Highway Safety Analytics and Modeling: Techniques and Methods for Analyzing Crash Data

Front-Matter_2021_Highway-Safety-Analytics-and-Modeling
Highway Safety Analytics and Modeling
Copyright_2021_Highway-Safety-Analytics-and-Modeling
Copyright
Dedication_2021_Highway-Safety-Analytics-and-Modeling
Dedication
Preface_2021_Highway-Safety-Analytics-and-Modeling
Preface
Chapter-1---Introduction_2021_Highway-Safety-Analytics-and-Modeling
1 . Introduction
1.1 Motivation
1.2 Important features of this textbook
1.3 Organization of textbook
1.3.1 Part I: theory and backbround
1.3.2 Part II: highway safety analyses
1.3.3 Part III: alternative safety analyses
1.3.4 Appendices
1.3.5 Future challenges and opportunities
References
Chapter-2---Fundamentals-and-data-colle_2021_Highway-Safety-Analytics-and-Mo
2 . Fundamentals and data collection
2.1 Introduction
2.2 Crash process: drivers, roadways, and vehicles
2.3 Crash process: analytical framework
2.4 Sources of data and data collection procedures
2.4.1 Traditional data
2.4.1.1 Crash data
2.4.1.2 Roadway data
2.4.1.3 Traffic flow data
2.4.1.4 Supplemental data
2.4.1.5 Other safety-related data and relevant databases
2.4.2 Naturalistic driving data
2.4.3 Disruptive technological and crowdsourcing data
2.4.4 Data issues
2.5 Assembling data
2.6 4-stage modeling framework
2.6.1 Determine modeling objective matrix
2.6.2 Establish appropriate process to develop models
2.6.3 Determine inferential goals
2.6.4 Select computational techniques and tools
2.6.4.1 The likelihood-based method
2.6.4.2 The Bayesian method
2.7 Methods for evaluating model performance
2.7.1 Likelihood-based methods
2.7.1.1 Maximum likelihood estimate
2.7.1.2 Likelihood ratio test
2.7.1.3 Likelihood ratio index
2.7.1.4 Akaike information criterion
2.7.1.5 Bayes information criterion
2.7.1.6 Deviance information criterion
2.7.1.7 Widely applicable information criterion
2.7.1.8 Bayes factors
2.7.1.9 Deviance
2.7.2 Error-based methods
2.7.2.1 Mean prediction bias
2.7.2.2 Mean absolute deviation
2.7.2.3 Mean squared prediction error
2.7.2.4 Mean squared error
2.7.2.5 Mean absolute percentage error
2.7.2.6 Pearson Chi-square
2.7.2.7 Coefficient of determination Rα2
2.7.2.8 Cumulative residuals
2.8 Heuristic methods for model selection
References
Chapter-3---Crash-frequency-modelin_2021_Highway-Safety-Analytics-and-Modeli
3 . Crash–frequency modeling
3.1 Introduction
3.2 Basic nomenclature
3.3 Applications of crash-frequency models
3.3.1 Understanding relationships
3.3.2 Screening variables
3.3.3 Sensitivity of variables
3.3.4 Prediction
3.3.5 Causal relationships
3.4 Sources of dispersion
3.4.1 Overdispersion
3.4.2 Underdispersion
3.5 Basic count models
3.5.1 Poisson model
3.5.2 Negative binomial model
3.5.3 Poisson-lognormal model
3.5.4 Other Poisson-mixture models
3.6 Generalized count models for underdispersion
3.6.1 Conway–Maxwell–Poisson model
3.6.2 Other generalized models
3.7 Finite mixture and multivariate models
3.7.1 Finite mixture models
3.7.2 Multivariate models
3.8 Multi-distribution models
3.8.1 Negative Binomial–Lindley model
3.8.2 Other multi-distribution models
3.9 Models for better capturing unobserved heterogeneity
3.9.1 Random-effects/multilevel model
3.9.2 Random-parameters model
3.9.2.1 Random parameters
3.9.2.2 Random parameters with means as a function of explanatory variables
3.10 Semi- and nonparametric models
3.10.1 Semiparametric models
3.10.2 Dirichlet process models
3.10.3 Nonparametric models
3.11 Model selection
References
Chapter-4---Crash-severity-modeling_2021_Highway-Safety-Analytics-and-Modeli
