Joint Analysis of Crash Frequency by Severity Based on a Random Parameters Approach

Zhaoming Chen, Wenyuan Xu, Youyang QuSustainability 2023 · DOI

TL;DR

This paper tackles the challenge of "unobserved heterogeneity" in road safety models, which can lead to inaccurate predictions of crash frequency and severity. By developing and comparing fixed, basic random, and improved random parameters models, the authors demonstrate that accounting for unobserved factors significantly enhances model accuracy and provides more reliable insights into crash causes.

What Problem Does This Paper Address?

Estimating road safety models is tricky because many factors influence crash frequency and severity, and not all of them can be observed or measured. This "unobserved heterogeneity" can lead to biased results, meaning our understanding of what causes crashes and how severe they are might be wrong. Traditionally, crash frequency and severity are analyzed separately, which misses crucial correlations. While joint models exist, they often struggle to account for this unobserved heterogeneity. This paper aims to address this by applying and extending the random parameters approach to jointly model crash frequency and severity, specifically focusing on how to better capture unobserved heterogeneity.

Methods

The core of this research lies in comparing three modeling approaches:

  • Fixed Parameters Model: A standard approach where all model parameters are assumed to be constant across all observations.
  • Basic Random Parameters Model: This model introduces "random parameters" that vary across observations, allowing for unobserved heterogeneity. The parameters are assumed to follow a normal distribution.
  • Improved Random Parameters Model: This is the novel contribution. It builds upon the basic model by allowing the means of the random parameters to vary (heterogeneous means) and by incorporating the correlation between these random parameters. This is achieved by expressing the random parameters as linear functions of other variables and using Cholesky decomposition to model their covariance.

Additionally, the paper proposes a method for calculating "marginal effects" (the impact of a one-unit change in an independent variable on the outcome) for these complex models, which is crucial for interpreting the results and informing safety countermeasures.

Datasets

The study utilizes crash data from five freeways in Heilongjiang and Liaoning provinces, China, covering January 2015 to December 2019. This dataset includes 12,097 crashes categorized into minor, general, and serious. Road design information, pavement condition data, and traffic volume data were also collected and integrated to create 3974 homogeneous road segments for analysis.

Key Insights and Ideas

  • Unobserved Heterogeneity is a Big Deal: Ignoring unobserved factors in safety models leads to significantly biased predictions, with marginal effects sometimes being over 100% wrong.
  • Random Parameters are Better: Even a basic random parameters model significantly outperforms traditional fixed parameters models in terms of statistical fit and predictive accuracy.
  • Improved Random Parameters are Even Better: Accounting for both heterogeneous means and correlations among random parameters further boosts model performance, leading to more accurate predictions and a better understanding of crash mechanisms.
  • Marginal Effects Need Careful Calculation: The way marginal effects are calculated is critical, especially for complex models like random parameters. The proposed method is vital for accurate safety analysis.
  • Joint Modeling is Key: Analyzing crash frequency and severity together, rather than separately, is essential for a comprehensive understanding and for developing effective safety strategies.

Key Results

  • Statistical Fit: The improved random parameters model showed the best statistical fit, as indicated by lower AIC and BIC values, and significantly better Chi-square statistics compared to both the fixed and basic random parameters models.
  • Predictive Performance: The improved random parameters model significantly reduced prediction errors. For minor, general, and serious crashes, it achieved reductions of 40–100% in mean prediction error, 3.7–8.3% in mean absolute error, and 7.6–8.9% in root mean square error compared to the basic random parameters model.
  • Bias Reduction: The random parameters approach, especially the improved version, substantially reduced the bias in marginal effects compared to the fixed parameters model. However, some bias still remained when heterogeneity in means and correlations were not fully accounted for.
  • Safety Factor Insights: The study identified several factors influencing crash frequency and severity, such as traffic volume, segment length, tunnel segments, speed limits, number of lanes, curvature, and road pavement condition. The random parameters approach provided more nuanced insights into how these factors interact.

Limitations and Future Work

The authors acknowledge several limitations:

  • Temporal Instability: The use of five-year aggregated data might mask temporal variations in safety effects. Future research could explore time-specific safety countermeasures.
  • Normal Distribution Assumption: The study assumes random parameters follow a normal distribution. Future work could investigate other distributional forms or explore heterogeneous variances of random parameters.

The paper suggests that future research should focus on investigating time-specific safety factors and exploring alternative distributional assumptions for random parameters.

Summary updated

Abstract

Unobserved heterogeneity is a major challenge in estimating reliable road safety models. The random parameters approach has been proven to be an effective way to account for unobserved heterogeneity but has rarely been used for crash frequency by severity level. In this paper, a fixed parameters model, a basic random parameters model, and an improved random parameters model, allowing for heterogeneity in the means and correlation of random parameters, are estimated and comparatively evaluated. To quantitatively analyze the impact of explanatory variables on the crash frequency of various severity levels, the calculating method of marginal effects for estimated models is proposed. The results indicate that (1) the basic random parameters model statistically outperforms the fixed parameters model, and the statistical fit can be further improved by introducing heterogeneous means and correlation of random parameters; (2) for the predictive performance, the basic random parameters model is more accurate than the fixed parameters model, and the improved random parameters model can further reduce the mean error, mean absolute error, and root mean square error by 40–100%, 3.7–8.3%, and 7.6–8.9%, respectively; (3) ignoring the unobserved heterogeneity or neglecting the heterogeneity in the means and correlation of random parameters may result in biased safety inferences, and the maximum bias of marginal effects can easily exceed 100 percent; and (4) the safety effects of explanatory variables are thoroughly discussed and the potential safety countermeasures are provided. The random parameters approach and the method for calculating marginal effects proposed in this study are expected to provide a new methodological alternative for the joint analysis of crash frequency by severity and should be helpful in uncovering the mechanism of crash occurrence and the resulting injury severity accurately.
BibTeX
@inproceedings{chen2023joint,
  author = {Zhaoming Chen and Wenyuan Xu and Youyang Qu},
  title = {{Joint Analysis of Crash Frequency by Severity Based on a Random Parameters Approach}},
  booktitle = {Sustainability},
  year = {2023},
  doi = {10.3390/su152115484},
}
Figure