What the evidence says
The strongest-design estimate on this page — a meta-analysis restricted to controlled before-after and controlled interrupted time series studies (Cohn 2020) — finds no measured effect of red-light cameras on total crashes. The pooled estimate is a 2% reduction, but the 95% confidence interval runs from a 9% reduction to a 7% increase, crossing zero, and heterogeneity across the underlying studies is high (I²=90.4%). The review itself calls the effect on total crashes "unclear."
Weaker-design sources disagree on direction. A review of high-income-country studies (Aeron-Thomas and Hess, via Fisa 2022) reports that red-light cameras decrease total casualty crashes, though no magnitude is given. An overview of before-after and time-series studies (Erke 2009, via Fisa 2022) reports the opposite direction for crashes overall: an increase of about 15% across all crash types combined. That same overview reports crash-type-specific effects moving in different directions — right-angle crashes decrease by about 10%, while rear-end crashes increase by about 40%. Retting (2003) separately reports that most reviewed studies found increases in rear-end crashes following camera installation, without a magnitude, consistent with Erke's rear-end figure.
Retting's review also reports that injury crashes overall (not total crashes) decrease 25-30% as a result of camera enforcement, with right-angle injury crashes decreasing at signalized intersections. That estimate mixes reductions in right-angle collisions against increases in rear-end collisions into a net figure — it measures injury severity, not total crash counts, so it does not resolve the total-crash question directly.
Data from fixed cameras in Colombia (Tavakkoli 2022) is mixed on total crashes: one analysis found a 57% increase in total crashes and injury crashes in the intervention group, while another found a 5-6% reduction. These cameras enforced multiple traffic violations rather than red-light running exclusively, which the source notes makes attribution to red-light enforcement specifically unclear.
Taken together, the evidence base for total crashes under matched or controlled designs specifically is thin: only Cohn's meta-analysis draws on controlled before-after and controlled interrupted time series designs, and it returns a wide, statistically insignificant estimate. Everything else on this page measuring total or overall crashes — the Aeron-Thomas review, the Erke overview, and the Colombia studies — uses before-after, time-series, or quasi-experimental designs the sources themselves flag as more susceptible to confounding and secular trends.
Where the evidence splits
Cohn's meta-analysis of controlled designs finds no measured effect on total crashes (2% reduction, confidence interval crossing zero), while Erke's overview of before-after and time-series studies finds a 15% increase in all crash types combined. The claim notes attribute part of this split to design: Erke's underlying studies are before-after and time-series, not matched or controlled, and are described as susceptible to confounding and secular trends, whereas Cohn's pooled estimate is drawn specifically from controlled before-after and controlled interrupted time series studies.
The Aeron-Thomas and Hess review reports a decrease in total casualty crashes, directly opposing Erke's reported 15% increase in overall crashes. The sources do not state a design difference between these two reviews that would explain the opposite directions — both are described as reviews/overviews of high-income-country studies — so this split is unexplained by the sources on this page.
The two Colombia studies disagree on both direction and magnitude for total crashes (a 57% increase versus a 5-6% reduction). The source attributes this partly to differing designs — a quasi-experimental difference-in-differences study versus an observational before-and-after study with a comparison group — and partly to the cameras enforcing multiple types of violations rather than red-light running exclusively, which the source says makes attribution to red-light enforcement unclear.