Legacy Book

You are here

Lane Width Crash Modification Factors for Curb-and-Gutter Asymmetric Multilane Roadways: Statistical Modeling

THOBIAS SANDO, PH.D., P.E., PTOE
GEOPHREY MBATTA, PH.D.
REN MOSES, PH.D., P.E.

School of Engineering
University of North Florida
1 UNF Drive
Jacksonville, Florida 32256
Phone: 904-620-1142
Fax: 904-620-1391
Email: t.sando@unf.edu

Department of Civil Engineering
FAMU-FSU College of Engineering
2525 Pottsdamer Street, Room 129-B
Tallahassee, FL 32310
Phone: (850) 410-6587
Fax: (850) 410-6587
Email: mbattageo@eng.fsu.edu

Department of Civil Engineering
FAMU-FSU College of Engineering
2525 Pottsdamer Street, Room 129
Tallahassee, FL 32310
Phone: (850) 410-6191
Fax: (850) 410-6142
Email: moses@eng.fsu.edu

KEYWORDS: Crash modification factors, lane width, asymmetric roadways

Abstract

This study developed lane width crash modification factors (CMFs) for urban curb-andgutter multilane roadways with asymmetric lanes, i.e., outside lane wider than inside lane. The roadway segments used were urban fourlane with a raised median (4D) and with a twoway left-turn lane (5T). Three crash categories were evaluated: KABCO (Fatal (K), incapacitating-injury (A), non-incapacitating injury (B), possible injury (C) and property damage only crashes (O)), KABC (Fatal (K), incapacitating-injury (A), non-incapacitating injury (B), and possible injury crashes (C)), and PDO (property damage only) crashes.

A cross-sectional method was used as it was the most practical and feasible for this study. Six-year (2004 to 2009) of segment crashes were examined. The analysis involved statistical modeling using the negative binomial model, whose coefficients were used to develop multiplicative CMF equations for a combined effect of variable inside and outside lane width.

In summary, the results show that reducing the inside lane width from 12 ft to 11 ft does not affect estimated crash frequency of four-lane with a raised median segments for all three crash categories, and PDO crashes for fourlane with a two-way left-turn lane segments.

However, narrowing the inside lane width appears to be associated with increased estimated KABCO and KABC crashes for four-lane with a two-way left-turn lane sections. The results also suggest that widening the outside lane from the baseline 12 ft causes a reduction in estimated crash frequency for all three crash categories (KABCO, KABC, and PDO) for both four-lane with a raised median and fourlane with a two-way left-turn lane segments.

Introduction

This paper presents the process that was used to develop crash modification factors (CMFs) for urban multilane curb-and-gutter asymmetric lanes in the state of Florida. The CMFs developed in this study describe change in safety when a typical section measuring 12 ft for both inside and outside lanes is changed to an asymmetric section, also known as an atypical configuration. An example of this would be changing a 12 ft inside and outside lane to an 11 ft inside lane and 13 ft outside lane. Most of these roadway configurations result from retrofitting, by repainting, or widening of the roadway rendering the outside lane a shared lane for bicycles and motor vehicles.

Literature Review

There are several studies that were conducted to develop lane width crash modification factors (CMFs) for two-lane rural highways (Griffin and Mak (1987), Zeeger et al. (1987), Harwood et al. (2000), Harwood et al. (2003), and Harkey et al. (2008)). These studies are also cited in the Highway Safety Manual, HSM (2010). They were all conducted on two-lane rural highways. Separate CMFs were reported for roadways with average annual daily traffic (AADT) less than 400 motor vehicles per day and for roadways with AADT greater than 2000 motor vehicles per day. These CMFs indicate that widening of lanes reduce a specific set of related accident types, namely single-vehicle run-offroad accidents, multiple-vehicle head-on, opposite-direction sideswipe, and same-direction sideswipe collisions (Harkey et al. (2008)& Highway Safety Manual (2010)). This decrease was relative to 12 ft lane width, which was considered the base line of comparison.

In another lane width study, Lord and Benneson (2007) developed lane width CMFs for two-lane rural highway frontage roads for the state of Texas. Rural frontage roadways differ from rural two-lane roadways because they have restricted access along at least one side of the road, a higher percentage of turning traffic, and periodic ramp-frontage-road terminals with yield control. The results showed increased crash frequency as lane width decreased from 12 ft to 9 ft.

CMFs reported in the Highway Safety Manual (2010) for rural multilane roadways were developed by the study that was conducted by Harkey et al. for the National Cooperative Highway Research Program (NCHRP) (2008). An expert panel was used to develop CMFs for rural multilane roadways. Lane width CMFs developed by modeling crashes for multilane highways are absent. Furthermore, it is clear that CMFs reported in the HSM (2010) for rural multilane (both divided and undivided) highways may not apply to urban multilane roadways. This is due to the difference in traffic operations and level of activities surrounding urban highways.

In a recent lane width study, Potts et al. (2007) investigated the relationship between lane width and safety for roadway segments on urban and suburban arterials. The study by Potts et al. did not develop CMFs. The study did not find any indication of safety risk on urban and suburban arterials when lane width narrower than 12 ft was used.

