This latest blog in our ongoing research into co-response programs in New Jersey focuses on the methods other programs have used and what we can learn from these research efforts about how to best study these programs. Co-response programs, like New Jersey’s ARRIVE Together program, are designed to pair local law enforcement with emergency medical staff to address mental health crises.
Studies typically focus on diversion from hospitalizations and criminal justice as main outcomes, often operationalized by reductions in arrests and jail bookings. Safety for police officers and service users is another common benefit identified in co-response programs and measured by examining the use of force rates and injuries, but not as commonly reported in studies. The use of force is largely considered to be affected through de-escalation techniques.
Other typical outcomes include improved linkage to community mental health services—identified as one of the most empirically supported benefits across evaluations—and an increased likelihood that individuals who were in mental health crisis engage with follow-up services rather than avoiding them. Co-response is presumed to provide a more person-centered, compassionate encounter, compared to encounters with police that can be traumatic and provoke fear.
Generally, outcomes measured are those for which there exist administrative data, while effects for the same outcomes vary in size across studies. The variation suggests that context matters greatly, like the availability of mental health services in a jurisdiction. The stated or desired benefits of co-response programs include a wide range and perhaps it may not be optimal to try to achieve all of them in a single intervention. A more targeted focus when designing co-response programs may lead to more efficient use of resources and more effective outcomes.
Research evaluating co-response programs is large and growing but continues to predominantly rely on non-experimental studies. While researchers often take great care to perform rigorous evaluations and seek to identify credible causal estimates of program effectiveness using statistical adjustments, the evidence from observational studies is less robust and more circumspect. Among the main reasons is that 911 calls that co-response teams respond to are not randomly assigned, leaving non-experimental studies vulnerable to selection bias.
Randomized controlled trials or field experiments address these problems more credibly, by design. Yet trials have to contend with another set of challenges. The generally small numbers of treated and control cases in trials are unable to detect small but meaningful effects of co-response programs. They are also costly, have limited external validity, and must confront serious ethical concerns when randomizing a likely beneficial treatment—which is what happens when responding to a 911 call about a person in crisis.
A comprehensive and most insightful evaluation of a program would combine experimental and careful non-experimental designs, but these cannot be done well without detailed and complete longitudinal administrative records. In addition, the critical element for evaluations is to ensure an adequate control group to which co-response programs are compared. Evaluations would also benefit from a qualitative perspective that seeks to understand how service users, mental health professionals, and police officers perceive the program—both in terms of what works and what does not work. Finally, long term assessments are rare in existing studies, yet they are necessary to understand how durable co-response interventions are and whether they reduce the revolving door of criminal justice contact for vulnerable people in crisis.