Ask The Experts

In all science, measurement is the process of assigning numbers to objects in a systematic manner. The scientist interested in measurement must always answer two questions:

1. (representation problem) How is the assignment of numbers to objects justified?
2. (uniqueness problem) To what degree is this assignment unique?

Reliability is an index of the consistency of a measure and addresses the representation problem. Validity is an index of the truth of a measure and is related to the uniqueness problem. Good measures are both reliable and valid.

Good research must also be generalizable. This means that results can correctly be applied to real-world systems. Generalizability depends upon three factors: subject representativeness, variable representativeness, and setting representativeness (see Kantowitz, 1992 for detailed explanations of these terms.) We can't guarantee that a measure, even if reliable and valid, will work properly unless it is observed in a research setting that is generalizable.

Without getting bogged down in technical details (see Kantowitz, 1992 if you want to slog through details), the best way to select a measure that will work is to be guided by theory. It is poor science to select a measure just because it is easy to obtain. It is almost impossible to select a single measure that captures all the essential characteristics of a complex system, such as a driver in a vehicle. Theory must be used to select a set of measures that are useful and appropriate.

Selecting Measures for Driver Distraction

It might seem that the best way to measure driver distraction would be simply to ask drivers if they were distracted by some event. This is called obtaining a subjective opinion. We can make this process appear even more scientific by asking the driver to rate (perhaps on a five-point scale from 1-5) how distracted they were. This is called a rating scale. Unfortunately, people are not always able to give subjective ratings in a consistent manner (see Nygren, 1991). Even with a lot of fancy statistical treatments, it can be difficult to interpret subjective ratings. They are used because they are easy to obtain and because sometimes they can be correlated with better measures of distraction.

The best measures are objective rather than subjective. This includes measures of how the vehicle is located on the roadway, how hard the driver is pushing on the brake pedal, and how long it takes the driver to react to a signal. Physiological measures are also objective but they are best for determining long-term states of the driver, such as fatigue, rather than specific reactions to particular signals.

Since distractions are related to driver attention, theories of attention can help us select the best measures. An important class of measures require the driver to perform another task, called a secondary task, while driving (Kantowitz & Simsek, 2000). If the driver is distracted, there is less attention available to perform the secondary task. So objective performance on the secondary task can be interpreted, using a theory or model of attention, as an index of driver distraction. For example, a secondary task might require a driver to push a button on the steering wheel when an auditory tone is heard inside the vehicle. The time from the onset of this tone until the driver pushes the button, called reaction time, would be a measure of distraction. If reaction time is high, the driver was distracted when the tone came on. If reaction time is low with a rapid response to the tone, we can rule out distraction.

However, there is no unique secondary task for measuring driver distraction. Many secondary tasks have been studied and several are useful (Kantowitz & Simsek, 2000). Some typical secondary tasks would include memorizing telephone numbers, doing mental arithmetic and pressing buttons when signals are presented inside the vehicle. But most of these secondary tasks are scored either by reaction time or by proportion of correct responses. So the best measures of driver distraction are time and/or correct responses provided a secondary task has been selected that meets the criteria explained in the first section of this answer.

Conclusion

There is no single best measure of driver distraction. Objective measures are better than subjective measures. Secondary-task measures of driver distraction offer the best opportunity for success because they can be related to theories of attention. Even so, it is not simple to select the most appropriate secondary task.

References

Kantowitz, B.H. (1992) Selecting measures for human factors research. Human Factors, 34, 387-398.

Kantowitz, B.H. & Simsek, O. (2000, in press) Secondary-task measures of driver workload. In P. Hancock & P. Desmond (Eds) Stress, workload and fatigue. Mahwah, NJ: Erlbaum.

Nygren, T.E. (1991) Psychometric properties of subjective workload measurement techniques: Implications for their use in the assessment of perceived mental workload. Human Factors, 33, 17-34.

(Answered by Barry  Kantowitz, UMTRI)

A. The Indiana based, "Tri-Level Study of the Causes of Traffic Accidents" published by NHTSA in 1975 remains one of the classic works in attempting to define causal factors in crashes. It tells us that about 90% of crashes include human factors as direct causes. Of these, approximately 50% were characterized as recognition errors, 40% as decision errors, and 10% as performance errors. These factors were derived from detailed analyses of crashes investigated by police and by trained in-depth crash investigators. Analysts were drawn from several disciplines. To my knowledge, the level of detail captured in this study has never been replicated.

