The premise of Reimaging Public Safety (RPS) is that police are asked to do too much and to perform tasks for which they are unsuited or untrained. To test these propositions, you might ask questions like: How do police spend their time? How many requests do they receive for different types of service? Is their response effective? Answers to these questions are important to the development of policy and the design of effective programs. And to answer them, it is quite natural — even essential — to turn to data.
Because every jurisdiction varies, we can’t know what data you have or help with all data-related challenges. But we can offer a useful model, with lessons that may apply to multiple data sets. In the following tutorial, we take a close look at Calls for Service (CFS), which are a useful, though incomplete, reflection of a police department’s role and actions in a community.
Calls for Service data (CFS), typically maintained by the police department, record every assignment to which an officer is dispatched, as well as other actions that an officer performs. Because police are the primary, all-purpose responders in most jurisdictions, these data will tell us a lot about the demands placed on local government and how government responds to those demands.
Entries in a CFS database may include: 911 calls and responses; instances in which other agencies (e.g., fire department, schools, etc.) call for police assistance; instances in which police officers identify a problem and initiate a response; and other activities that officers perform, such as writing reports, vehicle maintenance, or patrolling neighborhoods. In short, CFS data illustrate the spectrum of activities that officers perform and log.
Most CFS data are broken down into the categories that 911 dispatchers and call takers log in a Computer Aided Dispatch (CAD) system. CAD technology is used by 911 professionals to gather essential information during a 911 call, such as a caller’s location, the availability of nearby first responders, and prior calls made to specific locations. CAD categories are different for each jurisdiction, with some jurisdictions having hundreds of categories and some having thousands. The categories can be obvious (“person with gun”) or opaque (“citizen standby”).
To put it bluntly, CFS data categories are not always logical or understandable, and have often accreted over time without any comprehensive effort to establish an intelligible structure. With CFS data sets often relying on vague descriptors like “public threat,” “unknown trouble,” “overdue person,” “walkaway,” and “agency assist,” it’s no wonder that the resulting picture of police activities can be blurry at best, and nearly meaningless at worst. That’s why establishing clear, simple, user-friendly data categories has been one of our major priorities, and one of our most important recommendations for any jurisdiction considering alternative response.
To meet this challenge, we conducted conversations with community members, consulted with experts, explored publicly-available data, and then, at the conclusion of this process, developed a model with approximately 30 categories, expressed in everyday language and applicable to almost any police department.
Using this simplified model, we can then go one-by-one through almost any jurisdiction’s existing codes and re-group them. For example, one jurisdiction had nearly 50 different categories that we categorized as "traffic enforcement.” These included several geographic-specific traffic details, as well as calls coded as truck inspection, off road vehicles, child restraint, and drunk driver, among others. Under our “burglar alarm” category we included duress alarm, silent alarm, panic alarm, verified alarm, ATM alarm, false alarm, and bank robbery alarm, among others. Of course, reasonable people can disagree about the right grouping and category names. But, ultimately, our goal is to collapse the data into usable categories to gain a fuller understanding of why people call 911, how police respond, and the possibilities for alternative response.
These narrow distinctions among the existing categories may be useful for police departments and other responders — indeed, we recognize there may be an inherent tension between the needs of the agencies (e.g., operational needs, jurisdiction-specific processes) and those of policymakers. Aggregating them, however, allows policymakers to see the full volume of certain calls, such as mental health-related incidents. Understanding the order of magnitude of certain call types is essential to designing and standing up alternative response models.
In other instances, our insight into the data is limited by overly broad, opaque categories that lack more precise descriptions, such as:
What are these calls actually about? It is nearly impossible to discern without more detail. One can get that detail by reading the 911 call narratives, but that is very time-consuming and difficult work.
This graph shows all the calls made to 911 in a mid-sized United States city over a 4-year period. Note that this does not include activity that was initiated by police or other agencies, but just calls made to 911 by community members. For example, when looking at 911 calls only, “welfare checks” make up about 7% of calls. But when we include police activity that is initiated by non-911 calls (self-initiated by officer, the police department, or other agencies), then welfare checks only make up about 5% of total calls for service. Also notice that our pie chart does not match our categories perfectly — that is because each city has idiosyncrasies in its data coding that makes a one-to-one match difficult.
Take your data with a grain of salt. We'll go into more detail later, but, in short, there are real limits on what the data can show us. It may be incomplete or inaccurate. The categories may be confusing. Access to certain data fields may be restricted due to privacy. You may want information that's contained in multiple databases, which don't talk to each other. Data provides important insight into why people call for help and how police respond, but are only one component of the reimagining process.
