TRL421 and TRL511: Propaganda or science?
We've been reading these two reports from the Transport Research Laboratory with a critical eye.

  • Group 1: Roads which are very hilly, with a high bend density and low traffic speed. These are low quality roads.
  • Group 2: Roads with a high access density, above average bend density and below average traffic speed. These are lower than average quality roads.
  • Group 3: Roads with a high junction density, but below average bend density and hilliness, and above average traffic speed. These are higher than average quality roads.
  • Group 4: Roads with a low density of bends, junctions and accesses and a high traffic speed. These are high quality roads.

Figure 2 from TRL511 showing how accidents might be related to speed for four different road types.

TRL421 "The effects of drivers' speed on the frequency of road accidents" and TRL511 "The relationship between speed and accidents on rural single-carriageway roads" are both detailed scientific reports prepared by the Transport Research Laboratory (TRL) for the Road Safety Division of the then DETR now department for Transport (DfT).

Both reports were authored by M C Taylor and A Baruya, with D A Lynam for TRL421 and J V Kennedy for TRL511.

The stated general objective for each report is to investigate and determine the role of speed in road accidents in the UK. They follow on from an earlier claim that "a 1 mph reduction in mean traffic speed leads to a 5% reduction in accidents".

General conclusions of the reports

Both reports claim to establish a certain statistical relationship between mean vehicle speed and accident frequency on UK roads. They purport to confirm and add detail to the earlier claim of "a 1 mph reduction in mean traffic speed leads to a 5% reduction in accidents". We don't need to go into great detail, for reasons which we hope will become apparent.

Statistical modelling

Both TRL 511 and TRL421 base their approach on statistical modelling. In practice this means using a computer program to find ways of classifying the data into groups related numerically to one another. The numbers used are sometimes real and sometimes generated. Input parameters are real, for example the results of observations. But a series of variables are chosen and assigned weights (or "importance factors"). Models of this type are not created automatically, and the first results are useless. The users of the model fuss and tweak until the required results are obtained. If the greatest possible integrity and the best possible data is used then such models are capable of excellent results, with few doubts remaining about their findings.

Faster roads are safer

It's well know that UK motorways are our safest roads. TRL511 immediately goes one stage further and the very first finding is that the frequency of accidents is lower on faster roads.

Figure A2 from TRL 511. This could be a plot of accidents against speeds. The thin trend line could show that accidents tend to reduce as speed tends to increase. In fact this is generated sample data used to illustrate a point.

Figure A1 from TRL 511. This is a plot of the same data as above, but this time they have sub classified the data into four groups, each showing a rising trend.

So the first challenge facing the authors of TRL511 was to find a way of classifying the data to ensure that groups were related to roads, and that each road group showed a rising trend of accidents against speed. They were free to choose any parameters they wanted to sub classify roads, and ended up with a number of groups.

A very curious and important parameter which they chose was accident rate, saying that it was a good indication of "road quality". We're extremely suspicious of the choice, since one of the things we wish to learn about is accident rate. 

Since they already know that faster roads are safer it is vital to create suitable subcategories of roads in order to even suggest (let alone prove) that faster traffic causes more accidents. This sub classification is easily the greyest area of the work, and the most open to tuning and manipulation.


TRL 421 and TRL511 both purport to find a strong statistical link between speed and accidents. But neither report makes any attempt whatsoever to establish that the link is a causal one.

What is causality, and why is it important?

Suppose we observe a population of British adults and discover that taller people are more likely to be wearing trousers. Does that mean the if we give away trousers it will make people taller? Or perhaps if we travel to Africa and observe a tribe of pygmies that they are unlikely to be wearing trousers? Maybe we could restrict peoples' growth in some way and influence the patterns of trouser wearing in the future? No. The simple fact is that men are likely to be taller than women and also more likely to be wearing trousers. There's a link. We can prove it with sampling, but the link is not a causal one.

Without discovering what causes taller people to be more likely to be wearing trousers we've learned little or nothing from the observation, and we would obviously be daft to try to reduce average height by collecting up trousers.

But that hasn't stopped the TRL from making wild claims of accident reduction which might be expected from speed reductions. We have to ask what their motivations were in writing the reports when they decided to wilfully ignore a complete lack of evidence for causality for the purported statistical link. 

The Assumption of Causality

What are the possible explanations for this oversight or leap of faith?

  • They were paid to create the reports as evidence to support a predefined case.
  • They were so very confident of causality that they felt it safe to assume it, but lacked the courage to say that they were doing so.
  • They were grossly unscientific and therefore incompetent in their approach.
  • They have evidence of causality, but didn't bother to present it.
In the absence of evidence you'll have to decide for yourself which item in the list above is most likely.

There might well be another external variable which causes both excessive speeds and accidents. Why the TRL didn't mention such a basic idea is anyone's guess.

But we think it's very easy to explain the general purported link between accidents and speeding. Crashing is usually the result of "bad driving". Driving at a speed inappropriate to the conditions is both bad driving and risks causing an accident. The TRL probably did observe incautious drivers at inappropriate speeds and they probably did observe that these drivers have an increased accident risk. But to infer that slowing them down by a few mph will reduce their accident risk is clearly a leap too far. The elevated accident risk didn't come from speed, rather both the inappropriate speed and the accident risk come from poor driving practices. The link between speed and accidents is highly likely to be non causal. It's quite possible that a proportion of the link is causal, but they would then need to investigate what proportion is causal before they tried to draw conclusions. If there is a proportion of the observed link which is causal, we would expect it to be a small proportion.

