category: literaturenote citekey: pulugurthatrafficanalysiszonelevel2013 title: Traffic analysis zone level crash estimation models based on land use characteristics authors: "Pulugurtha, Srinivas S.; Duddu, Venkata Ramana; Kotagiri, Yashaswi" year: 2013 date: 2013-00-00 2013 doi: 10.1016/j.aap.2012.06.016 publication: "Accident Analysis & Prevention" url: "https://doi.org/10.1016/j.aap.2012.06.016" zotero_key: P5V88A56 zotero_storage: KW75PZ8B collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Pulugurtha, Srinivas S."

status: converted

Contents lists available at SciVerse ScienceDirect

Accident Analysis and Prevention

j ou rnal home page: www.elsevier.com/locate/aap

Traffic analysis zone level crash estimation models based on land use characteristics

Srinivas S. Pulugurtha^a,∗, Venkata Ramana Duddu^b,1, Yashaswi Kotagiri ^b,¹

^a Department of Civil and Environmental Engineering, Center for Transportation Policy Studies, The University of North Carolina at Charlotte, 9201 University City Boulevard, Charlotte, NC 28223-0001, USA
^b Department of Civil and Environmental Engineering, The University of North Carolina at Charlotte, 9201 University City Boulevard, Charlotte, NC 28223-0001, USA

a r t i c l e i n f o

Article history: Received 14 June 2010 Received in revised form 8 June 2012 Accepted 18 June 2012

Keywords: Crash Estimation Model Traffic analysis zone (TAZ) Land use characteristics GLM Negative binomial Safety

a b s t r a c t

The objective of this paper is to develop crash estimation models at traffic analysis zone (TAZ) level as a function of land use characteristics. Crash data and land use data for the City of Charlotte, Mecklenburg County, North Carolina were used to illustrate the development of TAZ level crash estimation models. Negative binomial count models (with log-link) were developed as data was observed to be over-dispersed. Demographic/socio-economic characteristics such as population, the number of household units and employment, traffic indicators such as trip productions and attractions, and, on-network characteristics such as center-lane miles by speed limit were observed to be correlated to land use characteristics, and, hence were not considered in the development of TAZ level crash estimation models. Urban residential commercial, rural district and mixed use district land use variables were observed to be correlated to other land use variables and were also not considered in the development of the models. Results obtained indicate that land use characteristics such as mixed use development, urban residential, singlefamily residential, multi-family residential, business and, office district are strongly associated and play a statistically significant role in estimating TAZ level crashes. The coefficient for single-family residential area was observed to be negative, indicating a decrease in the number of crashes with an increase in single-family residential area. Models were also developed to estimate these crashes by severity (injury and property damage only crashes). The outcomes can be used in safety conscious planning, land use decisions, long range transportation plans, and, to proactively apply safety treatments in high risk TAZs. © 2012 Elsevier Ltd. All rights reserved.

1. Introduction

According to National Highway Traffic Safety Administration (NHTSA), more than 5.5 million reported crashes occurred in the United States during 2009 (NHTSA, 2009). Over 33,000 people were killed and 2.2 million people were injured in these crashes. Federal agencies have made reducing crashes by considering safety every time and at every stage of a project a top priority (Federal Highway Administration – FHWA, 2011).

As traffic increases with growth in population, the conflicts that arise because of human interaction, off- and on-network characteristics also increase. Understanding the causes of the crashes, identifying appropriate solutions, and, proactively adopting or using them helps improve traffic safety.

Several authors in the past have examined the role of traffic volume on crashes at link-level or intersections. Examples include Levine et al. (1995a,b), Ivan et al. (2000), Abdel-Aty and Radwan

Traffic volumes from calibrated traditional four-step planning process depend on estimated trip productions and trip attractions of each traffic analysis zone (TAZ) in the study area. TAZs represent geographic locations of trip origins and destinations. Each TAZ is characterized by land use and demographic characteristics that allow planners to estimate the number oftrips likely to be produced

^(2000), Kim and Yamashita (2002), Noland and Oh (2004), Greibe (2003), Ladron de Guevara et al. (2004), Aguero-Valverde and Jovanis (2006), Caliendo et al.(2007), Naderan and Shahi(2010) and Pulugurtha and Nujjetty (2012). Data collected fromthefield and/or outputs fromwell-calibratedtraditionalfour-stepplanningprocess are potential sources of traffic volume at link level. While the data from field are collected based on the need or as a part oftraffic count programs and are available for selected links/intersections, traffic volumes are available as outputs for all major roads from traditional four-step planning process. Resource constraints (includes funds) often limit practitioners from collecting traffic volume data for all links/intersections in the transportation network but instead rely on outputs from calibrated traditional four-step planning process. These link volumes are not applicable for area-level modeling.

^∗ Corresponding author. Tel.: +1 704 687 1233.

E-mail addresses: sspulugurtha@uncc.edu (S.S. Pulugurtha),

vduddu@student.uncc.edu (V.R. Duddu), ykotagir@student.uncc.edu (Y. Kotagiri). ¹ Tel.: +1 704 687 1250.

from and attracted to that zone (Meyer and Miller, 2001). Achieving homogenous demographic/socio-economic characteristics, minimizing the number of intra-zonal trips and recognizing physical, political, jurisdictional, and historical boundaries are some criteria adopted in defining TAZs (Baass, 1981).

The trips produced from or attracted to a TAZ are generally estimated as a function of demographic/socio-economic and land use characteristics. The demographic and socio-economic characteristics used to estimate trip productions and trip attractions at TAZ level depend on land use characteristics (example, population of a TAZ depends on the number of dwelling/household units and the average household size of the TAZ). Further, transportation infrastructure (on-network characteristics) or project related decisions often depend on traffic volumes from calibrated traditional four-step planning process which as stated above depend on socio-economic/demographic/land use characteristics.

As trips produced from and attracted to a TAZ increases, the traffic volume increases on links in the TAZ (particularly, in TAZs with commercial activity, offices, mixed use development, and residences that are typical activity generators). New developments or increasing the intensity or changing thenature of existing economic activity will have implications on safety (Kim et al., 2006). Land use planning decisions therefore can influence demographic/socioeconomic characteristics of an area (say, TAZ), and hence traffic volume and crashes. While this appears obvious, the patterns in terms of crashes and the underlying use of land are difficult to describe (Kim and Yamashita, 2002).

It is widely known and accepted that traffic crashes are location specific and are influenced by land uses and network characteristics within its vicinity. Research even has shown that on- and off-network characteristics within 0.25-mile accessible distance can be used to statistically explain and estimate pedestrian crashes (Pulugurtha and Sambhara, 2011) and all crashes (Pulugurtha and Nujjetty, 2012) at signalized intersections. Such models based on data within the vicinity of a location could be used to estimate crashes and improve safety at the specific location. Typicallocationspecific improvements are relatedto engineering, education and enforcement and are generally short-term projects. However, developing aggregate (relatively macroscopic) area level models based on land use characteristics (or other variables) would help incorporate safety into the transportation and land use planning process, and possibly derive maximum benefit.

