category: literaturenote citekey: maneinfluencenetworktrafficsignal2018 title: "Influence of on-network, traffic, signal, demographic, and land use characteristics by area type on red light violation crashes" authors: "Mane, Ajinkya S.; Pulugurtha, Srinivas S." year: 2018 date: 2018-00-00 2018 doi: 10.1016/j.aap.2018.08.006 publication: "Accident Analysis & Prevention" url: "https://doi.org/10.1016/j.aap.2018.08.006" zotero_key: AGGVCW9I zotero_storage: CE7GGGE5 collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Mane, Ajinkya S."

status: converted

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Influence of on-network, traffic, signal, demographic, and land use characteristics by area type on red light violation crashes

Ajinkya S. Manea , Srinivas S. Pulugurthab,⁎

^a Student of Infrastructure & Environmental Systems (INES) Ph.D. Program, The University of North Carolina at Charlotte, 9201 University City Boulevard, Charlotte, NC 28223-0001, USA
^b Professor of Civil & Environmental Engineering Department, Director of IDEAS Center, The University of North Carolina at Charlotte, 9201 University City Boulevard, Charlotte, NC 28223-0001, USA

ARTICLE INFO

Keywords: Red light violation Crash Area type Demographic Land use Traffic

ABSTRACT

The focus of this research paper is on extraction of predictor variables pertaining to on-network, traffic, signal, demographic, and land use characteristics, by area type, and examining their influence on the number of red light violation crashes. Data for the city of Charlotte, North Carolina was extracted and used for analysis. Three different sets of signalized intersections were selected in the three different area types - Central Business District (CBD), urban, and suburban areas. Each set is comprised of sixty signalized intersections (total 180 signalized intersections). The number of red light violation crashes from January 2010 to December 2014, within the vicinity of each selected signalized intersection, was considered as the dependent variable to develop crash estimation models for each area type. The crash estimation models by area type were compared with the crash estimation model developed considering all the 180 signalized intersections together. Different predictor variables were found to be significant at a 95% confidence level in three different areas. Log-link model with Negative Binomial distribution was observed to best fit the data used in this research. Findings indicate that enforcement, either manually or using red light running cameras (RLCs), at signalized intersections with high traffic volume in the CBD area; at signalized intersections with high traffic volume, high all-red clearance time, near high density of horizontal mixed non-residential and open space/recreational type land uses in urban area; at signalized intersections with high traffic volume, speed limit on the major approach, the number of lanes on the minor approach, and all-red clearance time and areas surrounded with horizontal mixed non-residential and retail type land use in suburban areas, would lead to a reduction in the number of red light violation crashes.

1. Introduction

An estimated 165,000 people were injured and 800 people were killed, annually, in crashes due to red light running in the United States (Insurance Institute for Highway Safety (IIHS, 2007). Retting and Williams (1996) reported two red light runners per hour in, total, 234 h of data collection. Their study concluded that the red light runners are drivers under the age of thirty years, with poor driving history, less likely to use seat belts, and drove small and older cars. Retting et al. (1998) reported red light violations at the rate of 5.2 runners/hour and 1.3 runners/hour for two different sites. Porter and England (2000) stated that unbuckled and non-Caucasian drivers are more likely to run and violate red light.

Most of the past studies on red light violation have analyzed the safety effect of red light running cameras (RLCs). Maisey (1981), South et al. (1998), Kamyab et al. (2002), Garber et al. (2007), Tay and de Barros (2009), (2011), Pulugurtha and Otturu (2014) investigated red light running violations and/or the effectiveness of RLCs in reducing red light running violations or crashes. Retting and Kyrychenko (2002) found that injury crashes reduced by 29%, while right-angle crashes reduced by 32% after the implementation of RLCs. On the contrary, Maisey (1981), South et al. (1998) reported an increase in rear-end collisions after the implementation of RLCs. Weldegiorgis and Jha (2009) stated that factors such as driver behavior, approach speed, and intersection geometry contribute to red light violations.

Studies in the past concluded that younger drivers are prone to traffic violations while older drivers are prone to crashes at intersections (McGwin and Brown, 1999; Chapman et al., 2014). Similarly, Shell et al. (2015) found that teens taking driver education are less likely to be involved in crashes or traffic violations in the first two years of driving. Alver et al. (2014) studied the relationship between demographic characteristics, socio-economic characteristics, traffic rule

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

^⁎ Corresponding author.

Table 1 Summary of Past Research on Factors Attributed to Red Light Violation Behavior and Crashes.

Category	Study	Contributing Factors
Behavior	Van der Horst et al. (1986)	Yellow change time
	Baguley (1988)	High traffic volume
	Retting and Williams (1996)	Profiling of violators: young drivers, drivers of small and older cars, drivers with poorer driving record, and drivers who are less likely to use seat belts
	Kamyab et al. (2000)	High traffic volume
	Porter and England (2000)	Large intersections, and high traffic volume, time period (3 pm to 6 pm), safety belt use, and ethnicity
	Schattler et al. (2003)	All-red clearance and yellow change time
	Kulanthayan et al. (2007)	Day (weekday / weekend), camera enforcement, type of vehicle (four-wheel / two-wheel vehicles), cycle length, and type of traffic signal
	Zimmerman (2007)	Dilemma zone, additional time for trucks at a signalized intersection
	Gates and Noyce (2010)	Dilemma zone, vehicle type, time of the day and platooning
Crashes	Mohamedshah et al. (2000)	Number of lanes on both streets, left- and right-turn lanes, entering/crossing traffic flow, traffic control type (fully actuated, semi actuated, and pre-timed), speed limit (a potential variable but was not present in the dataset)
Behavior and crashes	Bonneson et al. (2002) Datta et al. (2000), Schattler and Datta (2004)	Traffic volume of the subject approach, traffic volume of the cross-street, and red light running rate All-red clearance and yellow change time

violations, and crashes among young drivers. Four different traffic violations, red light violation, seat-belt violation, speed limit violation, and driving under the influence of alcohol or drugs, were considered in their study. Their results indicate that crash rate increased to 38.3% for drivers who received at least one traffic citation.

Studies in the past have also reported factors contributing to red light violation behavior and red light violation crashes. Table 1 summarizes factors contributing to red light running behavior and red light violation crashes from selected past research studies.

Van der Horst et al. (1986) found that extending yellow change time from three to four seconds would decrease red light runners (car drivers) by 50%. Datta et al. (2000) found that introducing a well-designed all-red clearance time following the yellow change time would significantly reduce red light violations and right-angle crashes at urban signalized intersections. Schattler and Datta (2004) concluded that at urban intersections, all-red clearance time would decrease the probability of right-angled crashes even after red light violation. A similar result was observed in the Schattler et al. (2003) study. Zimmerman (2007) observed that if yellow change time was extended by 1.5 s, it would improve intersection safety by reducing the number of trucks by 47% in the dilemma zone. Gates and Noyce (2010) observed that tractor-trailer and single-unit trucks are more likely to be involved in red light running compared to passenger car. Also, the number of red light violations are observed to be higher during peak hours compared to off-peak hours.

