Accident Modification Factors for Traffic Engineering and ITS Improvements (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

34 Introduction This chapter includes a detailed description of each AMF verified, modified, or developed in this research effort. Table 18 includes the listing of AMFs and the source of each AMF. The possible sources include the following: • Literature Review. Completed research was discovered and critically reviewed. The assessment revealed that AMFs existed for the given treatment with an LOPC of either high or medium high. • EB Before-After Evaluation. Before-treatment and after- treatment crash data were acquired for locations where the treatment of interest had been installed. The latest statisti- cal methodologies (i.e., EB) for conducting before-after studies were applied to produce AMFs. • Reanalysis of Existing/Supplemental Data. Data from prior before-after evaluations were acquired and reanalyzed using the more rigorous EB methodology. In many cases, supplemental data were acquired to enhance the evaluation. • Analysis-Driven Expert Panel. A panel of knowledgeable researchers and practitioners was convened to review crit- ical research studies and reach a consensus on AMFs for a given treatment. In some cases, the AMF was developed through further analysis by one of the NCHRP project research teams sponsoring the expert panel meeting. • Cross-Sectional Model. A new analysis was conducted in which a cross-sectional model was produced and used to derive AMFs for a specific treatment. AMF Summaries For each AMF listed in Table 18, a summary of the research from which the AMF was developed is given below. Each sum- mary includes the AMF mean estimate(s) with standard errors shown in parentheses, the LOPC, the study methodology, a description of the sites used in the study, and supplemental comments and footnotes to describe the study results and applicability. Table 19 presents a glossary of acronyms used in the AMF summaries. C H A P T E R 5 Compilation of Recommended AMFs

Treatment Source of AMF Intersection Treatments Install Roundabout Literature Review Add Exclusive Left-Turn Lane Literature Review/Expert Panel Add Exclusive Right-Turn Lane Literature Review/Expert Panel Install Traffic Signal at Urban Intersection Literature Review Install Traffic Signal at Rural Intersection EB Before-After Evaluation Remove Traffic Signal (Urban Environment) Literature Review Modify Signal Change Interval Literature Review Prohibit Right Turn on Red Expert Panel/Further Analysis Modify Left-Turn Phase Expert Panel/Reanalysis of Existing/Supplemental Data Replace 8-in. Signal Heads with 12-in. Signal Heads Reanalysis of Existing/Supplemental Data Replace Single Red Signal Head with Dual Red Signal Heads Reanalysis of Existing/Supplemental Data Convert Nighttime Flash Operation to Steady Operation Reanalysis of Existing/Supplemental Data Convert to All-Way Stop Control Literature Review Convert Stop Control to Yield Control Literature Review Install Red-Light Cameras Literature Review Add Intersection Lighting Expert Panel Increase Pavement Friction on Intersection Approach Reanalysis of Existing Data Roadway Segment Treatments Narrow Lane Widths to Add Lanes Literature Review Add Passing Lanes (Two-Lane Roads) Literature Review Add Two-Way Left-Turn Lane (TWLTL) Literature Review/Expert Panel Change Lane Width Literature Review Change Shoulder Width and/or Type Literature Review Flatten Horizontal Curve Literature Review Improve Curve Superelevation Literature Review Add Shoulder Rumble Strips Literature Review Add Centerline Rumble Strips Literature Review Install/Upgrade Guardrail Literature Review Convert Undivided Four-Lane Road to Three-Lane and TWLTL (Road Diet) Reanalysis of Existing Data Increase Pavement Friction on Roadway Segment Reanalysis of Existing Data Change Median Width Cross-Sectional Model Change Roadside Sideslope Expert Panel Add/Remove On-Street Parking Expert Panel Add Roadway Segment Lighting Expert Panel Miscellaneous Install Raised Medians at Crosswalks Literature Review Reduce Mean Travel Speed Reanalysis of Existing Data *AMFs are listed in order of their presentation in this report. Table 18. List of AMFs developed or modified in NCHRP Project 17-25.* AADT Average annual daily traffic ADT Average daily traffic AMF Accident modification factor B/A Before/after EB Empirical Bayes HOV High-occupancy vehicle HSIS Highway Safety Information System LTL Left-turn lane LOPC Level of predictive certainty MH Medium high PDO Property damage only RTL Right-turn lane RTOR Right turn on red SD Superelevation deficiency SPF Safety performance function TWLTL Two-way left-turn lane vpd Vehicles per day Table 19. Glossary of acronyms for AMF summaries.

TREATMENT: Install Roundabout METHODOLOGY: Empirical Bayes Before-After All Crashes Injury Crashes All Crashes Injury Crashes Multilane - Urban/Suburban (prior control - stop sign) All Crashes Injury Crashes All Crashes Injury Crashes All Crashes Injury Crashes STUDY SITES: • Treatment sites included 55 intersections that were converted to roundabouts (36 were previously two-way stop-controlled, 10 were all-way stop-controlled, and 9 were controlled by signals). • The roundabouts were in rural, suburban, and urban environments. • Single-lane and multilane roundabouts were included; traffic volumes at the treatment sites in the after condition ranged from 2,668 vpd to 58,800 vpd. 0.29 (0.04) All Sites 55 0.65 (0.03) 0.52 (0.05) 0.24 (0.03) 0.22 (0.06)9 Single Lane - Rural (prior control - two-way stop-controlled) Single/Multilane - Urban/Suburban (prior control - signal) 16 9 0.22 (0.07) 0.13 (0.03) 0.82 (0.08) 0.28 (0.09) FOOTNOTES: A Ulf and Jörgen - 1999 (78 ). COMMENTS: • A non-significant increase of 3% was found for 10 sites which were all- way stop-controlled prior to conversion to a roundabout. • The authors were not able to determine the safety effects for pedestrians and bicyclists, but refer the reader to the positive results that have been found in Scandinavian evaluations.A • No evidence was found to indicate roundabouts result in more difficulties for older drivers. CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: High AMF 0.44 (0.06) Single Lane - Urban/Suburban (prior control - two-way stop-controlled) No. of Improved Sites REFERENCE: Rodegerdts et al. - 2007 (8 ). 11

