II. METHOD

A. Data

Data on fatal crashes and population were obtained from two Federal sources: the Fatality Analysis Reporting System (FARS) of the National Highway Traffic Safety Administration13 and the U.S. Census Bureau.14 FARS is a census of all fatal traffic crashes within the United States that involve a motor vehicle traveling on a public road and result in a death within 30 days of the crash. The numbers of drivers age 16 and ages 20 to 29 involved in fatal crashes in each State for each month from 1994 through 2004 were obtained from FARS. Cases where two 16-year-old drivers were involved in the same crash were counted as two events. Midyear population estimates for each State from 1994 to 2004 were obtained from the U.S. Census Bureau.

Information on GDL programs and their effective dates was obtained from the Insurance Institute for Highway Safety (IIHS),15 State government Web sites, and personal contacts with State personnel. IIHS has been tracking GDL programs since 1996, the earliest year in which any State adopted graduated licensing programs. Three lists were obtained from IIHS: a list of components of graduated licensing programs for each State in 1996; a list including enacted, effective dates and details of licensing amendments for States that have changed their programs since 1996; and a list of components of graduated licensing programs for each State in 2005 (IIHS 2005). State government Web sites were used to confirm the programs, resolve inconsistencies, and in some cases obtain the dates of changes in the programs.

Although each GDL program has some distinct features, the main provisions in the various GDL programs generally fall into seven categories: minimum age for a learner permit; mandatory waiting period before applying for intermediate license; minimum hours of supervised driving; minimum age for intermediate license; nighttime restriction; passenger restriction; and minimum age for full licensing (Table 1). To avoid small numbers, related provisions were collapsed into dichotomous variables, for example, “nighttime driving restrictions, yes/no.”

The seven components of GDL programs were coded into quarters of the year based on their effective month. For example, in Alabama unsupervised driving was prohibited from midnight to 6 a.m. beginning October 1, 2002; the first three quarters of 2002 were coded as “not exposed to the nighttime driving restriction” and the fourth quarter coded as “exposed.” If a restriction became effective at any time during a quarter, that entire quarter was coded as “exposed.” Quarters were used rather than calendar years because GDL programs became effective at different times of the year. The unit of analysis therefore became the State-quarter, which represents one State having a specified combination of GDL components for a given quarter of the year.

Excluded from analysis were four quarters after the effective date of each GDL program or component, because licensing restrictions would not affect teenagers who already had their licenses when legislation took effect. After a restriction goes into effect, it can be as long as a full year before all 16-year-old drivers in a State are driving under that restriction. Four quarters before the effective date were also excluded because some teenagers might hasten to get their licenses before the law changed, leading to an increased number of crashes in those quarters. Thus a total of eight quarters were excluded from the analysis for each GDL program changed or enacted.

Table 1. Definition of Seven GDL Components Studied1

Minimum age for learner permit

  • Minimum age 15 ½ years for obtaining a learner permit
  • Reference: less than 15 ½ years

Mandatory waiting period

  • Minimum 3 month waiting period after obtaining a learner permit before applying for an intermediate license
  • Reference: no mandatory waiting period of at least 3 months

Minimum hours of supervised driving

  • Minimum 30 hours of supervised driving
  • Reference: no required supervised driving or required less than 30 hours

Minimum entry age for intermediate stage

  • Minimum age 16 years for obtaining intermediate stage license
  • Reference: less than 16 years

Minimum age for full licensing

  • Minimum age 17 years for full licensing
  • Reference: less than 17 years

Nighttime restriction

  • Any nighttime restriction
  • Reference: no nighttime restriction

Passenger restriction

  • Any passenger restriction
  • Reference: no passenger restriction
1The values of the specific attributes defining the components studied (shown in Table 1 above) used to distinguish weaker from stronger State GDL programs were chosen to allow large enough numbers in each category for meaningful analysis.  They are not to be interpreted as representing ideal or even recommended GDL component attribute values, and they are not intended to provide guidance on what would be required for a State's GDL program to be considered effective.  For example, while State requirements for at least 30 hours of supervised driving practice or an initial stage of at least 3 months are associated with fewer 16-year-old driver crashes, this result does not suggest that either of these is ideal.  Sixty or 100 hours of supervised driving may well be preferable to 30 hours, but the available data did not allow question to be addressed.  Also, the reader should be aware that there may be some excellent restrictions that do not match the seven program components that were analyzed.  For example, North Carolina does not require a specified number of hours of supervised driving; rather, an intermediate license is not issued until at least a year after the initial provisional license, during which all driving must be supervised. It is possible that the impact of this provision could be even stronger that what was analyzed.