1.1. Limitations of past research and need for this research

Most of the past studies have investigated the effectiveness of RLCs, while a few researchers focused on identifying on-network (the number of lanes, width of intersection, speed limit, etc.), traffic, and signal (phasing and timing plan) characteristics that influence red light violations and associated crashes. The spacing between intersections and width of intersections in Central Business District (CBD) area is comparatively lower than the spacing between intersections and width of intersections in urban and suburban area. Likewise, the speed limit, network density, traffic volume, and signal timings also differ by area type. While on-network, traffic, and signal characteristics were observed to have an influence on the number of red light violation crashes, the influence of these characteristics, by area type, on the number of red light violation crashes were not explored much in the past.

Different land use types such as office buildings, commercial centers, industrial areas, and recreational areas attract people from different parts of the city. High-rise office buildings are typically located in the CBD area but not in the suburban area. On the other hand, singlefamily residential areas prevail more in the suburban area. High-rise commercial centers and mixed land uses are typically located in the CBD area, while horizontal mixed use or commercial centers are typically located in urban and suburban areas. The demographic characteristics as well as the number of trips generated and attracted by each land use type vary by area type. Therefore, traffic volume and congestion within the vicinity of each land use differ by area type. As stated in Kamyab et al. (2000), an increase in traffic volume and congestion influence drivers' aggressive behavior and hence, the number of traffic violations. In other words, the demographic and land use characteristics vary by area type and could also influence the number of red light violation crashes.

Literature review indicates that majority of past studies on red light violation were conducted considering urban signalized intersections. As on-network, traffic, signal, demographic, and land use characteristics not only vary by area type but influence the number of red light violation crashes, examining the relationship between the number of red light violation crashes and predictor variables such as on-network, traffic, signal, demographic, and land use characteristics, by area type, helps proactively identify where they could potentially happen for enforcement or implementation of remedial measures (implementation of manual or automated enforcement; signal timing changes; etc.). The findings from such a research are vital to identify, prioritize, and allocate limited available transportation funds, either manually or using RLCs.

2. Methodology

The city of Charlotte, North Carolina was considered as the study area to gather data and conduct analysis. The methodology includes the following steps.

1 Identify study signalized intersections by area type
2 Identify predictor variables contributing to red light violation crashes
3 Generate buffers and conduct geospatial analysis
- a Extract the number of red light violation crashes
- b Extract on-network, traffic, and signal characteristics
- c Extract demographic characteristics
- d Extract land use characteristics
4 Check for multicollinearity between predictor variables
5 Develop crash estimation models by area type
6 Validate the models

Each step is explained next in detail.

2.1. Identify study signalized intersections by area type

In this research, the city of Charlotte was divided into three different areas: CBD area, urban area, and suburban area. Freeways (I-277 and I-77), which surround the Charlotte downtown/uptown area were considered as boundary lines for the CBD area. The CBD area primarily comprises of high-rise office buildings, apartments/condominiums, businesses, and related commercial activity. Intersections are closely spaced, road widths and speed limits are relatively lower, and network density is higher in the CBD area. The outer freeway (I-485; beltway) was considered as the extreme boundary line for the suburban area. The suburban area primarily comprises of residential areas, institutions/ schools, commercial centers, and recreational areas. The area between the CBD area and the suburban area was considered as the urban area. The density of land uses in the urban area is higher than in the suburban area, but lower than in the CBD area. Likewise, spacing between intersections and number of lane miles is higher in the urban area than in the CBD area, but lower than in the suburban area. Road widths and speed limits are comparable in both, urban and suburban areas.

Sixty signalized intersections in each area type (total 180 signalized intersections) were randomly selected to conduct analysis and develop crash estimation models. These signalized intersections were ensured to be spatially distributed in the respective areas (Fig. 1). In addition, crash estimation models were developed by considering all the selected 180 signalized intersections.

2.2. Identify predictor variables contributing to red light violation crashes

As stated previously, the on-network, traffic, signal, demographic, and land use characteristics vary by area type. On-network characteristics are those that exist on the road network. They include speed limit, the number of lanes, the number of approaches, presence of median, road width, etc. These factors may influence the number of red light violation crashes. In this research, the speed limit and the number of lanes were considered as the predictor variables.

Traffic volume, yellow change time, and all-red clearance time are traffic and signal characteristics that may influence the number of red light violation crashes. These characteristics were also considered as predictor variables.

The traffic volume and the number of crashes (in general) could increase with an increase in population and the number of households. Moreover, different land use patterns could attract people from

Fig. 1. Selected signalized intersections in the City of Charlotte, North Carolina.

Table 2 Predictor Variables.

On-network, traffic and signal characteristics
Traffic Volume (TV)	Average traffic volume per year, including all the approaches
Speed Limit	Speed limit was categorized as the speed limit on the major approach (SLMA) and the speed Limit on the minor approach (SLMI)
# of Lanes	# of lanes was categorized as the number of lanes on the major approach (NLMA) and the number of lanes on the minor approach (NLMI)
# of Approaches (NAPR)	Total number of approaches at the signalized intersection
Average Yellow Change Time (YCT)	Time required for driver to comfortably decelerate to stop or to clear the signalized intersection.
All-Red Clearance Time (AR)	Time required to clear the vehicles in the signalized intersection.
Land use characteristics (square feet)
Civic/Institutional (Civic)	Major cultural, educational, medical, government, athletic institutions
Horizontal Mixed Use Residential & Non-Residential (HMRNR)	Typical block consists of multiple buildings. Each building is used as single-use buildings. Residential and compatible non residential uses are placed in the separate buildings.
Horizontal Mixed Use Non-Residential (HMNR) Industrial (IND)	Typical block consists of single-use buildings, only for non-residential uses such as commercial, institutional, etc. Areas with manufacturing, processing and assembling of parts, distribution centers and transportation terminals; specialized industrial operations
Large lots of residential (LLoR)	Large area utilized for residential
Multi-family (MF)	Areas with a variety of housing types; 12–43 dwelling units per acre
Office (OFF)	Areas conducive to establishment and operation of offices, institutions, and commercial activities not involving in sale
Open Space/Recreational (OS)	Area utilized for recreational purposes such as parks
Parking (Park)	Area utilized for parking purpose
Retail (Retail)	Area utilized for retail shops
Single-family – Attached (SFA)	Area with primarily single-family housing where houses have one common wall with the adjacent house
Single-family – Detached (SFD)	Independent residential area for single-family. Houses are not connected.
Transportation (Trans)	Area utilized for transportation purpose such as air transport, water transport, and rail transport
Utility (Utility)	Areas utilized to provide the services for everyone in a community

Vertical Mixed Use (VMU) Combines residential and compatible non-residential uses within the same building. Typically, non-residential uses are provided on the lower floors and residential uses on the upper floors

Warehouse/Distribution (WH) Area utilized for manufacturing and wholesale trade/distribution process

different parts of the city. As these demographic and land use characteristics within the vicinity of a signalized intersections may influence the number of red light violation crashes, they were also considered as predictor variables.

Total Household Units (HU) Total number of household units

Population (PL) Total population

The selected predictor variables for analysis and model development in this research are summarized in Table 2.

