Skip to contentUnited States Department of Transportation - Federal Highway Administration FHWA Home
Research Home
Report
This report is an archived publication and may contain dated technical, contact, and link information
Publication Number: FHWA-HRT-09-039
Date: April 2010

Pavement Marking Demonstration Project: State of Alaska and State of Tennessee-Report to Congress

Chapter 2. Safety Impact Assessment of Wider Pavement Markings

Previous | Table of Contents | Next

Longitudinal pavement markings provide a continuous amount of information to drivers by enabling them to safely select the appropriate lane and maintain the appropriate lane position. This is true in both day and night conditions. It is believed that increasing marking visibility will better enable drivers to maintain the appropriate lane position, resulting in an improvement in safety. In recent years, the use of wider pavement markings is one method by which transportation engineers have been trying to increase safety, as it is believed that wider pavement markings benefit drivers by increasing the visibility of the pavement markings.

The MUTCD defines the purpose of longitudinal pavement markings as the delineation of the vehicle path along the roadway. Variations in longitudinal markings are achieved by altering the color, pattern, and width, which all contribute to identifying the proper path for a driver.(3) It should be noted that while the MUTCD defines standard longitudinal pavement markings as having a width of 4–6 inches, for this report, any pavement markings that are wider than 4 inches are considered wider pavement markings.

Across the United States, the use of 4-inch markings is the basic application, and wider lines are used when deemed necessary. As part of a study conducted by Hawkins and Gates in 2001, the results from a nationwide survey indicated that 58 percent (29 States) used wider pavement markings to some degree.(10) All 50 States responded to this survey, providing a solid baseline for establishing usage. The survey results also indicated that the various States’ primary reasons for using markings wider than 4 inches were to improve visibility and thereby improve safety.

The 2001 study also found that there was limited research on the safety effects of using a 6-inch-wide pavement marking versus using the standard 4-inch-wide pavement marking.(10) The existing research did not provide conclusive results on the benefits of wider markings, and the results of various studies often conflicted. Despite these inconclusive findings, a 2007 statewide survey conducted as part of this study shows that the use of wider pavement markings is on the rise.

Introduction

For this effort, the safety aspect of wider lines was addressed using a dual approach, including a multistate retrospective crash study focusing on pavement marking width and a crash surrogate study conducted in Tennessee.

The retrospective crash study included a national survey of wider marking practices used to identify States that knew where and when they had installed wider markings. Crash data from those States were pooled to conduct a robust statistical analysis of the safety impacts of wider markings.

The crash surrogate study focused on the operational aspects (e.g., change in deceleration profiles approaching and transiting curves, change in mean speed, change in speed variability, mean lateral placement, and lateral placement variability) of vehicles when negotiating horizontal curves on two-lane highways that were marked with 4- and 6-inch pavement marking edge lines. The dual approach provided a comprehensive analysis on the effectiveness of wider lines with the intent of developing conclusive results.

Multistate Retrospective Analysis of Wider Edge lines

This section summarizes the safety analysis efforts associated with various pavement marking widths on rural two-lane highways. A general description of the data collection approach is provided, followed by the results of two analyses of the data. The two analyses are a cross sectional safety comparison between rural two-lane segments with 5- and 4-inch edge lines and a before-after analysis of rural two-lane segments on which the edge line width was changed from 4 to 6 inches.

Data

An electronic survey was distributed to identify States that installed pavement markings wider than 4 inches on all or some of their State-owned highways. It was sent through several media, including the following:

  • A list of State transportation agency representatives which was manually developed using rosters for the AASHTO Subcommittee on Safety Management and the Subcommittee on Traffic Engineering, as well as other research team contacts with pavement marking responsibilities.
  • A listserv for the AASHTO Subcommittee on Traffic Engineering.
  • A listserv for the Institute of Transportation Engineers Traffic Engineering.
  • A listserv for the National Committee on Uniform Traffic Control Devices Markings Technical Committee.
  • A listserv for the Transportation Research Board Traffic Control Devices Committee.

Several rounds of follow-up telephone calls were made to those States that were identified as having current or previous experience with wider lines. State traffic engineers, district traffic engineers, maintenance engineers, and staff from other safety-related agency branches were contacted to determine the following:

  • Whether locations of the wider lines could be determined (by route number and linear reference).
  • Whether the use of wider lines was extensive on roadway segments (i.e., not spot treatments).
  • Whether approximate dates of wider line installations were known.
  • Whether sufficient crash, traffic, and roadway databases existed in formats that could be merged with each other and pavement marking information.

The convergence of affirmative answers in all four areas was rare. Required data were most readily available in Illinois and Michigan.

Illinois

Illinois has varying pavement marking practices across its nine districts. The minimum line width in district 6 is 5 inches. This width includes edge lines on both sides of the traveled way, skip lines, and other types of centerline markings. In district 3, edge lines and centerlines are 4 inches, while white skip lines and yellow skip lines on two-lane highways are 6 inches. The pavement marking practices date back 15+ years before the availability of reliable crash and roadway data for a before-after analysis. A cross sectional analysis approach is possible using more current crash, traffic, and roadway data. Additional detail is provided in the analysis section below.