Unfortunately, the Tri-level Study was conducted long before the current plethora of in-vehicle technologies were developed. Still, the report cites driver inattention, internal distraction, improper lookout and excessive speed among the most prevalent causal factors.

The more recent 1997 NHTSA report, "An Investigation of the Safety Implications of Wireless Communications in Vehicles" examines current databases for indications of technology-use based causal factors in crashes. As explained in my testimony at the Public Meeting, these databases rely heavily on police accident reports to recognize the use of cell phones (and other devices) as pre-crash factors. Given the widespread use of small, easily concealed, handheld phones, it is extremely difficult for law enforcement personnel to detect such use in the absence of witness statements or other physical evidence. Because cell phone use is not illegal, there is little incentive for officers to inquire about, or to note such use on their reports. The introduction of other devices such as fax machines and navigational aids is so recent, that a body of data (even of poor data) has not yet been developed.

Police reports will never be able to adequately assess technology use as a causal factor. Highway safety researchers face the same challenges, and generally conduct their investigations days after the crashes occur. A crash investigation-generated statistical basis for safety decisions regarding in-vehicle devices will always be lacking the required rigor. None of the other nations which have passed laws regulating the use of in-vehicle technologies did so on the basis of statistics.

For those few crashes in the FARS and NASS data for 1996-1997 which were determined to be technology related, the citations issued to recognized cell phone-using drivers were primarily for inattention, failure to yield, run off the road, and excessive speed. For the in-depth investigations conducted by Dynamic Science in support of the report, the overriding factor was driver inattention.

Clearly then, driver inattention is a recognized and significant factor in highway crashes. The question then becomes, "What causes driver inattention?" Any driver can tell you that there are many causes - roadside activities, crying children, handling CDs, eating, drinking, shaving, whatever humans can invent.

Current NHTSA sponsored databases indicate that about 30% of crashes are caused by driver distraction. I am not familiar with comparable data from other countries, and refer you to the National Center for Statistics and Analysis and the Bureau of Transportation Statistics.

In Japan, a one-month study of cell phone use by drivers was conducted by police in June of 1996, prior to the adoption of their law banning hand held phone use. They studied 129 crashes and determined that drivers were generally dialing a phone or responding to a call at the times of their crashes. This would indicate that biomechanical distraction (handling the phone) is a serious issue in Japan. Both crash investigation and human factors data in the U.S. show that it is the cognitive distraction of being involved in conversation that constitutes the greatest risk for drivers.

The question of potential benefits of in-vehicle, distraction-related technologies is of great interest at this time. The cell phone industry and the law enforcement community tout the benefits of immediate emergency notifications. Such calls can and should be made from a stopped vehicle, which makes the issue of driver distraction a moot point. The human factors research cited in the 1997 report includes one study that indicated that conversation may help offset fatigue among professional truck drivers. It certainly can be argued that rest is the best cure for driver fatigue, and adding a recognized cognitive distraction to an impaired drowsy driving situation may be a poor solution. In fact, a great deal of attention is focused on fatigued commercial vehicle drivers, and I have not heard anyone suggest that we should issue cell phones to such drivers to improve their performance.

The merits of other in-vehicle technologies such as navigational devices, and night vision systems will have to be judged based upon human factors studies - at least for the short term. It takes years to be able to develop a statistically reliable crash data set for emerging technologies of any kind as we have seen from recent experience with air bags and antilock brakes. But the absence of statistics should never be used as an excuse for inaction when a problem has been recognized. Cell phones are not essential devices for driving. In fact, in my opinion, they are an unnecessary and dangerous source of driver distraction. Our first priority must always be safety. The design and development of new technologies must not be driven by profit, or even by convenience. The devices must be shown to at least not degrade driving performance if they cannot be shown to enhance driving safety.

(Answered by Frances  Bents, Dynamic Sciences, Inc.)