2.3. Generate buffers and conduct geospatial analysis

Demographic characteristics

Buffers were generated around each signalized intersection using "buffer" feature in ArcGIS. Further, "intersect" feature in ArcGIS was used to extract data by overlaying the generated buffers on individual geospatial layers pertaining to crashes, on-network, traffic, demographic, and land use characteristics. The width of the buffer to capture data is different for the number of red light violation crashes, on-network, traffic, demographic, and land use characteristics.

2.3.1. Extract the number of red light violation crashes

Crash data from January 2010 to December 2014 was used in this research. Crashes related to red light violation were extracted from the main crash database based on the contributing factor (red light violation). The selected crash data was then spatially overlaid on the street network. A buffer width of 100 feet was generated around each selected signalized intersection to extract the total number of crashes (related to red light violation) that occurred on all the approaches of each selected signalized intersection. The 100-feet buffer width was selected considering typical width of the signalized intersection (60–80 feet) in the study area. Considering 100-feet buffer width also minimizes concerns related to overlapping of buffers. The number of red light violation crashes at each intersection (within 100 feet) was considered as the dependent variable.

2.3.2. Extract on-network, traffic, and signal characteristics

The data pertaining to on-network (speed limit and the number of lanes), traffic, and signal characteristics was obtained from the city of Charlotte Department of Transportation (CDoT).

2.3.2.1. Speed limit. Typically, a signalized intersection has one major approach and one minor approach. Therefore, for each signalized intersection, speed limit on the major approach and the minor approach were considered separately in the analysis.

2.3.2.2. Number of lanes. Different approaches have different number of lanes. The number of lanes on the major approach and minor the approach were considered separately in the analysis.

2.3.2.3. Traffic volume. The number of crashes at a signalized intersection are related to the traffic volume (exposure). In this research, similar to crash data (five years), average traffic volume during the five-year period was considered for analysis. Here, the average AADT for five years from all the approaches was considered to represent traffic volume for an intersection.

2.3.2.4. Yellow change time. Yellow change time depends on the width of the signalized intersection and approach speed limit. It was computed for each approach and used to compute the average yellow change time for the signalized intersection. Eq. (1) was used to compute the yellow change time (North Carolina Department of Transportation (NCDOT, 2004). The minimum yellow change time is 3.0 s.

Yellow Change Time = $$t + \frac{V}{2a + 64.4 \times g}$$ (1)

where, t = perception reaction time, typically 1.5 s, V = design speed, in feet per seconds, a = deceleration rate, typically 11.2 feet /seconds² and, g = grade.

Fig. 2. Spatial overlay of census block and land use data on a 0.5-mile buffer around a signalized intersection.

2.3.2.5. All-red clearance time. Each signal is equipped with an all-red clearance time. It is provided to clear the vehicles in the signalized intersection. As per the traffic signal design manual of the North Carolina Department of Transportation (NCDOT), all-red clearance time was computed for each signalized intersection using Eq. (2). The minimum all-red clearance time is 1.0 s.

All-red Clearance Time = $$\frac{W}{V}$$ (2)

where, W = width of the intersection, in feet, and, V = design Speed, in feet per seconds.

2.3.3. Extract demographic characteristics

Demographic characteristics within the vicinity of each selected signalized intersection were extracted by generating 0.5-mile buffers and then overlaying census data (Fig. 2(a)).

It was assumed that population and the number of households are uniformly distributed in each census block. Therefore, the proportion of census block area that falls in a buffer was considered to estimate the aforementioned predictor variables (Eq. (3)).

$$P{i} = \sum{j} \frac{A{j,i}}{A{j}} P_{j}$$ (3)

where, Pi = total population or the number of households in buffer "i", Aj,i = actual area of census block "j" in the buffer "i", Aj = area of census block "j", and, Pj= total population or number of households in census block "j".

2.3.4. Extract land use characteristics

Land use characteristics were extracted within 0.5-mile buffer generated around each selected signalized intersection, using 'intersect' feature in GIS. Fig. 2(b) illustrates overlaying of existing land use layer on a 0.5-mile buffer around a signalized intersection. All the land use areas were represented in square feet.

2.4. Check for multicollinearity between predictor variables

The selected predictor variables may be correlated to each other. To identify the influence of predictor variables and accurately estimate the dependent variable (the number of red light violation crashes), only one predictor variable is selected between the two correlated predictor variables, at a time for model development. In addition to a strong correlation, the included and excluded predictor variables pair must have a meaningful relationship (example, traffic volume and speed limit or the number of lanes).

The multicollinearity between predictor variables and collinearity between dependent and predictor variables was checked by computing Pearson correlation coefficients for each dataset. The range of Pearson correlation coefficient is from −1 to +1. The cut-off value of Pearson correlation coefficient was set up as −0.3 and +0.3 (i.e., Pearson correlation coefficient value less than or equal to −0.3 or greater than or equal to 0.3 are assumed to be correlated to each other) to avoid multicollinearity between the predictor variables (at least at a 95% confidence level).

Either of the two correlated predictor variables could be selected for model development. This leads to multiple models with combinations of predictor variables.

2.5. Develop crash estimation models by area type

Crash estimation models by area type (CBD, urban, and suburban area) were developed to investigate the factors contributing to the number of red light violation crashes by area type.

The model selection depends on linear or non-linear correlation between the dependent and predictor variables. Generalized linear model with Negative Binomial distribution, log-normal distribution, Gamma distribution, and Poisson distribution are generally used to develop crash estimation models. The best model was selected based on the Quasi Likelihood under Independence Model Criterion (QIC) and Corrected Quasi Likelihood under Independence Model Criterion (QICC). The lower the QIC and QICC, the better is the goodness of fit. Moreover, the difference between QIC and QICC should be generally low for a good model. In addition, the predictor variables with p-value less than 0.05 are considered as significant at a 95% confidence level.

2.6. Validate the models

Each crash estimation model was validated using Mean Absolute Deviation (MAD). MAD was computed using Eq. (4). The values of MAD close to zero resembles the ability of models to predict the number of crashes close to the actual number of crashes (Oh et al., 2003).

$$MAD = \frac{1}{n} \sum{i=1}^{n} |\hat{c}{j} - c_{j}|$$ (4)

where, n = the number of observations, ĉj= the predicted number of red light violation crashes, and, cj= the observed number of red light violation crashes.

Table 3 Number of Red Light Violation Crashes.

Year	CBD		Urban		Suburban
	# of Crashes	# of Injuries	# of Crashes	# of Injuries	# of Crashes	# of Injuries
2010	184	215	221	259	226	267
2011	150	173	200	228	206	235
2012	153	149	194	206	192	198
2013	192	213	270	296	271	298
2014	180	222	231	283	233	282
Total	859	972	1,116	1,272	1,128	1,280

3. Results

The results obtained from analysis and model development are presented in this section.

3.1. Descriptive analysis

The total number of red light violation crashes and the associated number of injuries over the five-year period (January 2010 to December 2014) are presented in Table 3. The maximum number of crashes was observed in the year 2013, in all the area types. Moreover, the total number of crashes from year 2010 to year 2014 is ∼30% more in urban and suburban areas when compared to the CBD area. Descriptive statistics pertaining to all the variables considered in this research are presented in Table 4.

