incidents or domestic violence.

    “The effects are smaller for mental health crises and disturbances and are minimal for calls related to violent incidents or domestic violence.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Response times decreased for both program and police responses, but the decline is nearly twice as large for HEART responses.

    “Response times decreased for both program and police responses, but the decline is nearly twice as large for HEART responses.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Conditional parallel trends holds.
    • The authors state this as an assumption: No anticipation holds.
    • The authors state this as an assumption: No spillovers holds.
    • The authors state this as an assumption: Call classification is consistent.
    • The authors state this as an assumption: The comparison group is exogenous.

  • May combine two claims.

    Findingown data

    Police responses increase crime reports from their pre-program baseline, particularly for the 'other types' category (7.3 pp, SE = 0.9) and violent or domestic violence incidents.

    “In contrast, police responses increase crime reports from their pre-program baseline, particularly for the”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingreasoning

    The MVPF is infinity, meaning the program pays for itself through fiscal externalities.

    “implies an MVPF of infinity, which means that the program pays for itself through fiscal externalities.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    There are no reductions in crime reports or arrests among police responses.

    “We observe no reductions among police responses”
    • Also stated in: § 1 Introduction · plain
      “There is no comparable change for calls receiving police responses.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Conditional parallel trends holds.
    • The authors state this as an assumption: No anticipation holds.
    • The authors state this as an assumption: No spillovers holds.
    • The authors state this as an assumption: Call classification is consistent.
    • The authors state this as an assumption: The comparison group is exogenous.

  • May combine two claims.

    Findingown data

    For all outcomes, HEART responses are associated with reductions in use of force, but these reductions are not statistically significant.

    “For all outcomes, HEART responses are associated with reductions in use of force, but these reductions are not statistically significant.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Qualified cases tend to occur in lower-income areas than unqualified cases, but there is little difference on this metric between the HEART- and police-handled cases.

    “Qualified cases tend to occur in lower-income areas than unqualified cases, but there is little difference on this metric between the HEART-and police-handled cases.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Over the longer horizon (more than 31 days), HEART responses increase the likelihood of new calls by 4.3 pp (SE = 0.0077), slightly more than the 3.2 pp increase from police responses (SE = 0.0060).

    “Over the longer horizon (more than 31 days), HEART responses increase the likelihood of new calls by 4.3 pp (SE = 0.0077), slightly more than the 3.2 pp increase from police responses (SE = 0.0060).”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Co-response is deployed most densely in high-crime areas, while civilian phone responses are more evenly distributed but remain concentrated in moderate-crime areas.

    “Co-response is deployed most densely in high-crime areas, while civilian phone responses are more evenly distributed but remain concentrated in moderate-crime areas.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    A HEART response significantly reduces arrests, particularly for trespass or nuisance (-2.2 pp, SE = 0.4) and the 'other types' category (-2.4 pp, SE = 0.5), with smaller effects for disturbances and violent incidents.

    “Arrest patterns follow a similar trend: A HEART response significantly reduces arrests, particularly for trespass or nuisance (-2.2 pp, SE = 0.4) and the”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    After adjusting for SDB, the estimated WTP is $80, six times the direct cost of HEART.

    “After adjusting for SDB, we estimate a WTP of $80, six times the direct cost of HEART.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Reductions for disturbance and mental health calls are smaller, while effects on violent or domestic violence incidents are negligible.

    “The reductions for disturbance and mental health calls are smaller, while the effects on violent or domestic violence incidents are negligible.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Civilian in-person responses most effectively reduce arrests but are the most expensive at $98 per hour, increasing labor costs by over $30 per call.

    “Civilian in-person responses most effectively reduce arrests but are the most expensive at $98 per hour, increasing labor costs by over $30 per call.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian phone and in-person responses both reduce crime reports and arrests, though the magnitude of the effect varies.

    “Civilian phone and in-person responses both reduce crime reports and arrests, though the magnitude of the effect varies.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Program responses reduce arrests and crime reports but increase requests for backup, particularly for disturbance (4.7 pp, SE = 0.9) and mental health calls (6.7 pp, SE = 0.5).

    “While program responses reduce arrests and crime reports, they increase requests for backup, particularly for disturbance (4.7 pp, SE = 0.9) and mental health calls (6.7 pp, SE = 0.5), a result that highlights a key operational trade-off.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Arrests followed similar trends, with the program driving larger reductions, particularly for nonindex crimes.

    “Arrests followed similar trends, with the program driving larger reductions, particularly for nonindex crimes.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    The mean WTP is $102.91 per year, more than eight times the program's per-resident cost.

    “with the mean valuation being $102.91 per year-more than eight times the program's per-resident cost.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian phone responses are the least expensive at $32.50 per hour, reducing labor costs by nearly $49 per call.

    “Civilian phone responses are the least expensive at $32.50 per hour, reducing labor costs by nearly $49 per call.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The city's Black and Hispanic residents are concentrated in central and eastern Durham, where violent crime rates are also highest.

    “Figure 2 shows that the city's Black and Hispanic residents are concentrated in central and eastern Durham, where violent crime rates are also highest.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Differences in racial composition across the case types are minor in magnitude and not statistically significant.

    “Differences in racial composition across the case types are minor in magnitude and not statistically significant.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Backup requests initially rose (by 4.1 pp relative to a 3% baseline) but declined as the program matured.

