Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

In Bland's discussion of regression to the mean, there are several sections which detail examples of regression to the mean, which I understand to be an unavoidable consequence of not taking the mean of several measurements per subject. In the section entitled Comparison of two methods of measurement, Bland notes research which compared measured weights with self-reported weights. The conclusions of the research were that overweight people tended to underreport their weight whereas underweight people tended to overreport their weight. Bland says this is expected due to regression to the mean.

From my understanding, this comparison paradigm is the same as that used to establish the Dunning-Kruger effect which states that those who are unskilled believe they are more skilled than they really are, and those who are highly skilled believe they are less skilled than they really are.

It seems that, if Bland is correct, the Dunning-Kruger effect is more a statistical artefact and not necessarily reality. I've had a cursory read of the papers in support of the Dunning-Kruger effect and it seems that they do indeed rely on a single measurement for both "instruments" (i.e. self-reported vs measured).

Am I missing something or is there actually strong evidence of the Dunning-Kruger effect which is robust with regard to regression to the mean?

share|improve this question

"Just an artefact" sounds like the effect could be entirely due to regression to the mean. That is a very strong statement. In fact, Kruger & Dunning already addressed this in section 4.1.3 in their original paper ("regression effect" here stands for "regression to the mean"):

Despite the inevitability of the regression effect, we believe that the overestimation we observed was more psychological than artifactual. For one, if regression alone were to blame for our results, then the magnitude of miscalibration among the bottom quartile would be comparable with that of the top quartile.

And that simply is not the case; their Figure 1 shows that the bottom performers overestimate their performance by more than the top performers underestimate theirs.

In addition, Kruger & Dunning explicitly ran additional studies (studies 3 and 4 in their original paper) to address this, and the results from their studies 3 and 4 are consistent with an actual underlying psychological effect.

Of course regression to the mean will have an effect in the basic Dunning-Kruger setup, and one could conceive a permutation test approach to estimate how much of the Dunning-Kruger setup is due to it. Nevertheless, my impression is that there is something more there.

share|improve this answer
Thanks for noting that. I've changed the title. I didn't intend to imply just, but I didn't think about the title too much. I mean perhaps mostly. – post-hoc May 25 '14 at 11:58

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.