3.2. Examine the correlation between predictor variables

A correlation matrix was developed for each area type and for all the signalized intersections irrespective of area type. As an example, Table 5 summarizes Pearson correlation coefficient matrix computed using all the signalized intersections data. It was observed that some of the predictor variables were highly correlated to each other. For example, traffic volume, speed limit on the major approach, and speed limit on the minor approach are highly correlated to each other. As stated previously, to minimize the effect of multicollinearity, only one of these three predictor variables were considered in a model. As any of these three predictor variables could be considered, possible combinations of predictor variables were identified for the development of models. The possible combinations of selected predictor variables for each area type and complete dataset are summarized in Table 6. As an example, traffic volume was considered in Model 1, speed limit on the minor approach was considered in Model 2, and speed limit on the major approach was considered in Model 3, when developing models using all the signalized intersections data.

3.3. Crash estimation models

The number of red light violation crashes was considered as the dependent variable. A total of twelve crash estimation models were developed in this research. In all the developed models, Negative Binomial distribution-based log-link model was found to be a better fit than log-normal, Gamma or Poisson distribution-based log-link models.

When excluding correlated predictor variables, caution must be exercised to ensure that the estimated coefficients represent the combined effects of both the included and excluded predictor variables (Tay, 2017). In addition to examining strong and meaningful relationships between included and excluded predictor variables, the computed standard errors, QIC, and QICC for the twelve models were compared with those obtained for models considering all the predictor variables (irrespective of correlations). The computed standard errors, QIC, and QICC are lower for the twelve models than those obtained for models considering all the predictor variables.

Predictor variables with p-value greater than 0.05 were removed, one after another, when developing models. All the predictor variables with p-value less than 0.05 were considered as significant (at a 95% confidence interval) in the final model for each dataset. The general form of the negative binomial log-link distribution based model is represented as follows.

$$Y = EXP^{(\beta + \alpha_1 \times X_1 + \alpha_2 \times X_2 + \dots + \alpha_n \times X_n)}$$ (5)

where,

Y = predicted number of red light violation crashes,

α1, α2, …, αⁿ = coefficients of predictor variables, and,

X1, X2, …, Xn = predictor variables considered for model development.

3.3.1. CBD area

Three crash estimation models were developed for the CBD area. The results obtained are summarized in Table 7. The results obtained indicate that the number of red light violation crashes increase as the traffic volume increases. However, areas with large lots of residential, retail, utility, and industrial type land use within the vicinity of a signalized intersection have a negative effect on the number of red light violation crashes.

3.3.2. Urban area

The results obtained from the three crash estimation models developed for urban area are summarized in Table 8. The results obtained from Model 1 indicate that the number of red light violation crashes increase as the traffic volume and all-red clearance time increase. Likewise, both, Model 2 and Model 3 indicate that the number of red light violation crashes increase as the areas with horizontal mixed nonresidential and open space/recreational type land use increase within the vicinity of a signalized intersection. However, areas with large lots of residential, warehouse/distribution, and industrial type land use have a negative effect on the number of red light violation crashes.

3.3.3. Suburban area

The models developed for suburban area are summarized in Table 9. The results obtained indicate that the number of red light violation crashes increase as the traffic volume, speed limit on the major approach, the number of lanes on the minor approach, and all-red clearance time increase within the vicinity of a signalized intersection. Further, the number of red light violation crashes increase as areas with horizontal mixed non-residential and retail type land use increase within the vicinity of a signalized intersection. However, areas with single-family attached, single-family detached, and transportation type land use have a negative effect on the number of red light violation crashes.

3.3.4. All signalized intersections

Crash estimation models were developed by considering all the 180 randomly selected signalized intersections. The results obtained are presented in Table 10. The results obtained indicate that the number of red light violation crashes increase as the traffic volume, speed limit on the minor approach, the number of lanes on the minor approach, and areas with horizontal mixed non-residential and retail type land use increase within the vicinity of a signalized intersection. However, areas with industrial, large lots of residential, single-family attached, and transportation type land use have a negative effect on the number of red light violation crashes.

3.4. Model validation

Each area type model was validated by randomly selecting 10 signalized intersections in the respective areas. In addition, 30 signalized intersection were randomly selected for validation of complete dataset

Table 4 Descriptive Statistics.

Variables	N	CBD				Urban				Suburban
		Min.	Max.	Mean	SD	Min.	Max.	Mean	SD	Min.	Max.	Mean	SD
# Crashes	60	2	30	7.65	6.09	2	26	6.98	4.98	2	29	6.32	5.19
Traffic Volume	60	11,700	64,000	28,811.67	13,395.27	16,500	153,000	50,995.83	24,295.30	18,700	174,000	71,093.33	41,939.61
Speed Limit on the Major Approach	60	25	55	33.50	5.39	35	45	39.33	4.91	30	55	44.17	5.06
Speed Limit on the Minor Approach	60	25	35	29.08	3.74	25	45	33.33	4.66	25	45	35.83	5.46
# of Lanes on the Major Approach	60	2	7	4.12	1.03	2	6	4.53	0.79	2	7	4.58	1.27
# of Lanes on the Minor Approach	60	1	7	2.90	1.08	1	5	2.75	1.07	1	6	2.67	1.17
Yellow Change Time	60	3.00	3.63	3.15	0.17	3.00	3.95	3.37	0.26	3.13	4.28	3.55	0.27
All-Red Clearance Time	60	1.00	2.66	1.56	0.35	1.19	2.27	1.65	0.27	1.19	2.69	1.87	0.32
Civic / Institutional	60	254,398.00	5,878,422.83	3,342,770.44	1,789,053.70	0.00	6,341,518.07	2,929,702.45	1,855,887.42	0.00	6,394,974.05	2,120,006.58	1,755,776.83
Horizontal Mixed Use Residential & Non-Residential	60	0.00	118,103.06	40,852.51	51,980.63	0.00	223,278.53	21,057.36	48,593.25	0.00	27,866.67	2,956.20	8,341.75
Horizontal Mixed Use Non-Residential	60	0.00	1,565,545.52	397,896.75	365,021.90	0.00	2,143,985.21	414,594.11	502,554.75	0.00	2,231,985.01	363,938.64	502,167.02
Industrial	60	0.00	5,664,335.24	544,377.90	1,063,859.30	0.00	5,804,397.68	616,315.63	1,180,094.55	0.00	4,351,364.30	533,015.15	994,555.81
Large Lots of Residential	60	0.00	960,354.56	53,833.89	172,563.62	0.00	3,659,272.62	200,428.68	581,828.08	0.00	3,297,486.10	367,681.23	735,645.05
Multi-family	60	0.00	11,142,309.35	1,985,797.82	1,608,239.40	8,957.11	7,268,312.44	1,723,486.34	1,436,299.20	0.00	7,486,753.76	2,196,695.22	2,043,201.23
Office	60	0.00	3,030,190.50	1,126,151.82	740,476.84	0.00	3,767,315.20	1,230,128.82	930,743.24	0.00	7,757,025.70	1,192,769.77	1,337,266.67
Open Space / Recreation	60	912.79	5,282,417.54	829,294.68	831,171.13	0.00	4,518,764.78	930,169.29	872,017.18	0.00	4,243,499.02	998,271.91	889,121.75
Parking	60	0.00	2,841,049.05	1,026,864.72	958,257.33	0.00	2,638,204.48	510,244.72	807,845.54	0.00	2,003,667.72	202,007.54	415,918.27
Retail	60	0.00	5,380,914.18	933,930.46	745,264.06	0.00	8,771,509.35	1,806,618.13	1,508,168.42	0.00	7,408,903.41	2,179,961.61	1,801,904.00
Single-family Attached	60	0.00	1,388,892.85	199,693.95	281,364.27	0.00	1,092,918.80	244,121.01	280,718.71	0.00	2,115,459.34	197,877.16	392,879.57
Single-family Detached	60	0.00	10,542,842.52	2,956,778.84	3,395,938.48	3,641.96	10,444,370.27	3,267,718.65	2,713,917.96	0.00	14,069,605.22	3,957,467.24	3,838,236.01
Transportation	60	0.00	1,730,455.37	496,043.77	450,324.36	0.00	2,942,897.52	352,494.20	538,863.30	0.00	6,199,481.80	250,880.04	834,313.51
Utility	60	0.00	940,531.65	125,806.28	154,769.14	0.00	567,933.68	89,192.57	117,216.42	0.00	624,869.12	77,224.29	123,883.99
Vertical Mixed Use	60	0.00	2,197,209.47	866,151.00	819,917.81	0.00	2,032,848.74	441,082.59	632,955.28	0.00	1,731,480.37	223,652.94	370,779.75
Warehouse Distribution	60	0.00	4,592,676.78	594,300.38	883,998.59	0.00	5,685,340.07	743,654.24	1,164,114.17	0.00	6,129,426.78	700,656.13	1,257,063.39
Population	60	472.15	19,336.44	10,915.47	6,114.53	547.99	63,008.94	10,501.72	9,803.55	954.82	65,561.64	12,101.61	13,863.15
# Households	60	189.46	13,279.96	6,473.03	4,195.23	196.25	19,827.32	5,265.20	4,725.83	394.36	30,323.35	5,444.06	6,185.06

