Crash and Risky Driving Involvement Among Novice Adolescent Drivers and Their Parents

Abstract

Objectives. We compared rates of risky driving among novice adolescent and adult drivers over the first 18 months of adolescents' licensure.

Methods. Data-recording systems installed in participants’ vehicles provided information on driving performance of 42 newly licensed adolescent drivers and their parents. We analyzed crashes and near crashes and elevated g-force event rates by Poisson regression with random effects.

Results. During the study period, adolescents were involved in 279 crashes or near crashes (1 involving injury); parents had 34 such accidents. The incidence rate ratio (IRR) comparing adolescent and parent crash and near-crash rates was 3.91. Among adolescent drivers, elevated rates of g-force events correlated with crashes and near crashes (r = 0.60; P < .001). The IRR comparing incident rates of risky driving among adolescents and parents was 5.08. Adolescents’ rates of crashes and near crashes declined with time (with a significant uptick in the last quarter), but elevated g-force event rates did not decline.

Conclusions. Elevated g-force events among adolescents may have contributed to crash and near-crash rates that remained much higher than adult levels after 18 months of driving.

Motor vehicle crashes are the leading cause of death and disability among adolescents, whose crash involvement is much higher than that of older, more experienced drivers.¹ Crash risk is highest early in licensure, declining rapidly for approximately 6 months and then slowly for years before reaching stable, adult rates.^2–4 Crash risk among young drivers is particularly high under complex driving circumstances, such as late at night and with adolescent passengers.⁴ A range of explanations for the high crash rates among novices have been proposed, including the complexity of learning to drive safely and adolescent propensity for risky behavior.⁴

Learning basic vehicle management requires only a few hours of instruction and practice,⁵ but judgment consistent with safe driving is thought to develop only with substantial driving experience.⁶ Novices make mistakes on relatively simple problems when first learning a complex task (e.g., algebra, a musical instrument, golf), but over time they learn to process information quickly, become better judges of their abilities, make fewer errors, and resolve problems more efficiently.⁷ Accordingly, novice drivers make more judgment-related driving errors than do experienced drivers.⁸

Elevated g-force events are one of several important measures of risky driving, along with speeding, close following, distraction, and impairment. Elevated g-force events result from late braking, rapid starts, sharp turns, and yaw. These events are dangerous because they increase the potential for loss of vehicle control, reduce the time available to respond to hazards,⁹ and render a vehicle less predictable for other road users, thereby reducing safety margins. They can also be uncomfortable for passengers.¹⁰

Adolescents are thought to take more risks than adults in general¹¹ and while driving,⁴ which provides frequent opportunities for adolescent risk taking typified by elevated g-force events (e.g., hard stops, rapid turns). High rates of elevated g-force events may reflect risk taking or skill and judgment deficiencies regarding how best to navigate turns and when to initiate stopping. If elevated g-force events are attributable to inexperience and associated deficiencies in vehicle management, judgment, and attention, then rates should decline with experience. If they do not decline with experience they are less likely to be attributable to lack of skill and judgment and more likely to reflect a risky driving style.

Most previous studies on risky driving relied on driver self-reports, providing only estimates of the prevalence of crashes and risky driving.¹² Other studies used police reports on crashes and driving offenses, which can be incomplete and are prone to transcription errors.¹³ Instruments installed in vehicles enable continuous assessment of the g-forces exerted by the vehicle as it maneuvers, crash involvement, and miles driven, allowing accurate calculation of rates and providing many analytic and interpretation advantages over simple frequencies or estimated exposure.⁹

In-vehicle systems that monitor g-force events are commonly used by commercial trucking companies to reduce risky driving.¹⁴ These systems have been adapted for use with novice adolescent drivers to provide feedback to the driver in the form of a blinking light when the vehicle exceeds an established g-force threshold. This information can also be stored electronically and shared with parents. Several studies have demonstrated that feedback about elevated g-force events can reduce these events.^15–17

Despite great interest in risky driving among adolescents, little objective information about its prevalence is available. Accordingly, it is of interest to compare the driving performance of novice adolescent drivers with that of adults. Parents, who drove the same vehicles over the same period and geographic area as their newly licensed adolescents, would be expected to exhibit stable, consistent, and relatively low-risk driving behavior. To our knowledge, ours was the first observational study conducted in the United States that compared novice and adult driving by continuously recording driving performance and mileage. We examined the relationship between per mile rates of novice adolescent and adult crashes and near crashes and risky driving during the first 18 months of licensure. We hypothesized that the rates of adolescent crashes and near crashes and risky driving would decrease with experience but that over the study period the averages would be higher among novice adolescent drivers than among parents.

