«by Johnathon P. Ehsani A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy (Health Behavior ...»
Inclusion Criteria States were excluded from the sample if they introduced multiple GDL components simultaneously with the component of interest, or had a learner license age below 15. Because at least two years of data post-implementation were required to estimate the effect of a component, states introducing GDL components after December 2007 were also excluded from the sample.
Data and Measures Monthly counts of fatal crashes involving at least one teen driver (aged 16 or 17 years) in cars, trucks/pickups, vans/minivans, and sport utility vehicles were obtained for the contiguous period 1990 to 2009 from the Fatality Analysis Reporting System (FARS) for the states being analyzed (National Highway Traffic Safety Administration 2010).
FARS is a yearly census of fatal traffic crashes within the 50 States, the District of Columbia, and Puerto Rico. Every vehicle crash on a public roadway that results in at least one fatality is recorded in the FARS database with information retrieved from police accident reports (Guarino and Champaneri 2010). Fatalities are included in FARS if the victim dies within 30 days of being injured in a crash on a U.S. public road involving a vehicle with an engine (National Highway Traffic Safety Administration 2010). Ideally, data from all injury crashes (not just fatal crashes) occurring in each candidate state would also be included; however, only a limited number of states make their injury crash data available to researchers (National Highway Traffic Safety Administration 2011), so such an approach could not be taken for this study.
Ideally, fatal crash rates would be based on the number of licensed teen drivers, however, licensure data reported by the Federal Highway Administration underreport the actual number of licensed teens, and licensure data are difficult to obtain from individual states (Insurance Institute for Highway Safety 2006). Miles driven by each teen would also be ideal, but are also unavailable and are difficult to measure. Therefore, crash rates were based on the number of teens in the overall population. Annual population estimates by state and age were obtained from the U.S. Census Bureau (Bureau of the Census. U.S. Department of Commerce 1999; Bureau of the Census. U.S. Department of Commerce 2010). Monthly values were interpolated using cubic spline curves; which are the smoothest curve that exactly fits a set of data points (Bartels, Beatty et al. 1998).
Age-group-specific monthly fatal crash involvement rates of 16- and 17-year-old drivers per 100,000 population were calculated using monthly fatal crash counts and monthly population estimates. Data for drivers younger than 16 years were excluded because only a few states allow unsupervised driving by 15-year-olds (Insurance Institute for Highway Safety 2012), resulting in data that were too sparse to permit meaningful analysis (National Highway Traffic Safety Administration 2010).
Covariates Comparison population The monthly fatal crash rate for drivers age 25 to 54 was used as a covariate representing crashes for the typical adult driving population. Applying the identical method used to estimate 16- and 17-year-old fatal crash rates, age-group-specific monthly fatal crash rates of 25- to 54-year-old drivers per 100,000 population were calculated using monthly fatal crash counts and monthly population estimates. The purpose of the comparison population is to adjust for variability in the teen driver crash rates due to extraneous factors affecting drivers of all ages and to test the effect of GDL against a comparison population of persons unaffected by GDL. Although time series analyses control for pre-existing secular trends in crash rates, the inclusion of the crash rates of another age group as a historical covariate to control for unmeasured factors that affect all drivers enhances the validity of the findings.
Gas prices An inverse relationship between gas prices and fatal crashes has been identified for drivers of all ages (Sivak and Schoettle 2010); however, research suggests teen driving behavior may be more sensitive to higher gas prices, relative to older drivers (Morrisey and Grabowski 2010). Monthly national average gas prices, obtained from the U.S. Energy Information Administration (U.S. Energy Information Administration 2011), were used as a covariate in the analyses to adjust for their effect on the amount of driving exposure and resulting crash risk level.
GDL laws For each state, indicator variables were included for GDL components that were introduced before or after the learner license requirements being studied.
Analytical Method To estimate the effects of each GDL component, monthly fatal crash rates per 100,000 population of 16- and 17-year-olds were analyzed using Auto-Regressive Integrated Moving Average (ARIMA) interrupted time series analysis (McCleary and Hay
1982) for each state that independently introduced or changed the GDL component of interest. Interrupted time-series analyses compare observations before and after some identifiable event, with the goal of evaluating the impact of the intervention. The transfer function relates an intervention to its effect on fatality rates. In this analysis, the transfer function has two parameters. The first parameter, !, is the magnitude of the asymptotic change (rise or fall) in level after the intervention. The second parameter, !, reflects the onset of the change. If the null hypothesis that ! is 0 is retained, there is no impact of the intervention. If ! is significant, the size of the change is ! (as a percentage) (Tabachnick and Fidell 2007). For these analyses, ! was fixed at 0, meaning the anticipated change in fatal crash rates would be abrupt and lasting, referred to as a sudden impact permanent change model.
Analytical strategy For each state, the models were estimated using the natural logarithm of the monthly fatal crash rate per 100,000 population. Using the natural logarithm, the coefficient representing the intervention effects (!) is directly interpretable (using the formula 100 x [e! – 1]) as the percentage change in the post-intervention series relative to the pre-intervention series (McDowall, McCleary et al. 1980). Results presented are based on the models using the natural logarithm of fatal crash rates as the primary outcome variable.
The analyses were conducted in three stages. First, a linear regression model was estimated for the teen driver crash rates and the covariates: adult crash rates, gas prices, and GDL laws. Second, the model for each state was statistically adjusted for trends and seasonal variation. Autoregressive and moving average orders were identified using auto-correlation and partial-auto-correlation functions of the series residuals. Finally, the original regression model was re-estimated with the inclusion of the autoregressive or moving average orders identified in the second stage. Outliers were also detected and controlled for in the final model. Analyses were conducted using the SCA Time Series and Forecasting System, a specialized time-series analysis software package (Scientific Computing Associates 2011).
