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.
2The second Strategic Highway Research Program (SHRP 2) naturalistic driving study (NDS) comprises the largest set of naturalistic driving data ever collected in a single study, with more than 3,100 participants from six geographically dis- persed sites across the United States. Volunteer participants were observed for more than three years in all, producing some 50 million miles of naturalistic driving data from kine- matic and other sensors, as well as from four-channel video (Dingus et al. 2014). The combined video is illustrated in Fig- ure 1.1. Sensors included accelerometers in three dimensions, a global positioning system, forward radar, and others. The SHRP 2 data set is expected to help a generation of researchers address a myriad of crucial transportation safety and other questions. However, to properly analyze these data and meaningfully interpret results, one must first understand the degree to which the data sample can be used to study the target population of interest: all drivers in the United States. Simply stated, the primary goal of the SHRP 2 NDS was to collect and archive the largest set of naturalistic driving data in history. More specifically, adequate samples of key driver, vehicle, and environmental characteristics were identified as integral to a thoughtful examination of driver behaviors, par- ticularly those associated with crash-related events. Additional layers of challenge were presented by the studyâs financial and operational constraints, all of which necessitated savvy strate- gic planning in the precollection phase. In addition, nimble management of the ongoing study was required as novel chal- lenges continually emerged as the study progressed. Specifically, study objectives had to be attained within the confines of a budget that allowed for recruitment and data col- lection at only six sites, each of which had to be selected care- fully to maximize sample quality and variety while minimizing operational obstacles. In addition, budgetary limitations, along with the naturalistic essence of the study, precluded the provi- sion of study vehicles to participants. The resulting reliance on volunteer drivers using their own vehicles introduced the pri- mary impediment to both fulfilling original sample design goals and maintaining optimal operational efficiency, particu- larly with regard to successful instrumentation of 2,000 eligible vehicles with data acquisition equipment. That is, attracting interested male and female participants across the driving age spectrum with vehicles well suited for the study (i.e., in terms of facilitating the collection of vehicle network data) required numerous midstudy modifications, both to study equipment and to the way in which sampling goals were undertaken. Representativeness A representative sample is one from which results can be gen- eralized to the larger population (Spence et al. 1976). One method of obtaining a representative sample is to select the sample at random from the population of interest (i.e., so that every member of the population has an equal chance of being chosen [Spence et al. 1976]). However, a truly random selec- tion is extremely difficult to achieve in an actual study involv- ing human subjects, and it was not possible in Project S31 for several reasons. First, participants had to live in one of the six preselected data collection sites, illustrated in Chapter 2, so all other drivers in the United Statesâthe vast majority of the nationâs driversâdid not have a chance to participate. Second, SHRP 2 NDS participants, similar to participants in virtually all ethically conducted studies, were required to grant their informed consent before data could be collected from them. As a result, SHRP 2 participants could not be selected at random. Instead, they were selected only from drivers living in the six data collection sites who were willing to grant their informed consent to participate. Thus, SHRP 2 data constitute a conve- nience sample. It has been argued that it is not possible to determine the representativeness of such a convenience sample (Anderson et al. 2003). But it is still instructive to compare the sample with the target population on as many relevant parameters as are available. Equipped with such comparison data, the analyst may be better able to judge whether the sample is adequately C h a p t e R 1 Introduction
3 population of interest, that is, all drivers, vehicles, and driving conditions in the United States. More specifically, this report addresses four questions: 1. How do the selected sites in aggregate compare to the United States as a whole? 2. How do the participants sampled compare to all drivers in the United States? 3. How do the vehicles sampled compare to the entire U.S. fleet? 4. How do police-reported crash rates observed in the study compare to crash rates in the United States as a whole? Data from the SHRP 2 NDS sample are also compared, when possible, with other geographical areas of interest, including the aggregated SHRP 2 recruitment counties and the SHRP 2 states. approach First, the report recapitulates at a high level the original sample design, including how the six data collection sites were selected, as well as how recruitment and screening progressed through- out the study. Second, information is presented about each of the data collection sites. Third, three categories of study dataâ participant characteristics, vehicle fleet composition, and crash ratesâare compared with local or national comparison data sets, each from as many perspectives as feasible. Taken together, this information should provide the reader with sufficient information to assess how SHRP 2 data can be used for analyz- ing a variety of research questions. characteristic of the population of interest for any specific pur- pose and whether any reweighting of the data is appropriate for specific research applications. Indeed, the degree to which a sample is characteristic of the population of interest may vary widely according to the particular research question being addressed and the exact subset of data being extracted from the larger data set for analysis (Ramsey and Hewitt 2005). Goal of this Report The goal of this report is to help researchers and other con- sumers of the data better understand from a variety of per- spectives the relation of the SHRP 2 NDS data to the main Figure 1.1. Quad-video image.
