"Just keep up with traffic": a 60-year-old myth on trial

In 1964, David Solomon of the US Bureau of Public Roads published a study of crashes on main rural highways in America. The scale was enormous for its day: the speeds of roughly 10,000 crash-involved vehicles were compared against roughly 29,000 control vehicles measured alongside the same roads at different times of day. Solomon grouped vehicles into 10 mph speed bands, worked out what share of crashes fell into each band relative to how common that band was in normal traffic, and compared everything to the modus speed of each road, meaning the most common speed of the traffic on it, which his team established by driving with the flow and sampling it.

The result, plotted, made one of the most famous shapes in road-safety research. Picture a valley. The floor of the valley, the lowest crash rate, sat not at the most common traffic speed but about 6 mph, roughly 10 km/h, above it. Climb the valley wall in either direction, towards vehicles travelling much faster than the flow or much slower, and the crash rate rose steeply, with the extreme deviations in both directions, beyond about 50 km/h from the most common speed, showing substantially higher crash rates. That is the U-curve, and read at face value it says something startling: the slow outliers looked as crash-prone as the fast ones, and the safest place to sit looked to be slightly above the speed of the traffic around you.

Four years later, Julie Cirillo replicated the approach on American interstate roads and found the same shape again, this time with the valley floor about 12 mph, roughly 20 km/h, above the most common speed. One detail of her method deserves to travel with that finding: Cirillo analysed only crashes involving two or more vehicles moving in more or less the same direction, so single-vehicle crashes and opposing-traffic collisions were outside her data by design. Still, two large studies, one shape. By the early 1970s the U-curve was established wisdom, and the practical advice drawn from it, keep up with the flow and never mind the limit, has been repeated in cars, canteens and internet forums ever since.

Before we take the curve apart, credit where it is due. Solomon's study was a serious, pioneering piece of work, and the fast arm of his U has held up: every study in the record reviewed here, of every design, has found that vehicles travelling much faster than the traffic around them crash more. The question this guide is really about is the other arm. And the other arm has a problem that begins with a single methodological detail: Solomon's crash-involved speeds were not measured. They were estimated from notes in police crash reports, written after the event, often from statements and guesswork. Keep that in mind as we move to what happened when researchers went looking.

Exhibit A: the turning-vehicle artefact, 1970

In 1970 the US Research Triangle Institute, taking advantage of better measurement equipment, revisited the speed-and-crash question with 200 crash-involved vehicles. Instead of relying on police notes, the RTI reconstructed pre-crash speeds from physical evidence and the accounts of those involved, and measured control speeds with detection loops in the road. Their first-pass result looked familiar: a U-shape again, though already much flatter than Solomon's or Cirillo's. Then they looked closer at what was actually inside the slow half of their crash data, and found the detail that changes everything: 44% of the analysed crashes involved manoeuvring vehicles, meaning vehicles turning, entering or leaving the traffic stream. A car waiting to turn right across traffic, or pulling out of a side road, is travelling slowly by definition at the moment it collides. Counting those crashes as "slow-vehicle crashes" quietly converts the risk of the manoeuvre into an apparent risk of low speed itself.

When the RTI reanalysed its own data with the manoeuvring crashes set aside, the increased risk of the slow-moving vehicles came out much lower than Solomon's and Cirillo's results had indicated. Note the careful wording, because it matters: the slow arm did not disappear in that reanalysis, and nobody serious claims it did. What the reanalysis showed is that a substantial part of the slow-side risk in the famous curves was an artefact of counting turning and merging vehicles as slow travellers. Solomon and Cirillo had included manoeuvring vehicles in their analyses; strip them out and the valley wall on the slow side loses much of its height.

Exhibit B: the matched studies, 1997 and 2001

The modern test of the question came from Craig Kloeden and colleagues at the University of Adelaide, in two case-control studies that fixed both of the old defects at once. In the 1997 urban study, on Adelaide roads with a 60 km/h limit, 151 crash-involved vehicles were compared against 604 control vehicles. Crashes only counted if someone was admitted to hospital or killed, and the pre-crash speed of each case vehicle was reconstructed by experts from physical evidence using computer reconstruction, not read off a police officer's note. Most importantly, each case was matched to controls travelling through the same location in the same direction, in the same area, at the same time of day, in comparable weather and light, in the same type of vehicle. The 2001 companion study did the same on rural roads with limits of 80 to 120 km/h, with 83 cases and 830 controls, ten per case.

The result, in both studies: vehicles travelling faster than the traffic around them showed increased crash risk, rising steeply with every extra increment of speed. Vehicles travelling slower than the traffic around them showed no increased risk. No U. A rising curve on one side, and on the other side, nothing the data could distinguish from the risk of simply driving along. The review this guide is built on, by Letty Aarts and Ingrid van Schagen at the Dutch national road-safety institute SWOV, weighed the whole literature in 2006 and put the two halves of the question in plainly different categories: on the fast side, every study of every era points the same way, while for slower-moving vehicles, in the review's words, "the evidence is inconclusive". Older studies found a slow-side risk; the newer, better-controlled ones did not.

