How scientists collect positive evidence rather than test theories: There's a myth out there that has gained the status of a cliché: that scientists love proving themselves wrong, that the first thing they do after constructing a hypothesis is to try to falsify it. Professors tell students that this is the essence of science.
Yet most scientists behave very differently in practice. They not only become strongly attached to their own theories; they perpetually look for evidence that supports rather than challenges their theories. Like defense attorneys building a case, they collect confirming evidence.
In this they're only human. In all walks of life we look for evidence to support our beliefs, rather than to counter them. This pervasive phenomenon is known to psychologists as "confirmation bias." It is what keeps all sorts of charlatans in business, from religious cults to get-rich-quick schemes. As the philosopher/scientist Francis Bacon noted in 1620: "And such is the way of all superstition, whether in astrology, dreams, omens, divine judgments, or the like; wherein men, having a delight in such vanities, mark the events where they are fulfilled, but where they fail, though this happen much oftener, neglect and pass them by."
Just as hypochondriacs and depressives gather ample evidence that they're ill or ill-fated, ignoring that which implies they are well or fortunate, so physicians managed to stick with ineffective measures such as bleeding, cupping and purging for centuries because the natural recovery of the body in most cases provided ample false confirmation of the efficacy of false cures. Homeopathy relies on the same phenomenon to this day.
Moreover, though we tell students in school that, as Karl Popper argued, science works by falsifying hypotheses, we teach them the very opposite-to build a case by accumulating evidence in support of an argument. The phrase "confirmation bias" itself was coined by a British psychologist named Peter Wason in 1960. His classic demonstration of why it was problematic was to give people the triplet of numbers "2-4-6" and ask them to propose other triplets to test what rule the first triplet followed. Most people propose a series of even numbers, such as "8-10-12" and on being told that yes, these numbers also obey the rule, quickly conclude that the rule is "ascending even numbers." In fact, the rule was simply "ascending numbers." Proposing odd numbers would have been more illuminating.
An example of how such reasoning can lead scientists astray was published last year. An experiment had seemed to confirm the Sapir-Whorf hypothesis that language influences perception. It found that people reacted faster when discriminating a green from a blue patch than when discriminating two green patches (of equal dissimilarity) or two blue patches, but that they did so only if the patch was seen by the right visual field, which feeds the brain's left hemisphere, where language resides.
Despite several confirmations by other teams, the result is now known to be a fluke, following a comprehensive series of experiments by Angela Brown, Delwin Lindsey and Kevin Guckes of Ohio State University. Knowing the word for a color difference makes it no quicker to spot.
One of the alarming things about confirmation bias is that it seems to get worse with greater expertise. Lawyers and doctors (but not weather forecasters who get regularly mugged by reality) become more confident in their judgment as they become more senior, requiring less positive evidence to support their views than they need negative evidence to drop them.
The origin of our tendency to confirmation bias is fairly obvious. Our brains were not built to find the truth but to make pragmatic judgments, check them cheaply and win arguments, whether we are in the right or in the wrong.
Matt, an acclaimed author and former Science and Technology Editor for the Economist blogs at www.rationaloptimist.com. This is the first of three columns on the topic of confirmation bias - first published in the Wall Street Journal.
No comments:
Post a Comment