CHAPTER 2

Research Approach

Two NDS datasets—SHRP 2 and the National Institutes of Health (NIH)–funded SPDS—were employed to evaluate how exposure to driving in more diverse traffic and road environments during the learner’s permit phase and early independent phase is associated with teen driver behavior in later independent driving. Both datasets are described in the following and summarized in Table 1.

Description of the SHRP 2 NDS

The SHRP 2 NDS collected driving data on approximately 3,500 participants for a period of either 12 or 24 months of continuous driving performance data. This analysis presented in this report, which is focused on teen drivers, used a subset of this larger dataset that included a cohort of 254 drivers aged 16–17 (average age of 16.7 years at recruitment). These teen drivers drove for up to 24 months, accumulating a total of ∼1,800,000 miles, and were involved in 149 crashes. Participation was highest in the first year of the study, with a decline in retention over the second year.

For the SHRP 2 NDS teen dataset, there were 201 drivers (out of 254) with a full 12 months of data collection. Many of these drivers’ vehicles were not instrumented until later in their first year of driving. Based on participants’ self-reported data on how long they had held their licenses (at recruitment) and recruitment date, the research team determined there were 55 participants whose vehicles were instrumented within the first 6 months of independent driving. Note that the SHRP 2 NDS recruiting protocols did not require teen driver participants to be within a certain number of weeks of getting their license, and therefore, data are missing for this initial 6 months of driving. Figure 1 shows the distribution of months of data collection during the first 6 months of licensure that was collected in the SHRP 2 NDS database. Nineteen of the 55 participants’ vehicles were instrumented in the fourth or fifth month of driving, which left the team with 1 month of data from the first 6 months of independent driving for approximately 35% of this sample of participants (see Figure 1).

Description of the SPDS NDS

The SPDS NDS collected both learner’s permit driving and independent driving data from 82 teenage drivers (average age of 15.6 years at recruitment). All participants were recruited within 3 weeks of obtaining their learner’s permit. The average learner’s permit duration among participants was 10.35 months, and the teenagers drove a total of 18,686 trips and ∼110,000 miles during the learner’s permit driving period; nine crashes occurred and were recorded. During the independent driving phase of data collection, the teenagers drove a total of ∼380,000 miles, and the dataset includes a total of 69 crashes for the learner’s permit and independent driving

Table 1. Overview of SHRP 2 and SPDS datasets analyzed.

phases combined. The following analyses include all 82 participants’ learner’s permit driving and independent driving data.

Institutional Review Board and Data Use License Process

The research team successfully applied for Institutional Review Board (IRB) approval through the Virginia Tech IRB. Upon receipt of approval, the research team then successfully applied for data use licenses (DULs) to obtain access to the SHRP 2 and SPDS NDS datasets for teenage drivers (16- and 17-year-olds). Note that the SHRP 2 NDS has a formal process for all interested researchers to obtain a DUL (restricted public access) whereas the SPDS NDS dataset is restricted to NIH-approved researchers (restricted access).

Overview of Data Reduction

Using the trigger thresholds developed from previous NDSs, event databases were created for both the SHRP 2 and the SPDS NDSs. Similar data coding protocols as those developed for previous studies were used for both SCE and baseline coding to ensure comparable sampling and coded variables across both studies. Virginia Tech Transportation Institute (VTTI) software was used to scan the files of participating drivers to look for kinematic thresholds that were indicative of a probable SCE. Once triggered events were created, each one was reviewed by a trained

coder to determine validity. Valid events were categorized into one of four SCE types operationally defined as follows:

1. Crash: Any contact that the subject vehicle has with an object, either moving or fixed, at any speed. Also included are non-premeditated departures of the roadway where at least one tire leaves the paved or intended travel surface of the road.
2. Near-Crash: Any circumstance that requires a rapid evasive maneuver by the subject vehicle, any other vehicle, pedestrian, cyclist, or animal to avoid a crash.
3. Crash-Relevant Conflict: This refers to any circumstance that requires an evasive maneuver on the part of the participant vehicle or any other vehicle, pedestrian, cyclist, or animal that is less urgent than a rapid evasive maneuver (as defined above in near-crash) but greater in urgency than a “normal maneuver” to avoid a crash. A crash-avoidance response can include braking, steering, accelerating, or any combination of control inputs.
4. Non-Conflict: This refers to any incident or maneuver within the bounds of “normal driving” behaviors and scenarios that is accurately represented by the time series data that created the flagged event. The driver may react to situational conditions and events, but the reaction is not evasive, and the situation does not place the subject or others at elevated risk.