Alternatives to Near Misses: Braking Time & Unsafe Distances

If a long period of time is not practical for the evaluation, then a short duration on-road evaluation in an instrumented vehicle or a driving simulator could be used. The data from such an evaluation, however, include the effects of learning to use both the vehicle and the in-vehicle ITS, of being watched, and of performing contrived driving scenarios. For simulators, there are also fidelity issues to consider. Data from this method include: obstacle avoidance and lane maintenance. Obstacle avoidance is measured in two ways: braking time and occurrence of unsafe distances. Olson and Sivak (1986) measured the time from the first sighting of an obstacle until the accelerator was released and the driver contacted the brake. Their data were collected in an instrumented vehicle driven on a two-lane rural road. Drory (1985) used the same measure in a simulator to evaluate the effects of different types of secondary tasks. Burger, Smith, Queen, and Slack (1977) used the brake reaction time distance between the cohort vehicle and the subject driver's vehicle. In addition they also calculated the minimum area surrounding a vehicle that should have been clear of other vehicles at the initiation of a specific maneuver and through the completion of the maneuver. This measure is similar to near misses described previously. To simplify the analysis in a later study, Burger, Mulholland, Smith, Sharkey, and Bardales (1980) used 60-foot criterion for gaps during lane changes. More recently, Korteling (1994) used car-following performance distance. In a series of on-road tests at Veridian, vehicle decelerations greater than 0.4 g were used to indicate unsafe following behavior.

Measuring Lane Maintenance

The risk of lane infringement and run-off-the-road accidents has been inferred from lane exceedances. This measure has already been used to evaluate in-vehicle ITS. For example, based on findings in a study of the safety aspects of Cathode Ray Tube (CRT) touch panel controls in automobiles, Zwahlen, Adams, and DeBald (1988) stated, "the probabilities of lane exceedence during the operation of a CRT touch panel (driving at 40 mph, along a straight, level, smooth roadway; under ideal driving conditions) are 3% and 15% for lane widths of 12 feet and 10 feet, respectively, which are unacceptable from a driver safety point of view." Summala, Nieminen, and Punto (1996) used lane exceedances to evaluate location of a display in an automobile cockpit. Imbeau, Wierwille, Wolf, and Chun (1989) reported that the variance of lane deviation increased if drivers performed a display reading task. The data from both these studies were collected in a driving simulator. A similar measure, Time-to-Line-Crossing (TLC), was developed to enhance preview-predictor models of human driving performance. TLC equals the time for the vehicle to reach either edge of the driving lane. It is calculated from lateral lane position, the heading angle, vehicle speed, and commanded steering angle (Godthelp, Milgram, and Blaauw, 1984). Godthelp (1986) reported, based on field study data, that TLC described anticipatory steering action during curve driving.

Eye Glance Measures

When data can be collected in only a single car and only on the driver (not the vehicle), glance behavior has been used to infer safety impacts. Glance duration has long been used to evaluate driver performance. For example, in an early study, Mourant and Rockwell (1970) analyzed the glance behavior of eight drivers traveling at 50 mph on an expressway. As the route became more familiar, drivers increased glances to the right edge marker and horizon. While following a car, drivers glanced more often at lane markers. Burger, Beggs, Smith, and Wulfeck (1974) discussed the importance of considering long-duration glances away from the forward scene during safety evaluations and suggested using 2.00 sec as the definition of a long-duration glance. In research more relevant to evaluating the safety impacts of in-vehicle systems, Zwahlen, Adams, and DeBald (1988), cited previously, investigated the eye scanning behavior when driving in a straight path while operating a simulated CRT touch panel display (radio and climate controls). Similarly, Imbeau, Wierwille, Wolf, and Chun (1989), also cited previously, used time glancing at a display to evaluate instrument panel lighting in automobiles. Not unexpectedly, higher complexity messages were associated with significantly longer (+0.05s more) glance times. Kurokawa and Wierwille (1991) found, in a study of control label abbreviation effects, that labels could produce small but reliable reductions in number of glances to the instrument panel. Fairclough, Ashby, and Parkes (1993) used glance duration to calculate the percentage of time that drivers looked at navigation information (a paper map versus an LCD text display), roadway ahead, rear view mirror, dashboard, left-wing mirror, right-wing mirror, left window, and right window. Data were collected in an instrumented vehicle driven on British roads. The authors concluded that this "measure proved sensitive enough to (a) differentiate between the paper map and the LCD/text display and (b) detect associated changes with regard to other areas of the visual scene" (p. 248). These authors warned, however, that reduction in glance durations might reflect the drivers' strategy to cope with the amount and legibility of the paper map. These authors also used glance duration and frequency to compare two in-vehicle route guidance systems. The data were collected from 23 subjects driving an instrumented vehicle in Germany. The data indicate, "as glance frequency to the navigation display increases, the number of glances to the dashboard, rear-view mirror and the left-wing mirror all show a significant decrease" (p. 251). Based on these results, the authors concluded, "Glance duration appears to be more sensitive to the difficulty of information update. Glance frequency represents the amount of. "Visual checking behavior" (p. 251).