Illinois is a participating State in the Highway Safety Information System (HSIS). The HSIS is a multistate database managed by the University of North Carolina Highway Safety Research Center and Lendis Corporation, under contract with the FHWA. Participating HSIS States were selected based on their data quality and the ability to merge electronically coded crash-related and highway infrastructure-related files. The HSIS database is often the first data alternative for highway safety research with national sponsorship and geometric design components, including research efforts associated with production of the Highway Safety Manual and SafetyAnalyst.(11)

Illinois crash and roadway inventory files were obtained from HSIS from 2001 through 2006. Crashes were located by county, route number, and milepost, while roadway segments were defined by county, route number, beginning milepost, and ending milepost. Crashes were assigned to appropriate roadway segments and counted using a variation of a Statistical Analysis Software® (SAS) code provided by the HSIS lab manager. Over 115 different crash type variations were originally counted. The number was reduced to the following 14 types after a number of preliminary model estimation runs and research team decisions related to the most relevant crash counts for this analysis:

  • Total number of crashes.
  • Total number of fatal plus injury (F + I) crashes.
  • Total number of property damage only (PDO) crashes.
  • Total number of day crashes.
  • Total number of night crashes.
  • Total number of F + I crashes during the day.
  • Total number of F + I crashes during the night.
  • Total number of wet weather crashes.
  • Total number of crashes during wet weather at night.
  • Total number of single vehicle crashes.
  • Total number of single vehicle crashes in wet weather conditions.
  • Total number of crashes with at least one driver 55 years old or older.
  • Total number of opposite direction crashes (includes opposite direction sideswipe and head-on collisions).
  • Total number of fixed object crashes.

Roadway segments and associated crash counts for rural two-lane highways were identified using area type and roadway classification indicators. Rural two-lane segments coded with presence of traffic signals, stop signs, or yield signs were deleted from the database to minimize the influence of intersection presence on the analysis. Additional segments coded as having extremely short segment lengths or atypical rural two-lane highway features (e.g., medians, auxiliary lanes, etc.) were also eliminated. Finally, segments that showed any change in physical features during the observation period (2001–2006) were deleted to minimize the influence of any major reconstruction project on the analysis results. The final rural two-lane dataset for Illinois consisted of 3,439 segments (1,581.1 mi)—2,810 segments (1,321.4 mi) with 4-inch edge lines and 629 segments (259.7 mi) with 5-inch edge lines. Six years of data (2001–2006) were available for each segment. Descriptive statistics for the primary segment variables considered in the analysis are summarized in Table 1 and Table 2.

Table 1. Descriptive statistics for continuous Illinois segment variables.

Segment Variable

2,810 Segments with
4-Inch Edge Lines

629 Segments with
5-Inch Edge Lines

Minimum

Maximum

Average

Minimum

Maximum

Average

Length (mi)

0.12

5.45

0.47

0.12

2.51

0.41

Average daily traffic (vehicles per day)

100

25,900

3,300

100

11,100

2,180

Daily commercial traffic (trucks per day)

0

4,500

390

0

1,000

260

Lane width (ft)

8

16

11.7

9

16

11.5

Shoulder width (ft)

0

14

6.5

0

12

5.9

Paved shoulder width (ft)

0

12

3.7

0

12

4.3

Table 2. Descriptive statistics for categorical Illinois segment variables.

Segment Variable

2,810 Segments with
4-Inch Edge Lines

629 Segments with
5-Inch Edge Lines

Frequency

Percent

Frequency

Percent

25 mi/h posted speed

1

< 0.1

1

0.2

30 mi/h posted speed

43

1.5

16

2.5

35 mi/h posted speed

80

2.8

27

4.3

40 mi/h posted speed

72

2.6

14

2.2

45 mi/h posted speed

116

4.1

34

5.4

50 mi/h posted speed

76

2.7

8

1.3

55 mi/h posted speed

2,422

86.2

529

84.1

Presence of horizontal curve
sharper than 2.5 degrees

223

7.9

44

7.0

Michigan

Edge lines in Michigan are currently 6 inches wide on all State-owned roadways (except for those with curbs and gutters). The change was made from 4-inch edge lines on almost all of the State-owned systems during 2004. A Michigan Department of Transportation (MDOT) pavement marking engineer estimated that 6-inch lines were installed on 95 percent of applicable mileage in 2004, with the remainder installed in early 2005. A before-after analysis was possible with the timing of the change. The widespread switch from 4- to 6-inch edge lines minimized the concern of selection bias or regression to the mean. However, it also did not allow a before-after analysis using comparison sites within the same State. The research team examined several comparison site alternatives. Additional detail is provided in the analysis section below.

Michigan crash data for 2001–2006 were obtained from the Michigan State Police Traffic Crash Reporting Unit. MDOT provided roadway inventory files for those same years. Crashes were located by county, route number, physical reference number, and milepost. Roadway segments were defined by county, route number, physical reference number, beginning milepost, and ending milepost. Crashes were assigned to appropriate roadway segments and counted using SAS®. Counts for 12 of the 14 crash types available for Illinois were also available for Michigan data analysis. Crash type 14—the total numbers of fixed object crashes—were not available, and for crash type 12—total number of crashes with at least one driver 55 years old or older—the change in number of older drivers from the before to the after period was not known. A count for total number of single vehicle crashes during night was included in the Michigan data, making a total of 13 crash types analyzed for Michigan.

