8. Psychoacoustic Studies of Dolphins and Whales
333
trials use the same magnitude and the subject's performance on these trials
is calculated as a percentage correct or as the signal-detection measure d'
(see below). A different magnitude is then selected (often randomly) for
the next block of trials, and so forth. Plotting the changes in percent correct
or d' as a function of signal magnitude produces a psychometric function.
The threshold is then chosen to be some conventional level of accuracy,
often either 50% correct or 75% correct. By comparison, the staircase
method yields a level corresponding roughly to 71 % correct. Curve-fitting
techniques are required to interpolate the threshold magnitude of the signal
when none of the tested magnitudes actually yields the threshold accuracy.
A cumulative normal ogive is typically used for this purpose.
Whichever psychophysical method is used, the animal can indicate its
decision either by making one explicit response on signal-present trials and
a different explicit response on signal-absent trials, or by responding explicitly on signal present trials and refraining from responding on signal absent
trials. A two-response measure with dolphins, for example, might require
the animal to touch one paddle on the left side of the animal when the signal
has been detected and to touch a different paddle on the other side when
no signal has been detected. A single response measure might require the
animal to place its head in a hoop listening station to start a trial and then
to back out of the hoop and touch a paddle when the signal has been
detected, but to remain in the hoop if no signal has been detected. This procedure is called a go/no-go technique because the subject goes to respond
on signal-present trials, but does not go on signal-absent trials. Other
responses, including vocalizations, have also been used with this procedure.
1.2 Signal Detection Theory and the Threshold
Although widely used in psychophysical studies, the notion of a threshold
is itself problematical. It implies a sharp transition between sensation and
no sensation as signal magnitude decreases. It further implies that the perceiver is in exactly one of two states on any given trial. Either the observer
detects the signal if its magnitude is sufficient, or it fails to detect the signal
and is thereby unaffected by it. Furthermore, the data produced by psychophysical experiments rarely correspond to the all-or-nothing pattern
predicted by traditional threshold theory. Rather, the probability of detection tends to increase as a continuous ogive function of the magnitude of
the stimulus. An ogive, or S-shaped curve, could result from a threshold
detector if there is some variability in the detector (e.g., if the threshold is
normally distributed around some mean value), in the signal, or in the
amount of "noise" present in the environment. On this argument, the ogive
results from averaging a set of thresholds that vary about some value.
A more compelling reason to suspect the validity of the threshold notion
is the finding that expectancy and payoff can have very powerful effects on
the detections reported by observers. Some observers may be biased to
333
trials use the same magnitude and the subject's performance on these trials
is calculated as a percentage correct or as the signal-detection measure d'
(see below). A different magnitude is then selected (often randomly) for
the next block of trials, and so forth. Plotting the changes in percent correct
or d' as a function of signal magnitude produces a psychometric function.
The threshold is then chosen to be some conventional level of accuracy,
often either 50% correct or 75% correct. By comparison, the staircase
method yields a level corresponding roughly to 71 % correct. Curve-fitting
techniques are required to interpolate the threshold magnitude of the signal
when none of the tested magnitudes actually yields the threshold accuracy.
A cumulative normal ogive is typically used for this purpose.
Whichever psychophysical method is used, the animal can indicate its
decision either by making one explicit response on signal-present trials and
a different explicit response on signal-absent trials, or by responding explicitly on signal present trials and refraining from responding on signal absent
trials. A two-response measure with dolphins, for example, might require
the animal to touch one paddle on the left side of the animal when the signal
has been detected and to touch a different paddle on the other side when
no signal has been detected. A single response measure might require the
animal to place its head in a hoop listening station to start a trial and then
to back out of the hoop and touch a paddle when the signal has been
detected, but to remain in the hoop if no signal has been detected. This procedure is called a go/no-go technique because the subject goes to respond
on signal-present trials, but does not go on signal-absent trials. Other
responses, including vocalizations, have also been used with this procedure.
1.2 Signal Detection Theory and the Threshold
Although widely used in psychophysical studies, the notion of a threshold
is itself problematical. It implies a sharp transition between sensation and
no sensation as signal magnitude decreases. It further implies that the perceiver is in exactly one of two states on any given trial. Either the observer
detects the signal if its magnitude is sufficient, or it fails to detect the signal
and is thereby unaffected by it. Furthermore, the data produced by psychophysical experiments rarely correspond to the all-or-nothing pattern
predicted by traditional threshold theory. Rather, the probability of detection tends to increase as a continuous ogive function of the magnitude of
the stimulus. An ogive, or S-shaped curve, could result from a threshold
detector if there is some variability in the detector (e.g., if the threshold is
normally distributed around some mean value), in the signal, or in the
amount of "noise" present in the environment. On this argument, the ogive
results from averaging a set of thresholds that vary about some value.
A more compelling reason to suspect the validity of the threshold notion
is the finding that expectancy and payoff can have very powerful effects on
the detections reported by observers. Some observers may be biased to