The primary focus of this paper therefore is to understand the role of various land use characteristics on crashes, examine their correlation with selected on-network characteristics (includes traffic indicators) and develop TAZ level crash estimation models as a function of land use characteristics. It provides a framework to understand and incorporate safety in long range transportation planning and land use planning decisions (rather than analyzing crash data, identifying causes/solutions, and implementing location specific countermeasures as an "after-thought"). The outcomes can also be used in safety conscious planning and to proactively apply safety treatments in possible high risk TAZs (similar to traffic impact studies conducted to maintain operational performance on roads or at intersections near new developments) by planners, developers and safety experts.Inaddition, by aggregating the transportation system at the TAZ level, some of the difficulties caused by 'lumpiness' of random events that one see across intersections or across road segments are reduced (Washington et al., 2006).

2. Literature review

As stated previously, traffic volume depends on demographic/ socio-economic, land use and on-network characteristics. A coherent discussion on the effect ofthese characteristics on crashes is presented next.

Noland and Quddus (2003), Ladron de Guevara et al. (2004) and Wier et al. (2009) observed that population density and the number of employees (employment density) play a significant role in predicting crashes. Noland and Quddus (2003) also observed that traffic casualties in England are higher in areas (wards) with higher levels of social deprivation. Kim et al. (2006) modeled using a uniform 0.1-mi2 grid structure and found statistically significant relationships among population, job counts, economic output and crashes. Aguero-Valverde and Jovanis (2006) concluded that the counties with higher percentages of the population below the poverty line and those with younger and older drivers have significantly increased crash risks. Quddus (2008), based on non-spatial negative binomial models and spatial Bayesian hierarchical models at census ward level, observed that households with no cars and total employment are statistically significant variables to predict crashes.

Researchers have also examined the role of land use characteristics and area type on crashes in the past. Levine et al. (1995a,b), based on data for Honolulu, Hawaii, observed that most of the crashes occur closer to employment centers than residential areas. Their study also concludedthat crashes inrural andsub-urbanareas are most likely to be fatal or serious injury crashes that can be related to night-time driving and alcohol. Contrary to the observations by Levine et al. (1995), Kim and Yamashita (2002) found that residential neighborhoods, which have higher traffic volumes only during the peak hours, tend to have higher numbers of crashes than commercial centers that have high traffic volumes throughout the day. However, business and commercial areas are strongly associated with increased injury and fatal crashes (Kim et al., 2010).

Literature also documented research on predicting crashes based on on-network characteristics and roadway geometry. Ivan et al.(2000) observed that volume-capacity ratio, segments with no passing zones, the shoulder width, the number of intersections (or intersection density) and driveways (or driveway density) are significant variables that can explain single-vehicle crashes, whereas daylight conditions, the number of intersections and driveways are significant variables that can explain multi-vehicle crashes. Greibe (2003) found that road environment variables, minor side roads, parking facilities and speed limit are significant variables that can explain crashes on road links while only vehicle traffic flow was found to be a significant variable for road junctions. Noland and Oh (2004) examined that an increase in the number of lanes and road widths may lead to an increase in the number crashes, while an increase in shoulder width may lead to a decrease in the number crashes. Noland and Oh (2004) observed that changes in vertical andhorizontal curvatureshaveno statistical association withtraffic safety, while Caliendo et al.(2007) observed that degree of horizontal curves, shoulder width and median width play a significant role on frequency of crashes. Aguero-Valverde and Jovanis (2006) concluded that counties with increased road mileage and density have significantly increased crash risks. Wier et al. (2009) found that the proportion of arterial streets without transit may have a bearing on crashes. Recently, Kim et al. (2010) observed that accessibility measures such as the number of bus stops and the number of intersections are associated with an increase in all types of crashes.

Different levels of crash severity were taken into account while investigating the associations between crash occurrence and explanatory variables in the past. Noland and Quddus (2003), Ma et al. (2008), Abdel-Aty et al. (2010), Huang and Abel-Aty (2010), Naderan and Babaei (2011), and Pei et al. (2011) reviewed the mathematical formulation of integer, count and multilevel data structures (that describe the severity and nature of crash) into crash estimation models. Kim et al. (2010) concluded that job counts and the number of people living below the poverty level

Table 1 Description of land use variables.

Land use category	Description
Mixed use development (MUDEV)	Areas with residential and compatible non-residential uses less than 10 acres to serve the residents of the planned community
Mixed use district (MUDIS)	Areas with residential and compatible non-residential uses greater than 10 acres to serve the residents of the planned community
Urban residential (UR)	Single-family to higher-density residential development nearer to the employment core
Industrial (IND)	Areas with manufacturing, processing, and assembling of parts, distribution centers, and transportation terminals; specialized industrial
Business (BUS)	Areas with retailing of merchandize to serve a large trade area, warehousing, wholesaling, etc.
Urban residential commercial (URC)	Areas with residential, retail, office, recreational, and cultural uses
Multi-family	Areas with a variety of housing types; 12–43 dwelling units per acre
Office district (OD)	Areas conducive to establishment and operation of offices, institutions, and commercial activities not involved in sales
Single-family (SF)	Areas with primarily single-family housing; 3–8 dwelling units per acre
Institutional (INS)	Major cultural, educational, medical, governmental, religious, athletic, and other institutions
Neighborhood service district (NSD)	Mixed-use areas focusing on neighborhood retail and service activities
Right-of-way (ROW)	Right of way areas of interstates, major and minor thoroughfares and other roads
Commercial center (CC)	Shopping centers and individual retail establishments larger than 70,000 ft2 of floor area
Innovative (INN)	Non-traditional and new type of land use
Planned unit development (PUD)	Area with a variety of type, design, and arrangement of structures
Rural district (RURD)	Areas rural in nature
Research district (RESD)	Areas with high research, development and technology manufacturing operations and professional employment
Manufactured house (MH)	Areas with homes manufactured in a factory, transported as a whole unit and used at the site to build a house
Residential mobile (RM)	Areas with manufactured home and mobile home parks

are significantly associated with injury crashes, pedestrian crashes and bicycle crashes. A TAZ level non-parametric safety analysis by Siddiqui et al. (2012) showed that the total number of intersections per TAZ, light truck productions, total roadway segment length with 35 mph posted speed limit, total roadway length with 65 mph posted speed limit, and non-home based work productions are significant variables influencing severe crashes.

