The Behavioral Contributors to Highway Crashes of Youthful Drivers.

Abstract

The per-mile crash rate of drivers under age 20 is over five times that of the adult population in general, while that of 16-year-old novices is approximately ten times that of adults. Reports of over 2,000 non-fatal crashes involving young drivers were analyzed for behavioral crash contributors as a step in orienting preventive efforts. The great majority of non-fatal crashes resulted from errors in attention, visual search, speed relative to conditions, hazard recognition, and emergency maneuvers, with high speeds and patently risky behavior accounting from but a small minority. The pattern of errors for novices did not differ significantly from that of more experienced youth.

As a group, young drivers present a crash risk far exceeding that of any other age group. [Williams 1996]. On a per-mile basis, the non-fatal crash rate of drivers under age 20 is over five times that of drivers in the 30 – 65 year age range. The rate for 16-year-old novices is over ten times that of adults and almost three times that of 18-year-olds. The sharp drop in crash rate over the first few years of driving time testifies to the benefits of experience and maturity in leading to safer operation.

SOURCES OF RISK

The characteristics of young novice drivers and their crashes have been the subject of extensive research, well summarized by Lonero et al (1995) and Simpson (1996) among others. The wide array of crash contributors can be divided into those resulting primarily from lack of driving experience and those stemming from the immaturity of youth itself. Experience-related contributors to risk include lack of skill in handling the car, difficulty in sharing attention between control of the vehicle and the demands of traffic, unfamiliarity with rules of the road and safe driving practices, and difficulty in recognizing road and traffic hazards. Contributors more closely associated with lack of maturity include youthful thrill-seeking, impulsiveness, the need to impress, and resistance to authority, among others.

The public’s view of the risk presented by young drivers is greatly influenced by the serious crashes that gain newspaper coverage. Because speed is a major determinant of crash severity, it should be no surprise that a high percentage of highway deaths involve speeding [Williams, Preusser, Ulmer & Weinstein 1995]. Since driving at high speed tends to be associated in the public’s mind with youthful exuberance and thrill-seeking, much of the effort to address the young driver crash problem has focused upon changing attitudes toward the risks of driving. Yet fatalities are but a part of the crash picture for this age group. Among youthful drivers, injuries outnumber deaths almost 100 to 1 [NHTSA 1999]. Achieving the greatest benefit to young drivers, as well as society in general, will require addressing those factors that contribute to all crashes.

EXPERIENCE AND CRASH RATE

Research has shown crash rates to vary with both experience, as measured by amount of driving, and maturity, as generally reckoned in terms of age. However, there are reasons to believe that experience-related factors play the stronger role over the first years of driving. The almost two-thirds decline in crash rate from age 16 to age 18 seems more readily attributable to an initial learning curve than to rapid maturation. Even where licensing is delayed to age 18, a similar decline occurs over the next two years [Twisk 1996]. In motorcycling, where age and experience are relatively uncorrelated and the two can be separated statistically, the per-mile crash rate drops more steeply with experience than with age [McKnight and Robinson 1990]. Finally, the effect of experience is evident in the fact that 16-year old females, who compile approximately half as much driving as males of the same age, show a third higher per-mile crash rate [Williams 1996]. At ages where the annual mileages of the two sexes becomes similar, so does the crash rate. Regardless of the relative roles played by experience and maturity, efforts to reduce crashes by exploiting the lessons of experience appear more likely to achieve success than efforts to accelerate maturation.

ANALYSIS OF CRASHES

It seems reasonable to believe that much of the severe drop in crash rate over the first years of driving arises out of situations in which novices are exposed to potential danger and are able to acquire through those experiences the knowledges, skills and attitudes needed to cope with similar situations in the future. Efforts to develop safer driving through instructional processes would be greatly aided by insight into the specific dangers youthful, inexperienced drivers are most likely to encounter. Unfortunately, the primary source of insight into dangers faced by drivers comes from situations in which drivers fail to handle them adequately and are involved in crashes. Crash reports prepared by qualified investigating officers have served a source of information in efforts to improve roads, vehicles and drivers.