    “backup requests spiked in the early months but declined as the program matured”
    • Also stated in: § 1 Introduction · plain
      “Backup requests initially rose by 4.1 pp relative to a 3% baseline, but this increase was concentrated in the early months post-rollout, diminishing over time.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    While point estimates suggest potential benefits, the effects are not statistically distinguishable from zero.

    “Overall, while the point estimates suggest potential benefits, the effects are not statistically distinguishable from zero.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    HEART responses consistently reduce crime reports, with the largest declines for the 'other types' category (-5.7 pp, SE = 0.9) and trespass or nuisance (-3.6 pp, SE = 0.5).

    “HEART responses consistently reduce crime reports, with the largest declines observable for the”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Other crime reports, covering 9% of cases, declined by 5.9 pp for HEART responses, with little change for police responses.

    “Other crime reports, covering 9% of cases, declined by 5.9 pp for HEART responses, with little change for police responses.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Co-response shows no discernible effect and police-only cases register a negative shift of 1.9 pp in the probability of follow-up calls with the same nature code.

    “Co-response shows no discernible effect and police-only cases register a negative shift of 1.9 pp.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Over the analyzed period from 2021 to 2024, the total direct cost of the program amounted to $17,433,076, equivalent to $1,191 for each of the 14,637 HEART responses in the sample.

    “Over the analyzed period from 2021 to 2024, the total direct cost of the program amounted to $17,433,076, equivalent to $1,191 for each of the 14,637 HEART responses in our sample.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The effects of co-response are more variable: the estimate for total time is large and negative (-126.2, SE = 27.7), while the estimates for the other outcomes are imprecise and not statistically significant.

    “The effects of co-response are more variable: The estimate for total time is large and negative (-126.2,SE = 27.7),while the estimates for the other outcomes are imprecise and not statistically significant.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    The direct operational cost of a HEART response is $1,191, more than offset by fiscal savings, resulting in a net government cost impact of -$902.

    “In turn, the direct operational cost of a HEART response is $1,191, a figure more than offset by the fiscal savings, resulting in a net government cost impact of -$902.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Relative to comparison cases, cases assigned a civilian phone responder have a significantly lower likelihood of crime report issuance by 18.5 pp (SE = 0.046) and a total on-scene time that is nearly 59 minutes shorter (SE = 9.13).

    “For example, relative to comparison cases, cases assigned a civilian phone responder have a significantly lower likelihood of crime report issuance by 18.5 pp (SE = 0.046) and a total on-scene time that is nearly 59 minutes shorter (SE = 9.13).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    HEART introduction led to an overall cost reduction, with total crime cost savings of approximately $2,093, largest components being quality-of-life improvements, reductions in tangible costs, and smaller perpetrator work losses.

    “Our analysis reveals an overall cost reduction, with the largest components of the savings being quality-of-life improvements, reductions in tangible costs, and smaller work losses on the part of perpetrators, leading to a total crime cost savings of approximately $2,093.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Effects for other groups are smaller and not statistically significant.

    “The effects for other groups are smaller and not statistically significant, suggesting that the program's largest impact is on populations historically overrepresented in arrests.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    Civilian-led crisis response (HEART) reduces crime reports, arrests, and response times relative to calls not qualified for a program response.

    “This paper evaluates Durham's HEART program, providing evidence that civilian-led crisis response reduces crime reports, arrests, and response times relative to the outcomes of calls not qualified for a program response.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Cases with a police-only response or co-response show no robust effects across key outcomes, and the estimates are more sensitive to deviations from parallel trends.

    “In contrast, cases with a police-only response or co-response show no robust effects across key outcomes, and the estimates are more sensitive to deviations from parallel trends.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of qualified and unqualified calls would have followed similar trends over time.

  • May combine two claims.

    Findingown data

    Across most outcomes, the TSLS estimates are directionally consistent with the main results but generally smaller in magnitude and less precise.

    “Across most outcomes, the TSLS estimates are directionally consistent with the main results but generally smaller in magnitude and less precise.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingreasoning

    For unbounded WTP ranges, the triangular distribution gives $-10.50 as the conditional mean on (-∞, 3) and $127.11 as the conditional mean on (108, ∞).

    “This gives $-10.50 as the conditional mean on (-∞, 3) and $127.11 as the conditional mean on (108, ∞).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: For unbounded WTP ranges, the conditional distribution of WTP is triangular.

  • May combine two claims.

    Findingown data

    The main analysis finds that HEART reduced the total response time by 5.4 minutes (SE = 1.03), while the TSLS estimate is -21.6 minutes (SE = 5.12).

    “For example, the main analysis finds that HEART reduced the total response time by 5.4 minutes (SE = 1.03), while the TSLS estimate is -21.6 minutes (SE = 5.12).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Across most of the outcomes, the TSLS results are qualitatively consistent with the DiD estimates but generally of larger magnitude and lower precision.

    “Across most of the outcomes, the results are qualitatively consistent with the DiD estimates but generally of larger magnitude and lower precision.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Black individuals, men, and people aged 25-39 had the highest baseline arrest rates, while Hispanic, white, female, and younger individuals had lower rates.

