ABSTRACT
Techniques aimed at manipulating both sensitivity and response bias in absolute identification of loudness are explored in order to illuminate features of two Thurstonian-based models: the range theory of Braida and Durlach and the attention-band theory of Green and Luce. Evidence is first offered that signal range affects both the sensory and the criterion variance in the Braida-Durlach model. We then turn to studies of more local changes in sensitivity and response bias by means of experimental manipulations in which overall range is held constant.
One-trial sequential analyses strongly support the idea of shifting category boundaries, although we lack a simple rule to characterize the nature of these shifts as a function of experimental design. Changes in d' as a function of the relation of the present to the previous signal are, at best, minor. These one-trial sequential analyses do not support the subsidiary hypothesis of the Green-Luce theory that attention tracks the previous signal on a trial-by-trial basis.
A number of additional studies, all using some form of prolonged signal clustering, do however generate substantial local changes in sensitivity. This phenomenon can be interpreted as evidence of an attention mechanism that is somewhat sluggish in tracking signals. The experiments are of three types: probability clustering in which nonuniform distributions of trial independent schedules are used; intensity clustering into narrow (e.g., 10 dB) intensity bands but with outlier signals to maintain the range; and clustering that arises because of sequential dependencies within the signal presentation schedule. All methods provide evidence of substantial changes in d' when a signal is part of a cluster as compared to when it is not. Several alternative accounts of these phenomena are explored and rejected.
