# When is higher confidence predictive of less accuracy?

Normally, when an individual is more confident in a particular response (e.g., memory decision, general knowledge answer), he or she is also more likely to be accurate. There are also studies in which higher confidence is generally demonstrated to be unassociated with higher accuracy.

Are there any studies which demonstrate that the more confident an individual is, on average, the more likely he or she is to be incorrect?

-
As a specific example, it can be seen in the articles referenced here, confidence in driving ability appears to show no significant correlation with accuracy. – Ben Brocka Feb 22 '12 at 16:09
Perhaps in questions where there is no correct answer. All answers are insufficient, so the more confident you are the more wrong you are. – Preece Feb 22 '12 at 22:33
I might be misreading you, but you start by mentioning studies which show no correlation. So you are questioning these studies? In that case could you please reference them, so others can see which ones you are talking about and whether more recent ones are available? Would more recent studies which also show no correlation satisfy you? – Steven Jeuris Feb 23 '12 at 8:49
Hi Steven, I'm actually curious about the negative relationship, not the zero one (i.e., as is in the last sentence up there). I will link some studies, but Ben's link is one example of no relationship between confidence and accuracy. – Andy DeSoto Feb 23 '12 at 16:54

Kruger and Dunning (1999) conducted a study in which people performed a variety of 'tests' in different domains, such as logical reasoning, grammar, and humor. After taking a test, participants were asked to judge their performance relative to other test-takers, and to estimate their objective score on the test (how many questions were answered correctly).

What they found was that the subjects who performed the worst actually overestimated their performance the most. If you take their performance estimate as a proxy for confidence, then it supports your hypothesis-- but only to a relative degree. That is to say, participants who performed poorly did not estimate their ability to be better than good performers and vice versa; however, those who performed the worst were more confident in their ability. It's a bit confusing, and D&K can explain it better-- there's actually a paragraph in their article (second to last) which relates their findings to work on overconfidence.

The explanation that Kruger and Dunning give for this effect is that the metacognitive ability to assess one's performance is a correlated with actual performance on the same task. Thus, somewhat paradoxically, those who perform poorly may be unaware of their poor performance, due to their inability to predict how well they did.

Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77, 1121-1134.

-
Interesting article. I wonder if that is any more than just a result of the imperfect correlation between actual ability and perceived ability (I've often seen correlations in the .2 to .6 range), along with the inherent floor and ceiling effects of such scales. I.e., if you have low ability, you have more scope to over estimate. If you are high ability, then you have minimal scope to overestimate because you are already high. – Jeromy Anglim Feb 23 '12 at 0:43
@JeromyAnglim that was my first thought too (re: ceiling effects, regression to the mean), though they discuss this in attempt to rule out that explanation. i think there's still some room for debate, but it does have 1000+ citations, so a lot of people seem to buy into it, for whatever that's worth. – Jeff Feb 23 '12 at 0:49
Interesting. It's been a while since I've read the Kruger and Dunning article, I'll have a read and perhaps post a specific question, as I find the issue quite interesting. – Jeromy Anglim Feb 23 '12 at 1:24
Thanks! I'll have to think about this a bit more. I'm familiar with the Dunning-Kruger effect, but I hadn't yet made the connection between those findings and a confidence-accuracy correlation. Thanks. – Andy DeSoto Feb 23 '12 at 16:56