Efforts to improve the performance of drivers through crash analysis have been hampered by the inability to extract from automated traffic records information bearing upon the behavioral shortcomings that may have contributed to individual crashes. Most states record in automated crash files little of driver behavior beyond codes corresponding to traffic violations. These rarely identify the specific errors leading to crashes. The most systematic attempt to identify the behaviors contributing to crashes was the Indiana Tri-Level Study [Treat et al. 1977] in which scientists performed on-site analyses of 2,258 crashes and in-depth investigations of an additional 420 crashes. It was one of the first studies to point to inattention and lack of visual search as the leading contributors to crashes. While the experience of drivers making up Tri-Level Study sample was not recorded, only 20% of drivers were under age 20, making it unlikely that any sizable proportion consisted of novices.

Although automated state traffic records contain little information relating to behavioral crash contributors, most of the hard copy reports from which the records are generated provide narrative descriptions prepared by the police responding to incidents and making out the reports. The descriptions vary considerably in detail from state to state, as well as among jurisdictions and investigating officers within states. Yet, most are capable of shedding some light upon the mistakes leading to crashes. An analysis of these reports could help in identifying the behaviors contributing to the crashes of young drivers, including those just beginning to drive.

OBJECTIVE OF STUDY

The overall objective of the study to be described was to identify behaviors contributing most frequently to the crashes of young drivers, as a step in development of a driver education program oriented specifically toward reduction in those crashes. The identification process focused upon two sets of crashes: those of young drivers in general, and those of novices in particular.

Young drivers

The specific crashes experienced by drivers under age 20 provide a potentially fruitful target for efforts to lower vulnerability of young drivers. Knowing the relative frequency of the various crash-involved behaviors will help in prioritizing efforts to make the best possible use of available safety resources.

Novice drivers

Any crashes in which novices are over-represented relative to young drivers with some experience would be those that undergo the quickest decline, presumably indicating the influence of skill acquistion more than maturation. Behaviors underlying those crashes would represent particularly appropriate targets of preventive efforts.

METHODS

The procedure employed in identifying the behavioral contributors to the crashes of young drivers was one of examining the narrative descriptions of crashes for samples of drivers under age 20 and coding the crash-related behaviors in a way that would permit statistical analysis. To identify any subset in which novices were over-represented, crashes involving 16-year olds were compared with samples involving drivers in the 18-19 year age range.

STUDY SAMPLE

Samples of 1,000 crashes involving young drivers at each of two experience levels were considered necessary to gain acceptably reliable estimation of frequencies for behavioral errors leading to crashes. While a national crash sample drawn form many states would have provided the most accurate picture of crash related behaviors, the costs of accessing hard copy records are highly sensitive to the number of state record keeping systems involved, far more so than to the number of individual records accessed in one state. However, the inclusion of at least two states was believed necessary to furnish some measure on interstate variability and thus provide some indication as to the reliability of the crash figures gained (the most informative analysis of crash-related behaviors conducted thus far, Treat et al. [1977] was based upon reports from but one state). The authors had access to hard copy reports of crashes from the 1997 files of two states, California and Maryland. If the results obtained from these fairly dissimilar states were to show high agreement, they could be viewed as providing reasonably reliable estimates of behavior frequencies in general; if they did not, then it would be necessary to expand the sample to additional states.

In each of the two states, quotas of 1,000 crash reports, equally divided between novice and experienced youthful drivers, were established. The reports called for were those prepared by police summoned to the crash scene, not those filed by drivers involved in the crashes meeting specified reporting thresholds. The latter do not provide narrative descriptions, nor could they be trusted as furnishing objective descriptions of contributing behaviors. Generally speaking, police are rarely dispatched to crash scenes unless someone is injured or damage is such as to immobilize vehicles in traffic. In Maryland, one half of the sampled crashes involved drivers who were 16 years of age, and therefore novices at the time the crash occurred; the other half were 18 year old drivers, having up to two years experience. In California, where initial licensing is delayed somewhat, very few 16-year-olds were crash involved, and the age for novices was extended to 17, while the interval for experienced drivers was extended to age 19.

Obviously, a clearer delineation of experience effects could have been achieved by classifying crashes on the basis of miles driven, or at least years of driving. However, crash reports do not provide this information. Comparing drivers in their first year of licensed operation with drivers at least two years over the licensing age could be expected to reveal any effects of experience powerful enough to yield the decline in crashes occurring over the 16 - 18 year age span. California provided numbers exceeding the specified quota for the older age group to allow separation of experience levels within this group should such be desired. However, the process of determining driving experience would have been very costly and the absence of differences in the pattern of behavioral contributors across the two age/experience groups, to be noted later, made it unlikely that separating the sample by driving experience would have been informative.

