The study, published in Community Mental Health Journal, is the first empirical test of a 911-to-crisis-line call routing intervention of this kind. Using a quasi-experimental technique known as synthetic control analysis, Leonard Swanson, Aaron Gottlieb, and Margaret Thomas compared Mesa’s suicide call rates after the intervention against a weighted blend of twenty comparable cities across the United States. Their findings indicate that embedding a crisis line specialist at the 911 center reduced police responses to suicide calls by roughly 22 to 48 calls per quarter per 100,000 residents relative to the synthetic control. Translated into officer time, that reduction amounts to an estimated 228 to 491 police hours saved per quarter in a city of just over half a million people.

The logic behind the intervention rests on a mismatch in how Americans reach help during a mental health emergency. The 988 Suicide and Crisis Lifeline, launched nationwide in July 2022, offers around-the-clock counseling through a network of roughly 200 local call centers, and fewer than 2 percent of its calls ever require an emergency dispatch. Yet 911 remains the number people actually know and dial. An estimated 240 million 911 calls are placed every day across the country, and some analyses suggest that as many as one in five involve a mental health component. Even under a conservative estimate, mental health-related 911 calls would vastly outnumber the roughly six million calls 988 receives annually. Survey research underscores the awareness gap: only about one in five Americans knew of 988, while more than half said they would call 911 in a crisis.

Mesa’s path to reform began years before the embedded specialist arrived. In 2019, the Mesa Police Department and Solari, the nonprofit crisis line that answers 988 calls for Arizona, drafted a policy allowing 911 call-takers to route suicide calls to the crisis line when the caller expressed suicidal thoughts without plans or means and consented to the transfer. Dispatchers received an hour of training on identifying eligible calls. A similar policy for broader mental health calls followed in August 2020. But the decisive change came on June 5, 2021, when Solari placed a licensed counselor inside the emergency communications center itself. The embedded specialist did not take calls. Instead, the counselor sat alongside 911 professionals during their shifts, listened for the types and severity of calls Solari could handle, coached dispatchers in real time, and translated abstract routing policies into concrete, call-by-call decisions.

That physical presence, the researchers argue, may have been the ingredient that previous paper policies lacked. Emergency communication centers are overwhelmingly police-operated, and 911 professionals frequently default to law enforcement responses out of concern for legal liability and a lack of trust in alternative resources. Call-takers feel personally responsible when a caller expresses suicidal ideation, and strict call-taking protocols often leave little room for nuance. By working shoulder to shoulder with dispatchers, learning their procedures, norms, and coding practices, the embedded specialist built the interpersonal and institutional trust needed to make call routing a routine practice rather than a risky exception. Once the program was established, a behavioral health technician with a lower credentialing burden could assume the role, making it financially sustainable for other jurisdictions.

To measure the effect, the team turned to the synthetic control method, an approach originally developed in comparative politics and increasingly used in criminology. Rather than hunting for a single comparison city with matching pre-intervention trends, which is notoriously difficult, the method constructs a weighted combination of donor cities whose pre-intervention suicide call rates collectively mimic Mesa’s. The researchers scoured public data portals from all 369 U.S. cities with populations above 100,000, searching for jurisdictions that published computer-aided dispatch or calls-for-service data containing suicide-related call types. Twenty cities made the final donor pool. The resulting synthetic Mesa drew most heavily from Everett, Seattle, and Tacoma, with smaller contributions from New Orleans and Norman, and tracked Mesa’s actual suicide call rates closely before the intervention, with differences never exceeding 18 calls per 100,000 people per quarter.

After the specialist was embedded in mid-2021, the trajectories diverged. Mesa’s suicide call rate fell below its synthetic counterpart in quarter after quarter, with the gap widening to nearly 48 calls per 100,000 people by late 2022 before narrowing somewhat in 2023. To rule out chance, the researchers ran placebo tests, constructing synthetic controls for each donor city as if it too had received the intervention. Mesa’s observed decline exceeded nearly all of these placebo estimates, reaching conventional significance thresholds in several post-intervention quarters, including one quarter in which Mesa’s difference was larger than every placebo site. The team also conducted robustness checks, dropping individual cities to test for outsized influence and examining non-suicidal call trends for confounding influences, and the core result held.

The implications extend beyond officer hours. Police responses to mental health crises can escalate situations that might otherwise be resolved peacefully, sometimes ending in arrest, use of force, or death. Officers predominantly transport people in crisis to emergency departments, where patients can wait weeks for psychiatric placement in a practice known as ED-boarding, and where connections to follow-up care are often tenuous. Crisis lines, by contrast, can de-escalate a caller over the phone, activate a mobile crisis team, and link the person to ongoing mental health resources, all without a uniformed response. Deflecting calls upstream, before they enter the criminal-legal or acute care systems, is precisely the kind of intervention that crisis system frameworks from the Substance Abuse and Mental Health Services Administration envision, and the quantifiable reduction in police labor may give law enforcement administrators a pragmatic incentive to support such programs.

