Even if it's just about sharing insight, it is interesting that people were receptive to the insight, especially since Kahneman & Tversky had one study that implied that people are not. K&T presented people with two arguments, the correct argument explaining why Linda is more likely to be a bank-teller and an incorrect argument saying that she's more likely to be a feminist bank teller because that sounds more like her, and a majority (65%) found the incorrect argument more convincing.
Still, this new paper is vastly overstating the implications of their study. They used the transparent within-subjects version of the Linda problem, but the between-subjects version (where each person is asked to consider either "bank teller" or "feminist bank teller," not both) is more like the kind of problems that we run into in real life. More generally, the problem is that people think about likelihood in terms of representativeness when they should be thinking extensionally; the conjunction fallacy is just one particularly egregious consequence of that bias. This just shows that there are some cases where groups of people can overcome that problem.
The proportion of wrong answers in groups is not dropping faster than it should - that is, if you assume that any group containing a single insightful responder answers correctly, then the odds are no lower than they should be. In fact they are slightly too high. 58% for singles vs. 48% for pairs vs. 25% for trios. Or with incentives, 33% for singles vs. 13% for pairs vs. 10% for trios. We can't assume an effect of deliberation rather than a sharing of individual insight.
We hear a lot about wealth not being a fixed pie. Why don't we hear the same about rationality?
Even if it's just about sharing insight, it is interesting that people were receptive to the insight, especially since Kahneman & Tversky had one study that implied that people are not. K&T presented people with two arguments, the correct argument explaining why Linda is more likely to be a bank-teller and an incorrect argument saying that she's more likely to be a feminist bank teller because that sounds more like her, and a majority (65%) found the incorrect argument more convincing.
Still, this new paper is vastly overstating the implications of their study. They used the transparent within-subjects version of the Linda problem, but the between-subjects version (where each person is asked to consider either "bank teller" or "feminist bank teller," not both) is more like the kind of problems that we run into in real life. More generally, the problem is that people think about likelihood in terms of representativeness when they should be thinking extensionally; the conjunction fallacy is just one particularly egregious consequence of that bias. This just shows that there are some cases where groups of people can overcome that problem.
The proportion of wrong answers in groups is not dropping faster than it should - that is, if you assume that any group containing a single insightful responder answers correctly, then the odds are no lower than they should be. In fact they are slightly too high. 58% for singles vs. 48% for pairs vs. 25% for trios. Or with incentives, 33% for singles vs. 13% for pairs vs. 10% for trios. We can't assume an effect of deliberation rather than a sharing of individual insight.
The "Linda" example they used has dated quite a lot in the 25 years since Kahneman and Tversky used it.