    “The top panel shows arrest rates before the launch of the program by demographic group: Black individuals, men, and people aged 25-39 had the highest baseline rates, while Hispanic, white, female, and younger individuals had lower rates.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian in-person responses lead to a statistically significant decline in crime reporting and an increase in time spent on scene, consistent with their more service-oriented role.

    “Civilian in-person responses lead to a statistically significant decline in crime reporting and an increase in time spent on scene, consistent with their more service-oriented role.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The program significantly reduced arrests among Black individuals (-2.4 pp, p < 0.01), men (-2.2 pp, p < 0.01), and people aged 25-39.

    “The bottom panel presents DiD estimates, showing that the program significantly reduced arrests among Black individuals (-2.4 pp, p < 0.01), men (-2.2 pp, p < 0.01), and people aged 25-39.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Findingown data

    HEART reduces criminalization and arrests, leading to cost savings from averted crimes and reductions in public safety expenditures.

    “Motivated by our consistent finding that the HEART program reduces criminalization and arrests, leading to cost savings from both averted crimes and reductions in public safety expenditures, we incorporate these effects by constructing an index that captures the fiscal externalities that lead to changes in public safety costs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For overall crime reporting, the main estimate is a 9.4 pp reduction (SE = 0.005), but it is 7.9 pp under TSLS (SE = 0.024).

    “For overall crime reporting, the main estimate is a 9.4 pp reduction (SE = 0.005), but it is 7.9 pp under TSLS (SE = 0.024).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Civilian phone increases need backup by 0.076 (se=0.018); civilian in-person increases by 0.044 (se=0.003).

    “Mean Dep.= 0.03 β Phone = 0.076 (se=0.018)β In-Person = 0.044 (se=0.003)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Backup requests significantly increase for civilian phone response, while no meaningful change is observed for the other response types.

    “Backup requests significantly increase for civilian phone response, while we observe no meaningful change for the other response types.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Findingown data

    Civilian phone response reduces hourly costs by $48.901 (se=2.214); civilian in-person increases by $30.910 (se=1.427).

    “Mean Dep.= 56.29 β Phone = -48.901(se=2.214)β In-Person = 30.910(se=1.427)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian phone response reduces total time by 25.890 min (se=2.440); civilian in-person increases total time by 5.668 min (se=1.018).

    “Mean Dep.= 50.02 β Phone = -25.890(se=2.440)β In-Person = 5.668 (se=1.018)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    After the program was launched, violent crime reports declined by 0.8 pp for police responses to qualified calls and by 3.9 pp for calls with HEART responses, with p < 0.01 for both coefficients.

    “after the program was launched, this figure declined by 0.8 pp for police responses to qualified calls and by 3.9 pp for calls with HEART responses, where p < 0.01 for both coefficients.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Property crime reports dropped by 0.8 pp from their pre-program baseline of 1% with HEART responses and by 0.4 pp under police responses.

    “Property crime reports dropped by 0.8 pp from their pre-program baseline of 1% with HEART responses and by 0.4 pp under police responses.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian phone response yields clear reductions in response time, while the estimates for other models, including police, are imprecise and overlap with zero.

    “In terms of response time, civilian phone response yields clear reductions, while the estimates for other models, including police, are imprecise and overlap with zero.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Findingown data

    Arrests for violent crime, with a 3% baseline rate, fell by 1.3 pp for qualified calls with police responses and 1.5 pp for calls with HEART responses.

    “Arrests for violent crime, with a 3% baseline rate, fell by 1.3 pp after the program's launch for qualified calls with police responses and 1.5 pp for calls with HEART responses.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Civilian phone responders initially issued more requests for backup when the program was first rolled out, but these needs diminished over time.

    “Although civilian phone responders initially issued more requests for backup when the program was first rolled out, these needs diminished over time.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Crime reports and arrests fell after the program's implementation, with the largest effects for violent and other crimes.

    “Crime reports and arrests fell after the program's implementation, with the largest effects for violent and other crimes.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Findingown data

    TSLS estimates show that civilian in-person response reduces crime reports by 0.0672 (SE = 0.0262) and increases total time by 9.590 minutes (SE = 5.325).

    “Civilian In-Person -0.0672 * * 0.0122 9.590 * 0.0585 * * * (0.0262) (0.0169) (5.325) (0.0119)”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The instrument (implementation of HEART interacted with eligibility) satisfies monotonicity: it does not decrease the probability of a HEART response for any call.
    • Rests on an assumption: The instruments affect outcomes only through HEART response (exclusion restriction).

  • May combine two claims.

    Findingown data

    Backup requests rose by 4.1 pp (SE = 0.003) for calls with HEART responses, more than doubling the baseline.

    “but rose by 4.1 pp (SE = 0.003)more than doubling the baseline-for calls with HEART responses.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The average willingness to pay is $102.91 per year ($0.28 per day) or $80 per year ($0.21 per day) after adjusting for social desirability bias.

    “The average WTP is $102.91 per year ($0.28 per day) or $80 per year ($0.21 per day) after adjusting for SDB.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian responses are associated with significantly lower call priority, with phone diversions showing the largest decrease.

    “and are associated with significantly lower call priority, with phone diversions showing the largest decrease.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Co-response increases hourly costs by $11.617 (se=1.869); police-only reduces by $3.409 (se=0.866).

    “Mean Dep.= 56.29 β Co-Response = 11.617(se=1.869)β Police = -3.409(se=0.866)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Police handled 103,256 calls; 8,909 received a civilian in-person response; 2,857 a civilian phone response; and 2,871 a co-response.