Based on the summary of the literature review, two main observations need special attention. First, the average lane width was used in all previous studies that developed CMFs for lane width. While averaging may apply to symmetric lane configurations, such as 12 ft inside lane and 12 ft curb lane, it may be too simplistic for asymmetric sections, which have wider curb lanes and narrow inside lanes. Second, all existing CMFs for lane widths were developed for rural highways. None of the CMFs reported in previous studies were developed to specifically address the safety consequences of lane width in urban roadways. These two observations are the motivation of this study as it employs individual lane measurements instead of the average of aggregated lane width and focuses on urban segments, helping to fill the knowledge gap that exists in lane width CMF development.

Rural and urban highways differ in their crosssectional geometric configurations. Rural highways have shoulders while urban highways considered in this study have curb-andgutter. The shoulder provides room for road users to veer to the right if they are on the outside lane to avoid a crash while curb-andgutter roadways causes a constraint for lateral movement to the right of the travel lanes. Also, bicyclists do not share a lane with motorists on rural highways. They ride on the shoulder to the right of the white stripe. This study presents the analysis of urban wide curb lanes, i.e., outside lanes widths greater than 12 feet, to accommodate bicyclists and motorists on the same lane.

Research Objective

The objective of this study was to evaluate the safety of urban multilane roads with atypical lane width configurations, i.e., outside lane width greater than the standard lane width (12 ft) and narrower inside lane narrower than 12 ft. This objective was accomplished by developing Crash Modification Factors for two types of atypical multilane urban cross sections namely urban four-lane roadways with divided median (4D) and urban four-lane with two-way left-turn lane (5T). Crash Modification Factors quantify the change in expected average crash frequency (crash effect) caused by implementing a particular treatment. The value of Crash Modification Factor below 1 indicates treatment causes crash reduction while Crash Modification Factor greater than 1 indicates that the treatment is expected to result in an increased number of crashes. a Crash Modification Factor of 1 represents no effect on safety.

Data Collection

Roadway Data

Databases Used

Roadway characteristics inventory (RCI): This database was used to identify the type of road configuration and roadway characteristics including: total surface lane width, number of lanes, shoulder type, and traffic characteristics. All four-lane with a raised median and four-lane with a two-way left-turn lane were filtered and further analyzed using Florida Department of Transportation (FDOT) as-built plans.

FDOT scanned copy of as-built plans: FDOT archives scanned copies of as-built plans for state maintained roadway projects. The database has most of the roadway plans for completed projects and projects that are under construction. The advantage of as-built plans over RCI database is that they show individual lane width while roadway characteristics inventory database shows the total surface width. From the as-built plans, using roadway ID obtained from roadway characteristics inventory database, four-lane with a raised median and fourlane with a two-way left-turn lane multilane roadways with asymmetric lanes were verified by examining individual lane width. Additional data obtained from the as-built plans include individual lane width, type of median, number of driveways, number of median openings, and approximated segment length.

Comparison Sites

Comparison sites were roadways with standard lane width of 12 ft for both inside and outside lanes. They were obtained by using a technique suggested by Bonneson and Pratt (2008). At first, sites adjacent to selected asymmetric lanes on the same roadway (figure 1a) were chosen to ensure that the pairs were homogenous to the asymmetric segments. However, it was not possible to get sufficient data using this technique. Therefore, the selection was expanded to consider parallel (figure 1b) and intersecting (figure 1c) the selected asymmetric segments. Parallel and intersecting segments were considered only if their roadway characteristics were similar to the paired asymmetric segments and had a comparable average annual daily traffic (AADT). The attributes used for the selection of comparison sites were number of lanes, median type, posted speed limit, degree of curve, type of shoulder (curb), and type of onstreet parking.

Verification of Roadway Geometric Information for Asymmetric and Comparison Segments

Two issues emerged when reviewing the asbuilt drawings. First, it was discovered that most of as-built plans are not updated regularly. Second, there was inconsistency in the way the curb lane was measured. Therefore, as-built plans measurements were verified by performing field measurements for segments with asymmetric lanes and comparison sites. A total of 918 road segments were verified. After field verification, 454 segments (49.5% of all segments) were dropped as their characteristics differed those recorded on RCI and as-built drawings, hence did not qualify for analysis. A minimum of 100 segments is recommended for modeling (Agrawal and Lord, 2006). After field verification, both 4D and 5T were found to have enough segments for modeling with a total of 224 and 240 segments, respectively, for both segments with asymmetric lanes and comparison segments.

Crash Data

Statewide crash data was obtained from crash analysis reporting (CAR) database, an electronic repository of crashes maintained by FDOT. The data was from 2004 to 2009. The location of each crash was linearly referenced to the Florida Department of Transportation roadway system using the milepost system indexed by the roadway identification number (Roadway ID). Data was filtered to remain with mid-block crashes only. All crashes that occurred within a radius of 250 ft from the center of intersections were discarded.