For evidence of causality, we should look to proper investigation of individual accidents. TRL323 gathered data from 8 Police forces and discovered that excessive or inappropriate speed was a definite causal factor in 6.0% of the population of definite causal factors. But such a figure, which also includes inappropriate speed within the speed limit, couldn't possibly be used to justify concentrating our national road safety policy on "Speed Kills".

Ignoring the downside

The conclusions drawn and the accident reductions mentioned talk optimistically about reducing the average speed across the entire road network by 2 mph. They presume that this might be achieved by waving the government magic wand, and that no negative effects will result from the changes. This is a very major concern indeed, and we have a web page that specifically lists and considers the new dangers that might be caused by an attempt to reduce speeds nation wide.

The cynical view: A recipe to gain support for a planned unpopular policy

The two reports are similar in general method... 

First gather some data, then plug it into a statistical model, set up a series of variables based on the physical characteristics of the environment and adjust until the desired result is achieved.

Don't bother to try to establish a causal link between speed and accidents because it doesn't matter... after all if more accidents happen in faster traffic on similar roads (after playing with the numbers for a few weeks) it'll be quite obvious that slowing traffic down with the government magic wand will reduce accidents.

Don't worry about real accident causation, we all know that at least 30% of accidents are caused by speeding drivers, and anyway speed is easy to measure. Just remember "speed kills".

Don't bother to factor in traffic composition (HGV / car ratio for example). Don't bother to factor in weather. Do use "accident rate" as a highly weighted "road quality factor" input variable because it helps to create the intended result and no one will notice.

Don't stop to wonder why different roads of the same created class have higher speeds. No one will notice that these roads shouldn't actually be similarly classified. Dismiss the very clear negative correlation between speeds and accidents as soon as possible. We wouldn't want anyone to realize that faster roads are safer. 

When you've finished playing with imaginary numbers and your paymasters are satisfied with the result, draw some wild conclusions assuming a causal relationship and ignoring all other factors. "If we slowed all traffic down by 2 mph, 200 lives each year could be saved". Don't bother to consider the means of speed reduction or any side effects it might have. Especially don't bother to consider that at lower speeds more drivers will be inattentive or fall asleep.

Before publication, do your best to make it appear techie, so that it's harder for lay people to criticize. And don't forget to mention that speed is a major contributory factor in road accidents as many times as possible, because the more we say it the more people will believe it. Oh, and just in case the public wants to read it, make it expensive and not too generally available.

SafeSpeed concludes:

Statistical modelling is a technique wide open to endless manipulation. These two reports make wild and unnecessary claims about lives saved by speed reductions with no supporting evidence of causality whatsoever. As a consequence, we couldn't even begin to trust the authors to have operated the statistical models impartially. The most important question we must ask is: Why did they make wild and unnecessary claims?

The whole approach is one of smoke and mirrors. Real data, fed to a highly manipulatable computer model which is claimed to reveal that accident risk increases with speed. And this despite the very obvious fact that accident risk is lower on faster roads.

Does anyone really think that gradually reducing the speed of a single vehicle in lane 3 of a British Motorway would result in a gradually reducing risk? That would mean (for example) that at 20 mph the crash risk would be less than at 70 mph. On suitable roads with fast moving traffic it should be obvious that the crash risk of exceptionally slow vehicles is increased. Yet they expect us to believe that no such risk increase would be present on fast rural roads? The true shape of the speed crash risk curve is clearly "U" shaped with exceptionally fast and exceptionally slow vehicles at elevated crash risk.

Both reports are pure propaganda, with absolutely no scientific value. We don't dare imagine how this came about, but prefer to believe that some misguided souls are trying to make the roads safer by giving justification to a plan that they sincerely believe in.

It's nothing short of a national scandal that we are being subjected to a national roll out of speed camera technology based in part on the garbage contained in these two reports. Indeed, simply the fact that taxpayers' money was used to pay for them is scandal enough.

Now back to the conclusions in TRL421: Assuming that the causal relationship exists, they go on to say "If we slowed all traffic down by 2 mph, 200 lives each year could be saved". Assuming again that they can obtain the reduction with the government's side-effect-free magic wand would this be the best we could expect from a road safety strategy? To reduce annual UK road deaths from around 3,400 to around 3,200? Surely this would be a complete waste of the opportunity to improve road safety? Shouldn't we be planning changes that might save over 1,000 each year? 

What next?

We wrote to the TRL about this. You can see the letters (here). When that didn't work, we wrote to the Chairman of the TRL too. (click here) The we wrote to PACTS about it. (click here). What should we try next?

new Comments

Safe Speed encourages comments, further information and participation from our visitors. See our (forums).

Read about our comments policy (here).

Many pages (including this one) have a specific associated forum topic. You can (view) or (add comment) to the forum topic for this page. Posting in the forum requires simple registration.

We have a strict editorial policy regarding factual content. If any fact anywhere on this web site can be shown to be incorrect we promise to remove it or correct it as soon as possible.
Copyright © SafeSpeed 2002, 2003, 2004
Created November 2002. Last update 8/03/2004