Understanding Police Enforcement: A Multicity 911 Analysis Report Submitted to Arnold Ventures

By p.S. Rebecca Neusteter, Megan O’Toole, Mawia Khogali, Abdul Rad, Frankie Wunschel, Sarah Scaffidi, Marilyn Sinkewicz, Maris Mapolski, Paul DeGrandis, Daniel Bodah, and Henessy Pineda

At least 240 million calls to 911 are made each year.1 Responding to these calls takes up a sizable amount of police officers’ time, even though relatively few calls stem from crimes in progress. Despite their prevalence in police work, little research about the nature of 911 calls or how police respond is available. Basic information, such as the number of calls and reasons they are made, how call volumes vary across different call types, and what happens from the time a call is placed to when an officer arrives on the scene, is unknown. The 911 call system plays a critical role in policing practice and should be studied, not only to measure performance but also to aid in decision-making processes, inform strategic decisions, and understand opportunities to advance call processing and alternative responses.2 The current study was designed to define the landscape of 911 calls for police service and answer fundamental questions about how communications personnel and police respond to them. To begin, the study explores 911 call processing by examining what happens when 911 calls are answered and what training, protocols, standards, and management possibilities exist at each stage of 911 call processing. The study also examines how accurately 911 calls are categorized and handled when received by public safety personnel. Questions about the overall volume and rate of 911 calls for service, typical response time, and ordinary duration of responses to 911 calls, as well as how these vary by the call type, time, and location are also considered. To understand how characteristics of 911 calls impact police officers in the field, the study analyzes what proportion of officers’ activities represent responses to 911 calls versus those proactively initiated by officers. The study examines how 911 calls are resolved by identifying the categories of dispositions and their frequency, as well as how they vary by call volume, type, time, and location. The ultimate outcomes of police contacts initiated by 911 calls are also reviewed to understand what factors have the greatest contribution to 911 call responses. In addition, the current research examines communications systems among call-takers, dispatchers, and police officers in the field to determine whether all information relevant to outcomes is being effectively conveyed. The study further explores whether it is possible to improve outcomes for police and civilians by  identifying 911 calls that may be handled more appropriately by a response other than sending sworn officers. The following research activities provided details about the 911 landscape to address these questions: 1) 2) 3) 4) 5) an examination of prior research on 911 calls in the policing context; an analysis of 911 call and computer-aided dispatch (CAD) data to identify 911 call types, processing, and outcomes in Camden County (NJ) and Tucson (AZ) police and public safety communications departments; the development of a system processing map to trace calls from receipt through closure, which was achieved using data from focus groups, interviews, audio analysis of a sample of 911 calls, and field observations in Camden County Police Department (CCPD), Camden County Communications Center (CCCC), Tucson Police Department (TPD), and Tucson Public Safety Communications Division (PSCD); an examination of publicly available 911 call and CAD data from Detroit, New Orleans, and Seattle; and a convening of police, emergency communications practitioners, and other stakeholders to contextualize these findings and explore alternatives to sworn police response. The Vera Institute of Justice’s (Vera’s) review of the existing literature on 911 calls for service (detailed in Chapter 2) reveals a need for innovation in this space, as well as more research exploring key features of the system (such as call volumes, types, and outcomes at the national, state, and local levels). Since the birth of 911 in the late 1960s and its congressionally mandated national deployment in 1999, the emergency communications field has become professionalized and transformed by new technologies, such as Enhanced 911 (E911) and Next Generation 911 (NG911).3 However, much remains to be learned about how 911 calls are processed, how personnel are trained, and where opportunities for alternative responses need development or can be expanded. As a first step toward understanding how 911 calls are processed, Vera created a system processing map. This map (given in Chapter 3) shows that, when a community member calls 911, the caller relays information to a call-taker at a public safety communications center. The call-taker gathers relevant information from the caller; determines whether the call requires a response by fire, police, medical personnel, or a combination thereof; enters information and categorizes the call using a CAD system; and may give the caller instructions about what to expect or actions to take. The information the call-taker enters into the CAD system is sent to the appropriate dispatcher for further action. The dispatcher assigns officers to respond to the call based on the priority level of the reported incident, the narrative information entered in the CAD system by the call-taker, and available police resources. The dispatcher may, during this process, reclassify the call type or priority level. The assigned patrol officers then respond to the location given in the call, where they may take a report, provide instructions, resolve conditions found there, call for other resources, or take law enforcement action. Vera’s analysis of Camden and Tucson data shows that, with slight variations, this core set of actors and actions defines the landscape of 911 call processing. Within this system, call codes, training, and standards exist to guide the actions of call-takers and dispatchers; however, codes, training, and standards are not uniform across 911 call systems, and opportunities exist to improve outcomes by diverting appropriate calls to non-law enforcement responders. Vera’s detailed analysis of CAD data and 911 audio recordings from Camden and Tucson sheds further light on how the 911 system operates (presented in Chapters 4 and 5). • As many as half of CAD records may be of limited reliability due to lack of call type specificity and other call information omitted from the narrative. • Officers spend a substantial proportion of their time responding to calls for service, few of which are related to crimes in progress, let alone serious crime in progress. • Most calls do not relate to serious or violent crime; instead, the most frequent calls involve nuisance complaints and low-level crimes. • Trends across the departments differed. In 2016 and 2017, TPD officers spent most of their time responding to 911 calls for service, whereas CCPD officers engaged primarily in proactive police activity. (As explained in chapter 5, this finding may be a function of differences in departmental record keeping.) These observations of Camden and Tucson are further