Evaluating and addressing the effects of regression to the mean phenomenon in estimating collision frequencies on urban high collision concentration locations

Jinwoo Lee, Koohong Chung, Seungmo Kang

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

Two different methods for addressing the regression to the mean phenomenon (RTM) were evaluated using empirical data: 1 The Empirical Bayes (EB) method, which combines observed collision data and Safety Performance Functions (SPF) to estimate expected collision frequency of a site.2 Continuous Risk Profile (CRP), which estimates true collision profile constructed after filtering out the noise.Data from 110 miles of freeway located in California were used to evaluate the performance of the EB and CRP methods in addressing RTM. CRP outperformed the EB method in estimating collision frequencies in selected high collision concentration locations (HCCLs). Findings indicate that the performance of the EB method can be markedly affected when SPF is biased, while the performance of CRP remains much less affected. The CRP method was more effective in addressing RTM.

Original languageEnglish
Pages (from-to)49-56
Number of pages8
JournalAccident Analysis and Prevention
Volume97
DOIs
Publication statusPublished - 2016 Dec 1

    Fingerprint

Keywords

  • Continuous risk profile
  • Empirical bayes method
  • Random noise
  • Safety performance function
  • The regression to the mean

ASJC Scopus subject areas

  • Human Factors and Ergonomics
  • Safety, Risk, Reliability and Quality
  • Public Health, Environmental and Occupational Health
  • Law

Cite this