What We’ve Learned from Analyzing Data to Reimagine 911

We’re taking on the challenging task of analyzing data from our country’s 911 system so that communities and cities are empowered to advocate for change
  • Project Manager, Reimagine 911 & Projects Branch ,
    Code for America
  • Principal Solutions Engineer, Reimagine911,
    Code for America
a person on the phone typing

Improving our public safety must start at the root of all emergency response: the 911 system. We have to fundamentally understand it before we can reimagine it—but studying this system isn’t as easy as it seems. When we started this work during the 2021 National Day of Civic Hacking, it looked deceptively approachable. In around four hours, volunteers parsed through available data from 327 Public Safety Answering Points (PSAPs, also known as 911 call centers). Crowdsourcing allowed us to analyze around 5% of the country’s 911 data in a single day—more work than one researcher could do in a year!

But we quickly learned that not all open data is created equal—and we weren’t seeing the full picture. Hundreds of cities and countries make some (but importantly, not all) of their 911 call data publicly available. There aren’t any guidelines for what data should be made publicly available, and cities have very different ways of classifying, sorting, and publishing their data.

So why not focus on just one city? Put simply, if we can show a city that one of their peers has found a better path, that’s a powerful lever for change. Everyone who is involved in the 911 system that we’ve consulted agrees that police are often not the most appropriate responders.. But altering the system to redirect to non-law enforcement responders is a heavy lift for any agency. The Reimagine 911 project is focused on helping cities develop more emergency response alternatives by identifying and understanding opportunities visible in the data. We believe that if we can provide standardized data that allows for cross-jurisdictional comparison, we can provide a new level of transparency that cities and community groups alike can use to advocate for change.

If we can show a city that one of their peers has found a better path, that’s a powerful lever for change.

But first—why is the 911 system so difficult to understand?

The 911 system was created in 1968 as a response to the civil rights protests. While it has evolved into a critical emergency response network, administrative authority still resides with local law enforcement agencies. 911 is so fragmented that there are 5,748 PSAPs—and no guarantee that any two dispatch centers use the same technology or methods, or that their systems can work together. 

The first and greatest challenge facing Reimagine911—and all researchers—is that 911 data is inconsistent across jurisdictions. So, project members are standardizing 911 data to give researchers, administrators, and advocates an apples-to-apples view of the different ways cities  operate their 911 systems.

Reimagining 911 is challenging work

We began by leveraging the power of crowdsourcing to perform an online search for open 911 data in 382 cities. We prioritized the nation’s 100 largest cities and all cities with Code for America Brigades. To ensure a modest level of geographic distribution, we also included capital cities from every state. In the end, 128 of these cities had 911 data publicly available. Since we’re not 911 experts, we’ve educated ourselves through conversations with our partner Transform911 and many interviews with city agencies, police departments, community advocates, consultancies, and national standards bodies, all of which are reflected in this Blueprint for Change.

One of our greatest realizations has been that call types are a foundational—and flawed—way to analyze the 911 system. When a 911 call is received, a “call type” is assigned to the record. All subsequent actions and protocols are organized around these call types. But call types are more subjective than you might imagine. A 911 operator usually works a 12-hour shift of back-to-back judgment calls. How much can a researcher responsibly extrapolate from a 60-second call between two strangers (a 911 caller, and a 911 operator)? And how much more subjective does it become if/when the caller is calling on behalf of another person? 

While the call types that 911 operators assign may not always reflect the event, they do tell us what they understood the problem to be. This is critical because all downstream decisions can be traced back to how an agency chooses to respond to each perceived event. Since we want 911 call takers to have more choices for handling perceived events, the call types are still highly relevant despite being a subjective and sometimes incomplete reflection of the actual emergency.