Data Analysis

Crash Rate Analysis for Segments With Asymmetric Lanes Configuration

Categories of outside lane width were formed by grouping ranges of lane widths as follow: 11.8 ft–12.2 ft formed a 12 ft category; 12.3 ft– 12.7 ft formed a 12.5 ft category; 12.8 ft–13.2 ft formed a 13 ft category; 13.3 ft–13.7 ft formed a 13.5 ft category; and 13.8 ft–14.2 ft formed a 14 ft category. It should be noted that for the 12 ft category of outside lane width, the inside lane width was also12 ft (comparison sites). However, for all other lane width categories, the inside lane width was fixed to 11 ft. These categories were used for an explanatory analysis whose results are reported in table 1.

FIGURE 1 Criteria Used in Selection of Comparison Segments

FIGURE 1 Criteria Used in Selection of Comparison Segments

As can be seen in table 1, crash categories are described using three acronyms, i.e., KABCO, KABC, and PDO crashes, derived from the Highway Safety Manual naming convention. KABCO stands for fatal (K), incapacitating (A), non-incapacitating (B), possible injury (C) and PDO (O) crashes while KABC represents fatal (K), incapacitating (A), non-incapacitating (B), and possible injury (C) crashes. PDO is used for crashes that result in property damage only.

Figure 2 is a graphical representation of the results shown in table 1, depicting the relationship between crash rate per million vehicle miles (mvm) and the outside lane width. The two graphs presented in figure 2 show an increase of crashes when outside lane width increased from 12 ft (with inside lane width of 12 ft) to 12.5 ft (with an inside lane of 11 ft). There is a discernible pattern of decreased crash rate as the outside lane width is increased from 12.5 ft to 14 ft with a fixed inside lane width of 11 ft. This trend was observed for all three crash categories, i.e., KABCO, KABC, and PDO crashes.

Statistical Modeling

Selection of the Statistical Model

Two regression count models used to analyze crash data are Poisson and Negative Binomial. Poison regression distribution requires the mean and variance of the dependent variable to be equal. For most crash data, the variance of the crash frequency exceeds the mean and, in such case, the data would be overdispersed. The Highway Safety Manual (2010) specifically calls for the use of the Negative binomial model in lieu of Poisson model because the degree of overdispersion in a negative binomial model is represented by a statistical parameter, known as the overdispersion parameter that is estimated along with the coefficients of the regression equation. The larger the value of the overdispersion parameter, the more the crash data vary as compared to a Poisson distribution with the same mean.

TABLE 1 Crashes Rate for Different 4D Outside Lane Width Categories

Inside lane width Outside lane width Exposure (mvm) KABCO crashes PDO crashes KABC crashes KABCO crashes/mvm PDO crashes/mvm KABC Crashes/mvm
(ft) (ft)
12 *12.0 1,631.24 739 313 426 0.45 0.19 0.26
11 12.5 297.81 219 84 135 0.74 0.28 0.45
11 13 595.56 372 164 208 0.62 0.28 0.35
11 13.5 482.45 162 63 99 0.34 0.13 0.21
11 14 254.45 45 19 26 0.18 0.07 0.1

Crashes rate for different 5T outside lane width categories

Inside lane width Outside lane width Exposure (mvm) KABCO crashes PDO crashes KABC crashes KABCO crashes/mvm PDO crashes/mvm KABC Crashes/mvm
(ft) (ft)
11 12.5 215.94 187 79 108 0.87 0.37 0.5
11 13 209.46 153 56 97 0.73 0.27 0.46
11 13.5 88.16 59 21 38 0.67 0.24 0.43
11 14 120.84 71 24 47 0.59 0.2 0.39

*Comparison sites with inside lane width of 12.0. All other categories have inside lane width of 11.0 ft.

Selection of the Function

FIGURE 2 Graphs of Outside Lane Width and Crashes Rate by Severities

Chart of 4D Outside Lane Width Vs Crash Rate (Crashes/MVM)

Chart of 5T Outside Lane Width Vs Crash Rate (Crashes/MVM)

The first step toward development of predictive models is the selection of the functional form. Normally, the function is determined empirically after several runs of different variable combinations which correlate the dependent variable (outcome variable) to the model covariates. Different functions were considered and fitness of resulting models was assessed. After several trials of different combination of variables, the function based on Negative Binomial (NB) model presented as equation 1 was selected.

(1) µ i = βoLi(ADT)β1 eΣni=2 xjiβi

Equation 1 was simplified to provide a linear relationship between the dependent variable and covariates by taking natural logarithm on both sides. The resulting formula is presented as equation 2.

