A few days ago I received an email about my recent post on priors. A reader said he was very surprised that I accepted the claim that saturated fat is not bad for the heart. He said the evidence that saturated fat leads to heart disease is overwhelming and that there is no significant evidence contradicting these findings, other that the activity of groups financed by the meat and dairy industry.
I was also surprised. I had used the New Scientist article on saturated fats only as an example to illustrate the point that even the most well established hypotheses can be challenged by new evidence. I had not said that the evidence proved that saturated fats are ok – nor did the article. New Scientist – not a glossy fashion magazine – referred to a couple of recent studies, published in reputable academic journals (here and here), that shed some guarded doubts on the strength of the received wisdom. The article also said that the papers had been strongly criticised (see here and here) and, after a thorough and informative evaluation of the issue, concluded:
So while dietary libertarians may be gleefully slapping a big fat steak on the griddle and lining up a cream pie with hot fudge for dessert, the dietary advice of the 1970s still stands – for now. In other words, steak and butter can be part of a healthy diet. Just don’t overdo them.
The main point of my post – perhaps the main point of my entire blog – is that, since most evidence is inconclusive, priors matter, and that neglecting them – pretending they do not exist or they are not needed – is a major and consequential fallacy. Prior indifference does not do away with priors – it just sweeps them under the carpet. One may think he is avoiding them, but all he is doing is inadvertently assuming they are 50/50.
Indeed, to be ‘blinded by evidence‘ means that one reads the New Scientist article, understands that there are two camps to the hypothesis and concludes – perhaps aided by the flippant finale – that ‘the truth is in the middle’. That is a mistake: new evidence joins a tug of war where – as it is the case here – one side may already be much stronger. That strength should therefore be reflected in the priors against which the new evidence is evaluated. Only if the new evidence is extraordinarily strong itself – always a possibility – will it be sufficient to counterbalance or even overturn the priors. If not, all it can do is standing against them, waiting for reinforcement that may or may not come in the future.
At the same time, solid priors should help us avoid the opposite mistake: distrust. When the tug of war is in the balance, each side is prone to believe the other side is cheating. The temptation is strongest when one side is losing: the harder the evidence on the other side, the sharper the urge to call it fake – it is indeed the staple weapon of conspiracy theorists.
But when we are on the winning side – when the evidence about what we believe is well established – we should resist our basic instincts. So doubting the bona fide of the authors of the two papers in question – or, worse, accusing them of being on the payroll of industry – is unnecessary as well as gratuitous. Unless, of course, one has strong evidence. If not, his priors are wrong!