THE IMPACT OF AUDITORY TASKS
(AS IN HAND-FREE CELL PHONE USE)
ON DRIVING TASK PERFORMANCE

ICBC TRANSPORTATION
SAFETY RESEARCH
November, 2000

1. Background

From October, 1999 to February, 2000 the Insurance Corporation of British Columbia (ICBC) conducted closed-course driving experiments at Boundary Bay Airport (the Pacific Education Centre site) intended to test the impact of in-vehicle telephone use on driving performance.

Previous studies reported in the literature (notably Redelmeier and Tibshirani, 1997) have suggested a link between in-vehicle telephone use and driving safety, not just in terms of the physical task of operating the phone but -- most importantly -- in relation to the competing cognitive requirements of driving and conversing. With the recent proliferation of cell phones and their use by drivers to transform cars into secondary work environments, a legitimate concern has been raised as to the potential impact on crash risk due to distraction from the driving task.

Research in the area of attention -- switching has shown that task complexity plays a key role (Ranney et al, 2000; Lee et al, 2000). If the primary task (Eg. related to driving) is not very complex, then a certain amount of attention can safely be diverted to a secondary cognitive task (such as listening and responding to messages) without significantly impairing performance in the primary. For example, Noy et al (1999) found no effect of cell phone use on driver ability to maintain lateral position under normal traffic conditions.

But what about situations requiring critical choices? Some traffic situations require fairly complex and quick decision-making on the part of the drivers (making left turns through gaps in on-coming vehicle streams, for example). How might performance in such situations be affected by talking on a cell phone, and could crash risk increase as a result?

2. Research Methodology

ICBC researchers, together with engineers from MacInnis Engineering Associates limited, designed an 807m. elongated closed-course driving circuit to address the above questions. Three different driving tasks were set up. First was a commonly-encountered traffic light situation where a 3.5 second amber light was triggered from one of two vehicle positions vis-a-vis the stop line location: the closer position made it easier for the vehicle to run through the amber light (prior to the light turning red) than did the farther position. Second, offset pop-up targets were activated on one of the straight legs of the circuit in such a way as to create the need to weave between them. Two cases were presented: a short weave space and a longer space. And third, a left-turn task was set up which required the subjects to press the accelerator pedal of their appropriately-positioned but immobilized vehicle so as to indicate which gaps in an approaching vehicle stream (eight other vehicles continuously looping around the circuit) they would accept.

The subjects were exposed to activation of the first two tasks on a periodic basis as they drove at an average of about 50km/h. around the circuit. They were also required to respond verbally to taped messages which played in the vehicles and which were triggered so as sometimes to coincide with the physical driving task activations and sometimes not. Each subject completed both the signal and weave tasks 12 times while exposed to the messages and 12 times when no messages were playing. For the left-turning test, the message sequences were initiated randomly and subjects were presented with about 50 gaps during message presence and about 50 when no messages were being played.

The messages were of two types: verbal/semantic and spatial/imagery. The later required subjects to visualize the elements involved while the former required more abstract associations. In each taped message (following a brief instructional section) a criterion or contextual statement was presented (Eg. "means happiness" or "smaller than a dime"), followed by a string of target words separated by a 1.0-1.5 second gap. During this gap the subjects were required to state ("yes" or "no") whether or not the word met the criterion defined in the contextual phrase. Their responses were recorded automatically and the initial audio was activated by triggering switches installed on the test course pavement (similar to those which were used to activate the signal light and targets) so the message operation was entirely "hands-free" from the subjects' perspective.

In order to ensure that subjects always attempted both the driving and message tasks to the best of their ability and in the best possible time, they were informed that their scores in both would be tallied and their overall results would give them an equivalent number of chances to win a $1,500 draw. Since only 41 subjects were tested (30 male, 11 female, 7 aged 19-24, 25 aged 25-44, and 9 aged 45-70) the chances of winning with a high score were very good and thus the subjects should have been highly motivated to perform.

The test vehicle (a 1991 Honda Accord) was fully instrumented to record brake and acceleration pedal movement, and continuous speed over time. A number of performance-related measures relevant to each driving situation were defined (such as average deceleration/ acceleration, reaction time from event trigger to brake or accelerator pedal movement, etc.). Each of these were employed as dependant variables in a series of repeated measures analyses (using the SAS GLM proc.) where items such as driver age and gender, and weather/road condition were utilized as independent variables. The Wilk's Lambda test statistic was used as a measure of significance level (p<.05).