4 . Crash-severity modeling
4.1 Introduction
4.2 Characteristics of crash injury severity data and methodological challenges
4.2.1 Ordinal nature of crash injury severity data
4.2.2 Unobserved heterogeneity
4.2.3 Omitted variable bias
4.2.4 Imbalanced data between injury severity levels
4.3 Random utility model
4.4 Modeling crash severity as an unordered discrete outcome
4.4.1 Multinomial logit model
4.4.2 Nested logit model
4.4.3 Mixed logit model
4.5 Modeling crash severity as an ordered discrete outcome
4.5.1 Ordinal probit/logistic model
4.5.2 Generalized ordered logistic and proportional odds model
4.5.3 Sequential logistic/probit regression model
4.6 Model interpretation
References
Chapter-5---Exploratory-analyses-of-safe_2021_Highway-Safety-Analytics-and-M
5 . Exploratory analyses of safety data
5.1 Introduction
5.2 Quantitative techniques
5.2.1 Measures of central tendency
5.2.1.1 Mean
5.2.1.2 Median
5.2.1.3 Mode
5.2.2 Measures of variability
5.2.2.1 Range
5.2.2.2 Quartiles and interquartile range
5.2.2.3 Variance, standard deviation and standard error
5.2.2.4 Coefficient of variation
5.2.2.5 Symmetrical and asymmetrical data
5.2.2.6 Skewness
5.2.2.7 Kurtosis
5.2.3 Measures of association
5.2.3.1 Pearson's correlation coefficient
5.2.3.2 Spearman rank-order correlation coefficient
5.2.3.3 Chi-square test for independence
5.2.3.4 Relative risk and odds ratio
5.2.4 Confidence intervals
5.2.4.1 Confidence intervals for unknown mean and known standard deviation
5.2.4.2 Confidence intervals for unknown mean and unknown standard deviation
5.2.4.3 Confidence intervals for proportions
5.2.4.4 Confidence intervals for the population variance and standard deviation
5.2.5 Hypothesis testing
5.2.5.1 Decision errors
5.2.5.2 Two-tailed hypothesis test
5.2.5.3 One-tailed hypothesis test
5.2.5.4 Hypothesis testing for one sample
5.2.5.5 Hypothesis testing for two samples
5.2.5.6 Hypothesis testing for multiple samples
5.3 Graphical techniques
5.3.1 Box-and-whisker plot
5.3.2 Histogram
5.3.3 Bar graphs
5.3.4 Error bars
5.3.5 Pie charts
5.3.6 Scatterplots
5.3.7 Bubble chart
5.3.8 Radar/web plot
5.3.9 Heatmap
5.3.10 Contour plot
5.3.11 Population pyramid
References
Chapter-6---Cross-sectional-and-panel-stud_2021_Highway-Safety-Analytics-and
6 . Cross-sectional and panel studies in safety
6.1 Introduction
6.2 Types of data
6.2.1 Time-series data
6.2.2 Cross-sectional data
6.2.3 Panel data
6.3 Data and modeling issues
6.3.1 Overdispersion and underdispersion
6.3.2 Low sample mean and small sample size
6.3.3 Underreporting
6.3.4 Omitted variables bias
6.3.5 Endogenous variables
6.3.6 Unobserved heterogeneity
6.4 Data aggregation
6.5 Application of crash-frequency and crash-severity models
6.5.1 Functional form
6.5.1.1 Flow-only models
6.5.1.2 Flow-only models with CMFs
6.5.1.3 Model with covariates
6.5.2 Variable selection
6.5.3 Crash variance and confidence intervals
6.5.4 Sample size determination
6.5.5 Outlier analysis
6.5.6 Model transferability
6.6 Other study types
6.6.1 Cohort studies
6.6.2 Case-control studies
6.6.3 Randomized control trials
References
Chapter-7---Before-after-studies-in-highw_2021_Highway-Safety-Analytics-and-
7 . Before–after studies in highway safety
7.1 Introduction
7.2 Critical issues with before–after studies
7.2.1 Regression-to-the-mean
7.2.2 Site selection bias
7.3 Basic methods
7.3.1 Simple before–after study
7.3.2 Before–after study with comparison groups
7.4 Bayesian methods
7.4.1 Empirical Bayes method
7.4.1.1 Step 1—collect data for the treatment and comparison groups