Compounding this, “open” data doesn’t mean it’s easy to use. Open data is generally sanitized for public consumption, but there is no guidance or consistency for how this is done. For example, location information may be anonymized at many levels: by street address, street block, police beat, census block, or by dropping a decimal number from the latitude and longitude, meaning there is considerable nuance to understanding and using open 911 data. Different cities also share different types of data. Datasets rarely provide enough context to understand whether two cities are sharing the same data and describing it with different language, or actually sharing different datasets.

a graph of call types
A frequency distribution of the specific terms that different cities used to define records (or parts of records) in their dataset.

The questions we’re asking now

Can similar terms be grouped together? Not always. For example, take “Calls” and “Dispatched Calls.” While these seem very similar, Dispatched Calls are the subset of all Calls for which a 911 call taker actually dispatched a responder. Since the overall volume represented by these two types is inherently different, there are some types of analyses that cannot be made directly. 

Can cities using the same terms be compared? Again, not always. For example, some cities indicate whether an incident was reported by the public (via a 911 call) versus an incident where the response was initiated by law enforcement; other cities group these incidents together without indication (all under the same label of “call” records), making direct comparisons inappropriate.

Can we standardize these datasets? Not without significant efforts. Data format greatly impacts its usefulness. For example, 27 cities provide their data as web pages or portable document format (PDF) files, which makes it significantly harder to analyze. When cities provide application programming interface (API) access to their data, it opens up the possibility of collecting information programmatically, rather than manually. About two-thirds of cities had this kind of availability.

a graph of file types
A frequency distribution of the file formats that different cities used to publish their data.

Can we use all the data we’re collecting? It depends. Not every city’s “open data” comes with permission to use it. Of the cities we’ve reviewed, 52 (34%) have open licenses and 99 (66%) have no license or a closed license. We also found that 911 data sometimes have multiple licenses. 

Can we see historical trends? Yes, but it’s often variable and somewhat unreliable. It should be simple to determine how many years of data is available for a given city: open the dataset, sort by the event date, and record the oldest date. But a significant percentage of cities only display data for a fixed number of days—the last seven  days, for example. For these cities, the historical record could perhaps be created, but the onus is on the consumer to copy and store the open data before it is removed. This will generally require scraping software, and most likely will need to be configured separately for each city that provides data this way. 

a pie chart
A chart of the ways data is available: for a set number of days, continuining from a fixed date, or continuing from the present.

Why are we hopeful about this work at this moment?

Future historians might pick 2022 as the year 911 began to shift.. This July, the number 988 was dedicated to suicide prevention and behavioral health calls. If 988 is creating a window of opportunity for new policies and processes in the emergency response system, the next step is to identify 911 calls that should be diverted to other agencies. Funneling those calls to the right resources requires them to be standardized. 

What’s next for Reimagine 911?

We completed our exploratory phase of our Reimagine 911 work and are beginning a three-month prototyping phase known as an Impact Sprint. During this, we’ll produce a standardized set of open 911 data for 10 cities. This will mark the first time a standardized set of open 911 call data of this size has been available to review. We hope it will be a proof of concept demonstrating that more cities—hopefully many more cities—can also be included. As part of the Impact Sprint, we’ll also develop resources and a playbook for how Brigades and other community organizations can help their local 911 stakeholders improve the 911 data that they are sharing openly—because when we crowdsource tasks like these, we’re not only able to get communities involved in the local conversations about public safety, but also able to make a big impact across the country.

Do you want to get involved with our Reimagine911 work? We’re looking for people to help us categorize and standardize some tens of thousands of 911 call types. This cannot be automated, and the number of people needed to accomplish this has prevented other universities and organizations from doing this work in the past. This is one of the things that makes Code for America—and our vast volunteer network—well suited to tackle this particular problem. Fill out the volunteer interest form now.

Related stories

Using Data and Teamwork to Reimagine 911
July 21, 2022
Leading the Field: Rebecca Neusteter
August 30, 2021
Volunteer Spotlight: Reduce, Reuse, Record
August 10, 2022