B. Analysis

We examined States that had GDL programs at any time during 1994-2004, with the objective of determining whether some programs appeared to be associated with lower fatal crash involvement rates than others. This was a cross-sectional analysis, examining the experience of 16-year-olds in each of the study years, rather than a cohort analysis, which would have followed 16-year-olds in 1994 until they were 26 years old in 2004.

The analyses included 43 States in the continental United States; of these, 36 had GDL programs for at least part of the studied period. The District of Columbia was excluded from analysis since its crash data were heavily influenced by neighboring States. Maine, New Hampshire, Rhode Island, Utah, and Virginia were excluded because they changed their laws more than twice between 1994 and 2004, thus complicating any analysis.

The number of person-years for each quarter in each State was estimated using the mid-year population of 16-year-olds divided by four. The same method was used to calculate comparison group data, i.e., person-years in each State-quarter for drivers ages 20-24 and 25-29. Within the 11-year period 1994 through 2004, 1,480 State-quarters were examined.

The association between GDL programs and fatal crash incidence was assessed using negative binomial regression models based on generalized estimating equations (GEE).16 17 The negative binomial distribution approximates the counts of fatal crashes within State-quarters and the GEE approach takes into account the correlation among quarterly counts of fatal crashes in a given State. Statistical software SAS® was used for the analysis.18

Independent Variables
The independent variable of primary interest was the presence or absence of GDL and its provisions. Three different approaches were used to characterize the GDL programs. In the first approach, whether a State included an intermediate phase in its licensing system was used to determine the presence or absence of GDL programs as a dichotomous variable. The reference group for this comparison was State-quarters without three-stage GDL programs. This comparison is useful because it tells you the combined effect of all GDL programs and reflects the results for all States. Some States, however, have relatively weak GDL requirements, and the results therefore underestimate the impact of stronger programs.

In the second approach, the licensing system for young drivers in each State-quarter was characterized on the basis of how many of the seven GDL components studied were contained in the licensing restrictions, regardless of which specific components are included in the count. The reference group was State-quarters that did not meet the requirements of any of the seven components that we examined. This approach made it possible to examine the impact of programs that only partially met the GDL definition. This was a simple count of components that did not depend upon the effect of each one. It addressed the question, “How many components are needed in order to have an effective program?”

The third approach helped to identify the specific components of a good GDL program. As with the second approach, categorization of programs was based solely on their components, without considering whether the programs included an intermediate phase that would have qualified them as GDL programs. Again, the reference group was State-quarters that did not meet the requirements of any of the seven components. In this approach, the licensing systems for young drivers were grouped based on combinations of the four GDL program components not related to age of licensing: minimum waiting period of at least three months before applying for an intermediate license, minimum supervised driving of 30 hours, any nighttime restriction, and any passenger restriction. One program grouping, for example, included all State-quarters with a combination of a waiting period of at least three months, a nighttime driving restriction and a passenger-carrying restriction. Programs with only age restrictions were treated as a separate category. If a program grouping existed in fewer than 50 State-quarters then it was not treated as a separate category but was combined with other groupings that occurred too infrequently for separate analysis.

Three models based on the three approaches described above were fitted for each of the three age groups studied: drivers age 16, 20 to 24, and 25 to 29. The focus of our study was on 16-year-old drivers. However, fatal-crash-involved drivers ages 20 to 24 and 25 to 29 were also analyzed to allow comparison between GDL-exposed and non-GDL-exposed drivers. Theoretically, GDL programs will not affect the older age groups and therefore their Incidence Rate Ratios (IRRs) should be equal to one.

Dependent Variable
The outcome variable was the natural logarithm (the unit used for all negative binomial models) of the number of fatal crashes involving any drivers in our target age groups in a given State-quarter. States and quarters as dummy variables and year as a continuous variable were included in each model. The State variables controlled for State-specific unmeasured variations that might affect fatal crash counts, such as weather, traffic environment, regulations other than GDL, and socioeconomic conditions. The quarter variables controlled seasonal variations and the year variable controlled for variation in fatal crash counts over the time period studied. The sum of person-years in each State-quarter was considered as exposure and was included in each model.

Fatal crash involvement rates of drivers were used rather than driver fatality rates because they allowed more cases to be included for study. Person-years based on the population of 16-year-olds were used because the licensing rate might have changed due to implementation of GDL programs. Also, the overall benefit to the population, regardless of the licensing rate, was of primary interest.