Blind Choice Paradigm

What it is

The flaw it answers is Keith Chen and Jane Risen's. In the standard free-choice design, a person rates objects, chooses between two, and rates them again, and the chosen one rises while the rejected one falls. If the first ratings are an imperfect measure of preference and the choice is partly guided by preference, that pattern appears whether or not anything inside the person moved. Blinding the choice removes the second assumption. Two other repairs are named alongside it: one that measures the artefactual spread and subtracts it, and one that looks for preference change in items never directly compared.

In effect

The cleanest version used holiday destinations. Tali Sharot, Cristina Velasquez and Raymond Dolan had people rate them, then make a choice in which the two items were presented for two thousandths of a second and the strings presented were in fact nonsense, so nothing about the choice could have been guided by preference. A control condition had a computer make the decision instead. Preferences shifted after the blind choice and not after the computer's.

An earlier version by Louisa Egan, Paul Bloom and Laurie Santos did the same job twice over, with a genuinely blind choice for young children and an illusion of choice for capuchin monkeys, where the experimenter was in fact assigning at random. Egan and colleagues published it as a correction to their own 2007 free-choice design, which is creditable and worth saying in those terms.

What it does not say

It does not say the corrected designs settled the question. The children-and-monkeys study has a measurement problem at the other end: post-choice preference was read off a further blind choice for the children, which is not a good measure of preference, and off ten subsequent sighted choices for the monkeys, during which preference could have moved on its own. Keise Izuma and Kou Murayama's verdict is that the manipulations adequately address the critique while the findings are not sufficiently compelling, which is a precise and unusual thing for a review to say.

The destinations study, and the neural study that uses the subtract-the-artefact control instead, are the two whose effects are significant only on a one-tailed test. No sample size was reached for either and none is printed here. As of that 2013 review the corrected-design evidence base appeared to consist of four studies in total, which is a striking fact if it is still true. And blinding a choice does not make it an ordinary choice: a decision made about objects you cannot see is a long way from buying a sofa.

---