METHODS

We recruited a sample of newly licensed adolescents and at least 1 of their parents through driving schools and local media in the metropolitan areas of Blacksburg and Roanoke, Virginia. In that jurisdiction, adolescents can receive a provisional driver's license allowing unsupervised driving at the age of 16 years and 3 months, but they can take no more than 1 passenger for 6 months and cannot drive between midnight and 4:00 AM. We excluded identical twins because it would be difficult for coders to differentiate participants. We also excluded adolescents diagnosed with attention deficit disorder or attention deficit hyperactivity disorder because they are thought to be at unusually high risk for motor vehicle crashes. We identified adolescents with these conditions by asking parents whether their child had been diagnosed with either of them.

Our adolescent sample comprised 22 adolescent girls and 20 adolescent boys with an average age of 16.4 years (±0.3) from 9 different high schools (3 participants were homeschooled). Our parent sample had 21 men and 34 women; 42 were primary parents (1 parent of each adolescent participant was required to participate), and 13 were second parents who also drove the instrumented vehicles and agreed to participate in the study. Participating youths received $75 per month for the 18-month study period in compensation for the on-road data collection. They also received $20 per hour for other parts of the study, such as completing questionnaires, as well as a bonus of $450 for completing all aspects of the study.

Vehicle Instrumentation

We installed in participants’ own vehicles a sophisticated driving data acquisition system designed at the Virginia Tech Transportation Institute.^9,18 This system consisted of a computer that received and stored data from accelerometers to assess kinematic data, a global positioning system to assess mileage, and video recorders. We installed cameras strategically so that they could continuously monitor the driver's face, the dashboard, areas reachable by the driver's hands, and the forward and rearward roadway. Two other cameras provided periodic still shots of the vehicle interior (blurred to protect the anonymity of passengers). Audio was recorded only if the drivers pressed an incident button that allowed the driver to record for 30 seconds.

We installed the data acquisition system in participant vehicles within 3 weeks of provisional licensure, and data were collected for 18 months, during June 2006 to September 2008. With participants’ permission, data were downloaded every few months from the computers installed in the trunks of the vehicles; research assistants physically swapped hard drives, usually late at night and unobserved by the study participants.

Measures

Driver identity and time of day.

Coders viewed video data for each vehicle trip (defined as ignition on to ignition off), and the identity of the driver was coded (adolescent participant, parent participant, other driver). Coders categorized night driving according to visual observation of the ambient natural lighting at the start of the trip. We defined late night as 10:00 PM to 6:00 AM.

Crashes and near crashes.

Coders identified crashes and near crashes by viewing video footage of highly elevated g-force events recorded by accelerometers. We defined crashes as contact with an object at any speed in which kinetic energy was measurably transferred or dissipated. Near crashes were close calls that were similar to crash events but did not result in actual contact. In previous studies, analyses have demonstrated similar kinematic signatures and contributing factors between crashes and near crashes, with the primary difference being a successful evasive maneuver for near crashes.^18,19 Although data on near crashes may somewhat underestimate the risk of contributing factors, they have been shown to serve as valid surrogates for crashes.¹⁹ Because crashes are relatively rare and the expense of naturalistic methods limits sample sizes, combining crashes and near crashes into a single dependent variable greatly increases analytic options.^9,18,19

Risky driving.

We defined risky driving by the following g-force events: longitudinal deceleration or hard braking (≤ −0.45 g), longitudinal acceleration or rapid starts (≥ 0.35 g), lateral negative or hard left turn (≤ −0.50 g) and lateral positive or hard right turn (≥ 0.50 g) accelerations, and yaw (±6° within 3 seconds). Yaw is a measure of correction after a turn and is calculated as the ∆v between an initial turn and the correction; in short, it is the g-force exerted when the vehicle swerves. For each measure, the specific g-force at which an event was counted was sufficient to make passengers uncomfortable, modest enough to be sensitive to skill improvements, and common enough to provide stable rates when aggregated over 3-month periods. Following the method of Wåhlberg,²⁰ we created a composite variable by counting each event over (or below) the threshold set for each of the 5 individual measures and conducted analyses on the composite measure.

