2011Transportation Research Board 90th Annual MeetingTransportation Research BoardRequires access

Hierarchical Bayesian Modeling for Before-and-After Studies

Andrew Nolan Olsen, Grant G. Schultz, Daniel J. Thurgood, C. Shane Reese

Open publisher page 13 citations

Abstract

Hierarchical Bayesian methods are a powerful extension of commonly used empirical Bayesian methods. Bayesian methodology combines prior information with information from the data to create a posterior distribution. This fundamental principle extends to the development of Poisson hierarchical regression, which allows count data to be modeled as a function of one or more co-variates. This framework is extremely useful in before-after traffic safety studies because both the type of crash and co-variates included in the analysis may be changed in an extremely flexible manner. No calibrated factors are required, which reduces time and effort necessary for analysis. Hierarchical Bayesian methods also require fewer data in order to achieve valid results. With these distinct advantages associated with hierarchical Bayesian modeling, Departments of Transportation (DOTs) and other entities could be benefited immensely by the use of Bayesian methodology in their research. The relatively recent onset of these methods as they apply to traffic studies, however, can make implementation of new Bayesian methods seem daunting. This paper seeks to help bridge the gap between cursory familiarity with Bayesian methods and thorough understanding of the foundations of Bayesian analysis and how to apply them to traffic studies. To facilitate the presentation of Bayesian methods as they apply to before-after studies, an application evaluating the efficacy of cable barriers on Utah highways is presented. The foundational principles of Bayesian statistics discussed may be applied to essentially any situation as the fundamentals of hierarchical modeling are understood and implemented by researchers.

About this research paper

What this paper is about

Hierarchical Bayesian methods are a powerful extension of commonly used empirical Bayesian methods. Bayesian methodology combines prior information with information from the data to create a posterior distribution. This fundamental principle extends to the development of Poisson hierarchical regression, which allows count data to be modeled as a function of one or more co-variates. This framework is extremely useful in before-after traffic safety studies because both the type of crash and co-variates included in the analysis may be changed in an extremely flexible manner. No calibrated factors are required, which reduces time and effort necessary for analysis. Hierarchical Bayesian methods also require fewer data in order to achieve valid results. With these distinct advantages associated with hierarchical Bayesian modeling, Departments of Transportation (DOTs) and other entities could be benefited immensely by the use of Bayesian methodology in their research. The relatively recent onset of these methods as they apply to traffic studies, however, can make implementation of new Bayesian methods seem daunting. This paper seeks to help bridge the gap between cursory familiarity with Bayesian methods and thorough understanding of the foundations of Bayesian analysis and how to apply them to traffic studies. To facilitate the presentation of Bayesian methods as they apply to before-after studies, an application evaluating the efficacy of cable barriers on Utah highways is presented. The foundational principles of Bayesian statistics discussed may be applied to essentially any situation as the fundamentals of hierarchical modeling are understood and implemented by researchers.

Why it matters

OpenAlex reports 13 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

Hierarchical Bayesian methods are a powerful extension of commonly used empirical Bayesian methods. Bayesian methodology combines prior information with information from the data to create a posterior distribution. This fundamental principle extends to the development of Poisson hierarchical regression, which allows count data to be modeled as a function of one or more co-variates. This framework is extremely useful in before-after traffic safety studies because both the type of crash and co-variates included in the analysis may be changed in an extremely flexible manner. No calibrated factors are required, which reduces time and effort necessary for analysis. Hierarchical Bayesian methods also require fewer data in order to achieve valid results. With these distinct advantages associated with hierarchical Bayesian modeling, Departments of Transportation (DOTs) and other entities could be benefited immensely by the use of Bayesian methodology in their research. The relatively recent onset of these methods as they apply to traffic studies, however, can make implementation of new Bayesian methods seem daunting. This paper seeks to help bridge the gap between cursory familiarity with Bayesian methods and thorough understanding of the foundations of Bayesian analysis and how to apply them to traffic studies. To facilitate the presentation of Bayesian methods as they apply to before-after studies, an application evaluating the efficacy of cable barriers on Utah highways is presented. The foundational principles of Bayesian statistics discussed may be applied to essentially any situation as the fundamentals of hierarchical modeling are understood and implemented by researchers.

Key concepts: Bayesian probability, Bayesian hierarchical modeling, Computer science, Variable-order Bayesian network, Bayesian average, Bayesian statistics, Bayesian experimental design, Bayesian linear regression

Related papers

Back to paper searchBrowse research topicsOriginal source
Hierarchical Bayesian Modeling for Before-and-After Studies — Research Paper | ScholarLens