Roadway segments and associated crash counts for rural two-lane highways were identified using an area type indicator and a variable for total number of through lanes. Similar data screening techniques and criteria as those employed for Illinois data were used for Michigan, including those for intersections, atypical rural two-lane highway features, and observed changes in physical features during the observation period. The final rural two-lane dataset for Michigan consisted of 253 segments (851.5 mi). Each segment was observed for 3 years from 2001–2003 with 4-inch lines and for 2 years from 2005–2006 with 6-inch lines. Descriptive statistics for the primary segment variables considered in the analysis are summarized in Table 3 and Table 4.

Table 3. Descriptive statistics for continuous Michigan segment variables.

Segment Variable

Minimum Segments

Maximum Segments

Average Segments

Length (mi)

0.04

12.69

3.37

Average daily traffic before period

197

17,633

4,497

Average daily traffic after period

299

18,597

4,433

Daily commercial traffic (trucks per day)

20

2,100

360

Lane width (ft)

10

12

11.5

Shoulder width (ft)

3

12

8.1

Paved shoulder width (ft)

0

11

4.2

Table 4. Descriptive statistics for categorical Michigan segment variables.

Segment Variable

Frequency

Percent

25 mi/h posted speed

5

2.0

30 mi/h posted speed

1

0.4

35 mi/h posted speed

4

1.6

40 mi/h posted speed

3

1.2

45 mi/h posted speed

10

4.0

50 mi/h posted speed

4

1.6

55 mi/h posted speed

226

89.3

Level terrain

165

65.2

Rolling terrain

88

34.8

Analysis

Two types of analyses of Illinois and Michigan data were conducted. The first was a cross sectional safety comparison of rural two-lane segments with 5-inch edge lines to similar segments with 4-inch edge lines in Illinois. The second was a before-after analysis of rural two-lane segments in Michigan in which the edge line width was changed from 4 inches to 6 inches in 2004.

Analysis of Illinois Rural Two-Lane Highway Crash Data

In Illinois, data screening reduced the rural two-lane data set to 3,439 segments (1,581.1 mi), consisting of 2,810 segments (1,321.4 mi) with 4-inch edge lines and 629 segments (259.7 mi) with 5-inch edge lines. Crashes occurring at the segments with 4-inch edge lines were compared to crashes occurring at the segments with 5-inch edge lines. The types of crashes analyzed are listed in Table 5. The table shows the average crash rates computed as crashes per million vehicle miles of travel averaged over the segments considered in the study for Illinois rural two-lane highways. It is categorized by edge line width.

Table 5. Average crash rate (in million entering vehicles) per 1-mi segment of each roadway type.

Crash Type

4 Inches

5 Inches

Total

1.76

1.86

F + I

0.44

0.33

PDO

1.32

1.53

Daytime

0.74

0.64

Nighttime

0.87

0.98

Daytime F + I

0.26

0.19

Nighttime F + I

0.15

0.13

Wet

0.19

0.14

Wet night

0.10

0.08

Single vehicle

1.31

1.55

Single vehicle wet

0.14

0.12

Single vehicle night

0.79

0.94

Older driver (55 years old or older)

0.40

0.38

Opposite direction

0.04

0.05

Fixed object

0.34

0.30

The crash rates shown in Table 5 might be useful if all the segments included in the study are identical except for edge line width, segment length, and annual average daily traffic (AADT) and also if crashes increase linearly with AADT. However, the road segments were different not only in edge line width, segment length, and AADT, but also in other roadway characteristics such as lane width, shoulder width, presence of curves, etc. Also, the relationship between crashes and AADT was not necessarily linear. As a result, the effects of edge line width may not have been estimated correctly by the differences in simple crash rates between 4- and 5-inch edge line segments.

In order to separate out the effect of edge line width from other important roadway characteristics, a negative binomial regression model was developed from the data. The general form of the expected number of crashes in a negative binomial regression model can be given as follows in Figure 1:

Figure 1.  Equation.  General form of a negative binomial regression.  Mu subscript i equals the exponential of parentheses beta subscript zero plus beta subscript 1 times X subscript 1i plus beta subscript 2 times X subscript 2i and continuing in similar fashion to beta subscript k times X subscript ki end parentheses.

Figure 1. Equation. General form of negative binominal regression.

Where Mu subscript i is the expected number of crashes at segment i, X1i, …, Xki are the covariates/predictors corresponding to roadway characteristics of segment i, and beta subscript zero, beta subscript 1, beta subscript 2,…, beta subscript k are the regression coefficients. A model that included edge line width, lane width, shoulder width, presence of horizontal curve (1: present, 0: not present), and log of AADT as predictors and the log of the segment length as an offset variable provided the closest fit to the Illinois data.

Table 6 shows the estimates of the negative binomial regression model coefficients. The regression coefficient for edge line width was negative and statistically significant at Alpha = 0.05, which indicates a positive safety effect of wider edge lines (i.e., a smaller number of crashes is associated with wider edge lines) for the following crash types: F + I (-0.3555), daytime (-0.1710), daytime F + I (-0.3684), nighttime F + I (-0.2900), wet (-0.2953), single vehicle wet (-0.2560), and fixed object crashes -0.2808). Note that an Alpha = 0.05 indicates that there is a 95 percent probability that the observed differences are not due to chance. It can also be observed that the signs of the coefficients for lane width, shoulder width, log of AADT, and curve presence are consistent with intuition. For example, the negative signs of lane width and shoulder width coefficients imply that crashes tend to decrease as lane width or shoulder width increases, and the positive sign of curve presence implies that crashes tend to increase when there is a curve or curves as compared to when there is no curve.