Differences Between Simulator and On-Road Driver Performance

Olson and Sivak (1984), cited previously, used both laboratory and field studies to evaluate the effects of glare from rearview mirrors on driver performance. The laboratory study implied a reduction in seeing distance of 50% but, in the field study, the loss even at the highest glare level was only 15%. Korteling (1990) used the RT of correct responses and error percentages to compare laboratory, stationary, and on-road driving performance. RTs were significantly longer in on-road driving than in the laboratory.

Summary

If near misses cannot be collected then the following measures have been used to infer safety impact: braking time, distance to following vehicle, distance to obstacle, vehicle deceleration, probability of lane exceedence, and glance duration. If comparative data (i.e., in-vehicle ITS present versus absent) cannot be collected, then the following criteria have been used to infer safety impact:

• Braking time less than the time required to brake prior to hitting the obstacle
• Distance to following vehicle, less than braking distance
• Distance to obstacle, less than braking distance
• Vehicle deceleration, greater than 0.4 g
• Probability of lane exceedence, less than 3% for 12 foot lane and 15% for 10 foot lane
• Glance duration, less than or equal to 2 seconds

(Answered by Valerie  Gawron, Veridian Engineering)

As a result of these investigations, it was concluded that although talking on the phone still caused accidents, the majority were caused by trying to pick up the call and secondly trying to place a call. The risk would be greatly reduced if the phones were to be hands-free, so the National Police Agency decided to put a ban on using the phone (or any hand held transmission device) with the exception of hands-held phones. A very good article describing the National Police Agency's ban can be found at the following link (http://www.drivers.com/cgi-bin/go.cgi?type=ART&id=000000273&static=1)

An extensive campaign on national TV, radio and newspapers preceded the ban that began November 1999, so it is safe to assume that it would be difficult to make excuses as to not having known of such a ban. The National Police Agency did a survey for the first month (i.e.; November 1999) and compared this with the month before (October 1999) and the same month the year before when there was no ban in place. Results found that in the month after the revised Road Traffic Law went into effect, the number of traffic accidents caused by drivers using cellular phones that resulted in fatalities or injuries fell by about 75 percent. Another survey was conducted for the half year from November 1999 to May 2000 and compared that with the same period in the previous year. The agency revealed that in this first 6-month-period, when the use of cellular phone while driving was banned, the number of accidents involving the use of cellular phones decreased by 60%.

My guess is, not everyone changed over to a hands-free phone, although there was an increase in demand for these devices. My personal view for reasons that accidents went down are:

1. Since most drivers knew it was against the law to use a hand held phone, they just simply refrained or only used it in very restricted instances.
2. Knowing it was against the law, when they did use it, they used it very carefully, which helps a lot.
3. In reporting to police, excuses such as, "I was using the phone" no longer seemed appropriate.

I would view that in Japan, with the statistics as those in 1997 and 1998, the decrease in accident rate compared to before the ban will stabilize at around 40%. Of course, the statistics cited above apply to Japanese drivers, and since the traffic situation and the way phones are used in Japan and in the US is quite different, the same statistics may not generalize to the US.

(Answered by Hiroshi  Tsuda, Nissan)

(Question submitted to Michael Goodman, and Response prepared by Thomas Ranney, Transportation Research Center; and Elizabeth Mazzae, NHTSA)
(Answered by Michael  Goodman, NHTSA)

A. First, a working assumption: a 'properly-trained' driver is one who has learned strategic (trip planning), tactical (situational awareness), and operational (vehicle maneuvering) skills to criterion levels not attainable by a 'poorly-trained' (or untrained) driver.

Next, one's concentration on 'the act of driving,' as exemplified by where one directs one's attention, how quickly and appropriately one responds to safety threats, etc., can reasonably be expected to change with experience, as specific behaviors are reinforced in some situations but not in others. Slowing down and checking carefully to the sides as one approaches an intersection where sight distance is limited by a structure, vegetation, etc., is reinforced often enough so that this training lesson sticks. (The partial reinforcement schedule for such behavior in fact makes it extremely likely to persist, to the motorist's advantage.) An untrained driver who happens to behave in this manner is similarly reinforced, of course. Thus, to the extent that a novice driver is 'properly' trained, the initial months or years of driving should be characterized by superior allocation of attention (i.e., looking where you should, when you should) relative to an untrained driver who must (hopefully) learn the same lessons through trial and error.

The differences in how effectively drivers attend to potential hazards (as well as their susceptibility to distractions) as a function of training may not be so evident over time, however. Some hazards manifest themselves very infrequently, such as trains encountered at at-grade crossings. As a result, slowing down sufficiently to effectively check to the sides before crossing the tracks may be reinforced so rarely that the 'properly trained' driver behaves no more safely than the untrained driver after some time. This may not be exactly what the question implied, by "susceptibility to distractions," though.