The study also exposes a stubborn data problem that hampers research in this field. The researchers had hoped to analyze whether Mesa’s broader mental health-related call rate declined as well, but definitions and coding practices for mental health calls vary so widely across jurisdictions that no meaningful cross-city comparison was possible. Some calls coded as mental health-related are not, and many genuinely mental health-related calls are buried under categories like check welfare or public disturbance. Suicidality, with its clearer concepts of ideation, threats, and means, proved far more comparable across cities. The authors note that this coding ambiguity itself shapes outcomes: when 911 professionals lack confidence in identifying mental health concepts that require clinical judgment, they tend to default to dispatching officers, and the absence of a standard definition inhibits empirical evaluation of deflection programs nationwide.

The authors are careful about causal claims. Mesa’s earlier routing policies laid groundwork that the embedded specialist built upon, and other cities may not replicate the results by embedding a counselor without first establishing transfer policies and training. The synthetic control method absorbed the effects of those earlier policies into the pre-intervention baseline, which strengthens the case that the observed decline reflects the embedded specialist’s distinctive contribution. Still, the association is striking: a single embedded clinician, working within a police-operated dispatch center, coincided with hundreds of police hours redirected away from suicide calls each quarter. As cities across the United States and abroad experiment with alternatives to police-led crisis response, the Mesa model suggests that the most powerful lever may not be a new mobile team or a retrained officer corps, but a counselor with a headset sitting in the room where the first call for help arrives.

Subject of Research: Embedding a mental health crisis line specialist in a 911 emergency communications center to reduce police responses to suicide calls

Article Title: Deflecting Suicide Calls From Law Enforcement With an Embedded Crisis Line Specialist at a 911 Center: A Synthetic Control Analysis

Article References: Swanson, L., Gottlieb, A., & Thomas, M. (2026). Deflecting Suicide Calls From Law Enforcement With an Embedded Crisis Line Specialist at a 911 Center: A Synthetic Control Analysis. Community Mental Health Journal. https://doi.org/10.1007/s10597-026-01726-2

Image Credits: AI Generated

DOI: 10.1007/s10597-026-01726-2

Keywords: 911, 988 crisis line, suicide prevention, mental health crisis response, police deflection, synthetic control method, emergency communications center, Mesa Arizona, crisis intervention, mobile crisis teams, law enforcement, community mental health

The central methodological challenge the researchers faced is one that haunts all ex-ante policy evaluation. Conventional tax studies examine what happened after a reform using observed post-policy data. Nigeria’s 2026 reform, by contrast, has produced no such observations, making retrospective evaluation impossible. Rather than abandoning quantitative analysis altogether, the team constructed a simulation-based counterfactual framework that combines twenty-five years of observed macroeconomic data, from 2000 to 2024, with scenario-based projections running from 2025 to 2035. Throughout the paper, the authors are careful to insist that their estimates are conditional projections, dependent on stated modelling assumptions, rather than forecasts or realised causal effects. This explicit humility is itself a departure from much of the fiscal policy literature, which often blurs the line between historical association and forward prediction.

The empirical foundation rests on a four-country comparative panel comprising Nigeria, South Africa, the United Kingdom, and Canada, chosen deliberately to span different stages of fiscal maturity and institutional development. The United Kingdom and Canada represent advanced economies with deep, well-administered tax systems; South Africa occupies an intermediate position; and Nigeria, with its narrow tax base, chronic compliance challenges, and constrained administrative capacity, stands in as the reforming developing economy. The dependent variable throughout is real GDP per capita growth, while the key fiscal predictors are the tax-to-GDP ratio, a measure of overall revenue mobilisation, and the direct tax share, which serves as a structural proxy for the progressivity of a tax system because direct taxes on personal income and corporate profits track ability-to-pay principles more closely than indirect taxes.

Using a two-way fixed effects panel model with Driscoll-Kraay standard errors to guard against cross-sectional and temporal dependence, the study finds that both fiscal variables carry positive and statistically significant associations with growth. A one-percentage-point increase in the tax-to-GDP ratio is associated with roughly 0.017 percentage points of additional annual real GDP per capita growth, while a one-percentage-point rise in the direct tax share corresponds to approximately 0.008 percentage points. The magnitudes are modest, and the authors repeatedly caution that reverse causality, where faster growth generates higher tax revenue, cannot be fully ruled out in an observational panel of only four countries. Still, the pattern is consistent with theoretical expectations drawn from Ramsey’s Optimal Tax Theory, Romer’s Endogenous Growth Theory, and Musgrave’s Ability-to-Pay Principle, which jointly suggest that well-designed taxation, embedded in effective institutions, can support long-run prosperity.

