Can Acoustic Gunshot Detection Technology Reduce Crime? Meta-Analysis Weighs Evidence

Across hundreds of American cities, networks of microphones perch on rooftops and utility poles, listening around the clock for the acoustic signature of gunfire. The premise is seductively simple: detect a shooting within seconds, triangulate its exact location, and send officers racing to the scene before the shooter escapes — and, over time, drive gun violence down through sheer technological vigilance. Now the most comprehensive assessment of the technology to date has delivered a sobering verdict. In a systematic review and meta-analysis published in the American Journal of Criminal Justice, criminologists Jessica Huff, Brenna Dunlap and Roderick L. Pearson Jr. synthesized the available quantitative literature on acoustic gunshot detection technology (GDT) and found that, deployed on its own, it does not reduce crime. Pooling 44 effect sizes from eight studies that reported sufficient data, the team calculated a relative incident rate ratio (RIRR) of 1.02, with a 95 percent confidence interval spanning 0.90 to 1.16 — a range that includes the value of 1.0, meaning no effect whatsoever. If anything, the point estimate leaned toward a non-significant 2.36 percent increase in crime.

The technology, marketed most prominently under the ShotSpotter brand, works by exploiting physics that gunfire produces involuntarily. When a bullet leaves a gun barrel, it generates a muzzle blast wave — a sharp, distinctive pressure pulse that propagates outward at the speed of sound. Arrays of acoustic sensors deployed over a coverage area register these pulses, and software identifies the shot’s geographic origin by triangulating the precise activation times of individual sensors. Live-fire validation suggests the core engineering largely works as intended: an early test in Redwood City, California, recorded an 81 percent gunshot detection rate with locations accurate to within 41 feet. More recent field comparisons have been kinder to the sensors than to the public. In Kansas City, Missouri, and Durham, North Carolina, sensor-derived locations were consistently more precise than those provided by 911 callers, who frequently reported shots on entirely different street segments or offered only vague addresses. For departments plagued by chronic underreporting — a survey in Portland, Oregon, found that only 11.5 percent of residents who heard gunshots called 911 — that precision is precisely the appeal.

But precision is not infallibility, and the review devotes sustained attention to the twin failure modes of any detection system: false positives and false negatives. Field researchers in Virginia discovered that sensors activated for 50 percent of bottle rockets fired in Newport News and for 75 percent of bottle rockets and 35 percent of firecrackers set off in Hampton; in Dallas, officers responding to an alert found teenagers igniting firecrackers directly beneath a sensor. In St. Louis, 5 percent of activations were ultimately attributed to fireworks. False negatives are harder to catch but arguably costlier, because a missed shot is a shooter who walks away. In live-fire trials, the system failed to activate for 76 percent of 9-millimeter and .45-caliber test shots in Newport News and 66 percent of .22-caliber rounds in Hampton. An audit in New York City found the system missed between 7 and 25 confirmed shootings per month in Manhattan, while in Las Vegas it missed 10.5 percent of confirmed shootings. The sensors also detect only outdoor shots, and their sensitivity fluctuates with season, time of day, weather, background noise and local topography.

To organize what the technology can and cannot accomplish, the authors constructed a four-stage logic model tracing how GDT could theoretically reduce gun violence. Stage 1 is detection and rapid response: a shot is fired, sensors activate, and officers speed to an accurate location. Stage 2 is incapacitation: officers either catch the shooter on scene or harvest evidence — spent casings, forensic ballistic matches, firearm traces — that leads later to arrest, prosecution and conviction. Stage 3 is deterrence: visible police presence and public announcements about the system raise a would-be shooter’s perceived risk of being caught. Only in Stage 4 does crime actually fall. Each link in that chain, the evidence shows, is weaker than assumed. Just 3 percent of activations in Las Vegas involved a suspect still on scene when officers arrived, making immediate capture statistically marginal. Deterrence, meanwhile, is undermined when cities keep sensor locations secret, as Camden, New Jersey, has done — a strategy of dubious value given the small, inconspicuous sensors. A survey in Chicago found that just 40 percent of residents knew the city even used the technology.

The intermediate steps of the model fare better, but unevenly. Gunshot detection dramatically reshapes police workload: dispatches rose 287 percent in Dallas, gunshot-related calls for service climbed 259 percent in Philadelphia, and shooting calls more than doubled in Durham, Denver, Milwaukee and Richmond, California. Response times do improve — Winston-Salem officers were dispatched a median of two minutes faster to sensor alerts than to citizen calls, and Kansas City officers arrived 93 seconds sooner — but speed carries opportunity costs elsewhere. In Chicago, sensor deployment delayed dispatches to citizen-reported life-threatening emergencies by one minute and officer arrivals by two. Evidence collection shows genuine gains: gun recoveries rose 11.2 percent in Chicago’s covered areas, the share of Phoenix shootings with recovered casings jumped from 8.6 to 25.2 percent, and in Tampa a quarter of alerts produced entries in the national ballistic-matching network, half of which yielded investigative hits. Arrests, however, remain stubbornly rare. Only about 3 percent of alerts led to arrests in Las Vegas, Phoenix and Winston-Salem; in St. Louis, just one alert out of 890 produced an arrest; in Omaha, the odds of arrest for sensor-initiated calls were 20 percent lower than for citizen-reported gunfire. Studies in Chicago, Kansas City and Hartford found no improvement in case clearance.

