Analysis of left-turn crash using Poisson Regression model at unsignalized intersections

Liyan Qin, Yan Lu, Shuai Zhang2018 · DOI

Abstract

This study explores the safety performance of short left-turn lanes at unsignalized intersections, where standard lengths are often impractical due to space constraints or high traffic volumes. Analyzing six years of crash data (2006–2011) from Houston, Texas, the research examines 39 substandard lanes and 13 lanes meeting standard requirements. A Poisson Regression model was developed to correlate traffic and geometric attributes with crash frequency (rear-end, sideswipe, and object-motor vehicle accidents). The study also calculates Crash Modification Factors (CMFs) to predict safety outcomes for lane length adjustments. Results indicate that short left-turn lanes significantly increase crash risks, leading to recommendations for improving safety at unsignalized intersections.

Introduction

Urban geometric design, guided by standards like AASHTO’s "Green Book," dictates specific lengths for left-turn lanes to ensure safe deceleration and queuing. However, in dense urban environments, physical constraints often prevent engineers from implementing these ideal lengths. While left-turn lanes are known to enhance safety generally, the specific risk associated with "substandard" or short lanes has remained under-quantified. This research addresses this gap by modeling the relationship between lane length, traffic volume, and crash frequency to help engineers make informed trade-off decisions.

Methodology

To determine the most accurate statistical relationship, the researchers compared four modeling approaches for count data:

  1. Poisson Regression: The baseline for modeling discrete event counts.
  2. Negative Binomial (NB) Regression: An alternative used if data exhibits overdispersion (variance exceeding the mean).
  3. Zero-Inflated Models (ZIP and ZINB): Specialized models for datasets with an excess of zero-crash sites.

Following rigorous statistical validation, including overdispersion tests and the Vuong statistic, the Poisson Regression model was selected as the most appropriate fit for the Houston dataset. Additionally, the study utilized Crash Modification Factors (CMFs)—multiplicative values that estimate the change in expected crash frequency resulting from specific geometric modifications.

Data Collection

The study utilized data from Houston, Texas, spanning 2006 to 2011. Data sources included:

  • Texas Crash Record Information System (CRIS): For historical accident records.
  • City of Houston and TxDOT Records: For Average Daily Traffic (ADT) and geometric specifications.

The final sample consisted of 52 median left-turn lanes at four-leg unsignalized intersections. Of these, 75% (39 lanes) were classified as "short" relative to Green Book standards. The sample lanes featured lengths between 140 and 450 feet and were located on roads with speed limits ranging from 30 to 40 mph.

Key Findings and Model Results

The analysis identified three primary safety failure modes for short lanes:

  1. Speed Differentials: High speed variance between turning vehicles and through-traffic.
  2. Deceleration Constraints: Inadequate space for safe braking.
  3. Lane Overflow: Queued vehicles spilling into through-lanes, creating hazards.

The calibrated Poisson model is expressed as: $$E(y_i|X) = e^{(-2.9155 + 0.2208V_i - 4.1993L_i)}$$ where $V_i$ represents traffic volume and $L_i$ represents relative lane length.

Statistical Significance

Both traffic volume ($p=0.0048$) and relative lane length ($p=0.0015$) were found to be highly significant predictors. The model demonstrates that as lane length decreases relative to the standard, crash potential increases exponentially. The derived CMFs provide a practical tool for engineers to quantify the safety benefits of lengthening existing short lanes.

Conclusion and Future Directions

The study confirms that short left-turn lanes at unsignalized intersections pose a measurable safety risk. While engineers must often balance safety with economic and spatial constraints, this research provides a data-driven basis for those decisions.

Limitations of the study include a relatively small sample size (52 lanes) and a narrow speed limit range (30–40 mph). Future research should expand the dataset to include a broader variety of urban contexts and higher-speed corridors to enhance the model's generalizability. Additionally, a comprehensive cost-benefit analysis integrating mobility and safety would further assist urban planners in optimizing intersection designs.

Summary updated

Abstract

The purpose of this study is to explore the safety performance of short left-turn lanes at unsignalized intersections, where it is often impractical to provide the standard lengths either when the available length between two adjacent opening is inadequate or when heavy left-turn volumes leads to demanding lane length. Left-turn crash data of six years from Houston, Texas of the years 2006-2011 were collected. There were thirty-nine lanes shorter than the standard requirements and thirteen lanes meeting the requirements. A Poisson Regression model was developed to relate traffic and geometric attributes to the total count of rear-end, sideswipe, and object-motor vehicle accidents at a left-turn lane. Furthermore, Crash Modification Factors (CMFs) were calculated for future applications in projecting the crash frequency given a specific change of the lane length. The results shows that short left-turn lanes have negative safety impacts. The study finally puts forward the opinion to improve the present conditions at the left-turn unsignalized intersection.
BibTeX
@misc{qin2018analysis,
  author = {Liyan Qin and Yan Lu and Shuai Zhang},
  title = {{Analysis of left-turn crash using Poisson Regression model at unsignalized intersections}},
  year = {2018},
  doi = {10.2991/iceesd-18.2018.94},
}
Figure