Academic journal article Human Factors

The Effects of Correlation and Response Bias in Alerted Monitor Displays

Academic journal article Human Factors

The Effects of Correlation and Response Bias in Alerted Monitor Displays

Article excerpt


Many modern systems, such as nuclear power plants and advanced aircraft, are so complex that a human cannot control the system without the assistance of automation. The task demands placed on the human in such systems exceed the available resources. One common solution to this problem is to automate parts of the system. This solution can be so successful that the fundamental nature of the human's task can change from controller to supervisor. As long as the system is functioning normally, the human supervisor does not need to intervene.

An important aspect of this automation is its ability to alert the supervisor when a suspected problem occurs. Although the exact manner of detecting a problem is system-dependent, the interaction between the alarm and human supervisor can be generalized across different systems. Sorkin and Robinson (1985; see also Robinson & Sorkin, 1985) developed a model that has been reasonably successful in predicting how a human and an automated alarm interact, called the contingent criterion model.

In the contingent criterion model, the human/ automated alarm system is viewed as two cascaded signal detection systems in which the automated alarm forms the first signal detection system and the human forms the second. The human observes information that may overlap with the automated alarm's observations completely, partially, or not at all. The human's criterion depends on whether or not the automated alarm is signaling an abnormal event. The criterion will usually be more conservative when the automated detector has not signaled an abnormal event compared with when it has. Thus the human is assumed to have two criteria that are independent of the automated alarm's criterion.

As an example, Sorkin and Robinson (1985) discussed a smoke detector. The smoke detector samples information from near the ceiling; based on its information, the alarm either sounds or does not sound. At the same time, the human is sampling the air approximately 1.7 m from the floor. Because heat tends to rise, this set of information only partially overlaps with the set of information sampled by the smoke detector. In the absence of a signal from the smoke detector, the human may be conservative and require a strong smell of smoke before deciding that the house is on fire. However, if the smoke detector is sounding, the human may be more liberal and require only a weak smell of smoke before calling the fire department. Thus the criterion that the human adopts is contingent on whether or not the automated alarm is signaling an abnormal situation.

Because the automated alarm and the human are viewed as two signal detection devices, each will have a sensitivity parameter and criterion parameter(s). The sensitivity parameters are usually constrained: The automated alarm will be as sensitive as financial and technical constraints allow, and the human will be as sensitive as training, practice, fatigue, and so forth allow. However, the alarm's criterion can be easily set to any value, and the human's criteria could, in theory, be changed to practically any values given sufficient training and practice. Thus the design of automated alarms should focus on the criterion parameters that are free to vary.

Two basic questions must be answered when designing an automated alarm display. The most fundamental question is whether or not an automated alarm display will decrease the workload of the human. Although it may seem intuitive at first that an automated alarm system would automatically reduce workload, they do not always do so. Pollack and Madans (1964) found few benefits of an automated alarm display; system performance with a partially valid automated alarm was often not appreciably higher than system performance with a chance level automated alarm. In a few conditions with the partially valid alarm, the sensitivity measure of the human plus automated alarm system was appreciably lower than the sensitivity of the automated alarm system by itself. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.