Composition of the final crash sample by experience, gender and state appears in Table 1. Samples of reports were drawn at random from across all 12 months of the year. In each state, some reports had to be excluded due to the lack of information as to age or gender, insufficiently detailed narratives, or lack of any contributing behavior on the part of the sampled driver. The distribution by state and age was established by quota; the distribution by gender is the result of random sampling within each state/age category. Males constituted approximately 60% of the crash sample in each state. The same distribution occurred in each age/state category except the California 17-year-old group, in which males made up only 54% of the sample. The lack of data as to numbers in the licensed population or amount of driving bars any inferences as to relative risk by gender.

Table 1.

Distribution of Sample by Age, State, and Gender

	California		Maryland		Total
	17 yr.	19 yr.	16 yr.	18 yr.
Male	291 13.6%	414 19.5%	265 12.5%	291 13.7%	1261 59.3%
Female	247 11.6%	267 12.5%	176 8.3%	177 8.3%	867 40.7%
	538 25.3%	681 32.0%	441 20.7%	458 21.5%
Total	1219 57.3%		909 42.7%		2128

IDENTIFICATION OF BEHAVIORS

Accident narratives were reviewed for information capable of providing valid inferences as to the behavior of the drivers that contributed to each crash. The definition of a contributing behavior was one that, if performed correctly, would have reduced the likelihood of a crash occurring. This definition includes both behaviors that would avoided creating a crash situation and behaviors that would have protected drivers from situations created by others, i.e. defensive behaviors. Legal culpability was not an issue.

To permit analysis, behavioral contributors had to be coded in some manner. An initial classification system was developed with the following broad categories:

Driving Preparation	Visual Search	Emergencies
Basic Control	Communication	Driver Condition
Rules of the Road	Maintaining Space	Vehicle Condition
Attention	Adjusting Speed

These broad categories describe the scope of the classification system. Each category was divided into subcategories and each of these into specific behaviors to furnish a set of 214 potential behavioral accident contributors. The classification was not intended as a taxonomy of crash-related behavior but rather a convenient way of sorting the types of behavior expected in the narratives. The specific codes making up each category are too numerous to present here and many were not involved in any of the crashes reviewed. Those codes that were used will be described, with corresponding frequencies, in the Results section.

Crash reports were initially analyzed and coded by research assistants, who entered short descriptions of each behavioral contributor along with the associated code. The first three hundred cases were also analyzed separately by the senior staff, and the results used to clarify code definitions and provide guidance to the assistants. The remaining cases were then processed by the assistants and the results for each case then reviewed by the senior staff.

In addition to behavior codes and descriptions, the date of the crash and the driver’s date of birth were recorded in order to establish age at the time of the crash. Gender was also recorded to allow any gender-related crash characteristics to be considered in possible targeting of educational efforts. These data elements were recorded outside of the behavior coding process in order to keep knowledge of age or gender from influencing the coding process. The accident reports also included many other characteristics of crashes, including severity, drivers, vehicles, roadways, natural environment and others that might be of interest. However, cost considerations dictated limiting data collection to elements that could be directly applied to the educational development process funding the analysis.

RESULTS

Analysis of the 2,138 crashes yielded 2,774 behaviors identified as contributing to them, many crashes involving more than one contributor. High correlations of behavioral contributors across age levels, states and gender, to be presented later, allow the results to be pooled in quantifying behavioral contributors. The few significant differences that emerged within these variables will be noted after the general findings have been presented.

CONTRIBUTING BEHAVIORS

The percent of crashes involving each of the behaviors and behavior categories are summarized in Table 2. Since many crashes involved more than one behavior, the percentages add up to more than 100%.

Table 2.

