Readback

Does it actually reduce crime — or move it?

Every randomized trial of patrol ALPR finds the same thing: more plates checked, more stolen cars recovered — and no measurable drop in crime. The strongest evidence for fixed networks is an 11% auto-theft reduction. Everything bigger is marketing.

Every study and audit on this page is American 🇺🇸. That isn't an editorial choice — it's the finding. No Canadian police service has published a peer-reviewed evaluation of its ALPR programme or a hit-rate audit, and there is no Canadian equivalent of California's State Auditor review. So the only rigorous evidence about a technology now deployed across Canada comes from another country's oversight bodies. Canadian entries are flagged 🇨🇦 wherever they exist — on this page, the only Canadian data point is an absence.
StudyFinding
🇺🇸Lum et al. (2011) — RCT, 30 hot spots, Virginia No significant effect on any crime outcome no deterrence
🇺🇸Taylor, Koper & Woods (2012) — RCT, Mesa AZ 3× more stolen-vehicle hits, 2× recoveries — no decline in auto theft rates no deterrence
🇺🇸Koper, Lum et al. (2021) — 785 patrols, 33 hot spots More recoveries, no effect on likelihood/timing/seriousness of next crime no deterrence
🇺🇸RAND (2014) — literature review Anecdotes only; "lack of rigorous outcome evidence" no deterrence
🇺🇸Mourtgos & Adams (2025) — 216 agencies, fixed Flock networks Motor vehicle theft down 11% after deployment (quasi-experimental) modest effect
🇺🇸Shjarback & Sarkos (2025) — Atlantic City fixed network No reduction in violent crime; reductions in shootings, auto theft, property crime modest effect

Displacement: does crime just move?

The 40-year CCTV literature (Piza et al. 2019, 80 studies) — the closest proxy for fixed ALPR — finds a modest overall decrease driven almost entirely by vehicle crime in car parks. Violent crime doesn't respond. Displacement to nearby areas is rarer than feared, but so is city-wide benefit: effects stay local. When York Region's deputy chief cites a 6% crime reduction within 1 km of cameras, the honest question is what happened at 2 km — and no Canadian service publishes that analysis.

Flock's "reduces crime by up to 70%" came from San Marino, CA — whose own police chief later said the figure isn't accurate. Forbes found the company handpicked and oversimplified data in four flagship cities. The academic who oversaw Flock's "10% of US crime" study later said the data was "too varied and incomplete… to do any type of meaningful statistical analysis."

Hit rates: what a billion scans actually catch

Audited jurisdictionHit rate
🇺🇸LAPD (CA State Auditor)~0.1% — 99.9% of stored images matched nothing
🇺🇸Central Marin, CA0.02% of ~4M plates
🇺🇸Piedmont, CA0.0013% — 2M scans, 26 leads
🇺🇸High Point, NC0.08%
🇺🇸Rhinebeck, NY0.01%
🇺🇸LAPD Inspector General hotlist audit1 in 3 alerts were FALSE (stale hotlist data)

Toronto has never published a hit-rate audit. At 1.25M scans/day capacity and typical hit rates, roughly 999 of every 1,000 records the system keeps for 7 days belong to drivers suspected of nothing.

When it's wrong

At least 27 documented wrongful stops in the US since 2018 — including the Gilliam family in Aurora, CO: four Black children held at gunpoint because an ALPR matched their SUV's plate to a stolen motorcycle from another state ($1.9M settlement). Vallejo, CA trials found 37% of stationary ALPR hits were misreads. These are system failures — stale hotlists, no plate-state verification — not one vendor's bug.

Sources

Every claim on this page, cited — expand for the full list