Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
5 C H A P T E R 2 Research Approach Overview of NDS Data Resources, Strengths, and Limitations Work began in this project with a scan of the NDS database, using the SHRP2 InSight tool (https://insight.shrp2nds.us/), to gain an appreciation of the resources available to researchers, their strengths, and their limitations. NDS data are categorized according to four primary âselectionsâ pertaining to drivers, trips, events, and vehicles. While an exhaustive inventory of the data collected for teen SHRP2 NDS participants would very substantially exceed the scope of this project, our team completed an overview focusing on specific variables within each data selection that a recent literature review and input from the Young Driver Subcommittee (TRB ACH60) suggested may be productive to include in future analyses. Beginning with the Driver Data Selection, our initial focus was on driver demographics (gender, ethnicity, household income, living arrangements, etc.), and driving history (age received license, average miles driven per year, etc.). While noting that individual test scores and responses to questionnaires are not available at the driver level to researchers without executing a DUL, this data selection is comprised of 17 tables characterizing the SHRP2 NDS participants in terms of: ⢠Self-reported demographic characteristics and driving knowledge (Driver Demographic Questionnaire, Driving History Questionnaire, Driving Knowledge Survey); ⢠Visual, perceptual, and cognitive abilities (Visual and Cognitive Tests which include near and far acuity, day and night contrast sensitivity, depth perception, color, glare, and peripheral vision, as well as Visualizing Missing Information, Visual Search [Trail-making] A & B, Useful Field of View; Connerâs Continuous Performance Test, Clock Drawing Assessment); ⢠Physical condition (Physical Strength Tests which include hand strength and raw walk time); ⢠Psychological profile (Barkleyâs ADHD Screening Test, Risk Perception Questionnaire, Risk Taking Questionnaire); ⢠Self-reported health condition (Medical Conditions & Medications, Sleep Habits Questionnaire); and ⢠Two electronic online questionnaires administered after completion of in-vehicle data collection (Medical Conditions and Medications â Exit and Driver Exit Interview). The Trip Data Selection includes the Trip Summary table and the Time Series Database. The Trip Summary table captures summarizes the characteristics of each continuous trip traversed by each driver in the SHRP2 NDS, including variables that provide information about the type of trip, and vehicle kinematic data such as minimum and maximum speed, time of day, trip duration, number of brake activation, etc. The Time Series Database includes all variables collected continuously by the in-vehicle data acquisition system. The Event Data Selection includes the Event Detail table and the Post-Crash Interview. The Event Detail table lists all crashes, near crashes, and baseline events that were identified and analyzed by trained SHRP2 NDS video coders; this spans 6,772 events attributed to teen drivers. Variables included in the Event Detail table provide information on driver behaviors, environmental conditions, and roadway characteristics of
6 trip segments that were identified as a safety critical event or a balanced-sample baseline. Safety critical events (crashes, near crashes) were identified by participant report (either by data collection site report or pressing the critical incident button included on the vehicle data acquisition system), were reported by the vehicleâs data acquisition systemâs internal algorithm for kinematics and other triggers, or were identified by the analyst manually reviewing video for other purposes (e.g., driver ID, baseline event coding). The baseline sample was randomly stratified by vehicle and proportion of time driven over 5 mph. The Post- Crash Interview table, meanwhile, includes variables collected via an electronic questionnaire administered after a crash occurred that describes self-reported circumstances surrounding the incident (e.g., incident time and date, passenger demographics, environmental conditions, and driver behaviors). The Vehicle Data Selection, finally, contains the Vehicle Detail Table, which provides information describing each vehicle that was instrumented in the SHRP2 NDS study, as well as study site location. This Selection also includes how many and what types of vehicles were instrumented, the number of vehicles by model year, beginning mileage, and calendar month, and the amount of data collected during the study. The sheer breadth of the SHRP2 NDS suggests it will be an unprecedented resource for teen behavioral analyses supporting countermeasure development and evaluation. Specifically, this resource unifies data in the Driver Data Selection describing key characteristics (responses on scales measuring risk perception, risk taking, sensation seeking) for a large sample of teens, providing good statistical power, with objective measures of driver performance (in the Trip Data Selection) recorded on a continuous basis over an extended period. New insights regarding crash causation seem certain to emerge