Example of the confounding nature of awareness of the track in an adaptive staircase method for determining the effects of training on ILD discrimination thresholds. The learning curve represents data from one subject on a single-staircase 3D1U training paradigm. The arrow highlights the first session after the subject became aware of the adaptive algorithm. Errorbars represent one standard error of the reversal points in each session.
Example of a dual-staircase run. The first staircase (stair 1) is denoted with a gray line, and the second with a black line (stair 2). In this example, the staircases diverge after around 30 trials.
Examples of psychometric functions fitted using probit analysis for the constant-stimulus method (A) and the dual-staircase method (B). In (B), the black line and circles represent the first staircase, and the gray line and triangles represent the second staircase.
Bias in the adaptive thresholds derived from psychometric functions (A) and reversals (B) as a function of corresponding constant-stimulus score. The fits are for the non-constrained linear regression models. The crosses indicate thresholds measured for a single subject. Dashed lines indicate 95% confidence intervals for the linear regression fits. In (A), the confidence intervals intersect and are mostly below the unity line, indicating that adaptive thresholds calculated from psychometric functions were generally biased low compared with corresponding constant-stimulus thresholds obtained from the same subjects. In (B), the unity line is completely encompassed by the confidence intervals, indicating that adaptive thresholds were relatively unbiased when calculated from reversals.
Comparison of adaptive thresholds calculated from psychometric functions and from reversals for both the bias data (A) and the learning data (B). The dashed lines indicate 95% confidence intervals for the fitted functions; the confidence intervals for the learning data are so close to the function itself that they are only barely visible. For both sets of data, the fitted function lies mostly beneath the unity line, indicating that, when calculated from psychometric functions, the adaptive thresholds from both the bias investigation and the learning experiment exhibited similar degrees of bias compared with reversal-derived thresholds.
Mean absolute [(A) and (B)] and normalized percentage change [(C) and (D)] thresholds for each training method as a function of training session. In panel (B) black dashed lines indicate days for which no data were available for a given subject. The thick black line in each plot indicates the group mean for that day. The relative curvature of the mean learning curve for the adaptive data [(B) and (D)] can clearly be seen.
Pre- and post-test thresholds for all subjects. Results in each of the stimulus conditions (4 or 0.5 kHz carrier, lateralized left or right) are shown in separate panels, as indicated. Subjects trained on the 4 kHz left condition are shown in black, and untrained controls in gray. Subjects trained with constant-level stimuli (const) are shown by the circles, and subjects trained using the adaptive staircase paradigm (stair) are plotted as triangles. In both the trained 4 kHz left and untrained 0.5 kHz left conditions, the trained subjects clearly tend to cluster below the unity line, indicating that post-test thresholds were lower than pre-test thresholds. In contrast, control listeners in all three conditions, as well as trained listeners in the 4 kHz right condition, exhibit thresholds that cluster around the unity line, indicating little change in thresholds between pre-test and post-test.
Box plots showing distribution of threshold changes (post-test minus pre-test threshold expressed as % of pre-test score) for the trained and control cohorts in each of the stimulus configurations. It is clear that only the trained listeners in the 4 kHz left and 0.5 kHz left conditions showed a negative threshold change between pre-test and post-test.
Offset binaural cue values used (± indicates lateralization favoring right or left hemifield).
Article metrics loading...
Full text loading...