(continued on next page)

Table 5 Correlation Matrix - All Signalized Intersections.

IND	−0.226 −0.257 −0.279 −0.217 −0.238 −0.175 −0.151 −0.003 −0.083 0.361 0.642* 0.183 0.080 0.008 1	PL
HMNR	−0.211 −.0217 −0.230 −0.072 −0.016 −0.026 −0.106 0.226 0.226 0.232 0.184* 0.140 0.082 0.031 0.041 1	WH
HMRNR	−0.204 −0.222 −0.383 −0.148 −0.180 0.512 0.293 0.534 0.244** 0.004 0.018 0.059 0.057 0.053 0.027 0.104 1	VMU
Civic	−0.369 −0.274 −0.215 −0.266 −0.196 −0.474 −0.460 −0.004 −0.111 −0.101 0.296 0.240 0.686 0.211 0.675 0.308 0.406 1	Utility
AR	−0.348 −0.400 −0.404 −0.216 −0.395 −0.074 −0.071 −0.126 −0.094 −0.109 −0.056 0.227 0.288 0.356 0.261* 0.187 0.028 0.091 1	Trans
Y.C.T.	−0.528 −0.285 −0.616 −0.601 −0.279 −0.165 −0.165 −0.129 −0.102 −0.079 −0.076 0.263 0.247 0.398 0.192 0.457 0.225* 0.162 0.066 1	SFD 1
	−0.046 −0.009 0−.043 −0.083 −0.056 −0.050 −0.085 −0.032 −0.114 −0.096 −0.111	0.439** SFA 1
NLMI	0.063 0.079 0.059 0.133 0.008 0.057 0.015 0.018 0.002 0.009 1	−0.157* −0.084 Retail 1
NLMA	−0.171* −0.086 −0.055 −0.059 −0.035 −0.077 −0.025 −0.042 −0.114 −0.108 −0.007 0.284 0.214 0.126 0.064 0.113 0.077 0.077 0.066 0.045 0.022 0.003 1	−0.327 −0.306 −0.605** Park 1
SLMI	−0.202 −0.198 −0.198 −0.199 −0.067 −0.045 −0.123 −0.029 −0.049 −0.049 −0.133 0.467* 0.152 0.153 0.159 0.036 0.090 0.018 0.122 0.043 0.022 0.007 0.091 1	−0.234 −0.060 0.281 0.014
SLMA	−0.210 −0.235 −0.225 −0.088 −0.077 −0.143 −0.050 −0.031 −0.041 −0.081 0.586 0.500 0.201 0.230* 0.183 0.023 0.136 0.082 0.012 0.020 0.008 0.006 0.034 0.026 1	OS 1 −0.204 −0.463 −0.134 0.266 0.205 OFF
TV	−0.237 −0.335 −0.199 −0.323 −0.188* −.196 −0.108 −0.062 −0.087 −0.124 −0.047 0.526 0.665 0.247 0.263 0.621 0.345 0.269 0.352* 0.149 0.004 0.015 0.045 0.104 0.007 1	1 −0.134 −0.058 −0.119 −0.131 0.234** 0.042 MF 1
# Crashes	−0.066 −0.080 −0.007 −0.138 −0.105 −0.105 −0.033 −0.054 −0.100 −0.004 −0.121 0.253 0.174 0.148 0.165* .195 0.093 0.007 0.029 0.145 0.074 0.013 0.130 0.020 0.026 0.030 1	−0.254* −0.166 −0.152 −0.013 0.171 0.020 0.061 LLoR 1
Correlation Pearson	# Crashes HMRNR HMNR NLMA Utility SLMA NAPR Y.C.T. Retail NLMI Trans SLMI Civic LLoR VMU Park OFF IND SFA SFD WH HU AR MF TV OS PL	Correlation # Crashes HMRNR Pearson HMNR NLMA SLMA NAPR Y.C.T. Retail NLMI SLMI Civic LLoR Park OFF IND SFA SFD AR MF TV OS

| PL | | | | | | 1 | 0.923** | | |---------|-------------|----------|----------|----------|----------|----------|----------|--| | WH | | | | | 1 | −0.245** | −0.261** | | | VMU | | | | 1 | −0.261** | 0.313** | 0.574** | | | Utility | | | 1 | 0.120 | −0.048 | −0.043 | −0.004 | | | Trans | | 1 | 0.085 | 0.142 | 0.337** | −0.138 | −0.060 | | | SFD | | −0.242** | −0.232** | −.614** | −0.045 | −0.414** | −0.515** | | | SFA | | −0.109 | −0.138 | −0.295** | 0.055 | −0.115 | −0.157* | | | Retail | | −0.117 | −0.184* | −0.294** | 0.064 | 0.366** | 0.169* | | | Park | | 0.178* | 0.138 | 0.960** | −0.260** | 0.241** | 0.515** | | | OS | | −0.302** | −0.155* | −0.217** | −0.262** | −0.220** | −0.272** | | | OFF | | −0.093 | 0.040 | 0.316** | −0.171* | 0.600** | 0.625** | | | MF | | −0.151* | −0.248** | −0.107 | −0.148* | −0.034 | −0.054 | | | LLoR | | −.0073 | 0.068 | −0.272** | 0.076 | −0.227** | −0.275** | | | Pearson | Correlation | Trans | Utility | VMU | WH | PL | HU | |

TV: Traffic Volume, SLMA: Speed Limit on Major Approach, SLMI: Speed Limit on Minor Approach, NLMA: # of Lanes on Major Approach, NLMI: # of Lanes on Minor Approach, YCT: Average Yellow Change Time, AR: All-Red Clearance Time, Civic: Civic/Institutional, HMRNR: Horizontal Mixed Use Residential & Non-Residential, HMNR: Horizontal Mixed Use Non-Residential, IND: Industrial, LLoR: Large Lots of Residential, MF: Multi-family, OFF: Office, OS: Open Space/Recreational, Park: Parking, SFA: Single-family – Attached, SFD: Single-family – Detached, Trans: Transportation, VMU: Vertical Mixed Use, WH: Warehouse/Distribution, PL:

Population, HU: Total Household Units. * Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).