from analyses targeting the Event Data Selection, particularly with respect to the sources and consequences of driver distraction. At the same time, certain limitations must be acknowledged. For example, investigations focused on driver distraction may need to supplement the data elements already coded for a given event by viewing in- vehicle video of the actual behavior of a driver and his/her passenger(s); to do this will require physically traveling to one of two âsecure data enclavesâ as such data are protected from release to researchers. Similarly, other than for events, much of the information that defines the driving context such as weather conditions, visibility level, and traffic conditions is not available at the trip level, and must be coded by researchers themselves based on external camera views. All data extractions to allow analyses at the individual driver level must be performed by the SHRP2 data custodian under the terms of a Data Use License (DUL), for which a fee will be assessed. Even with a DUL, many variables of interest in the NDS cannot be read directly from a table, but must be derived by the data custodian; and certain information such as the driverâs date of birth is considered PII and cannot be shared with researchers. Last, if a research question concerns behavior for a particular maneuver type (e.g., left turns at intersections) or a particular type of facility (e.g., freeway driving) it is necessary to supplement NDS information with the Roadway Information Database (RID), a separate SHRP2 resource, using the LinkID variable.
7 Priority Research Questions A project activity that followed directly from the consideration of strengths and limitations of the NDS database for teen behavioral analysis was to identify priority research questions to which such analyses might be oriented, and that may be informed through naturalistic data. Without such a framework any particular content or attribute of the database can be considered a âstrengthâ only in an abstract sense, i.e., not necessarily being relevant to the present research agenda. To identify research questions, our project team initially drew upon a recent literature review undertaken for the National Highway Traffic Safety Administration (Mastromatto, Lococo and Staplin, 2018). This review highlighted a number of demographic and individual differences that have significant associations with teen crash involvement, plus contextual factors that have been implicated as influences on a teenâs âreadiness to drive.â Next, our team also carried out a brief survey of fifteen (15) members and friends of the Young Driver Subcommittee of the TRB Committee on Operator Education and Regulation (ACH60), to gain expert perspective on the level of priority for questions emerging from the literature review, and to augment this list with additional research questions or topics they perceived to be of high priority. The results of this exercise are shown below, by priority level from highest to lowest for ranked questions, plus supplemental questions suggested by subcommittee members. 1. Does the diversity of traffic/road environments teens are exposed to in their early driving experience predict crash and/or near crash involvement during unsupervised driving? 2. How does the impact of crash contributing factors (such as speeding, or distracted driving) change as teens gain driving experience? 3. To what extent are internal and external distractions risk factors in teen crashes and near crashes? 4. To what extent are driving risks increased (or different) for vulnerable teen populations, such as those on the autism spectrum, or diagnosed with ADD or ADHD? 5. How well do teens comply with GDL restrictions, and how does risk change for those who are noncompliant? 6. Do newly licensed teens have higher crash rates in States where the entry age for unsupervised driving is 16 or younger than in States with a later entry age (e.g., 16.5 or 17)? 7. How reliable are teen self-report data regarding various risky driving behaviors (e.g., speeding, cell phone use)? 8. Does chronological age vs. experience behind the wheel serve as a better predictor of crash (event) involvement and/or severity? ⢠How do driving performance factors change as teens gain experience (e.g., improved visual search, smoother acceleration-braking)? ⢠To what extent are driving risks increased (or different) for teens from households with lower vs higher socio-economic status teens, for those living in more rural/remote areas vs urban-suburban, and/or according to race and ethnicity? ⢠What explains the sharp decline in crash rates over the initial months of driving? ⢠In what (and how many) instances can a 'near crash' be considered a success (i.e., a crash avoided) rather than a failure? ⢠How does young driversâ risk of crash involvement (or near crash involvement) differ when driving in locations (specific roads or segments) where he or she drives regularly vs less frequently? ⢠What is the increased risk associated with carrying one or more passengers? ⢠What is the frequency and what are the circumstances of drowsy driving? ⢠What is the frequency of speeding and how does the increased risk for teens compare to the increased risk for adult drivers?
