Integration of Analysis Methods and Development of Analysis Plan (2012)

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.

45 A p p e n d i x A introduction Advances in data collection techniques, such as instrumen- tation suites that capture naturalistic driving behavior, allow problems relating to transportation and driving behavior to be examined in a way that was not previously possible. In recent years, capturing naturalistic driving behavior has become more feasible and cost-effective. Naturalistic refers to a method of observation that captures driver behavior in a way that does not interfere with the various influences that govern those behaviors. This in-vehicle method allows for the observation of drivers in their own environment and can pro- vide deeper insight into the factors affecting driving safety. Naturalistic driving studies provide data that are most likely to generalize to actual driving situations; however, such studies provide the least control, making it difficult to identify causal mechanisms without ambiguity. Naturalistic driving studies by their nature collect a large amount of data and have the potential to produce some of the world’s largest databases. The size and challenge of identifying mechanisms underly- ing driving safety from these databases complicate the data analysis techniques used to address research questions related to driver safety. The spatial, dynamic, and temporal nature of driver behavior adds to the complexity of such analyses. Sifting through the plethora of data collected in such studies can be extremely demanding and will provide little insight if data are improperly sampled, integrated, and analyzed. Reviews of Safety projects S01 and S05 The outcomes of Safety Project S01, Development of Analysis Methods Using Recent Data, and Safety Project S05, Design of the In-Vehicle Driving Behavior and Crash Risk Study, were reviewed in this phase. The S01 contractors included the University of Minnesota Center for Transportation Stud- ies, the Pennsylvania Transportation Institute, the University of Michigan Transportation Research Institute, and the Iowa State University Center for Transportation Research and Education. From these reports, 56 specific research questions were extracted. In addition, 392 questions were extracted from the S05 report from the Virginia Tech Transportation Institute (VTTI). These research questions were initially compiled into sep- arate matrices. Each specific question (described in Appen- dices B and C) was separated into categories for classifying the variables corresponding to those factors identified in Figure A.1: environmental, driver, vehicle, roadway, and nondriving activities. This systems-based perspective frames the specific research questions to identify how multiple fac- tors influence the risk of collisions. These factors are not independent of each other but may actually have different contributions depending on the inter- actions of different factors (Donmez et al. 2006; Neyens and Boyle 2006). Categories are functionally defined below. Driver characteristics include attention, perception, situa- tion assessment, and motor control (Lee 2006). These charac- teristics vary among people and are influenced by individual differences such as age and driving experience. Drivers’ psychological functioning also varies across time as a func- tion of fatigue and alcohol or drug impairment (which are identified as the driver state in Figure A.1). In addition, nondriving-related activities, especially driver distraction, also influence driver attention, perception, situation assess- ment, and motor control. Technology, such as cell phones, MP3 players, and Internet connectivity, makes possible a wide range of nondriving activities that can distract drivers. The effect of such technology on crash risk depends on more than the vehicle characteristics alone. It is very dependent on how drivers engage the technology relative to the road- way characteristics (Lee 2006). Crash risk often depends on the interaction of driver and roadway characteristics, as demonstrated by the overrepresentation of older drivers in intersection crashes. Summary of Phase I of SHRP 2 Project S02

46 Vehicle characteristics influence driver behavior; for exam- ple, advanced braking systems influence the braking effec- tiveness of the driver. Rear-end collision avoidance systems have been shown to have a safety effect in reducing crash fre- quency (Lee et al. 2002). Other technologies, such as adaptive cruise control or crash-warning systems, change the driving task more fundamentally and may lead drivers to disengage from the driving task and lose situation awareness (Stanton and Young 2005; Young and Stanton 2004). The interaction between the driver and the vehicle is a critical aspect of the safety of the driver–vehicle system. Roadway characteristics influence the safety consequences associated with changes of vehicle state. A narrow lane or shoulder magnifies the safety consequences of a deviation from the center of the lane. Isolated road characteristics such as shoulder width and curve geometry can influence crash risk, but the interaction of these factors with driver charac- teristics may have a more powerful influence on risk. Lane width and shoulder treatment influence lane-keeping behav- ior, and drivers’ ability to anticipate curves based on signage and topography, as well as other factors, influence road- departure crashes. Environmental characteristics influence the safety conse- quences of driver characteristics, vehicle dynamics, demands of nondriving tasks, and roadway characteristics (Lunenfeld and Alexander 1990). The environment includes traffic den- sity, ambient lighting, weather conditions, and pavement sur- face condition. These elements not only represent situational factors considered (or ignored) by the driver in planning behavior (or triggering errors), but they also define bound- aries for the operation of the vehicle for a particular roadway (e.g., ice reduces the speed at which a vehicle can negoti- ate a curve