Percent of crashes attributable to deficiencies in specific driving behaviors

Behavior	%	Behavior	%	Behavior	%
BASIC CONTROL	8.0	SEARCH AHEAD	19.1	ADJUSTING SPEED	20.8
Lane keeping	2.6	Distance	3.1	Traffic/road conditions	8.7
Turning path	1.3	Roadsides	4.3	Curves	6.1
Braking	1.3	Before left turns	4.8	Slick surfaces	2.3
Turning speed	0.7	Car ahead	3.1	Slick curves	1.5
Other	2.1	Left-turning vehicle	2.9	High speed	0.7
RULES OF THE ROAD	5.6	Next lane	0.9	Other	1.5
Traffic lights	1.7	SEARCH TO SIDE	14.2	MAINTAINING SPACE	9.8
Stop signs	1.3	Intersection: burdened	7.7	Following distance	5.8
Lane use	1.5	Intersection: privileged.	5.5	Crossing & entering	1.4
Passing	0.6	Sight obstructed	0.8	Side clearance	1.3
Other	0.5	Other	0.2	Overtaking	1.1
ATTENTION	23.0	SEARCH TO THE REAR	9.4	Other	0.2
Maintaining attention	18.6	Slowing	3.0	COMMUNICATION	1.2
Avoiding distractions	3.8	Backing	2.1	Interpreting signals	0.8
Attention sharing	.07	Periodically	2.1	Signaling intent	0.3
DRIVER CONDITION	4.8	Changing lanes	1.5	Signaling presence	0.1
Alcohol impairment	2.4	Other	0.7	EMERGENCIES	9.4
Fatigue	1.7			Swerving	5.6
Other	0.7	OTHER	0.9	Skid recovery	1.4
VEHICLE CONDITION	1.5			Braking	1.0
				Tire failure	0.7
				Brake failure	0.7

Search

It is apparent that inadequate visual search, previously known as a major crash contributor, was involved in more crashes than any other form of behavior, contributing to a total of 43.5 % of crashes across all individual sub- categories. What is particularly revealing about the present analysis is the specific patterns of failure that were involved, including search when crossing intersections, both as the burdened vehicle (7.7%) and the privileged vehicle (5.5%), when making left turns (4.8%) or approaching a left turning vehicle (2.9%), looking far enough ahead (3.1% ) or checking the roadsides ahead (4.3%), watching the vehicle ahead for sudden slowing (3.1%) or checking the mirror when slowing abruptly (3.0%), checking behind in lane changes (1.5%) and watching vehicles in the next lane for abrupt moves (.9%).

Attention

The second most common shortcoming involved lack of attention, which contributed to 23% of all crashes. The single largest individual contributor was failure to maintain attention ahead (18.6%), typically a situation in which traffic slowed or stopped ahead and the driver failed to notice in time to avoid a rear-ender. Distractions accounted for 3.8% of crashes while poor attention sharing made up the rest of the category.

Speed

Driving too fast for conditions contributed to 20.8% of all crashes, the single biggest subcategory being failure to adjust to traffic or road conditions (8.7%). Others included taking curves too fast (6.1%), and not adjusting to slick surfaces (2.3%) or slick curves (1.5%). While speed was a clearly frequent contributor to crashes, only .7% involved high rates of speed (over 70 mph.).

Space

Failure to allow enough space from other vehicles contributed to 9.8% of crashes, with lack of adequate following distance not surprisingly being the largest contributor (5.8%). Also contributors were not allowing enough space when crossing traffic at intersections (1.4%), and not allowing space from cars to the side when passing (1.3%), or overtaking (1.1%). No cases of failure to allow sufficent passing distance on two or three lane roads were reported.

While such instances may have occurred, they apparently did not result in crashes.

Emergencies

For the purposes of analysis, “emergencies” were situations requiring quick control responses on the part of the driver, failure to do so adequately contributing to 9.4% of crashes. In most reported cases (5.6%), a quick swerve would have prevented a crash. Other failures were inability to recover from a skid (1.4%) to brake properly in an emergency (1.0%), or to effectively handle the failure of tires or brakes (.07% each).

Basic Control

Lack of control over the car’s motion contributed to 8% of crashes, with simply staying in lane being the single largest problem (2.6%), followed by path in turns (1.3%), braking (1.3%), and speed in turns (.7%). Other contributors were problems with accelerating, shifting, turning, slowing and backing, which contributed to 2.1% of crashes.