Once these classifications were complete, VTTI’s trained data coders then further reviewed the crashes, near-crashes, and crash-relevant conflicts and coded a variety of variables, as listed in the appendix of this report (see Table A-1). The reduction protocols used for both the SHRP 2 and SPDS NDSs were identical.

As noted above, trained coders completed the data reduction for this study. Data reduction procedures at VTTI follow a standard quality assurance/quality control workflow. This workflow has four phases: protocol development, reductionist training, data reduction, and post-reduction. More information on this process can be found in Description of the SHRP 2 Naturalistic Database and the Crash, Near-Crash, and Baseline Data Sets (Hankey et al. 2016).

Study Design

To answer the research questions focused on teen driving exposure and driving diversity, the research team first identified and operationally defined the independent variables of interest and then calculated the pertinent dependent variables. Key to the planned analyses was how driving exposure and diversity of driving experience during the learner’s permit and early independent driving phase would be operationalized. The independent and dependent variables used in these analyses are described in detail in the following.

Independent Variables

The independent variables for analyses included the phase of driving experience (learner’s permit, early independent, later independent) and the classification of driving diversity, such as roadway/traffic environments and route familiarity. These independent variables allowed the research team to assess overall driving exposure, driving diversity, and driving behaviors that may impact safety outcomes. Key independent variables analyzed to address the research objectives are discussed in the following and summarized in Table A-1 in the appendix.

Driving Phase

The independent variable, driving phase, allowed for the analysis of three driving phases—learner’s permit, early independent, and later independent driving for the teenage drivers studied (Figure 2). For the SPDS, participants were recruited within 3 weeks of obtaining their learner’s

permit (supervised practice driving) and data were collected from this time (within 3 weeks of obtaining a learner’s permit) through the first 12 months of independent driving. Three driving phases were identified: (1) learner’s permit (i.e., practice driving), (2) early independent driving, and (3) later independent driving. Learner’s permit or practice driving was the data collection period when the driver held their learner’s permit. Early independent driving was the first 6 months of independent driving, and later independent driving was months 6 through 12 of independent driving.

For the SHRP 2 NDS, no data were collected during the learner’s permit period. Sixteen- and seventeen-year-old participants were recruited post licensure, but recruitment was not based upon timing of licensure, so the amount of driving data collected during this critical first year of driving is not constrained. Operational definitions for early independent driving were also any driving that occurred during the first 6 months of independent driving. Later independent driving was defined as months 7 through 12 post licensure (see Figure 2).

Driving Exposure

Given that this research project attempted to determine how driving exposure at the beginning of the process of learning to drive may impact later independent driving, the research team wanted to develop an independent variable that allowed assessment of teen driving performance based upon how much driving occurred during the earliest driving periods possible. For the SPDS, the researchers calculated normalized driving hours during the learner’s permit phase (total driving hours divided by total months) for each participant. Those participants who obtained less than the median value of normalized driving time were assigned to the less-driving-exposure group and those who obtained more than the median value of normalized driving time were assigned to the more-driving-exposure group.

Figure 3 shows how the distribution of normalized driving hours per month (total driving hours divided by total months in the learner’s permit phase) for each participant was separated by the median value to classify teen SPDS drivers into two distinct exposure groups: less driving exposure (below the median value) and more driving exposure (above the median value).

A similar procedure was used for the SHRP 2 NDS study; however, the exposure group was based on the early independent driving phase, or first 6 months of independent driving. Again, those participants who obtained less than the median value of normalized driving time during the first 6 months post licensure were assigned to the less-driving-exposure group, and those who obtained more than the median value of normalized driving time were assigned to the more-driving-exposure group (see Figure 4).

Route Familiarity

The learner’s permit period requires the presence of a supervisory driver, usually a parent or guardian, in the car with the learner driver (Williams & Tilson 2012). Thus, the supervisor plays an important role in guiding the routes undertaken during the learner’s permit period, impacting the exposure to diverse roadways to increase overall driving familiarity (Ehsani & Tefft 2021; Mirman et al. 2014). Driving familiarity is developed through exposure to diverse driving situations, meaning not only routes that differ in their roadway classification, but also driving on those routes under varying day/night, weather, and traffic conditions. Further, it is assumed that, for all drivers, repeated trips on the same route increase familiarity with that route.

Route familiarity was calculated using a dichotomous metric of a given trip based on the overlap of Global Positioning System (GPS) data to other trips and trip lengths. This unique metric considers portions of the trip that may be familiar to the driver (e.g., the initial part of the trip through their neighborhood) and identifies portions of the trip on roadways that they may never have traveled before. A trip was considered familiar if 70% of the GPS data overlapped with a prior trip and the trip length was within 0.02 miles of a prior trip. This measure was developed by Ehsani and Johns Hopkins University and applied to other teenage drivers (Zhu et al. 2024).

