Small-sample characterization of stochastic approximation staircases in forced-choice adaptive threshold estimation

Area of Science:

• Psychophysics
• Sensory perception research
• Quantitative psychology

Background:

• Up-down staircases are common for adaptive threshold estimation.
• Recent debates question their efficiency and usability in forced-choice tasks.
• Small-sample convergence properties of stochastic approximation (SA) staircases require investigation.

Purpose of the Study:

• To evaluate the small-sample convergence properties of stochastic approximation (SA) staircases.
• To compare the performance of SA staircases against classical up-down staircases.
• To determine optimal parameters for reliable SA staircase implementation in forced-choice experiments.

Main Methods:

• Simulation techniques were employed to analyze SA staircase convergence.
• Experimental parameters, including algorithm type, initial stimulus intensity, and step size, were varied.
• Performance was assessed based on accuracy and reliability across different psychometric function spreads and target percent-correct levels.

Main Results:

• Accelerated SA staircases, with appropriate initial stimulus intensity and step size, ensure accurate convergence.
• SA staircases demonstrate accuracy independent of the psychometric function's spread.
• SA staircases reliably target percent-correct levels away from the psychometric midpoint and outperform fixed step-size up-down methods.

Conclusions:

• Optimized stochastic approximation (SA) staircases provide accurate and reliable sensory threshold estimation.
• SA staircases are suitable for forced-choice experiments, particularly when targeting specific performance levels.
• The findings support the adoption of SA staircases for practical sensory threshold estimation.