without sliding). Thus, research questions must be framed to consider the relevant context of the driving environment. Feedback loops in Figure A.1 show the influence of past events on a driver’s response to future events. One example is how drivers’ awareness of their own state can influence their behavior—people adapt and drive more conservatively if they perceive themselves to be impaired. Likewise drivers’ attention and perception are strongly influenced by expec- tations, such as typical cues that indicate horizontal curves (Lunenfeld and Alexander 1990). These expectations reflect both very recent experiences that may have occurred only seconds before and long-term exposure to similar situations (driver expectation) over months or years. It is therefore important to consider research questions that relate these human factors to driver behavior and crash risk. Safety consequences can be viewed from two perspectives: inside-out and outside-in. Inside-out refers to a driver-centric perspective and focuses on driver-related data relative to safety critical events. For example, an increase in visual attention demand (e.g., eyes-off-road time, complexity of the envi- ronment, or driving task) has been associated with dimin- ished vehicle control. The outside-in perspective focuses on environment-related data relative to safety critical events. Crash migration is a consideration because traffic improve- ment projects tend to improve the acute problem (an inside- out perspective) without attending to changes that may occur in surrounding areas (an outside-in perspective). This con- sideration has been examined by Griffith (1999), Shen and Gan (2003), and Smith and Ivan (2005). Their studies have Perceptual, cognitive, and motor characteristics Driver state Environmental Driver Distractions Vehicle state Proximity to safety boundaries Nondriving activities Dynamics and crash characteristics Vehicle Intersection incidents Lane departures Roadway departures Safety consequences Geometry, vehicles, and control devices Roadway Traffic density, weather, season, and time of day Figure A.1. The dynamic relationships between driver, vehicle, roadway, and environment and the resulting safety consequences.

47 typically centered on rumble strips and as such have not seen definitive effects. However, the benefit of the naturalistic driv- ing study is the ability to observe whether changes in travel patterns and behavior result from changes in the system (e.g., the implementation of speed bumps, traffic signals, new roadway networks, and construction zones). Most studies on traffic improvements focus on before and after interven- tions of the acute area, but such micro examination does not provide insights into any larger safety problem that may have occurred due to changing travel patterns. The outside-in per- spective is analogous to a bird’s-eye-view of the vehicle in that it can allow observation of the contributions to crashes that arise from factors beyond the driver’s control. The data collected from this study can be aggregated to dif- ferent levels (e.g., event, trip, or driver levels). For example, if the data are aggregated to the trip level, questions regard- ing trip-specific research questions can be addressed. At each level of data reduction some variables are consistent across the whole trip (e.g., the driver’s gender), and other variables are not consistent across the trip (e.g., distraction). The cat- egorization of the factors addressed in each question (from Figure A.1) was refined further to identify static and dynamic characteristics within each category of variables. Dynamic characteristics are variables that can vary during the time period of interest (e.g., a trip or an event). For exam- ple, a driver can be distracted for only a portion of the trip; hence distraction is a dynamic variable. The curvature of the road, the type of pavement markings, and the average vehicle speed would also change within a trip. Static characteristics are variables that are relatively con- stant across the time period of interest. For example, at the trip level, driver’s age, the vehicle’s make and model, and the environment (e.g., region or state) are static character- istics. At the event level, the driver’s age and gender and the vehicle’s make and model, as well as the road type, pavement markings, and visibility, are static. Using the dynamic–static perspective identifies a combina- tion of driver factors and nondriving activities that can be col- lapsed into a single category that represents both dynamic and static influences at the trip and event levels (see Figure A.2). Cells in Figure A.2 noted as N/A (not applicable) have no static or dynamic variables at the specified level of aggregation. At the trip level, there are no static roadway variables because, from the driver’s perspective, the roadway is expected to change continuously within a trip. However, at the event level (e.g., lane departure) the roadway variables are static (e.g., road curvature, pavement markings) during the event. Likewise, at the event level, there are no dynamic environmental vari- ables. In this context an event is a 5- to 15-second period sur- rounding a notable state change. The distinction of dynamic and static is based on the time constant of the variables rela- tive to the time period over which the data are aggregated. As an example, visibility can change on the order of minutes, making it a dynamic variable at the trip level. But at the event Driver Vehicle Roadway Environment Dynamic Static Dynamic Static Dynamic Static Dynamic Static Distraction Age Gender Speed Road type Pavement markings Make Model VisibilityN/A Region State (a) Trip level Driver Vehicle Roadway Environment Dynamic Static Dynamic Static Dynamic Static Dynamic Static Gaze location Age Gender Lane position N/AMake Model N/ARoad type Pavement markings Region State Visibility (b) Event level Figure A.2. Examples of dynamic and static factors that relate to driver, vehicle, roadway, and environmental variables at the trip level and event level.

