Safety Study - Update on Multi-State Retrospective Crash Study

Preliminary Illinois Data Analysis

January 2008
The Illinois crash data for years 2000 through 2005, obtained from 20,000 segments corresponding to roughly 5,000 miles for each year, were analyzed to assess the safety effect of wider edge lines. Because the data are cross-sectional in nature, (i.e., all of the roadways in one district (District 6) have wider edge lines while the roadways in other districts have standard edge lines), Before-After analysis could not be performed. Instead, the data were analyzed by applying the Negative Binomial Regression models on the count of crashes of various types while accounting for temporal correlations that may exist in yearly crash counts obtained from the same road segment into the analysis so as not to overestimate the effect of wider edge lines. Intersection related crashes were excluded from the analysis.

Types of crashes analyzed were Total accidents, Fatal and Injury accidents, Night crashes, Wet crashes, Wet night crashes, Accidents involving an older driver (>=55 years old), Night crashes involving an older driver, and Run-Off-the-Road crashes. The Negative Binomial Regression models included as many relevant variables that are available in the database as possible such as Edge line width, Skip line width, Lane width, Shoulder width, Median width, Area type indicator (1= Rural, 2=Urban), Access control (0=uncontrolled, 1: partial control, 2: full control), Log of AADT, Log of Heavy commercial vehicle volume, Number of lanes, and Number of special lanes, as predictors and the segment length as an offset variable.

The regression coefficient of Edge line width was negative and statistically significant, which indicates a positive safety effect of wider edge lines (i.e., a smaller number of crashes is associated with wider edge lines) for most crash types analyzed, i.e., for Total accidents, Fatal and Injury accidents, Night crashes, Wet crashes, Wet night crashes, and Run-Off-the-Road crashes. For Wet crashes, Night crashes, Wet night crashes, and Run-Off-the-Road crashes, the coefficients of Edge Line width were larger in absolute value than that for Total accidents, which suggests that the effects of wider edge lines are larger for those relevant crashes. The signs of regression coefficients for other predictors were also consistent with intuition, i.e., Lane width, Shoulder width, Median Width, Access control, and Number of special lines have statistically significant negative coefficients and Log of AADT and Log of Heavy commercial vehicle volume have a statistically significant positive coefficient.

These findings are preliminary and are subject to change upon further analyses of additional state data. Specifically, it needs to be noted that the information on additional delineation and guidance measures as well as the roadway curvature information were not available for the Illinois data, 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 and the roadway curvature are the same (or averaged out) for the sites with/without wider edge lines.

November 2007
With respect to the retrospective multi-state crash analysis, the 2000 Illinois data have been compiled. Because the data are cross-sectional in nature, (i.e., all of the roadways in one district (District 6) have wider edge lines while the roadways in other districts have standard edge lines), Before-After analysis could not be performed. Instead, the data were analyzed by applying the Negative Binomial Regression models on the count of crashes of various types. Intersection related crashes were excluded from the analysis. Types of crashes analyzed are Total accidents, Fatal and Injury accidents, Night crashes, Wet crashes, Wet night crashes, Accidents involving a older driver (>=55 years old), Night crashes involving a older driver, etc. The Negative Binomial Regression model include Edge line width, Lane width, Access control (0=uncontrolled, 1: partial control, 2: full control), and Log of AADT, as predictors and the segment length as an offset variable. In all of the crash types mentioned above, the regression coefficient of Edge line width was negative and statistically significant, which indicates a positive safety effect of wider edge lines (i.e., a smaller number of crashes is associated with wider edge lines). For Wet crashes, Night crashes, Wet night crashes, and Accidents involving an older driver, the coefficients of Edge Line width were larger in absolute value than that for Total accidents, which suggests that the effects of wider edge lines are larger for those relevant crashes. The signs of regression coefficients for other predictors were also consistent with intuition, i.e., Lane width and Access control have statistically significant negative coefficients and Log of AADT have a statistically significant positive coefficient. It needs to be noted that the information on additional delineation and guidance measures were not available for the Illinois data, and could not be incorporated into the analysis. Therefore, the above observations are based on the assumption that the effects of those additional delineation and guidance measures are the same (or averaged out) for the sites with/without wider edge lines.