For the meta-analysis itself, the team followed the PRISMA framework for systematic reviews. Consulting a university librarian, they searched eight databases for the terms ‘acoustic gunshot detection’ and ‘ShotSpotter’ in January 2025, then mined reference lists and federal award records from the Bureau of Justice Assistance and the National Institute of Justice. The searches surfaced 630 studies, plus 75 more from references and funding lists; after deduplication, 397 records were screened on title and abstract, 88 underwent full-text review, and just 14 studies met the inclusion criteria: quantitative analyses of crime outcomes with the detection system itself as the intervention. The eligible studies spanned 13 communities between 2008 and 2024 — seven government final reports, five peer-reviewed publications, one book chapter and one doctoral dissertation — and were graded on the Maryland Scientific Methods Scale, with seven scoring 3, five scoring 4 thanks to statistical matching or synthetic controls, and two scoring 2. Eight studies reported enough information to compute relative incident rate ratios, a measure the authors argue captures place-based treatment effects more faithfully than Cohen’s d. The 44 resulting estimates were pooled in a multilevel random-effects model with restricted maximum-likelihood estimation, homogeneity testing and cluster-robust variance checks.

The headline result was stark. The pooled relative incident rate ratio of 1.02 (standard error 0.07, p = 0.72) means gunshot detection technology was associated with no statistically significant change in crime of any type. Multilevel meta-regression moderator analyses found that neither crime type nor methodological rigor shaped the outcome: the technology did not perform better for shootings than for other offenses, and the strongest studies were no more flattering than the weakest. What did emerge was substantial between-study heterogeneity, flagged by a statistically significant Q statistic of 143.85 — evidence that some jurisdictions saw apparent declines while others saw increases, likely reflecting inconsistent implementation rather than any hidden average effect. Even within single cities, results contradicted themselves. A Miami-Dade County evaluation that applied four modeling approaches and three spatial estimation methods across two time periods produced scattered statistically significant increases amid predominantly null findings, while a Chicago analysis tracking eleven phases of expanding coverage found both significant decreases and significant increases in individual phases even though the aggregate effect was null. Across designs, cities and crime categories, insignificance dominated.

The findings land amid genuine policy turbulence. Chicago spent $49 million on ShotSpotter between 2018 and 2024 before terminating its contract over privacy concerns, fears of racial profiling and the thin evidence of benefit — and is now weighing a return. Federal agencies have funneled more than $20 million into implementing and evaluating the technology, which has spread to hundreds of cities since the 1990s. Part of the gap between hope and evidence lies in what the sensors actually produce. In Washington, D.C., and Oakland, California, only 12.4 percent of sensor-identified gunshots coincided with a 911 call or crime report. Confirmation rates varied wildly: 80 percent of alerts were verified as true shootings in Las Vegas, versus just 9 percent in Chicago, and shots detected in Kansas City’s covered area were 15 percent more likely to be dismissed as unfounded than those in matched comparison areas. In St. Louis, every 100 sensor activations yielded 0.9 founded gun crimes, against 7.6 per 100 citizen calls — meaning alerts often carried less actionable information than the 911 calls they were meant to supplement.

The authors’ prescription is not to rip out the microphones but to stop expecting them to fight crime alone. They advocate a strategic crime control approach in which problem analysis, evidence-based responses, partnerships, strategic planning and built-in accountability mechanisms surround the technology — treating sensor alerts as raw intelligence to be channeled into forensic systems such as the National Integrated Ballistic Information Network rather than as a self-executing solution. There are hints this can work: when Chicago integrated the sensors with cameras and crime databases in its Strategic Decision Support Centers, the odds of clearing nonfatal shootings rose by 19 to 24 percent. But integration is not automatic. Officers in Tampa admitted “cheating the system” by logging firecrackers without searching for evidence, and crushing call volumes in Kansas City left officers unable to conduct thorough initial canvasses. The deeper lesson, the researchers conclude, is that achieving incapacitation or deterrence requires comprehensively weaving detection systems into organizational practice — and that any city weighing a multimillion-dollar contract should do so against an evidence base that, after fourteen studies and nearly three decades of deployment, still shows no reliable reduction in crime. The sensors hear every shot. Whether anyone is caught for it remains, overwhelmingly, up to the humans.