On this score, it is important to remember that training can have a strong impact on what a driver CAN do, but does not necessarily determine what he WILL do. An individual who has received relatively more extensive driver training may be expected to more rapidly find, understand, and react appropriately to the most safety-critical information in a given situation than an untrained or poorly trained individual. Training teaches drivers what to expect in the way of potential hazards, so they may be anticipated and recognized sooner, and responded to more effectively. This gained efficiency in visual search, except in extremely high demand situations (e.g., high-speed, high-volume traffic; or adverse weather conditions), will result in 'spare capacity.'

That is, while the untrained (especially novice) motorist is likely to experience the driving task as sufficiently demanding that his or her full attention is required to perform it, the highly-trained driver will perceive the difficulty of the driving task as being easier-even routine--especially when driving on familiar routes. And with this perception that one's full attention is not necessary to meet the demands of the driving task, the susceptibility to distraction increases.

This does not suggest that training is unnecessary or counterproductive. With experience, the same perceptions of spare capacity evolve. And for novices, I would expect safety benefits of training--especially to the extent it is focused on the 'tactical' aspects of driving, situational awareness and hazard recognition--to be measurable for at least several years. But to reiterate, it is the pattern of reinforcement for everyday behavior that ultimately controls how often and to what a driver pays attention.

At the moment, what seems to me to be the most interesting research approach in this area would be a comparison of the attentional behaviors and hazard avoidance responses, obtained unobtrusively under completely naturalistic (on-road) driving conditions, between groups selected to permit study of the interactions between experience, amount/type of training, and functional ability level.

(Answered by Loren  Staplin, Scientex Corp.)

Q. Figures that mobile phone use in cars involves a four-fold increase in crash risk are now commonly quoted. If this is true, where are all the crashes? There has been a massive increase in cell phone use in automobiles, but has there been a concomitant increase in crash rates?

A. The estimates to which you refer were made in an epidemiological study by a researcher at the University of Toronto. This study was able to examine crashes in detail, and by obtaining cell phone records, was able to draw an "association" between the use of the cell phone and the crash. While causality could not be established by this approach, the relationships were strong and was the basis for establishing the increase in crash risk for both hand-held and hands-free phones. Note also that the lack of crash data does not mean there is not a problem. The data does not exist because it is not collected by the state authorities. This situation may soon change as the various jurisdictions examine the issue more closely. You should also note that other research has consistently shown the relationship between wireless phone use and a deterioration in safety relevant driving performance. I would suggest that you read some of the research papers that are included on the web site.
(Answered by Michael  Goodman, NHTSA)

Long answer: Speech recognition would reduce the manual and visual distractions associated with dialing a cellular telephone. It would allow drivers to keep their hands on the wheel and eyes on the road; however, it would not eliminate the cognitive distractions. Telephone conversations with hands-free phones demand driver attention, particularly complex conversations. Similarly, interacting with a speech-based operating system can increase driver reaction times to roadway events. Because the commands to dial a phone are not complicated the cognitive distractions might be minimal, but speech-recognition in an automotive environment may be prone to errors and recovering from these errors could draw drivers attention away from the road. In addition, even a perfect speech recognition system might distract drivers if the dialog structure is not well-designed. A poorly designed dialing system could lead the driver to make errors and recovering from these errors could pose a cognitive distraction.

Other considerations (an even longer answer): The question implies that if the distractions associated with dialing a telephone were eliminated then the use of a cellular telephone while driving would be safe or at least appreciably safer than using a standard cellular telephone while driving. Completely eliminating the distractions associated with dialing might not affect the overall safety consequences of using a cellular telephone. Several studies suggest that the primary distraction associated with cellular telephones is the conversation and not the dialing.

Because speech recognition technology makes cellular telephone use SEEM much less distracting than manually pushing the buttons, it may encourage people to make calls that they wouldn't otherwise make. This would lead to more telephone calls and increase the total potential for distraction, even though the speech recognition technology might reduce the distraction associated with placing each call.

Thinking beyond the ability of speech recognition technology to dial the number, developers may take advantage of this technology and introduce a range of features that could be substantially more distracting. With speech recognition, it would be possible to allow the driver to search for numbers using an electronic "yellow pages". It would also be possible to allow drivers to search through electronic business cards to find a number. These features might encourage drivers to do things they would be unlikely to do (hopefully) with a standard cellular telephone, but that could be very distracting even with speech recognition.