TREATMENT: Add Exclusive Left-Turn Lane METHODOLOGY: Empirical Bayes Before-After/Analysis-Driven Expert Panels Rural Stop-Controlled Intersection (four legs) 25 0.72 (0.03) 0.52 (0.03) Rural Stop-Controlled Intersection (three legs) 36 0.56 (0.06) — Rural Signalized Intersection (four legs) 0.82D 0.67D Rural Signalized Intersection (three legs) 0.85D — Urban Stop-Controlled Intersection (four legs) 9 0.73C (0.03) 0.53C (0.04) Urban Stop-Controlled Intersection (three legs) 8 0.67 (0.12) — Urban Signalized Intersection (four legs) 39 0.90 (0.01) 0.81 (0.13) Urban Signalized Intersection (three legs) 0.93D — Rural Stop-Controlled Intersection (four legs) 24 0.65 (0.03) 0.42 (0.04) Rural Stop-Controlled Intersection (three legs) 11 0.45C (0.08) — Urban Stop-Controlled Intersection (four legs) 9 0.71C (0.04) 0.50C (0.06) Urban Stop-Controlled Intersection (three legs) 0.65F — Urban Signalized Intersection (four legs) 39 0.91 (0.01) 0.83 (0.02) Urban Signalized Intersection (three legs) 0.94F — Rural Stop-Controlled Intersection (four legs) 23 0.63 (0.07) 0.40 Rural Stop-Controlled Intersection (three legs) 35 0.38C (0.15) — Urban Stop-Controlled Intersection (four legs) 7 0.74 (0.07) 0.55 Urban Signalized Intersection (four legs) 35 0.87E (0.03) 0.76E FOOTNOTES: A AMF (both approaches) = AMF (one approach) x AMF (one approach). B Project-Related Accidents - All accidents involving one or more vehicles that had made, were making, or intended to make the specific left-turn maneuver(s) for which the left-turn lane(s) being evaluated were installed. C AMF based on comparison group evaluation. D Recommended AMF based on analysis-driven expert panel results (rural two-lane roads) from Harwood et al. - 2000 (9 ). E AMF based on yoked comparison evaluation. F Recommended AMF based on analysis-driven expert panel results from NCHRP 17-25/17-26 expert panel on urban/suburban arterials. Total Intersection Accidents (all severity levels, all accident types) No. of Improved Sites One Approach BothA Approaches REFERENCE: Harwood et al. - 2002 (12 ); NCHRP Project 17-25 research results AMF Project-Related Accidents (all severity levels)B *LOPC considered to be MH for AMFs derived by analysis-driven expert panels. CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: High* Fatal and Injury Intersection Accidents (all accident types) STUDY SITES: • Included rural and urban sites located in eight states – Illinois, Iowa, Louisiana, Minnesota, Nebraska, North Carolina, Oregon, and Virginia • 199 treatment sites where a left-turn lane (LTL) was added, as well as 300 similar intersections that were not improved during the study period and used for comparison and reference sites. • All improvements were made during the years 1989 through 1998. Mean duration of before and after periods were 6.7 years and 3.9 years, respectively. COMMENTS: • The study applied two alternative evaluation approaches (B/A with yoked comparisons and B/A with a comparison group) and recommended that the EB evaluation results be used if statistically significant. If not, it was recommended that statistically significant comparison group results be used, followed by statistically significant yoked comparison results. The authors note that results from either comparison method may be "overly optimistic." • Stop-controlled locations had stop signs on the minor road approaches. • Mean total entering ADT for rural stop-controlled, rural signalized, urban stop-controlled, and urban signalized improved sites were 9,700 vpd, 17,800 vpd, 15,500 vpd, and 26,800 vpd, respectively. • All tests of statistical significance in this report were performed at the 5% significance level (95% confidence level). Only statistically significant results are shown.

TREATMENT: Add Exclusive Right-Turn Lane METHODOLOGY: Empirical Bayes Before-After/Analysis-Driven Expert Panel Rural Stop-Controlled Intersection (four legs) 28 0.86 (0.05) 0.74 Rural Signalized Intersection (four legs) 0.96B (0.02) 0.92B Urban Signalized Intersection (four legs) 18 0.96 (0.02) 0.92 Urban Signalized Intersection (three legs) 0.96C — Urban Stop-Controlled Intersection (four legs) 0.86C 0.74C Rural Stop-Controlled Intersection (four legs) 29 0.77D (0.07) 0.59D Rural Signalized Intersection (four legs) 0.91B (0.03) 0.83B Urban Signalized Intersection (four legs) 17 0.91 (0.03) 0.83 Urban Signalized Intersection (three legs) 0.91C — Urban Stop-Controlled Intersection (four legs) 0.77C 0.59C CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: High* Fatal and Injury Intersection Accidents (all accident types) STUDY SITES: • Included rural and urban sites located in eight states – Illinois, Iowa, Louisiana, Minnesota, Nebraska, North Carolina, Oregon, and Virginia. • 108 treatment sites where a right-turn lane (RTL) was added, as well as 300 similar intersections that were not improved during the study period and used for comparison and reference sites. • All improvements were made during the years 1989 through 1998. Mean duration of before and after periods were 6.7 years and 3.9 years, respectively. *LOPC considered to be MH for AMFs derived by analysis-driven expert panels. FOOTNOTES: A AMF (both approaches) = AMF (one approach) x AMF (one approach). B Authors recommend that the AMFs for urban signalized intersections be applied to rural signalized intersections. C Recommended AMF based on analysis-driven expert panel results from NCHRP 17-25/17-26 panel on urban/suburban arterials. D AMF based on comparison group evaluation. Total Intersection Accidents (all severity levels, all accident types) No. of Improved Sites One Approach BothA Approaches REFERENCE: Harwood et al. - 2002 (12 ); NCHRP Project 17-25 research results AMF COMMENTS: • The study applied two alternative evaluation approaches (B/A with yoked comparisons and B/A with a comparison group) and recommended that the EB evaluation results be used if statistically significant. If not, it was recommended that statistically significant comparison group results be used, followed by statistically significant yoked comparison results. The authors note that results from either comparison method may be "overly optimistic." • Stop-controlled locations had stop signs on the minor road approaches. • Mean total entering ADT for rural stop-controlled, rural signalized, urban stop-controlled, and urban signalized improved sites were 9,700 vpd, 17,800 vpd, 15,500 vpd, and 26,800 vpd, respectively. • All tests of statistical significance in this report were performed at the 5% significance level (95% confidence level). Only statistically significant results are shown.