(180 signalized intersection) models. These signalized intersections were not considered for model development. The MAD values closer to zero indicate the best crash estimation model. The results obtained from validation are summarized in Table 11. The computed MAD values of all the models are close to zero. Therefore, all the models are acceptable.

3.5. Effect of predictor variables

Positive and negative coe fficients of predictor variables in nonlinear models indicate the role of each predictor variable on red light violation crashes. The results obtained indicate that, in most of the cases, the sign of predictor variable was found to be consistent with respect to each area type. For example, in CBD area, the sign of tra ffi c volume coe fficient was positive in all the developed models (Table 7). Likewise, in urban area, the sign of horizontal mixed non-residential coe fficient was same, in both models 2 and 3 (Table 8). The e ffect of each signi ficant predictor variable from the developed models are summarized in Table 12

The results obtained indicate that in all the area types, tra ffic volume was found to be significant and has a positive effect on the number of red light violation crashes. In urban and suburban areas, allred clearance time and areas with horizontal mixed non-residential type land use have a positive e ffect on the number of red light violation crashes. Similarly, in suburban area, speed limit on the major approach, the number of lanes on the minor approach, and areas with horizontal mixed non-residential and retail type land uses have a positive e ffect on the number of red light violation crashes. Likewise, areas with horizontal mixed non-residential, open space/recreational, and retail type land uses have a positive e ffect on red light violation crashes when complete dataset was considered for model development.

The positive e ffect indicates that these factors contribute more to the number of red light violation crashes than predictor variables with negative coe fficients (areas with industrial, large lots of residential, retail, and utility type land uses in CBD area; areas with industrial, large lots of residential, and warehouse/distribution type land uses in urban area; areas with single-family attached, single-family detached, and transportation type land uses in suburban area; areas with industrial, large lots of residential, single-family attached, and transportation type land uses in the complete dataset).

From Table 12, various predictor variables were observed to in fluence the number of red light violation crashes. Datta et al. (2000) stated that a well-designed all-red clearance time would reduce red light violation and right-angle crashes. Similar results were observed from this research. The results obtained indicate that all-red clearance time in fluences the number of red light violation crashes at signalized intersections in the urban and suburban areas; could be associated to the width of the signalized intersections in these areas. Therefore, optimized all-red clearance time should be provided for signalized intersections in urban and suburban areas. However, its influence at signalized intersections in the CBD area is statistically insignificant.

Mohamedshah et al. (2000) stated that tra ffi c volume, speed limit, and the number of lanes on the major and minor approach contribute to the number of red light violation crashes. Similar results were observed from this research. Tra ffic volume (exposure) was observed to have an in fluence on the number of red light violation crashes at signalized intersections in all the three area types. However, tra ffic volume is relatively higher at signalized intersections in the suburban area when compared to CBD and urban areas, indicating possible in fluence of other predictor variables. The findings indicate that speed limit on the major approach and the number of lanes on the minor approach in fluence the number of red light violation crashes at only signalized intersections in the suburban area (in addition to tra ffic volume). This could be due to higher speed limits and permitted turning movements with high wait time from the minor approaches in suburban areas. Typically, the variation between the speed limit on major approach and

Table 6 Selected Predictor Variables to Develop Models by Area Type.

Predictor Variables	CBD Area			Urban Area			Suburban Area			All
	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3
Traffic Volume (TV)	✓	✓	✓	✓	✓	✓	✓			✓
Speed Limit on the Major Approach (SLMA)	✓							✓	✓			✓
Speed Limit on the Minor Approach (SLMI)											✓
# of Lanes on the Major Approach (NLMA)		✓		✓	✓		✓				✓	✓
# of Lanes on the Minor Approach (NLMI)									✓	✓		✓
Average Yellow Change Time (YCT)
All-Red Clearance Time (AR)				✓				✓
Civic/Institutional (Civic)				✓			✓		✓
Horizontal Mixed Use Residential & Non			✓		✓	✓					✓	✓
Residential (HMRNR)
Horizontal Mixed Use Non-Residential (HMNR)			✓	✓	✓	✓	✓			✓	✓	✓
Industrial (IND)			✓		✓	✓						✓
Large lots of residential (LLoR)	✓			✓	✓					✓	✓	✓
Multi-family (MF)	✓	✓			✓				✓		✓	✓
Office (OFF)			✓		✓	✓		✓			✓	✓
Open Space/Recreational (OS)							✓			✓		✓
Parking (Park)
Retail (Retail)		✓			✓	✓			✓			✓
Single-family – Attached (SFA)	✓			✓	✓	✓	✓		✓		✓
Single-family – Detached (SFD)								✓		✓
Transportation (Trans)							✓	✓	✓		✓
Utility (Utility)		✓							✓		✓
Vertical Mixed Use (VMU)		✓
Warehouse/Distribution (WH)	✓			✓						✓
Population (PL)
# Households (HU)

Table 7 Crash Estimation Models – CBD Area.

Parameters	Model 1		Model 2		Model 3
	B (S.E.)	p-value	B (S.E.)	p-value	B (S.E.)	p-value
Intercept	1.315 (0.19)	< 0.05	1.581 (0.20)	< 0.05	1.444 (0.20)	< 0.05
Traffic Volume	2.473E-05 (< 0.01)	< 0.05	2.282E-05 (< 0.01)	< 0.05	2.157E-05 (< 0.01)	< 0.05
Industrial					−1.709E-07 (< 0.01)	< 0.05
Large Lots of Residential	−9.450E-07 (< 0.01)	< 0.05
Retail			−1.45E-07 (< 0.01)	0.02
Utility			−1.050E-06 (< 0.01)	0.03
QIC	23.578		23.386		23.375
QICC	27.920		29.367		27.589

minor approach at signalized intersections in suburban area is higher than in the CBD and urban areas. Hence, higher priority is given for major approaches than minor approaches. The high traffic volume, high speed limit on the major approach, and more green time for the major approach traffic at signalized intersections in suburban area influence some drivers from minor approach to take risk and violate the signal.

An increase in area with retail type land use within 0.5 miles from a signalized intersection in suburban areas increases the number of red light violation crashes. This could be attributed to increased traffic activity at such signalized intersections. However, in CBD area, an increase in area with retail type land use within 0.5 miles from a signalized intersection decreases the number of red light violation crashes (could be attributed to the presence of mixed use activity and alternate mode users). Likewise, an increase in areas with utility type land use in CBD area, areas with warehouse/distribution land use in urban area, and areas with single-family detached land use in suburban area decrease the number of red light violation crashes.