Linear as well as non-linear distributions were used in developing crash estimation models in the past. Wood (2002) described the underlying mechanism of generalized linear crash (or count) models and practical resolution to address the 'low mean value' problem which is barrier to goodness of fit testing. Typical nonlinear distributions that were considered in the past to develop crash estimation models included Poisson distribution (Ivan et al., 2000; Wood, 2002, 2005; Caliendo et al., 2007; Ma et al., 2008) and negative binomial distribution (Abdel-Aty and Radwan, 2000; Wood, 2002, 2005; Noland and Quddus, 2003; Ladron de Guevara et al., 2004; Kim et al., 2006; Aguero-Valverde and Jovanis, 2006; Caliendo et al., 2007; Quddus, 2008; Pulugurtha and Sambhara, 2011; An et al., 2011; Pulugurtha and Nujjetty, 2012).

2.1. Limitations of past research

Most researchers have focused on crash estimation models based on demographic characteristics, socio-economic characteristics, and on-network (includes traffic volume) characteristics or on the effect of these variables on crashes. A few researchers have considered land use characteristics but along with other predictor variables. A strong correlation may exist between land use characteristics and these other predictor variables (in particular, when considered at an area or TAZ level).

While literature documents vast research on safety models for segments and intersections (location specific), not many authors have looked at area or TAZ level crash estimation models for planning purposes. As stated earlier, area-wide models would lower difficulties caused by 'lumpiness' of random events across intersections or across road segments (Washington et al., 2006).

Researchers who have considered area level, spatial proximity or trip generation data (Naderan and Shahi, 2010; Abdel-Aty et al., 2010; An et al., 2011) for safety conscious planning did not examine the direct relationship between land use characteristics and crashes at TAZ level. This is particularly important if a strong correlation exists between land use characteristics and demographic/socio-economic and network characteristics. Identifying land use characteristics that contribute to safety problems and developing crash estimation models based on land use characteristics (focus of this research) at TAZ level would help in making better long range transportation plans and land use decisions.

3. Study area, data and database to develop models

The City of Charlotte, Mecklenburg County, North Carolina was considered as the study area. Crash data, land use data, street centerline network and TAZ layer (with planning variables data, and tripproductions andattractions)for the year 2005 was obtainedina GIS formatfrom the City of Charlotte Department of Transportation (CDOT).

While traffic volume may be appropriate when modeling at point or link level, vehicle miles traveled would be more appropriate when modeling at area level. However, traffic volume data are not available for all the links in the network to calculate vehicle miles traveled. Even outputs (traffic volume at link level) from traditional four-step planning process are limited to major roads which are part of regional transportation planning model network. Collector roads and local roads (which comprise a vast portion of road network) are nottypically part of regionaltransportationplanning model network. Hence, trip productions and attractions of a TAZ are considered as traffic indicators for modeling purposes in this research. As stated previously, trip productions and attractions greatly influence link volumes on links in a TAZ.

Data for 1057 TAZs in the study area were considered to develop and validate TAZ level crash estimation models. A total of 10,726 reported crashes occurred in these TAZs during the year 2005. It was observed that there were 24 fatal crashes, 3522 injury crashes and 7180 PDO crashes in the study area during the year 2005. Of the initially considered 1057 TAZs, 24 TAZs were excluded from further analysis and model development as they had open land area or unknown type land area. Data for 765 TAZs with a total of 9799 crashes (includes 20 fatal crashes, 3227 injury crashes and 6552 PDO crashes) were selected for the analysis and development of models, while data for 268 randomly selected TAZs (∼35% when compared to 765 TAZs) was used to validate the models developed.

3.1. Spatial analysis and autocorrelation

The crash data (shapefile – points) was overlaid on the TAZ layer to extract and estimate the total number of crashes and the number of crashes by severity (fatal crashes, injury crashes, and property damage only – PDO crashes) for each TAZ in GIS environment. Fig. 1 shows the number of crashes in each TAZ during the year 2005 in the study area.

The land use data (shapefile – polygons) was overlaid on the TAZ layer using "intersect" feature in GIS environment to estimate the area of each land use characteristic for each TAZ. Table 1 briefly describes the land use characteristics considered in this research.

The street centerline network (shapefile – lines) was overlaid on the TAZ layer using "intersect" feature in GIS environment to summarize the total length (center-lane miles) for each TAZ.

As crashes in each TAZ's are location specific, spatial autocorrelation (Goodchild, 1986) was investigated to evaluate whether there is any influence of crashes in a TAZ on its adjacent TAZ's. Moran's I was calculated in GIS environment to measure spatial autocorrelation. An inverse distance (Euclidean distance) method was used to indicate a decrease in the influence with an increase in the distance from crash location. For the data considered in this research, the Moran's I value was calculated to be 0.07 (very low) with a Z-score of 48.38 and a P-value less than 0.01. In general, Moran's I closer to +1 indicates highly positive spatial autocorrelation (clustering or nearby areas tend be similar), closer to −1 indicates highly negative spatial autocorrelation (dispersion or nearby areas tends to be dissimilar), and closer to 0 indicates zero spatial autocorrelation (more random and independent). From the results obtained, it was felt that a crash or crashes in a TAZ are relatively random and do not influence other crashes or in adjacent TAZs (or TAZs in close proximity). This could be attributed to the criteria that are usually adopted to define TAZ boundaries with homogeneous and unique characteristics.

3.2. Examination of correlation between independent variables

Table 2 summarizes the minimum, maximum and mean value of crash and land use characteristics for 765 TAZs considered to develop models in this paper.

As stated earlier, the primary focus of this paper is to better understand the role of land use on crashes and develop crash estimation models as a function of land use characteristics at a TAZ level. As literature documents an association between selected demographic, socio-economic and network characteristics and crashes, one needs to test the correlation between demographic/socio-economic characteristics (population, the number of household units and employment), traffic indicators (trip productions and attractions), and on-network characteristics (center-lane miles by speed limit) and land use characteristics to minimize any possible bias that might arise due to eliminating these variables. Further, correlation may exist between various land use variables. Statistical tests were therefore conducted by computing Pearson correlation coefficient to examine the correlation between population, number of household units, employment,trip productions,trip attractions, center-lane miles in eachTAZ and landuse characteristics.Inthis research,two variables were considered to be strongly correlated to each other if the computed Pearson correlation coefficient is less than −0.2 or greater than +0.2 (significance value less than 0.01 for the considered data).

Table 3 summarizes computed Pearson correlation coefficients between allthe selected independent variables. Population of a TAZ was observed to be significantly correlated to single-family (SF) residential, multi-family (MF) residential, planned unit development (PUD) and innovative (INN) land use areas. The total number of household units in a TAZ was observed to be significantly correlated to single-family (SF) residential, multi-family (MF) residential and innovative (INN) land use areas. Employment or job count was observed to be significantly correlated to industrial(IND), office district (OD), mixed use development (MUDEV) and research district (RESD) land use areas.

