TRL 511 – The Need for Speed (Reduction)

High-level summary:

What is TRL 511?

TRL 511 is the report from the Transport Research Laboratory used by the DfT and Councils like Oxfordshire to justify reducing speed limits on the open road from 60mph to 50mph, and then enforcing them with cameras.

What does it say?

The report's conclusion is that reducing average traffic speeds will slash accidents, no matter what caused them. It supports this by comparing accident rates and mean speeds on different rural roads across the country.

What does it REALLY say?

Reading TRL511 in detail reveals that the initial assessment of 174 rural roads showed that accidents FELL with higher speeds rather than rose, because there were fewer hazards on the roads where people drove faster.

How did the authors turn the results around like this?

The report writers managed to turn this predictable conclusion on its head by splitting up the roads by hazard density into four different groups, and claiming that, within each of these groups, accidents ROSE with speed.

Is this approach valid?

The methodology used to reach this conclusion is flawed because it is invalid to compare accident rates between roads that carry different amounts of traffic, even if the roads are otherwise similar.

What use is TRL511?

TRL511 is a contrived report that really shows that more accidents happen where there are more hazards, and this is nothing to do with the average speed of traffic.

How has this distorted road safety policy?

The way to address road safety is to deal with the actual cause of real accidents. The authorities are using reports like TRL 511 to avoid telling the public what really causes crashes, using this statistical chicanery to falsely blame the responsible behaviour of the majority of motorists.

What effect has this had on road safety and wider society?

Various authorities then hide behind baseless and vacuous statements like "a 1mph reduction in average speed causes a 5% fall in accidents" to dismiss reasonable complaints about nonsensical speed limits and enforcement that is unrelated to safety. This has led to falling driving standards and mass breakdown in respect for law and order.

Background

When they legalised speed cameras in 1992, the authorities clearly intended from day one to use them to drive down general speeds by blanketing roads with them and reducing speed limits. Some of the worst examples of cameras being placed on straight rural roads and urban dual carriageways date from the very earliest days of the grey boxes. Some of these have even been taken down by camera partnerships, as they do not meet new visibility guidelines. Wholesale limit reductions began at the same time – the first batch of Oxfordshire rural 50mph limits dated from the opening of the M40 in the early 90's.

But they had a problem – all the road safety research that had been done supported the setting of speed limits at sensible levels, so there was no intellectual backing for their blanket speed reduction policy.

Quickly, they cobbled together a report known as "Finch et al." full of impressive looking statistical analysis of historical, international examples of speed limit changes. This created the absurd claim that a 1mph reduction in average speed would always cut accidents by 5%.

At a stroke, any concept of there being a "correct" and "sensible" absolute level at which to set a speed limit was swept away. However low a speed limit was, "Finch et al" would always justify lowering it further.

Finch et al is almost embarrassingly contrived, violating the most basic principles of statistical reporting.

Finch et. Al. is still quoted regularly as the authoritative piece of research that it is manifestly not. But the speed reduction brigade clearly felt they were skating on thin ice, so they commissioned two further studies to back up their position.

TRL 421 and TRL511 are the illegitimate love twins of "Finch et al", dealing with urban and rural roads respectively.

TRL 511 – What it Says

Lets look at TRL 511, which is being used to justify more rural 50mph limits on perfectly good roads.

The authors chose 174 rural roads or between 1km and 10km in length, all subject to the National Speed Limit of 60mph, and a mix of A, B, C and unclassified.

Then they measured the speed in ONE location on each of these roads for TWO WHOLE DAYS.

Armed with this less than sophisticated data and a sheaf of STATS 19 forms, they counted the number of accidents over a four-year period (about two per year per location), divided them by the measured traffic levels to get an accident rate.

Then they made an adjustment to this accident rate because it is well known that if you double traffic on a road, you never get twice as many accidents. Crucially, the formula they used to make this adjustment was a simplistic, linear one. This is the vital error in their argument.

They were then left with an adjusted accident rate from which all the known effects of different levels of traffic had been stripped away.

Then they plotted this accident rate on a graph against the average speed on each road.

One can only imagine their horror when they discovered that this accident rate FELL as speeds rose. Yes – the faster the road, the fewer accidents there were!

Any reasonable person would conclude that accidents are more likely where there are more hazards, but that the vast majority of drivers respond to those hazards correctly and this is why speeds are also lower in these places.

