This national panel study indicates that the adoption of an average GDL law was associated with a significant decrease in fatal crash involvements of 16- and 17-year-old drivers (under GDL restrictions) relative to involvements of their closest peer group, 19- and 20-year-old drivers (who were not subject to GDL restrictions), but were subject to the same underage drinking and underage drinking-and-driving laws. GDL laws were even more effective in reducing 16- and 17-year-old driver involvements in fatal crashes relative to adults aged 21 to 25 who were not subject to GDL, underage drinking, or underage drinking-and-driving laws. Our results appear to confirm those of Dee, Grabowski, and Morrisey (2005)
: that the benefits of GDL, though varying somewhat within the legal environment in which the laws are implemented, are not dependent on the existence of other traffic or underage drinking laws.
Our analysis does suggest, however, that the strength of the effect of GDL laws varies with their comprehensiveness or quality. The data from indicate that good GDL laws based on the IIHS rating system (IIHS 2010b
) produced greater reductions in the fatal crash ratios than did the average GDL laws. Conversely, our analysis of the effectiveness of the less-than-good laws found that (with one small exception) they did not appear to be effective in reducing 16- and 17-year-old driver fatal crashes. This finding of limited effectiveness of less-than-good GDL laws agrees with the results of Chen et al. (2006)
and McCartt et al. (2010)
. It should be noted, however, that the effectiveness of GDL laws varies with the extent to which its key components are implemented. Thus, the strength of the GDL laws provides additional evidence for their effectiveness. Nevertheless, the limited effectiveness of the less-than-complete GDL laws should not discourage enacting such limited provisions. Limited GDL laws may act as a stepping stone to the development of stronger legislation. In any case, the demonstrated benefit of GDL laws with most of the seven major components should encourage legislation (e.g., the STANDUP Act 2011
) and amendments to current legislation to create more comprehensive GDL laws in the states.
Three of the legal environment laws (primary seat belt, .08 BAC illegal per se, and use and lose) were somewhat collinear, and when analyzed concurrently with GDL laws, the combination of these laws reduced GDL’s measured effect on 16- and 17-year-old fatal crash involvements. However, the analysis of GDL in combination with the ZT law strengthened GDL’s measured relationship with crash involvements. Although we could not determine the reason for this relationship, this result likely occurred because states that produced early and strong ZT laws (before the federal legislation essentially mandating them) were also more likely to enact strong GDL laws. Another possible reason for the ZT law effect is that it may represent a surrogate measure of enforcement of youthful drivers, especially novice drivers. Whatever the reason, Dee et al. (2005)
found an effect of ZT similar to ours.
A significant problem in evaluating GDL laws that primarily affect crash involvements of 16- and 17-year-olds is accounting for the influence of general changes in the legal, roadway, and vehicle environment in which the GDL laws have been implemented. Aside from using population rates, most studies have attempted to compare the rates for the 16- and 17-year-old cohorts to which GDL regulations apply to the rates of older drivers not affected by GDL laws. A limitation in this approach is illustrated in , which presents the trends in fatal crash involvements for the three age groups included in this study. As displayed in , crash frequency declined for the 16–17 and 21–25 age groups but not for the 19 to 20 age group. In this study, we compensated for changes in crash frequency of the two comparison groups by using the crash ratio measure and by including a time (year) variable in our regression analysis. To ensure a conservative estimate of the public health benefits of GDL, we based our crash reduction benefits () on the pre-GDL period to avoid incorporating the lower denominator produced by the reduction in 20 to 25 age group crash rates. The effect of the downward trend in the crash rates of the 21- to 25-year-old drivers is illustrated in , which shows the estimated percentage of reductions in 16- and 17-year-old fatal crashes assuming no change versus the actual observed change in the 16–17/21–25 ratio. The no-change estimate for the average GDL law (6.9%) is similar to the value (6.8%) reported by Dee et al. (2005)
for the analysis of GDL laws alone. However, when Dee and colleagues introduced alcohol safety laws into their study, the reduction was decreased to 5.8%.
Fatality Analysis Reporting System data, 50-state total crash involvement by age group, years 1990–2007
Coefficients from time-series regression models testing the effects of the passage of selected covariates on the variance of age group ratios.a
Some of the limitations in this study include:
- A number of the sample sizes for 16- and 17-year-old driver involvements in fatal crashes in the state-by-year analyses were small, increasing the year-to-year variance in the ratios of fatal crashes for age groups. This may have accounted for some statistically insignificant findings. It would have been advantageous to have state crash files for 1990 to 2007 for all states and the District of Columbia because these files include crashes of all severities (fatal and nonfatal). Consequently, these files would have a larger number of cases that would have increased the power of this study. Although we used covariates accounting for the existence in the states of primary seat belt laws, ZT laws, .08 BAC per se laws, and use and lose laws (Fell et al. 2009) that could have had an effect on 16- and 17-year-old driver involvements in fatal crashes, there certainly may have been other factors in specific states that could also have had an effect on these young drivers. These effects include (a) traffic enforcement intensity of GDL restrictions by police, (b) publicity surrounding GDL laws, and (c) parental influence on driving restrictions, to mention a few. Measures for these factors were not readily available at the state level.
- Use of fatality ratios is a well-known technique to control for changes in crash exposure over time (i.e., Voas et al. 2007b; Romano et al. 2008). This approach, however, only adjusts for changes that similarly affect both the numerator and denominator (i.e., the 16- and 17-year-olds as well as older age groups). Changes in crash exposure that occur differently across age groups may not have been fully accounted for by this approach. Our adult (aged 21 to 25) comparison group exhibited a large reduction in crash involvements over the period of our study. The use of a crash ratio and a time trend in the analysis may not have fully compensated for the adult trend.
- Because we relied on fatal crash ratios to account for changes in crash exposure, we could not use the driver as the unit of analysis. By aggregating data (i.e., by using state/year as the unit of analysis), we were forced to leave some potentially relevant explanatory variables out of our models. The role of variables, such as gender or race/ethnicity, were not considered in this study.
- We used the IIHS GDL grades from September 2007, the last year included in our time-series analysis, to categorize the GDL law for each year an individual state had a GDL law on the books. Thus, our results specific to these grades are the most conservative we could have provided. This is because states improve their GDL laws over time, if they change at all. We did not see any weakening of the GDL laws. For example, from 2007 to 2008, the IIHS GDL grade changed for one state (Minnesota), improving from “M” (marginal) to “G” (good). Given this trend, the effect of G laws compared to non-G laws may be underestimated in our results, but this was the most conservative way to handle these particular data.
Despite the limitations inherent to the use of crash ratios computed over an aggregated unit of analyses, we believe that the control in crash exposure provided by those ratios (and the consequent reduction in bias) more than compensates for the limitations described herein.