Uncertain Data and Confident Decisions: Winnipeg's Safety Performance Functions and Network Screening Project

This paper presents modelling techniques for addressing three data reliability issues encountered in the City of Winnipeg safety performance functions (SPF) and network screening project: (A) uncertain traffic volume data, (B) non-uniform collision under-reporting linked to segment length, and, (C) missing traffic volume data for minor roads of an intersection. The first issue relates to uncertain traffic volume data. The City of Winnipeg uses short-term count stations and has a form of traffic data known as the weekday average daily traffic (WADT). The issue of data uncertainty is addressed by estimating the amount of error in the volume data that goes into the SPF and accounting for this error in the SPF development using a Monte Carlo-based modelling approach. The simulation approach maps traffic volume uncertainty to SPF parameter uncertainty. The second issue relates to non-uniform under-reporting linked to segment length. The non-uniform under-reporting of segment collisions resulted in a global and localized model bias. We introduce a new technique to detect, quantify, and correct this bias by using residuals analysis to stratify the population. The third issue relates to modelling intersections with missing minor street flow volumes. We apply an approach that uses the functional class of the intersections as proxies for the missing flow volumes. For each issue, we demonstrate quantitatively that good modelling results can be obtained despite input data limitations. Key conclusions are: (A) for traffic volume measurement errors of up to 30%, Monte Carlo analysis shows that the ability to create reliable SPFs is not affected; (B) residuals analysis to stratify a population according to non-uniform under-reporting can essentially eliminate global and local model bias, and (C) using a readily available proxy for a missing predictor variable can improve predictive ability by almost 50% (measured by mean absolute deviation) when compared to omitting that predictor variable entirely.

• Record URL:
  https://www.cite7.org/wpdm-package/2016-kelowna-compendium/
• Corporate Authors:

  INSTITUTE OF TRANSPORTATION ENGINEERS. DISTRICT 7 CANADA

  ,    
• Authors:
  • Ghanney, B
  • Milligan, C
  • PETERNIAK, R
  • Hearson, M
• Conference:
  • Kelowna 2016 - CITE Annual Meeting and Conference - Technical Compendium
• Publication Date: 2016

Language

• English

Media Info

• Media Type: Web
• Pagination: 1 PDF file, 901 KB, 16p.
• Monograph Title: Kelowna 2016 - CITE Annual Meeting and Conference - Technical Compendium

Subject/Index Terms

• TRT Terms: Conferences; Crash rates; Errors; Information processing; Mathematical analysis; Mathematical models; Safety; Traffic counts; Urban areas
• Geographic Terms: Canada
• ITRD Terms: 1612: Accident rate; 6471: Analysis (math); 8018: Canada; 8525: Conference; 8655: Data processing; 6440: Error; 6473: Mathematical model; 1665: Safety; 689: Traffic count; 313: Urban area
• Subject Areas: Safety and Human Factors; I71: Traffic Theory; I72: Traffic and Transport Planning;

Filing Info

• Accession Number: 01616270
• Record Type: Publication
• Source Agency: Transportation Association of Canada (TAC)
• Files: ITRD, TAC
• Created Date: Nov 15 2016 4:46PM