The short answer
The strongest evidence supports targeted speed enforcement as one tool among many. The record does not justify treating every camera, every location, every fine structure and every revenue practice as equally effective.
This project does not claim that cameras never work. It challenges a common leap: because some automated enforcement can improve some safety outcomes, any scale of deployment and any financial design must be accepted without a vote.
The evidence is favorable, but the effect is not a blank check.
What is supported
Systematic reviews and location studies commonly report less speeding and reductions in some injury crashes near speed-camera sites.
What varies
Effects differ by road type, baseline crash history, camera visibility, enforcement threshold, follow-up period and whether nearby crashes are displaced.
Regression to the mean
Cameras are often placed where recent crashes were unusually high. Some decline may occur even without intervention, so stronger studies use comparison locations and longer periods.
Location-level accountability
A city should publish before-and-after speeds, crashes and severity for every site instead of relying only on a citywide total.
Starting sources: Cochrane review; Federal Highway Administration.
Crash type and severity matter.
Red light camera studies often report fewer right-angle crashes, which can be severe, alongside increases in rear-end crashes. A headline that says “crashes fell” or “crashes rose” can conceal that tradeoff. Signal timing, especially adequate yellow intervals, is a critical part of the analysis.
Starting source: Insurance Institute for Highway Safety overview.
A dramatic decline is not the same as a causal estimate.
DC traffic deaths declined from 52 in 2024 to 25 in 2025. That is good news. It is also a single annual comparison during a period with multiple policy, engineering and enforcement changes.
To attribute the decline to STEER or camera expansion, DC would need a design capable of separating camera effects from roadway changes, traffic exposure, police activity, vehicle technology, weather and normal year-to-year variation. The public should see location-level results and a transparent methodology.
If a location is safer, publish the before-and-after speeds, crashes, severity and exposure. If the data are convincing, the program should withstand both scrutiny and a public vote.
The Washington Post reported both the 2025 decline and that deaths remained near the level recorded a decade earlier: March 2026 analysis.
Questions people ask
Are cameras safer than police stops?
The concern behind this question is real. MPD's own stop data shows 73.2 percent of police stops involve Black residents, in a city that is roughly 44 percent Black. A roadside encounter with an armed officer carries risks a camera does not. We take that seriously and we do not dismiss it.
But the District never actually made that trade, and its own records show it. The substitution claim makes a prediction that can be checked: as camera enforcement grows, ticket-producing police stops should fall. They did not.
Ticket-producing stops across six years, indexed to 2020: 100, 87, 89, 89, 77, and 94 in 2025. There is no trend. In 2025, with camera citations at an all-time high, MPD made more ticket-producing stops than in 2022, 2023 or 2024.
The cleanest way to put it: between 2022 and 2025 the District added 1,560,079 camera citations, a 105 percent increase. Over the same period ticket-producing police stops changed by plus 2,269. The marginal 1.56 million camera citations displaced no police stop.
Nor was enforcement merely moved from officers to machines. Counting both, total traffic enforcement contact went from 1,530,891 in 2022 to 3,096,215 in 2025. The District did not reallocate enforcement. It doubled it. 2025 also had the most police stops of any year in the series.
There is a second problem, and it is the more serious one. A camera can only replace the kind of stop it can perform, a ticket for speeding or running a signal. It cannot perform an investigatory stop. And the investigatory stop is the category that grew. The investigatory share of MPD stops was about 41 percent from 2020 through 2023 and about 50 percent in 2024 and 2025. In MPD's data the two categories look very different: ticket-only stops are 62.3 percent Black, while non-ticket investigatory stops are 85.0 percent Black. The gap widens by district. In the Second District, where 47.1 percent of stops involve Black residents, 68.6 percent of stops are ticket-only. In the Seventh District, where 92.2 percent of stops involve Black residents, only 30.9 percent are. Across the seven districts the correlation between the share of stops involving Black residents and the share that are ticket-only is negative 0.84.
Read plainly, that means automated enforcement can only ever substitute for the least racially disparate category of stop, and cannot touch the most disparate one. The kind of stop a camera can replace is the kind that stayed flat. The kind it cannot replace is the kind that grew.
The consequence shows up in the overall figure. The Black share of all MPD stops ran 74.3, 72.0, 71.2, 70.2, 73.0 and 75.7 percent across 2020 to 2025. 2025 is the most racially disparate year of stops in the series.
And that is a composition effect rather than a within-category one. The Black share inside each stop type barely moved: ticket-only stayed in the low sixties, investigatory stayed near 85 percent. Overall disparity rose because the mix shifted toward the category that was already 85 percent Black.
Two limits we want to state ourselves. We do not claim cameras caused stops to rise or the mix to shift, and nothing here would support that. And the composition shift is abrupt, dating to October 2023, which is also the signature of a recording change rather than a change in conduct. We have asked MPD what changed and will not interpret that break until we know. The first two findings do not depend on it.
A fair defender can still argue stops would have grown faster without cameras. Citywide data cannot refute a counterfactual. The test that would settle it compares stops near the 108 cameras installed in 2024 against matched control corridors. We have not run it, because MPD publishes stop locations as block descriptions without coordinates. If it shows stops falling near new cameras, cameras did substitute for police contact, and we will publish that.
One limit we want to state ourselves: none of this shows that cameras caused stops to rise, and we do not claim it. What it shows is that the prediction the substitution argument makes did not happen. A stronger test would compare stops near the 108 cameras installed in 2024 against matched control corridors. We have not run it yet, because MPD publishes stop locations as block descriptions without coordinates. If that test shows stops falling near new cameras, cameras did substitute for police contact, and we will publish that result.
And the policy choice is not limited to cameras or police stops. Street design, signal timing, narrower lanes and raised crossings reduce dangerous behavior without either one. Georgia added one second to yellow intervals and violations fell roughly 80 percent. That removes the ticket and the stop.
Sources: MPD Stop Data, 229,064 records covering 2023 to 2025, and 279,936 records covering July 2019 to 2022; DC monthly moving violation files, all twelve months of each year. Both published by the District on Open Data DC and retrieved 25 August 2026 through unauthenticated public API calls that anyone can repeat. One caution for anyone re-running this: MPD's stop file begins 22 July 2019, so 2019 contains only 5.4 months and must not be used as an annual baseline. Our full analysis, including every limit of these comparisons, is available on request.
Does questioning DC's statistics mean opposing safety?
No. Demanding a valid causal claim is part of taking safety seriously.
What does the Camera Tax Project support?
Engineering first, targeted and publicly approved enforcement, income-scaled fines, no poverty penalties, transparent site-level evaluation and safety-dedicated revenue.