Crashes fell 21% after ending mandatory hook turns in Tainan. What does the study actually prove?

“Crashes fell 21% after two-stage left turns were removed.” Read in isolation, that sentence invites two opposite reactions: the rule has finally been disproved, or the statistic was selected to support a policy.

The 21% is not a poll or a simple before-and-after count. It comes from a 2026 paper in Transportation Research Part A that treats Tainan’s staggered reform as a quasi-natural experiment. Before interpreting the result, however, we need to say precisely what was removed.

Tainan removed a mandate, not the option to make a hook turn

On selected one-way, two-lane roads, Tainan allowed scooters into the inside lane and ended the requirement to make a two-stage left turn. Existing waiting boxes were generally retained. A rider could make a direct left turn or choose the two-stage route according to conditions.

Tainan’s 2026 policy report says 52 roads in ten districts were opened in 2024, followed by 27 roads across the remaining fifteen districts in 2025. The scope was the city’s one-way, two-lane roads. The conclusion cannot simply be extended to three-lane arterials, express roads, complex junctions, or roads that still prohibit scooters from the inside lane.

The study found more than one declining number

The improvement was concentrated in injury crashes; fatal incidents did not show the same clear reduction. The study therefore supports a fall in overall conflict and injury risk, not a claim that ending hook turns prevents every severe crash.

Why this is stronger than a basic before-and-after comparison

Comparing one year before and one year after would leave traffic volume, weather, enforcement, construction, and pandemic effects mixed into the result. The paper uses the phased introduction and policy boundary in a difference-in-discontinuity design, comparing affected roads with similar roads that did not change at the same time.

That design is more credible for causal inference than “this year versus last year,” but a city is still not a laboratory. The result best answers what happened on this group of Tainan roads during this reform. It does not answer every intersection in Taiwan in advance.

Why might crashes decline instead of rise?

Two-stage turns are intended to keep scooters from crossing several traffic streams at once. They can also concentrate scooters at a corner, where right-turning cars, through traffic, and riders entering a waiting box interact. When the inside lane becomes legal and the left-turn path becomes continuous, some abrupt lane changes and crossflows may disappear.

That mechanism is consistent with the result, not proof that every junction behaves the same way. A high-speed road with poor visibility, heavy vehicles, no left-turn lane, or no protected phase presents a different risk. Road design cannot be replaced by either one national rule or a rider’s improvisation.

Why Taiwan’s transport ministry did not announce a nationwide repeal

Responding in June 2026, the Ministry of Transportation and Communications neither rejected the Tainan finding nor endorsed immediate nationwide removal. It called for reviewing speed limits, dedicated left-turn lanes, and protected signal phases, then opening intersections case by case. Complex or high-speed junctions may retain two-stage turns.

That answer is less satisfying than choosing a side, but it matches the reach of the evidence. The next useful evidence will come from other cities and other road geometries: do they show the same change in crash types after reform?

What should a rider do today?

Taiwan has not ended two-stage left turns nationwide. Where signs or lane markings prohibit scooters from the inside lane, or an intersection instructs riders to make a two-stage turn, the requirement still applies. Local lists can change, so a remembered rule or a short video is not enough.

Use the intersection in front of you: check signs and markings first, then whether a direct left is permitted. On an unfamiliar road, do not cut across several lanes at the last moment. A waiting box that remains in place is still an option; giving riders a choice is useful precisely because the safer path can depend on current conditions.

What the study really changes is the burden of proof

Mandatory hook turns have often been treated as safe by default, leaving reformers to prove that removal cannot cause harm. Tainan’s evidence breaks that one-way assumption: the mandate itself may create conflicts, and longevity is not a substitute for evaluation.

Turning 21% into “remove the rule everywhere now” would make the opposite error. A better policy publishes the road geometry, traffic exposure, and crash outcomes for each reform, then adjusts rules to measured risk. This study does not settle every intersection. It does require every rule to show its results.