July 2, 2009

The rather extended metaphor got subbed out, but a lot of the geekery stayed in…

Tom Whipple

The Times

Imagine that Britain is one large version of Grand Theft Auto. Imagine now that the media is a spotty 14 year-old, happily mowing down pedestrians on a gleeful shooting spree. For almost a decade, from 1997 to 2004, newspapers cheerfully spread the idea that the MMR vaccine caused autism. They trumpeted compelling anecdotes from teary Islington mothers, made fun of those silly science boffins in white coats, and sold a lot of papers. The game was fun.
Then something they hadn’t thought of happened: in 2004, in the real world, MMR uptake dropped to 80 per cent of two year-olds – from 92 per cent in 1997. Doctors – including one of those who had originally posited the autism link – warned that children were at risk of dying. The game, it seemed, was interactive. The pedestrians were real and it wasn’t so much fun.

Since 2004, broadly speaking, newspapers have accepted they went too far. Stories about MMR are now more likely to rubbish a link with autism, and papers have congratulated themselves on being responsible. But has it worked? Can papers be a force for public good, as well as bad?

In 1997, 17 articles in the UK mentioned MMR. In 2004 there were 1442. If you plot this data against the uptake of MMR, there is a very strong negative correlation (-0.73, for the geeks). This implies, as you would expect, that the more the media question the jab, the fewer people take it. Since 2004, you would expect something different. If more articles supported the scientific line, then the correlation should change. It should move closer towards zero.

What has happened? Well, in the last five years MMR uptake has slightly increased. But there has also been a corresponding decrease in the number of articles. In fact the correlation between the two is almost unchanged from before (-0.82). This is not good. One obvious conclusion is, it doesn’t matter what the articles say: the more there are, the worse it is for MMR uptake.
Last month we learnt that there is a measles outbreak in Wales. The temptation is for newspapers to start writing about MMR again but this time to encourage vaccination. Perhaps, though, the best thing we can do about it is just shut up.

Advertisements

June 26, 2009

The New York Times has some characteristically excellent graphics, charting Michael Jackson’s billboard ranking versus other artists. Scarily, it seems that Mariah Carey is just as successful as the King of Pop, RIP

http://www.nytimes.com/interactive/2009/06/25/arts/0625-jackson-graphic.html?hp

June 16, 2009

Below is the latest of my stats columns…I’ve decided to start posting them online, after the paper decided not to (when it does, they are in the Women/lifeandstyle section).

The Times, June 15

That it would be a personal tragedy is not in doubt. But if Gordon Brown is forced from office – an office he waited a decade to reach – is his plight Shakespearean or Greek? Statistics can help.

 

The protagonist in a Shakespearean tragedy is brought down by a character flaw. In a Greek tragedy he is destroyed by fate. To claim that Brown’s tragedy is Greek would be to say that he was doomed from the moment he assumed power. The argument might go as follows: after ten years of Labour, the country inevitably wanted change, and thus Brown would inevitably fail.

 

As a statistical hypothesis, this theory relies on the assumption that prime ministers who take office to replace an opposing party (as Blair and Thatcher did) do better than those who take office after succeeding a previous member of their own party (as Brown and Major did). This seems sensible, but what do the statistics say?

 

There have been 25 prime ministerships since 1900. 13 of those were like Blair, taking over from a different party. 12 were like Brown, taking over from a fellow party member.

 

Those who initially gained power from a general election served an average four years and six months; those who didn’t, served four years and two months. There is very little difference. It appears that Brown’s tragedy was not in fact ordained.

 

Now that we know the play is Shakespearean, all that remains is to assess the magnitude of the tragedy. Is it an epic worthy of Lear, or is it more like the first half hour of Bambi? Even if Brown goes this month, he will have survived two years. Andrew Bonar Law, Alec Douglas-Home and Anthony Eden all served less. In previous centuries it was common to serve less than a year.

 

But perhaps longevity is not the only measure of success. After all, John Major lasted seven years.

