Behavior: Basic Psychology of the User - Foundations for Designing User-Centered Systems - page 81

Table 4.2 Types of
responses to a signal

Response Signal present

Yes No

Yes Hit False alarm (FA)

No Miss Correct rejection (CR)

reporting seeing nothing where there is nothing there is called a correct rejection,
whilst reporting seeing something when it is not there is called a false alarm.
Human visual behavior often includes searching, scanning, and monitoring.
Vigilance tasks, which are a subset of these tasks, usually involve an extended time
period between the signals that the user is supposed to recognize. While it should
be possible to moderate the performance of these tasks based on the dimensions
used in the table, it is often useful to analyze these situations using Signal
Detection Theory (SDT) (Swets 1973 ; Swets et al. 1961 ; also see Wickens et al.
2014 or Wickens 2002 ).
SDT was developed to explain the task of identifying enemy planes on hard to
read radar displays. It has been applied to a wide range of classification and
decision making tasks. Swets ( 1973 ), for example, used SDT to summarize how
well doctors could classify cancer tumors from X-ray films. It can also be applied
within computer games to explain the potential problems when players search for
enemies or resources.
Figure 4.1 shows how the key parameters in SDT relate to the ability to dis-
tinguish the object of interest (signal) from distracting items (noise). The set of
signals is assumed to be normally distributed about a point some distance away
from 0, whilst the set of noise is normally distributed around 0 (indicating no valid
signal strength). d' represents the distance between the mean of the noise and the
mean of the signal, and is an inherent property of an observer and stimuli.
Observers set a threshold (here, beta, b, some authors call this lambda, k, or yet
another Greek character) as a parameter that is inherent in the observer. Obser-
vations above the threshold are classified by the observer as signal and observa-
tions below are classified as noise. Thus, the parts of the signal above the threshold
are hits and the parts of the noise distribution are false alarms (FA). Observations
below the threshold are either correct rejections (CRs) if they are noise or misses if
they were part of the signal.
Observers can often adjust the threshold to take account of the relative costs of
the four responses in Table 4.2 . The cost of false alarms and misses will influence
where the threshold is set. Where misses are costly, the threshold will be to the left,
classifying more signal and more noise as positive responses. Where false alarms
are expensive compared to misses, the threshold will be moved to the right,
eliminating more of the noise but also part of the signal.
Signal detection theory is often described in terms of sensitivity and bias. The
sensitivity (which is the same as d') refers to the separation between the signal and
noise distributions. Where the separation is small, and the two distributions