    “police handling 103,256 calls, while 8,909 receive an civilian inperson response, 2,857 a civilian phone response, and 2,871 co-response.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    TSLS estimates show that civilian phone response reduces crime reports by 0.185 (SE = 0.0457) and total time by 58.95 minutes (SE = 9.130).

    “Civilian Phone -0.185 * * * -0.00259 -58.95 * * * 0.0622 (0.0457) (0.0178) (9.130) (0.0408)”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The instrument (implementation of HEART interacted with eligibility) satisfies monotonicity: it does not decrease the probability of a HEART response for any call.
    • Rests on an assumption: The instruments affect outcomes only through HEART response (exclusion restriction).

  • May combine two claims.

    Findingown data

    Civilian phone response reduces crime reports by 0.035 (se=0.010); civilian in-person reduces by 0.073 (se=0.004).

    “Mean Dep.= 0.16 β Phone = -0.035(se=0.010)β In-Person = -0.073(se=0.004)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Civilian phone response increases arrest reports by 0.013 (se=0.003); civilian in-person reduces by 0.019 (se=0.002).

    “Mean Dep.= 0.05 β Phone = 0.013 (se=0.003)β In-Person = -0.019(se=0.002)”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The number of local governments implementing behavioral health crisis response programs has grown over time.

    “figure shows the growth in the number of local governments implementing behavioral health crisis response programs over time.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The expanding footprint of civilian in-person responses over time may indicate a growing reliance on nonpolice crisis intervention and an extension of alternative response models beyond the most crime-dense neighborhoods.

    “The expanding footprint of civilian in-person responses over time may indicate a growing reliance on nonpolice crisis intervention and an extension of alternative response models beyond the most crime-dense neighborhoods.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    HEART responses shift cases away from formal criminal processing but require greater coordination and resources to manage crises effectively.

    “HEART responses shift cases away from formal criminal processing but require greater coordination and resources to manage crises effectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Civilian crisis teams can divert low-risk calls from the criminal legal system, reducing formal enforcement without compromising safety.

    “they can divert low-risk calls from the criminal legal system, reducing formal enforcement without compromising safety.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    HEART is both welfare-enhancing and self-sustaining, delivering substantial economic and social returns.

    “Together with the high WTP, these findings suggest that HEART is both welfare-enhancing and self-sustaining, delivering substantial economic and social returns.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Civilian teams handle emergencies without escalation and may encourage continued, constructive use of emergency services.

    “Taken together, the results suggest that civilian teams handle emergencies without escalation and may encourage continued, constructive use of emergency services.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Civilian crisis response models can reduce contact between individuals in crisis and the criminal justice system and racial disparities in enforcement, even without explicit targeting.

    “These results underscore the potential for civilian crisis response models to reduce contact between the individuals in crisis and the criminal justice system and racial disparities in enforcement, even without explicit targeting.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Relative to a police response, HEART responses do not increase the risk of escalation and may foster willingness to continue to engage with emergency services without increasing use of force or violent outcomes.

    “These results suggest that relative to a police response, HEART responses do not increase the risk of escalation and may foster willingness to continue to engage with emergency services without increasing the use of force or violent outcomes.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Despite the existence of over 100 crisis response programs nationwide, little is known about how effective they are, how well they integrate with other emergency services, and which models work best.

    “Despite the existence of over 100 such programs nationwide (Community Safety Workgroup, 2025), we still know little about how effective they are, how well they integrate with other emergency services, and which models work best”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Community Safety Workgroup, 2025

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Study design and conductown data

    Demographic and economic characteristics are broadly similar across the case groups, with small differences in racial composition and median household income.

    “Demographic and economic characteristics are broadly similar across the case groups, with small differences in racial composition and median household income.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    The 2020 murder of George Floyd accelerated efforts to rethink public safety, spurring a wave of crisis response programs.

    “The 2020 murder of George Floyd accelerated efforts to rethink public safety (Akbar, 2020;Kaba et al., 2021;Phelps, 2024), spurring a wave of crisis response programs aimed at reducing police involvement in mental health, substance use, and quality-of-life emergencies.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Akbar, 2020 source
    • cites Kaba et al., 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Phelps, 2024

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationown data

    Public valuation of HEART is high and credible, with robust support across valuation methods and program types.

    “Together, these estimates point to a public valuation of HEART that is both high and credible; indeed, we find robust support across valuation methods and program types.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The fiscal savings from HEART fully offset its direct costs, meaning the program effectively funds itself through reductions in criminal justice expenditures.

    “This would suggest, in other words, that the fiscal savings from the program fully offset its direct costs, meaning that the program effectively funds itself through reductions in criminal justice expenditures.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductnone given

    Overall, the program effects remain negative and meaningful under reasonable assumptions.

    “Overall, the program effects remain negative and meaningful under reasonable assumptions;”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of qualified and unqualified calls would have followed similar trends over time.
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Interpretationreasoning

    These results highlight the key distinction between phone and in-person crisis response: while both reduce crime reports, in-person teams more effectively prevent escalation without requiring additional support from law enforcement.