Trips productions from and trip attraction to a TAZ were observed to be significantly correlated to industrial (IND), multifamily (MF) residential, office district (OD) and institutional (INS) land use areas.

The total number of center-lane miles was observed to be significantly correlated to mixed use district (MUDIS), single-family (SF) residential, multi-family (MF) residential and planned unit development(PUD)land use areas. The total number of center-lane miles with minor roads (local and collector roads with speed limit equal to 25 mph or 35 mph) was observed to be significantly correlated to mixed use district (MUDIS), single-family (SF) residential, multifamily (MF) residential and planned unit development (PUD) land use areas, while major roads (arterials and expressways with speed limit greater than or equal to 40 mph but less than 60 mph) was observed to be significantly correlated to business (BUS) land use. In general, total center-lane miles of major roads were found to be highly correlated to total center-lane miles of all roads.

The urban residential commercial (URC), rural district (RURD) and mixed use district (MUDIS) land use variables were observed to be significantly correlated to urban residential (UR), planned unit development (PUD) and single-family (SF) land use variables, respectively.

Therefore, omitting demographic, socio-economic and network characteristics as well as urban residential commercial (URC), rural district (RURD) and mixed use district (MUDIS) land use variables will not only minimize any possible multicollinearity but will help examine the direct effect of selected land use characteristics on crashes.

4. Selection of distribution function to develop models

A mathematical relationship between the dependent variable and independent variables is the basis of modeling process. The nature of data plays a vital role in developing a model. As discussed in the "Literature Review" section, linear as well as non-linear distributions (Poisson with log-link, negative binomial with log-link, and log-normal) were used to develop crash estimation models in the past. Poisson distribution is generally used to model count data if the mean is equal to its variance. However, if data are overdispersed, the variance will be greater than mean. In such a case, a negative binomial distribution is used in the modeling process. These negative binomial distribution models can take into account the effect of unobserved heterogeneity due to omitted variables or variables that were not considered among TAZs.

Eq.(1) shows the relationbetweenthe variance andmeanincase of a negative binomial distribution (McCullagh and Nelder, 1989).

$$Variance = \sigma^2 = \mu + \alpha \mu^2 \tag{1}$$

where is the standard deviation of crashes, is the estimated mean number of crashes and ˛ is the negative binomial dispersion parameter.

Fig. 1. Spatial distribution of total crashes per TAZ in Charlotte, North Carolina.

Linear as well as non-linear distributions (Poisson with log-link, negative binomial with log-link, and log-normal) were developed using SPSS® (SPSS, 2008) software package. Independent variables with Wald Chi-square value less than 1 or the level of significance greater than 0.05 (95% confidence level) were considered to have a statistically insignificant effect on crashes. These parameters were examined for each independent variable of each modelto eliminate those that have a statistically insignificant effect on crashes as well as to minimize specification error.

The quasi-likelihood under independence model criterion, QIC (to select the model with best working correlation structure) and the corrected version of quasi-likelihood under independence model criterion, QICC (to choose the best subset of predictors) were used as statistical parameters to assess the goodness of fit. The lower the QIC and QICC, the better is the goodness of fit. Also, the difference between QIC and QICC should be generally low for a good model.

The computed variance based on crash data for the study TAZs in this paper was observed greater than the mean number of crashes (shown in Table 2) in all the three cases (total crashes, injury crashes and PDO crashes). The data was observed to be over-dispersed (˛ greater than zero) and not spatially correlated indicating that a negative binomial distribution is more suitable to estimate TAZ level crashes based on land use characteristics in this study.

5. TAZ level crash estimation models

Crash estimation models were developed separately considering the total number of crashes, the total number of injury crashes

Table 2 Summary of land use characteristics.

Variable	N	Minimum	Maximum	Mean	Standard deviation
Number of TAZs	765
Dependent variables
Total number of crashes (TC)	765	0.000	113.000	12.809	14.726
Number of injury crashes (IC)	765	0.000	37.000	5.100	4.218
Property damage only crashes (PDO) Independent variables (in mi2)	765	0.000	76.000	10.127	8.565
Mixed use development (MUDEV)	765	0.000	0.221	0.002	0.011
Mixed use district (MUDIS)	765	0.000	1.386	0.019	0.100
Urban residential (UR)	765	0.000	0.155	0.001	0.009
Industrial (IND)	765	0.000	3.658	0.057	0.181
Business (BUS)	765	0.000	0.660	0.018	0.042
Urban residential commercial (URC)	765	0.000	0.023	0.000	0.001
Multi-family (MF)	765	0.000	0.328	0.029	0.050
Office district (OFFD)	765	0.000	0.178	0.008	0.022
Single-family (SF)	765	0.000	1.919	0.256	0.297
Institutional (INS)	765	0.000	0.882	0.006	0.041
Neighborhood service district (NSD)	765	0.000	0.048	0.001	0.004
Right-of-way (ROW)	765	0.000	0.086	0.001	0.006
Commercial center (CC)	765	0.000	0.315	0.005	0.027
Innovative (INN)	765	0.000	0.118	0.002	0.011
Planned unit development (PUD)	765	0.000	0.472	0.008	0.046
Rural district (RURD)	765	0.000	0.005	0.000	0.000
Research district (RESD)	765	0.000	0.904	0.004	0.048
Manufactured house (MH)	765	0.000	0.004	0.000	0.000

and the total number of PDO crashes as the dependent variable. It was observed that the 765 TAZs data considered for modeling have only 20 fatal crashes (out of 9799 crashes). Also, only 2.61% of all the TAZ's considered have fatal crashes. This percent is too low to even test zero-inflated models. So, a crash estimation model for fatal crashes was not developed due to fewer numbers of fatal crashes and statistically insignificant sample size observed in the study TAZs.

All land use characteristics shown in Table 2, excluding urban residential commercial (URC), rural district (RURD) and mixed use district (MUDIS), were considered as the independent variables. Though data was observed to be over-dispersed, linear, Poisson with log-link and log-normal distributions were also tested in addition to negative binomial with log-link distribution. QIC and QICC obtained for negative binomial with log-link distribution based models were observed to be the lowest implying that these models would result in lower specification errors than when compared to other models tested in this research. Therefore, results obtained considering negative binomial with log-link are only discussed in this paper.

The statistical parameters along with the coefficient, Wald Chisquare and significance value for each variable in the preliminary model to estimate the total number of crashes per year in a TAZ are shown in Table 4. It was observed that a few land use characteristics have a significance value greater than 0.05 (at a 95% confidence level). In other words, these land use characteristics do not have a statistically significant effect on the total number of crashes in a TAZ.