(2) ln(µi) = ln(βo) + ln(Li) + β1ln(ADT) + Σni=2 xjiβi

Where

AADT = is an average annual daily traffic over six years of study period

Li = segment length

µ i = mean number of crashes for six year period for site i

x1, x2 ,…… xn = explanatory variables

βo, βi ,……., βn = regression coefficients to be estimated

Selection of Explanatory Variables

Previous studies have found that roadway cross-section variables such as lane width, median width, median type, grade, segment length, and degree of curve have contribution to occurrences of crashes (Zeeger et al. (1987) & Harkey et al. (2008)). Mauga and Kaseko (2010) found median opening density and driveway density to have contributed to the increase in crashes in urban multilane roads. Also, AADT and posted speed limit have been widely reported as important variables in crash modeling (Harkey et al. (2008), HCM (2010), & Mauga and Kaseko (2010)). In this study, two main explanatory variables—AADT and segment length were considered to be key variables that relate number of crashes to predictors. In addition, inside and outside lane width were considered as study variables and were given equal importance as key variables. Due to the nature of sites used for this study, other variables including posted speed limit (mph), median width (ft), degree of curve (degree), and driveway density (number of driveway/ 0.1 mile) and median opening density were also added in the model. Number of median opening density was found to be irrelevant for four-lane with a two-way left-turn lane configuration as the configuration does not restrict turning at any point. However, it was important for four-lane with a raised median configuration as turning to access adjacent properties is only through median openings.

It is worth noting that in previous studies, driveway density and median opening density had been expressed in terms of number of driveways/mile or number of openings/mile. However, for this study, segment lengths were ranged from 0.01 mile to 0.52 mile for fourlane with a two-way left-turn lane configuration with 0.1 mile being a mean value. In order to avoid having high values for driveway density and median openings, the mean segment length of 0.1 mile was used to scale the median openings and driveway density.

Negative Binomial (NB) Model Selection and Evaluation

The negative binomial model was developed to analyze the influence of the independent variables on three response variables i.e., KABCO, KABC, and PDO crashes. Model results were tested at 0.05 level of significant. All insignificant variables were removed to form a reduced model. A reduced model was re-run and tested again at the same level of significant.

Thereafter, a comparison of the full and reduced models was performed using two information-theoretic approach indicators i.e., Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The general criterion for comparison is that the model with a smaller value of AIC and BIC is considered to be better. The values of BIC and AIC for the reduced models were smaller than those of the full models for all three response variables (KABCO, KABC, and PDO) crashes. The reduced model was then selected for all three response variables.

Model Results for 4D Segments

Model results for four-lane with a raised median segments are reported in table 2. The results revealed an increase in KABCO, KABC, and PDO crashes as outside lane widths is decreased. The increase in crashes was significant when tested at 95% confidence level with p-values of 0.010, 0.044 and 0.0153 for the outside lane width. The effect of the inside lane width was insignificant, therefore the coefficient was removed. Also, the increase in median opening density resulted in the increase of KABCO, KABC, and PDO crashes. This was evident as the p-values of 0.0167, 0.0165, and 0.0315 for KABCO, KABC, and PDO crashes were observed. These results were consistent to the Mauga and Kaseko (2010) study which observed the increase in injury crashes with an increase in median opening density.

TABLE 2 Results for Urban Four-Lane Roadway With Divided Median (4D)

4D Segments—metadata for the KABCO, KABC, and PDO crashes

Variable Mean Standard deviation Minimum Maximum
AADT 37,510 7,383 25,100 52,500
Segment length (length) 0.17 0.1 0.01 0.64
Outside lane width (ft) 12.63 0.7 12 14
Inside lane width (ft) 11.68 0.4 11 12
Median opening density 0.68 1.3 0 10.6
Driveway density (drive way/0.1mile) 1.16 2.1 0 19

KABCO crashes

Parameter Estimate Standard error p-values Comment
Intercept -33.1729 4.8553 <.0001 Significant
Log of AADT 3.7903 0.4586 <.0001 Significant
Log of length 0.3505 0.126 0.0054 Significant
Outside lane width (ft) -0.3591 0.1395 0.0101 Significant
Median opening density 0.1713 0.0716 0.0167 Significant

Deviance (value/df): 1.12
Over-dispersion parameter k: 1.51
BIC: 1198.95
AIC: 1177.82
Pearson χ2 (value/df): 1.04

KABC crashes

Parameter Estimate Standard error p-values Comment
Intercept -31.8083 5.321 <.0001 Insignificant
Log of AADT 3.5618 0.5011 <.0001 Significant
Log of length 0.3944 0.1391 0.0046 Significant
Outside lane width (ft) -0.3113 0.1546 0.044 Significant
Median opening density 0.1921 0.0802 0.0165 Significant

Deviance (value/df): 1.05
Over-dispersion parameter k: 1.71
BIC: 1002.31
AIC: 981.98
Pearson χ2 (value/df): 1.01

PDO crashes

Parameter Estimate Standard error p-values Comment
Intercept -38.6478 5.6048 <.0001 Insignificant
Log of AADT 4.2327 0.5342 <.0001 Significant
Log of length 0.3038 0.1444 0.0354 Significant
Outside lane width (ft) -0.3743 0.1544 0.0153 Significant
Median opening density 0.1584 0.0736 0.0315 Significant

Deviance (Value/df): 1.01
Over-dispersion parameter k: 1.51
BIC: 869.59
AIC: 849.26
Pearson χ2 (value/df): 1.03

Model Results for Four-Lane With a Two-Way Left-Turn Lane Segment

Table 3 presents the model results for four-lane with a two-way left-turn lane segments. According to the results reported in table 3, both KABCO and KABC crashes increased with reduced lane width for both lanes (inside and outside). The results were significant at 95% confidence level. For KABC crashes, p-values of 0.0184 and 0.0294 for inside and outside lane, respectively, were observed while for KABCO crashes, p-values for inside and outside were 0.0493 and 0.0106, respectively.

