Click for next page ( 9

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement

Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 8
8 as expressed in pretested algorithms (e.g., longitudinal associated with crashes or crash surrogates? And what ana- deceleration < -0.5 g and forward time-to-collision < 3.0 s). lytical models can be developed to predict crash or crash Vision-, attention-, and distraction-based questions. surrogates? Continuous video of the participant's face permits the scoring of glance direction and duration which can be ana- The list of research questions will continue to expand as lyzed to address these areas of relevance. This can be accom- the full implications of the data are analyzed and understood. plished via manual frame-by-frame video data reduction, Experience with other naturalistic studies, and the expecta- but machine-vision-based techniques are being developed tion for the SHRP 2 NDS, suggests that the data will be use- at VTTI to enable some of the glance-related scoring to be ful for addressing many questions for years after the data performed in an automated fashion. are collected. The specific research questions are discussed Speed- and speeding-based questions. A set of redundant in greater detail in Appendix A of the S02 Phase 1 report sensors provides continuous vehicle speed information. (Boyle et al. forthcoming). In addition, onboard Global Positioning System (GPS) sensor data can be combined with GIS-based information Sample Design Plan on speed limits to permit researchers to look at speeding- related behaviors (e.g., actual speed relative to the posted There are two broad aspects to the NDS sample design: speed limit) and safety-related events. (1) contractor-site selection and (2) participant-vehicle Crash-countermeasure-based questions. Crash counter- sample design. The first refers to the process of determin- measures can be evaluated by examining actual crash-related ing the number and location of the data-gathering sites and events and superimposing different countermeasure algo- the contractors who will manage each; the second refers to the rithms to evaluate their relative effectiveness. Similarly, methods and factors used to recruit and select participants. these same algorithms can be applied to non-event baseline These two aspects ultimately work toward the definition of a epochs to determine their relative propensity for creating unified plan that specifies the quantity and characteristics of false alarms. participants to be recruited at and across all site locations. Passing-maneuver-based questions. Passing maneuvers From the standpoint of robust experimental design, it is can be detected in the data stream (i.e., via the yaw sensor) desirable to ensure that there is good representation in the and differentiated from swerves on the basis of turn signal participant sample of the basic drivervehicle variables of rel- status and visual validation of the passing maneuver. evance and interest. In this study, the primary variables of Multifactor/Multivariate questions. All of the above types interest are age, gender, and vehicle type. of questions (and many more not listed) can be looked at It has been well documented that age is a strong indicator in any combination desired by a researcher. For instance, a of driving risk, with the youngest and oldest drivers having researcher may be interested in looking at a speeding-related an elevated crash rate per mile driven compared with other countermeasure to evaluate whether it may have been pos- age groups. Exposure details were carefully considered in an sible to prevent any speeding-related crash events observed. effort to understand the implications of driver age and gender in terms of crash rates per mile and per numbers of licensed The research questions were grouped by the SHRP 2 Safety drivers. Since the target data collection budget was already Project S02 contractor into the following high-level categories: established, the following are some of the major trade-offs that were considered: How does driver distraction influence crash likelihood? How does driver fatigue affect crash likelihood? Representative versus risk-prone sample. Do we strive How do aggressive driving behaviors influence crash like- for a more representative sample, which enhances general- lihood? izability of the data, or do we strive for a sample we believe What is the influence of driver impairment (e.g., alcohol) to be more crash-prone (i.e., one emphasizing the extremes on crash likelihood? of the age range) to observe more safety-related incidents How do driver interactions with roadway features influ- of interest? ence the likelihood of lane departure crashes? Overall costs of participant pay versus the ability to How do driver interactions with intersection features (con- enhance participant attraction and retention via mean- figuration and operations) influence crash likelihood? ingful compensation amounts. It is reasonable to expect How do advanced driver support systems influence crash that the greater the compensation, the greater the recruit- likelihood? ment uptake and retention rates would be. However, this What variables or pre-event factors are the most effective is constrained by the project funding available for this pur- crash surrogate measures? What explanatory factors are pose. It was ultimately determined that participants would