48 level, visibility is considered static because it changes very slowly within the time period of the event. development of Global Research Questions Research questions from the S01 and S05 projects were inde- pendently organized into five groups based on common themes related to • Similar safety outcomes; • Similar explanatory variables; • Relationship to crash surrogate measures; • Dynamic and static characteristics; and • Combinations of these four subjects. Based on the commonalities within each group of research questions, a broader set of global research questions (GRQs) was generated that represents the summation of the individ- ual specific research questions (Figure A.3). All original S01 or S05 questions were unchanged, except to correct typographi- cal errors. The S01 and S05 questions were combined to form one matrix that represented all of the research questions. The outcome of this task was 27 GRQs (see list in Table A.1 on p. 52). Several S01 contractors also provided feedback on the categorization and wording of the global questions. Because the GRQs were developed based on the specific S01 and S05 questions in each grouping, some of the global ques- tions may not actually be as broad as one might expect. For example, one global question (“How does the turn-lane con- figuration influence behavior and crash risks?”) is specific to turn lanes as a result of the nature of the seven site-specific questions from the S05 contractors that did not seem to fit in other global questions but were inclusive of turn lanes (e.g., protected and unprotected phases, turn lanes, and bays). During this process, the wording of the GRQs and the clas- sification of the original research questions were revised to ensure proper grouping and representation of the variables included in each S01 and S05 question. For example, at one point the specific questions related to GRQ 18 all began with “What is the relative risk of. . . .” However, after further exam- ination of each question in this group, the team recognized that the major issue being addressed relates to the relationship of various factors. Thus, the global question changed from “What is the relative risk of specific factors given the driver’s involvement in nondriving-related activities?” to “What are the interrelationships of environmental, road, and driver fac- tors with nondriving-related activities?” decision Tree for prioritization These research questions can be addressed from many broad perspectives, and they capture several analytical approaches so that researchers who enter the process at a later date will be able to dive right in. For this project, a decision tree was used to filter the global questions from a safety perspective, which is the most relevant consideration for this overall research program. Several prioritization schemes or sets of criteria based on one of three perspectives (applied, basic, or meth- odological) are possible for ranking research questions: 1. Applied—Questions focused on countermeasure develop- ment or evaluation, such as GRQ 2, which relates to road- way features and lane keeping. This perspective will provide immediate results that are of greatest interest to policy mak- ers and state and federal departments of transportation. 2. Basic—Questions leading to new information on driver behavior, such as GRQ 1, which relates to the effects of dynamic driver characteristics on crashes. This perspec- tive may well lead to better understanding of important issues about which little is known currently, such as the role of driver behavior on the likelihood of engaging in distractions. While at the core of the whole project, answers to questions from the basic perspective may not lead to immediate countermeasures, but they will be of greatest interest to researchers and policy makers. 3. Methodological—Questions relating to surrogate develop- ment and data reduction needs. This perspective will also consider whether stand-alone data sets can be developed that will address a range of analytical needs or whether each detailed analysis requires separate data reduction. Questions from the methodological perspective are vital for address- ing both applied and basic questions as there is currently no Figure A.3. Process of generating representative research questions through identifying similarities between research questions.