3. Results

The results presented below represent only those specifically concerning the presence/ absence of messages in relation to driving performance measures. Other significant main effects of such things as driver age and gender are not covered in this report.

3.1 Traffic Signal Task

Table 1 presents the significant (p<.05) main effects of message presence/absence from the repeated measures analysis.

Where subjects decided to "run" the short-trigger light (as opposed to stopping) they appeared to require a greater initial speed to precipitate the decision when the message was present than when it was not. They were more likely to stop (62.6%) when attending to the message than when not (47.9%). Also, once the decision had been made to go through the amber, the subjects accelerated harder in the message than in the non-message situation thus giving themselves more of a time cushion in avoiding the red light.

Where subjects decided to stop for the long-trigger amber light, the presence of the messages had the effect of producing lower average deceleration owing to a faster brake reaction time which also produced a longer estimated "arrival time" (the time to reach the stop line based on maintenance of the initial speed prior to braking -- also known as the "time-to-collision" or TTC -- see Tijerina, 2000).

The net effect of the messages on subject driving performance in the traffic signal task seemed to be to encourage conservative or anticipatory behaviour. This is consistent with drivers acting according to expectations concerning the primary task (Tijerina, 2000). But there was an indication at least in the case of subjects choosing to stop at the long-trigger light, that such over-compensation was not operative under wet pavement conditions. Under wet conditions the arrival time (TTC) was not longer with the messages than without; suggesting that the additional level of complexity associated with estimating brake requirements under adverse conditions may have made the subjects attention-switching strategy less effective.

3.2 Pop-up Target Task

Table 2 presents the significant (p<.05) main effects of message presence/absence from the repeated measures analysis.

The results for the target tests were very straightforward. The presence of the messages had a highly significant impact on deceleration prior to and, speed through, the weave manoeuvre. With the messages playing, subjects confronted with the more critical -- short space -- weave situation reduced their speed markedly less than they did when not having to divert attention to the auditory task. While there was no adjacent traffic flow to restrict how widely the vehicles could swerve in completing the manoeuvre, nevertheless it was apparent that the subjects in "real life' would have been dealing with a lower factor of safety for adjacent lane encroachment during the divided attention conditions.

There were no significant interactions of message presence/absence with subject/environment variables in explaining driving performance variable values.

3.3 Left-Turn Task
Table 3 presents the significant (p, .05) main effects of message presence/absence from the repeated measures analysis.

The most important outcome of the left-turn task was that subjects when exposed to the messages accepted significantly shorter gaps (both in terms of distance and time) than they did when not exposed. This translated into a shorter through-vehicle arrival time (TTC) associated with the messages which has implications for crash risk. The difference in arrival time for the trailing through vehicle would have reduced the lag or "space cushion" available between this and the turning (subjects') vehicle by over 1.5 metres -- a difference which could be important in determining whether or not a collision occurs. But it was found that the impact of message on gap acceptance and TTC was primarily related to wet pavement conditions. Under such conditions the turning lag would be reduced by fully 3.0 metres at the average through-vehicle speed, were this vehicle not to react in time.

4. Discussion

The overall effect of the messages in the cases of the traffic signal task was to produce a more conservative response on the part of subject drivers. With the message/response task activated, the drivers were more likely to stop (as opposed to running the light) than when no message was presented in the short-trigger situation where the choice was more ambiguous than it was for the long-trigger. Similarly, a higher initial speed was associated with the decision of subjects under these conditions (message on, short-trigger) to run the light than was the case when no message was playing. This can reasonably be interpreted as a conservative response if drivers under the message-on condition felt they needed more of a speed "cushion" to support their decision to run the light than was the case when no message was playing.

Similarly, when the signal light activation situation heavily favoured one choice alternative (i.e., stopping under the long-trigger situation) drivers who stopped while under exposure to the cell message tended to react earlier than when stopping and not exposed, which enabled lower average deceleration and greater "arrival time" (less chances of pedestrian "conflict"). This could imply anticipation of the event and predetermination of response strategy.

The effect of the secondary (message) task on the traffic signal response was consistent with the driver acting according to expectations concerning the primary task (as per Tijerina, 2000). The traffic signal task modelled a situation encountered by drivers dozens of times every day and for which they have undoubtedly established workable coping strategies which allow them to successfully "multi-task". It is easy to see how drivers could make an a-priori decision to favour stopping as opposed to running an amber light, in order to divert some attention to responding to a cell phone, although such a strategy cannot be taken to mean that such phone use has an inherent safety benefit.