TREATMENT: Install Traffic Signal at Urban Intersection METHODOLOGY: Empirical Bayes Before-After All Crashes Right-Angle Crashes Rear-End Crashes Four-Leg Intersections All Crashes Right-Angle Crashes Rear-End Crashes CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: High STUDY SITES: • Included sites located in five states – California, Florida, Maryland, Virginia, Wisconsin – and Toronto. • Three-leg intersection data included 22 treatment sites (converted from stop to signal control) and 118 reference group sites (99 stop-controlled and 19 signalized intersections). • Four-leg intersection data included 100 treatment sites (converted from stop to signal control) and 295 reference group sites (96 stop-controlled and 199 signalized intersections). • An additional reference group was developed from the HSIS California urban data and included 1,418 stop-controlled and 799 signalized intersections.A • Minor street traffic volumes for the treatment sites ranged from 911 to 3,952 vpd; major street volumes ranged from 11,739 to 24,584 vpd. AMF 0.86 (0.32) 0.66 (0.45) 1.5 (0.51) FOOTNOTES: A The Highway Safety Information System (HSIS) is a multistate safety database that contains accident, roadway inventory, and traffic volume data for a select group of states and is sponsored by the FHWA. Three-Leg Intersections No. of Improved Sites REFERENCE: McGee, Taori, and Persaud - 2003 (18 ) COMMENTS: • AMFs are for crashes involving fatalities and injuries only; property-damage-only (PDO) crashes were excluded from the analysis. • AMFs were developed using data from urban intersections. The authors do not recommend that these results be applied to rural intersections. • The study notes that the results could be adapted (i.e., reversed) to assess the safety of removing a traffic signal. The authors of the study do not have as much confidence in using the results in this way. 22 100 0.77 (0.22) 0.33 (0.20) 1.38 (0.39)

TREATMENT: Install Traffic Signal at Rural Intersection METHODOLOGY: Empirical Bayes Before-After 0.56 (0.03) 0.23 (0.02) 1.58 (0.14) 0.4 (0.05) AMF Level of Predictive Certainty: High CRASH TYPE STUDIED AND ESTIMATED EFFECTS ECONOMIC ANALYSI S Three-Leg and Four-Leg Intersections Combined N o. o f Improved Sites 45 FOOTNOTES: A The Highway Safety Information System (HSIS) is a multistate safety database that contains accident, roadway inventory, and tra ffic volume data for a select group of states and is sponsored by the FHWA. 0.27 (0.001) STUDY SITES: • Included sites located in Minnesota and California. Data were acquired from the Highway Safety Information System . A • Three-leg intersection data included six treatment sites (converted from stop to signal control) and 1,927 stop-controlled reference group sites. • Four-leg intersection data included 39 treatment sites (converted from stop to signal control) and 1,661 stop-controlled reference group sites. • An additional reference group was developed using 84 signalized intersections to develop a more sophisticated procedure for evaluating the potential safety effects of a contemplated signal conversion. • Minor street traffic volumes for the treatment sites ranged from 101 to 10,300 vpd; major street volumes ranged from 3,261 to 29,926 vpd REFERENCE: NCHRP Project 17-25 research results All Crashes AMF All Crashes Right-Angle Crashes Rear-End Crashes Left-Turn Crashes COMMENTS: • The authors of the study do not recommend that the results be adapted (i.e., reversed) to assess the safety of removing a tra ffic signal. • The treatment benefits are greater on higher volume intersections and are greater where the ratio of expected right-angle cr ashes to rear-end crashes is higher. There is little difference between the effects on three-leg vs. four-leg sites or on sites with two lanes on the major vs. four lanes. Thus, the overall crash frequency AMFs can be assumed to apply to all rural site types. • Economic analysis was conducted to determine if the increase in rear-end crashes negated the decrease in other, generally mo r e severe, collision types. The economic analysis may be used to develop AMFs for total crashes, which account for the differences in injury severity that occur with different collision types. The AMF for all crash severities would be 0.27. Three-Leg and Four-Leg Intersections Combined AMF

TREATMENT: Remove Traffic Signal (Urban Environment) METHODOLOGY: Empirical Bayes Before-After All Crashes Right-Angle and Turning Crashes Rear-End Crashes Pedestrian Crashes Fixed-Object Crashes Light Condition (all severities) Day Night Injury Severity (all collision types) Severe Minor 0.69A 0.76 (0.18)199 FOOTNOTES: A The AMF for fixed object crashes was based on the classical estimate (i.e., expected number of crashes in the after period is based on count of crashes in the before period as opposed to the EB estimate of before-period crashes). Type of Collision (all severities) No. of Improved Sites REFERENCE: Persaud et al. - 1997 (19 ) COMMENTS: • The authors note the inability to account for year-to-year variation in traffic volumes, but nonetheless express confidence in the results. • It is important to note that this study was for one-way streets in an urban environment. There are no comparable studies for two-way streets or for intersections in rural environments. 199 0.78 0.70 0.47 (0.10) CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: High STUDY SITES: • 199 treatment sites and 71comparison sites in Philadelphia. • Treatment sites were unwarranted signals and mostly changed from signal control to all-way stop control between 1979 and 1988. • All intersections were at one-way streets in non-arterial streets in an urban environment. • Crash data were acquired for the years 1978 through 1992. • Traffic volumes were often estimated from upstream and downstream AADTs due to the sparse volume data available. AMF 0.76 (0.38) 0.76 (0.35) 0.71 (0.06)199 0.82 (0.12)