49 consensus regarding the most effective analytical techniques for extracting information from naturalistic data. The goal was to prioritize the GRQs with an emphasis on safety and the application of mitigation strategies. The deci- sion tree (Figure A.4) emphasizes questions that require data about drivers, those that have the potential to support safety interventions, and those that address large-scale morbidity and mortality consequences. Each GRQ is evaluated using the questions in this decision tree to determine a priority rank- ing. The priorities range from lowest priority (ranking of 8) to highest priority (ranking of 1). This decision tree will retain all questions regardless of ranking as it relates to both the basic science and analytical perspectives. This decision tree is designed to account for any ques- tions that can be addressed in a naturalistic study regardless of questions developed as part of S01 and S05. That is, it is designed to account for any safety-related future questions that may be generated as the SHRP 2 project progresses and as additional research questions are developed. The following sections discuss the questions that are represented as nodes in the decision tree. Decision Node A: Is the Question Safety Relevant and Focused? Questions that are not safety relevant or are too broad to address are unranked because they are out of the scope this project. The question progresses to the next node in the decision tree if the unit of analysis or outcomes relate to crash risk or driver behav- ior (which can then be related to some safety consequence). Some of the research questions raised were too broad to be ranked with the remaining questions in this decision tree. Others have an indirect relationship to safety or are not directly linked to the goal of the SHRP 2 Safety effort: quan- tifying crash risk as a function of driver, vehicle, roadway, or environmental factors. These questions have been included in an unranked priority listing and should be reviewed by possible users since they do contain ideas for future research using the SHRP 2 naturalistic driving study data. Decision Node B: Does the Question Relate to a Potentially High Number of Fatalities? If the question does not relate to a high number of fatal crashes, it is given a priority of 7. The criterion for determin- ing high numbers of fatalities is whether the issue is known to relate to a specific crash type that encompasses a high number of fatalities or to one of the following factors related to mor- bidity and mortality in motor vehicle crashes (Evans 2004): • Speeding; • Alcohol; • Safety belt usage (which relates to injury severity); • Driver inattention; and • Fatigue. Factors related to high numbers of fatalities but that are not driver behavior–related will progress to the next node of the decision tree for consideration. None of the GRQs included in this report filtered out at this level. However, the goal is to use this decision tree as a future theoretical framework, and number of fatalities would be an important factor to consider. Decision Node C: Does the Question Require Data Beyond What Are Currently Available? If the research question can be addressed using existing data sources (e.g., crash data, existing simulator data, or road studies), then it is given a priority of 6. However, if the ques- tion is related to factors that have not yet been captured (e.g., aggressiveness), are underreported or inaccurately reported in the current crash databases as a result of judgments made by the officials reporting (e.g., presence of driver distraction factors), or cannot be examined in controlled studies (e.g., changes in weather conditions), then the question progresses to the next node in the decision tree. Decision Node D: Does the Question Require Data About Driver Behavior? The value of naturalistic studies is their ability to capture information about the driver. Thus, if the question relates to information about the driver’s behavior (e.g., scan behavior, steering wheel movements), then the question progresses to Node E and is given a higher priority. Otherwise, it stops here and is given a priority of 5. Factors related to the driver can include driver action (e.g., braking, scanning behavior, speeding, drinking, not wearing a restraint) or cognitive state (e.g., inattention, alcohol impair- ment, fatigue). Driver behaviors like these are not accurately measured in most crash investigations in that they are based on after-the-fact judgments rather than direct observations. In addition, since a driver’s performance and behavior can change over time, there is a need for longer-term measurements. Decision Node E: Are Naturalistic Data the Best Way to Address This Question? There are many ways to capture information on crashes, including test tracks, simulators, and observational data (e.g., crash databases). If a naturalistic study is not the best alterna- tive, then the question stops here and is given a priority of 4. If the question is best addressed using naturalistic data, then it progresses to Node F and is given a higher priority. This node is included to ensure that time- and evidence-based

50 Research question Does the question require data about driver behavior (e.g., scan behavior, steering wheel movements)? Priority 6Is the question safety relevant and focused? Does the question relate to a potentially high number of fatalities? Does the question require data beyond what are currently available? Are naturalistic data the best way to address this question? Can straightforward intervention be implemented? Will answers to this question provide broad insights into driving safety? Priority 7 Priority 4 Priority 5 Unranked Priority 3 Priority 2 Priority 1 (highest) Yes No No NoYes No Yes No No No Yes Yes Yes Yes A B C D E F G Figure A.4. Decision tree for prioritization of the research questions based on a safety perspective.

