Alan Taylor's Thesis
Auditory psychophysics in birds:
The effects of unique noise on sensitivity
Submitted for the degree of Doctor of Philosophy in Psychology
Victoria University of Wellington, New Zealand
The performances of observers in auditory experiments are likely to be
affected by extraneous noise from physiological or neurological sources and also
by decision noise. Attempts have been made to measure the characteristics of
this noise. In particular its level relative to that of masking noise provided
by the experimenter. This study investigated an alternative approach, a method
of analysis which seeks to reduce the effects of extraneous noise on measures
derived from experimental data.
Group-Operating-Characteristic (GOC) analysis was described by Watson (1963)
and investigated by Boven (1976). Boven distinguished between common and unique
noise. GOC analysis seeks to reduce the effects of unique noise.
In the analysis, ratings of the same stimulus on different occasions are
summed. The cumulative frequency distributions of the resulting variable define
a GOC curve. This curve is analogous to an ROC curve, but since the effects of
unique noise tend to be averaged out during the summation, the GOC is less
influenced by extraneous noise. The amount of improvement depends on the
relative variance of the unique and common noise (k). Higher levels of
unique noise lead to greater improvement.
In this study four frequency discrimination experiments were carried out with
pigeons as observers, using a three-key operant procedure. In other experiments,
computer-simulated observers were used.
The first two pigeon experiments, and the simulations, were based on known
distributions of common noise. The ROCs for the constructed distributions
provided a standard with which the GOC curve could be compared. In all the cases
the analysis led to improvements in the measures of performance and increased
the match of the experimental results and the ideal ROC.
The amount of improvement, as well as reflecting the level of unique noise,
depended on the number of response categories. With smaller numbers of
categories, improvement was reduced and k was underestimated. Since the
pigeon observers made only "yes" or "no" responses, the results for the pigeon
experiments were compared with the results of simulations with known
distributions in order to obtain more accurate estimates of k.
The third and fourth pigeon experiments involved frequency discrimination
tasks with a standard 450 Hz and comparison frequencies of 500, 600, 700, 800
and 900 Hz, and 650 Hz, respectively. With the multiple comparison frequencies
the results were very variable. This was due to the small number of trials for
each frequency and the small number of replications. The results obtained with
one comparison frequency were more orderly but, like those of the previous
experiment, were impossible to distinguish from those which would be expected if
there was no common noise.
A final set of experiments was based on a hardware simulation. Signals first
used in the fourth pigeon experiment were processed by a system made up of a
filter, a zero-axis crossing detector and a simulated observer. The results of
these experiments were compatible with the possibility that the amount of unique
noise in the pigeon experiments overwhelmed any evidence of common noise.
09 Nov 2009 11:39 AM