Consistency

One thing that has often irked me is the criticism politicians get for making U-turns. What’s wrong with changing your mind if new information comes to light?
A new paper by Armin Falk and Florian Zimmermann of the University of Bonn sheds light on my puzzlement. People, they say, value consistency in themselves and in others as a way of signalling intellectual strength.
And this value is an intrinsic one. We don’t just like consistency because it is a means to better decision-making. We value it even if it gets in the way of rational decisions.
Falk and Zimmermann established this experimentally. They got subjects to estimate the number of peas in a bowl. Subjects were first asked to make an estimate. They were then told that the wisdom of crowds means that average estimates are often more accurate than individual estimates, and told what the average estimate was. Subjects were then invited to give a second estimate of the number of peas, being paid according to the accuracy of that estimate.
They found that subjects did not update their first estimates by very much in light of that new information. As a result, they ended up making more inaccurate estimates, on average, than people who had merely been given information about the average estimate and then invited to make an estimate without having made an earlier guess.
A desire for consistency, they conclude:

induces subjects to act consistently at the cost of neglecting valuable information and receiving lower payoffs…Early statements on a matter can make people ignore valuable information.

This is consistent (!) with a classic study (pdf) by Lee Ross and colleagues back in 1979. They found that when supporters and opponents of the death penalty were given the same evidence to consider - some it of supportive of the penalty and some not - their beliefs became more polarized. This contradicts Bayes’ theorem, but is consistent with people wanting to behave consistently, wanting to maintain their pre-established position.
There are three aspects of Falk and Zimmermann’s experiment that are especially significant:
1. There was no ambiguous information that might have been interpreted to corroborate subjects initial estimates. In the real world however - as in Ross’s experiment - there is often such noisy information. This means the desire for consistency will, in practice,often be reinforced by the confirmation bias, our tendency to interpret new data in a way that corroborates our priors.
2. The number of peas in a bowl is a simple factual matter. An opinion on it is therefore not constitutive of our identity in the way that our political opinions are. It should, therefore, be susceptible to change in the face of new evidence. That even such “neutral” opinions can be easily fixed means we should be especially pessimistic about the prospects for changing other opinions.
3. Subjects in this experiment had little concern with a reputation for consistency. Politicians, however, want to establish such a reputation. This will strengthen the desire for consistency.  
The message here is a depressing one - it’s that rational cool-headed political dialogue is very difficult, simply because once we have taken a position, evidence does not shake it.

Comments

[william]

" What’s wrong with changing your mind if new information comes to light?"

I agree, nothing. The problem with the current lot is that the new information is invariably that their original idea was shit.

[Sean]

I always start a discussion with "on the big issues off the day over your lifetime, on which ones have you been wrong or changed your mind"

mine is GM and the Eu btw, to name but two.

I dont think you can have a serious discussion with anyone regardless of viewpoints, without them being able to answer that question in under 30 seconds.

[CS Clark]

"They were then told that the wisdom of crowds means that average estimates are often more accurate than individual estimates, and told what the average estimate was. Subjects were then invited to give a second estimate of the number of peas, being paid according to the accuracy of that estimate."

Did they control by giving people false information and then telling them, falsely, that a theory proved that the information provided falsely was more likely to be true, to see if the same people who adjusted positively according to good information would also adjust negatively according to bad information?

Maybe consistency is just a bullshit filter.

[David Friedman]

One possibility is that consistency is valued because it makes people's behavior more predictable, which in turn lowers the cost to other people of dealing with them.

[David Ellis]

Following up on #1. I don't think these people are changing their minds because of new information about the substantive issue but because greater political opposition than expected has meant that they can no longer get away with an increasingly obviously self-serving policy.

Consistency is value because you could get eaten by a saber toothed tiger if your mate fell asleep on watch.

[David Ellis]

In other words consistency is a genetically selected trate as without it you were unlikely to survive. Of course, in this political, class riven, dog eat dog world, self-serving, alientated, what's-in-it-for me society, pragmatism rules and in a crisis we have feverish zig-zagging. This must eventually give way to a regime of violence or socialism.

[Andrew]

Whilst the conclusion about desire for consistency is plausible, the pea study has another game-theoretic interpretation.

The social pay-off for changing your estimate to the group average is zero. You get no personal credit if you were off on your first estimate, and none if you were right on your first and changed it to the group mean.

The potential pay-off for sticking to a far-from-mean guess is that you could be proven more right than the group. Sticking to this guess makes it look like you had reason to value your judgement over the average. You are clearly high value.

It may therefore be worthwhile to pursue a low probability of a high reward over a certainty of a zero reward. This would be an extrinsic rather than intrinsic preference for consistency.