ShotSpotter: Good or Bad?
Microphones that detect gunshots
Most gunfire is not reported to police. Some estimates say “only 20% of shootings are reported to law enforcement.”
Imagine two different scenarios. In the first, you overhear a loud crack that might be gunfire - but maybe it’s a car backfiring or fireworks or something else. You’re not sure. Are you going to call the police? In the second scenario, imagine living in a rough neighborhood where you heard gunfire yesterday, you heard gunfire today, and you expect to hear gunfire tomorrow. On top of that, you’re not a fan of the police, you don’t know exactly where the gunfire is coming from, and you don’t want the cops stopping by at your place. Are you going to call the police?
Enter: ShotSpotter.
What is ShotSpotter
ShotSpotter is a system of microphones that are deployed throughout a city to automatically detect gunfire and report it to police. If three or more microphones hear a sound they classify as a gunshot the system can triangulate1 the precise location of the gunshot.
When the system detects a likely gunshot a recording is given to a human analyst who will confirm that the sound is a gunshot. Within 60 seconds of the sound being detected the human analyst needs to complete their assessment and, if it’s a gunshot, report it to police.
ShotSpotter is deployed in 84 metro areas across 34 states. Feedback on the system and its efficacy is mixed.
Why might this be good?
In research Gunshot Detection Technology (GDT), like ShotSpotter, led to faster police and Emergency Medical Services response times. The linked paper looks at differences in response times for first responders responding to either a 911 call or an alert from ShotSpotter and finds that first responders arrive a median of 93 seconds faster for ShotSpotter alerts relative to 911 calls.
Partly the faster response can be attributed to automated systems reacting faster and with more confidence than humans. Where a human might pause to consider if they should call 911, or struggle to find their phone, the automated system will react instantly. Also, ShotSpotter can give police a precise geographic coordinate whereas a human might only offer a vague location where they heard the sound.
Intuitively, you might expect that faster response times would lead to more suspects apprehended and more lives saved. If the police arrive on scene after a shooting faster, it stands to reason they would be more likely to apprehend the shooter and/or save the victim.
However, there is not evidence for these assumptions. The paper I linked recounts a different study which found no difference in mortality rates for victims who were picked up as a result of a GDT alert versus a 911 call. As we will see later, in practice, many police departments using ShotSpotter do not seem to apprehend more shooters.
ShotSpotter’s website lists “lives saved” in 9 different cities - and the claims range from 13 saved in Philadelphia to 125 in Chicago. As I mentioned above, ShotSpotter is deployed in 84 different metro areas - how many lives are being saved in other cities? Your guess is as good as mine.
Why might this be bad?
Bad 1 - False Positives
A 2006 test of GDT found an accuracy of 99.6% across 234 shots in 23 locations. GDT like ShotSpotter very likely detects the vast majority of all gunfire.
In practice, despite the high accuracy, most ShotSpotter events do not result in any evidence that a gun related crime has taken place. A Chicago Inspector General report from 2021 found that of 41,830 ShotSpotter events where officers went to the location - officers only found evidence of a gun crime at 4,556 (11%) locations. Indicating most (89%) of the time no evidence of a gun crime is even found.
Even finding evidence of a gun crime is not necessarily a ShotSpotter success. The Chicago Inspector General’s report gives an archetypal example where officers respond to a ShotSpotter alert and observe a vehicle traveling at a high rate of speed. The officers stop the speeding vehicle and search it finding a gun that the driver is not permitted to have. This case counts as one of the 11% of ShotSpotter alerts resulting in evidence of a gun related crime, but it’s not certain, or even argued, that ShotSpotter detected that particular gun that was recovered being fired.
What I’m trying to describe is that a ShotSpotter alert summons police to an area and the police often stop the first suspicious person in that area and sometimes find evidence that person has committed a crime. If you replaced ShotSpotter with a system that randomly dispatched police to a location and they stopped the first suspicious person they saw on their way - how many crimes would they detect? Would it be better, worse, or about the same as ShotSpotter?