Table 6. Estimates of regression coefficients of the negative binomial regression model applied to Illinois rural two-lane highway crash data for 6 years (2001–2006).

Crash Type Intercept Edge Line Width Lane Width Shoulder Width Log AADT Curve Presence Dispersion Pearson
Chi-Square/
Degrees of Freedom

Total

-5.3007

-0.0398

-0.0675

-0.0133

0.8645

0.2521

0.4288

1.3101

F + I

-5.9759

-0.3555

-0.0882

-0.0417

0.9748

0.6070

0.5978

1.2853

PDO

-5.8323

0.0397

-0.0633

-0.0066

0.8458

0.1260

0.4501

1.2267

Daytime

-7.3511

-0.1710

-0.1026

-0.0359

1.1449

0.2547

0.5737

1.4866

Nighttime

-4.8929

-0.0239

-0.0475

-0.0014

0.6752

0.2945

0.4196

1.1137

Daytime F + I

-7.5377

-0.3684

-0.0885

-0.0471

1.1190

0.3579

0.8243

1.3217

Nighttime
F + I

-5.7133

-0.2900

-0.0845

-0.0369

0.7619

0.9276

0.3630

1.0843

Wet

-7.2627

-0.2953

-0.0849

-0.0212

0.9853

0.3638

0.7133

1.1082

Wet night

-6.7358

-0.2458

-0.0552

-0.0023

0.7465

0.4562

0.6720

1.1026

Single vehicle

-3.6780

-0.0196

-0.0403

-0.0076

0.5624

0.3590

0.4031

1.1220

Single vehicle wet

-5.1418

-0.2560

-0.0337

-0.0175

0.5767

0.5359

0.7081

1.0961

Older-driver (55 years-old or older)

-7.4711

-0.0940

-0.0525

-0.0176

0.9571

0.1654

0.5371

1.3095

Opposite direction

-14.7025

0.1768*

-0.1019

-0.0051

1.5046

0.6268

0.3489

1.1148

Fixed object

-5.0044

-0.2808

-0.0216

-0.0651

0.6937

0.6994

0.5051

1.2885

Note: Significant (at Alpha = 0.05) effects are shown in bold. There was an extreme outlier in the opposite direction crash data for a 0.27-mi segment with 5-inch edge lines, which greatly affected an estimate of the edge line width coefficient for opposite direction crashes. When this outlier was removed, the opposite direction coefficient for edge line width changed from 0.3295 to 0.1768 and became insignificant.

For Illinois, raised reflective pavement markers (RRPM) are used statewide, and rumble strips are used on interstates statewide. However, information on additional delineation and guidance measures (other than RRPM and rumble strips) was not available and could not be incorporated into the analysis. Therefore, the above observations are based on the assumption that the effects of the variables not in the database, such as those additional delineation/guidance measures, are the same (or averaged out) for the segments with and without wider edge lines.

Analysis of Michigan Rural Two-Lane Highway Crash Data

In Michigan, changes to 6-inch edge lines occurred in 2004 for about 95 percent of the road segments statewide. Before-after evaluations were conducted with 3 years of before data (2001–2003) and 2 years of after data (2005–2006) obtained from 253 segments corresponding to 851.5 mi of rural two-lane highways. Crashes that occurred during the before period were compared to crashes that occurred during the after period. The types of crashes analyzed can be viewed in Table 7, which shows the average crash rates computed as crashes per million vehicle miles of travel averaged over the segments considered in the study for each of the before and after periods.

Table 7. Average crash rate (in million entering vehicles) per 1-mi segment of Michigan rural two-lane highways for each of before (2001–2003) and after (2005–2006) periods.


Crash Type
Period
Before After

Total

3.06

3.00

F + I

0.44

0.40

PDO

2.63

2.60

Daytime

1.29

1.22

Nighttime

1.41

1.41

Daytime F + I

0.29

0.25

Nighttime F + I

0.12

0.12

Wet

0.28

0.24

Wet night

0.14

0.12

Single vehicle

2.26

2.24

Single vehicle wet

0.21

0.19

Single vehicle night

1.29

1.29

Opposite direction

0.08

0.07

It can be observed from Table 7 that crash rates decreased overall. However, this direct comparison of before-after crash rates is valid only when it can be absolutely assured that there have been no changes from before to after periods other than edge line width and traffic volumes and that the relationship between crashes and traffic volumes is linear. Both of these assumptions are often violated when the crash data of multiple years are analyzed. There will almost always be changes over time in weather, vehicle fleet, driver characteristics, economic conditions, etc., and crashes may increase with traffic volume in a nonlinear fashion.

