8. Psychoacoustic Studies of Dolphins and Whales
335
absent trials. In general, stronger sensation levels are increasingly likely to
have come from the signal-plus-noise distribution rather than the noiseonly distribution.
Instead of a threshold, signal detection theory posits a criterion sensation
level. If the perceived sensation exceeds this criterion then the subject
reports "signal present." If it is less than the criterion, then the subject
reports "signal absent." The location of the criterion depends on the prior
probability of signal-present trials and on the relative payoffs for making a
signal-present response. Signal detection theory thus specifies that the
optimal response depends on the prior probability of signal-present trials,
on the magnitude of the sensation perceived, and on the relative payoffs
for making a signal-present response. The sensitivity of the subject to the
signal is represented by the parameter d' and the bias by the parameter ~.
Signal detection theory has not been used extensively in psychophysical
studies, largely for historical reasons. It is just as reasonable, for example,
to use a conventional level of d', say d' = 1.0 (e.g., Nelson 1994) in place of
a conventional threshold at 75% correct to compare studies, but this has
only rarely been used. Further, the d' measure is well suited to the method
of constants where percentage correct responses can be translated directly
into proportions of hits and false alarms, but it is difficult to apply to the
more commonly used staircase method. Because the signal magnitude
changes on every trial, it is difficult to estimate the proportions of hits and
false alarms.
Even if it is not used extensively, signal detection theory is still valuable
to our understanding of psychophysics for reminding us of the importance
of false alarms (false positive responses) and bias. An animal could, for
example, get every signal-present trial correct and still be responding with
chance accuracy. Further, animals differ somewhat in their apparent willingness to make false alarms or have been trained specifically to avoid
making them (e.g., Johnson 1967). Some animals appear to be very conservative decision makers and others appear to be more liberal. These biases
can affect the animal's measured thresholds, making comparison among
animals difficult.
1.3 Electrophysiological Psychophysics
Psychophysical methods are intended to determine the level of sensation
that derives from a stimulus of known physical magnitude. The assumption
is that these sensations derive from the pattern of functional response of
the underlying nervous system. Recent technological advances have made
it possible to examine these neurological responses more directly using
auditory evoked potentials. For example, Popov and Klishin (1998) used the
auditory brain stem response of a common dolphin (Delphinus delphis) to
measure its hearing sensitivity as a function of frequency. The auditory brain
stem response is a pattern of electrical activity emanating from the brain
335
absent trials. In general, stronger sensation levels are increasingly likely to
have come from the signal-plus-noise distribution rather than the noiseonly distribution.
Instead of a threshold, signal detection theory posits a criterion sensation
level. If the perceived sensation exceeds this criterion then the subject
reports "signal present." If it is less than the criterion, then the subject
reports "signal absent." The location of the criterion depends on the prior
probability of signal-present trials and on the relative payoffs for making a
signal-present response. Signal detection theory thus specifies that the
optimal response depends on the prior probability of signal-present trials,
on the magnitude of the sensation perceived, and on the relative payoffs
for making a signal-present response. The sensitivity of the subject to the
signal is represented by the parameter d' and the bias by the parameter ~.
Signal detection theory has not been used extensively in psychophysical
studies, largely for historical reasons. It is just as reasonable, for example,
to use a conventional level of d', say d' = 1.0 (e.g., Nelson 1994) in place of
a conventional threshold at 75% correct to compare studies, but this has
only rarely been used. Further, the d' measure is well suited to the method
of constants where percentage correct responses can be translated directly
into proportions of hits and false alarms, but it is difficult to apply to the
more commonly used staircase method. Because the signal magnitude
changes on every trial, it is difficult to estimate the proportions of hits and
false alarms.
Even if it is not used extensively, signal detection theory is still valuable
to our understanding of psychophysics for reminding us of the importance
of false alarms (false positive responses) and bias. An animal could, for
example, get every signal-present trial correct and still be responding with
chance accuracy. Further, animals differ somewhat in their apparent willingness to make false alarms or have been trained specifically to avoid
making them (e.g., Johnson 1967). Some animals appear to be very conservative decision makers and others appear to be more liberal. These biases
can affect the animal's measured thresholds, making comparison among
animals difficult.
1.3 Electrophysiological Psychophysics
Psychophysical methods are intended to determine the level of sensation
that derives from a stimulus of known physical magnitude. The assumption
is that these sensations derive from the pattern of functional response of
the underlying nervous system. Recent technological advances have made
it possible to examine these neurological responses more directly using
auditory evoked potentials. For example, Popov and Klishin (1998) used the
auditory brain stem response of a common dolphin (Delphinus delphis) to
measure its hearing sensitivity as a function of frequency. The auditory brain
stem response is a pattern of electrical activity emanating from the brain