51 data (which can only be captured in a naturalistic study) are elevated in priority. Decision Node F: Can a Straightforward Intervention Be Implemented? If the GRQ can be addressed with a straightforward interven- tion, then it progresses to Node G. A straightforward inter- vention is defined here as an intervention for which a solution is (more or less) known, even though the actual implemen- tation may be easy or difficult. Potential interventions can include infrastructure improvements, in-vehicle system enhancements, educational programs, and policy implica- tions (e.g., installing rumble strips, enhancing an in-vehicle technology, or developing training programs). However, if no interventions could be developed based on the GRQ, then it is given a priority of 3. Decision Node G: Will Answers to This Question Provide Broad Insights into Driving Safety? If the GRQ provides some fundamental understanding of the basic mechanisms of motor vehicle crashes and driving behavior that can be generalized to other situations beyond the specific question addressed in the study, then it is given a priority of 1 (the highest priority); otherwise it is given a priority of 2. prioritized Global Research Questions Table A.1 prioritizes the GRQs listed above by using the deci- sion tree. Thus, the nonsafety-relevant questions are located in the lowest portion of the table. Those questions that can be answered using means outside of the SHRP 2 project are located in the lower half of the table. Full GRQs can be found in Appendix B and Appendix C. The priority summary shows how each of the global questions filtered through the decision tree. Since the deci- sion tree was designed to address a broad set of research questions, there were several priority levels (Priorities 6 to 8) that were not represented from this sample. This was a result of the largely infrastructure-based questions from S01 and S05. Future questions will likely emerge as part of the S02 refinement. If those new questions fit into the exist- ing global questions, then the priority will already be set. If a new question is outside of the current set of global ques- tions, a new high-level question would be created and run through the decision tree. Some of the GRQs provide insights to driver behavior or driving performance but were left unranked because they are not directly related to safety. These questions are as follows: • How does the speed that a driver selects influence the driv- er’s other behaviors or actions? • How do roadway features influence driver performance and behavior? • How does driver fatigue affect driving performance? • How do the number and type of passengers influence the driver’s behavior? • How does inattention affect driving behavior? Although some readers may feel that these questions deserve a higher priority, other similar questions directly related to safety were ranked more highly based on the deci- sion tree. These include the following: • How does driver fatigue influence the likelihood and type of crashes? (Ranked Priority 1.) • How does driver distraction influence crash likelihood? (Ranked Priority 1.) • How do roadway features influence crash likelihood? (Ranked Priority 2). In addition, one GRQ (Question 9) consists of two questions: • GRQ 9a: What variables or pre-event factors are the most effective crash surrogate measures? • GRQ 9b: What explanatory factors are associated with crashes or crash surrogates and what analytical models can be developed to predict crash or crash surrogates? These questions are related to crash surrogate measures, effective associations between variables, and to analytical models based on these associations. Thus GRQ 9a and 9b are grouped such that this connectivity is salient. Challenges and Limitations: Verification of Research Themes with a Lexical Analysis Accurately capturing the central themes of the research ques- tions represented a major challenge that arose when develop- ing the GRQs from the diversity of questions from the S01 and S05 contractors. The questions from the S01 contrac- tors were developed from an analytical perspective, while the S05 contractors used a data-user perspective. In both cases, a large number of specific research questions were reviewed to identify representative GRQs. The resulting global ques- tions reflect the subjective judgment of several researchers and thus may reflect cognitive biases and limits. At the least,

52 Table A.1. Initial Prioritization of the Global Research Questions Based on a Safety Perspective Question General Rating Initial Priority Global Research Question 1 1* How do dynamic driver characteristics, as observed through driver performance measures, influence crash likelihood? 2 1* What impacts do roadway countermeasures have on lane-keeping performance? 3 1* How does driver distraction influence crash likelihood? 4 1* How do aggressive driving behaviors influence crash likelihood? 5 1* How does driver fatigue influence the likelihood and type of crashes? 6 1* How do advanced driver support systems influence crash likelihood? 7 1* What is the influence of driver impairment on crashes and driver errors? 8 1* How does the turn-lane configuration influence behavior and crash risks? 9a 1* What variables or pre-event factors are the most effective crash surrogate measures? 9b 1* What explanatory factors are associated with crashes or crash surrogates, and what analytical models can be developed to predict crashes or crash surrogates? 10 No broad insights 2* How do roadway features influence crash likelihood? 11 2* How do signage, lighting conditions, and other traffic control–related countermeasures influence crash likelihood and driver performance? 