Raja Parasuraman Catholic University of America
P. A. Hancock Liberty Mutual Insurance Co.
Alarms and warning systems are proliferating with the rapid pace of automation at work and in the home ( Parasuraman & Mouloua, 1996). Alarms are meant to signal threats or emergencies that demand an immediate response. But often they turn out to be false. Excessive false alarms abound in civil ( Patterson, 1982) and military aviation ( Tyler, Shilling, & Gilson, 1995), nuclear power plants ( Seminara, Gonzalez, & Parson, 1977), and other high-technology industries ( Pate'-Cornell, 1986). As a result, people tend to ignore warning systems much of the time. Often they may turn off an alarm before verifying that it was indeed false. Can warning systems be designed so as to be useful? We describe a quantitative, analytical solution to this problem that provides step-by-step procedures for designing effective warning systems.
We describe the problem and the proposed solution in the context of collision-warning systems for motor vehicles ( Parasuraman, Hancock, & Olofinboba, 1997), although our analysis applies to any warning system. Rear-end collisions constitute approximately 25% of all traffic accidents that are reported to the police ( Knipling et al., 1993). Inattention is one of the leading causes ( Wang, Knipling, & Goodman, 1996). Collision-warning systems have been developed in response to this problem. Proponents claim that such systems can alert inattentive motorists in time to take evasive action.
Numerous collision warning system systems are currently being developed (ITS America, 1996). Several have already entered the market. An example is the Forewarn™ system by Delco Electronics. This system uses very high-frequency microwave object-detection sensors to measure the position and relative speed of objects ahead of (77 GHz) and behind (24 GHz) the vehicle. Caution and emergency alarms are issued at driver-preset distances. Alarms are visual (icons on a head-up display), auditory (audio system), or tactile (pulsed vibration of the brakes).
What has fueled the explosive growth of collision-warning systems? Clearly, the annual toll on the road of approximately 40,000 fatalities and more than 6 million collisions ( National Safety Council, 1996) represents a compelling problem in need of a solution. Unfortunately, perfect detection of impending collisions cannot be achieved.
Although many apparently accurate systems have been fielded, a crucial fact that is neglected is that collisions, particularly ones that cause loss of life or disabling injury, are very rare events. The frequency of impending collisions, or those that are avoided by appropriate action on the part of the driver, is generally higher than that of actual collisions. Nevertheless, the base rate or prior probability of either actual or impending collision events is likely to be very low. Collision-warning systems are designed to detect a subset of these events, that is, those scenarios that would result in a collision if the driver did not take appropriate action. Thus, for most drivers, the base rate of traffic events leading to impending collision is likely to be quite low. This fact has major implications for the design and implementation of warning systems, as we demonstrate analytically.