Speech recognition technology may slightly decrease the overall distraction associated with cellular telephones by making dialing the telephone less distracting, but it may also encourage drivers to place more calls and may lead to new functionality that could be quite distracting. Unless properly implemented speech-recognition technology may have the counter-intuitive effect of increasing driver distraction and degrading driving safety.
(Answered by John  Lee, University of Iowa)

• automation systems that can augment the driver's driving activities by providing additional "eyes and ears";
• automation systems that can partially substitute for the driver's driving activities;
• automation systems that can completely replace the driver's driving activities.

The first category of automation systems represent collision or safety warning systems, using sensors to detect hazardous driving conditions and then processing the sensor outputs to determine when the driver needs to be warned. The warnings could be auditory (tones, buzzers, synthesized speech), haptic (vibration or torque applied to steering wheel, vibration or pressure to gas pedal or seat cushion), kinesthetic (application of brake pulse) or visual (lights on instrument panel, in mirrors or head-up display). The auditory, haptic and kinesthetic warnings could be very effective at catching the attention of a distracted driver IF they are well designed to elicit the "correct" emergency response from the driver. The visual warnings are less likely to help, since the distracted driver is not necessarily going to notice them.

A variety of these systems have been introduced to the market for commercial trucks and buses in the U.S., to help avoid forward collisions, run-off-the-road crashes and side collisions during lane changes. However, the passenger car market has not yet seen any of these (except for short-range warnings to assist in parking, which are not really relevant to the driver distraction issue). A few such systems have recently been introduced in high-end cars in Japan.

The second category of systems, providing control assistance to the driver, present a more complicated picture relative to driver distraction. The most prominent of these systems is adaptive cruise control (ACC), which uses a forward ranging sensor such as a radar to measure the distance and closing rate to the leading vehicle and then uses that information to adjust the speed of the equipped vehicle so that it maintains an "appropriate" separation behind the leading vehicle. Another system that has been proposed by some people is a lane keeping assistance system, which would provide an active torque to the steering wheel to tend to keep the vehicle centered in the lane, providing the driver the impression of driving in gentle ruts in the pavement.

The ACC systems may be able to improve safety by encouraging drivers to follow at somewhat longer separations from other vehicles than they do today, and they may be able to reduce rear-end crashes caused by inattentive drivers overtaking slower vehicles. However, if drivers become overly reliant on the ACC and do not really understand its limitations (inability to sense stopped vehicles, road debris, and animal intrusions and inability to respond to aggressive cut-ins or abrupt stops of preceding vehicles), it has the potential to exacerbate the driver distraction problem. This could even encourage drivers to engage in more non-driving tasks than they do now while driving, which would be most unfortunate. I am not aware of any definitive data to confirm or refute these hypotheses, which are in urgent need of testing by drivers who do not know that they are being tested for these issues. Primitive ACC systems have been on the passenger car market in Japan for several years, while capable ACC systems were introduced in Europe last year and are likely to be available in the U.S. within the next year on select high-end cars and heavy trucks. The lane keeping assistance systems would pose substantially more serious concerns for driver distraction and are not under serious consideration as products at this time, as far as I can tell. Any attempt to combine lane keeping assistance with ACC has the potential to be disastrous, because it would present the driver with a simulacrum of automated driving, which some drivers would be tempted to abuse by ignoring their driving responsibilities.

The third category of automation systems, which completely take over the driving function, raise an additional set of issues. These systems are not subject to distraction themselves, so while they are in use the driver distraction problem per se becomes a moot issue. The driver can turn his/her attention to other issues, or "tune out" completely, without raising safety concerns. However, the important issue then becomes how to re-engage the driver's attention at the end of the automated drive so that s/he can take over driving from the exit of the automated highway facility to his/her final destination. There are also some longer-term challenges associated with the possible decrement of driving skills or driving attentiveness by drivers who do a large fraction of their travel in the automated mode, but still need to do considerable conventional driving. It is important that they not carry over their expectations for performing other activities during the automated drive into their conventional manual driving behavior. The fully automated driving capabilities are likely to become available only to transit bus and commercial truck drivers on specially equipped facilities within the coming decade; passenger car drivers will probably need to wait until the decade after.