The land use characteristics with a significance value greater than 0.05 were omitted and the model was re-run. The statistical parameters along with the coefficient, standard error Wald Chisquare and significance value for each variable in the final model with only land use characteristics that have statistically significant effect on the total number of crashes are shown in Table 5. Similarly, crash estimation models were developed to predict injury and PDO crashes in a TAZ. Results obtained are also summarized and shown in Table 5.

The coefficient for all the land use characteristics other than single-family (SF) residential is positive for the models to estimate total number of crashes and PDO crashes. This implies that an increase in the area of these land use characteristics will lead to an increase in the total number of crashes. An increase in single-family (SF) residential land use area will result in a decrease in the total number of crashes and PDO crashes.

The coefficient for all the land use characteristics other than single-family (SF) residential and industrial (IND) was positive for the model to estimate the number of injury crashes in a TAZ. This implies that an increase in the area of these land use characteristics will lead to an increase in the number of injury crashes. An increase in single-family (SF) residential and industrial (IND) land use areas will result in a decrease in the number of injury crashes.

The institutional (INS), industrial (IND) and research district (RESD) land use characteristics have a significant effect in estimating the total number of crashes and/or injury crashes but not PDO crashes. This shows that, the presence of these characteristics in a TAZ might contribute to an increase in the severity of a crash (more injury crashes).

The negative binomial dispersion parameter value (˛) was lower for the model to estimate the number of injury crashes than when compared to the models to estimate the total number of crashes and the number of PDO crashes. This indicates that injury crashes are less dispersed when compared to the total number of crashes and PDO crashes. QIC and QICC were reasonably close for all the three final models.

5.1. Validation of models

The models developed were validated using data for 268 randomly selected TAZs that were not considered in the development of models. The null hypothesis was defined as "the observed number of crashes is equal to the estimated number of crashes" while the alternate hypothesis was defined as "the observed number of crashes is not equal to the estimated number of crashes". The null hypothesis was not rejected if the computed Chi-square statistic using Eq. (2) (McCullagh and Nelder, 1989) was lower than the critical Chi-square statistic at a 95% confidence level.

Chi-square statistic = $$\sum_{L}^{N} \frac{(O - E)^2}{E + \alpha E^2}$$ (2)

where N is the number of TAZs for validation of models, O is the observed number of crashes in a TAZ, E is the estimated number

Table 3Pearsoncorrelation coefficients – summary.

		# ho hold use		Trip	Trip	Cen	lane ter-	Mi les
Var iab le	Pop ulat ion	s unit s	Em ploy t men	pro duc tion s	attr acti ons	All ds roa	Min or ds roa	Maj ors ds roa	MU DE V	MU DIS	UR	IND	BU S	UR C	MF	OD	SF	INS	NS D	RO W	CC	INN	PUD	RU RD	RE SD	MH
ulat ion Pop	1.00
# ho hold use unit s	0.95	1.00
Em ploy t men	-0.0 6	-0.0 6	1.00
Trip duc tion pro s	0.32	0.31	0.81	1.00
Trip acti attr ons	0.32	0.31	0.81	1.00	1.00
All ds roa	0.60	0.57	0.05	0.22	0.22	1.00
Min oad or r s	0.64	0.61	0.03	0.21	0.21	0.98	1.00
Maj oad or r s	0.04	0.00	0.12	0.09	0.09	0.45	0.26	1.00
MU DE V	-0.0 9	-0.0 8	0.24	0.15	0.15	0.03	0.0 1	0.09	1.00
MU DIS	-0.0 1	-0.0 1	-0.0 4	-0.0 4	-0.0 4	0.37	0.38	0.09	-0.0 1	1.00
UR	0.03	0.06	0.05	0.06	0.06	0.0 1	0.0 1	0.00	0.08	0.00	1.00
IND	-0.1 3	-0.1 4	0.43	0.21	0.21	0.09	0.06	0.17	0.06	-0.0 1	-0.0 3	1.00
S BU	0.03	0.04	0.15	0.17	0.17	0.07	0.03	0.21	0.0 1	-0.0 3	-0.0 4	0.0 1	1.00
UR C	0.02	0.08	0.00	0.02	0.02	-0.0 1	-0.0 1	0.0 1	0.08	-0.0 1	0.34	-0.0 2	-0.0 2	1.00
MF	0.54	0.60	0.00	0.21	0.21	0.30	0.30	0.10	-0.0 6	-0.0 7	-0.0 3	-0.0 1	0.1 1	-0.0 4	1.00
OD	0.05	0.07	0.31	0.33	0.33	0.10	0.07	0.19	0.13	-0.0 2	-0.0 2	0.0 1	0.14	-0.0 1	0.07	1.00
SF	0.37	0.33	-0.1 8	-0.0 6	-0.0 6	0.66	0.68	0.16	-0.1 1	0.25	-0.0 6	-0.1 0	-0.0 8	-0.0 6	0.04	-0.0 2	1.00
INS	0.18	0.03	0.12	0.21	0.20	0.09	0.07	0.1 1	-0.0 2	-0.0 2	-0.0 2	-0.0 1	0.04	0.0 1	0.05	0.02	-0.0 3	1.00
NS D	-0.0 1	-0.0 1	0.07	0.03	0.03	0.09	0.09	0.06	-0.0 1	0.04	0.0 1	0.03	0.02	-0.0 1	0.04	0.12	0.13	0.08	1.00
RO W	0.00	0.00	-0.0 3	-0.0 2	-0.0 2	0.1 1	0.07	0.18	-0.0 1	0.0 1	-0.0 1	-0.0 3	0.00	-0.0 1	-0.0 4	0.00	0.12	-0.0 1	-0.0 1	1.00
CC	0.00	0.00	0.04	0.09	0.09	0.19	0.17	0.15	-0.0 1	0.16	-0.0 2	-0.0 4	0.06	0.00	0.0 1	0.03	0.04	0.04	-0.0 1	0.03	1.00
INN	0.19	0.22	-0.0 4	0.05	0.05	0.1 1	0.12	-0.0 1	-0.0 2	-0.0 3	0.00	-0.0 6	-0.0 3	-0.0 1	0.07	0.00	0.12	-0.0 2	-0.0 3	-0.0 2	0.04	1.00
PUD	0.22	0.20	-0.0 4	0.05	0.05	0.24	0.26	0.03	-0.0 3	0.05	-0.0 1	-0.0 4	0.00	-0.0 1	-0.0 1	0.00	0.1 1	-0.0 2	-0.0 2	0.1 1	0.08	0.09	1.00
RU RD	0.15	0.12	-0.0 1	0.03	0.03	0.12	0.13	0.00	-0.0 1	-0.0 1	0.00	0.03	-0.0 1	0.00	0.00	-0.0 1	0.05	-0.0 1	-0.0 1	0.00	-0.0 1	-0.0 1	0.36	1.00
RE SD	-0.0 2	-0.0 1	0.30	0.15	0.15	0.05	0.04	0.06	0.13	-0.0 1	-0.0 1	-0.0 3	-0.0 1	-0.0 1	0.02	0.03	-0.0 2	0.08	-0.0 1	-0.0 1	-0.0 2	-0.0 2	-0.0 2	0.00	1.00
MH	-0.0 3	-0.0 3	-0.0 1	-0.0 2	-0.0 2	0.00	-0.0 3	0.1 1	-0.0 1	-0.0 1	0.00	0.06	-0.0 2	0.00	-0.0 2	-0.0 1	0.04	-0.0 1	-0.0 1	0.00	-0.0 1	-0.0 1	-0.0 1	0.00	0.00	1.00

Table 4 Preliminary model parameters to estimate the total number of crashes in a TAZ.

