The use of multiple observers in signal detection theory: a method to remove the effect of unique noise from experimental data

Abstract

This thesis reports an investigation of the use of multiple observer groups in the theory of signal detectability. When an experimenter conducts a psychophysical experiment his data contain the effect of noise which is unique to each observer and can systematically affect his results. Since the experimenter cannot adequately measure this noise he cannot assess its impact on his data. The mathematical theory of multiple observers suggests that it is possible to remove the effect of this unique noise from signal detection data by using multiple observer groups. Three experiments designed to demonstrate this result are reported. The results suggest that the group-operating-characteristic (GOC) curve of Watson can, if sufficient observers are run, be used to remove the effect of unique noise from experimental data. Applications of multiple observers to the measurement of the ratio of unique noise variance to common noise variance are considered. It is shown that the criteria adopted by the observers in the group affect both the performance of the group and the estimate of the ratio of unique noise to common noise.