The most striking result in the historical analysis concerns institutional quality, which records by far the largest estimated coefficient at 0.521, with a p-value of 0.001. Gross fixed capital formation also matters, carrying a coefficient of 0.012. The implication is that taxation does not operate in a vacuum: it functions within a broader institutional and investment environment that determines whether mobilised revenue translates into productive outcomes. This finding aligns with a substantial body of literature showing that progressive tax reforms deliver macroeconomic benefits mainly where governance, administrative efficiency, and expenditure quality are strong, and that in settings like Nigeria, weak enforcement and institutional constraints can blunt or even negate the growth dividend of higher revenue collection.

To probe whether the taxation-growth relationship is nonlinear, the researchers estimated a threshold regression that allows the association between tax effort and growth to differ above and below an estimated tax-to-GDP tipping point. The results suggest that the positive association nearly doubles above the threshold, rising from approximately 0.041 to 0.072 percentage points of growth per additional percentage point of tax-to-GDP. Economies operating above an estimated fiscal threshold, in other words, historically extracted greater growth value from each unit of additional tax effort. The authors stress that with only four countries in the panel, these threshold estimates are exploratory rather than definitive, and bootstrap inference was used to assess the stability of the estimated break point. Nevertheless, the finding resonates with the idea that fiscal capacity exhibits increasing returns once administrative systems mature.

The heart of the paper, however, is its synthetic control simulation of the 2026 reform. The Synthetic Control Method traditionally compares a treated unit’s observed post-intervention outcomes with a weighted combination of untreated donors, selected so that the pre-treatment trajectories match closely. Here, because Nigeria’s post-2026 outcomes do not yet exist, the authors invert the logic: they simulate Nigeria’s reform trajectory under explicitly defined assumptions, including a gradual rather than immediate adjustment of the tax-to-GDP ratio and direct tax share from 2026 onward, and compare it against a no-reform counterfactual built from the three comparator economies. Weights on the donor countries were chosen to minimise pre-2025 distances in growth, fiscal structure, investment, and institutional quality, and placebo-adjusted checks were run to confirm that the projected gap for Nigeria is qualitatively distinguishable from gaps generated for the donors themselves.

Under these assumptions, the synthetic control estimate suggests a short-run deviation of approximately minus 0.48 percentage points in annual real GDP per capita growth relative to the counterfactual benchmark, with a 95 percent confidence interval running from minus 0.92 to minus 0.10. The placebo-adjusted estimate of minus 0.44 tells a nearly identical story. The authors interpret this projected negative gap not as evidence that the reform will contract the economy, but as a plausible transitional pathway: fiscal reforms of this scale typically coincide with implementation costs, administrative adaptation, and private-sector adjustment before longer-term gains emerge. In a country where tax administration capacity and compliance infrastructure are still evolving, such friction is unsurprising. The team also emphasises that alternative assumptions, external shocks, or stronger-than-expected implementation could produce entirely different trajectories.

As a validation layer, the researchers deployed a machine-learning T-learner, a technique that estimates separate predictive functions for baseline and reform scenarios and compares their outputs. Crucially, the T-learner was used exclusively as an out-of-sample predictive validation tool, trained with rolling time-series cross-validation and tuned within training folds only to prevent information leakage, and never as a source of causal inference. It produced a projected change of minus 0.39 percentage points, directionally consistent with the synthetic control result, though its wider confidence interval of minus 0.81 to 0.02 spans zero. The model achieved an out-of-sample root mean squared error of 0.61 and a cross-validated R-squared of 0.50, indicating meaningful predictive power without obvious overfitting. The directional agreement between a structural econometric method and a machine-learning approach strengthens confidence in the internal consistency of the simulation framework, even as it underscores the inherent uncertainty of any forward-looking exercise.

The policy implications are likely to be debated in Abuja and beyond. The study suggests that raising tax revenue alone will not guarantee immediate macroeconomic improvement unless accompanied by complementary institutional reforms: stronger tax administration, expanded digital compliance systems, better taxpayer education, reduced administrative leakage, and, critically, transparent and productive allocation of new revenue toward infrastructure, education, healthcare, and capacity building. The authors also recommend gradual sequencing of the reform to give households, firms, and administrators time to adapt. They acknowledge the limitations of their design candidly, from the four-country panel and the proxy nature of the direct tax share to the absence of micro-level behavioural data. Yet by integrating structural econometrics, counterfactual simulation, and machine-learning validation within a single transparent framework, the study offers a template for evaluating fiscal reforms in real time, a contribution that may prove as influential as its specific findings about Nigeria’s tax experiment.