Bad 2 - Racism
A Wired article recently examined the locations of ShotSpotter microphones and determined that the populations being observed by ShotSpotter are disproportionately black (relative to total population). ShotSpotter defenders would say that ShotSpotter is deployed in areas with lots of violence and the demographic consequences are unintentional.
You could take this in two ways. First, if ShotSpotter is bad, then it is black people who will disproportionately suffer from it. Second, if ShotSpotter is good, then black people would disproportionately benefit. Or, maybe ShotSpotter is both.
Bad 3 - Cost Benefit
ShotSpotter isn’t free - it costs 65-90k per square mile of coverage. A report from Atlanta describes their decision to discontinue ShotSpotter.
At $70,000 per square mile; the system cost $280,000 in total (one of the five square miles was provided at no additional cost.) During the 2019 implementation, that created a cost ratio of $56,000 per gun recovered or $56,000 per arrest, since there were only five (5) arrests and five (5) guns recovered. This cost ratio illustrates a lack of operational efficiency since this ratio is on par with the cost of one sworn APD FTE.
Essentially, the system costs a lot of money and recovers relatively few guns. The Atlanta report argues that the money would be better spent paying the salaries of police officers.
There may be benefits to the ShotSpotter system, but if the benefits are not better than other available choices (e.g. hiring an additional officer) then ShotSpotter could still be bad relative to alternatives.
Bad 4 - Big Brother
Microphones can hear things other than gunshots. They could hear people talking for example. Do you want an unaccountable private company working closely with the police to be spying on the public from numerous undisclosed locations without oversight?
The idea that ShotSpotter microphones are going to be picking up conversations isn’t just worrying about some possible dystopian future it’s already happened. In 2011 ShotSpotter microphones recorded not only gunshots but also a conversation between the shooter and the victim where, helpfully, the shooter’s name was said. This evidence was used at trial to help convict the shooter.
ShotSpotter, after being rejected by the city of Toronto, sought to minimize the “Big Brother” concern by having a third party conduct a “Privacy audit” and they adopted the suggestions of the auditors to mitigate privacy concerns.
One example of a mitigation - before the audit the sensors recorded 72 hours of audio at a time. The auditors suggested that number be reduced to 30 and ShotSpotter agreed. This is good, because if the police decide they want an audio record of some spot, they need to ask for it within 30 hours - it’s some kind of limit. But, this is also bad because it’s an example of ShotSpotter unilaterally deciding to alter their data retention policy. ShotSpotter made an update and their system switched from storing 72 hours of records to storing 30 hours. Who is to say they won’t make another update and switch to storing audio for 100 hours? Or indefinitely?
Imagine combining microphones distributed across the city, ShotSpotter’s parent company also makes indoor acoustic monitoring sensors too by the way, with text to speech. ShotSpotter could offer police a searchable transcript of a covered area. That seems like something that could be abused.
Final Judgement
I wouldn’t want ShotSpotter in my community.
From everything I’ve read, I’m not persuaded there is a net benefit to using ShotSpotter. It’s involved with a relatively small number of arrests, but it also sends the police out for no purpose most of the time. If I had to guess, I’d guess it is slightly useful, but probably not worth the cost.
My biggest concern is the potential for misuse. If I trusted the police completely then I would say that the police chief should pick the tools to use - ShotSpotter included. Of course, I don’t trust the police and I’m concerned that they currently use ShotSpotter as an excuse to go to locations and search or arrest people they find suspicious. I’m concerned that this technology might develop further and be more widely deployed and become part of a surveillance state apparatus.
BUT my opinion might be different if I lived in a high crime area. If I lived in a high crime area where I was constantly afraid of being murdered or robbed then having the police randomly deploy to arrest suspicious people might seem like a pretty good deal. Assuming, that is, the police didn’t consider me suspicious.
If I were king, my decree would be that people should have to vote to adopt ShotSpotter - but only people in the affected communities should vote. That is, if you live in a neighborhood that is to receive, or not, ShotSpotter coverage, your vote on the matter should count. These are the people who will suffer from the increased surveillance and who might benefit from it too.
Maybe more accurate to say “pseudo-range multiliterate” rather than “triangulate” - but ShotSpotter documentation says “triangulate”.