Impact of incomplete feedback on response bias

The goal of most laboratory studies is to produce a bias-free estimate of sensitivity. Forced-choice procedures with trial-by-trial feedback, equal a priori probability of signal occurrence, and a balanced pay-off matrix were designed to assess listener sensitivity with response bias forced to zero. (Green, 1960; Green and Swets, 1964; 1988). Real-world detection tasks rarely provide N-interval forced-choice opportunities. The probability of signal occurrence may vary from near zero in vigilance tasks to near 100% for some clinical testing. Further, many real-world listening situations do not lend themselves to trial-by-trial feedback. Davis (2015, 2017) conducted a single interval yes-no, tone-in-noise experiment measuring response bias with incomplete feedback. Liu (2020) used Bayesian analysis to account for participant variability in the Davis data. Results from different subjects were combined to draw inferences about the effect of each experimental condition on listener bias. As expected, complete feedback drives the response criteria toward the unbiased point, and incomplete feedback conditions result in various degrees of deviation from the optimal criterion. Critical incomplete feedback conditions included responses on signal trials, “yes” responses, and correct responses. Implications for the design of experiments intended to approximate listening conditions in realistic detection tasks are discussed.