Tom Whipple
387 words
26 May 2009
The Times
1
5
English
(c) 2009 Times Newspapers Limited. All rights reserved

While using taxpayers’ money to pay for dredging of moats and duck ponds may tell you something about individual MPs, it is difficult to deduce anything broader about the state of the Commons from such stories. To see what really went on, we need numbers.

In the past two weeks, second-home expenses claims of more than 100 MPs have been broken down and published. Many as-yet-untarnished MPs claim that the House has been let down by bad apples. Statistically, they are arguing that there are two expenses populations: the crooked and the clean. Some of the accused MPs make the counter-claim that they are just unlucky; that they follow the rules in the same way as all 675 MPs. Statistically, they argue that there is one MP population, made up of the crooked … and the caught.

There is a way to test this. Parliament already publishes the second-home expenses of every MP, without individual breakdowns. So we have always had data for the whole MP population. Now we can also identify a subset whose expenses, broken down, look dodgy. The question is, do these sets, as the untarnished MPs claim, act differently? Do the dodgy MPs spend more? A crude way to test if two sets of data are similar is to see if they share the same average and variance, which is the movement about that average.

For all MPs, the average second-homes claim was £16,500 in 2004 and £18,000 last year. For those MPs whose expenses claims have been criticised, the average claim is £1,500 above that. The variance is also different. It seems that the “crooked and clean” hypothesis should be accepted.

But the expenses furore concentrated on second-home allowances — so only those who could justify second homes were scrutinised. As the overall MP population also includes those with no second home (who claim £0), we are not comparing like with like. What if we remove those 50, mainly London-based, MPs? Then the average rises by £1,500 and the variances equalise. The populations are the same, after all.

Does this mean that all MPs are equally corrupt? Or uncorrupt? That, alas, is as far as the numbers will take us.

Tom Whipple
378 words
19 May 2009
The Times
1
6
English
(c) 2009 Times Newspapers Limited. All rights reserved

As environmental policies go, it is a tricky sell: buy more cars to save the planet. But the government scrappage scheme — offering £2,000 towards a new car if you trade in a ten-year-old model — claims to do just that. It may be designed to boost the economy but, Lord Mandelson assures us, the benefits will be felt in happy polar bears as well as happy bankers.

The premise is this: ten years ago, an average car belched out 190g of CO2 per kilometre; now it gently wafts 158g over the same distance. So new cars are better. Environmentalists counter that, while this may be true, there is also a CO2 cost in just making a new car. And here is the problem. Emissions data is a routine requirement for all cars. There is no such requirement for production data — we must trust manufacturers.

Volkswagen estimates that making a car takes about five tonnes of CO2. Ford estimates six tonnes. The Society of Motor Manufacturers and Traders, meanwhile — which lobbied for the scrappage scheme — prefers just 0.6 tonnes. Someone is wrong; but let us go with the majority view. It would take 190,000 kilometres of driving before the advantage in lower CO2 emissions outweighed an initial six-tonne disadvantage in production. As the average UK car covers 14,500km a year, that equals 13 years of driving.

Sceptical environmentalists may have yet another gripe. It was reported yesterday that, according to “one Citroën dealership in Reading”, most of those taking the £2,000 are pensioners. Now, Badstats would normally prefer a broader sample — perhaps surveying a Toyota garage in Maidenhead as well — but, assuming that the story is correct, how would this affect the calculation? According to the RAC, over-70s drive about half as far as the national average — thus requiring 26 years to recover the CO2 on a new car. This leads to another, delicate, question: how old do you have to be before death is a more probable outcome than being around for 26 years? The answer is 56. The new car scrappage scheme: excellent for the environment, provided both car and owner can last for a quarter of a century.

Migrant Workout

June 2, 2009

TomWhipple
422 words
12 May 2009
The Times
1
6
English
(c) 2009 Times Newspapers Limited. All rights reserved

If I compare Migration Watch, the “independent non-political think-tank”, to the BNP, it is — you must understand — only for illustrative purposes. Both face a similar problem. The BNP wants to convince people that it is a serious political party. Migration Watch, which campaigns for stricter immigration control, wants to convince people that it is a serious academic think-tank.

It has had moderate success. When, last week, it published a report stating that each illegal immigrant given legal status would cost £1 million over his or her lifetime, it featured on the front page of the Daily Express.