Overall, the predictor variables and their effect on red light violation crashes differ by area type. The developed crash estimation models or safety performance functions (SPFs) by area type can be used to predict the number of red light violation crashes at a signalized intersection, by area type, within the city of Charlotte boundary. Also, these SPFs can be used to identify and prioritize signalized intersections for safety improvements.

Table 8 Crash Estimation Models – Urban Area.

Parameters	Model 1		Model 2		Model 3
	B (S.E.)	p-value	B (S.E.)	p-value	B (S.E.)	p-value
Intercept	−0.071 (0.48)	0.88	0.900 (0.17)	< 0.05	1.297 (0.17)	< 0.05
Traffic Volume	1.216E-05 (< 0.01)	< 0.05	1.173E-05 (< 0.01)	< 0.05	1.157E-05 (< 0.01)	< 0.05
All-red Clearance Time	0.898 (0.28)	< 0.05
Horizontal Mixed Use Non-Residential			3.591E-07 (< 0.01)	< 0.05	2.200E-07 (< 0.01)	< 0.05
Industrial					−1.207E-07 (< 0.01)	< 0.05
Large Lots of Residential	−2.808E-07 (< 0.01)	< 0.05
Open Space/Recreational			2.564E-07 (< 0.01)	0.02
Warehouse/Distribution	−1.349E-07 (< 0.01)	< 0.05
QIC QICC	14.693 22.921		16.359 22.250		18.015 24.196

4. Conclusions

Red light running violation is a major contributing factor of crashes at signalized intersections. This paper investigates red light violation crashes by categorizing the study area into CBD, urban, and suburban areas. The on-network, traffic, signal, demographic, and land use characteristics were considered as predictor variables. Crash estimation models were developed by randomly selecting sixty signalized intersections in each area. Models were also developed using data for all the 180 signalized intersections.

As the physical property of predictor variables vary by area type, different predictors were observed to be significant and influence the number of red light violation crashes in the three different area types. CBD, urban, and suburban areas have different land use patterns that contribute to the number of red light violation crashes. For example, industrial, large lots of residential, retail, and utility type land uses are observed as significant predictor variables in the CBD area. They attract people from all over the study area. These areas seem to have a direct but negative influence on the number of red light violation crashes. On the other hand, in an urban area, areas with horizontal mixed nonresidential land use influence the number of red light violation crashes. Likewise, in suburban area, the number of red light violation crashes are higher at signalized intersections with high traffic volume, speed limit, and areas surrounded with horizontal mixed non-residential and

Table 9 Crash Estimation Models – Suburban Area.

Parameters	Model 1		Model 2		Model 3
	B (S.E.)	p-value	B (S.E.)	p-value	B (S.E.)	p-value
Intercept	1.084 (0.14)	< 0.05	0.080 (0.62)	0.90	−0.262 (0.44)	0.55
Traffic Volume	8.03E-06 (< 0.01)	< 0.05
Speed Limit on Major Approach			0.022 (0.01)	0.05	0.031 (0.01)	< 0.05
# of Lanes on Minor Approach					0.152 (0.61)	0.01
All-red Clearance Time			0.542 (0.21)	< 0.05
Horizontal Mixed Use Non-Residential	4.537E-07 (< 0.01)	< 0.05
Retail					1.374E-07 (< 0.01)	< 0.05
Single-family – Attached	−3.937E-07 (< 0.01)	0.04			−4.021E-07 (< 0.01)	0.02
Single-family - Detached			−5.975E-08 (< 0.01)	0.01
Transportation			−1.609E-07 (< 0.01)	0.01
QIC QICC	16.090 21.966		22.386 29.944		17.098 24.708

Table 10 Crash Estimation Model – All Signalized Intersections.

Parameters	Model 1		Model 2		Model 3
	B (S.E.)	p-value	B (S.E.)	p-value	B (S.E.)	p-value
Intercept	1.047 (0.09)	< 0.05	1.351 (0.07)	< 0.05	1.566 (0.20)	< 0.05
Traffic Volume	6.720E-06 (< 0.01)	< 0.05
Speed Limit on Minor Approach			0.02 (< 0.01)	< 0.05
#of Lanes on Minor Approach					0.120 (0.06)	0.03
Horizontal Mixed Use Non-Residential	2.361E-07 (< 0.01)	< 0.05	1.673E-07 (< 0.01)	< 0.05
Industrial					−1.026E-07 (< 0.01)	< 0.05
Large Lots of Residential	−2.409E-07 (< 0.01)	< 0.05	−2.051E-07 (< 0.01)	< 0.05	−1.948E-07 (< 0.01)	< 0.05
Open Space/Recreational	1.279E-07 (< 0.01)
Retail					7.106E-08 (< 0.01)	0.05
Single-family - Attached			−3.249E-07 (< 0.01)	< 0.05
Transportation			−1.486E-07 (< 0.01)	< 0.05
QIC QICC	63.303 71.776		67.438 77.823		69.289 75.294

Table 11 Model Validation – Results Summary.

Area Type	Model 1	Model 2	Model 3
CBD	0.81	0.77	0.78
Urban	0.26	0.54	0.24
Suburban	0.29	0.28	0.33
All	0.61	1.28	1.29

retail type land use. In addition, traffic volume is the most significant predictor variable related to the number of red light violation crashes. It was found to be a significant predictor variable in all the developed models.

Based on the findings, manual enforcement or RLCs can be an effective measure at signalized intersections with high traffic volume in the CBD area. Implementing manual enforcement or installing RLCs at signalized intersections with high all-red clearance time, near high density of horizontal mixed non-residential and open space/recreational type land uses in urban area; at signalized intersections with high traffic volume, speed limit on the major approach, the number of lanes on the minor approach, all-red clearance time and areas surrounded with horizontal mixed non-residential and retail type land use in suburban areas would lead to a reduction in the number of red light violation crashes.

The all-red clearance time could be checked and optimized, if found inadequate, at signalized intersections with higher number of red light violation crashes in urban and suburban areas. Likewise, the speed limit on the major approaches, and possible influence of traffic, signal phases and signal timings attributed to minor approaches could be examined at signalized intersections with higher number of red light violation crashes in suburban areas.

Data for the city of Charlotte were used for this research. Similar studies should be conducted using data for other cities to increase the

Table 12 Effect of Predictor Variables.

Predictor Variables	CBD Area	Urban Area	Suburban Area	All
Traffic Volume	P	P	P	P
Speed Limit on the Major Approach			P
Speed Limit on the Minor Approach				P
# of Lanes on the Minor Approach			P	P
All-red Clearance Time		P	P
Horizontal Mixed Use Non-Residential		P	P	P
Industrial	N	N		N
Large Lots of Residential	N	N		N
Open Space / Recreational		P		P
Retail	N		P	P
Single-family - Attached			N	N
Single-family - Detached			N
Transportation			N	N
Utility	N
Warehouse/Distribution		N

P: Positive EffectN: Negative Effect.

understanding of the role of predictor variables on the number of red light violation crashes. Further, geospatial characteristics within 0.5 miles were only captured and used in the analysis. Considering other buffer widths and detailed demographic and socio-economic characteristics (age, gender, ethnicity, income groups, etc.) warrant further investigation.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Acknowledgement

The authors sincerely thank the staff of the city of Charlotte Department of Transportation (CDoT) for their help with providing data required for the study.