    “These results highlight the key distinction between phone and in-person crisis response: While both reduce crime reports, in-person teams more effectively prevent escalation without requiring additional support from law enforcement.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    To ensure data quality, we incorporated several validation steps throughout the survey, including multiple attention checks, a comprehension check confirming understanding of HEART services, and a bot-detection question.

    “To ensure data quality, we incorporated several validation steps throughout the survey, including multiple attention checks, a comprehension check confirming understanding of HEART services, and a bot-detection question that prompted respondents to copy and paste copyrighted content into a text box.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Given per-response WTP of $5,872, HEART generates positive net benefits, implying an MVPF of infinity, as fiscal savings fully cover operational costs.

    “Furthermore, given the per-response WTP of $5,872, the program generates positive net benefits, implying an MVPF of infinity (∞), as the fiscal savings fully cover the operational costs.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    We estimate WTP using a preregistered contingent valuation (CV) survey.

    “We estimate WTP using a preregistered contingent valuation (CV) survey, a widely adopted approach for valuing nonmarket goods.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Civilian phone responses appear to encourage repeat engagement with emergency services for similar crises, while in-person teams are associated with a small reduction in such calls; police-centered approaches show either no impact or a suppressive effect on repeat calling.

    “In short, civilian phone responses appear to encourage repeat engagement with emergency services for similar crises, while its in-person teams are associated with a small reduction in such calls.Police-centered approaches show either no impact or a suppressive effect on repeat calling.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Arrests drop more under a HEART response, suggesting that traditional policing may result in unnecessary contact with law enforcement, especially for groups with higher baseline arrest rates.

    “Arrests drop more under a HEART response, suggesting that traditional policing may result in unnecessary contact with law enforcement, especially for groups with higher baseline arrest rates.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    These findings reflect the efficiency of remote interventions and the more challenging, time-intensive character of the interventions by in-person crisis teams.

    “These findings reflect the efficiency of remote interventions and the more challenging, time-intensive character of the interventions by in-person crisis teams.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The broader pattern confirms that the program consistently lowers crime reports, arrests, and response times, while the effects for police are less clear.

    “However, the broader pattern confirms that the program consistently lowers crime reports, arrests, and response times, while the effects for police are less clear.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    For total response time, the estimates show greater variability, particularly for co-response, but the posttreatment effect for civilian phone response remains robust.

    “For total response time, the estimates show greater variability, particularly for co-response, but the posttreatment effect for civilian phone response remains robust.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of qualified and unqualified calls would have followed similar trends over time.
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Interpretationown data

    These results suggest that civilian models-particularly phone response-generate consistent reductions in enforcement outcomes without evidence of overreaction or compromised safety.

    “These results suggest that civilian models-particularly phone response-generate consistent reductions in enforcement outcomes without evidence of overreaction or compromised safety.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The stabilization of backup requests from civilian phone responders over time suggests that the initial challenges in remote crisis management were ameliorated, but the smaller impact on arrests for responses of this kind indicates the limits of phone-based interventions in fully deescalating certain situations.

    “The stabilization of backup requests from civilian phone responders over time suggests that the initial challenges in remote crisis management were ameliorated, but the smaller impact on arrests for responses of this kind indicates the limits of phone-based interventions in fully deescalating certain situations.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Appendix Figures A.3 and A.4 present event-study estimates that show no significant differences in pretreatment trends, supporting the validity of the DiD design.

    “Appendix Figures A.3 and A.4 present event-study estimates that show no significant differences in pretreatment trends, supporting the validity of the DiD design.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of treatment, the outcomes of qualified and unqualified calls would have followed similar trends over time.
    • The authors state this as an assumption: In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

  • May combine two claims.

    Interpretationreasoning

    The decline in backup requests for civilian phone responses suggests better call triage and more confidence in remote handling of crises, which may have mitigated the need for escalation.

    “Over time, backup requests for civilian phone responses declined, suggesting better call triage and more confidence in remote handling of crises, which may have mitigated the need for escalation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The pretrends appear flat across all outcomes, supporting the parallel trends assumption for a causal interpretation.

    “The pretrends appear flat across all outcomes, supporting the parallel trends assumption for a causal interpretation.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Study design and conductown data

    The results show no significant pretrends in crime reports, arrests, or backup requests, confirming that the observed post-treatment effects are not driven by preexisting differences.

    “The results show no significant pretrends in crime reports, arrests, or backup requests, confirming that the observed post-treatment effects are not driven by preexisting differences.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Roth (2022) source
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Study design and conductown data

    The program reduces response time but increases backup requests, with no evidence of overestimation.

    “The program reduces response time but increases backup requests, though we see no evidence of overestimation.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Study design and conductown data

    Arrests also decline, with the bounds ruling out large positive effects.

    “Arrests also decline, with the bounds ruling out large positive effects.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Interpretationreasoning

    The alternative forms of emergency response offered by the program lead to a shift in the handling of behavioral health crises and quality-of-life incidents away from police without increasing crime.

    “The alternative forms of emergency response offered by the program lead to a shift in the handling of behavioral health crises and quality-of-life incidents away from police without increasing crime.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of the program's introduction, trends in outcomes for calls qualified for HEART and comparison calls with a police response would have remained parallel.
    • The authors state this as an assumption: The instrument significantly predicts the likelihood of a HEART response (relevance).
    • The authors state this as an assumption: Potential outcomes are independent of HEART assignment, conditional on the instrument (independence).
    • The authors state this as an assumption: The instrument affects the outcome only through its impact on treatment probability (exclusion restriction).
    • The authors state this as an assumption: The instrument consistently increases the likelihood that a call would receive a HEART response (monotonicity).