Statistical Analyses

We analyzed counts of crashes, near crashes, and elevated g-force events with Poisson regression models that incorporated the logarithm of the mileage driven as an offset.²¹ The unit of analysis was the trip. We included participant-specific random effects to allow each individual to deviate randomly from the general trend. Preliminary analyses indicated a positive correlation between elevated g-force rates of adolescent drivers and their parents (Spearman's ρ = 0.26 for the composite measure), so we included an additional random effect in the models to account for the correlation within families. The models appeared to provide a reasonable balance between goodness of fit and parsimony.

Poisson regression with random effects provided estimates of median incidence rates (IRs) of crashes and near crashes per 10 000 miles and elevated g-force events per 100 miles. An IR represented the median rate for drivers in the specific population (adolescents or parents). We tested possible changes over time since licensure (in quarters). We compared IRs for parents and adolescents according to incident rate ratios (IRRs), which we calculated by quarter and overall. An IRR of 1.0 indicated no difference, 0.5 indicated 50% lower risk, and 1.5 indicated 50% greater risk for adolescent drivers than for their parents. We calculated individual IR profiles for crashes and near crashes and g-force events through estimated random effects, a byproduct of the model-fitting procedure.

RESULTS

Among the data-collecting vehicles were 30 sedans, 5 minivans, and 7 SUVs; 76% were less than 10 years old. Mean mileage among adolescent participants during the 18-month period was 6384 miles (range = 1881–14 865 miles; median = 5934 miles). Adolescent participants drove 310 to 350 miles per month for the first 6 months and 360 to 410 miles per month for the last 6 months. Parents drove the vehicles a mean of 3000 miles (median = 1058 miles).

Crashes and Near Crashes

Adolescent participants experienced 37 crashes and 242 near crashes, and their parents were involved in 2 crashes and 32 near crashes during the 18-month study period. Four of the adolescent driver crashes were reported to police, and 1 crash resulted in injury; none of the adult driver crashes were reported to police. Adolescent drivers had 64 crashes and near crashes in the first quarter, 53 in the second, 42 in the third, 44 in the fourth, 30 in the fifth, and 46 in the sixth.

Table 1 shows the crash and near-crash median IRs per 10 000 miles for adolescents and parents for the 18-month period and by quarter. Table 1 also shows the IRRs comparing adolescent and parent IRs. The IR for adolescents was highest in the first 2 quarters and higher than for parents in each quarter. The overall IR was 7.43 for adolescents and 1.90 for parents per 10 000 miles, yielding an IRR of 3.91. The P values indicate the significance of change in the quarterly values over time according to Poisson regression analyses with random effects. The quarterly IRs for adolescents varied significantly over time (P = .019).

TABLE 1.

Median Incidence Rates and Rate Ratios for Crashes and Near Crashes per 10 000 Miles and G-Force Events per 100 Miles Among Newly Licensed Adolescent Drivers and Their Parents, Blacksburg and Roanoke, VA, June 2006–September 2008

Event	Overalla	Quarter 1	Quarter 2	Quarter 3	Quarter 4	Quarter 5	Quarter 6	Pb
Crashes and near crashes
Adolescents, IR	7.43	9.15	11.68	7.48	5.29	3.90	7.09	.019
Parents, IR	1.90	1.25	2.94	3.55	1.39	0.94	0.98	.168
IRR	3.91	7.32	3.98	2.11	3.80	4.13	7.21	.422
G-force events
Rapid starts
Adolescents, IR	1.02	1.33	1.20	0.87	0.80	0.85	0.80	.034
Parents, IR	0.23	0.33	0.27	0.16	0.20	0.17	0.10	.005
IRR	4.35	3.98	4.47	5.53	4.08	4.92	8.05	.435
Hard stops
Adolescents, IR	0.91	1.14	1.20	0.96	0.90	0.96	1.01	.122
Parents, IR	0.20	0.24	0.16	0.23	0.19	0.15	0.19	.220
IRR	4.58	4.73	7.39	4.09	4.68	6.64	5.46	.144
Hard left turns
Adolescents, IR	0.70	0.62	0.63	0.72	0.79	0.80	0.82	.931
Parents, IR	0.02	0.02	0.02	0.03	0.03	0.02	0.02	.974
IRR	29.02	29.77	28.00	24.21	27.12	33.93	35.51	.988
Hard right turns
Adolescents, IR	0.31	0.23	0.29	0.41	0.39	0.33	0.50	.022
Parents, IR	0.01	0.01	0.01	0.02	0.02	0.02	0.01	.270
IRR	23.29	18.75	35.42	17.61	21.08	21.98	76.08	.283
Yaw
Adolescents, IR	0.14	0.15	0.13	0.16	0.13	0.14	0.15	.888
Parents, IR	0.04	0.05	0.05	0.03	0.03	0.03	0.03	.609
IRR	3.50	2.86	2.71	5.24	4.48	4.65	4.34	.679
Composite
Adolescents, IR	4.16	4.52	4.65	4.37	3.87	4.44	4.79	.510
Parents, IR	0.82	0.93	0.82	0.83	0.75	0.67	0.61	.484
IRR	5.08	4.84	5.66	5.29	5.15	6.57	7.85	.455