If there is a road with a straight bit and a dangerous bend, speeds will be slower around the bend but that is nonetheless where the accidents will occur, because one or two drivers out of thousands will attempt to take it too fast.

This is blindingly obvious, but for those who commissioned the TRL report, it does of course give:

THE WRONG ANSWER!

Undeterred, the authors split the sample roads into four different types of roads, based on the density and nature of hazards such as bends, hills and minor junctions. They then assumed that all of these roads were identical both with each other and along their entire lengths. They called them "homogeneous" roads.

Lo and behold, when they looked at each of these four types of "homogeneous" road separately, they found that within each of these types they got the answer they wanted – a nice little graph showing how accident rates increase with average speed.

They finished the report by showing how many lives could be saved by reducing rural speeds in accordance with their cosy little mathematical equation.

A CRITICAL ASSESSMENT

This report contains no assessment whatsoever as to the causes of any of these accidents, and whether excessive speed was put down as a causal factor on the STATS 19 forms, which were available to the writers of TRL511.

I can think of no rational reason why the authors of any report on road safety would ignore this vital information and approach the study in the way it was actually done. Other than, of course, the need to arrive at a pre determined answer to support existing policy.

All of the roads, even when they are split into these homogeneous groups, will vary along their length, with some straight bits and some bits with bends and junctions. Speeds on the straight bits will be higher, but accidents on the slower, bendy bits will be more numerous. That doesn't change. And the average speed of thousands of
drivers in on single location still has nothing to do with the actual speed of the one or two drivers who crash somewhere else on the road.

But that's realty – these people don't recognise such things, and they would just point to their equations and say they demonstrated that the link was there, and that it was statistically significant!

There is, however, a systematic flaw in their reasoning which points to an erroneous explanation for their results. Put simply, they have shot themselves in the foot.

THE FLAW

They have allegedly created four groups of roads that are so comparable, so "homogeneous", that both speeds and accident rates can be directly compared between roads in the same group. This in itself is contrived and unlikely – these sorts of roads vary by the hour and by the 100 yards – but we accept that these variations would be random and wouldn't explain the result they obtained in the report, so let's bear with them for a moment.

If these roads were that similar to one another, then why on earth would the average speed of traffic be different? Why do they have a variable to measure at all? If the hazard spread and road width is the same, why should Lincolnshire drivers choose a different speed from Gloucestershire ones?

By claiming that the roads are identical, the authors of the report have left only one reason for the average speed to vary – traffic levels. As a road gets busier, the average speed will drop, even in free flowing conditions, as the faster drivers are more likely to be held up by slower ones and overtaking opportunities diminish.

Now the authors of the report had gone to great lengths to strip out flow difference effects from their calculations – they made a great play of trying to isolate the mean speed variable by stripping away "masking factors" like traffic levels and road geometry, leaving mean speed as the only variable.

The way they did this was to assume that a doubling of traffic levels would increase accidents not by 100% but by 65%, and adjust the accident figures accordingly.

HERE'S THE PUNCH LINE

They have already made an adjustment to the accident rate to strip out the effects of different levels of traffic. That INCLUDES any effects that might accrue from changes in speed caused by changes in traffic volumes!

There aren't any variables left. All they have done is to prove that 0=0! Their methodology strips out the very thing they are trying to measure, and they should end up showing a flat graph, with no variation between speeds and accidents.

The fact that they don't can mean only one of two things:

1 There is some other factor causing the difference in speeds between different but "homogeneous" roads, in which case they are not homogeneous and it is invalid to split them into the four groups without finding out more about the causes of speed variation. One can only speculate what these causes might be – different driver demographics, vehicle types (tractors!), traffic density profiles through the day and so on. Any one of these could also affect accident frequency on a systematic "common cause" basis.

2 Their 65% equation for adjusting accident rates for traffic flows is over simplistic and contains a systematic error which is small enough to be masked by the reductions in hazards in the overall numbers but is revealed when the variations in hazards are reduced in the four homogeneous groups.

Evidence to support the assertion that the analysis in TRL511 is fatally flawed comes from the TRL itself. TRL's Published Project Report PPR026 (Accident Analysis on Rural Roads - a technical guide), contains the following statement in paragraph 4.15:

"The relationship between accidents and vehicle flows is not a linear one (e.g. see Walmsley and Summersgill, 1998). For this reason, it is recommended that roads with very different flow levels are not studied together."