NORC is a subcontractor to Dunlap & Associates on this study sponsored by the National Highway Traffic Safety Administration (NHTSA) of the US Department of Transportation (DOT). We will examine data bases such as the Behavioral Risk Factor Surveillance Survey, National Hospital Discharge Survey, Emergency Department Data, Health Interview Survey and others to identify, analyze, and categorize the geographic, demographic (e.g. income, education, language), social-cultural and psychological characteristics of at-risk populations (i.e., those groups with seat belt use below the national average or unrestrained fatalities above the national average). This also includes a focus on characteristics of non-users and part-time user groups. Almost all of our research efforts involve coordinating with National, State, local, and private entities of all kinds to achieve our project objectives.
We will supply NHTSA with a list of potential partners and how we expect each may be able to assist in future program activities. We will produce a final report summarizing the characteristics of the newly identified hard-to-reach populations, including an analysis of the applicability of NHTSA's current strategies used with current partners to the newly identified at-risk groups. We will provide an honest evaluation of whether existing NHTSA techniques and partners are capable of addressing the problem subpopulations, or if new strategies and partners are needed to have a meaningful impact. Finally, we will review and identify any recommendations for moving forward with any current public health strategy which could be adopted with these populations. We expect to find a number of possible solutions that will be applicable to one or more of the identified subpopulations of interest. We will work with NHTSA to develop recommendations on how best to proceed to address this important problem.
Classification and Regression Tree (CART) analysis was conducted to categorize the group of people that tend to not wear a seat belt. CART analyses are machine-learning methods for constructing prediction models from data. An outcome measure, non-seat belt use, was modeled by partitioning the data into subgroups that were predictive of the outcome. Each subgroup was further split into smaller subgroups based on their predictive power of the outcome. The resulting partitioning was represented graphically as a decision tree. We set the cut-off point for our results at a 10% chance of not wearing a seat belt.
Many of the outcomes of interest were present in multiple data sets. Similar to the bivariate analyses, results from each data set identified consistent findings, strengthening the suggestion that associations between the outcomes and non-seat belt use exist. Individual CART models were successfully constructed on nine of the data sets. In summary, people who are less educated, drive a truck, and are less self-disciplined or care less about personal health are less likely to wear a seat belt. Characteristics like smoking, alcohol/drug usage, and obesity have significant associations with non-seat belt usage.