“The Perfection Premium”, 2020-09-10 (; backlinks; similar):
This research documents a perfection premium in evaluative judgments wherein individuals disproportionately reward perfection on an attribute compared to near-perfect values on the same attribute.
For example, individuals consider a student who earns a perfect score of 36 on the American College Test to be more intelligent than a student who earns a near-perfect 35, and this difference in perceived intelligence is substantially greater than the difference between students whose scores are 35 versus 34. The authors also show that the perfection premium occurs because people spontaneously place perfect items into a separate mental category than other items. As a result of this categorization process, the perceived evaluative distance between perfect and near-perfect items is exaggerated. Four experiments provide evidence in favor of the perfection premium and support for the proposed underlying mechanism in both social cognition and decision-making contexts.
[Keywords: perfection, categorization, numerical cognition, social cognition]
…In four experiments, we find that even when the objective numerical gap between two values is equal, people perceive the difference between individuals and items to be greater if one has a perfect attribute value or rating. For example, the perceived difference in intelligence of two students scoring100% versus 99% on an exam exceeds the perceived gap between students scoring 99% versus 98%, even though the scores differ by 1% in both cases.
[Part of this is just a ceiling effect: if one hits the ceiling on a test by scoring a perfect score rather than falling short slightly, that represents a lower bound—the person scores at least that high, and so likely scores higher, and if the test is not an extremely good one, then potentially arbitrarily much higher.
For example, if someone scores 128 on an IQ test with a ceiling of 130 (+2SD), another 129, and another scores the max of 130, then the expected scores are 128/129/136, and the expected differences are not 1/1/1 but 1/1/7. (You can calculate the truncated normal expectation using
truncNormMean(2)in my dog cloning page).]
View PDF: