Empirical Bayes Approach to the Estimation of "Unsafety:" the Multivariate Method.

Details

• Creators:
  Hauer, E.
• Corporate Creators:
  University of Toronto
• Contributors:
  Viner, John
• Corporate Contributors:
  United States. Federal Highway Administration. Office of Safety and Traffic Operations R&D ; Canada. Transport Canada
• Subject/TRT Terms:
  Bayes' Theorem Estimation Theory Safety Statistical Analysis
• Publication/ Report Number:
  FHWA-RD-90-006
• Resource Type:
  Tech Report
• Right Statement:
  Public Domain
• Geographical Coverage:
  United States
• Edition:
  Final Report
• Contracting Officer:
  Viner, John
• Corporate Publisher:
  United States. Federal Highway Administration. Office of Safety and Traffic Operations
• Abstract:
  There are two kinds of clues to the unsafety of an entity: its traits (such as traffic, geometry, age or gender) and its historical accident record. The essence of the Empirical Bayes (EB) approach to the estimation of unsafety is that it uses both clues. How this is accomplished is described. To estimate the unsafety of an entity using the EB approach, information is needed about the mean and the variance of the unsafety of similar entities which form its reference population. The Method of Sample Moments has been used for this purpose in the past. It suffers from three shortcomings. First, to yield usable estimates a very large reference population is required. Second, the choice of reference population is to some extent arbitrary. Third, entities in the chosen reference population usually cannot match the traits of the entity the unsafety of which is estimated. To alleviate these shortcomings the Multivariate Method for estimating the mean and variance of unsafety in reference populations is offered. Its logical foundations are described and its soundness is demonstrated. The use of the Multivariate Method makes the EB approach to unsafety estimation applicable to a wider range of circumstances, it makes the decision about what entities to include in the reference population less arbitrary and it yields better estimates of unsafety. The applications of the EB and Multivariate Methods to tasks of identifying deviant entities and estimating the effect of interventions on unsafety are discussed and illustrated by numerical examples.
• Format:
  PDF
• Funding:
  DTFH61-86-C-00075
• Collection(s):
  Federal Highway Administration
• Main Document Checksum:
  urn:sha-512:6bd9599ad12a5dd18fdfa6c1c8fba55d42dff7351b02ede744c5f5ffe6393fe5ce0d3a9e881eb384793bec19e81c6b2d81026cb1f877fa92c71da37f462726fd
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/88405/dot_88405_DS1.pdf
• File Type:
  [PDF - 2.39 MB ]