TREATMENT: Modify Signal Change Interval METHODOLOGY: Before-After with Control Group All Crashes Multiple-Vehicle Crashes Rear-End Crashes Right-Angle Crashes Pedestrian/Bicyclist Crashes Accident Type (injury crashes only) All Crashes Multiple-Vehicle Crashes Rear-End Crashes Right-Angle Crashes Pedestrian/Bicyclist Crashes 0.63 STUDY SITES: • Included crash data from 40 treatment intersections and 56 control intersections in Nassau County and Suffolk County, New York. • All intersections were standard four-leg junctions. • The treatment sites were randomly selected for the signal timing change, eliminating the site-selection bias. • Six years of crash data were used in the analysis (October 1991 through October 1997), with 3 years each in the before and after periods. • Analysis included only "reportable" crashes, which require an injury or a minimum of $1,000 in property damage in New York. CRASH TYPE STUDIED AND ESTIMATED EFFECTS 1.12A (0.16) 0.96A (0.18) 0.63 0.91 1.06A (0.22) AMF Level of Predictive Certainty: Medium-High AMF 0.92 (0.09) 0.95A FOOTNOTES: A Results were not significant at a 90% confidence level (P > 0.10). AMF of 1.0 recommended for these accident types. Accident Type (all severities) No. of Treated Sites REFERENCE: Retting, Chapline, and Williams - 2002 (22 ) COMMENTS: • IMPORTANT NOTE - Both the yellow change interval and the red clearance interval were adjusted at the treatment sites to conform to the Institute of Transportation Engineers Determining Vehicle Change Intervals: A Proposed Recommended Practice (95). In some cases, this meant an increase in the interval, while in others, the interval was decreased. Thus the AMFs do not reflect the effects of increasing only the change and clearance intervals. • AMFs are based on the odds ratios. • Yellow change intervals at the treatment sites ranged from 3 to 4 seconds in the before period and 2.6 to 5.4 seconds in the after period. Red clearance intervals ranged from 2 to 3 seconds in the before period and 1.1 to 6.5 seconds in the after period. •Authors acknowledge that the results do not account for variables such as geometry, traffic volume, and other signal parameters such as cycle length and number of phases 40 1.08A (0.17)40 0.88 (0.09)

TREATMENT: Prohibit Right Turn on Red METHODOLOGY: Analysis-Driven Expert Panel REFERENCE: NCHRP Projects 17-25 and 17-26 research results COMMENTS: • Expert panel on urban/suburban arterials considered this AMF function to be the best estimate for the prohibition of right turn on red (RTOR). • The AMF was derived from a simple before-after analysis of intersections in Alabama and South Carolina after the passage of laws in both states that permitted RTOR.A The results were presented in terms of the effect on total crashes at an intersection if RTOR was permitted (AMF = 1.067). Making an assumption that most of the intersections were four-leg locations, the AMF for each approach becomes 1.016. • The inverse of the Clark AMF was derived to reflect the prohibition of RTOR (1/1.016 = 0.984). ACCIDENT MODIFICATION FUNCTION AMF Level of Predictive Certainty: Medium-High FOOTNOTES: A Clark, Maghsoodloo, and Brown - 1983 (79 ) AMF = (0.984)n n = number of signalized intersection approaches where RTOR is prohibited Note: AMF applies to total intersection crashes. where:

TREATMENT: Modify Left-Turn Phase METHODOLOGY: Empirical Bayes Before-After/Analysis-Driven Expert Panel Left-Turn Crashes Total Crashes Left-Turn Crashes Total Crashes AMF COMMENTS: • There was evidence that non-left-turn crashes increased following the change to protected-only left-turn phasing. Further research is necessary to determine the specific reasons for the effect on non-left-turn crashes. However, it seems reasonable to speculate that introducing a protected left- turn phase will tend to increase mostly rear-end crashes (which are in general less severe compared to left-turn crashes) because of the increased number of phases (and therefore dilemma zone opportunities) and the increase in queues that results from reduced green time available for all traffic not protected by the introduced phase. This also implies that the measure would be most effective overall where there is a relatively high frequency of left-turn crashes. Change from Permissive to Permissive/Protected Phasing 35 0.84 B (0.02) 1.00B FOOTNOTES: ** statistically significant at the 95% confidence level. A AMF of 0.99 was not statistically significant; AMF of 1.00 recommended. B Recommended AMF based on analysis-driven expert panel results from NCHRP Projects 17-25/17-26 panel on urban/suburban arterials. Primary source of information was the study by Lyon et al. - 2005 (21 ), which included 35 four-leg intersections in Toronto. 12 0.99A (0.07) AMF Level of Predictive Certainty: Medium-High CRASH TYPE STUDIED AND ESTIMATED EFFECTS REFERENCE: NCHRP Project 17-25 research results Accident Type No. of Sites 0.01 (0.01)** Change from Permissive or Permissive/Protected to Protected-Only Phasing STUDY SITES: • The treatment sites included 12 signalized intersections in Winston- Salem, NC. Among those 12 sites, the left-turn phase was changed from permissive to protected-only at eight sites and from permissive/protected to protected-only at four sites.

TREATMENT: Replace 8-in. Signal Heads with 12-in. Signal Heads METHODOLOGY: Empirical Bayes Before-After Right-Angle All FOOTNOTES: A Statistically significant at the 95% confidence level. B AMF for all crashes was 0.97 but was not statistically significant; AMF of 1.00 is recommended. COMMENTS: • There is evidence of an increase in non-right-angle crashes that almost offsets the decrease in right-angle crashes. It is possible that the increased signal head size encouraged more drivers to stop at the red light and probably led to an increase in rear-end crashes. Rear-end crashes are generally less severe compared to right-angle crashes. An economic analysis in which the decreased angle collision costs are combined with the increased rear-end collision costs revealed a reduction of about $11,800 per intersection-year in the overall crash harm due to this treatment. Collision costs developed in FHWA Study by Council et al. in 2005 (72 ). AMF Level of Predictive Certainty: Medium-High CRASH TYPE STUDIED AND ESTIMATED EFFECTS REFERENCE: NCHRP Project 17-25 research results Accident Type No. of Sites AMFSTUDY SITES: • Treatment sites included 26 signalized intersections from Winston-Salem, NC. • Safety performance functions were developed with data from 60 signalized intersections in Winston-Salem using data from 1991 to 2004. 26 0.58 A (0.07) 0.97B (0.06)