Note. IR = incidence rate; IRR = incidence rate ratio. Adolescents had 37 crashes and 242 near crashes, and their parents had 2 crashes and 32 near crashes during the study period.

^aOverall IRR for crashes and near crashes and overall and quarterly IRRs for g-force events were statistically significant from 1 (P < .001).

^bDerived from likelihood ratio test that an IR or IRR was constant over time; individual IRR's were different from 1 (P < .001).

We observed a general decrease for adolescents over the first 5 quarters, with a significant increase from the fifth to the last quarter (P = .026; Figure 1). Similarly, when we treated quarter as a continuous covariate (i.e., we used the log scale), we found that the IR for adolescent crashes and near crashes decreased significantly over time (P = .001). Figure 1 provides the profiles for each adolescent participant and shows that most followed the same pattern as the overall or typical adolescent but that a small proportion of adolescent drivers experienced relatively high crash rates. We detected no significant change in the parents’ rate, and the relatively higher rates in quarters 2 and 3 accompanied low mileage. Moreover, the adolescent–parent IRRs did not significantly differ over time.

FIGURE 1.

Median incident rates for crashes and near crashes among adolescent and adult drivers and adolescent driver variability in crashes and near crashes for six 3-month periods over the first 18 months of adolescent licensure: Blacksburg and Roanoke, VA, June 2006–September 2008.

Note. CNC = crashes and near crashes; IR = incidence rate. Incident Rate Ratio = 3.91

Adolescent Risky Driving Validity

Coder evaluations of the video footage of a sample of elevated g-force events determined the percentages of valid events and events attributable to potholes or other road conditions. Of 3166 events evaluated, coders determined 3080 (97.3%) to be valid. Of the valid events, coders found 86.8% to be attributable in part to driver error, most often poor speed management such as late or inadequate braking. Because of this evidence of the validity of the sample, we used all events in analyses.

The Spearman's rank correlation coefficients between the rates of the individual risky driving measures and the composite measure ranged from 0.60 to 0.84 for adolescent drivers (P < .001 for all measures) and from 0.36 to 0.80 for parents (P = .006 for hard left turns, .001 for hard right turns, and < .001 for all other measures), providing reasonable internal consistency (standardized α = 0.78 for adolescents and 0.65 for the 44 adults who drove a vehicle ≥ 100 miles) for the composite measure of risky driving. For adolescent drivers, the Spearman's rank correlation coefficients between the individual measures of risky driving and crashes and near crashes ranged from 0.27 (for rapid starts) to 0.76 (for hard stops) and was 0.60 for the composite measure (P = .08 for rapid starts, and < .002 for all other measures). In preliminary analyses reported elsewhere, we found that crashes and near crashes were many times more likely among drivers with higher rates of risky driving in the previous month.²²

Risky Driving Among Adolescents and Parents

Median IRs for the 5 elevated g-force measures and the composite measure are shown in Table 1 and Figure 2 for adolescents and parents by quarter. The P values in Table 1 indicate the significance of change in the quarterly values over time. For adolescents, the IRs did not vary systematically over time, although the data showed a decline in rapid starts and an increase in hard right turns after the second quarter. For all g-force events and the composite value, the IRRs were significant and did not vary significantly over time. The overall IRR was 5.08 for the composite measure; for the individual measures IRRs ranged from 3.50 for yaw to 23.29 for hard right turns and 29.02 for hard left turns (the high IRRs were attributable mainly to very low rates among parents). We observed considerable variability among adolescent drivers; a few had extremely elevated g-force rates (Figure 2).