Traffic controls

Instances in which drivers evidently saw but failed to comply with traffic controls contributed to 5.6% of crashes. These included failing to heed traffic lights (1.7%), stop signs (1.3%), rules governing lane use (1.5%), and passing restrictions (0.6%). The remaining cashes primarily involved crosswalk violations and responses to police.

Driver Condition

Crashes attributable in some degree to the condition of the driver amounted to 4.8% of cases, and included alcohol impairment (2.4%) and fatigue (1.7%), with a very few attributable to illegal drugs and vision. Vehicle Condition - Vehicle-related problems (1.5%) involved primarily maintenance of brakes and tires to prevent the failures resulting in situations leading to the emergencies noted earlier.

Signals

A surprisingly small number of crashes (1.2%) involved signaling problems and most of these arose from failure to interpret signals correctly (.8%). The other half consisted of failure to signal turns and stops (.3%), and not using emergency flashers to make the presence of the vehicle known (.1%).

AGE DIFFERENCES

The second of two research questions giving rise to the study addressed possible influence upon behavioral lapses of the experience gained in the first years of driving. A simple product-moment correlation of .96 between the frequencies of the various crash-related behaviors across the two age levels indicates that the contributors to crashes of young drivers are highly similar across the 16-19 year age span. While differences between the two age levels in frequencies of various behaviors were small, they were significant taken as a whole (•² =61.46, p < 001). The significance of age differences for individual behaviors was tested by the method of Adjusted Standardized Residual, which is the observed minus expected frequency divided by an estimate of its standard error. Those behaviors showing significant age differences appear in Table 3. Behaviors contributing to a greater percent of novice crashes were lack of search prior to left turns, not watching the car ahead, driving too fast for conditions, and failure to adjust to wet roads. On the other hand, the older drivers had a significantly greater percent of crashes involving following too closely and alcohol impairment. It is noteworthy that, while the novices were over-represented in crashes resulting from unsafe speeds, they were involved in only two of the 15 crashes occurring at very high speed.

Table 3.

Percent of crashes by AGE GROUP for behaviors showing significant age differences.

Behavior	16-17	18-19	Sig	Behavior	16-17	18-19	Sig
Search: Before Left turn	5.7%	3.3%	.01	Speed: slick surfaces	2.7%	1.7%	.05
Search: Car ahead	3.5%	2.2%	.05	Following distance	3.7%	5.8%	.05
Speed: safe for conditions	9.4%	6.8%	.05	Alcohol Impairment	0.7%	2.6%	.01

STATE DIFFERENCES

The similarity of information gained from the two states was assessed through the interstate agreement on the frequencies of the various behavioral contributors. A correlation between the two sets of frequencies of .89 indicates sufficiently high agreement to allow results from the two states to be combined, as mentioned earlier. However a significant overall difference between the two states (•² =219.68, p <001) permits examination of individual behaviors manifesting significant interstate differences as shown in Table 4. The data are presented simply to illustrate the small magnitude of interstate differences; the nature of the differences between but two of fifty states is not relevant to the objectives of the study.

Table 4.

Percent of crashes by STATE for behaviors showing significant state differences

Behavior	CA	MD	P	Behavior	CA	MD	P
Lanekeeping (straight)	3.5%	1.0%	.05	Search to the side	1.3%	0.3%	.01
Maintaining attention	15.5%	18.9%	.05	Search to the rear	4.3%	0.8%	.01
Avoiding distractions	4.3%	2.4%	.05	Speed for conditions.	9.5%	6.2%	.05
Search: Distance	1.4%	4.7%	.05	Speed in curve	4.3%	7.2%	.01
Search: Roadsides	4.8%	2.9%	.05	Speed on wet surface	1.0%	3.6%	.01
Search: Before left turns	2.8%	6.3%	.01	Speed in wet curve	0.8%	2.1%	.01
Search: Car ahead	2.0%	3.9%	.01	Following distance	5.7%	3.8%	.05
Search: Left- turning vehicle	3.7%	1.3%	.01	Emergency: swerve	6.2%	3.8%	.01

None of the differences amounted to more than a few percentage points and the similarities between states greatly exceeded the differences.

GENDER DIFFERENCES

While gender differences were not specifically under investigation, knowledge of any such differences could be helpful in targeting prevention efforts. As with age, the patterns of behavioral contributors were highly similar, with a correlation of .95 across the two sexes. Here again a significant overall difference in crash frequencies was found (•² =68.38, p < 001). Individual behaviors showing significant differences appear in Table 5.