TREATMENT: Convert Nighttime Flash Operation to Steady Operation METHODOLOGY: Empirical Bayes Before-After Nighttime Angle All Nighttime FOOTNOTES: A Statistically significant at the 95% confidence level. B Statistically significant at the 90% confidence level. STUDY SITES: • The treatment sites included 12 signalized intersections from Winston- Salem, NC. • Data were available for 518 intersection-months before the change and 516 intersection-months after the change. • Safety performance functions were developed with data from 60 signalized intersections in Winston-Salem using data from 1991 to 2004. COMMENTS: • Number of sites is low and hence results should be treated with caution. • The reduction in nighttime angle and nighttime crashes (about 35%) is lower than the reductions shown in other simple before-after studies indicating that bias due to regression to the mean was probably significant in those studies. AMF Level of Predictive Certainty: Medium-High CRASH TYPE STUDIED AND ESTIMATED EFFECTS REFERENCE: NCHRP Project 17-25 research results Accident Type No. of Sites AMF 12 0.66 A (0.18) 0.65B (0.15) TREATMENT: Replace Single Red Signal Head with Dual Red Signal Heads METHODOLOGY: Empirical Bayes Before-After Right-Angle All AMF Level of Predictive Certainty: Medium-High CRASH TYPE STUDIED AND ESTIMATED EFFECTS COMMENTS: • Adding additional signal heads does not seem to be effective in reducing right-angle or total crashes. With only eight sites, the sample size is small. Thus, the results have to be treated with caution. REFERENCE: NCHRP Project 17-25 research results Accident Type No. of Sites AMF 8 1.05 (0.13) A 1.18 (0.11)A STUDY SITES: • Treatment sites included eight signalized intersections from Winston- Salem, NC. • Safety performance functions were developed with data from 60 signalized intersections in Winston-Salem using data from 1991 to 2004. FOOTNOTES: A AMF was not statistically significant; AMF of 1.00 is recommended.

TREATMENT: Convert to All-Way Stop Control METHODOLOGY: Before-After Analysis w/Likelihood Functions All Crashes Right-Angle Crashes Rear-End Crashes Left-Turn Crashes Pedestrian Crashes Crash Severity (all collision types) All Crashes Injury Crashes FOOTNOTES: A Includes all sites from the four regions. Type of Collision (all severities) No. of Improved Sites REFERENCE: Lovell and Hauer - 1986 (30 ) COMMENTS: • Analysis included the reanalysis of datasets from San Francisco, Philadelphia, and Michigan to correct for regression to the mean bias and a new analysis of data from Toronto. Likelihood functions were used to combine the results from the various cities. • The AMF for All Crashes (all collision types and all severities) was vetted by an expert panel on rural two-lane roads and included as the recommended AMF for this treatment within FHWA's Interactive Highway Safety Design Model — IHSDM (Harwood et al. - 2000 [9]). 0.53 0.29 (0.06) 360A CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: Medium-High STUDY SITES: • Included data from three urban regions (San Francisco, Philadelphia, and Toronto) and one rural region (Michigan). • The number of treatment sites in which an intersection was converted to all-way stop control in each region is as follows: - San Francisco: 49 sites (from two-way stop control). - Philadelphia: 222 sites (one-way streets, prior traffic control not stated). - Michigan: 10 sites (from two-way stop control). - Toronto: 79 sites (from two-way stop control). AMF 0.53 0.28 (0.03) 0.87 (0.13) 0.80 (0.52) 0.61 (0.08) 360A

TREATMENT: Convert Stop Control to Yield Control METHODOLOGY: Before-After with Control Group All Crashes 141 A STUDY SITES: • Treatment sites were converted from stop control to yield control; comparison sites were stop-control intersections. The numbers of each type in each city were as follows: - Saginaw, MI (53 treatment sites, 42 control sites). - Pueblo, CO (69 treatment sites, 15 control sites). - Rapid City, SD (19 treatment sites, 8 control sites). • The conversions took place between 1982 and 1987. • The number of years of crash data included in each before and after period ranged from 1 to 2 years, depending on the city and year of conversion. FOOTNOTES: A Includes all sites from the three cities. CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: Medium-High AMF 2.37 Total Accidents (all severities) No. of Treated Sites REFERENCE: McGee and Blankenship - 1989 ( 31 ) COMMENTS: • AMF computed from the cross product ratio (odds ratio) of the before and after crash frequencies at the treatment and control sites. • No additional AMFs were provided. The authors do indicate that the proportion of fatal or injury accidents does not appear to increase with the conversion, nor is there a change in the distribution of collision types. • The authors note that the probability of an increase in crashes is greater with higher volumes, either major street volume, minor street volume, and/or the combination of the two volumes.

TREATMENT: Install Red-Light Cameras METHODOLOGY: Empirical Bayes Before-After Rear-End Crashes Right-Angle Crashes Injury Crashes Only Rear-End Crashes Right-Angle Crashes All Crash Severities Total Crashes Rear-End Crashes Right-Angle Crashes Injury Crashes Only Total Crashes Rear-End Crashes Right-Angle Crashes REFERENCES: Persaud et al. - 2005 (25 ) and Council, Persaud et al. - 2005 (26 ) COMMENTS: • Economic analysis was conducted to determine if the increase in rear-end crashes negated the decrease in right-angle collisions. Results showed there was a net economic benefit that ranged from $39,000 to $50,000 per year per site where red-light camera systems were installed. • The economic analysis may be used to develop AMFs for total crashes, which account for the differences in injury severity that occur with different collision types. The AMF for all crash severities would be 0.91, while the AMF for injury crashes only would be 0.86. 1.24 (0.12) 0.84 (0.06)132 STUDY SITES: • Included data from seven jurisdictions across the U.S. for 132 treatment intersections where red-light cameras had been installed. • The reference group included similar signalized intersections in each jurisdiction that were not equipped with red-light cameras, which were used to develop SPFs and to investigate possible spillover effects. • A second reference group of unsignalized intersections was used to account for time trends and to calibrate the SPFs. 1.02 (0.008) 0.711 (0.006) All Crash Severities No. of Treated Sites AMF Level of Predictive Certainty: High AMF ECONOMIC ANALYSIS Economic AMF CRASH TYPE STUDIED AND ESTIMATED EFFECTS 1.15 (0.03) 0.75 (0.03)132 0.72 (0.006) 0.86 (0.005) 0.91 (0.004) 1.09 (0.007)

