Improving calibration without training: the role of task information

Abstract

Abstract Medical students estimated probabilities that medical school applicants selected randomly from a defined population would be offered a place on the basis of information about eight characteristics of each one. Estimates were biased in favour of acceptance and this was unaffected by mere provision of base rate information. However, a first experiment showed that provision of this information reduced miscalibration after each candidate had been discussed within small groups of participants. A second experiment showed beneficial effects of providing people with the range of probability responses derived from a statistical model of how candidates' characteristics influenced whether or not they were offered a place. Numerical and graphical representations of this information both improved calibration but did so by having different effects on the distribution of participants' probability responses. Copyright © 2003 John Wiley & Sons, Ltd.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 19 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.