7.4.1.2 Step 2—develop a regression model from the comparison group
7.4.1.3 Step 3—estimate the EB for the before period
7.4.1.4 Step 4—calculate rtf
7.4.1.5 Step 5—estimate the predicted value for the after period
7.4.1.6 Step 6—calculate the estimated value for the after period
7.4.1.7 Step 7—calculate the variance for the predicted and estimated values
7.4.1.8 Step 8—calculate the difference and index
7.4.1.9 Step 9—calculate the variance for the difference and index
7.4.1.9.1 Caution with the EB method
7.4.2 Bayes method
7.4.2.1 Step 1—calculate Rc
7.4.2.2 Step 2—predict π
7.4.2.3 Step 3—estimate θ
7.4.2.4 Step 4—estimate δ
7.4.2.5 Step 5—determine the significance of θ and δ
7.5 Adjusting for site selection bias
7.5.1 Example application for estimating θadj
7.5.1.1 Step 1—calculate the naïve estimate
7.5.1.2 Step 2—estimate the value of the variables inside Eq. (7.45)
7.5.1.3 Step 3—calculate the adjusted safety index (Eq. 7.45)
7.6 Propensity score matching method
7.7 Before–after study using survival analysis
7.8 Sample size calculations
7.8.1 Factor influencing sample size calculations
7.8.2 Sample size estimation using known crash counts for both time periods
7.8.3 Sample size based on the variance and ratio rd (before period)
References
Chapter-8---Identification-of-hazardous_2021_Highway-Safety-Analytics-and-Mo
8 . Identification of hazardous sites
8.1 Introduction
8.2 Observed crash methods
8.2.1 Crash frequency method
8.2.2 Crash rate method
8.2.3 Rate quality control method
8.2.4 Equivalent property damage only method
8.2.5 Severity index method
8.2.6 Composite safety score
8.3 Predicted crash methods
8.3.1 Potential for improvement using predicted crashes
8.3.2 Level of service of safety
8.4 Bayesian methods
8.4.1 Empirical Bayes method
8.4.2 Bayes method
8.5 Combined criteria
8.6 Geostatistical methods
8.6.1 Clustering methods
8.6.1.1 K-means clustering
8.6.1.2 Ripley's K-function
8.6.1.3 Nearest neighborhood clustering
8.6.1.4 Moran's I index
8.6.1.5 Getis-Ord general G∗(d)
8.6.2 Kernel density estimation
8.7 Crash concentration location methods
8.7.1 Sliding window method
8.7.2 Peak searching method
8.7.3 Continuous risk profile
8.8 Proactive methods
8.8.1 Identify focus crash types and facility types
8.8.2 Develop risk factors
8.8.3 Screen and prioritize candidate locations
8.9 Evaluating site selection methods
8.9.1 Site consistency test
8.9.2 Method consistency test
8.9.3 Total rank differences test
8.9.4 Total score test
8.9.5 False identification test
8.9.6 Poisson mean differences
References
Chapter-9---Models-for-spatial-data_2021_Highway-Safety-Analytics-and-Modeli
9 . Models for spatial data
9.1 Introduction
9.2 Spatial data and data models
9.3 Measurement of spatial association
9.3.1 Global statistics for spatial association
9.3.1.1 Getis–Ord general G∗(d)
9.3.1.2 Moran's I
9.3.2 Local indicators of spatial association
9.3.2.1 Local Gi∗(d)
9.3.2.2 Local Moran's Ii
9.4 Spatial weights and distance decay models
9.5 Point data analysis
9.5.1 First- and second-order process
9.5.2 Kernel density estimation
9.5.3 Ripley's K-function
9.5.4 Cross-K function
9.6 Spatial regression analysis
9.6.1 Spatial econometrics methods
9.6.1.1 Spatial autoregressive model
9.6.1.2 Spatial error model
9.6.2 Generalized linear model with spatial correlation
9.6.2.1 Generalized linear mixed model
9.6.2.2 Hierarchical Bayesian model
9.6.3 Modeling local relationships in crash data
References
Chapter-10---Capacity--mobility--and-s_2021_Highway-Safety-Analytics-and-Mod
10 . Capacity, mobility, and safety
