Different adaptive-threshold algorithms yield different threshold estimates
Evidence
The final MAA threshold estimate was calculated based on the median of the last three lower reversal points during a run. On average, this method provides a threshold value, which is lower than that produced by e.g. Levitt ( 1971 ) or Neher et al. ( 2011 ). However, data inspection indicated that both of these alternate methods could lead to misleading results if a test subject had been performing unreliably during parts of the run. — Dijk, Pim van, Physiology, Psychoacoustics and…, 3052
Categorization
Category: Craft and Method
- SEED: Angle for an article or course note: how the choice of threshold estimator in psychophysics quietly changes the numbers reported — a methodology-caveat piece.
Related
- Median-based MAA thresholds differ from Levitt-style staircase estimates (similarity 0.89)
- Median-of-reversals MAA estimation is more robust than Levitt when subjects perform unreliably (similarity 0.88)
- Median-of-reversal-point threshold estimates are more robust to unreliable runs (similarity 0.88)
- Adaptive staircase averaging of lower reversal points yields lower MAA thresholds than Levitt’s method. (similarity 0.84)
- Adaptive threshold methods trade fast converging results against convergence risk (similarity 0.81)
- Different staircase-analysis methods yield different minimum audible angle estimates (similarity 0.81)
- The 3-down one-up adaptive rule tracks 79.4 percent correct thresholds (similarity 0.81)
- One-up-two-down adaptive staircases can report lower thresholds than other estimation methods (similarity 0.81)
- Accuracy and function-shape information trade off against speed in threshold measurement (similarity 0.80)
- Unconverged adaptive measurement runs are excluded as unusable data (similarity 0.79)
- MAA thresholds are measured with 3-AFC staircases tracking the 70.7% point (similarity 0.79)