This is an impressive calculation to make. You have to project 40 years into the future, consider the differential earning power of people with different qualifications — as well as predicting changes in benefits. One crude way of doing it would be to look at firstgeneration immigrants who arrived in the Sixties, and see how they have fared.

But that would require a lot of work. How did Migration Watch do it? Simple. It assumed that each immigrant is a 25-year-old male who over 40 years will never be promoted and never earn above the minimum wage, who will marry and have two children, but whose wife will never work.

Was this misleading? “No,” says Sir Andrew Green, the Migration Watch chairman. “The example was only for illustrative purposes.” So we can’t make any wider conclusions? “It was for illustrative purposes.”

When your report said that “an amnesty for illegal immigrants would cost taxpayers, on average, an extra £1 million over the lifetime of each immigrant,” was that correct? “It was for illustrative purposes.” The report refers to a person who may or may not exist, who seems to have been chosen because he represents a worst reasonable-case scenario. There are two conclusions. Either Migration Watch is incompetent or it is malicious.

But there is one final calculation that could resolve the dilemma. Migration Watch knows that a feckless immigrant costs more if he lives in London rather than outside: it estimates £1.1 million versus £0.9 million. It is difficult to produce a nationwide figure from this data, as it requires knowledge of the settlement patterns of illegal immigrants. But Migration Watch was diligent. It carefully weighted the data, summed across the population densities, and reached the figure of £1 million.

Only kidding. It added 1.1 to 0.9 and divided by two.

Tom Whipple
374 words
10 April 2009
The Times
1
7
English
(c) 2009 Times Newspapers Limited. All rights reserved

his week, Bad Statistics would modestly like to suggest a new clause in the press code of conduct. If a paper reports on an annual statistic one year, it should do so the next. We will outline our case with the statistician’s oldest metaphor: the coin toss.

When you toss a coin, there are three possible outcomes: heads, tails or standing on its edge. The same is true with annual statistics. The 2009 UK widget output can be higher than 2008, lower or exactly the same. Given the scale of our widget-making industry, the chances of it being exactly the same is like a coin landing on its edge.

So even if a statistic — teen pregnancies, say — stays at the same level, it will probably have either decreased slightly (not a news story) or increased (preferred terms are “soared” or “spiralled”). After ten years of small but steady decline, this year teenage pregnancies in England and Wales “spiralled” (© the Daily Mail) from 40.9 per 1,000 15 to 17-year-old girls, to 41.9.

This was an especially useful statistic, as it coincided with news that a 13-year-old boy had fathered a child. Will teenage pregnancy rates be reported next year? Well, that is entirely dependent on whether they rise.

Even if they don’t, something else — truancy rates perhaps, or glue-sniffing — is bound to oblige. A more sophisticated variation of this technique has proved effective in crime statistics.

Has crime inconveniently fallen this year? No matter, lots of different crimes are measured, each equivalent to a coin toss; one of them is bound to have come up heads. So in 2007, when crime fell, we wrung our hands about an increase in “alcohol-related violence between 3am and 6am”. In 2008, when crime also fell, pre-dawn attacks had inexplicably dropped from the news agenda to be replaced with concerns about school-age burglars.

This is genius. If every social phenomenon in the UK remained constant every year for ever more, the press would still be able to report that Britain is going to hell in a handcart 50 per cent of the time..

Down with arts

June 2, 2009

Tom Whipple
283 words
20 March 2009
The Times
1
5
English
(c) 2009 Times Newspapers Limited. All rights reserved

When Cambridge University announced this week that applicants must have an A* at A level, it was accused of favouring the privileged few who get intense coaching at independent schools. Some years ago, when one of the colleges said that it would look favourably on state school applicants, it was accused of social engineering. Whatever Cambridge does, it will be criticised. But it could broaden intake, maintain a fair applications system and help the economy at the same time. Statistics have the answer.