Both KABCO and KABC crashes were significantly correlated to driveway density. The increase in driveway density resulted in the increase in KABCO and KABC crashes. Pvalues of 0.0334 and 0.0007 for KABCO and KABC crashes, respectively, were observed. This finding is consistent to the results reported by Mauga and Kaseko (2010) which observed the increase in injury crash rate as driveway densities were increased. However, with respect to PDO crashes, the inside lane width and driveway densities were found to be insignificant not only at 95%, but also at 90% confidence level.

The influence of AADT was found to be significant for all three response variables (KABCO, KABC and PDO crashes). The model yielded p-values of 0.0001, 0.0001 and 0.0001 for KABCO, KABC and PDO crashes, respectively, for four-lane with a raised median segments. P-values of 0.0001, 0.0001, 0.0111, for KABCO, KABC and PDO crashes, respectively, were observed for four-lane with a twoway left-turn lane segments. In all three cases for the four-lane with a two-way left-turn lane segments, the model coefficient for segment length was approximately 1.000, therefore was used as an offset variable.

Developing Crash Modification Factors

Method Used

The Highway Safety Manual (2010) provides a list of methods that can be used for developing Crash Modification Factors. The most preferred methods are controlled experiments and the Empirical Bayes method using the beforeand-after data. Due to the difficulty in obtaining the exact date that a treatment was implemented, the before-and-after analysis was not feasible for this study. Another method recommended by the Highway Safety Manual (2010), i.e., the cross-sectional method was therefore adopted as it does not require the "before" period data for analysis. Instead, it employs the treatment and comparison sites of "after" period data for analysis. It is the same method that was used by Lord and Bonneson (2007) to estimate Crash Modification Factors for rural frontage roads in Texas. The method estimates Crash Modification Factors by using coefficients developed from regression models, for this case, coefficients reported in tables 2 and 3. Crash Modification Factors for each specific response variable follow an exponential relationship shown in equation 3.

(3) CMFi=eβi[xi-yi]

Where

xi = range of values or a specific value investigated (e.g. lane width, etc.)

yi = baseline conditions or average conditions for the variable

βi = regression coefficient

TABLE 3 Results for Urban Four-Lane Roadway With TWLTL (5T)

5T Segments—metadata for the KABCO, KABC, and PDO crashes

Variable Mean Standard deviation Minimum Maximum
AADT 22078 7118 7480 43929
Segment length (length) 0.1 0.07 0.01 0.52
Outside lane width (ft) 12.6 0.7 12 14
Inside lane width (ft) 11.5 0.5 11 12
Median width (ft) 12.1 1 10 14
Driveway density (drive way/0.1mile) 5 3 0 24

KABCO crashes

Parameter Estimate Standard error p-values Comment
Intercept 7.9121 6.8456 0.2478 Insignificant
Log AADT 1.019 0.2412 <.0001 Significant
Driveway density 0.0504 0.0237 0.0334 Significant
Outside lane width (ft) -0.5887 0.2305 0.0106 Significant
Inside Lane width (ft) -0.6318 0.3214 0.0493 Significant

Deviance (value/df): 1.11
Over-dispersion parameter k: 1.07
BIC: 999.53
AIC: 979.56
Pearson χ2 (value/df): 1.38

KABC crashes

Parameter Estimate Standard error p-values Comment
Intercept 5.6102 6.7447 0.4055 Insignificant
Log AADT 1.1963 0.2523 <.0001 Significant
Driveway density 0.0823 0.0243 0.0007 Significant
Outside lane width (ft) -0.4978 0.2285 0.0294 Significant
Inside lane width (ft) -0.7452 0.3162 0.0184 Significant

Deviance (value/df): 1.09
Over-dispersion parameter k: 0.87
BIC: 795.99
AIC: 815.96
Pearson χ2 (value/df): 1.4

PDO crashes

Parameter Estimate Standard error p-values Comment
Intercept -1.2237 3.948 0.7566 Insignificant
Log AADT 0.9114 0.3587 0.0111 Significant
Outside lane width (ft) -0.4092 0.1835 0.0258 Significant

Deviance (value/df): 0.9
Over-dispersion parameter k: 2.14
BIC: 676.39
AIC: 659.75
Pearson χ2 (value/df): 1.14

Lane Width Crash Modification Factors

The lane width of 12 ft for both inside and outside lanes was considered as a base condition. Since CMFs are multiplicative factors when used to predict crash frequencies, based on the results presented in tables 2 and 3, and equation 3, the resulting CMFs were derived as:

  • four-lane with a raised median Crash modification functions
    CMF for KABCO crashes (4) CMFKABCO = e-0.36[xoutside-12]
    CMF for KABC crashes (5) CMFKABC = e-0.31[xoutside-12]
    CMF for PDO crashes: (6) CMFPDO = e-0.37[xoutside-12]
  • four-lane with a two-way left-turn lane Crash modification functions
    CMF for KABCO crashes: (7) CMFKABCO = e-0.59[xoutside-12] , e-0.63[xoutside-12]
    CMF for KABC crashes: (8) CMFKABC = e-0.50[xoutside-12] , e-0.75[xoutside-12]
    CMF for PDO crashes: (9) CMFPDO = e-0.41[xoutside-12]

It can be noted that the coefficient of the inside lane width is not included in the Crash Modification Factors of all three crash categories for fourlane with a raised median segments (equations 4 to 6) and Crash Modification Factor of PDO crashes for four-lane with a two-way left-turn lane segments (equation 9). This is because the inside lane width was not significant for these particular cases as explained in previous sections.

Crash Modification Factor Curves and Interpretation

Figures 3 and 4 show Crash Modification Factor curves for KABCO, KABC, and PDO crashes for four-lane with a two-way left-turn lane segments and four-lane with a raised median segments, respectively. The six curves were developed by substituting lane widths of 11 ft and 12 ft for the inside lane and varying lane widths of 12.5 to 14 for the outside lane, in equations 4 through 9. The base CMF of 1.00 corresponds to the segments with the inside and outside lane width of 12 ft each.

When considering different combination of inside and outside lane widths, the following observations were made. For four-lane with a twoway left-turn lane configuration (figure 3), apart from the base condition (12 ft for both inside and outside lanes), Crash Modification Factor of 1.0 for KABCO crashes was observed for a combination of 11 ft inside and 13 ft outside lane and a pair of 11 ft inside and 13 ft outside lane width. The combination of 11.5 ft and 13 ft resulted to Crash Modification Factor of 0.75, which indicates reduced estimated average KABCO crash frequency in comparison to the base condition. For KABC crashes, the combination of 11 ft inside/13 ft outside lane width and 11.5 ft inside/12.5 ft outside lane width yielded Crash Modification Factors greater than 1.00, which indicates an increase in estimated KABC crashes. On the other hand, the combination of 11.5 ft and 13 ft resulted in a CMF Crash Modification Factor smaller than 1.00, which indicates a reduction in estimated KABC crashes. With respect to PDO crashes, it was outside lane width which had an effect on Crash Modification Factor. As the width increased to greater than 12 ft the Crash Modification Factor was less than one indicating reduction in PDO crashes.

For four-lane with a raised median configuration (figure 4), 60% reduction for KABCO and PDO crashes was observed as the Crash Modification Factor decreased from 1.0 to 0.4 as the outside lane width increases from 12 ft to 14.5 ft. The crash reduction of 66% was observed for KABC crashes as the outside lane width widens from 12 ft to 14.5 ft.

FIGURE 3 Graphs of CMFs for Four-Lane With a Two-Way Left-Turn Lane Segments

Graph of Outside Lane Width Vs KABCO CMFs for 5T Road Type

Graph of Outside Lane Width Vs KABC CMFs for 5T Road Type

Graph of Outside Lane Width Vs PDO CMFs for 5T Road Type

FIGURE 4 Graph of CMFs for four-lane with a raised median segments

Graph of Outside Lane Width Vs KABCO CMFs for 4D Road Type

Graph of Outside Lane Width Vs KABC CMFs for 4D Road Type

Graph of Outside Lane Width Vs PDO CMFs for 4D Road Type

Summary

This study developed lane width crash modification factors for asymmetric urban multilane roadways. The roadway segments used were urban four-lane with a raised median (4D) and with two-way left-turn lane (5T). In total, the analysis reported in this study used 25 centerline miles of four-lane with a twoway left-turn lane segments and 39 centerline miles of four-lane with a raised median roadways.

Development of Crash Modification Factors followed a protocol described by the Highway Safety Manual (2010). The cross-sectional method was used. Negative binomial regression models were used to model the relationship between crash frequency and model variables. Variables considered in modeling included driveway density, median opening density, posted speed limit, inside lane width, outside lane width, median width, segment length, and average annual daily traffic (AADT). Six years (2004–2009) of segment crashes were examined. Three crash categories were evaluated: KABCO (Fatal (K), incapacitating-injury (A), non-incapacitating injury (B), possible injury (C) and property damage only crashes (O)), KABC (Fatal (K), incapacitating-injury (A), non-incapacitating injury (B), and possible injury crashes (C)), and PDO (property damage only) crashes.

The results of the safety analysis are summarized in table 4. These values are calculated using equations 4 through 9. A Crash Modification Factor of 1.00 indicates no influence in causing crashes while Crash Modification Factors smaller and greater than 1.00 indicate that a change of a variable from a base value causes a decrease and increase in crashes, respectively. According to the results depicted in table 4, for four-lane with a raised median segments, the effect of inside lane width is insignificant, indicating that the decrease of lane width from 12 ft to 11 ft does not cause an increase in crash frequency. According to the results, crashes decrease as the outside lane width is increased from 12 ft. This decrease is seen on all types of crashes analyzed in this study, i.e., KABCO, KABC, and PDO.