12 No straightforward intervention 3* How do static driver characteristics influence crash likelihood? 13 Naturalistic data not the best 4* How do static driver characteristics, as observed through driver performance measures, influence crash likelihood? 14 4* What are the relationships between driver behavior, performance, crash types, crash likelihood, and population-attributable risk for each factor contributing to crashes? 15 4* How do individual differences (e.g., age, gender, or speed selection) influence lane-keeping performance? 16 4* How do traffic and traffic volume influence intersection negotiation, lane-keeping performance, and crashes? 17 Crash or simulation sufficient 6 Do vehicle characteristics influence crash likelihoods or driver behaviors? 18 6 What are the interrelationships of environmental, road, and driver factors with nondriving- related activities (e.g., technology, OEM, or nomadic devices)? 19 6 How does seatbelt use vary with different levels of enforcement and in different jurisdictions? 20 Nonsafety related Unranked General or very high-level questions. 21 Unranked How else can naturalistic driving data be used? 22 Unranked How does the speed that drivers select influence other driver behaviors or actions? 23 Unranked How do roadway features influence driver performance and behavior? 24 Unranked How do the number and type of passengers influence the driver’s behavior? 25 Unranked How does driver fatigue affect driver performance? 26 Unranked How does inattention affect driver behavior and performance? 27 Unranked What nonsafety-related but useful information can be obtained from these data? *Indicates relevance to goals of SHRP 2.

53 these questions reflect one of several perspectives that might be used to aggregate the questions. Lexical analysis techniques provide a computational approach to understanding the content of the research ques- tions provided by the S01 contractors and Virginia Tech Transportation Institute. Word frequency and word co- occurrence can identify key themes, concepts, and their con- nections. Because lexical analysis identifies central themes of the research questions independently of the subjective method used to identify the global questions, it offers a means of verify- ing the relevance of the GRQs. This analysis used Leximancer, a software tool that employs a two-stage approach to lexical analysis (Smith and Humphreys 2006). The first stage is a semantic analysis and the second is a relational analysis. The semantic analysis uses a Bayesian co-occurrence metric to consider how frequently words occur together and how frequently they occur apart. The relational analysis uses the results of the semantic analy- sis and a naïve Bayesian algorithm to code segments of text. The result of this two-stage analysis is a statistical descrip- tion of the co-occurrence of concepts and the related text. This information provides the basis for identifying themes, which are labeled by a highly connected word that dominates the region. The number of themes is a parameter that can be adjusted in the analysis. The input to the analysis was a file containing only the research questions without any labeling or grouping. Co- occurrence and proximity of concepts reflect how frequently particular words occur within each research question. Figures A.5 and A.6 are the outcomes of the analysis for the S01 and S05 questions, respectively. The S05 contract questions are shown separately with six and eight themes as a comparison. The general themes are represented as circles (e.g., animals, lane, behavior) and concepts (e.g., vehicle, gap, driving) as points within the circles. More frequently occurring themes and concepts are shown as darker words and circles. The proxim- ity of concepts and themes represents how similar they are, as determined by how often they co-occur. For example, driver and behavior are similar themes by virtue of their spatial proximity, whereas influence, lane, and behavior are not. The five-theme grouping has four major themes (driver, behavior, influence, and lane) and one minor theme (relative). Concepts are identified from sets of words that tend to occur in similar contexts. For example, in the five-theme grouping on the left side of Figure A.5, the concept of lane reflects association of words such as vehicle, pedestrian, and travel. Map B, featuring eight themes for the same S05 questions, shows a generally similar pattern. In both Map A and B, the themes driver, influence, and lane dominate. The prominence of lane is somewhat surprising and reflects a substantial focus on the key activity of lane keeping in the development of these questions. These spaces show the centrality of the driver in the questions, with road, traffic influences, and behavior playing important roles. The themes in the high-priority research ques- tions are generally consistent with the themes in Figure A.5, particularly the themes of lane keeping, support systems, and driver behavior. Missing from the global questions is the gen- eral theme of traffic and drivers’ management of their position relative to other vehicles (e.g., gap acceptance and following distance). driver relative lane behavior influence influencedriver relative devices traffic lane crashdifferences differences countermeasures driving role travel pedestrians behavior driver risk relative system driving intersections road influence speed factors gap traffic lane vehicle role travel pedestrians crash vehicle lane crash red driversystem devices relative light gap factors change road speed influence intersections traffic behavior risk Map A Map B Figure A.5. Relationships among S05 contractor questions (five and eight themes).