October 2007
The research team worked with state agencies and FHWA to obtain roadway, traffic and crash data from selected states. The following progress was made:

• Roadway, crash, vehicle and occupant data were obtained from Illinois (through HSIS) for years 2000 through 2005;
• Roadway and crash data were obtained from Michigan DOT and Michigan State Police for years 2000 through 2006; and
• Crash data were obtained from Kansas DOT for years 2000 through 2007.

• 6 inch skip line and 4 inch edge line: 2016.94 miles (3847 observations);
• 4 inch skip line and 5 inch edge line: 1905.44 miles (4818 observations); and
• 4 inch skip line and 4 inch edge line: 11803.04 miles (31837 observations).

Merging will continue for remaining Illinois years and for Michigan and Kansas. Both Michigan and Kansas data will result in two sets of pavement characteristics:

• 4 inch center line and 4 inch edge line;
• 4 inch center line and 6 inch edge line.

September 2007
State transportation agencies were contacted to determine the availability of wider pavement marking data. Specifically, phone interviews of appropriate central and division/district office personnel were conducted to assess whether records (computerized, paper or memory) of wider pavement marking locations and installation-schedules existed. Discussions with transportation agencies have continued to determine the availability of wider pavement marking data and the ability to merge pavement marking information with crash, roadway and traffic data. Wider pavement marking information was compiled for Michigan, Illinois, Kansas and South Carolina. Michigan roadway and traffic data were obtained for years 2000 through 2006. Illinois roadway, traffic and crash data were obtained for years 2000 through 2005. Phone inquiries were made to 21 Texas districts regarding wider pavement marking practices. The team has confirmed that 4 TxDOT districts are using 6" wide edge lines. We are working on gathering respective location and timing information. Several LISTSERV requests were made to national committees and traffic engineering-related organizations. Responses are being investigated to identify other potential data sources.

August 2007
State transportation agencies were contacted to determine the availability of wider pavement marking data. Specifically, phone interviews of appropriate central and division/district office personnel were conducted to assess whether records (computerized, paper or memory) of wider pavement marking locations and installation-schedules existed. Discussions with transportation agencies from twelve states were initiated (and in some cases concluded) in August:

1. Illinois
2. Kansas
3. Kentucky
4. Massachusetts
5. Maryland
6. Michigan
7. Florida
8. North Carolina
9. South Carolina
10. South Dakota
11. Tennessee
12. Arizona

In many cases, striping or re-striping with wider markings is classified as a maintenance activity. Although general timings of changes in state pavement marking policies are known, the timings of changes on specific roadways are not. Policy changes occurred anywhere from 15 to 30 years ago in Illinois, Kentucky, Maryland, Florida, and Tennessee. Policy changes in Kansas and Michigan are more recent (i.e. 2004-2005). With information gathered from phone discussions to date, data availability and characteristics appear limited to the following:

• Before-after data on all state roadways in Michigan without comparison sites; two years of "after" data available;
• With-without data from Illinois; all roadways in one IL district with wider lines and roadways in other districts without wider lines; the data may be used for comparison groups and estimation of safety performance functions for before-after analysis; and
• Limited (i.e. several sites) of before-after data from Kansas and North Carolina.

Project summaries and meeting minutes from the FHWA Pooled Fund Study on Evaluations of Low-Cost Safety Improvements reference similar challenges regarding locations and dates of improvement installations. A data-entry form for States to record low-cost safety installations has been developed. The TTI project-team will contact the current FHWA representative for the Pooled Fund Study to determine the availability of information potentially useful to the wider lines safety study. Discussions with state transportation agencies will also continue.

Update on Crash Surrogate Study

Safety Study

Durability Study

PM Demo Main Page