Everywhere I look I see stuff about how ‘unreliable’ much science might be. For instance see here, here and here. In truth most of this is about the recent paper on the lack of reproducibility of some outputs [irony intended] in psychology journals. John Ionnnidis writes:
Multiple lines of evidence suggest this is a recipe for disaster, leading to a scientific literature littered with long chains of irreproducible results.
I do not have much original to add to this debate, but find none of it very surprising. Go back and read the ‘Double Helix’, or look at how David Hubel and Torsten Wiesel described how they changed the world. I think of it in the following terms.
If people use probability management as the sole marker of truth, you have left science — that is science as the attempt to discover the deeper structures that underpin reality — behind. Science is about reliable knowledge, and if what you do is not reliable, you have not been doing science. There is a lot of activity that is going on that that is not science. Much is Cargo cult science; much is testing; much is effectively the D of R+D where we are playing and abusing Type 1 and Type 2 errors (Neyman Pearson errors, rather than Fisher); much the sort of thing any large logistics or commercial company does, but they probably have more humility. And it is why I have argued that much medical research (sic) could be usefully outsourced to Amazon: its logistics, stupid. ( see for instance, Of supermarkets and medical science: getting the product lines right). The Feynman quote that I recall was: “When, during the Challenger enquiry, Richard Feynman was told that the chance of an error in a certain process was 1:105, he retorted, “if a man tells me the chance of failure is 1 in 10^5 I know he is full of crap”.
Science relies on truth seeking; integrity; and honesty. It is not just a career. You have to have an attitude to your subject. Arrogance before men; humility before your subject (Bronowski). Most good science is iterative. You want to see the world honestly because your next bit of work depends on how you have interpreted what you did yesterday. In one sense, you publish for yourself. The hardest judge must be you and your reputation, not some anonymous reviewer. Your peers are not the people who review your papers, but those people you do not know who decide whether you are in the books 20 years later. Think of dietary observational epidemiology: how much will survive?
When publication is no longer a matter of record, but of career advancement, we are doomed to endlessly inventing ways to discover who has been debasing our currency. Institutions matter, not so much the bricks and mortar ones, but the way science is organised. John Ziman foresaw the mess we are getting into in his book Real Science, and elsewhere:
What is more, science is no longer what it was when Merton first wrote about it. The bureaucratic engine of policy is shattering the traditional normative frame. Big science has become a novel way of life, with its own conventions and practices. What price now those noble norms? Tied without tenure into a system of projects and proposals, budgets and assessments, how open, how disinterest- ed, how self-critical, how riskily original can one afford to be?
He was of course talking about Robert Merton’s essay (J. Legal and Political Sociology 1, 115–262; 1942). He went on to say:
There is no going back to that world we have lost. Anyway, science has never had it so good as in this past half-century, and is still going great guns. But soon it will be offered a new contract with society. To renegotiate that contract with its eyes open, on even terms, science will need to understand itself much better. That understanding is going to require, not adherence to an obsolete ethos, but a sharp but sympathetic sociological self- analysis. That is the unfinished business that Merton’s little paper began.
This is where we are. We need to find such a new contract. Bruce Charlton had the right diagnosis a long time ago (Not even trying: the corruption of real science).