Try applying our template to your CFS data. We bet that you will find the picture of local police activity coming into much clearer focus, along with opportunities for reform and innovation. For each type of service, you’ll likely see the number of calls received, timing and location of these calls, duration of the police response, and whether the incident was still happening when the police arrived. Depending on how complete or accurate the data are, you may also be able to learn things like how many officers responded and whether the incident led to criminal charges or an arrest. We look at specific examples further below.
For reasons we explain in a moment, many jurisdictions have found that alternative responder programs designed for mental health calls are also well-suited to address welfare checks, and vice versa. Suppose your jurisdiction was already committed to alternative response for mental health calls and wanted to consider whether this same program could also handle welfare checks. How would you proceed?
First, let’s define our terms and categories.
A welfare check call is a request that officers check on an individual who may be in some type of need, often indicated by not returning phone calls, missing appointments, unexplained noises, and/or erratic or otherwise concerning behavior. Calls for welfare checks are often made by concerned family, friends, and neighbors.
Mental health calls refer to calls about individuals suffering from a suspected mental health crisis, as well as the public health challenges, nuisances, drug use, and potential threats of physical violence that originate from the crisis.
In one sample city, we identified 13 categories that we would consider mental health calls:
And 13 categories that we will consider a type of welfare check:
Next, dive into the data to learn how responses are currently conducted, and to assess the potential workload for mental health alternative responders. This entails reviewing call volume, the time and day of week calls come in, average police response time, and amount of time that officers spend on the scene.
We begin by tabulating the number of potential calls involved.
From 2018 - 2021, there were approximately 18,000 mental health calls to 911 in our sample jurisdiction. (Note that the actual number of referrals to an alternative responder would likely be lower, since calls entailing the threat of violence or use of a weapon would continue to be assigned to police.)
Over the same period and using the same methodology, there were approximately 86,000 welfare check calls to 911 in the jurisdiction.
While these records of past call volumes cannot project future demands with precision, they are still instructive. We obtain good order of magnitude estimates and learn that there are many times more welfare check calls compared to mental health calls, meaning that welfare checks could take up a considerable share of a responder’s time.
Next, we use the data to understand what currently happens in response to these calls. How long do officers spend on the scene and how quickly do they arrive?
By looking at these two charts, we can see that officers spend on average nearly two more hours at mental health calls than welfare check calls, and they arrive on the scene of mental health calls over twice as quickly. Arrival time can be a function of how a call is prioritized in a given jurisdiction. On this jurisdiction's scale of 0 – 9, with 0 being the highest priority, the majority of mental health calls were coded as level 2 and the majority of welfare checks calls were coded as level 3.
When considering an alternative response program, understanding the distribution of calls across the day and week may be as important as understanding total call volume, and will help determine whether a 24/7 response capacity is necessary.
We can see from the data that mental health calls and welfare checks calls have similar patterns — both are at their lowest overnight, and then pick up in the afternoon. Both call types are fairly evenly distributed across days of the week.
We began with a hypothetical: your community has decided to create an alternative responder program for mental health calls and is trying to determine whether welfare checks calls could be included as well. The data from our sample city tell us that:
Depending on the available data, it also may be possible to determine whether a call resulted in criminal charges, arrests, or a use of force. These would be important data points when considering what share of calls could appropriately be referred to alternative response, and what share should remain in the hands of police officers. (To be clear, just because force was used in a particular situation, does not mean force had to be used – it could be that a different responder could have resolved the same situation without force.)
Of course, even if you can have the same responder handle two different call types, does that mean you should? This may require information beyond the CFS data. To help inform the decision, you may want further detail about what happens on these calls, what a successful call outcome looks like and how it’s achieved, and the skillset, training, and equipment that best meet the needs of the caller.
These data together provide an excellent, though not complete, basis for informed analysis of options and decision-making.
For all its value, data analysis also has important limitations. CFS data cannot answer some of the most fundamental questions about how we currently respond to mental health and welfare checks: What information did the caller relay to the call taker? How do we account for the individual decisions of call takers, including any bias that may impact decision-making about how to code certain calls? What exactly did the officer do on scene? Was the caller’s issue resolved successfully? Did anyone get hurt?
Data can inform, but not determine, the skill sets, qualifications, and job titles suitable for alternative responders, nor can data speak to the legal authority responders may need to address mental health calls and welfare checks. Posing such questions helps us to identify data gaps, serves as a check against overconfidence, and allows us to consider what additional data could be collected once an alternative responder program is launched.
Ultimately, any successful program will require direct engagement with stakeholders both within and outside of government, and honest conversation about community values and goals. Good data can be a critical part of this larger process, helping to ensure that decisions and expectations are grounded in reality, and that outcomes are measured accurately.