The annual fatal crash rates across five year intervals for states that implemented GDL learner license requirements (duration of the learner license or number of supervised driving hours) independent of other GDL components during the period 1990 to 2009 are presented in Table 2.3. Teen drivers’ fatal crash rates were generally higher than adult drivers’ fatal crash rates for most years, with the exception of Rhode Island. There was considerable variation among states’ teen and adult fatal crash rates, with teen crash rates generally highest in Kentucky and South Carolina, and lowest in Hawaii and Rhode Island. Adult drivers’ crash rates were highest in South Carolina and lowest in Connecticut.
Teen drivers’ crash rates were typically highest in 1990, and usually twice the magnitude of the adult crash rate within the same state. Both teen and adult crash rates declined over the study period, although the decline was more pronounced among teens, such that by 2009, teen crash rates were lower or comparable to adult crash rates in most states in the study sample.
Learner License Period The results of the analysis of the effect of the learner license duration partially confirm the first hypothesis. Under certain conditions, a six-month learner license period was followed by a reduction in teen drivers’ fatal crash rates. Specifically, there was a significant reduction in fatal teen crash rates following the introduction of a six-month learner license in Virginia, Minnesota and Connecticut (Table 2.4). There were no significant changes in the fatal crash rates of 16- and 17-year-old drivers following the introduction of a learner license for the remainder of the states in the sample. Adult fatal crashes accounted for some of the variability in teen fatal crashes for each state in the sample, with the exception of Hawaii and Connecticut.
In Virginia, the introduction of the six-month learner license in July 1998 was associated with a modest, but statistically significant decline in teen drivers’ fatal crash rates (-5.5%). In Minnesota, the introduction of a six-month learner license in February 1997 was followed by a significant decline in teen drivers’ fatal crashes (-18.9%). In Connecticut, the introduction of the six-month learner license in January 1997 was followed by a significant decline in 16- and 17-year-old drivers’ fatal crash rates (Supervised Driving Hours Based on the statistical model’s findings, the second hypothesis is rejected - the introduction of required hours of supervised driving was not followed by a significant reduction of teen drivers’ fatal crashes in any state included in the sample (Table 2.5). In Minnesota, the introduction of 30 hours of required supervised driving practice corresponded with a significant 34.5% increase in 16- and 17-year-old drivers’ fatal crash rates (34.5%). The implementation of a required number of supervised driving hours was not associated with teen drivers’ crashes in Arizona, Kentucky, Maine, New Hampshire or Rhode Island. Adult fatal crashes predicted some of the variability in teens’ fatal crashes in each state except New Hampshire.
In three states: Connecticut, Minnesota, and Virginia, the implementation of a sixmonth learner license was associated with a significant decline in 16- to 17-year-old drivers’ fatal crashes. Learner license periods less than six months were not associated with a reduction in crashes. Required supervised driving hours did not result in a decline in teen drivers’ fatal crash rates in any state. In Minnesota, the introduction of 30 hours of required supervised driving was followed by an increase in teen drivers’ fatal crash rates.
While the decline in crashes for Connecticut, Minnesota, and Virginia was smaller than previously reported effects of a specific learner license period (Agent, Steenbergen et al. 2001; Ulmer, Ferguson et al. 2001; Mayhew, Simpson et al. 2003), determining why a decline occurred in these states and not others, requires an examination of the role of licensing age on learner license duration. Each one of these states introduced a six-month learner license, however, due to variations in the minimum licensing age in each state, the actual duration of teens’ learner license periods is likely different.
Williams outlines three distinct scenarios related to introducing a required learner license period (Williams 2007). In those states that require the same minimum age (e.g., 16 years) for both the learner license and the intermediate license, adding a learner license holding period guarantees license delay. In states that have a younger learner license age (e.g., 15 years 6 months) with the difference between the learner license minimum age and the intermediate license minimum age (e.g., 16 years) being the same as the required holding period (6 months), a delay in licensure may occur. Finally, states that have a younger learner license age (e.g., 15 years) with the difference between the learner license age and the initial license age (e.g., 16 years) significantly greater than the holding period (e.g., 3 months), a delay in licensure is unlikely to result just from an extension of the holding period for a learner license.
Using these criteria, the introduction of a learner license guaranteed license delay by six months in two study states (Virginia and Connecticut) and by three months in Hawaii and South Carolina. In Kentucky, a delay in licensure may occur because the difference between the learner license minimum age (16 years) and the intermediate license minimum age (16 years 6 months) was the same as the required holding period (6 months). In the remaining three states (Minnesota, Tennessee, and Utah), the learner license period was unlikely to cause a delay in licensure because the difference between the learner’s minimum age and the initial license minimum age was greater than the duration of the required holding period.
The extended learner license has been hypothesized to be a primary mechanism through which GDL reduced teen drivers’ crash rates (Williams 2007). Therefore, it would be expected that in those states where a learner license holding period guarantees license delay a reduction in crash rates would be more likely than in those states that did not guarantee delay. If we assume that a dose response relationship exists between the delay in licensure caused by the learner license and fatal crash rates, then Virginia and Connecticut were most likely to experience a decline, followed by Hawaii and South Carolina.
The results of this study provide evidence supporting the assumption of a dose response and Williams’ scenarios. The significant decline in crashes in Virginia and Connecticut suggests that a learner license period that guarantees a six-month delay in licensure saves teen drivers’ lives. The absence of a decline in Hawaii and South Carolina suggests that the delay in licensure should exceed three months in order to result in a significant decline in fatal crashes. The decline in crashes in Minnesota, where a delay in licensing was least likely, suggests that license delay may not be the only mechanism through which the learner license exerts safety effects.