Slope bias of psychometric functions derived from adaptive data

Area of Science:

• Psychophysics
• Statistical analysis
• Perceptual science

Background:

• Adaptive procedures are widely used for estimating sensory thresholds.
• Existing methods often exhibit an unexplained slope bias in threshold estimation.
• This bias affects various fitting techniques, including maximum-likelihood and Spearman-Kärber analysis.

Purpose of the Study:

• To demonstrate and explain the phenomenon of slope bias in adaptive psychometric procedures.
• To introduce a novel adaptive staircase procedure for direct measurement of both threshold and slope parameters.

Main Methods:

• Analysis of parametric and nonparametric maximum-likelihood fits.
• Application of Spearman-Kärber analysis to adaptive data.
• Development and testing of a new adaptive staircase procedure using double trials.

Main Results:

• Slope bias was demonstrated across multiple standard adaptive procedures (staircase, stochastic approximation).
• The bias was explained by serial data dependency inherent in adaptive procedures.
• The novel adaptive staircase procedure successfully measured both threshold and slope directly.

Conclusions:

• Serial data dependency is the cause of slope bias in adaptive threshold estimation.
• The newly developed adaptive staircase procedure offers a more comprehensive approach by estimating both threshold and slope.
• This advancement has implications for more accurate psychometric function fitting.