Variable	Coefficient	Wald Chi-square	Significance value
(Intercept)	2.223	914.34	<0.001
Mixed use development (MUDEV)	8.213	9.41	0.002
Urban residential (UR)	3.813	3.82	0.051
Industrial (IND)	0.005	0.01	0.936
Business (BUS)	7.670	1,325.03	<0.001
Multi-family (MF)	4.038	99.92	<0.001
Office district (OD)	7.123	102.04	<0.001
Single family (SF)	−0.426	23.13	<0.001
Institutional (INS)	0.256	4.10	0.043
Neighborhood service development (NSD)	0.114	0.00	0.972
Right-of-way (ROW)	6.537	657.24	<0.001
Commercial center (CC)	−0.665	1.09	0.296
Innovative (INN)	0.457	1.08	0.298
Planned unit development (PUD)	0.403	48.89	<0.001
Research district (RESD)	0.334	4.05	0.044

Table 5 Summary of parameter estimates of the models.

Variable	Coefficient	Std. error	Wald Chi-square	Significance value
Model to estimate the total number of crashes in a TAZ
(Intercept)	2.227	0.071	981.128	<0.001
Mixed use development (MUDEV)	9.035	2.778	10.573	0.001
Urban residential (UR)	8.484	2.905	8.531	<0.001
Business (BUS)	7.590	0.242	987.116	<0.001
Multi-family (MF)	3.979	0.441	81.371	<0.001
Office district (OD)	7.016	0.701	100.280	<0.001
Single-family (SF)	−0.415	0.077	29.208	<0.001
Institutional (INS)	0.262	0.124	4.487	0.03
Right-of-way (ROW)	6.404	0.293	478.605	<0.001
Planned unit development (PUD)	0.401	0.082	23.889	<0.001
Research district (RESD)	0.311	0.175	3.164	0.07
Dispersion parameter ˛= 1.24
Quasi-likelihood under independence model criterion (QIC) = 874.97
Corrected quasi-likelihood under independence model criterion (QICC) = 891.19
Model to estimate the number of injury crashes in a TAZ
(Intercept)	1.124	0.067	279.521	<0.001
Mixed use development (MUDEV)	7.700	2.703	8.114	<0.001
Urban residential (UR)	9.525	2.717	12.294	<0.001
Industrial (IND)	−0.149	0.068	4.826	0.028
Business (BUS)	7.456	0.264	796.941	<0.001
Multi-family (MF)	4.378	0.362	146.420	<0.001
Office district (OD)	6.068	0.636	91.107	<0.001
Single-family (SF)	−0.430	0.075	32.690	<0.001
Institutional (INS)	0.477	0.104	21.040	<0.001
Right-of-way (ROW)	7.168	0.311	530.793	<0.001
Research district (RESD)	0.568	0.110	26.442	<0.001
Dispersion parameter ˛= 0.48
Quasi-likelihood under independence model criterion (QIC) = 874.39
Corrected quasi-likelihood under independence model criterion (QICC) = 891.69
Model to estimate the number of PDO crashes in a TAZ
(Intercept)	1.844	0.075	606.170	<0.001
Mixed use development (MUDEV)	9.588	2.800	11.725	0.001
Urban residential (UR)	7.376	3.180	5.379	0.02
Business (BUS)	7.276	0.232	987.177	<0.001
Multi-family (MF)	3.651	0.472	59.866	<0.001
Office district (OD)	7.309	0.780	87.848	<0.001
Single-family (SF)	−0.426	0.079	28.931	<0.001
Right-of-way (ROW)	6.098	0.359	289.195	<0.001
Innovative (INN)	0.966	0.584	2.736	0.09
Planned unit development (PUD)	0.586	0.068	74.950	<0.001
Dispersion parameter ˛= 0.61
Quasi-likelihood under independence model criterion (QIC) = 875.03
Corrected quasi-likelihood under independence model criterion (QICC) = 888.91

Table 6 Validation of models – results summary.

Crash category	Sample size	Degree of freedom	Chi-square statistic	Critical Chi-square statistic (95%)	Reject null hypothesis
Total number of crashes	268	255	163	317	No
Number of injury crashes	268	256	304	318	No
PDO crashes	268	258	289	320	No

Table 7 Applicability of models using a hypothetical case.

| | Area of TAZ = 0.25 mi | | | | | | | | | |-----------------------|-----------------------|------------------|----|--------|--------|--|--|--|--| | | Case 1 | Case 2
Case 3 | | Case 4 | Case 5 | | | | | | % Area | | | | | | | | | | | MUDEV | 100 | 75 | 75 | 75 | 75 | | | | | | UR | 0 | 25 | 0 | 0 | 0 | | | | | | BUS | 0 | 0 | 25 | 0 | 0 | | | | | | OD | 0 | 0 | 0 | 25 | 0 | | | | | | SF | 0 | 0 | 0 | 0 | 25 | | | | | | Expected # of crashes | | | | | | | | | | | All | 89 | 86 | 81 | 78 | 49 | | | | | | Injury | 21 | 24 | 21 | 19 | 13 | | | | | | PDO | 69 | 61 | 60 | 60 | 37 | | | | | | Percent difference | | | | | | | | | | | All | | 3 | 9 | 12 | 45 | | | | | | Injury | | −12 | 2 | 10 | 40 | | | | | | PDO | | 13 | 13 | 13 | 47 | | | | |

Note: MUDEV, UR, BUS, OD and SF in the table are mixed use development, urban residential, business, office district and single-family land use areas, respectively.

of crashes in a TAZ from developed models, and ˛ is the negative binomial dispersion parameter.

Results obtained from validation of models are summarized in Table 6. The computedChi-square statistic was observedlower than the critical Chi-square statistic for all the three final models. Therefore, the null hypothesis was not rejected in all these cases. This indicates land use data considered in this research can be used to estimate the total number of crashes, number of injury crashes and PDO crashes at a 95% confidence level.

