Report 119 · Luxury Home Security
For fifteen years the honest answer was that nobody knew, because the only rigorous experiments tested a camera bolted to a patrol car. Two weeks ago that changed: a study of 216 agencies found vehicle theft fell 11 percent after fixed plate readers went live. It is the best evidence this technology has ever had. It also does not measure the thing a homeowners association is being sold, and the authors say so themselves.
The pitch arrives at the HOA meeting looking like a solved problem. Four solar-powered cameras on the entrance roads, a subscription, and every plate that enters the neighborhood gets read, timestamped, and checked against a hot list. Somebody says the word "deterrent." Somebody else says the technology "solves crimes." A vote gets taken.
I want to walk through what is actually known about that, because the evidence base moved significantly this month and almost nobody has noticed. It moved in the direction of the cameras working. It also, read carefully, moved in a direction that should make a private neighborhood more careful rather than less.
The number that gets quoted against plate readers comes from a 2010 report by the Center for Evidence-Based Crime Policy at George Mason University, written by Cynthia Lum, Linda Merola, Julie Willis and Breanne Cave, funded by the National Institute of Justice and the Navy's SPAWAR Atlantic. They ran a two-jurisdiction randomized controlled experiment in Alexandria and Fairfax County, Virginia. Their finding:
Our randomized controlled experiment mirrored the findings from the PERF experiments in that the use of LPR in autotheft hot spots does not appear to result in a reduction of crime generally or autotheft specifically, during the period of time measured.
That report also contains the single best sentence anyone has written about security technology, and it is a section heading: Efficiency Does Not Equal Effectiveness. Their argument is worth quoting because it generalizes far beyond plate readers:
Especially with law enforcement technologies, efficiency is often mistakenly interpreted as effectiveness, which can perpetuate a false sense of security and a mythology that crime prevention or progress is occurring.
A plate reader is unambiguously more efficient than an officer squinting at bumpers. It scans hundreds of plates a minute, with less discretion and therefore less room for bias, which the GMU team credited as a genuine advantage. Efficiency was never the question. Whether that efficiency turns into fewer crimes is a separate question, and it does not follow automatically.
Here is the part that everyone citing that 2010 study gets wrong, though. It tested readers mounted on patrol cars, driven through auto-theft hot spots for limited hours. That is not what is being installed at your gate. A patrol reader covers the places an officer happens to be. A fixed camera on an arterial road runs continuously and builds a searchable archive. Those are different interventions, and evidence about one is not evidence about the other.
Before the new study, the loudest numbers in this market came from the market. Flock Safety, the dominant supplier of fixed neighborhood cameras, publishes an analysis claiming that 10 percent of reported crime in the United States is solved with Flock, that 700,000 crimes a year are solved using its cameras, and a 9.10 percent increase in clearance rate per device per officer.
Read the method note. Those figures come from surveying Flock's own customers between April and June 2023, running regressions against FBI crime data, and then, in the company's words, "applying this framework to all of Flock's customers and estimating the totals." The customer sample was 123 respondents.
I am not calling that fraudulent. I am saying it is a vendor comparing agencies that bought the product against agencies that did not, which cannot separate the camera's effect from whatever made a department decide to buy cameras that year. Departments with rising vehicle theft and a new grant behave differently in a hundred ways at once. The academic literature has a blunt name for this class of evidence: "vendor-sponsored correlations that do not address selection into adoption."
On 16 August 2026, Scott M. Mourtgos and Ian T. Adams of the University of South Carolina posted a working paper that does the thing nobody had done: estimate effects across many agencies, using the staggered timing of adoption as the source of identification.
They took the first fixed-camera deployment date for 216 U.S. law enforcement agencies, matched it to monthly National Incident-Based Reporting System data from 2017 through 2023, and ran a Sun-Abraham interaction-weighted event study, which is the current standard tool for handling staggered rollouts where treatment effects differ across adopters. Roughly 184,000 agency-months in the incidence models.
The headline results:
Motor vehicle theft fell 11.0 percent after deployment (95 percent confidence interval −17.3 to −4.2, p = .002). Because a general crime decline would produce the same number, they also measured vehicle theft relative to the same agencies' other property crime, which fell 5.2 percent, while other property crime itself changed by −6.1 percent with a confidence interval that crosses zero. The decline is specific to vehicle theft, not a rising tide. It persists through twelve months, which argues against it being a transitory correction after the crime spike that triggered the purchase.