FIGURE 2.

Median incident rates for risky driving (composite measure) among adolescent and adult drivers and adolescent driver variability in risky driving for six 3-month periods over the first 18 months of adolescent licensure: Blacksburg and Roanoke, VA, June 2006–September 2008.

Note. IR = incidence rate. Incidence Rate Ratio = 5.08.

We repeated analyses comparing crashes and near crashes and risky driving rates between adolescent drivers and their parents with controls for driver gender, night driving, and month (as a measure of season), with almost identical results (not shown).

DISCUSSION

We conducted the first observational study in the United States to assess rates of risky driving and crashes and near crashes among novice adolescent drivers and adult drivers through continuous recording and objective assessment of driving performance and mileage. Our findings indicated that crash and near-crash rates were nearly 4 times as high among adolescents as among adults during the 18-month study period. Although adolescent rates declined significantly over time, they were higher than were adult rates in each quarter. Similarly, risky driving was 5 times as high among adolescents as among adults during the entire study period. Our data showed consistently higher rates than were reported in one of the rare studies that used vehicles equipped with data collection instuments¹⁷ and in a recent questionnaire study.²³ The data indicated that novice adolescent drivers reduced crash rates over time despite the persistence of their risky driving habits.

The observed reduction over time in adolescent crash and near-crash rates was consistent with the contention that these incidents decline with experience, presumably because of learning effects. The decline in crash and near-crash rates was most notable after 6 months, consistent with other reports.^2–4 Nevertheless, rates of crashes and near crashes among adolescent drivers remained several times as high as adult rates overall and after 18 months. A significant increase occurred from quarter 5 to quarter 6, consistent with other studies that have shown that after the initial decrease, crash risk increases in late adolescence and then declines in the following decade.⁴ Conversely, adolescent rates of risky driving (i.e., elevated g-force events) remained high relative to adult rates throughout the study, providing no support for the contention that risky driving declines with experience and that adolescents learn to reduce risky driving behavior. Despite some fluctuations in the estimated quarterly IRs, we detected little evidence that quarterly differences declined over time.

Adolescents’ persistent engagement in elevated g-force events may have contributed to the residual high crash and near-crash rates among adolescent drivers. The relative contributions to crash and near-crash rates of elevated g-force events and other factors, such as distraction or inattention, speeding, and errors of judgment, remain to be determined in future analyses. Adolescents’ risky driving may be directly responsible for some crashes and near crashes, but the main concern is that their risky driving style increases the likelihood of crashing by reducing safety margins. Our findings are consistent with the establishment of a risky driving style among adolescent drivers, as previously reported.^23–25 The observed persistence in risky driving may have reflected deficiencies in driving expertise, because expertise consistent with safe driving would be expected to develop fully only after years of driving experience.⁷ Alternatively, persistent risky driving may be attributable to aspects of adolescent development that may prevent novice drivers from recognizing elevated g-force events as risky or may motivate them to drive in ways that make driving challenging, fun, interesting, and dangerous.^26,27

Strengths and Limitations

Our study used objective measurement of driving outcomes and exposure among adolescents and their parents (serving as an adult comparison group) over an 18-month period of assessment. However, the small sample size, the regional nature of the sample, and the rarity of crashes (only 1 crash resulted in serious injury) limit generalization of the findings.

We observed an association between risky driving and crash and near-crash involvement over the study period. Our data suggest that risky driving increases crash risk by reducing the margin of safety.²² Although we are confident that our composite measure of elevated g-force events provided a good assessment of risky driving, we were unable to determine the extent to which events were unintentional and caused by poor judgment or intentional and attributable to self-selected style of driving.

Conclusions

Our sample of novice adolescent drivers experienced rates of risky driving that remained higher than were adult rates over the entire study period. Although adolescent driver rates of crashes and near crashes declined over time, they remained much higher than were parent rates throughout the first 18 months of driving. Our adolescent–parent comparison of risky driving demonstrated the magnitude and persistence of adolescent risky driving. Our data suggest that novice adolescent drivers maintain a risky style of driving that may have contributed to the sustained disparity in rates of crashes and near crashes between novice and experienced drivers.