To distinguish the effect of edge line width from the effects of other factors that might have also changed from the before to the after period, an EB approach for safety evaluation was employed.(12,13) The EB method estimated changes in crashes (due to wider edge lines) by comparing the observed number of after period crashes to the predicted number of crashes during the after period that would have occurred had wider edge lines not been installed, rather than to the observed number of before period crashes. Predicted crash frequencies by the EB method were obtained in such a way that they accounted for a potential nonlinear relationship between crashes and traffic volume (through the regression function called the Safety Performance Function (SPF)) as well as changes in general underlying trend caused by extraneous factors such as weather, vehicle fleet, and driver characteristics between the before and after periods. The SPF, which describes the relationship between crashes and traffic volume as well as other roadway characteristic variables such as lane width, shoulder width, and terrain, was derived from the before period from the Michigan data. The changes in general trend would typically have been estimated based on crash counts from road segments on which edge line width remained at 4 inches throughout the study period. Because no such segments remained in Michigan due to statewide installation of 6-inch edge lines during the study period, an alternative approach of deriving the trend factor based on another entity set was taken in which the general trend between the before and after periods was derived from the Illinois F + I crash data obtained from rural two-lane segments with 4-inch edge lines.(13) Using the Illinois data to provide a comparison group yielded results that were comparable to the cross sectional analysis conducted with the Illinois data. Additional analyses are being conducted to further verify this approach.

Table 8 presents the results of EB before-after evaluations based on the crash data in Michigan from 253 segments (851.5 mi) of rural two-lane highways. The observed number of after crashes over the segments, the predicted number of crashes during the after period that would have occurred without installing wider edge lines, and an estimate of the percent change in crashes from the before to the after period are shown in the table. As can be observed from the table, the EB before-after evaluations (using the before period Michigan data to develop the SPFs and the Illinois F + I crash data obtained from segments with 4-inch edge lines to derive a trend between the before and after periods) resulted in the following crash reduction estimates for rural two-lane highways in Michigan:

  • Total crashes: 7.1 percent.
  • F + I crashes: 17.1 percent.
  • PDO crashes: 5.4 percent.
  • Daytime crashes: 10.0 percent.
  • Nighttime crashes: 2.4 percent.
  • Daytime F + I crashes: 18.0 percent.
  • Nighttime F + I crashes: 11.7 percent.
  • Wet crashes: 24.4 percent.
  • Wet night crashes: 22.6 percent.
  • Single vehicle crashes: 2.0 percent.
  • Single vehicle wet crashes: 20.0 percent.
  • Single vehicle night crashes: -0.2 percent.
  • Opposite direction crashes: 14.9 percent.

All of these crash reduction estimates, except for nighttime, single-vehicle, and single-vehicle night crashes, were statistically significant at the 95 percent level.

Table 8. Results of EB before-after safety evaluations based on Michigan crash data with 3 years (2001–2003) of before and 2 years (2005–2006) of after data.

Crash Type Observed
After Crashes
Predicted
After Crashes with
4-Inch Edge Lines
Percent Reduction in Crashes

Total

6,077

6,541.2

7.1

F + I

811

977.5

17.1

PDO

5,266

5,563.1

5.4

Day

2,231

2,478.6

10.0

Night

3,149

3,277.4

2.4

Daytime F + I

498

607.1

18.0

Nighttime F + I

257

291.0

11.7

Wet

459

607.1

24.4

Wet night

243

313.7

22.6

Single vehicle

4,862

4,962.86

2.0

Single vehicle wet

353

440.691

20.0

Single vehicle night

2,923

2,916.34

-0.2

Opposite direction

165

193.8

14.9

Note: Statistically significant results (at 95 percent confidence level) are shown in bold.

Crash Surrogate Study

The crash surrogate study was designed to detect possible operational impacts of 4-inch versus 6 inch pavement marking edge lines on horizontal curves on RTLTW undivided highways. Three curve site selection criteria (curve radius, posted speed limit, and presence of paved shoulder) were identified through the literature review and team discussions as having the greatest potential impact on the effectiveness of wider edge lines. The crash surrogate study employed a before-and-after technique to reduce site-to-site variability using operational measures of effectiveness as surrogates for crashes. It was assumed that driver-to-driver (or vehicle-to-vehicle) variability would be less than variability caused by installation of wider lines. The literature review, combined with the expert opinion of the research team, led to the decision to study the impacts of wider pavement markings on horizontal curves exclusively. The operational measures of effectiveness that were studied included the following:

  • Change in deceleration profiles approaching and negotiating the curve.
  • Change in mean speed.
  • Change in speed variability.
  • Change in mean lateral placement.
  • Change in lateral placement variability.

Even with a before-and-after technique, there is the possibility that some uncontrolled extraneous factor may impact the data; hence, the research team chose to have comparison sites. Comparison sites are curves that have similar geometric and traffic flow characteristics to the treatment site curves and where the pavement marking width is left unchanged between the before and after periods. Use of comparison sites helped ensure internal validity of the study by reducing confounding between the effect of treatment and the effects of uncontrollable extraneous variables. Examples of uncontrollable extraneous variables in this measure of effectiveness study might have included changes in drivers, driver behavior, and observers between the before and after periods.

Study Site Selection

Based on a review of the literature regarding safety problem areas, all horizontal curve test sites were established on RTLTW highways. Approximately 60 potential sites within Tennessee were visited to assess the geometric and operational characteristics of the candidate curves (see Table 9).

Table 9. Safety-related controls for curve study.

Geometric

Operational

  • Lane width (10–12 ft).
  • Grade ( ≤ 4 percent).
  • Approach tangent length ( ≥ 0.25 mi).
  • Curve length (vehicle time in curve, t ≥ 3 s).
  • Ambient lighting (none).
  • Vehicle headway ( ≥ 5 s).
  • On-coming vehicles (none).
  • Approach speeds ( ≥ posted speed limit minus 10 mi/h).
  • Curve speeds ( ≥ posted advisory speed minus 10 mi/h).