References

Alver, Y., Demirel, M.C., Mutlu, M.M., 2014. Interaction between socio-demographic characteristics: traffic rule violations and traffic crash history for young drivers. Accid. Anal. Prevent. J. 72, 95–104.
Baguley, C.J., 1988. Running the red at signals on high-speed roads. J. Traffic Eng. Control 29 (7–8), 415–420.
Bonneson, J.A., Zimmerman, K., Brewer, M.A., 2002. Engineering Countermeasures to Reduce Red-Light-Running. Publication # FHWA/TX-03/4027-2. FHWA. United States Department of Transportation.
Chapman, E.A., Masten, S.V., Browning, K.K., 2014. Crash and traffic violation rates before and after licensure for novice California drivers subject to different driver licensing requirements. J. Saf. Res. 50, 125–138.
Datta, T., Schattler, K., Datta, S., 2000. Red light violations and crashes at Urban inter sections. Transportation Research Record: Journal of the Transportation Research Board, No. 1734. Transportation Research Board of the National Academies, Washington, D.C, pp. 52–58.
Garber, N.J., Miller, J.S., Abel, R.E., Eslambolchi, S., Korukonda, S.K., 2007. The Impact of Red Light Cameras (Photo-Red Enforcement) on Crashes in Virginia. Publication No. FHWA/VTRC 07-R2. United States Department of Transportation.
Gates, T.J., Noyce, D.A., 2010. Dilemma zone driver behavior as a function of vehicle type, time of day, and platooning. Transportation Research Record: Journal of the Transportation Research Board, No. 2149. Transportation Research Board of the National Academies, Washington, DC, pp. 84–93. (Accessed 20 June 2016). https:// connect.ncdot.gov/resources/safety/its%20and%20signals%20resources/its%20and %20signals%20unit%20design%20manual.pdf.
Insurance Institute for Highway Safety (IIHS), 2007. Red Light Cameras in Philadelphia All but Eliminate Violations. Status Report, vol. 42 No. 1.
Kamyab, A., McDonald, T., Stribiak, J.J., Storm, B., 2000. Red Light Running in Iowa: The Scope, Impact, and Possible Implications. Ames IA. Center for Transportation Research and Education, Iowa State University December.
Kamyab, A., McDonald, T., Stribiak, J.J., 2002. The scope and impact of Red light running in Iowa. the 81st Annual Meeting of the Transportation Research Board Compendium of Papers. Transportation Research Board of the National Academies,
Washington, D.C.
Kulanthayan, S., Phang, W.K., Hayati, K.S., 2007. Traffic light violation among motorists in Malaysia. IATSS Res. 31 (2), 67–73.
Maisey, G.E., 1981. The Effects of Mechanical Surveillance Device on Urban Signalized Intersection Accidents. Research and Statistics Report No.17. Road Traffic Authority, Perth, Australia.
McGwin Jr., G., Brown, D.B., 1999. Characteristics of traffic crashes among young, middle-aged, and older drivers. Accid. Anal. Prevent. J. 31 (No. 3), 181–198.
Mohamedshah, Y.M., Chen, L.W., Council, F.M., 2000. Association of selected intersection factors with Red light running crashes. Highway Safety Information System (HSIS) Summary Report. United States Department of Transportation.
North Carolina Department of Transportation (NCDOT), 2004. Traffic Management & Signal Systems Unit Design Manual.
Oh, J., Lyon, C., Washington, S., Persaud, B., Bared, J., 2003. Validation of FHWA crash models for rural intersections: lessons learned. Transp. Res. Rec.: J. Transp. Res. Board 1840, 41–49.
Porter, B.E., England, K.J., 2000. Predicting red-light running behavior: a traffic safety study in three urban settings. J. Saf. Res. 31 (1), 1–8.
Pulugurtha, S.S., Otturu, R., 2014. Effectiveness of red light running camera enforcement program in reducing crashes: evaluation using "before the installation", "after the installation", and "after the termination" data. Accid. Anal. Prevent. J. 64, 9–17.
Retting, R.A., Kyrychenko, S.Y., 2002. Reductions in injury crashes associated with Red light camera enforcement in Oxnard, California. The 81st Annual Meeting of the Transportation Research Board Compendium of Papers. Transportation Research Board of the National Academies, Washington, D.C.
Retting, R.A., Williams, A.F., 1996. Characteristics of red light violators: results of a field investigation. J. Saf. Res. 27 (1), 9–15.
Retting, R.A., Williams, A.F., Greene, M.A., 1998. Red-light running and sensible countermeasures: summary of research findings. Transportation Research Record: Journal of the Transportation Research Board, No. 1640. Transportation Research Board of the National Academies, Washington, D.C, pp. 23–26.
Schattler, K., Datta, T., 2004. Driver behavior characteristics at urban signalized inter sections. Transp. Res. Rec.: J. Transp. Res. Board 1862, 17–23.
Schattler, K., Datta, T., Hill, C., 2003. Change and clearance interval design on red-light running and late exits. Transp. Res. Rec.: J. Transp. Res. Board 1856, 193–201.
Shell, D.F., Newman, I.M., Córdova-Cazar, A.L., Heese, J.M., 2015. Driver education and teen crashes and traffic violations in the first two years of driving in a graduated licensing system. Accid. Anal. Prevent. J. 82, 45–52.
South, D.R., Harrison, W.A., Portans, I., King, M., 1998. Evaluation of the Red Light Camera Program and the Owner Onus Legislation. Victoria Transport: Road Traffic Authority, Hawthorn, Victoria, Australia.
Tay, R., 2017. Correlation, variance inflation and multicollinearity in regression model. J. Eastern Asia Soc. Transp. Stud. 12, 2006–2015.
Tay, R., De Barros, A., 2009. Minimizing red light violations: how many cameras do we need for a given number of locations? J. Transp. Saf. Secur. 1 (1), 18–31.
Tay, R., De Barros, A., 2011. Should traffic enforcement be unpredictable? The case of red light cameras in Edmonton. Accid. Anal. Prevent. J. 43 (3), 955–961.
Van der Horst, A., Richard, A., Wilmink, A., 1986. Drivers' decision-making at signalised intersections: an optimisation of the yellow timing. J. Traffic Eng. Control 27 (12), 615–617.
Weldegiorgis, Y., Jha, M.K., 2009. Driver behavior, dilemma zone, and capacity at red light camera equipped intersections. Transportation and Traffic Theory 2009: Golden Jubilee. pp. 481–494.
Zimmerman, K., 2007. Additional dilemma zone protection for trucks at High-speed signalized intersections. Transportation Research Record: Journal of the Transportation Research Board, No. 2009. Transportation Research Board of the National Academies, Washington, DC, pp. 82–88.

maneinfluencenetworktrafficsignal2018.md 96 KB Түүх Анхны өгөгдөл