8 Inventory of NDS Data Elements An inventory of SHRP2 NDS data elements was needed to operationalize measures/ measurement constructs invoked in the high priority research questions. Our team developed an inventory associated with key measurement constructs embedded within the preceding list of research questions, comprised of variable definitions and data elements from the tables displayed on the SHRP2 InSight website data dictionary, and also from the RID. The inventory included the variables necessary to define each measure, the SHRP2 table and variable needed, and any relevant calculations including what those calculations are (if known) and whether the variables can be defined by the researcher, or must be derived by the SHRP2 data custodian. It should also be noted that a large portion of the variable definitions use information from the Event Detail table. While this table includes a breadth of circumstantial and environmental data, it does so for a limited subset of trips, and only for a segment of each trip. The specific NDS variables included in the inventory are indicated by the cells checked in the matrix shown in Figure 1. The independent, dependent, and discriminating variables for each research question are shown with bolded text in the row where each appears. Columns in this matrix specify measures and measurement constructs that permit the operationalization of these variables. Clearly, this is not exhaustive â additional research questions would require additional columns keyed to new measures/constructs â but the matrix serves to illustrate how the inventory was constructed. As described in a later section, this inventory provided the starting point in developing each Data Specification submitted with the Research Needs Statement deliverables.
9 Figure 1. Inventory associated with key measurement constructs. Key Measures/Constructs for Teen Behavioral Analysis Components of Candidate Research Priorities to be Operationalized Using the NDS Cra sh Ri sk Ev en t S ev eri ty Ag e Ex pe rie nc e Po ssi ble AD HD Div ers ity of Tr aff ic En vir on me nts Div ers ity of Ro ad En vir on me nts Int ern al Dis tra cti on s Ex ter na l D ist rac tio ns GD L C om pli an ce Sta te Cra sh Co ntr ibu tin g F ac tor s Ris ky Be ha vio r Re spo ns es to Ris k T ak ing Q ue sti on na ire Dr ivin g P erf or ma nc e So cio eco mi c S tat us Ra ce Eth nic ity Nu mb er of Oc cu pa nts Dr ow sy Dr ivin g Sp ee din g Fre qu en t L oc ati on Dr ivin g C irc um sta nc es Vid eo Re vie w Does the diversity of traffic/road environments teens are exposed to in their early driving experience predict crash and/or near crash involvement? ⢠⢠⢠⢠⢠How do crash contributing factors change as teens gain driving experience? ⢠⢠⢠⢠To what extent are internal vs. external distractions risk factors in teen crashes and near crashes? ⢠⢠⢠⢠To what extent are driving risks increased (or different) for vulnerable teen populations, such as those on the autism spectrum, or diagnosed with ADD or ADHD? ⢠⢠⢠How well do teens comply with GDL restrictions, and how does risk change for those who are noncompliant? ⢠⢠⢠⢠Do teens have higher crash rates in States where the entry age for unsupervised driving is 16 than in States where the entry age is 16.5? ⢠⢠⢠How reliable are teen self-report data regarding various risky driving behaviors (e.g., speeding)? ⢠⢠⢠Does chronological age vs. experience behind the wheel serve as a better predictor of crash (event) involvement and/or severity? ⢠⢠⢠⢠How do driving performance factors change as teens gain experience? ⢠⢠⢠To what extent are driving risks increased (or different) for teens from households with lower vs higher socio-economic status teens and/or according to race and ethnicity? ⢠⢠⢠⢠⢠What explains the sharp decline in crash rates over the initial months of driving? ⢠⢠⢠⢠In what (and how many) instances can a 'near crash' be considered a success (i.e., a crashavoided) rather than a failure? ⢠⢠⢠How does young driversâ risk of crash involvement (or near-crash involvement) differ when driving in locations (specific roads or segments) where he or she drives regularly vs less frequently? ⢠⢠⢠What is the increased risk associated with carrying one or more passengers? ⢠⢠⢠What is the frequency and what are the circumstances of drowsy driving? ⢠⢠⢠What is the frequency of speeding and how does the increased risk for teens compare to the increased risk for adult drivers? ⢠⢠â¢
