Gaining a better understanding of the factors that affect the likelihood of a vehicle crash has been an area of research focus for many decades. However, in the absence of detailed driving data that would help improve the identification of cause and effect relationships with individual vehicle crashes, most researchers have addressed this problem by framing it in terms of understanding the factors that affect the frequency of crashes - the number of crashes occurring in some geographical space (usually a roadway segment or intersection) over some specified time period. This paper provides a detailed review of the key issues associated with crash-frequency data as well as the strengths and weaknesses of the various methodological approaches that researchers have used to address these problems. While the steady march of methodological innovation (including recent applications of random parameter and finite mixture models) has substantially improved our understanding of the factors that affect crash-frequencies, it is the prospect of combining evolving methodologies with far more detailed vehicle crash data that holds the greatest promise for the future.

http://ceprofs.tamu.edu/dlord/Papers/Lord-Mannering_Review.pdf

The statistical analysis of crash-frequency data: A review and assessment of methodological alternatives

https://www.ce.utexas.edu/prof/bhat/ABSTRACTS/AMAR_review_article.pdf

Analytic methods in accident research: Methodological frontier and future directions

Identifying crash “hotspots”, “blackspots”, “sites with promise”, or “high risk” locations is standard practice in departments of transportation throughout the US. The literature is replete with the development and discussion of statistical methods for hotspot identification (HSID). Theoretical derivations and empirical studies have been used to weigh the benefits of various HSID methods; however, a small number of studies have used controlled experiments to systematically assess various methods.

Using experimentally derived simulated data—which are argued to be superior to empirical data, three hot spot identification methods observed in practice are evaluated: simple ranking, confidence interval, and Empirical Bayes. Using simulated data, sites with promise are known a priori, in contrast to empirical data where high risk sites are not known for certain.

To conduct the evaluation, properties of observed crash data are used to generate simulated crash frequency distributions at hypothetical sites. A variety of factors is manipulated to simulate a host of ‘real world’ conditions. Various levels of confidence are explored, and false positives (identifying a safe site as high risk) and false negatives (identifying a high risk site as safe) are compared across methods. Finally, the effects of crash history duration in the three HSID approaches are assessed.

The results illustrate that the Empirical Bayes technique significantly outperforms ranking and confidence interval techniques (with certain caveats). As found by others, false positives and negatives are inversely related. Three years of crash history appears, in general, to provide an appropriate crash history duration.

Experimental evaluation of hotspot identification methods

Mixed logit model-based driver injury severity investigations in single- and multi-vehicle crashes on rural two-lane highways

Intersection crash prediction modeling with macro-level data from various geographic units.

Development of comprehensive accident models for two-lane rural highways using exposure, geometry, consistency and context variables

At a European level, road accident fatalities on “two-lane rural roads” constitute up to 56% of the total number of fatalities caused by road accidents. This paper presents a comprehensive procedure starting from a low-cost data survey and leading up to the implementation of an Accident Prediction Model (APM) specifically designed for two lane rural roads. Firstly, significant variables were identified for the segmentation of highways into homogeneous sections. More specifically, research efforts is used to divide the roads sample into segments in which all the highway characteristics (exposure, geometric, consistency and context-related variables) were constant and such to have a minimum length significant for the accident expectation. All the basic data were obtained from low cost surveys purposely set up. In particular, differential cinematic GPS surveys were used to perform horizontal alignments and Road Safety Inspections (RSIs) were conducted in order to quantify other safety concerns. Based on the identified highway characteristics, different APMs were defined using the Generalized Linear Model approach (GLIM). Considering significant parameters related to exposure, geometric, consistency and context factors as explanatory variables, four APMs with significant goodness-of-fit were selected. Model 1, which considers only exposure, was used to define the length of the homogeneous section in order to obtain a minimum number of expected accidents. Model 2 represents the best statistical fit. Model 3 was defined to avoid illogical results. Finally, even if Model 4 does not represent the best fit of the data, it has the best engineering value as Safety Performance Indicator because at least one significant parameter related to exposure, geometric, consistency and context factors is included in the model.

Safety Performance Indicators for Local Rural Roads: Comprehensive Procedure from Low-Cost Data Survey to Accident Prediction Model

https://cyberleninka.org/article/n/23237.pdf

Application of highway safety manual to Italian divided multilane highways

Defining a black spot is not as perceptual as it may appear. Unusually high crash counts do not necessarily indicate a real safety problem. When black spots on two-lane rural roads with low and medium traffic volumes are being identified, a reduced number of observed crashes is a critical issue that can emphasize weak results obtained with various safety indicators and checks. It is generally assumed that an empirical Bayesian (EB) estimate represents the best approach for the identification of black spots, but it is difficult to define quantitatively the precision of estimates that are arrived at with different procedures. Because traffic volume, segment length, and crash observation period are critical issues in the identification of black spots and can emphasize the quality of the results obtained with safety indicators and checks, definition is needed of the limits and errors that can be expected when inadequate indicators and procedures are used to identify black spots on two-lane rural roads. A Monte Carlo simulation was used to produce theoretical crash data similar to empirical data, and the data were used to define a priori hazardous sites and, therefore, to assess whether a method could correctly identify such sites. The accuracy and efficiency of procedures are compared on the basis of observed frequency of crashes, crash rate, EB estimation, and the potential for safety improvement (PSI). As a general rule, even if some arrangements can mitigate the lower performance of some indicators, the best practice is to use the indicators based on the EB estimation PSI to identify true positives.

Performance of Safety Indicators in Identification of Black Spots on Two-Lane Rural Roads

Safety performance functions (SPFs) are commonly calibrated with negative binomial regression in which a dispersion parameter that represents extra-Poisson variation is estimated. The primary use of this parameter is in empirical Bayes estimation for safety management applications such as treatment evaluation and network screening. It stands to reason that the importance of precise estimation of the dispersion parameter should be established. Recent research has suggested that the dispersion parameter is not constant but actually varies from site to site, depending on site characteristics such as segment length. In revisiting the dispersion parameter issue in this empirical investigation, previous research on this issue is reinforced by filling a number of knowledge gaps. First, light is cast on the dispersion parameter variation for SPFs for two-lane rural roads, an important entity type for which there is little or no knowledge in this regard. In this study, more precise model forms are investigated to represent the variation with respect to the key variable, segment length. This investigation confirms that the dispersion parameter is inversely related to segment length but reveals that it is not inversely proportional to segment length, as suggested in other research. Second, it was found that the dispersion parameter is smaller and the variation less pronounced with better-specified models. Finally, additional evidence is provided to suggest that dispersion parameter variation matters more for shorter segment lengths.

Giacomo Di Silvestro

Papers

Development of comprehensive accident models for two-lane rural highways using exposure, geometry, consistency and context variables

Safety Performance Indicators for Local Rural Roads: Comprehensive Procedure from Low-Cost Data Survey to Accident Prediction Model

Application of highway safety manual to Italian divided multilane highways

Performance of Safety Indicators in Identification of Black Spots on Two-Lane Rural Roads

Revisiting Variability of Dispersion Parameter of Safety Performance for Two-Lane Rural Roads