Functional Road Classification

This variable was obtained using mapping software data for both the SHRP 2 and SPDS datasets. The percentage of time on each type of road classification was calculated for these datasets.

Functional class (FC) defines a hierarchical network used to determine a logical and efficient route for a traveler. There are five levels of FC, and each street segment is tagged with an FC number indicating its level, defined in the following:

• FC 1: Very-long-distance routes between major cities. The “highest level” network comprises the FC 1 arterials, which are primarily controlled-access highways designed for very-long-distance travel linking major metropolitan areas and cities.
• FC 2: Primary routes between major and smaller cities and through metro areas.

• FC 3: Major routes between minor cities or towns and through city districts.
• FC 4: Routes connecting minor towns or villages and collecting local traffic in city districts.
• FC 5: Roads that are not efficient through routes. The “lowest level” and final category is FC 5, which comprises roads not considered to be arterials or transportation corridors.

Due to low frequencies of driving time, FC 1 and FC 2 were combined as a high-speed roadway variable and FC 3, FC 4, and FC 5 were combined for a more moderate- to slower-speed roadway variable.

Vehicle Access

Gershon et al. (2018b) demonstrated that teen drivers who had primary access to a vehicle (primary vehicle driver) drove more miles, sped more frequently, and had a higher crash rate compared to teens who shared a family vehicle (shared vehicle driver). This suggested that teens with more access to a vehicle during the initial months of licensure may face a greater risk of being involved in crashes. Thus, it is necessary to include vehicle access as a factor of driving exposure. The research team replicated Gershon’s operational definition of vehicle access for teens: if they were the driver of the vehicle for 50% or more of all trips in that vehicle, they were the primary driver; if they operated the vehicle for less than 50% of all trips, they were defined as sharing a family vehicle.

Time of Day

The time of day (TOD) variable was categorized into two levels: daytime and nighttime. Daytime was defined as any trip that started between the hours of 6:00 a.m. to 8:00 p.m. Any trip that started outside of this time period was defined as nighttime. Previous studies have shown crash risk is greater at nighttime compared to daytime for teenage drivers. Little is known about exposure to nighttime and daytime conditions during the learner’s permit and early independent driving periods.

Day of Week

The day of week (DOW) variable included seven days and was divided into two levels: weekday and weekend. Weekdays are Monday, Tuesday, Wednesday, Thursday, and Friday. Weekend days are Saturday and Sunday.

Sex

Sex was identified as a contributing factor to crash risk, with teen males having a higher crash risk than teen females (Gershon et al. 2018a). Study participants self-reported sex as female or male.

Passenger Presence

Data coders reviewed the camera view of the vehicle’s cabin to record whether passengers were present in the vehicle. Front seat passengers were also coded into four age groups: adults, children, teens, and no passengers.

Seatbelt Use

Data coders also recorded seatbelt use of both the driver and front seat passenger for these two studies. This coding was performed at the same time that passenger presence was reviewed and coded.

Dependent Variables

Crash and near-crash rates, kinematic risky driving (KRD), and percentage of time speeding were all identified as dependent variables of interest. These variables are described in more detail below.

Crash and Near-Crash

Crash events were defined as any contact with an object, either moving or fixed, at any speed in which kinetic energy was measurably transferred or dissipated. An object included other vehicles, roadside barriers, objects on or off the roadway, pedestrians, cyclists, or animals. Near-crashes were defined as events where the subject-vehicle drivers executed a rapid evasive maneuver to avoid a crash or departed the roadway. Crashes and near-crashes within each cohort were combined and redefined as SCEs. The crash/near-crash (CNC) rate was defined by the number of critical events divided by driving time in different driving phases and exposure groups.

KRD

This metric has been used in literature (Simons-Morton et al. 2011; Carney et al. 2010) and is typically defined as hard braking events and hard cornering events. For this study, the following a priori thresholds for KRD were used:

• Lateral acceleration (left/right) = ±0.5 g
• Longitudinal deceleration = −0.45 g
• Longitudinal acceleration = 0.35 g
• Yaw rate = ±6 degrees per second

The number of events per mile traveled was calculated and summed for each type of KRD event to arrive at the KRD rate per driver.

Percentage of Time Speeding

The percentage of time speeding was calculated by summing the amount of time that the vehicle speed was at least 10 mph greater than the posted speed limit and dividing by the duration of the trip. The percentage of time speeding per trip was then averaged by participant and independent variable condition.