10 The strengths of the NDS database notwithstanding, our research team considered other sources of data that could potentially be useful for operationalizing variables in teen behavioral analyses, and examined the potential for accessing these supplemental data sources. Three sources were considered in this regard: 1. State-level licensing data for more precise analyses involving teen driving experience. As one of the most persistent questions concerning teen safety remains the relative influence of advancing age/maturity level versus amount of experience behind the wheel, and given the limitations on gauging experience among teen NDS participants (only to the nearest 6 months, and only based on subjective questionnaire responses), State-level data defining individual driversâ GDL status and milestone dates could be extremely valuable as a predictor variable for a wide range of events and kinematic measures with safety significance. This data source was ruled out on the basis of confidentiality provisions for NDS participants, even given cooperation by a State DOT. 2. Other naturalistic teen driving studies with data at the individual driver level. The potential to aggregate SHRP2 NDS teen driver data with data from other naturalistic studies with common measures of behavior and safety would not only increase statistical power, but could make available data that inform analyses of, for example, in-vehicle distraction, that can only be obtained (for nonevent trips) under a SHRP2 DUL through a multi-step process. An analytic database was developed as part of the NIH- sponsored Young Driver Naturalistic Driving Study and should be publicly accessible. This data source was ruled out, however, because an investigation of the protocols and procedures involved in applying for data access were deemed outside the scope of this project. 3. Results of post-SHRP2 reductions of NDS data. A means of sharing verifiable NDS-sourced data between researchers is the Dataverse, which houses over 100 datasets. Each is listed on the InSight website with a brief abstract of the study it was originally used for. File metadata including a dictionary of all included variables is also available for each dataset. The Dataverse is searchable by publication year, keyword, data type, and author name. NDS data reductions performed by VTTI staff that are not associated with a published study as is the case with the contents of the Dataverse, but are of clear interest with respect to teen behavioral analysis, were also considered. With respect to the Dataverse, a search using the term âteen driverâ identified three potentially useful post-SHRP2 reductions of NDS data: ⢠Differences between younger and elderly driver behavior and how they interact with roadway infrastructure in a naturalistic setting; ⢠Teen Driver Crashes: Exposure Results from SHRP 2 Naturalistic Driving Study; and ⢠The Association of Teenage Personality and Psychosocial Factors with Crash Rates. Within the scope of this project we were unable to confirm that any of the above datasets analyzed individual driver age. Rather, age was included as a binned (categorical) variable. Accordingly, while these datasets could be quite valuable in other respects, without further processing by the SHRP2 data custodian they would be of limited utility for any teen behavioral analyses using within- or between-driver designs. 4. Most promising as a supplementary data source is another NDS data reduction by staff at the data custodian, i.e., the reduction of eye glance location on a frame-by-frame basis to produce a time series of eye glance locations for a behavioral epoch (crash, near crash, or baseline). This dataset was created via a manual data coding process, whereby a team of data reductionists used all available video context clues (using videos of the forward roadway, rear roadway, driverâs face, and driverâs hands/dashboard) to assess the driverâs glance behavior and assign the best glance location category to each video frame. The glance location data, while very context-reliant, and even though limited to crash, near crash and baseline events, hold particular appeal for analyses in which driver distraction serves as either an
11 independent or dependent variable. In terms of filling a critical gap in the NDS, the importance of being able to specify not only a teen driverâs glance direction but actual and instantaneous glance location is unrivaled as an allocation-of-attention measure. Narrowing the Recommended Research Agenda After gaining an understanding of the data resources within the NDS, then building an inventory of specific data elements well suited to operationalizing the key variables that define a set of candidate research priorities, the penultimate project task was to narrow the high priority research questions to those few for which a formal TRB Research Needs Statement would be developed. Our teamâs judgment of the perceived scientific value of analysis outcomes within the framework of the GDL model led to a final prioritization of the following four questionsâwith accompanying rationale. ⢠Does the diversity of traffic/road environments teens are exposed to in their early driving experience predict crash and/or near crash involvement? Studies have shown that supervised practice driving is characterized mainly by routine driving trips on familiar roads in relatively benign conditions (Goodwin et al. 2010; Ehsani et al. 2017). Teenagers seldom obtain supervised practice in potentially challenging situations such as highways, inclement weather, darkness, heavy traffic, or country roads (Goodwin et al. 2010). The total amount of supervised practice that teens obtain varies widely; but a majority fails to meet the requirements for nighttime driving (Ehsani et al. 2017). It is widely believed that driving