  • May combine two claims.

    Interpretationreasoning

    Civilian phone response generates the largest per-call savings because of lower wages and the smaller volume of calls receiving responses of this type, but its overall budget impact remains limited.

    “Civilian phone response generates the largest per-call savings because of the lower wages-since only one employee handles each call, in contrast to the multiperson teams used in the civilian in-person and co-response modelsand the smaller volume of calls receiving responses of this type, but its overall budget impact remains limited.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Co-response and police responses show smaller but similar effects, suggesting spillovers.

    “Co-response and police responses show smaller but similar effects, suggesting spillovers.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The first-stage F-statistic for the TSLS models is 143.8.

    “F-Stat 143.8 143.8 143.8 143.8”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    There are no meaningful changes in the outcomes of police-handled cases post-implementation, suggesting that the observed effects stem from the program itself rather than changes in officer behavior.

    “There are no meaningful changes in the outcomes of police-handled cases post-implementation, which suggests that the observed effects stem from the program itself rather than changes in officer behavior.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

  • May combine two claims.

    Interpretationown data

    Taken together, the descriptive results support the view that expanding civilian alternatives normalizes 911 use for everyday problems and, in doing so, may build public confidence in emergency services.

    “Taken together, these descriptive results support the view that expanding civilian alternatives normalizes 911 use for everyday problems and, in doing so, may build public confidence in emergency services.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Co-response falls in between, reducing costs modestly while maintaining police involvement.

    “Co-response falls in between, reducing costs modestly while maintaining police involvement.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    In 2022, the Durham Community Safety Department launched HEART to provide alternatives to a police-only response for nonviolent mental health crises and quality-of-life-related concerns, with the aim of reducing unnecessary police involvement.

    “In 2022, the Durham Community Safety Department launched HEART to provide alternatives to a police-only response for nonviolent mental health crises and quality-of-liferelated concerns, with the aim of reducing unnecessary police involvement 4 .”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [4] 4. Ang, D. (2020, 09). The Effects of Police Violence on Inner-City Students. The Quarterly Journal of Economics. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    HEART responders reported feeling safe in approximately 99% of incidents as of summer 2025.

    “As of summer 2025, HEART responders are required to submit post-response safety assessments to the Department of Community Safety, with data from the public dashboard indicating they reported feeling safe in approximately 99% of incidents.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites DCSD

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    The fiscal savings from HEART exceed its direct costs, indicating that it effectively funds itself through externalities.

    “Applying the marginal value of public funds framework, we find that the fiscal savings from HEART exceed its direct costs, which indicates that it effectively funds itself through externalities.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: In the absence of the program's introduction, trends in outcomes for calls qualified for HEART and comparison calls with a police response would have remained parallel.
    • The authors state this as an assumption: The instrument significantly predicts the likelihood of a HEART response (relevance).
    • The authors state this as an assumption: Potential outcomes are independent of HEART assignment, conditional on the instrument (independence).
    • The authors state this as an assumption: The instrument affects the outcome only through its impact on treatment probability (exclusion restriction).
    • The authors state this as an assumption: The instrument consistently increases the likelihood that a call would receive a HEART response (monotonicity).

  • May combine two claims.

    Interpretationown data

    HEART's unarmed civilian channels appear to bring a new subset of lower-stakes crises into the formal emergency system, while police-involved options remain focused on higher-priority incidents.

    “In practical terms, HEART's unarmed civilian channels appear to bring a new subset of lower-stakes crises into the formal emergency system, while police-involved options remain focused on higher-priority incidents.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Civilian in-person response is the most expensive model, reflecting the city's decision to prioritize higher-touch, in-person interventions, and its expansion significantly increases labor costs.

    “Civilian in-person response, by contrast, is the most expensive model, reflecting the city's decision to prioritize higher-touch, in-person interventions, and its expansion significantly increases labor costs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Pretrend tests for HEART and police responses on crime reports, arrests, total time, and need backup yield p-values greater than 0.05, supporting parallel trends.

    “Pretrend p-value = 0.37”
    • Also stated in: § Need Backup · plain
      “Pretrend p-value = 0.38”
    • Also stated in: § Need Backup · plain
      “Pretrend p-value = 0.12”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    In the period prior to the program's implementation, trends in outcomes for qualified and comparison calls are flat and parallel, supporting the identification strategy.

    “For the period prior to the program's implementation, the trends in outcomes for qualified and comparison calls are flat and parallel, supporting the identification strategy.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The sharp post-introduction divergence indicates the program's substantial impact on response rates, reinforcing the strength of the instrument.

    “The sharp post-introduction divergence indicates the program's substantial impact on the program's response rates, reinforcing the strength of the instrument.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study was approved by the IRB at Duke (IRB 2025-0376).

    “The study was approved by the IRB at Duke (IRB 2025-0376).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The first-stage coefficient on the interaction between post-introduction and qualification indicators is 0.147 (SE = 0.003), with an F-statistic of 1,874.9, confirming a strong first stage.