5.2. Applicability of the research

This subsection demonstrates how crash frequencies vary with the identified significant explanatory variables to understand the applicability of results obtained for long range transportation planning and land use planning decisions. An analysis was conducted by examining the variations in the estimated crash frequencies as a function of selected explanatory variables (area of land use characteristics) in the TAZ. A 0.25-mi2 TAZ was considered as a hypothetical case. In Case 1, the TAZ has mixed use development (MUDEV). In case of 2, 3, 4 and 5, 25% of TAZ area has urban residential (UR), business (BUS), office district (OD) and single-family (SF), respectively. The remaining 75% has mixed use development (MUDEV). All other characteristics were assumed to remain constant. Crashes were estimated using the models developed for the five different cases. Table 7 summarizes percent area of each selected land use and the estimated total number of crashes, number of injury crashes and number of PDO crashes for all the five cases considered in the analysis. The table also shows the percent difference in the number of crashes, number of injury crashes and number of PDO crashes when compared to Case 1.

Analysis indicate that a 0.25-mi2 TAZ (Case 2) with 25% urban residential (UR) area and 75% mixed use development (MUDEV) will have 12% more injury crashes than a TAZ (Case 1) with 100% mixed use development (MUDEV). However, the total number of crashes and number of PDO crashes will be lower in Case 2. On the other hand, a 0.25-mi2 TAZ (Case 3) with 25% business area (BUS) and 75% mixed use development (MUDEV) or a 0.25-mi2 TAZ (Case 4) with 25% office district (OD) and 75% mixed use development (MUDEV) will have at least 10% fewer crashes than a TAZ (Case 1) with 100% mixed use development (MUDEV). A 0.25-mi2 TAZ (Case 5) with 25% urban residential area (UR) and 75% mixed use development(MUDEV) willhave 40%fewer crashes (injury andPDO as well) than a TAZ (Case 1) with 100% mixed use development(MUDEV). In other words, a 25% reduction in mixed use development (MUDEV) area in a TAZ may result in 10–40% fewer crashes in the TAZ. While planning and developing office district (OD) instead business (BUS) may resultin very little to marginal benefits, planning and developing single-family (SF) residential area instead of office district (OD) or business (BUS) will result in 30% fewer crashes.

The above results indicate that a delicate balance between land use characteristics in a TAZ will lower expected crash occurrence and lead to improved safety on roads. The coefficients of each variable from models developed could be used to understand the hierarchical role of each land use variable on crashes. This information could be used in planning new developments, long range transportation planning and land use decisions. Where retrofitting or rezoning is not possible, more funds could be allocated to areas comprising land use variables with high coefficients to proactively deploy treatments and improve safety on roads. Developing a balanced system from land use perspective, indirectly, would also lower short-end trips, traffic congestion and make transportation network more sustainable.

6. Summary and conclusions

The development of traffic analysis zone (TAZ) level crash estimation models based on land use characteristics is presented in this paper. Crash data and land use data for 765 TAZs with a total of 9799 crashes (includes 20 fatal crashes, 3227 injury crashes and 6552 property damage only crashes) were selected for the analysis and development of models. The developed models to estimate the total number of crashes, the number of injury crashes, and the number of property damage only crashes were validated considering data for 268 randomly selected TAZs. The computed Chi-square statistic was observed lower than the critical Chi-square statistic at a 95% confidence level for all the models. This shows that estimates from the developed models are statistically meaningful and closer to real-world data.

Results obtained show that mixed use development area (with residential and compatible non-residential uses less than 10 acres to serve the residents of the planned community), urban residential area (nearer to the employment core), single-family residential area (with 3–8 dwelling units per acre), multi-family residential area (with 12–43 dwelling units per acre), business area (with large trade area, warehousing, wholesaling, etc.) and office district area (conducive to establishment and operation of offices, institutions and commercial activities not involving in direct sale) are strongly associated with crashes in a TAZ. These land uses are generally high activity generators.

Land use characteristics such as institutional area (with major cultural, educational, medical, governmental, religious, athletic and other institutions), industrial area (with manufacturing, processing, and assembling of parts, distribution centers, and transportation terminals) and research district area (with high research, development and technology manufacturing operations and professional employment) were observed to play a statistically significant role only in estimating the total number of crashes and the number of injury crashes in a TAZ. From the results obtained, it can be concluded that these land use characteristics in a TAZ are strongly associated with an increase in severe crashes.

The coefficient of all the land use characteristics excluding single-family residential area is positive in the final model for total number of crashes in a TAZ. An increase in the area of single-family residential land use tends to lower the total number of crashes in a TAZ.While the presence of single-family residential area and industrial area tends to lower the number of injury crashes in a TAZ, the presence of single-family residential area only tends to lower the number of property damage only crashes in a TAZ. It can, therefore, be concluded that presence of single-family residential area may have a neutralizing effect, possibly due to different behaviors adopted by drivers (such as cautious driving) or lower travel speed in these areas.

The coefficient for urban residential land use area and mixed use development land use area was observed to be the highest than when compared to any other land use characteristic area in case of all the models developed in this paper. This indicates that urban residential and mixed use development land use areas are strongly associated with the number of crashes in a TAZ. These land use areas are generally followed by business and office activity generators as areas that are strongly associated with higher number of crashes. TAZs with these mixed land use areas, which produce as well as attract trips of different modes of transportation throughout the day, need additional emphasis at planning and project implementation level to enhance traffic safety.

Overall, it can be concluded that strong associations could be established and crash estimation models developed based on land use characteristics to estimate the total number of crashes, the number of injury crashes and the number of property damage only crashes in a TAZ at a 95% confidence level. Maintaining a delicate balance between different land uses in a TAZ or area based on outcomes from this research would improve safety and maximize derived benefits. The developed methodology and results can be used to incorporate safety into long range transportation plans and land use decisions so as to minimize anticipated crashes in the future. The models developed using the methodology can also be used to examine the effect of changes in land use characteristics (new development or re-zoning) on safety.

Acknowledgments

The authors acknowledge and sincerely thank Charlie Jones, Steven Castongia, and Anna Gallup of the City of Charlotte Department of Transportation (CDOT) for their help with crash data, land use data, and demographics/socio-economic characteristics data.

Disclaimer: The contents of this paper reflect the views of the author(s) and not necessarily the views of the University or the City of Charlotte Department of Transportation (CDOT). The author(s) are responsible for the facts and the accuracy of the data presented herein.