As a result of these site visits, the researchers recommended that a total of 19 horizontal curves should be studied in Tennessee, with 10 treatment sites and 9 comparison sites. The black dots in Figure 2 represent the location of the 19 horizontal curve study sites. The researchers verified that no roadway improvements were planned for the 19 study sites for the duration of the study. While efforts were made to select only isolated horizontal curves, two of the horizontal curves were located within winding roadway segments. The speed limit along the winding roadway segments was 35 mi/h, so it was believed that the speed limit had greater influence on the approach speeds than the alignment.

Figure 2.  Chart.  Map of 19 curve study sites.  A regional map of Tennessee centered on Nashville identifies the locations of the 19 curve study sites; dots indicate the study sites.

Figure 2. Chart. Map of 19 curve study sites.

The researchers categorized the horizontal curves based on three factors that have been identified through the literature review and team discussions as having the greatest potential impact on the effectiveness of wider edge lines. The sites were selected based on the radius of the curve (two levels), the posted speed limit (two levels), and the presence of a paved shoulder (two levels). The study matrix that includes two by two by two levels of those factors is shown in Table 10. The curves were split into the treatment and comparison sites in such a way as to have comparisons for each combination of selection criteria. Note that sites for one of the eight combinations could not be identified.

Table 10. Study site matrix.

Speed Limit Curve Design Safety Rating1

Radius ≤ 700 ft

(Degree of Curvature ≥ ~8.0)

Radius ≥ 800 ft

(Degree of Curvature ≤ ~7.0)

Presence of Paved Shoulder2 Presence of Paved Shoulder2
Yes No Yes No

55 mi/h

1/1

2/2

2/1

1/1

50 mi/h

0

2/2

1/1

1/1

1 2/1 indicates that there will be at least two treatment sites and one comparison site for each category.

2 For this project, presence of a paved shoulder exists when there is at least 36 inches of usable pavement beyond the inside edge of the edge line. For this project, absence of a paved shoulder exists when there is less than or equal to 24 inches of usable pavement beyond the inside edge of the edge line.

Data Collection

Data were collected along the 19 rural horizontal curves using traffic classifiers. The before data collection took place over a 5-week period from August to September 2007, and the after data collection took place over a 5-week period from July through August 2008. Traffic classifiers were installed on a Monday and retrieved on a Thursday in the same week by two to four research team members. Approximately 96 hours of data were collected at each study site for the before and after periods.

During the before data collection period, the curves had 4-inch-wide pavement markings. During the after period, the edge lines were restriped with 6-inch-wide pavement markings along the edge lines but not the centerlines—centerlines were restriped with 4-inch-wide markings. Driver eye scanning studies showed that drivers used the adjacent pavement marking edge line to negotiate curves regardless of whether they were in the inside or outside lane.(14)

Every effort was made to minimize differences between the periods of data collection and pavement marking installations. The average retroreflectivity of the edge lines in the before period was 200 mcd/m2/lx, with none of the sites below 100 mcd/m2/lx, while the average edge line retroreflectivity for the after period was 288 mcd/m2/lx. The pavement markings for the after period were installed in late May 2008. After the pavement markings were installed, at least 1 month was provided to allow drivers to acclimate to the new markings.

Equipment Setup

When a vehicle passed through a particular curve, the traffic classifiers recorded the classification of the vehicle (i.e., passenger car or tractor trailer), the lateral position of the vehicle, and the speed of the vehicle. Piezoelectric road sensors were used in conjunction with traffic classifiers. The traffic classifiers enabled the researchers to collect raw data with a time stamp precision of 0.001 s.

Four traffic classifiers were used at each study site to track the movements of the vehicles traveling through the outside of each horizontal curve. These locations are defined as follows and in Figure 3:

  • Upstream (U) location: Positioned approximately 1,000 ft upstream of the curve warning sign location, this location was adjusted to avoid driveways, cross streets, or other factors (i.e., grade, horizontal curvature) that could impact the data collection effort.
  • Advance curve warning sign location: This location was positioned at the advance curve warning sign (or the location at which a sign would be located when no sign was present). If a wider edge line was installed in the after period, it was started approximately 500 ft in advance of the curve warning sign location.
  • Point of curve (PC) location: This location was positioned at the PC of the horizontal curve of interest. A second traffic classifier was also installed at this location to ascertain if an opposing vehicle passed through the study curve within ±7 s of a study vehicle traveling in the outside lane.
  • Midpoint of curve (MC) location: This location was positioned near the MC of interest.

Figure 3. Illustration. Horizontal curve traffic classifier layout.  The diagram provides identifiers for locations on a horizontal curve and shows the layout of traffic counters used for speed and lateral placement data collection.  Abbreviations on the diagram used for location identifiers are denoted as follows:  ST indicates the starting point of the edge line treatment, which is approximately 500 ft prior to the advance warning sign; U represents upstream and indicates a location approximately 500 ft upstream of the treatment location; W represents the location at the advanced curve warning sign; PC indicates the location at the point of curve; and MC indicates the midpoint of the curve. Traffic counters are located at the positions identified (from left to right on the curve) as MC, PC, W, ST, and U.

Figure 3. Illustration. Horizontal curve traffic classifier layout.

