EXECUTIVE SUMMARY

How reliable must traffic information be for motorists to trust and accept such advice? People are slow to accept and use new technology, even when the technology works reliably (Kantowitz, Becker, & Barlow, 1993). Can an in-vehicle Advanced Traveler Information System (ATIS) presenting real-time traffic information be commercially successful when some of the information it presents is incorrect?

A route guidance system is a driver decision aid that uses knowledge about a traffic network to provide advice that facilitates travel between an origin and a destination. There are many possible algorithms and heuristics to provide such support. A simple static algorithm may only calculate the path providing the shortest distance. More sophisticated heuristics might take travel times into account based upon historical data. The most powerful systems use real-time communication between the vehicle and a traffic information center to provide frequent updates on travel times and network bottlenecks. Route guidance systems can plot travel routings for the driver and some can update them if traffic conditions change or if the driver diverts from the plotted path. Thus, it is important for the system designer to be able to estimate the conditions that will maximize the probability that a driver will trust and follow ATIS suggestions.

The basic issue of the effects of traffic information reliability was first studied by Kantowitz, Kantowitz, and Hanowski (1994) using the Battelle Route Guidance Simulator (RGS), a part-task simulator that provides the driver with continuous real-time information and traffic reports. This is an improvement in methodology over earlier simulator studies that used discrete traffic images either projected from slides (Allen et al., 1991) or on a small computer (Bonsall, 1994). When traffic information was 100 percent accurate, drivers were able to reduce penalty costs associated with non-optimal route selection relative to an unreliable condition with 77 percent accurate information. However, drivers continued to use the simulated ATIS even when the system was unreliable. In this first experiment, a real existing traffic network, Seattle and its environs, was simulated. The present experiment extends and replicates these results by using three levels of information accuracy and two traffic networks.

The RGS was used to study: (1) the effects of information accuracy, and (2) familiarity of the driving environment. The simulator provided real-time traffic information and traffic video. Traffic information was either 100 percent, 71 percent, or 43 percent accurate. Drivers experienced either Seattle and its environs or an artificial setting that was topologically matched to Seattle. A total of 48 drivers was tested.

Three objective dependent variables were studied:

• Penalty cost--the amount charged drivers when they either encountered heavy traffic or selected a non-optimal route.
  
• Convergence--the agreement between the route selected and a baseline route drawn by the drivers prior to beginning their simulated journey.
  
• System query frequency--the number of times traffic information was requested.

Five subjective measures were studied:

• Trust--the degree of confidence the driver had with the system.
• Self-confidence--the degree of confidence the driver had on his own.
• Trust minus self-confidence--a measure that indicates acceptance of automation.
• Traffic expectations--what the driver expected traffic density to be on a link.
• Estimated link travel times--how long the driver thought it would take to complete a link.

Results showed for several dependent variables (e.g., penalty cost, trust, trust minus self-confidence) that while 100 percent accurate information yields best driver performance and subjective opinion, information that is 71 percent accurate remains acceptable and useful. Drivers are willing to tolerate some error in a simulated ATIS. However, when information accuracy drops to 43 percent, driver performance and opinion suffer. Thus, information accuracy above 71 percent is recommended to system designers. Future research is needed to evaluate information accuracy between 44 percent and 70 percent.

Drivers did not use simulated ATIS accurate information as effectively in the familiar setting as in the unfamiliar setting. Inaccurate traffic information was more harmful in the familiar setting. These results may imply that commercial success for in-vehicle ATIS will be easier to accomplish in unfamiliar settings, e.g., for use in rental vehicles for visitors, than in one's home city. Because drivers have greater self-confidence in familiar settings, they are more critical of ATIS advice and hold to a higher standard of user acceptability when they know the area geography. Thus, to achieve user acceptance, in-vehicle systems intended for purchase by a driver in a private passenger vehicle will likely have to meet higher standards than systems intended for commercial use.

Driver trust was decreased by inaccurate traffic information but recovered when accurate information was received. However, the more likely that information was inaccurate, the less the recovery. For the 43 percent accuracy condition, trust minus self-confidence became negative, implying that drivers would prefer their own solutions to those offered by the simulated ATIS device. So, although drivers do not demand perfect traffic information, high degrees of inaccuracy will cause drivers to ignore system advice, especially in familiar settings.