<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=648880075207035&amp;ev=PageView&amp;noscript=1">

Vehicle Size, Weight, and Injury Risk

High-Dimensional Modeling and Causal Inference with Bayesian Networks

DOI: 10.13140/2.1.1950.1768

Vehicle Size, Weight, and Injury Risk

This paper’s intent is to illustrate how Bayesian networks and BayesiaLab can help overcome certain limitations of traditional statistical methods in high-dimensional problem domains. We consider the vehicle safety discussion in the recent Final Rule, issued by the Environmental Protection Agency (EPA) and the National Highway Traffic Safety Administration (NHTSA) on future CAFE standards, as an ideal topic for our demonstration purposes.   

Although this paper is meant to focus on technique as opposed to the subject matter itself, our findings will inevitably generate new insights. However, it is not our intention to challenge the judgement of the EPA/NHTSA Final Rule. Rather, we plan to take an independent look at the overall problem domain while considering the rationale presented in the Final Rule. Instead of merely replicating the existing analyses with different tools, we will draw upon a broader set of variables and use alternative methods to create a complementary view of some aspects of this problem domain. Extending beyond the traditional parametric methods employed in the EPA/NHTSA studies, we want to show how Bayesian networks can provide a powerful framework for forecasting the impact of regulatory intervention. Ultimately, we wish to use Bayesian networks for reasoning about consequences of actions not yet taken. 

Admittedly, we will restate a number of the original research questions in order to better suit our expository requirements. Even though a macro view of this domain was required by EPA/NHTSA, i.e. societal costs and benefits, we believe that we can employ Bayesian networks particularly well for understanding high-dimensional dynamics at the micro level. Consequently, we examine this domain at a “higher resolution” by using additional accident attributes and finer measurement scales. Primarily for explanatory clarity, we also restrict our study to more narrowly defined contexts, i.e. vehicle-to-vehicle collisions, as opposed to all motor vehicle accidents. We also need to emphasize that all of our considerations exclusively relate to vehicle safety. We do not address any of the environmental justifications given in the EPA/NHTSA Final Rule. In that sense, we only focus on a small portion of the overall problem domain.

This paper is meant to portray a prototypical research workflow, presenting an alternating sequence of questions and answers. As part of this discourse, we gradually introduce a number of Bayesian network-specific concepts, but, each time, we only cross the proverbial bridge when we come to it. In the beginning chapters, we strive to provide a large amount of detail, including step-by-step instructions with many screenshots for using BayesiaLab. As we progress through this study, in later chapters, we try omitting some technicalities in favor of presenting the bigger picture of Bayesian networks as powerful reasoning framework.

Please Register to Download the Complete Tutorial (84 pages, 13.9MB)

2017 BayesiaLab Conference