(Answered by Steven  Shladover, California PATH)

Your first question on the maximum number of recommended information displays a HUD should feature can be answered simply: It depends. There is a tendency for designers to think of such displays as a panacea. That is, since it intuitively seems that providing head-up information is best, then everything should be displayed using a head-up presentation. One comprehensive source on guidelines for automotive HUD information content is a PhD dissertation by Steve Jahns at the University of Iowa (Steven K. Jahns, 1996. Information content and format recommendations for automotive head-up displays, PhD Dissertation. University of Iowa). The guidelines cited in David Curry's response to this same question are based on Jahns' work.

It is my personal opinion that if HUDs are used, they are best suited to display simple command information (e.g., turn-by-turn information for navigation). More complex information (such as a detailed map) can be more distracting than a head down display. Drivers also may feel over-confident in their glances to a HUD versus a dedicated head-down display (HDD). For instance drivers know that is dangerous to look away from the roadway when they look at a HDD, however, drivers may feel that a HUD is safer to look at even the information may be equally as demanding. Other status-based information is simply not important enough to require head-up presentation. For instance, a glance to the speedometer is a common occurrence, but not necessary a visually demanding task. Other driver status information such as telltales also are not critical enough for this type of display and may be more salient if flashed on the instrument panel. Unlike commercial and military aircraft, drivers need not react immediately to this type of information. The use of HUDs for crash avoidance information may also may be a detriment since the goal of crash warnings are to immediately orient the driver's attention to the hazard. Some other issues to consider before selecting a HUD as an information source include:

• Ambient light - Most drivers spend much of their time on-the-road during the day under high ambient light conditions. Cost limitations on current HUDs prevent salient information presentation during high ambient light conditions.
• Redundant information - Most, if not all information placed on HUDs in the past is redundant with the instrument panel. Designers need to consider the cost/benefit. Most HUDs to day are put on vehicles to increase the marketability of a vehicle.
• Perceptual capture- Although HUDs may be focused at a variety of distances in front of the vehicle, drivers still are required to perceptually capture the information, thus distracting them from the road (this is especially true for more attentionally demanding information). It is not possible to "look through the HUD" and see the environment ahead as well at the information display. We are "spot light" information processors-we are either looking at the HUD information or the outside environment. As a consequence, there are two distinct visual planes with HUDs and driving that independently require driver processing resources.

A list of specific literature that takes into account (1) emerging technologies (2) cognitive load (3) the line of sight, and (4) driver preferences and adaptability to such a system can be found at the following link: www.uiowa.edu/~ppc/hudrefs.html
(Answered by Daniel  McGehee, University of Iowa)


---

Wierwille and Tijerina (1998) using a narrative crash database from North Carolina were able to put together a simple regression model that relates eye glance behavior to crash rates. This model, although simple, is built upon actual crash data and reasonable assumptions. The model requires as input the following parameters:

• Average Glance Length
• Number of Glances, and
• Frequency of device use

The data for the fuel gage was present in the Wierwille and Tijerina article as were data on the frequency of using radio controls. I used additional data that we have gathered on-road over several years from a variety of studies and data that were present in other articles to generate a range representing the types of new devices that are coming onto the market. In addition to using these data for glance length and number of glances, I estimated that a typical frequency of use for such a device would be 20 times per week. This represents two times per commute trip and would probably be a reasonable estimate for a navigation system with traffic information or a mobile internet type of application. In contrast, the radio control use frequency was 56 times per week. From these data, the model predicted a crash rate of 7 to 32 times higher for the newer devices relative to the simple visual task of checking a fuel gage.

Reference
Wierwille, W.W.and Tijerina, L. (l998). Modelling the Relationship between Driver In-Vehicle Visual Demands And Accident Occurrence. In Vision in Vehicles VI. North Holland Press, Amsterdam.

(Answered by Thomas  Dingus, Virginia Tech Transportation Institute)

• "To insure timely driver detection and response to the HUD information, the number of items on the HUD should be kept to a minimum by including only that information which is required or useful for a given set of circumstances.
• To ensure that the impact on driver task performance is minimized, no more than four to five efficiently designed information items should be displayed on the HUD at any one time.
• If HUD information is only presented at very infrequent intervals (e.g., to indicate a system failure), the information may result in a prolonged "novelty" effect or a less than optimal driver response to HUD warning information. Provide enough HUD display information so that the driver is accustomed to scanning and responding to HUD information"

