Low Cost Safety Improvements
Pooled Funds Study

Safety of Lane/Shoulder Width Combinations on Two-Lane Rural Roads

Dr. Frank Gross, Vanasse Hangen Brustlin (VHB), Inc
Dr. Paul P. Jovanis, Penn State University

Overview

• Introduction
• Objective
• Study Design
• Methodology
• Data Collection
• Preliminary Results
• Conclusions
• Future Research

Background on Strategy

• Shoulder Paving/Widening
  • Proven strategy
• Pavement Width
  • What lane/shoulder width produces lowest crash odds?
  • Identified at Technical Advisory Committee June 2006
• Target crashes
  • Head-on
  • Run-off-road
  • Sideswipe
• Potential Difficulties
  • Confounding Variables
• Key to Success
  • Flexible modeling approach

Literature Review

• Crash Modification Factors in Highway Safety Manual
  • Key studies: Zegeer et al. (1981); Zegeer et al. (1988); Griffin and Mak (1987)

	Recommended CMF for Shoulder Width (Harwood et al., 2000)

• Few Studies Address Allocation of Total Width
  • "Road diets" change total number of lanes
    • Burden and Lagerwey (2001); Welch (1999)
  • Reallocation of width on urban freeways
    • Add lane by reducing lane and shoulder width
    • McCasland (1978); Urbanik and Bonilla (1987)
• Evaluate Re-allocation Without Other Changes

Objective

• Estimate Safety Effectiveness
  • For a given pavement width, what allocation of lane/shoulder width produces the lowest crash odds?
• Secondary Questions of Interest
  • Do effects vary by:
    • Traffic volume?
    • Speed limit?
  • For a given lane width, do effects vary as shoulder width increases?
  • Is the treatment economically feasible?

Methodology

• Case-Control Methodology
  • Cases: crash-involved segments for a given year
  • Controls: non-crash-involved segments for a given year
• Matching Variables
  • ADT and Segment Length
• Additional Covariates
  • Speed, District, Unpaved Shoulder, Curvature, and Grade

SW means shoulder width and LW means lane width

• Case-Control Methodology
  • Allows answer to primary and secondary questions
  • Regression-to-the-mean not an issue
  • Accounts for confounding variables
    • Matched design
    • Model covariates

Study Design

• Required Sample Size
  • Minimum: 15,094 segment-years
    • Detect 10 percent reduction in total crashes with 90 percent confidence
  • Desirable: 57,576 segment-years
    • Detect 5 percent reduction in total crashes with 90 percent confidence
• Assumption
  • 50 percent discordant pairs
• How Does Assumption Hold?
  • PA discordant pairs: 70 percent (LW) and 80 percent (SW)
  • WA discordant pairs: 66 percent (LW) and 84 percent (SW)

Data Collection-1

Data Collection-2

• Crash Data
  • 5 years of PA data
  • 6 years of WA data
• Roadway Data
  • (PA and WA)
    • Number of Lanes
    • Area Type
    • AADT
    • Segment Length
    • Speed Limit
    • Surface Width
    • Paved Shoulder Width
  • (WA only)
    • Horizontal Curvature
    • Vertical Curvature
  • (PA only)
    • Unpaved Shoulder Width
    • District

Data Collection-3

Data Collection -4

Evaluation Results (PA Total Crashes)

Evaluation Results (PA Target Crashes)

Evaluation Results (WA Total Crashes)

Evaluation Results (WA Target Crashes)

Example Odds Ratio

Conclusions

• Within Pavement Width
  • PA: Not many significant changes
    • Particularly for total crashes
  • WA: General decrease in crashes for narrow lane and wide shoulder
    • Target crashes in particular
    • Be aware of small samples
• Within Lane Width
  • General decrease in crashes as shoulder increases
    • Supports model results (consistent with prior studies)
• Preliminary Results!
  • Need to explore outliers
  • Answer secondary questions

Future Research

• Field-verify sites in PA
  • Preliminary data verification using PA video logs
• Evaluate Anomalies
  • Met with PennDOT to discuss results
  • Review PennDOT and WSDOT design guides
• Secondary Questions
  • Do effects vary by:
    • Traffic volumes?
    • Speed limit?
• Economic analysis

< Presentation 9 | Presentation 11 >