For four-lane with a two-way left-turn lane sections, the results show an increase in crashes as the inside lane width is reduced to 11 ft while the outside lane width is increased to 12.5 ft. This trend was observed for both KABCO and KABC crashes, but not for PDO crashes. However, the combination of 11.5 ft or more for the inside lane and 13 ft for the outside lane width resulted in the decrease in crashes for KABCO and KABC crashes. Crash Modification Factors for PDO crashes were found to be independent of the inside lane width, but dependent of outside lane width. Relative to outside lane width of 12 ft, the Crash Modification Factors for PDO crashes were found to decrease as the outside lane width increased.

As stated above, for four-lane with a raised median segments, narrowing the inside lane from 12 ft to 11 ft did not result in an increase in crash frequency for any of the three types of crashes. Also, for four-lane with a two-way left-turn lane segments, the decrease in inside lane width was not significant for PDO crashes. It was only significant for KABCO and KABC crashes, hence higher values of Crash Modification Factors for KABCO and KABC crashes for four-lane with a two-way leftturn lane. As far as four-lane with a two-way left-turn lane segments are concerned, higher Crash Modification Factor values for KABCO and KABC crashes might have been attributed to the type of median and might have less to do with the inside lane width. Having higher values of Crash Modification Factors for KABCO and KABC crashes (total crashes) on roads with a two-way left-turn lane is consistent with studies by Mauga and Kaseko (2010) and 15 studies reviewed by Gluck et al. (1999). These studies reported crash reduction that range from 3% and 57% for KABCO crashes on roads with raised median in comparison to segments with two-way left-turn lane. Mauga and Kaseko (2010) also found a decrease of 21% on KABC crashes for roads with raised median in comparison to those with two-way left-turn lane.

TABLE 4 Comparison of CMFs for 4D and 5T When Inside Lane Width is Fixed to 11 ft While Outside Lane Width Varies

[4D CMF]

(5T CMF)

Ratio (5T CMF)/(4D CMF)

Outside lane width range (ft) 11.8-12.2 12.3-12.7 12.8-13.2 13.3-13.7 13.8-14.2 14.3-14.7
Outside lane width (ft) 12 12.5 13 13.5 14 14.5
CMF for KABCO crashes [1.00] [0.84] [0.70] [0.58] [0.49] [0.41]
-1.88 -1.4 -1.04 -0.77 -0.58 -0.43
1.88 1.67 1.49 1.32 1.18 1.05
CMF for KABC crashes [1.00] [0.86] [0.73] [0.63] [0.54] [0.64]
-2.12 -1.65 -1.28 -1 -0.78 -0.61
2.12 1.92 1.75 1.59 1.44 0.95
CMF for PDO crashes [1.00] [0.83] [0.69] [0.57] [0.48] [0.40]
-1 -0.81 -0.66 -0.54 -0.44 -0.36
1 0.98 0.96 0.95 0.92 0.9

Table 4 also shows the ratio between the Crash Modification Factors developed for four-lane with a raised median and four-lane with a twoway left-turn lane segments with a fixed inside lane of 11 ft while outside lane width varied from 12.5 ft to 14.5 ft. The results revealed that with respect to KABCO crashes, the Crash Modification Factor for four-lane with a twoway left-turn lane segments, when the inside lane width is 11 ft and the outside lane width is 12 ft is 1.88 times that of four-lane with a raised median segments. The ratio decreases as the outside lane width increases from 12.5 ft to 14.5ft, where the four-lane with a two-way left-turn lane CMF is 1.05 times that of fourlane with a raised median segments. A similar trend was observed for KABC crashes as the ratio decreased from 2.12 to 0.95 as the outside lane width increased from 12 ft to 14.5 ft while keeping the inside lane width constant at 11 ft. As can be seen in table 4, for PDO crashes, the ratio of Crash Modification Factors for fourlane with a raised median segments to Crash Modification Factors for four-lane with a twoway left-turn lane segments is smaller than 1.0, indicating that for PDO crashes, a higher crash reduction is expected for four-lane with a two-way left-turn lane segments than for four-lane with a raised median segment when the outside lane width is widened while keeping the inside lane fixed at 11 ft.

When comparing a typical 12 ft inside and a 12 ft outside through lane width segment (a total of 24 ft) with an asymmetric segment of an 11 ft inside lane and a 13 ft outside through lane (also, a total of 24 ft), the results in table 4 show that a four-lane with a raised median asymmetric lane configuration would result in fewer crashes (See highlighted cells—CMFs of 0.70, 0.73, and 0.69 for KABCO, KABC, and PDO crashes, respectively). For four-lane with a raised median configurations, given a total of 24 ft pavement width for both lanes, the results presented in table 4 indicate that restriping a roadway 12 ft to an 11 ft inside and a 13 ft outside through lane would result in a decrease in crashes. For four-lane with a two-way left-turn lane sections, the results are mixed, showing a slight increase for KABCO and KABC crashes (CMFs of 1.04 and 1.28, respectively) and a reduction of PDO crashes (CMF of 0.66), when a typical roadway is retrofitted to an 11 ft inside and a 13 ft outside through lane, respectively. The results also show that as the width of outside lane increases, for both four-lane with a raised median and four-lane with two-way left-turn lane configurations, crashes decrease.