The Kloeden urban study did not just fail to find a slow-side risk; it quantified the fast side with a published equation. From the 2002 reanalysis of the Adelaide data, the risk of involvement in a serious crash rises exponentially with each km/h an individual vehicle travels above the average speed of the traffic around it. We have computed what that equation implies at round numbers, and the ladder is worth staring at. On an urban 60 km/h road, relative to a car moving with the traffic, a car 5 km/h faster carries roughly 1.9 times the risk of a crash serious enough to put someone in hospital. At 10 km/h faster the multiple is about 4.1. At 15 km/h faster, about 10. At 20 km/h faster, about 30. These are our computations from the published curve, not figures printed in the paper, and the crash measure throughout is hospital admission or worse. The rural equation from the 2001 study climbs too, but less steeply: one of the review's consistent findings, from every study design, is that the same extra speed costs more on minor and urban roads than on major and rural ones, which is exactly where learners and town drivers spend their time.

Different methods land in the same territory. Two large British self-report studies measured drivers' real speeds on real roads, then surveyed their crash histories: Maycock and colleagues in 1998, with 6,435 drivers, fitted a power function with an exponent of 13.1, and Quimby and colleagues in 1999, with 4,058 free-flowing drivers, an exponent of 7.8. Translated into the rule of thumb the authors used: each 1% increase in speed is associated with roughly a 7.8% to 13.1% increase in crash liability. Hold that loosely, for two reasons the review itself flags: crash histories were self-reported by questionnaire, with response rates under half, and the percentage rule is only approximately right for small speed changes. An earlier Australian self-report study by Fildes and colleagues found the same exponential pattern, again steeper on urban roads than rural ones. But notice what has happened: three self-report studies, two matched case-control studies and one reviewed literature all agree that risk climbs with speed at an accelerating rate. Nothing about the fast side is inconclusive.

Two loose ends complete the picture. First, speed variance at the level of whole roads: American work by Garber and Gadiraju found that roads where speeds are more spread out have more crashes, and, intriguingly, that crash rates and speed variance were lowest where the limit was set 5 to 10 mph below the road's design speed, which is roughly the opposite of the myth's spirit. British work by Taylor and colleagues found the same variance link on urban roads. Both carry a caveat the review is careful about: the variance in these studies was measured across a whole 24 hours, so it partly reflects the difference between rush-hour and free-flow speeds rather than cars weaving past each other in the same moment. Second, average speed: when a whole road's mean speed changes, crashes change with it, following the well-established power model in which injury crashes track roughly the square of the speed change and fatal crashes roughly the fourth power; that relationship, and the kinetic energy underneath it, has its own guide in Speed, risk and kinetic energy, so we will not restate it here.

Start with what the test actually requires. The RSA's own guidance on the driving test says you must "make safe and reasonable progress", and lists "progress speed" among the aspects assessed. On the tester's marking sheet, Progress is a heading in its own right, assessed at moving off, on the straight, overtaking, at cross junctions, at roundabouts, turning right, turning left, changing lanes and at traffic lights; Speed is a separate heading, assessed against road conditions, traffic conditions and the speed limit. The tester's marking guidelines give concrete examples of progress faults: sitting at a safe gap without taking it when turning right, waiting at a green light when it is clear to proceed, crawling into turns far below the speed the situation allows. Our guide to test faults is blunt about it: both exceeding the limit and driving too slowly for the conditions are faults, and a nervous candidate trundling 5 to 10 km/h under the limit will start collecting marks the moment traffic behind is forced to brake, swerve or overtake. Our test-day guide says the same: driving well below the limit on an open road is a fault.

Now the part the myth gets wrong about the law. Contrary to what many drivers believe, Irish law sets no general minimum speed limit. The speed-limit framework under the Road Traffic Act 2004, as amended, sets maximums only; there is no offence of simply driving below any particular speed. (Driving absurdly slowly without regard for anyone behind you is not beyond the law's reach, since the general careless-driving provisions exist, but there is no minimum-speed offence as such.) The one statutory speed-capability rule sits on motorways, and it is about the vehicle, not your speed on the day: under Article 7(2) of the Roads Regulations 1994, a vehicle may only use a motorway if, among other conditions, it is constructed, conditioned and loaded so as to be "capable of maintaining a speed of 50 kilometres per hour on the level" under its own power. That is why tractors and low-powered machines cannot use motorways; it is not a minimum speed imposed on you and your capable car. Learner-permit holders are excluded from motorways by a different condition of the same article, the requirement to hold a licence for the vehicle. For the full current speed-limit picture, defaults, recent changes and the penalty ladder, see our complete guide to Irish speed limits.

Notice that the myth only ever gets used in one direction. Nobody says "keep up with traffic" to make a crawler speed up to 50 in a 50; the phrase is deployed to justify 130 in a 120 because everyone else is doing it. That is why we framed the claim on trial precisely as matching the flow above the limit. Below the limit, "keep up with the traffic you can safely keep up with" is genuinely good advice, ours included: it is what reasonable progress means, it is what your tester wants to see, and nothing in this article argues otherwise. What the evidence removes is the excuse for the ceiling, not the existence of the floor. The U-curve's slow arm was the only scientific justification the over-the-limit version ever had, and it has not survived cross-examination.