supported through the findings from the open data sites—Detroit, New Orleans, and Seattle. Highlights from the five-city analysis demonstrate the following: • The most frequent incident type was noncriminal in nature. In four of the five sites, the most frequent incident type was some variation of a complaint or request for an officer to perform a welfare check. Across all sites, the most common priority types were nonemergency. • The five sites have a wide range of dispatcher and officer response times. The two sites (Detroit and New Orleans) that have response time available by priority level show that response times are faster in emergency incidents. Among call types, the fastest response times for dispatchers and officers were behavioral health incidents, medical emergencies, traffic stops, officer requests for help, area checks, and alarms. • Examining CAD events generated through 911 calls for service and those that are officer-initiated reveals that, in Tucson and New Orleans, 911 calls were most prevalent in the CAD system. However, in both Camden and Seattle, officer-initiated events accounted for most CAD entries. In Detroit, the proportions of CAD entries varied across the study period, shifting from being mostly 911 responses to mostly officer-initiated events. The findings across all sites suggest the need for future research and local conversations about whether certain types of 911 calls for service require responses by police. There are critical gaps in knowledge regarding the underlying needs, causes, and consequences for these resource intensive calls for service that do not involve a crime. The current research also produced initial empirical evidence of how data collected by call takers and dispatchers relates to officer activity on the ground (discussed in Chapter 6). In both Camden and Tucson, incidents labeled as violent were more likely to result in arrest than those labeled nonviolent. However, incidents categorized as nonviolent were more likely to result in arrest when initiated by police than when originating from a 911 call, revealing a divergence that suggests the need for additional research. To a large extent, mental health and medical incidents were diverted from criminal justice enforcement, potentially indicating that the focus on mental health awareness has the potential to pay dividends. Vera’s analysis also revealed the potential for gathering additional data in the 911 call context to advance broader insights, such as how to improve call-taker and dispatcher operations to support improvements in criminal justice outcomes and the integration of additional variables to permit more varied and appropriate responses to 911 calls. The research also sought to test the viability of data science methods known as Natural Language Processing (NLP) in order to understand if data contained within CAD narrative fields (which makes up much of the CAD data) appears frequently enough to merit developing mechanisms to capture and analyze this information in a more structured manner (e.g., to develop new structured CAD codes). Several key findings emerged from applying the NLP approach, methods, and techniques to Camden and Tucson’s 911 data (described in Chapter 7). The high-level results include the following: • The narrative fields in the CAD entries are essential to making accurate policing decisions. • Subjective bias can be injected into the narrative fields by call-takers, dispatchers, and officers. • Additional research is needed to understand why this detectable difference between the narrative field and the structured data exists; how call-takers, dispatchers, and officers use the narrative field; and how much cognitive load is placed on officers when consuming the narrative data as opposed to the structured data. This inquiry would require researchers to review the data manually and identify another method to compare structured and unstructured data fields prior to employing a computational/algorithmic approach. To further explore the empirical findings that resulted from the research activities, Vera hosted a national convening of law enforcement leaders and system stakeholders (summarized in Chapter 8). At the convening, researchers presented their findings, explored alternatives to enforcement, and collaborated to identify opportunities for reform. This convening was held in partnership with Arnold Ventures and George Mason University’s Center for Evidence-Based Crime Policy (CEBCP). Both research teams (Vera and CEBCP) presented their findings to explore implications of the research and spark innovations, particularly around alternatives to enforcement. Participants from 40 organizations across the country were in attendance, including representatives from 10 police departments, five public safety communications agencies, and 10 research organizations. The room was full and engaged. The convening’s energy and insights provided clear evidence that additional conversation and collaboration on the topic is both needed and wanted. This report concludes with a number of key policy recommendations and practitioner innovations (presented in Chapter 9), ranging from developing new protocols for how and if police departments should respond to unverified burglar alarms to providing de-escalation tactics trainings to 911 call-takers and dispatchers. Clear needs have emerged for better call taking and recording practices, as well as standardized codes and procedures, with the goal of improving procedural justice, customer service, and the safety and wellbeing of officers, community members, call-takers, and dispatchers. Many practical solutions exist, some of which are currently being implemented and tested and others that are on the cutting edge. One effort that is feasible and valuable in the immediate term is developing a national coalition to advance thinking, practice, research, and standardization. This can be achieved through the roundtable model that has successfully mobilized reform in many other areas of the justice system for the past several decades.4 Alternatives to police response and collaborative community responses have shown great promise for integration into 911 call processing. Additional investments in this research and practice can help inform taking them to scale in local jurisdictions nationally. Many opportunities exist, and needs abound—this research makes clear that the 911 system is both massive and neglected. Though much was accomplished through the course of this current research effort, in most places the 911 call-taking, dispatching, and police response continuum continues to operate as a ‘black box,’ and there is a pressing need for further investment and research. Myriad opportunities exist to further develop this work, including continued and expanded analysis of the data already in hand. Other opportunities to expand the national conversation with roundtables about national standards, best practices, and building coalitions for understanding practice and moving it forward present an immediate first step in continuing to meaningfully advance this work. The goal of this and future work is to enhance public safety, promote meaningful alternatives to 911, and eliminate unnecessary police response and enforcement.   

New York: Vera Institute of Justice, 2020. 316[.