As a general rule, the greater the number of items on the display, the more distraction potential the display will have. During simulator experiments which we sponsored, drivers with 7 or 8 items on the HUD glanced at the display with increased frequency and duration in comparison to displays with fewer items. Their speed maintenance and lane position performance were also reduced while using high information complexity displays. Based upon these results, it is recommended that a maximum of four or five information items be presented on the HUD at any one time. This will eliminate overload potential by providing a cap in the complexity level the HUD can attain. Furthermore, an attempt should be made to keep the number of items on the HUD as low as possible at any one moment in time. Driver reaction to new information items will be best if such items are added to an uncluttered display (containing, for example, only one other item). If the driver has to detect a change in one of several items, reaction time will increase. Basically, this is an endorsement of "by-exception" type of HUD information---in other words, telltales may be displayed on the HUD for system malfunctions, but multiple status indicators (e.g., engine temperature, oil pressure, etc), for the most part, would not be appropriate unless they were out of tolerance. Notable exceptions to this heuristic would be such items of frequently accessed information as vehicle speed.

Note: Material for this response was gathered from guidelines prepared by Steve Jahns and Tom Dingus at the Human Factors Research Group at the Center for Computer-Aided Design at the University of Iowa under Delco Electronics sponsorship.

(Answered by David  Curry, Delco Electronics Corporation)

We must then defer to human factors data. There are 3 types of distraction generally cited in the literature: visual, mechanical and cognitive.

It may be valid to assume that as cell phone users become more familiar with their equipment, they may spend less time looking at their device to turn on the power, or dial. They will still have to look at their phone if there are text messages, or other features. Therefore, there may be decreased visual distraction of a second or two for frequent users who can manually detect the power button and speed dial features.

With regard to mechanical distraction, the argument is that using a phone in a hands free mode (i.e., placing the phone in a holder of some sort) decreases driver distraction. The phone must still be dialed in some way, and calls sent out, but drivers would not be holding the phone to their ear. Frequent or casual cell phone users may decrease their mechanical distraction by using a holder, and keeping both hands on the wheel.

What seems to be most relevant to safe cell phone use is the cognitive distraction. I defer to the human factors experts who may have studied our ability to better multi-task as activities are practiced. But I would also caution that such practice would again more likely address the visual and mechanical aspects of cell phone use. Anyone of driving age has made numerous phone calls, using land lines, during their lifetimes. How do we respond to someone who is standing in front of us trying to capture our attention while we are on the phone? Often we wave them away, or interrupt our conversation on the phone to address the other person. Even after years of talking on the land line phone, our ability to concentrate on more than one activity doesn’t seem to improve. The activity that cell phone using drivers are not attending to is the driving task. I believe that this is a critical issue, and that non-essential technologies which do not help us operate our vehicles more safely should not be allowed.
(Answered by Frances  Bents, Dynamic Sciences, Inc.)

In 1989, when the first "accurate-to-the-exact-street" navigation system for the Japanese market came out, there was much discussion as to how much information should be shown to the driver while the car was in motion. There was also concern over operation of the navigation system, such as inputting destinations. After much debate, it was decided that the major automotive OEMs would get together and conduct research to form the basis for common guidelines that would ensure good usable products while ensuring safety.

Reviews of previous research and follow up experiments with various systems and loads were conducted to come up with what is called the JAMA guidelines. (JAMA: Japanese Automobile Manufacturing Association.) The guidelines have undergone couple of revisions as technology emerged, such as when communication of real-time traffic information became common.

I would not want to use the expression "learn", but rather address what is worth considering when developing and marketing such new in-vehicle systems. Below are personal views that I believe many of my colleagues share.

1. Human nature; Will the product (even if unintentionally) cause "human nature" to do what is not rationally safe? If the answer is yes, then consideration should be given as to how these systems are designed and marketed.
2. The Good and the BAD; Will the public benefit from these systems? If so, we should seek to ensure that the merits from these systems will be realized without getting overly cautious and killing the good in them. Therefore, guidelines must be practical. We cannot expect perfection.
3. Cooperation & Competition Without going against anti-trust issues, there should be good (honest) cooperation between OEMs so that logically and practically correct systems emerge and competition will be fought in areas where we will not sacrifice safety. Having certain restrictions will in many cases spawn new innovative design that are easier to use as well as being safer. This is healthy competition.
4. Timing is crucial. It is difficult to come to consensus once products come out in great numbers. After committing to a certain design, there could be a tendency for non-logical factors to dominate discussions. So it is better to come to a timely conclusion of a Grade-B solution rather than waiting forever for a Grade-A solution. In some cases, "Good is better than best, because best may never come."
5. Flexibility. Since technology evolves, we should be prepared to change guidelines to match these changes. There should be an institutional effort and climate that facilitates this making it possible to observe timing issues mentioned above (number 4).