TREATMENT: Add Intersection Lighting METHODOLOGY: Meta-Analysis/Analysis-Driven Expert Panel Total Crashes All Injury Crashes Total Crashes All Injury Crashes FOOTNOTES: A Elvik and Vaa (60 ) FMAsehsarC llA 0.96 0.94 AMF Level of Predictive Certainty: Medium-High REFERENCE: Elvik and Vaa - 2004 (60 ); NCHRP 17-25 Final Report; NCHRP 17-26 Final Report COMMENTS: • The meta-analysis results produced AMF estimates for reductions in fatal, injury and property-damage-only accidents of 0.36, 0.72, and 0.83, respectively.A • The NCHRP 17-25/17-26 expert panel on urban/suburban arterials recommended that the meta-analysis results be applied to intersections and that the fatal and injury results be combined into a single AMF for all levels of injury. • The NCHRP 17-26 Final Report includes a distribution of crashes by time of day and injury severity for different types of intersections. FMAsehsarC emitthgiN 0.79 0.71 STUDY SITES: • 38 studies were evaluated as part of the meta-analysis, including 14 U.S. studies.A • Distributions of crashes by injury severity and time of day were obtained from the HSIS data for the states of North Carolina and Minnesota. CRASH TYPE STUDIED AND ESTIMATED EFFECTS

TREATMENT: Increase Pavement Friction on Intersection A pp roach METHODOLOGY: Empirical Bayes Before-After 0.8 (0.03) 0.43 (0.03 ) 0.58 (0.03 ) 1.15 (0.05 ) 0.32 (0.04 ) Accident Type AMF No. of Treated Sites AMF Level of Predictive Certainty: High CRASH TYPE STUDIED AND ESTIMATED EFFECTS FOOTNOTES: A Bray - 2001 ( 80 ). All Crashes Wet-Road Crashes Rear-End Crashes STUDY SITES: • The treatment data for this analysis were from the Skid Accident Reduction Program (SKARP) developed by NY State DOT in 1995. A • Data were collected for 256 treated intersections and 3,993 reference intersections. Intersections were in both urban and rural locations. • For the treated sites, 73 were signal-controlled, 176 were stop-controlled, and 7 were yield-controlled. Fifty-seven were four-leg and 199 were three- leg intersections. • Sites are selected for treatment based on both a high proportion of wet- road accidents and low friction numbers. Dry-Road Crashes COMMENTS : • The treatment generally involved a 1.5-in. resurfacing or a 0.5-in. microsurfacing using non-carbonate aggregates. • Table 9 in Chapter 3 of this report provides additional AMFs by categories for traffic control and number of legs. REFERENCE: NCHRP Project 17-25 research results 256 Rear-End Wet-Road Crashes

TREATMENT: Narrow Lane Widths to Add Lanes METHODOLOGY: Empirical Bayes Before-After All Crashes Fatal, Injury, and PDO-Tow-Away Crashes Fatal and Injury Crashes Five-to-Six-Lane Conversions All Crashes Fatal, Injury, and PDO-Tow-Away Crashes Fatal and Injury Crashes FOOTNOTES: A The Highway Safety Information System (HSIS) is a multistate safety database that contains accident, roadway inventory, and traffic volume data for a select group of states and is sponsored by the FHWA. Four-to-Five-Lane Conversions No. of Treated Sites REFERENCE: Bauer et al. - 2004 (45 ) COMMENTS: • The treatment described here is the addition of a travel lane to an urban freeway by decreasing existing lane widths through restriping, converting all or part of the shoulder to a travel lane, or by using both in combination. In most cases, the shoulder conversion was done to add an HOV lane. Results are not applicable to other roadway types. • Other EB analyses conducted found: - Increase in sideswipe collisions at four-to-five-lane conversions and a decrease in such collisions at five-to-six-lane conversions. - Increase in crashes adjacent to on- or off-ramps for both types of conversions. Increase in crashes away from ramps for four-to-five-lane conversions, but a decrease in crashes away from ramps for five-to-six-lane conversions. • The authors also examined accident migration patterns upstream and downstream of the conversions. The findings suggest that the conversion projects may result in fewer crashes upstream and an increased number of crashes downstream, which may reflect the fact that the operational bottleneck has been shifted. 1.03* (0.05) 1.04* (0.06) 79 45 1.07* (0.07) 1.11 (0.05) STUDY SITES: • All treatment and reference sites were located on four freeways in Los Angeles and San Diego Counties, California. • The treatments included two project types: (1) four to five lanes and (2) five to six lanes. The first type included 79 sites and 36.4 miles, while the second included 45 sites and 12.5 miles. All conversions were made in 1993. • Crash data were acquired from the FHWA HSIS and included 2 years of before data and 7 years of after data.A • Traffic volumes at the treatment sites ranged from 77,000 vpd to 128,000 vpd. * Results for the five-to-six-lane conversions were not statistically significant. Recommended AMF of 1.0 recommended for five-to-six-lane conversions. AMF Level of Predictive Certainty: Medium-High AMF 1.11 (0.03) 1.10 (0.04) CRASH TYPE STUDIED AND ESTIMATED EFFECTS

TREATMENT: Add Passing Lanes (Two-Lane Roads) METHODOLOGY: Analysis-Driven Expert Panel One-way (single direction of travel) Two-way (short four-lane sections) CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: Medium-High Type of Passing Lane REFERENCE: Harwood et al. - 2000 ( 9 ) AMF A 0.75 0.65 COMMENTS: • Expert panel considered these AMFs to be the best estimates for the installation of passing lanes on rural two-lane roadways. Results are not applicable to other roadway types. • Expert panel notes that these AMFs are based on the assumption that the passing lanes are operationally warranted, and the length is appropriate for conditions. • The AMFs apply to total accidents within the passing-lane section of the roadway and do not include upstream or downstream accidents. • The AMF for short, four-lane sections does not apply to extended lengths of four-lane highways. FOOTNOTES : A Estimates are based on work by Harwood and St. John - 1984 (81) and Nettelbad - 1979 ( 82 ).