Sample Size

A power analysis was used to determine the sample size (the number of vehicles, n) needed to detect a practically important minimum difference in effects of increasing the pavement marking width and among the interaction effects between the pavement marking width and the day/night factor at each site. The procedures given in Wheeler, Nelson, and Bratcher et al. were used for the sample size calculation.(15–17) Because the necessary sample size varies with the desired significance level (Alpha), the desired power, the standard deviation (SD; Sigma) of the response variable, and the minimum difference of practical importance (Pressure), those values were predetermined before the sample size calculation. By convention, the desired significance level and the desired power were set to 0.05 and 0.90, respectively. Previous research indicates that the approximate SDs in speed and lateral placement in similar curves to those used in this study are 8 mi/h and 20 inches, respectively.(18) The minimum difference of interest before and after installation of wider lines was determined to be 3 mi/h for the mean speeds and 6 inches for the mean lateral placements based on engineering judgment and previous research.(18,19) It is believed that 6 inches is the minimum change in mean lateral position that would be a practically significant change for at least two reasons: (1) field experience has shown that striping installations vary in width as much as ±0.5 inches, and restriping can be misaligned by more than 1 inch, which may result in wide variability between pavement marking installations; and (2) previous research supported 6 inches.(18) The 3 mi/h minimum difference of interest was selected as a value between the values chosen by previous research because it is believed that a change of 3 mi/h would be the minimum change that would influence changing posted speed limits or advisory speeds.(18,19)

The minimum sample size (nspeed) necessary for detecting a mean speed difference (pressure) of 3 mi/h, with a Sigma in speed of 8 mi/h, before and after installation of wider lines at each site is shown in figure 4, where r is the number of levels of a factor.

Figure 4.  Equation.  Power analysis for sample size to detect a speed difference of 3 mi/h.  n subscript speed equals parentheses 3 times r times sigma divided by delta end parentheses squared.  That quotient equals parentheses 3 times 2 times 8 divided by 3 end parentheses squared equals 256.

Figure 4. Equation. Power analysis for sample size to detect a speed difference of 3 mi/h.

The minimum sample size (nlp) necessary for detecting a mean lateral placement difference (pressure) of 6 inches, with a Sigma of 20 inches, before and after installation of wider lines at each site is shown in figure 5.

Figure 5.  Equation.  Power analysis for sample size to detect a lateral placement difference of 6 inches.  n subscript l p equals parentheses 3 times r times sigma divided by delta end parentheses squared.  That quotient equals parentheses 3 times 2 times 20 divided by 6 end parentheses squared equals 400.

Figure 5. Equation. Power analysis for sample size to detect a lateral placement difference of 6 inches.

The minimum sample size necessary for detecting a mean speed difference of at least 3 mi/h in any two interactions means between pavement marking width and day/night at each site is shown in figure 6. In the figure, ν is the number of interaction degrees of freedom, c is the number of factor-level combinations for the factors that are involved in the interaction, k is the number of factors involved in the interaction, and Deltais the minimum difference of interest among the interaction effects.

Figure 6.  Equation.  Power analysis for sample size to detect a speed difference of 3 mi/h with two interactions.  n subscript speed equals 9 times sigma squared times parentheses nu plus 1 end parentheses times c divided by delta squared times 2 superscript k minus 2, that whole quotient times 1/2  That whole quotient equals 9 times 8 squared parentheses 1 plus 1 end parentheses times 4 divided by 3 squared times 2 superscript 2 minus 2, that whole quotient times 1/2 equals 256.

Figure 6. Equation. Power analysis for sample size to detect a speed difference of 3 mi/h with two interactions.

The minimum sample size necessary for detecting a mean lateral placement difference of at least 6 inches in any two interactions means between pavement marking width and day/night at each site is shown in Figure 7:

Figure 7.  Equation.  Power analysis for sample size to detect a lateral placement difference of 6 inches with two interactions.  n subscript l p equals 9 times sigma squared times parentheses nu plus 1 end parentheses times c divided by delta squared times 2 superscript k minus 2 times, that whole quotient times 1/2.  That quotient equals 9 times 20 squared parentheses 1 plus 1 end parentheses times 4 divided by 6 squared times 2 superscript 2 minus 2, that whole quotient times 1/2 equals 400.

Figure 7. Equation. Power analysis for sample size to detect a lateral placement difference of 6 inches with two interactions.

A sample size of 400 vehicles was selected to assure the power of the tests to be at least 0.90 for both mean speed difference and the mean lateral placement difference. Thus, the desired number of vehicles to be observed for each daytime and nighttime condition and for each before and after installation of wider lines at each site is at least 100 vehicles.

Statistical Analysis Methodology

A field experimental before-after study was conducted to compare 4-inch versus 6-inch pavement marking edge lines along isolated RTLTW roads. The researchers collected continuous quantitative data from traffic classifiers. Two primary treatments were studied: (1) curves marked with 4-inch-wide edge lines and (2) curves marked with 6-inch-wide edge lines. Other factors were the posted speed limit, the curve radius, the shoulder width, and the period of the day. The dependent variables were vehicle speed and vehicle lateral placement. The changes in mean speed, speed variance, 85th percentile speed, mean lateral position, and lateral position variance before and after installation of wider edge lines were the main interests of the study. In addition, the mean differences in the speed and lateral position between the different traffic classifier locations were also investigated, such as between the data collected at the PC and the MC. Evaluation criteria included the following:

  • Change in mean speed at each traffic counter location.
  • Change in speed variance at each traffic counter location.
  • Change in 85th percentile speed at each traffic counter location.
  • Change in mean lateral position at each traffic counter location.
  • Change in lateral position variance at each traffic counter location.
  • Mean difference in speed between traffic counter locations (i.e., between the PC and the MC counter locations).
  • Mean difference in lateral placement between traffic counter locations (i.e., between the PC and the MC counter locations).