54 Figure A.6 shows maps of the five and eight themes asso- ciated with the S01 questions. Map A is a set of five themes dominated by crash, terms, relationship, and vehicle. Map B shows eight themes dominated by similar themes of crash, evidence, vehicle, and road. In both maps, the space defined by these themes roughly separates into an area on the left of associated data and extracting meaning from data, as indicated by the concepts of elucidative, data, and relation- ship, and on the right of factors affecting safety, including vehicle, lane, and road. Many of the themes in Figure A.6 are not directly reflected in the global questions, as these themes reflect the basis for answering the questions. These themes may play a greater role in Phase 2 of this project, when the emphasis will shift to identifying promising analytic tech- niques that can provide evidence regarding the conditions and relationships that contribute to crashes. The themes, concepts, and their relationships are strikingly different for the two sets of questions. With the S05 questions, the driver and driver behavior hold a central position, whereas themes associated with understanding relationships play a dominant role in the S01 questions. The S01 questions also focus on road and roadway infrastructure to a much greater extent than do the S05 questions. The two sources of questions provide complementary perspectives on issues to be addressed in a naturalistic study. Any study should address questions from both sources. The two perspectives also demonstrate how perspectives from different constituencies can be radically dif- ferent. In this case, the difference may reflect the requirements of the particular contracts the S01 and S05 contractors were aiming to satisfy. A broader survey of stakeholders might produce important questions and perspectives not represented in the current set of questions. The lexical analysis only provides one alternate perspective and does not support any firm conclusions. In some cases, the themes reflect idiosyncratic word choices of the authors, such as the repeated use of a particular word (e.g., elucida- tive). In addition, some words have quite different mean- ings depending on their context, such as terms as used in the phrase in terms of. The co-occurrence algorithms that under- lie the lexical analysis do not always produce a meaningful interpretation of such phrases when they are used to label themes. These limits demonstrate the need for subject matter experts to create GRQs. At the same time, the lexical analysis suggests several themes, such as traffic and the broad analytic issue of identifying relationships from data, which should be considered in the next phase of this project. Specifically, the general groupings of behavior, crashes, and relationships/ influences suggest a central theme of identifying surrogate measures that effectively relate contributors to driving safety to safety-relevant changes in driver behavior. Conclusions The GRQs generated in this report were based on the specific questions posed by the S01 and S05 contractors, questions that arose from very different focuses. As the tables in Appen- dices B and C show, the S01 questions were more focused on transportation and infrastructure questions. The S01 ques- tions did include some driver-related questions, but not to the specificity provided by the S05 contractors. The systems-based crash terms relationship vehicle roadside evidence crash types vehicle conditions road lane occur conditions data crash road vehicle lane driver relationship evidence elucidative terms types plausible types behavior evidence elucidative terms relationship environmental driver road statistical roadside lanevehicle data conditions occur horizontal environmental Map A Map B Figure A.6. Relationships among S01 contractor questions (five and eight themes).