TREATMENT: Add Two-Way Left-Turn Lane (TWLTL) METHODOLOGY: Analysis-Driven Expert Panels where: ACCIDENT MODIFICATION FUNCTION AMF Level of Predictive Certainty: Medium-High REFERENCE: Harwood et al. - 2000 (9 ) FOOTNOTES: A Hauer - 2000 (83 ). AMF = 1 – 0.7PDPLT/D PD = driveway-related accidents as a proportion of total accidents PD estimated as: where: DD = driveway density (driveways per mile) PLT/D = left-turn accident susceptible to correction by TWLTL as a proportion of driveway-related accidents PLT/D = 0.5 (estimated by expert panel on the basis of work by Hauer)A COMMENTS: • Expert panel considered this AMF function to be the best estimate for the installation of a TWLTL without data on left-turn volumes within the TWLTL. • Expert recommends a minimum driveway density of 5 driveways/mile for the AMF to be applied; the AMF for any lesser density would be equal to 1.0. • Estimate function for driveway-related accidents is based on work by Hauer, which included a critical review of 14 studies conducted between 1964 and 1997.A The AMF function shown here is more conservative than the Hauer AMF. • Most of the studies reviewed by Hauer analyzed TWLTLs in urban and suburban areas. Hauer noted that the safety effects on rural roads should be at least as large as those on urban and suburban roads. Thus, the AMF shown here is applicable to rural and urban two-lane and multilane roads. • NCHRP Project 17-25 urban/suburban expert panel confirmed the adoption of this AMF for the road classes noted above. 2 2 0.0024DD0.0047DD1.199 0.0024DD0.0047DD

TREATMENT: Change Lane Width METHODOLOGY: Analysis-Driven Expert Panel where: where: ACCIDENT MODIFICATION FUNCTION AMF Level of Predictive Certainty: Medium-High f = factor for roadway type f = 0.75 for multilane undivided and 0.50 for divided AMF = (AMFRA – 1.0)PRA + 1.0 AMF = accident modification factor for total accidents AMFRA = accident modification factor for related accidentsB AMFRA is calculated by dividing the AMF for the after-improvement condition by the AMF for the before condition - each can be selected from the following table:C PRA = proportion of total accidents constituted by related accidents PRA = 0.35 (estimated from distribution of accident types) Rural Two-Lane RoadsREFERENCES: Harwood et al. - 2000 (9) and Harwood et al. - 2003 (51) Rural Multilane Roads AMF = f (AMFRA – 1.0)PRA + 1.0 COMMENTS: • The AMFs for ADTs greater than 2,000 are largely based on work by Zegeer et al.; AMFs for ADTs less than 400 are based on work by Griffin and Mak. AMFs for ADTs between 400 and 2,000 were based on expert panel judgment and an extensive critique of the literature by Hauer.A • If lane widths differ for the two directions of travel, the AMF should be determined for each direction and then averaged to obtain an AMF for the roadway. • The factors for rural multilane roads were developed by an expert panel. There is less confidence in the rural multilane AMF. • The AMFs developed do not apply to urban roadways. FOOTNOTES: A Zegeer et al. - 1988 (84); Griffin and Mak - 1987 (85); Hauer - 2000 (86). B Related accidents include single-vehicle run-off-road, multiple-vehicle head-on, and opposing- and same-direction sideswipe accidents. C Table developed in Harwood et al. - 2003 (51). < 400 400 to 2000 > 2000 9 ft 1.05 1.05+2.81x10-4 (ADT-400) 1.50 10 ft 1.02 1.02+1.75x10-4 (ADT-400) 1.30 11 ft 1.01 1.01+2.5x10-5 (ADT-400) 1.05 12 ft 1.00 1.00 1.00 Average Daily Traffic (ADT) Lane Width

TREATMENT: Change Shoulder Width and/or Type METHODOLOGY: Analysis-Driven Expert Panel where: AMF Level of Predictive Certainty: Medium-High AMF = (AMFWRA AMFTRA – 1.0)PRA + 1.0 AMF = accident modification factor for total accidents AMFWRA = accident modification factor for related accidents based on shoulder widthB AMFWRA is calculated by dividing the AMF for the after-improvement condition by the AMF for the before condition - each can be selected from the following table:C PRA = proportion of total accidents constituted by related accidents PRA = 0.35 (estimated from distribution of accident types) Rural Two-Lane and Multilane RoadsREFERENCES: Harwood et al. - 2000 (9) and Harwood et al. - 2003 (51) AMFTRA = accident modification factor for related accidents based on shoulder typeB AMFTRA is calculated by dividing the AMF for the after-improvement condition by the AMF for the before condition - each can be selected from the following table: ACCIDENT MODIFICATION FUNCTION COMMENTS: • Shoulder width AMFs for ADTs greater than 2,000 are largely based on work by Zegeer et al. (1988); AMFs for ADTs less than 400 are based on low-volume roads work by Zegeer et al. (1981). AMFs for ADTs between 400 and 2,000 were based on expert panel judgement and an extensive critique of the literature by Hauer.A • Shoulder Type AMFs are based on work by Miaou for differences in gravel and paved shoulders and Zegeer et al (1988) for differences in turf and paved shoulders. Composite shoulders are 50% paved and 50% turf; the AMFs are averages for the two types.A • If shoulder widths/types differ for the two directions of travel, the AMF should be determined for each direction and then averaged to obtain an AMF for the roadway. • The AMFs developed do not apply to urban roadways FOOTNOTES: A Zegeer et al. - 1988 (84); Zegeer, Dean, and Mayes - 1981 (87 ); Miaou - 1996 (88); Hauer - 2000 (89). B Related accidents include single-vehicle run-off-road, multiple-vehicle opposing- and same-direction sideswipe accidents. C Table developed in Harwood et al. - 2003 (51). < 400 400 to 2000 > 2000 0 ft 1.10 1.1+2.5x10-4(ADT-400) 1.50 2 ft 1.07 1.07+1.43x10-4(ADT-400) 1.30 4 ft 1.02 1.02+8.125x10-5(ADT-400) 1.15 6 ft 1.00 1.00 1.00 8 ft 0.98 0.98+6.875x10-5(ADT-400) 0.87 Average Daily Traffic (ADT) 0 1 2 3 4 6 8 10 Paved 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Gravel 1.00 1.00 1.01 1.01 1.01 1.02 1.02 1.03 Composite 1.00 1.01 1.02 1.02 1.03 1.04 1.06 1.07 Turf 1.00 1.01 1.03 1.04 1.05 1.08 1.11 1.14 Shoulder Width (ft)Shoulder Type Shoulder Width

