Second-hand Smoke...

and the power of gross statistical fraud.

by John Brignell   at Number Watch

The tobacco-banning movement has finally reached the stage of the last push in England. It is remarkable that it has taken so long, considering that it is driven by the combined forces of the political correctness movement, the zeal of the converts with a fear of their own cravings, those who get their kicks out of ordering others around and the socialist love of banning things.

If the age of science ever returns, the story will provide the classic example of the use of junk science to enforce a policy; though that is unlikely, as history is written by winners. The scientific basis of most of the claims comes from two studies, by the CDC and the EPA. The preliminary results of both were embarrassing: the CDC study showed that smokers lived longer than non-smokers, while the EPA one showed that passive smoking was harmless. The CDC solved their problems with a sequence of statistical fiddles, the most egregious of which involved treating all eighty and ninety year old smokers as having died prematurely of their habit.

The important one, however, was the EPA so called meta-study. They had spent four years developing an anti-tobacco policy when someone pointed out that they did not have any evidence that smokers caused harm to others. Their reaction to this has been covered frequently in these pages and the associated books, as well as widely elsewhere. The most astonishing was committing the statistical crime of changing the level of significance part way through a study, and reducing it to a level that is a record low, yet still achieving only a ludicrous RR of 1.19. The product of these studies were the baseless claims the 400,000 Americans die from smoking each year and 3,000 die from passive smoking. The British zealots  reduced these figures pro rata for population and then characteristically added a bit for luck. Some of the attempts by at evidence the zealots who have taken over the BMA are quite ludicrous. The efforts to rewrite history to cover up any challenges to the so-called evidence are legion (e.g. the vilification of the heroic Judge Osteen).

The proposed English smoking policy is classic Old Labour, with an unnecessary proliferation of ugly notices, criminalisation of an innocent minority, draconian policing with yet more burdens imposed on local authorities and hence more taxes, plus the recruitment of snoopers. Above all there is the irrelevance. We noted back in February 2001 how Labour back-benchers packed the house for the hunting ban and then walked out when serious matters were to be debated. We have national disasters in such areas as hard drugs and violent crime, yet we are going to dissipate resources on catching people having a quick drag. People have always taken pleasure in persecuting minorities, but in the world of PC some minorities are more persecutable than others.

More smoke and mirrors

But you cannot beat the anti-tobacco lobby when it comes to imagination. This one was found by seasoned number watcher Dennis Ambler. Half an hour in a smoky pub could trigger heart attack is the headline in the Independent. That exposure gets you less nicotine than you get by eating a potato. Since the infamous multiple statistical fraud by the EPA in its "Meta-study on Environmental Tobacco Smoke" there has been a scramble among epidemiologists to associate passive smoking with everything from dental caries and bad behaviour in children to infertility and premature death in adults. There is not a single respectable indication of statistical significance in any of them and what they really establish is that passive smoking is completely harmless. But once the EPA had broken the dam with a relative risk of 1.19 at a significance level of P<0.1, anyone could prove that anything causes anything; and they do.

When the EPA comes up with a fraudulent claim of such a magnitude, no other number gets a look in. This organisation has managed to surpass even its own levels of mathematical abuse with these new claims. There have been over one hundred and forty epidemiological studies of the purported relationship between passive smoking and lung cancer (they are listed in the new book by McFadden). In almost all of them the undemanding 95% significance range includes the relative risk value of  1.0, meaning that there is no significance at all. If you believe in marginal statistics, thirty of them show that passive smoking actually  protects against lung cancer.

It is clear that no one has come anywhere near to establishing a relationship. In fact, such an overwhelming amount of non-evidence can only lead to the conclusion that there isn't one.

Yet the EPA has the gall to claim not only that a correlation has been established, but that they are able to show a dose-response relationship. There are a number of implications of this claim.
1. Though no one has managed to establish one point on the curve, they have been able to establish a sufficient number to which to fit a trend.
2. In order to do this they must be able to establish the level of exposure to smoke to a remarkable accuracy.
3. They then claim that the trend is statistically significant. "P for trend" is a new fashion among epidemiologists. It was a year ago that Number Watch remarked on this new phenomenon and in the first month of this year begged for information on how they purport to calculate it. No one has offered any information.
4. As we have shown, you can fit a trend to any small number of random points, and the slope depends only on the outermost ones. Experience suggests that a trend is real only if it is self-evident to the naked eye.
5. The truly gross act is to combine all the probabilities to produce the one in a billion estimate. In virtually all the trials the probability of the result being wrong is not 0.05, since the RR of 1.0 is within the equivalent undemanding range. Add to this the effects of publication bias, the absence of any acceptable RRs, together with the problems listed above and you are forced to the conclusion that the probability of each study being wrong is close to 1.0.  Thus the probability of them all being wrong is also close to 1.0, which is rather different from the reciprocal of one billion.

The EPA, of course, will blithely ignore any such criticisms and carry on propagating its anti-scientific nonsense for its own solely political reasons.

Perhaps the most fundamental defect of the presentation of the risk of passive smoking is the failure to distinguish between relative and absolute risk. In a critical commentary, the Australian medical research scientist Raymond Johnstone noted that the annual death rate from lung cancer among the non-smoking wives of non-smoking men is around six per 100,000, whereas among the non-smoking wives of smoking men the corresponding figure is eight per 100,000. Now this may be reported as an increased (relative) risk of 33 per cent. Yet in absolute terms it amounts to an absolute (or exposure) risk of one in 50,000, which is, for practical purposes, negligible.

Johnstone's conclusion was that 'the most that one can say about the alleged link between passive smoking and lung cancer is that if there is one, then it is so small that it is difficult to measure it accurately and the risk, if any, is well below the level of those to which we normally pay attention'. The alarming estimates of deaths attributable to passive smoking result from multiplying miniscule risks of dubious validity by vast population numbers - an effective propaganda device but statistical sharp practice.

The intense moral fervour and political commitment now driving the campaign against passive smoking has created a climate inimical to serious scientific inquiry. In 2003 the British Medical Journal published a study of 120,000 adults in California over a 40-year period, which concluded that 'the results do not support a causal association between environmental tobacco smoke and tobacco-related mortality, though they do not rule out a small effect'.