55 approach used to classify the research questions by driver, vehicle, roadway, and environment factors as described in the introduction to this report (Figure A.1) provides a means to illustrate these gaps. For example, no GRQs consider the feed- back loops in Figure A.1. That is, no question addressed how driving behavior changes over time and how drivers adapt to changing conditions. Other research questions that may also be meaningful based on this analysis are the following: • How do advanced driver support systems mitigate crash like- lihood when a driver is impaired? (Estimated priority of 1.) • How does inattention interact with fatigue to influence crash likelihood? (Estimated priority of 1.) • How do surrounding vehicles adapt to the impairment of the driver? (Estimated priority of 1.) • How do drivers adapt their behaviors (and engagement) with driver support systems over time? (Estimated prior- ity of 3.) While it is unclear how many participants in the natural- istic driving study will have driver support systems in their vehicles, these drivers can still provide important insights. Using the systems-based approach to aggregate the spe- cific research questions posed by the S01 and S05 contractors offers insights into the commonalities and differences in the types of research questions that potentially can be addressed in a naturalistic study. The initial prioritization provides a means to rank these questions within the scope of this study. Questions that were considered too broad for ranking or not directly related to the goals of this effort are considered out of scope of the project. However, they may be useful in guiding future research that uses the SHRP 2 naturalistic driving data. References Donmez, B., L. N. Boyle, and J. D. Lee. 2006. The Impact of Driver Distraction Mitigation Strategies on Driving Performance. Human Factors: The Journal of the Human Factors and Ergonomics Society, Vol. 48, No. 4, pp. 785–804. Evans, L. 2004. Traffic Safety. Science Serving Society, Bloomfield Hills, Mich. Griffith, M. S. 1999. Safety Evaluation of Rolled-In Continuous Shoul- der Rumble Strips Installed on Freeways. In Transportation Research Record: Journal of the Transportation Research Board, No. 1665, TRB, National Research Council, Washington, D.C., pp. 28–34. Lee, J. D. 2006. Driving Safety. In Reviews of Human Factors and Ergonom- ics, Volume 1 (R. S. Nickerson, ed.), Human Factors and Ergonomics Society, Santa Monica, Calif., pp. 172–218. Lee, J. D., D. V. McGehee, T. L. Brown, and M. L. Reyes. 2002. Colli- sion Warning Timing, Driver Distraction, and Driver Response to Imminent Rear-End Collisions in a High-Fidelity Driving Simula- tor. Human Factors: The Journal of the Human Factors and Ergonom- ics Society, Vol. 44, No. 2, pp. 314–334. Lunenfeld, H., and G. J. Alexander. 1990. A User’s Guide to Positive Guidance, 3rd ed. Report FHWA-SA-90-017. Office of Safety and Traffic Operations Research and Development, FHWA, McLean, Va. Neyens, D. M., and L. N. Boyle. 2006. The Effect of Distractions on the Crash Types of Teenage Drivers. Accident Analysis and Prevention, Vol. 39, No. 1, pp. 206–212. Shen, J., and A. Gan. 2003. Development of Crash Reduction Factors: Meth- ods, Problems, and Research Needs. Transportation Research Record: Journal of the Transportation Research Board, No. 1840, Transporta- tion Research Board of the National Academies, Washington, D.C., pp. 50–56. http://trb.metapress.com/content/8q563j8373753294/ fulltext.pdf. Smith, A. E., and M. S. Humphreys. 2006. Evaluation of Unsuper- vised Semantic Mapping of Natural Language with Leximancer Concept Mapping. Behavior Research Methods, Vol. 38, No. 2, pp. 262–279. Smith, E. B., and J. N. Ivan. 2005. Evaluation of Safety Benefits and Potential Crash Migration Due to Shoulder Rumble Strip Installa- tion on Connecticut Freeways. In Transportation Research Record: Journal of the Transportation Research Board, No. 1908, Transporta- tion Research Board of the National Academies, Washington, D.C., pp. 104–113. http://trb.metapress.com/content/p4xu128h05516081/ fulltext.pdf. Stanton, N. A., and M. S. Young. 2005. Driver Behaviour with Adaptive Cruise Control. Ergonomics, Vol. 48, No. 10, pp. 1294–1313. Young, M., and N. Stanton. 2004. Taking the Load Off: Investiga- tions of How Adaptive Cruise Control Affects Mental Workload. Ergonomics, Vol. 47, No. 9, pp. 1014–1035. http://bura.brunel .ac.uk/bitstream/2438/659/1/2004%20Young%20%20Stanton%20 Ergonomics%20(preprint).pdf.