Cambridge records intake by subject and school type — 75 per cent of classicists come from independent schools, compared with 20 per cent of mathematicians. If classics were abolished, say, and maths expanded, the problem would be solved. This method has an obvious hitch: classics professors might decry it as equally pernicious social engineering. If only there were some way to argue that the subjects favoured by posh kids — the arts — also happened to be intrinsically useless? If only there were an axis of usefulness where classics came out worst and maths best? Well there is.

A year after graduation, Cambridge students are asked whether they are employed, in further education or neither. Plot the data on a graph and the conclusion is unmistakable: the more useless a degree, the more likely it is to be studied by the wealthy — 12 per cent of classicists aren’t in work or education a year later; the rate for mathematicians is half that.

The Government’s course of action is clear: to make the economy stronger and academia fairer, it needs to slash funding for the arts, and make everyone study statistics instead.

Tom Whipple
343 words
13 February 2009
The Times
1
5
English
(c) 2009 Times Newspapers Limited. All rights reserved

Every day thousands of Britons risk neurological damage, paralysis and even death for a temporary high. Some wealthy children — girls in particular — are enticed in before they are even teenagers.

Many are attracted by a gateway product that is perfectly legal: the Shetland pony.

Professor David Nutt, chairman of the Advisory Council on the Misuse of Drugs, which advises the Government, has argued that the risks of horse riding, which he claims accounts for around 10 deaths and 100 road accidents annually, are similar to the risks of taking Ecstasy.

Around 30 people die each year from taking the drug.

Should horse riding therefore be banned? That is a philosophical debate.

Is Professor Nutt correct? That is a statistical debate. This week, after studying Ecstasy use, his council recommended downgrading the drug from Class A to B. The advice was then ignored by the Home Secretary.

Why will no politician even engage with these conclusions? To ignore science is not to ignore just another annoying political argument, it is to ignore our best mechanism for understanding the world: it is to ignore reality.

Here is the reality: one climber dies on Everest for every ten who reach the summit; two motorcyclists die every year in the Isle of Man TT race; competitive male rugby players have around a one in ten chance of injury in each game. These activities are deemed acceptable and encouraged, despite probably carrying more danger than ecstasy use.

If the Government chooses to ignore statistics, it should not be allowed to cite selectively — as it inevitably will — health justifications for maintaining the classification. Its own committee believes that those are not sufficient to justify Class A status for Ecstasy.

The argument now requires a definition as to why risk in some contexts — mountaineering, sailing, horse riding — is laudable, while in others — bare-knuckle boxing, Russian roulette, drug use — it is not. And then, at last, we will have not a scientific argument, but a philosophical one. Discuss.

Fat Chance

June 2, 2009

330 words
6 February 2009
The Times
1
5
English
(c) 2009 Times Newspapers Limited. All rights reserved

Fact: by 2050, 90 per cent of Britons will be obese. But fear not; by 2100, everyone will be anorexic. As well as being obese. Predicting the future is tricky. But when Alan Johnson, the Health Secretary, launched the Change4Life anti-obesity campaign, he felt able to give a killer angle: if nothing is done, by 2050 nine adults out of ten will be overweight.

His figures are based on the idea that obesity will keep on rising over the next 42 years as it has in the past 14. There is a flaw in this plan, of which the authors of his study were aware: “Fitting straight lines to this data presents a problem as these would show some groups passing 100 per cent, which could not occur in real life.” So, if trends continued, more people would be obese than existed.

Instead they used: “A simple and convenient set of slowly varying, monotonic functions that are asymptotic to 0 and 1.” In other words, they fiddled it. This had a peripheral benefit: the calculations look very sciencey, with things such as p(t) = ½[1 + tanh(a+bt)].

But what about anorexics? Between 2002 and 2007 the number of hospital beds used in treating eating disorders increased by two thirds. If that rate continues, in 90 years everyone in the UK (including fat people) will be in hospital with an eating disorder. This is what we know: people are getting fatter; being fat is bad. We don’t know what will happen 42 years from now.

To get technical, why choose to make your trend asymptotic to 100 per cent of the population? Why not to 99 per cent, so excluding PE teachers and athletes? Why not to 85 per cent, in case 15 per cent of the population are genetically incapable of being fat? Perhaps it is worth making up numbers if it scares some fatties into slimming.