Learning to calibrate age estimates
- Jordan Suchow, Stevens Institute of Technology, Hoboken, New Jersey, United States
AbstractAge is a primary social category and, with little effort, we can quickly approximate it from photographs. Here, we analyze 1.5 million age judgments derived from a popular online website where participants estimate the age of a person depicted in a photograph, with feedback. We find that median age judgments across participants are linear in the actual age, with little bias. However, the slope is considerably less than one, such that the aggregate overestimates the age of younger people and underestimates the age of older people. Age estimates are found to be unbiased at 37.5 years, which coincides with the median age across all the depicted persons. These results are consistent with an account in which, over time, participants learn to calibrate an analogue magnitude to the learned distribution of encountered ages, combining photographic evidence with distributional information to arrive at an estimate that balances the two.
Return to previous page