conditions change when teenagers make the transition to unsupervised driving. The fact that a supervisor is no longer present means the teenager is, for the first time, fully in charge of the vehicle. The absence of the supervisor may encourage many kinds of expressive and/or impulsive behaviors that the teenager formerly avoided. Additionally, driving locations and conditions may change because trips that were previously elective are now compulsory. A newly licensed driver can no longer easily transfer driving to a supervisor simply because of inclement weather or because s/he is running late. Little is known about how driving exposureâboth the amount of driving and driving conditionsâchange when teens make the transition to unsupervised driving. One study that tracked 38 families during this transition found that teens drove more often in darkness and inclement weather once driving unsupervised (Goodwin et al. 2011). An important question is whether teenagers who are exposed to greater diversity of traffic and road environments early in their driving career have lower crash involvement than those who are exposed to less diversity. Supervised driving data are not available in NDS; however, it is possible to compare the diversity of experience in the early months of unsupervised driving with later months, and to examine the association of exposure to greater diversity with crashes and near crashes. The NDS allows researchers to examine the relationship between diversity of road types and crash and/or near crash involvement across time. Also worth noting is that while the Trip Summary table includes data on the percent of trips taken at various driving speeds, on roads with various speed limits, and on various road types, details like traffic flow and locality (e.g., open residential, commercial, industrial) are only recorded for trip segments in the Event Detail table, which includes at least one baseline trip segment for most young drivers (n = 2654) as well as each trip segment coded as an event. Analyses directed to this question will also inform researchers as to whether the risk of a crash or near crash for new drivers differs when driving in locations (specific roads or segments) where s/he drives regularly versus locations where s/he drives less frequently. Although police-reported crashes are relatively
12 rare in naturalistic studies of teenage drivers, anecdotal evidence from these studies suggests crashes often occur when teenagers are driving in locations and/or situations with which they are unfamiliar. The NDS database is well suited to examine this question. ⢠How do crash contributing factors (such as speeding or distracted driving) change as teens gain driving experience? Teenagers have very high population-level crash rates that decline sharply during the first 6-12 months of independent driving (Mayhew et al. 2003; McCartt et al. 2003). Crash rates continue to decline, albeit at a slower pace, for several more years. Exactly what new drivers are learning during this initial period that brings down their crash rates (risk) is not well understood. As noted by one young driver researcher: âUntil a better understanding of what changes during the first 18-24 months of driving is developed, traffic safety practitioners will be greatly handicapped in their efforts to develop programs, policies or other interventions to bring young driver crash rates more closely in line with those of experienced adult driversâ (Foss et al. 2011). Certain types of crashes appear to decline rapidly during the first year of unsupervised driving, such as making a left turn from a roadway or driveway into traffic, running off the road to the right, and failing to yield (Foss et al. 2011). Other types of crashes decline more slowly (e.g., rear-end collisions) or even increase over time (e.g., alcohol-related crashes). More research is needed to understand how crash contributing circumstances change during the initial months as novice drivers gain experience. The NDS includes information on a number of relevant factors that contribute to crashes or near crashes. This includes speeding, distracted driving, short headways to lead vehicles, high lateral and longitudinal acceleration rates, and others. Analyzing changes in performance with increasing teen driver experience, with an emphasis on crash contributing factors, is feasible using NDS data. However, there are limitations to this analysis path, specifically the age at which the participant received his/her license is self-reported and does not identify GDL-restricted licenses. SHRP2 offers a number of crash contributing factors common among young driver crashes, including but not limited to driver speed (including coder judgments about whether the speed was within the limit but too fast for conditions), time of day, roadway alignment, and others. Kinematic data such as brake activations and lateral and longitudinal acceleration/deceleration also support analyses of changes in driver performance with increased driving experience. Within-subject analysis is an option with this dataset as well. As one example, numerous