Dec 152016

Ronald Fisher was sceptical about the lady’s tea tasting prowess. But he was prepared to change his mind. If the lady could correctly identify 8 cups, he was willing to acknowledge her ability, or – in his unnecessarily convoluted but equivalent wording – to admit that her ability could not be disproved. He would have done the same with one mistake in 12 cups. In his mind, such procedure had nothing to do with scepticism. He would not let his subjective beliefs taint his conclusions: only data should decide.

This is a misconception. Data can do nothing by themselves. We interpret data into explanations and decide to accept or reject hypotheses, based on our standard of proof (PO), evaluation of evidence (LR) and prior beliefs (BO).

There is hardly a more appropriate illustration of this point than Fisher’s astonishing smoking blunder.

By the time of his retirement in 1957, Sir Ronald Aylmer Fisher (he had been knighted by the newly crowned Elisabeth II in 1952) was a world-renowned luminary in statistics and biology. In the same year, he wrote a letter to the British Medical Journal, in response to an article that had appeared there a month earlier, on the ‘Dangers of Cigarette-Smoking’. Fisher’s letter read ‘Alleged Dangers of Cigarette-Smoking‘:

Your annotation on “Dangers of Cigarette-smoking” leads up to the demand that these hazards “must be brought home to the public by all the modern devices of publicity”. That is just what some of us with research interests are afraid of. In recent wars, for example, we have seen how unscrupulously the “modern devices of publicity” are liable to be used under the impulsion of fear; and surely the “yellow peril” of modern times is not the mild and soothing weed but the original creation of states of frantic alarm.

A common “device” is to point to a real cause for alarm, such as the increased incidence of lung cancer, and to ascribe it urgent terms to what is possibly an entirely imaginary cause. Another, also illustrated in your annotation, is to ignore the extent to which the claims in question have aroused rational scepticism.

Amazing. Fisher’s scepticism at work again. But this time he would not change his mind. To our eyes, this is utterly dumbfounding. Today we know: the evidence accumulated by epidemiological research in establishing tobacco smoking as a major cause of multiple diseases is overwhelming, conclusive and undisputed. But it took a while.

This is from 1931. This, featuring a well-known actor, is from 1949:

And this is the other Ronnie in 1956.

Public attitudes to smoking have changed a great deal over the last hundred years – especially in the last few decades. I remember – to my horror – having to bear with my colleagues’ “right” to smoke next to me in the open space I was working on in the ’90s. And what about smoking seats in the back of airplanes? (By the way, why do some airlines still remind passengers, before departure, that ‘This is a non-smoking flight’?).

The adverse health effects of smoking were not as clear in the ’50s as they are today. But there was already plenty of evidence. Granted, the earlier studies had an unsavoury source: Adolf Hitler – a heavy smoker in his youth – had later turned into an anti-tobacco fanatic and initiated the first public anti-smoking campaign in Nazi Germany – that’s where the term ‘passive smoking’ (Passivrauchen) comes from. The link between tobacco and lung cancer was first identified by German doctors. But by 1956 it had been well established by the British Doctors’ Study, building on the pioneering research of Doll and Hill.

How could Fisher – a world authority in the evaluation of statistical evidence! – be so misguided? Let’s see. As a start, he was sceptical about the harmfulness of smoking. Did being a lifelong cigarette and pipe smoker have anything to do with it? Of course. But not – of course – if you asked him: he was prior indifferent – ready to let the data decide. His aggressive, confrontational character – he was apparently intolerant to contradiction and hated being wrong on any subject – did not help either. Besides, he was being paid some – probably small – fee by the tobacco industry. None of this, however, is very important. It is fair to assume that, as in the tea tasting experiment, Fisher was open to let evidence overcome his scepticism. The crucial point is that his scepticism (and stubbornness), combined with a suitably high standard of proof, meant that the amount of evidence he required in order to change his mind was likely to be particularly large.

Here is where Fisher failed: answering Popper’s question. The more sceptical we are about the truth of a hypothesis, the more we should try to verify it. But he didn’t: rather than taking stock of all the evidence that had already been accumulated on the harmful effects of smoking, Fisher largely ignored it, focusing instead on implausible alternatives – What if lung cancer causes smoking? What if there is a genetic predisposition to smoking and lung cancer? – with no indication on how to explore them. And he nitpicked on minor issues with the data, blowing them out of proportion in order to question the whole construct.

The parallel with the today’s global warming debate is evident, indicative and disturbing. It is what happens with all the weird beliefs we have been examining in this blog. The multiplicative nature of evidence accumulation is such that there are always two main ways to undermine a hypothesis: ignore or neglect confirmative evidence and look for conclusive counterevidence – dreaming it up or fabricating it when necessary.

What evidence do you require that would be enough to change your mind? Popper’s question invokes searching for disconfirmative evidence when we are quite convinced that a hypothesis is true. But if, on the contrary, we believe that the hypothesis is likely to be false, what is called for is a search for confirmative evidence – as much as we need to potentially overturn our initial conviction. Popper himself failed to draw this distinction, with his unqualified emphasis on falsifiability. And Fisher followed suit, as reflected in his null hypothesis flip and his verbal contortions to avoid saying ‘prove’, ‘accept’ or ‘verify’. The result was his embarrassing counter-anti-smoking tirade – definitely not the best end to a shining academic career.

PO=LR∙BO. Get any of these wrong and you are up for weird beliefs, embarrassing mistakes and wrong decisions. It can happen to anyone.

Print Friendly