OCR for page 8
9 be compensated by the nominal amount of $25 per month vehicles (SUVs; including crossover vehicles), and minivans. of participation. It is expected that only vehicle model-years later than 2002 will Total number of sites versus number of DAS units man- be targeted for recruitment to help ensure that only mechan- aged per site. It was ultimately determined that the maxi- ically sound vehicles with access to vehicle network data, such mum number of DAS units that could be supported was as speed and accelerator position, will be included in the 1,950. Then, the issue becomes how many data collection sample. The goal is for the DAS to support installation in sites must be established to manage those DAS units. If about half of the light-vehicle population on the road. too many sites are established, then the costs would be On the basis of U.S. light-vehicle sales from model years prohibitive. If too few sites are established, then each may 20002007, a list of the top 50 most popular vehicles were iden- be asked to manage more DAS units than resources at a tified for possible inclusion in the study. These top sellers are particular site may allow. represented by the members of either the Alliance of Automo- Explicitly including stratification variables in the experi- bile Manufacturers (AAM) or the Association of International mental design versus the difficulty of filling each cell. The Automobile Manufacturers (AIAM); these organizations rep- more variables formally included in the experimental design, resent a cumulative total of 89% of all U.S. light-vehicle sales the greater the risk of ending up with experimental cells for the selected time period. Relationships are being pursued with too few participants--or perhaps none at all. How- with original equipment manufacturers (OEMs) belonging to ever, if factors of interest are not formally included, then AAM and/or AIAM with the objective of having CAN Param- the actual sample may not include sufficient numbers for eter IDs (PIDs) available to obtain and interpret additional analysis, and they are likely to be unevenly spread across data from the onboard vehicle network for high-volume the other factors of interest. models; the types of data of interest include speed, wiper Total number of primary participants versus months of usage, brake actuation, accelerator position, and turn signal data collection per participant. It is desirable to have as usage, as well as steering data. Within the advanced technol- many participants as possible in the study, yet recruitment ogy group of participants, additional data will be collected is expensive, and it is costly to move a DAS from one vehi- regarding the usage of in-vehicle communication systems cle to another. and advanced infotainment systems. Additionally, infor- Total cost of data collection versus cost per data-year. mation about driver monitoring, feedback, and collision Just as it is desirable to have as many participants in the warning systems will be available from some participating study as possible, so too is it desirable to have as many years manufacturers. of data or data-years as possible (i.e., where a data-year is Using the research questions to guide the process, the equivalent to the amount of data generated by a single par- research team conducted an analysis to estimate the statis- ticipant over the course of a single year). Of course, there tical power afforded by the experimental design to detect a is a substantial cost for each study year. However, the cost per statistically significant effect associated with the various age data-year tends to diminish as the study period is extended, and gender groupings. The analysis generally indicated that thus making the study simultaneously more expensive yet the study was sufficiently powered to address the age by gender more cost-efficient. questions. This analysis was conducted on a single variable, Site recruitment size versus a contractor's ability to man- speed variability, where previous data could be used to estimate age the square mileage. The larger the size of a site's recruit- that variable's mean and standard deviation (both required ment area, the greater the probability of finding a sufficient for power estimation). Since its standard deviation was fairly number of participants. However, if a site is too spread out high relative to its mean, this variable represents a relatively geographically, then not only will this cause the contractor conservative estimate of the statistical power that can be difficulty in managing all the DAS and participant issues, but expected with various analyses. Still, it must be recognized it will also make it more difficult for those participants at that the power associated with each analysis will be depen- the most distant points from the installation site. dent upon the particular means and standard deviations of the measures used and the actual magnitude of the differences Passenger cars and light trucks will be the focus of this study observed. because these types of vehicles accounted for almost 95% of all The experimental design shown in Table 2.1 is based on the vehicles on the road in 2007, as reported by Ward's Automo- preceding investigations, and revisions were made throughout tive Group (2010), and 94% of all motor vehicle crashes, as the course of the S05 study design project. Note that there are reported by the National Highway Traffic Safety Administra- more data years than participants because some participants tion (2007). Vehicles that will be instrumented for the SHRP 2 will participate for the full two years of data collection instead NDS include the following: passenger cars (sedans, coupes, of just one. Also note that the sample design emphasizes the hatchbacks, and station wagons), pickup trucks, sport utility extremes of the age spectrum more than the middle-aged