Table 5.

Percent of crashes by GENDER for behaviors showing significant sex differences.

Behavior	Male	Fml	Sig	Behavior	Male	Fml	Sig
Search: Before left turns	3.6%	5.5%	.05	Speed: safe for conditions	9.9%	5.2%	.01
		%
Search: Intersection-burdnd.	6.2%	8.2%	.05	Fatigue	2.0%	0.8%	.05
						%
Search: Intersection-privlgd.	4.1%	6.5%	.05	Alcohol Impairment	2.4%	0.8%	.01

Males were relatively more often involved in crashes occurring at unsafe speeds and when the driver was alcohol impaired. Lapses by females tended to involve visual search at intersection before left turns and when crossing or turning at intersections, the latter including both situations in which they were required to yield and when the law accorded them the right of way.

DISCUSSION

The results of the crash analysis provided an enlightening description of the behavioral shortcomings leading to crashes in which young drivers are involved. The overwhelming majority of non-fatal crashes appears to result from failure to employ routine safe operating practices and failure to recognize the danger in doing so rather than what might be viewed as thrill-seeking or other forms of deliberate risk-taking. Only a very small minority of crashes involved what could be termed deliberately risky behavior, such as operating at very high speeds or engaging in what was characterized as reckless driving. Even in instances of deliberate law violations, such as running red lights and failing to come to a full stop at stop signs, most drivers claimed to have checked for traffic and believed the way to be clear as opposed to simply disregarding the particular traffic control. The information compiled as to the nature and frequency of various behaviors is not dissimilar to that reported in the tri-level study of Treat et al. (1977] in which visual search, speed, attention, and evasive action, in that order, led the categories of behavioral crash contributors.

A caveat is in order. The inferences as to behavioral crash contributors are based upon reports by investigating police officers and the project staff reviewing the circumstances surrounding each crash. Most of the inferences as to speed, space, signaling, traffic controls, and alcohol involvement were fairly evident from information obtainable from crash scenes. However, those involving search, attention, signaling and ways of handling emergencies were clearly inferential, based in some cases upon what participants said e.g. “I never saw him,” some on what the officer concluded from characteristics of the scene, e.g. the length of the skid marks, and some upon what analysts inferred from information provided in reports. In the case of search, it was rarely possible to tell whether drivers failed to look in the direction of other vehicles, or whether they did so and simply failed to notice another vehicle. For these reasons, the frequencies presented in the tables should be taken as representing general orders of magnitude rather than precise quantities.

What the crash analysis failed to do was to identify any sizeable subset of behavioral contributors that vary over the first few years of driving and could help account for the sharp decline in the per-mile crash rate that characterizes this period. The correlation between the crash frequencies of those age 16 - 17 and those age 18-19 year was high enough, and significant differences few enough, to consider the two groups to be vulnerable to the same general sets of behavioral deficiencies. Indeed, the patterns of crash contributors of both novice and experienced youth greatly resemble those of adults, as reported in the Indiana Tri-Level study. Whatever are the benefits of age and experience in preventing crashes, they appear to apply almost equally across almost all aspects of driving.

Had it been possible to subdivide the young driver sample on the basis of actual driving experience rather than age, larger differences might have materialized. However, given the substantial experience differences that are likely to exist between the age groups that were employed it seems probable that any strong effects of experience on the pattern of young driver crashes would have manifested themselves in something more than the small differences observed. The results support the observation of Williams et al. (1995) that “In most ways, the crash characteristics of 16-year old drivers are typical of those of teenage drivers in general; they are simply different in degree.” And as per-mile crash rates show, the degree is quite large.

(Presenter: James McKnight)

Carl Clark: Do you see a possibility of giving experience with simulators and should that begin to be the police driving requirements?

J. McKnight: You’re talking about visual search and so on. Yes, a lot of this could be handled by simulation. Right now, however, interactive computer generated simulation is really not quite up to this yet with the resolution of images and so on, but the way things are progressing, it isn’t going to be very long before they are. And much of this information is currently being applied to the development of programs at AAA [American Automobile Association], and I should mention that the AAA funded most of this research as an antecedent to developing and revising their driver education courses.