References

Baass, K.G., 1981. Design of Zonal Systems for Aggregate Transportation Models. Transportation Research Record 807. National Academy Press, Washington, DC, as cited in Meyer, M.D., Miller, E.J., 2001. Urban Transportation Planning, second edition. McGraw-Hill Science Engineering, New York, NY.
Abdel-Aty, M.A., Radwan, A.E., 2000. Modeling traffic accident occurrence and involvement. Accident Analysis and Prevention Journal 32 (5), 633–642.
Abdel-Aty, M.A., Siddiqui, C., Darwichi,A., Huang, H., 2010. Macroscopic safety investigationatthe traffic analysis zone level.In:Road Safety on Four Continents: 15th International Conference, Abu Dhabi, United Arab Emirates, March 28–30.
Aguero-Valverde, J., Jovanis, P.P., 2006. Spatial analysis of fatal and injury crashes in Pennsylvania. Accident Analysis and Prevention Journal 38 (3), 618–625.
An, M., Casper, C.T., Wu, W., 2011. Using travel demand model and zonal safety planning modelfor safety benefit estimation in project evaluation. In: Transportation Research Board 90th Annual Meeting Compendium of Papers DVD, Washington, DC.
Caliendo, C., Guida, M., Parisi, A., 2007. A crash-prediction model for multilane roads. Accident Analysis and Prevention Journal 39 (4), 657–670.
Federal Highway Administration (FHWA), 2011. Reducing Highway Fatalities. Available from: http://safety.fhwa.dot.gov/ (accessed 04.05.11).
Goodchild, M.F., 1986 Spatial Autocorrelation. Catmog 47, Geo Books. Available from: http://qmrg.org.uk/files/2008/11/47-spatial-aurocorrelation.pdf (accessed 04.05.11).
Greibe, P., 2003. Accident prediction models for urban roads. Accident Analysis and Prevention Journal 35 (2), 173–185.
Huang, H., Abel-Aty, M., 2010. Multilevel data and Bayesian analysis in traffic safety. Accident Analysis and Prevention Journal 42 (6), 1556–1565.
Ivan, J.N., Wang, C., Bernardo, N.R., 2000. Explaining two-lane highway crash rates using land use and hourly exposure. Accident Analysis and Prevention Journal 32 (6), 787–795.
Kim, K., Yamashita, E., 2002. Motor vehicle crashes and land use: empirical analysis from Hawaii. Transportation Research Record 1794, 73–79.
Kim, K., Brunner, I.M., Yamashita, E.Y., 2006. Influence of land use, population, employment, and economic activity on accidents. Transportation Research Record 1953, 56–64.
Kim, K., Punt, P., Yamashita, E.Y., 2010. Measuring influences of demographic and land use variables in Honolulu, Hawaii. Transportation Research Record 2147, 9–17.
Ladron de Guevara, F., Washington, S.P., Oh, J., 2004. Forecasting crashes at the planning level: simultaneous negative binomial crash model applied in Tucson, Arizona. Transportation Research Record 1897, 191-100.
Levine, N., Kim, K.E., Nitz, L.H., 1995a. Spatial analysis of Honolulu motor vehicle crashes: I. Spatial patterns. Accident Analysis and Prevention Journal 27 (5), 663–674.
Levine, N., Kim, K.E., Nitz, L.H., 1995b. Spatial analysis of Honolulu motor vehicle crashes: II. Spatial patterns. Accident Analysis and Prevention Journal 27 (5), 675–685.
Ma, J., Kockelman, K.M., Damien, P., 2008. A multivariate Poisson-lognormal regression model for prediction of crash counts by severity, using Bayesian methods. Accident Analysis and Prevention Journal 40 (3), 964–975.
McCullagh, P., Nelder, J.A., 1989. Generalized Linear Models, second edition. Chapman & Hall, London, United Kingdom.
Meyer, M.D., Miller, E.J., 2001. Urban Transportation Planning, second edition. McGraw-Hill Science Engineering, New York, NY.
Naderan, A., Babaei, M., 2011. Assessment of statistical models for aggregate crash prediction. In: Transportation Research Board 90th Annual Meeting Compendium of Papers DVD, Washington, DC.
Naderan, A., Shahi, J., 2010. Aggregate crash prediction models: introducing crash generation concept. Accident Analysis and Prevention Journal 42 (1), 339–346.
National Highway Traffic Safety Administration (NHTSA), National Center for Statistics and Analysis, 2009. Traffic Safety Facts 2009. Available from: http://www-nrd.nhtsa.dot.gov/pubs/811402ee.pdf (accessed 04.05.11).
Noland, R.B., Quddus, M.A., 2003. A spatially disaggregate analysis of road casualties in England. Accident Analysis and Prevention Journal 36 (6), 973–984.
Noland, R.B., Oh, L., 2004. The effect of infrastructure and demographic change on traffic-related fatalities and crashes: a case study of Illinois county-level data. Accident Analysis and Prevention Journal 36 (4), 525–532.
Pei, X., Wong, S.C., Sze, N.N., 2011. A joint-probability approach to crash prediction models. Accident Analysis and Prevention Journal 43 (3), 1160–1166.
Pulugurtha, S.S., Nujjetty, A.P., 2012. Assessment of models to estimate crashes at intersection: with and without using traffic volume. In: Transportation Research Board 91st Annual Meeting Compendium of Papers DVD, Washington, DC.
Pulugurtha, S.S., Sambhara, V.R., 2011. Pedestrian crash estimation models for signalized intersections. Accident Analysis and Prevention Journal 43 (1), 439–446.
Quddus, M.A., 2008. Modelling area-wide count outcomes with spatial correlation and heterogeneity: an analysis of London crash data. Accident Analysis and Prevention Journal 40 (4), 1486–1497.
Siddiqui, C.,Abdel-Aty, M., Huang, H., 2012.Aggregate nonparametric safety analysis of traffic zones. Accident Analysis and Prevention Journal 45, 317–325.
Washington, S., van Schalwyk, I., Mitra, S., Meyer, M., Dumbaugh, E., Zoll, M., 2006. Incorporating Safety into Long-Range Transportation Planning. NCHRP Report 546, Available from: http://onlinepubs.trb.org/onlinepubs/ nchrp/nchrp rpt 546.pdfhttp://onlinepubs.trb.org/onlinepubs/nchrp/nchrp rpt 546.pdf (accessed 04.06.11).
Wier, M., Weintraub, J., Humphreys, E., Seto, E., Bhatia, R., 2009. An area-level model of vehicle-pedestrian injury collisions with implications for land use and transportation planning. Accident Analysis and Prevention Journal 41 (1), 137–145.
Wood, G.R., 2002. Generalized linear accident models and goodness of fit testing. Accident Analysis and Prevention Journal 34 (4), 417–427.
Wood, G.R., 2005. Confidence and prediction intervals for generalized linear accident models. Accident Analysis and Prevention Journal 37 (2), 267–273.

pulugurthatrafficanalysiszonelevel2013.md 81 KB 文件歷史 原始文件