A three-step pattern in audiometric thresholds

Use of the audiogram as the gold standard index of hearing ability amplifies the consequences of error in threshold measurement. A Markov chain model of the audiometric procedure revealed a three-step pattern in the stimuli presented each trial. Monte Carlo simulations were used to generate threshold estimates for a simple listener model. Thresholds sorted by trial had a mean bias consistent with model predictions. An alternate scoring method is proposed that uses equal sampling of Markov states. The resulting threshold targets a specific probability of detection and has no systematic bias as a function of trial.