10.1 Introduction
10.2 Modeling space between vehicles
10.3 Safety as a function of traffic flow
10.4 Characterizing crashes by real-time traffic
10.5 Predicting imminent crash likelihood
10.6 Real-time predictive analysis of crashes
10.6.1 Binary logistic regression model
10.6.2 Conditional logistic regression model
10.6.3 A note about binary logit and conditional logistic regression models
10.7 Using traffic simulation to predict crashes
10.7.1 Cell transmission model
10.7.2 Fundamental diagram calibration
10.7.3 CTM simulation algorithm
10.7.4 Crash modeling
10.7.5 Crash prediction
References
Chapter-11---Surrogate-safety-measur_2021_Highway-Safety-Analytics-and-Model
11 . Surrogate safety measures
11.1 Introduction
11.2 An historical perspective
11.3 Traffic conflicts technique
11.4 Field survey of traffic conflicts
11.5 Proximal surrogate safety measures
11.5.1 Collision course
11.5.2 Time- and distance-based proximal surrogate safety measures
11.5.2.1 Time to collision family
11.5.2.2 Encroachment time family
11.5.2.3 Proportion of stopping distance
11.5.2.4 Other indicators
11.6 Theoretical development of safety surrogate measures
11.6.1 Block maxima using the generalized extreme value distribution
11.6.2 Peak over threshold using the GP distribution
11.6.3 Block maxima or peak over threshold
11.7 Safety surrogate measures from traffic microsimulation models
11.8 Safety surrogate measures from video and emerging data sources
References
Chapter-12---Data-mining-and-machine-learni_2021_Highway-Safety-Analytics-an
12 . Data mining and machine learning techniques
12.1 Introduction
12.2 Association rules
12.3 Clustering analysis
12.3.1 K-means clustering
12.3.2 Latent class cluster
12.4 Decision tree model
12.4.1 The CART model
12.4.2 Random forest
12.4.3 Gradient boosted trees
12.5 Bayesian networks
12.6 Neural network
12.6.1 Multilayer perceptron neural network
12.6.2 Convolutional neural networks
12.6.3 Long short-term memory—recurrent neural networks
12.6.4 Bayesian neural networks
12.7 Support vector machines
12.8 Sensitivity analysis
References
Appendix-A---Negative-binomial-regression-model_2021_Highway-Safety-Analytic
A - Negative binomial regression models and estimation methods
. Probability density and likelihood functions
. Poisson-gamma model
NB-2 model
NB-1 model
. Poisson-gamma model with spatial interaction
. Estimation methods
Maximum likelihood estimation
Monte Carlo Markov Chain estimation
MCMC Poisson-gamma model
MCMC Poisson-gamma-CAR model
Prior distributions for MCMC Poisson-gamma-CAR
References
Appendix-B---Summary-of-crash-frequency-and-crash_2021_Highway-Safety-Analyt
B - Summary of crash-frequency and crash-severity models in highway safety
Introduction
Crash-frequency modeling
Crash-severity modeling
Crash modeling by model type
References
Appendix-C---Computing-codes_2021_Highway-Safety-Analytics-and-Modeling
C - Computing codes
Negative binomial model
SAS code
R code
WinBUGS code
Negative binomial model with varying dispersion parameter
SAS code
WinBUGS code
Radom effects negative binomial model
SAS code
WinBUGS code
Random parameters negative binomial model
SAS code
WinBUGS code
Poisson-lognormal model
WinBUGS code
Negative binomial-Lindley model
WinBUGS code
Conway–Maxwell–Poisson distribution
R code
Multinomial logit model
SAS code
Nested logit model
SAS code
Appendix-D---List-of-exercise-datase_2021_Highway-Safety-Analytics-and-Model
D - List of exercise datasets
Index_2021_Highway-Safety-Analytics-and-Modeling
Index
A
B
C
D
E
F
G
H
I
K
L
M
N
O
P
Q
R
S
T
U
V
W
Z