Award Date

December 2015

Degree Type


Degree Name

Doctor of Philosophy (PhD)


Civil and Environmental Engineering

First Committee Member

Hualiang Teng

Second Committee Member

Mohamed Kaseko

Third Committee Member

Alexander Paz

Fourth Committee Member

Moses Karakouzian

Fifth Committee Member

Djeto Assane

Number of Pages



Safety analysis of freeway networks entails the quantification of crash frequency influencing factors which include roadway and traffic characteristics, environmental factors as well as human factors. This quantification can be used to detect locations with large impacts on the occurrence of crashes which in turn assist engineers and planners to improve safety levels of the network. Roadway characteristics are comprised of the physical elements of the road geometry such as section length, median and right shoulders, speed-exchange lanes, the number of main facility as well as geometry of the entrance from and exit to the main freeway facility. Traffic characteristics are comprised of traffic flow and vehicular volumes while environmental factors include weather conditions, pavement surface conditions, work zone areas conditions, and lighting conditions along the travel facility. Human factors are comprised of aging, aggressiveness while driving, mental stability, fatigue, alcoholism, acute psychological stress, suicidal behavior, drowsiness, and temporary distraction.

Variability in the crash frequency is captured by the interaction of the aforementioned factors either in a multiplicative or additive nature through the use of statistical model formulation. When all factors believed to influence the occurrence of crashes are included in a mathematical formulation and all the assumptions underlying the statistical model are met, variability in the crash frequency referred to as observed heterogeneity can be fully explained. However, not all information believed to generate crashes is available. Some of the factors are latent in nature and some are either not available at the time of analysis or require time and high cost to be established. When such conditions exist, a formulated model does not fully explain observed heterogeneity in the crash frequency. Lack of information to fully explain variability in crash frequency as a result of excluding some factors leads to unobserved heterogeneity problems which results in biased and inconsistent safety estimators.

Specifically, when observed crash counts are considered as clusters, analytical approach should consider the possibility of dependence within clustered crash counts. Correlation within clusters may be due to variation being induced by common unobserved cluster-specific factors. Ignoring cluster-effects increases the likelihood in drawing conclusion based on unrealistic inferences because safety estimator standard errors are likely to be underestimated and the usual conditional mean is no longer correctly specified. Cross sectional dependence may also arise when the crash counts have a spatial dimension due to contiguous freeway segments. Such conditions lead to what is known as spatial autocorrelation. This is the presence of spatial pattern in crash frequency over space due to geographic proximity whereby high values of crash frequency tend to cluster together in adjacent freeway segments or high crash frequencies are contiguous with low values of crash frequencies. When the distribution of crash frequency over space exhibit the aforementioned pattern, safety analysis techniques based on the distributional assumption of independence of crash frequency is violated.

This study has two objectives: First, analyze safety of freeway geometric features while accounting for the effect of unobserved influencing factors and cluster-specific effects; Second, analyze safety of freeway geometric elements in the presence of spatial autocorrelation due to geographical proximity effects. To achieve the first objective, four models are compared: Two are standard Poisson and Negative binomial regression models which do not account for cluster effects. The other two are mixed effects Poisson and Negative binomial regression models which in addition to fixed effects parts they account for the effects of randomness arising from heterogeneity and clustering.

The empirical results indicate that 13.9% of the variation in crash frequency is unaccounted for, which is an indication of the existence of unobserved factors influencing the occurrence of crashes. It is also revealed that weaving segments (EN-EX) had the highest between segment variance compared to non-weaving segments. More vehicles and short segments increased crash frequency while wider right shoulder decreased the crash frequency.

It is also observed that weaving segments decreased crash frequency compared to non-weaving segments. These results indicate that by allowing parameters to vary within the weaving and non-weaving segments it is possible to capture and quantify unobserved factors. Ignoring these factors results in biased coefficients because the estimate of the standard errors required determining inferential statistics will be wrong.

To achieve the second objective, Conditional Autoregressive models in Bayesian setting framework (CAR) is used. CAR models recognize the presence of spatial dependence which helps to obtain unbiased estimates of parameters quantifying safety levels since the effects of spatial autocorrelation is accounted for in the modeling process.

Based on CAR models, approximately 51% of crash frequencies across contiguous freeway segments are spatially autocorrelated. The incident rate ratios revealed that wider shoulder and weaving segments decreased crash frequency by factors of 0.84 and 0.75 respectively. The marginal impact graphs showed that an increase in longitudinal space for segments with two lanes decreased crash frequency. However, an increase of facility width above three lanes results in more crashes which indicates an increase in traffic flows and driving behavior leading to crashes. These results call an important step of analyzing contagious freeway segments simultaneously to account for the existence of spatial autocorrelation.


Bayesian spatial model; Freeway segments; Mixed effects Negative binomial; Mixed effects Poisson model; Second order spatial effects; Unobserved heterogeneity


Civil Engineering | Transportation

File Format


Degree Grantor

University of Nevada, Las Vegas




IN COPYRIGHT. For more information about this rights statement, please visit