    “The coefficient on the interaction between the post-introduction and qualification indicators is 0.147 (SE = 0.003), with a nF-statistic of 1,874.9,confirming a 1 strong first stage.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The joint F-test for the program's responses (Column (1)) is statistically insignificant (p = 0.125), indicating no correlation with call characteristics.

    “The result from the joint F-test for the program's responses (Column (1)) is statistically insignificant (p = 0.125), indicating no correlation with call characteristics.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The joint F-test for the instrument (Column (2)) is statistically insignificant (p = 0.162), supporting the assumption that the instrument is balanced.

    “The result from the joint F-test for the instrument (Column (2)) is also insignificant (p = 0.162), supporting the assumption that the instrument is balanced.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Across all specifications, the first-stage estimate remains positive and highly significant, ranging from 0.137 to 0.157.

    “Across all specifications, the first-stage estimate remains positive and highly significant, ranging from 0.137 to 0.157.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The field experiment was preregistered in the AEA RCT Registry (AEARCTR-0016060).

    “The field experiment was preregistered in the AEA RCT Registry (AEARCTR-0016060).”
    Quote in the paperNot a substantial claim
119 of 119

  • The authors state this as an assumption.

    Conditional parallel trends holds.

    “conditional parallel trends”

  • The authors state this as an assumption.

    No anticipation holds.

    “no anticipation”

  • The authors state this as an assumption.

    No spillovers holds.

    “no spillovers”

  • The authors state this as an assumption.

    Call classification is consistent.

    “We also assume consistent call classification”

  • The authors state this as an assumption.

    The comparison group is exogenous.

    “exogeneity of the comparison group”

  • The authors state this as an assumption.

    HEART responses do not influence the outcomes for other, untreated calls, for example they do not alter how police respond in comparison cases.

    “We assume SUTVA, meaning that HEART responses do not influence the outcomes for other, untreated calls-for example, they do not alter how police respond in comparison cases.”

  • The authors state this as an assumption.

    Treatment status in the post-period does not affect outcomes in the pre-period.

    “The noanticipation assumption requires that treatment status in the post-period not affect outcomes in the pre-period-that is, that Y i,pre (0) = Y i,pre (d) for all treated units.”

  • The authors state this as an assumption.

    In the absence of treatment, the outcomes of qualified and unqualified calls would have followed similar trends over time.

    “The parallel trends assumption states that, in the absence of treatment, the outcomes of qualified and unqualified calls would have followed similar trends over time:”

  • The authors state this as an assumption.

    In the absence of treatment, the outcomes of HEART-qualified and comparison calls would have followed similar trends conditional on observed covariates.

    “Specifically, we assume: E[Y i,post (0) -Y i,pre (0) | D i = 1, X i ] = E[Y i,post (0) -Y i,pre (0) | D i = 0, X i ]”

  • The authors state this as an assumption.

    For all calls that eventually received a HEART response after the program began, in the pre-period those same calls would have been handled by police.

    “In other words, for all calls that eventually received a HEART response after the program began, we assume that in the pre-period, those same calls would have been handled by police.”

  • The authors state this as an assumption.

    Call coding was consistent over time; HEART did not change how incidents were classified.

    “Fourth, we assume consistent call coding over time, that is, that HEART did not change how incidents were classified.”

  • The authors state this as an assumption.

    Comparison calls are not indirectly influenced by the presence of HEART and are never routed to HEART teams in practice.

    “Finally, we assume exogeneity of the comparison group: that comparison calls are not indirectly influenced by the presence of HEART and, importantly, are never routed to HEART teams in practice.”

  • The paper relies on this without support.

    Officers may adjust their behavior in response to the program's presence, so qualified police-handled calls are not used as a control group.

    “We do not use qualified police-handled calls as a control group because officers may adjust their behavior in response to the program's presence.”

  • The paper relies on this without support.

    Alternative crisis response programs have expanded rapidly, but their perceived value remains unquantified.

    “Although alternative crisis response programs have expanded rapidly, their perceived value remains unquantified.”

  • The authors state this as an assumption.

    For unbounded WTP ranges, the conditional distribution of WTP is triangular.

    “For the unbounded ranges, we assume that the conditional distribution of the WTP is triangular.”

  • The authors state this as an assumption.

    Overstaters state WTPs drawn randomly from the same distribution as truth-tellers, but have a true WTP of 0.

    “Next, assume that overstaters state WTPs drawn randomly from the same distribution as truthtellers, but have a true WTP of 0.”

  • The authors state this as an assumption.

    Truth-tellers honestly report their WTP.

    “Because truth-tellers honestly report their WTP, E[True WTP|Truth-teller] = E[Stated WTP|Truth-teller].”

  • The authors state this as an assumption.

    In the absence of HEART, outcomes for HEART-qualified calls and unqualified calls would have followed parallel trends.

    “supporting the parallel trends assumption for a causal interpretation.”
    “The figures plot potential violations of parallel trends based on Roth (2022).”

  • The authors state this as an assumption.

    The instrument (implementation of HEART interacted with eligibility) satisfies monotonicity: it does not decrease the probability of a HEART response for any call.

    “Table A.4: Tests of Monotonicity Assumption”
    “The interaction between the time since introduction and an indicator for calls qualifying for a HEART response serves as the instrument, with unqualified calls as the control group.”

  • The paper relies on this without support.

    The instruments affect outcomes only through HEART response (exclusion restriction).