Recommendations for Further Study

This study is not without limitations. The most preferred methods for developing Crash Modification Factors are controlled experiments and observational before-and-after studies. This study used a cross-sectional method. A beforeand-after method would have given more robust results but was not practical or feasible as exact dates when standard 12 ft lanes were retrofitted to create asymmetric lanes could not be obtained.

The results of this study are not without bias. The Highway Safety Manual protocol calls for use of homogeneous segments for obtaining crash modification factors. Hence segments tend to be shorter, rendering a small number of crashes per segment, potentially causing higher dispersion of data. Although sites were selected randomly from around the state of Florida, many potential sites were dropped from analysis because there was no homogeneous comparison sites, i.e., sites with similar variables except for a few variables considered in the model.

Lane width Crash Modification Factors for urban roadways do not exist. Therefore, there were no existing Crash Modification Factor equations to compare the results with. The robustness of CMFs developed by statistical modeling is improved by using homogeneous sites, i.e., sites with similar properties, whereas the only variables are AADT, segment length, and the treatment variable, for this case, lane width. This was not practical as it was not possible to get sufficient segments with similar properties such as the posted speed limit, median opening density, and driveway density. Also, due to limited data, area type was not considered as a variable. A much wider study is recommended, which will develop lane width separate Crash Modification Factors for residential, industrial, and central business district areas. This study did not model the effect of truck percentage due to lack of accurate data for truck traffic at studied sites. Future studies should consider truck percentage as it might have significant contribution to crash occurrences. Last but not least, further research is needed to calibrate the developed Crash Modification Factors to make them useful elsewhere other than Florida.

References

Agrawal, R. and D. Lord. Effects of Sample Size on Goodness-of Fit Statistic and Confidence Intervals of Crash Prediction Models Subjected to Low Sample Mean Values. In Transportation Research Record: Journal of the Transportation Research Board, No. 1950, Transportation Research Board of the National Academies, Washington, D.C., 2006, pp. 35-43.

Bonneson. A. J., and. P. M. Pratt. A Procedure for Developing Accident Modification Factors from Cross Section Data. Paper No. 08-0323. Present at the 87th Annual Meeting of the Transportation Research Board, Washington, D.C., January, 2008.

Gluck, J., H. S. Levinson, and V. Stover. Impacts of Access Management Techniques. NCHRP Report 420: Transportation Research Board, National Research Council,Washington, D.C., 1999.

Griffin, L. I., and K. K. Mak. The Benefits to Be Achieved from Widening Rural, Two-Lane Farm-to-Market Roads in Texas. Report No. IAC (86-87)-1039, Texas, Transportation Institute, College Station, TX., 1987.

Harkey, D.L., S. Raghavan, B. Jongdea, F.M. Council, K. Eccles, N. Lefler, F. Gross, B. Persaud, C. Lyon, E. Hauer, and J. Bonneson. Crash Reduction Factors for Traffic Engineering and ITS Improvement. NCHRP Report No. 617, Transportation Research Board, Washington, DC., 2008.

Harwood, D.W., E.R.K. Rabbani, K. R. Richard, H. W. McGee, and G. L. Gittings. Systemwide Impact of Safety and Traffic Operations Design Decisions for 3R Projects. NCHRP Report No 486, Transportation Research Board of the National Academies, Washington, D.C., 2003.

Harwood, D.W., F. M. Council, E. Hauer, W. E. Hughes, and A. Vogt. Prediction of the Expected Safety Performance of Rural Two-Lane Highways. Report No. FHWA-RD-99-207. Federal Highway Administration, U.S. Department of Transportation, December 2000.

Highway Safety Manual. Transportation Research Board of the National Academies, 2010.

Lord, D., and J. Bonneson. Development of Accident Modification Factors for Rural Frontage Road Segments in Texas. In Transportation Research Record: Journal of the Transportation Research Board, No. 2023, Transportation Research Board of the National Academies, Washing-ton, D.C., 2007, pp. 20-27.

Mauga, T., and M. Kaseko. Modeling and Evaluating Safety Impacts of Access Management Features in the Las Vegas, Nevada, Valley. In Transportation Research Record: Journal of the Transportation Research Board, No. 2171, Transportation Research Board of the National Academies, Washington, D.C., 2010, pp. 57-65.

Potts, I., Harwood, D., and K. Richard. Relationship of Lane Width to Safety for Urban and Suburban Arterials. In Transportation Research Record: Journal of the Transportation Research Board, No. 2023, Transportation Research Board of the National Academies, Washington, D.C., 2007, pp. 63-82.

Zegeer, C.V., D.W. Reinfurt, J. Hummer, L. Herf, and W. Hunter. Safety Effects of Cross-Section Design for Two-Lane Roads. Publication FHWA-RD-87-008. FHWA, U.S. Department of Transportation, 1987.

Updated: Friday, May 19, 2017