Rethinking the Dunning-Kruger effect: Negligible influence on a limited segment of the population

Abstract

Gignac and Zajenkowski (2020) recommended testing the Dunning-Kruger (DK) hypothesis with a combination of polynomial regression and LOESS regression, as the conventional approach to testing the hypothesis (i.e., quartile split) confounds regression toward the mean and the better-than-average effect. Building upon Gignac and Zajenkowski (2020), an insightful method to estimate the magnitude and prevalence of a DK effect is introduced based on comparing linear and LOESS regression predicted values. Based on simulated data specified to exhibit a plausible DK effect for cognitive abilities, the magnitude of the DK effect was empirically modeled. The effect peaked at a 20-point relative overestimation at an IQ of 55, impacting only 0.14% of the population, and decreased to a 7-point relative overestimation at an IQ of 70, affecting 2.3% of the population. Analysing two large field data samples (N ≈ 3500 each) from participants who completed intelligence subtests in grammar and logical reasoning, the DK effect was found to account for a maximum relative ability overestimation of 7 to 9 percentile points. Notably, this effect was confined to only ≈ 0.2% of the participants (IQ ≈ 55), all of whom scored at chance levels. Finally, low levels of conditional reliability (≈ 0.40) at distribution extremes were found to complicate interpreting results that superficially support the DK hypothesis. It is concluded that, when analyzed using appropriate methods, it is unlikely that the DK effect will ever be demonstrated as an unambiguously meaningful psychological phenomenon affecting an appreciable portion of the population.