drivers age 16-19 experienced multiple events (e.g., 1182 near crashes vs. 553 drivers overall); though, it should be noted that analyses would be limited to 1 to 2 years in half-year increments. ⢠To what extent does driving distraction contribute to teen crashes and near crashes? Distracted driving has become a growing concern over the past few decades with the advent of smartphones and other technologies that have the potential to divert attention from the task of driving. However, the contribution of distracted driving to crashes is not well established. It can be challenging for an officer to determine whether a driver was distracted at the time of crash. For that reason, it is widely believed that distractions are underreported in crash records. Given the limitations of crash data, researchers have turned to observational methods to examine the prevalence and increased risk posed by non-driving- related tasks. For example, a naturalistic study of 42 newly licensed teenage drivers in Virginia found the risk of a crash or near crash increased substantially when the driver was dialing a cell phone, reaching for a cell phone, or sending or receiving text messages (Klauer et al., 2014). Many other studies have
13 demonstrated that cell phone useâand texting in particularâis associated with decrements in driving performance (McCartt, Hellinga, & Braitman, 2006; Caird et al., 2008). Another concern specific to teenagers is the potentially distracting influence of teenage passengers. In another naturalistic study of 52 high school age drivers in North Carolina, loud conversation and horseplay among passengers was more strongly associated with near collisions and high g-force events than electronic device use and other potentially distracting behaviors (Goodwin et al. 2012). The NDS data provides an opportunity to address a number of questions related to teenagers and distracted driving: Which potentially distracting driver behaviors are most common among teenage drivers? Under what conditions do distracted driving behaviors most commonly occur (e.g., time of day, day of week, amount of traffic, presence of passengers)? Which distracted driver behaviors are most strongly associated with crashes and near crashes? How does driver behavior change in presence of teenage passengers?â The SHRP2 NDS objectively identifies driver distraction immediately before a crash or other event. This is a clear strength of this dataset, as most other crash data are derived from retrospective police reports. Each trip segment in the event â whether it is a crash, near crash, baseline event, etc. â includes up to three driver behaviors (including distraction) and up to three secondary tasks that identify what the driver was distracted by (e.g., cell phone use, passenger interaction, external distractions). Also included is the secondary task outcome variable, which signifies if the coder determined the secondary task contributed to the event sequence or severity. Event-level video coding was conducted on random, nonevent trip segments (âbalanced-sample baselineâ and âadditional baselineâ) and secondary tasks were coded, both for events (crashes, near crashes) and nonevents (baselines); this allows researchers to examine whether specific behaviors were likely to have an effect on event severity. Finally, the emerging reduced dataset containing eye glance location on a frame-by-frame basis for crash, near crash, and baseline epochs presents an unprecedented opportunity to investigate the antecedents and consequences of driver distraction. ⢠To what extent are driving risks increased (or different) for vulnerable teen populations, such as those on the autism spectrum or those diagnosed with ADD or ADHD? Young drivers with developmental disabilities may be at increased risk for crashes due to impairments commonly associated these conditions. In recent years, a growing body of research has examined driving risks for teens with autism spectrum disorder (ASD), attention deficit disorder (ADD), and attention deficit hyperactivity disorder (ADHD). Rates of licensure are substantially lower among teenagers with ASD than teenagers without ASD (Curry et al. 2018). Although all novice drivers are at increased risk because of inexperience, teenagers with ASD face additional challenges that include difficulties identifying social hazards (e.g., pedestrians), slower reaction times, and poorer situation awareness skills (Silvi et al. 2018). Similarly, research suggests teenagers with ADHD display a number of decrements in driving skill and are more likely to engage in distracted driving and to drive at higher speeds (Bishop et al. 2018). Medications may improve driving performance for teenagers with ADHD (Classen and Monahan 2013). The NDS provides an opportunity to examine this vulnerable teen population by accessing responses to Barkleyâs ADHD Screening Test (i.e., not ADHD diagnoses). One potentially useful approach is to compare predictor-criterion relationships for teenagers so categorized in the top and bottom quartiles, with respect to a variety of driving behaviors and crash/near crash risk.