    “using response-type-specific eligibility interacted with post-period indicators as instruments.”
    “The instruments are defined as the interaction between an indicator for post-introduction periods and response-type-specific eligibility, with unqualified calls serving as the control group.”

  • The authors state this as an assumption.

    In the absence of the program's introduction, trends in outcomes for calls qualified for HEART and comparison calls with a police response would have remained parallel.

    “The key identifying assumption for the first stage under DiD is that in the absence of the program's introduction, trends in outcomes for calls qualified for HEART and comparison calls with a police response would have remained parallel.”

  • The authors state this as an assumption.

    The instrument significantly predicts the likelihood of a HEART response (relevance).

    “Relevance requires that the instrument significantly predict the likelihood of a HEART response, meaning that qualified calls post-introduction must be likelier to have received the intervention.”

  • The authors state this as an assumption.

    Potential outcomes are independent of HEART assignment, conditional on the instrument (independence).

    “Independence assumes that the potential outcomes are independent of the HEART assignment, conditional on the instrument, indicating that the trends in outcomes would have been similar for qualified and comparison calls without the program.”

  • The authors state this as an assumption.

    The instrument affects the outcome only through its impact on treatment probability (exclusion restriction).

    “The exclusion restriction requires that the instrument affect the outcome only through its impact on treatment probability.”

  • The authors state this as an assumption.

    The instrument consistently increases the likelihood that a call would receive a HEART response (monotonicity).

    “Finally, monotonicity requires that the instrument consistently increase the likelihood that a call would receive a HEART response.”
25 of 25

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.

  • Arrest effects for civilian phone responses have opposite signs

    Major

    The paper says both fully civilian models reduce arrests, but its response-type results say civilian phone responses increase arrests by 1.3 percentage points.

  • Fiscal-savings account includes a benefit omitted from its stated cost measure

    Major

    Equation 8 defines total costs using seven tangible categories and does not include quality-of-life losses. The later fiscal-savings account names quality-of-life improvements among its largest components, so the stated cost measure and the savings used for the self-funding conclusion do not match.

  • Main violent-arrest estimate differs from Table A.5

    Minor

    The robustness discussion gives the main-model effect on violent-crime arrests as −1.3 percentage points. Table A.5 reports −0.009, or −0.9 percentage points, for that main-analysis outcome.

  • Main response-time estimate differs from the comparison table

    Minor

    The robustness discussion attributes a 5.4-minute reduction to the main analysis, but Table A.5 labels −2.592 minutes as its main-analysis estimate for total time.

  • Police violent-arrest estimates disagree across results

    Minor

    The offense-type discussion reports a 1.3-percentage-point decline for police responses, consistent with Table 2's −0.0128. Appendix Figure A.2 instead labels the police estimate −0.004, or 0.4 percentage points.

  • Co-response labor costs are described as falling despite a positive estimate

    Minor

    The labor-cost discussion calls co-response a cost-reducing model, whereas the reported estimate is an increase of $11.62 per call.

  • First-stage cross-effects are not consistently negative

    Minor

    The prose says the off-diagonal first-stage coefficients are consistently negative and that phone eligibility crowds out in-person responses. Table A.6 instead reports positive coefficients for phone eligibility predicting in-person and co-responses, and for in-person eligibility predicting co-response.

  • Other demographic arrest effects are described as insignificant

    Minor

    The narrative says effects for groups other than those highlighted are not statistically significant. Table 4 marks the HEART estimates for White, Hispanic, female, and other age groups as statistically significant.

8 of 8

Cites

The publisher hasn’t deposited this paper’s reference list.

Research works printed but not linked: 5

  1. [17]17. City of Durham (2023). Fiscal year 2023-24 adopted budget. Pages 238-239.
  2. [19]19. Cohen, M. A. (2000). Measuring the costs and benefits of crime and justice. Criminal justice 4(1), 263-315.
  3. [52]52. Phelps, M. S. (2024). The Minneapolis Reckoning: Race, Violence, and the Politics of Policing in America. Princeton University Press.
  4. [55]55. Rivera, R. G. and B. A. Ba (2023, 10). The effect of police oversight on crime and misconduct allegations: Evidence from chicago. The Review of Economics and Statistics, 1-45.
  5. [63]63. Stevenson, M. T. and S. G. Mayson (2022). Pretrial detention and the value of liberty. Virginia Law Review 108(3), pp. 709-782.

Cited by

Sorted by citations

Related

Authors
Ba, B., Chen, P., Cheng, T., Eies, M., & Holz, J.
Affiliations
  • National Bureau of Economic Research (US)
  • Duke University (US)
  • University of Michigan (US)
ORCID
Venue
National Bureau of Economic Research
Publisher
National Bureau of Economic Research
Published
2025-10
DOI
10.3386/w34344
OpenAlex
W4415286458
Cited by
2
References
0
Citations by year
2026: 2
Type
report
Language
en
Open access
Locations
National Bureau of Economic Research (publishedVersion)
Funders
  • Arnold Ventures
  • National Institute of Child Health and Human Development (HD065563)
  • Eunice Kennedy Shriver National Institute of Child Health and Human Development (P2CHD065563)
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
alternatives to policing, crime reduction, arrests, police response time, public trust in police, cost-effectiveness, fiscal sustainability, Durham, North Carolina

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.

Cited sources are checked by 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