But when the driving task moved away from the familiar and towards the more demanding, the effect of the cell message intervention on driver performance changed. In the more critical short-trigger weave situation (short spaces between targets), drivers took more time to respond in easing-up on the accelerator pedal when the messages were playing than they did under the no-message condition. They then made significantly less speed adjustments and ended-up going substantially faster through the weave manoeuvre than they did when not exposed to the messages. Thus the margin of safety in the short-weave task can be said the have been significantly reduced by the addition of the secondary "cell-phone" message task.

However, the most clear-cut impact of the messages was found in the left-turn task as had been anticipated based on the literature review. Listening and responding to the messages was associated with significantly riskier decision making (shorter accepted gaps) on the part of the subjects, and especially when the ambient test conditions reflected slippery road conditions. Drivers adjusted their gap acceptance decision making significantly for wet conditions (i.e., gaps increased in size and time) when not subjected to the message task but did not do so when attending to the messages. There was some evidence of attempted compensation for message task demand and adverse visibility in that these conditions tended to be related to acceptance of gaps in front of slightly (but significantly) lower-speed vehicles than was the case when visibility was clear and no messages were playing. However, such apparent compensation was evidently insufficient to prevent significant degradation of gap acceptance decision-making under adverse conditions as reflected in the significantly lower TTCs produced on wet pavement in the message listening/responding situation.

5. Conclusions

Listening and responding to relatively complex messages, as might occur when using a hands-free cellular telephone to conduct business or deal with important domestic issues, was found to significantly degrade driving performance in a series of driving tasks. The extent to which this degradation occurred seemed connected to the complexity of the driving manoeuvre: commonly encountered traffic signal-related choices tended to elicit conservative decision-making in the presence of the messages but the somewhat less common events of weaving and left turning demonstrated a significant negative impact of message attention.

Most importantly though, there was evidence that the problems associated with divided attention (driving and message attention/response) were exacerbated by adverse driving conditions. Attention to the secondary message task seemed to prevent the normal adjustment by drivers for potentially slippery road conditions in their decision-making.

While it was not possible to make a direct connection to crash risk from the experimental results, the nature of the driving performance degradations measured in relation to the presence of the message task clearly point to potential safety-related problems associated with such things as phone use while driving -- even if such use does not involve physical manipulation of the device.

5. References

Lee, J.D., Caven, B., Haake, S., Brown, T.L., 2000. Speech-based interaction with in-vehicle computers: the effect of speech-based E-mail on drivers' attention to the roadway. Internet Forum on Driver Distraction, NHTSA.

Noy, I., Cassidy, H., Brown, C.M., 1999. Quality of Driving with Cellular Telephones. Technical Memorandum TME 9901, Road and Motor Vehicle Regulation Directorate, Transport Canada, Ottawa.

Ranney, T.A., Mazzae, E., Garrott, R., Goodman, M.J., 2000. NHTSA driver distraction research: past, present and future. Internet Forum on Driver Distraction, NHTSA.

Redelmeier, M.D., Tibshirani, R.J., 1997. Association between cellular-telephone calls and motor vehicle collisions. The New England Journal of Medicine, 336 (7), 453-458.

Tijerina, L., 2000. Issues in the evaluation of driver distraction associated with in-vehicle information and telecommunications systems. Internet Forum on Driver Distraction, NHTSA.

Table 1

Significant Differences in Outcome Measures Related to Message Presence/ Absence During Traffic Light Tests

 

 

Condition	Initial Speed (km/h)	Deceleration (m/sec2)	Arrival Time (TTC) (seconds)	Acceleration (m/sec2)
Running short-trigger light i without message i with message	52.0 53.1			2.12 2.24
Stopping for long-trigger light iwithout message iwith message		0.341 0.324	2.66 2.73

Table 2

 

Significant Differences in Outcome Measures Related to Message Presence/ Absence During Pop-up Target Tests

Condition	Deceleration (m./Sec2)	Speed Through Targets (km/h)	Speed Reduction (Km/h)
Weaving through closely-spaced targets without messages with messages	0.365 0.328	23.4 25.9	27.8 25.2

 

 

Table 3

Significant Differences in Outcome Measures Related to Message Presence/ Absence During left-Turn Tests

Condition	Avg. Vehicle Approach Speed (km/h) (accepted gaps)	Avg. Gap Size Accepted (metres)	Avg. Gap Time Accepted (seconds)	Critical Gap (seconds)	Arrival Time (TTC) (seconds)
Without messages	53.4	57.3	4.01	3.23	4.34
With messages	52.9	55.2	3.90	3.07	4.23

[ Drive Now, Talk Later Index ]