The statistical analyses included descriptive statistics, graphical analysis, and hypothesis testing. The descriptive statistics calculations included minimums, maximums, ranges, means, medians, quartiles, and 85th percentile values. Boxplots, histograms, scatter plots, and cumulative distributions were used to investigate the distribution of the data and to identify any trends or outliers in the data that would impact the testing methods used to conduct the hypothesis testing. The analysis of variance, specifically a split-plot design analysis, was used to test equality of mean speed and equality of mean lateral position of vehicles before and after the installation of wider edge lines.

Analysis

The descriptive statistics are separated into several tables. Table 11 contains summary statistics with respect to the sample size. While each study site had ample volume to provide 100 vehicles for each condition, some of the sample sizes for the nighttime data were less than desired once the researchers removed all of the unusable data. Unusable data were defined based on the following criteria:

  • There was an opposing vehicle present.
  • The vehicle in question could not be tracked through the entire system of classifiers.
  • The speed data appeared unreasonable (e.g., the upper threshold was set at 100 mi/h because it was believed that vehicles would not be able to achieve that speed or higher within any of the study sites).
  • The lateral position data were outside the measureable range of the sensor traps (the measureable range was 9.19 ft).
  • The weather was questionable during the period of data collection (only curve 1 in the before condition had weather conditions that warranted the removal of data).

Table 11. Sample size summary.

Statistic

Before

After

Day

Night

Day

Night

Minimum

279

43

613

56

Mean

1,012

113

901

130

Median

890

84

828

100

Maximum

2,770

354

1,403

274


Table 12 shows summary statistics for the general trends. The values were calculated from the difference in the before and after period means and SD values. A positive value for a change in mean lateral placement would mean that drivers in the after period were driving closer to the centerline, while a negative value for the change in SD in lateral placement would indicate that the drivers were more centrally located within their respective lane of travel. Table 29 through Table 32 in appendix A contain the detailed mean and SD values for the speed and lateral position data collected between the before and after periods for all 19 study sites. Other statistics such as range and variance were investigated, but they are not reported herein because they did not enhance the information already provided through the mean and SD. There were no trends that would suggest that the installation of wider edge lines affected a driver’s selection of speed, but it appears that the installation of wider edge line markings in rural curves may have impacted a driver’s selection of lateral position through horizontal curves (with a slight shift toward the centerline once in the curve). However, there were no mean changes of speed that exceeded 3 mi/h or mean changes in lateral position that exceeded 6 inches, which were established as the practical statistically significant differences during the sample size calculations.

Table 12. Change in speed and lateral position statistics for the treatment sites.

Speed Limit Change in Statistical Measure Curve Design Safety Rating
Radius ≤ 700 ft
(Degree of Curvature ≥ ~8.0)
Radius ≥ 800 ft
(Degree of Curvature ≤ ~7.0)
Presence of Paved Shoulder Presence of Paved Shoulder
Yes No Yes No
Speed (mi/h) Lateral Position (inch) Speed (mi/h) Lateral Position (inch) Speed (mi/h) Lateral Position (inch) Speed (mi/h) Lateral Position (inch)

≥ 55 mi/h

Mean

1.6

3.8

-0.1

4.0

0.0

-0.4

0.2

-0.4

SD

0.1

0.0

0.4

-1.9

0.2

-0.4

-0.1

0.4

≤ 50 mi/h

Mean

0.0

0.0

-0.7

2.1

0.7

-1.5

-1.1

0.0

SD

0.0

0.0

0.1

1.6

-0.7

-1.1

-0.2

1.1

Chapter Summary

The retrospective crash analysis based on Illinois and Michigan rural two-lane highway data shows that there are positive safety effects of wider markings for relevant crashes as follows:

  • For Illinois, the negative binomial regression analysis based on the crash data aggregated for 6 years resulted in positive safety effect estimates for F + I, daytime, daytime F + I, nighttime F + I, wet, single vehicle wet, and fixed object crashes.
  • For Michigan, an EB before-after evaluation resulted in positive safety effect estimates for total, F + I, PDO, daytime, daytime F + I, nighttime F + I, wet, wet night, single vehicle wet, and opposite direction crashes.

At the same time, the crash surrogate study results support previous findings, which show that there are either no real vehicle operational impacts or, at most, only subtle vehicle operational impacts as a result of adding or widening edge line markings—even for narrow two-lane highways and day and night conditions.

It should be noted that additional work is being completed. For the retrospective crash analysis, researchers are analyzing the impacts of widening interstate highway markings from 4 to 6 inches. For the crash surrogate study, researchers are conducting a more thorough statistical analysis using multivariate analyses techniques.

 

Previous | Table of Contents | Next

ResearchFHWA
FHWA
United States Department of Transportation - Federal Highway Administration