Arrest clearance for vehicle theft rose 15.9 percent (+7.3 to +25.0, p < .001). This is the number a vendor would put on a billboard, and it is the number the authors themselves trust least. Clearance starts rising in the three months before the recorded deployment date. Their handling of that is a model of how to report an inconvenient result:
The increase could reflect preparatory work, date error, pilot cameras, or a broader anti-theft initiative that included camera purchase. Under the first three accounts, the effect is mistimed; under the last, it is partly confounded by selection. The data cannot distinguish these accounts.
Under a robustness method designed to test exactly this, the clearance effect's confidence interval includes zero. The theft-reduction result survives that same test; the clearance result does not.
Vehicle recovery barely moved. Among vehicles recorded as recovered, the median lag fell about 0.28 days, roughly seven hours. Across all stolen vehicles including those never recovered, nothing is distinguishable from zero: not recovery within one day, not within seven days, not restricted mean recovery time at 30 days. The authors explain why that makes mechanical sense. Most recovered vehicles are abandoned, crashed, or found by patrol or their owners within a few days anyway, and a plate-reader hit requires an active camera, a readable plate already on a hot list, and an officer free to respond.
Buried in the robustness checks is the finding I would put in front of any HOA board before the vote.
The primary estimate weights agencies by how much vehicle theft they had before deployment, so it answers the question "did theft fall where theft was concentrated?" When the authors instead weight by population, or give every agency-month equal weight, the estimates get smaller and, in their words, are "not distinguishable from zero." They are explicit about what that narrows the finding to: the decline "is concentrated in the jurisdictions carrying more of the MVT burden rather than representing a uniform effect across adopting agencies."
The average adopting jurisdiction did not show a detectable benefit. The aggregate burden fell because it fell in the places already carrying most of it.
A low-crime affluent subdivision is the opposite of that profile. Whatever this study establishes, it does not establish that a place with very little vehicle theft will see its very little vehicle theft go down.
And then the authors close the door on the neighborhood question outright:
We do not estimate how benefits or surveillance burdens vary with neighborhood camera coverage.
Every number in that paper describes a police agency's network, integrated with hot lists, with sworn officers who can act on an alert. The intervention was not four cameras owned by a homeowners association. I have found no rigorous study of that, and I am not going to manufacture one by analogy.
Strip away the marketing and a fixed plate reader does two distinct jobs.
Interception requires a hit while the car is still moving, on a list that already contains that plate, with someone available and willing to respond. Every link in that chain has to hold in real time.
Reconstruction requires only that the network recorded the vehicle. An investigator, later, searches historical reads. Nothing has to work in the moment.
Reconstruction is the job these cameras are genuinely good at, and it is the job that has nothing to do with your gate specifically. It is also the job a private camera does worst, because reconstruction has value in proportion to coverage, and a neighborhood entrance is one point on a map. The evidence pattern in the new study fits this exactly: effects show up in deterrence and identification, and not in the recovery of the average stolen car.
Notice what neither of those two jobs is. Neither one stops the burglary in progress at your house. A plate reader is an investigative sensor pointed at a road. It does not detect an intruder, it does not summon anyone to a door, and it does not care what happens on your property. That is not a criticism of the technology, it is a description of it. I have written about the actual evidence on whether gated communities are safer and about whether anyone comes when your alarm trips, and the plate reader does not touch either problem.
When an HOA installs these, the community becomes the operator of a surveillance system, which is a job with obligations most boards have not thought about.
The Electronic Frontier Foundation's Jason Kelley and Matthew Guariglia wrote about neighborhood plate-reader adoption in September 2020, and the detail that has stayed with me is not a policy argument, it is an HOA board member's own admission: asked whether the board had rules about who could search the system and how often, she said "it hadn't dawned on her that someone might use the system to track her neighbors' movements."
That is the whole governance problem in one sentence. A system recording every entry and exit reveals when your neighbor stopped leaving for work in the morning, when a car that is not their spouse's stays overnight, when the house is empty for eleven days. In a police department, access to that is at least nominally governed by policy, audit logs, and a chain of command. On an HOA board it may be governed by whoever has the password.
The Mourtgos and Adams paper does not dodge this either, even though effectiveness is what they set out to measure:
Effectiveness is not sufficient for policy. Fixed ALPR networks record the movements of every driver on covered roads, nearly all of whom are not suspected of an offense.
They note the databases have documented histories of misuse, and that the policy question turns on retention, audit logs, interagency sharing, and who is authorized to run a search. If your community installs cameras that share into a wider network, you have enrolled every resident and every visitor, including the ones who did not vote for it, into a system whose reach you do not control.
Not whether the cameras work. That question is too coarse to have an answer. These, in order:
What specifically are we expecting them to do? If the answer is "deter burglars from our houses," the honest reply is that no study supports it, including the good new one, which measured vehicle theft. If the answer is "give investigators a record after a vehicle crime," that is defensible and is what the evidence actually shows.
Who can search it, and is the search logged? If nobody can answer that, the system is not ready to be installed regardless of price.
How long are reads kept, and who else receives them? Retention and sharing are the two settings that convert a neighborhood tool into something else.
What happens to the alert at 3 a.m.? This is the same question I keep asking about every sensor. An alert nobody is contracted to act on is a log entry, not a security measure.
Disclosure, because I build in this category rather than only writing about it: I help design the AI security systems for a veteran-owned (SDVOSB) home-security company run by a fellow combat veteran and his father, who has three decades in the trade. I do not own that company and earn nothing from this link. Full policy here. I flag it here because this report argues against a product category that companies like theirs sell, which is roughly the opposite of a sales pitch.
The Mourtgos and Adams paper is a working paper, not a peer-reviewed publication. It was posted to CrimRxiv on 16 August 2026 and carries a working-paper date of 14 August 2026. I read the full text, and its methods and self-reported caveats are unusually careful, but it has not been through review and its numbers may change. Do not treat the 11.0 percent as settled.
Its deployment records cover a single vendor, Flock Safety. The authors flag the resulting contamination: comparison agencies may run other plate-reader systems, and treated agencies may have used them earlier, which would blur the contrast between groups. They also note adopting agencies are larger and more metropolitan than U.S. agencies generally, so the findings may not extend to small or rural jurisdictions.
I did not open the underlying NIBRS files or attempt to reproduce any estimate. The supplementary tables (S1 through S8) are referenced in the text I read; I did not inspect the tables themselves.
I did not open the Shjarback and Sarkos 2025 Atlantic City evaluation, which is peer-reviewed and directly on this topic. Its publisher's page returned an access error to me and its CrimRxiv copy returned a 403. I have cited it only as it is characterized inside the paper I did read, and I make no independent claim about its findings.
I have not verified the Flock Safety figures against any independent source. They are reproduced as the company's own claims, with the company's own stated method, and nothing more.
None of this rests on my own research. My published work is in microwave spectroscopy, and my background here is operational rather than criminological: I spent my Army career as a Counter-IED and Electronic Warfare Officer, which taught me a great deal about sensor networks and nothing about running a regression on crime data. The statistical claims belong to the cited authors.
The interesting thing about this month's study is not that it is good news for plate readers. It is how carefully its authors separated three results that a brochure would have merged into one.
Theft down, and the finding survives their toughest test. Clearance up, and they tell you it probably starts before the cameras did. Recovery, essentially unchanged, with an explanation of why that was always the likely outcome. Same technology, three different strengths of evidence, reported as three different strengths of evidence.
That is what an honest effectiveness claim looks like, and it is the standard to hold your vendor to. Ask which of your three questions their number answers. If the answer covers all three at once, it is a marketing number, and the 2010 warning applies: you are being sold efficiency and invited to hear effectiveness.
Scope note: this report summarizes what published evaluations measured about police-operated automated license plate reader networks. It is not an evaluation of any vendor's product, not a legal analysis of plate-reader use by private entities, and not advice on whether any particular community should install cameras. Laws on private surveillance, data retention, and sharing with law enforcement vary by state and locality; consult counsel for your own jurisdiction. The central quantitative source is a working paper that has not completed peer review, as stated in the body. Disclosure: the author helps design AI security systems for a veteran-owned home-security company, as stated in the body of this report, and does not own that company.
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.