TREATMENT: Flatten Horizontal Curve METHODOLOGY: Analysis-Driven Expert Panel where: ACCIDENT MODIFICATION FUNCTION AMF Level of Predictive Certainty: Medium-High REFERENCE: Harwood et al. - 2000 (9) FOOTNOTES: A Zegeer et al. - 1992 (90). LC = length of horizontal curve (miles); does not include spiral curve length R = radius of curvature (ft) S = 1 if spiral transition curve is present and 0 if no such transition exists COMMENTS: • AMF applies to total accidents on the curved roadway segment. • AMF was derived from the regression model developed by Zegeer et al.A • The AMF is applicable to rural two-lane roads only. C C 1.55L 0.012S R 80.21.55L AMF TREATMENT: Improve Curve Superelevation METHODOLOGY: Analysis-Driven Expert Panel FOOTNOTES: A Zegeer et al. - 1992 (90). COMMENTS: • AMF applies to total accidents occurring on curved roadway segments. • Expert panel noted there was no safety effect until the superelevation reached 0.01. • AMF was derived from the results of Zegeer et al.A • The AMF is applicable to rural two-lane roads only. 1.00 1.00 + 6(SD - 0.01) 1.06 + 3(SD - 0.02) < 0.01 0.01 < SD < 0.02 > 0.02 AMF Level of Predictive Certainty: Medium-High AMFSuperelevation Deficiency (SD) REFERENCE: Harwood et al. - 2000 (9)

TREATMENT: Add Shoulder Rumble Strips METHODOLOGY: Before-After with Comparison Sites All Single-Vehicle Run-Off-Road Crashes Injury Single-Vehicle Run-Off-Road Crashes Rural Freeways All Single-Vehicle Run-Off-Road Crashes Injury Single-Vehicle Run-Off-Road Crashes REFERENCE: Griffith - 1999 (53 ) STUDY SITES: • Included 55 treatment sites and 55 matched comparison sites from rural and urban freeways in Illinois. • The treatment sites covered 196 miles of rural freeway and 67 miles of urban freeway. • The treatment sites were not selected on the basis of accident history; thus, there was no selection bias. 55 COMMENTS: • Results for all freeways based on yoked comparison analysis; results for rural freeways based on comparison group method using 29 of the treatment sites. Results could not be developed for urban sites separately. • An analysis of multi-vehicle accidents showed the rumble strips to have no effect on such accidents. • The AMF is not applicable to other road classes (two-lane or multilane). All Freeways (Rural and Urban) No. of Improved Sites AMF CRASH TYPE STUDIED AND ESTIMATED EFFECTS 0.79 (0.10)29 AMF Level of Predictive Certainty: Medium-High 0.82 (0.07) 0.87 (0.12) 0.93 (0.16)

TREATMENT: Add Centerline Rumble Strips METHODOLOGY: Empirical Bayes Before-After All Crashes Frontal/Opposing-Direction Sideswipe Crashes Accident Type (injury crashes) All Crashes Frontal/Opposing-Direction Sideswipe Crashes FOOTNOTES: A The Highway Safety Information System (HSIS) is a multistate safety database that contains accident, roadway inventory, and traffic volume data for a select group of states and is sponsored by the FHWA. CRASH TYPE STUDIED AND ESTIMATED EFFECTS 0.85 (0.08) 98 REFERENCE: Persaud, Retting, and Lyon - 2003 (57 ) STUDY SITES: • Crash and traffic volume data were collected for 98 treatment sites, consisting of 210 miles, where centerline rumble strips had been installed on rural two-lane roads in the states of California, Colorado, Delaware, Maryland, Minnesota, Oregon, and Washington. • The average length of the treatment sites was 2 miles, and the traffic volumes ranged from 5,000 to 22,000 vpd. • The reference group of sites was developed from HSIS data for the states of California, Washington, and Minnesota.A Additional data were acquired from Colorado for SPF calibration for the Colorado sites. COMMENTS: • The authors note that the results cover a wide range of geometric conditions, including curved and tangent sections and sections with and without grades. • The results include all rumble strip designs (milled-in, rolled-in, formed, and raised thermo-plastic) and placements (continuous versus intermittent) that were present. • The AMF is not applicable to other road classes (multilane). Accident Type (all severities) No. of Improved Sites AMF AMF Level of Predictive Certainty: Medium-High 0.86 (0.05) 0.79 (0.12) 0.75 (0.15) 98

TREATMENT: Install/Upgrade Guardrail METHODOLOGY: Meta-Analysis Fatal Injury Crashes All Injury Crashes CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: Medium-High REFERENCE: Elvik and Vaa - 2004 (60) COMMENTS: • The results apply to the installation of guardrail along an embankment. The studies were not differentiated by roadway class. • The analysis also included an estimate for the change in accident rate, but the results were not significant. • Results were also included for changing to softer guardrails. However, specifics on the type of change in hardware were not indicated, and not all results were significant. Therefore, they are not included here. Run-Off-Road Accidents AMF 0.56 (0.10) 0.53 (0.05) STUDY SITES: • 20 studies were evaluated, including 12 U.S. studies (6 of which were conducted in 1982 or later).

TREATMENT: Convert Undivided Four-Lane Road to Three-Lane and TWLTL (Road Diet) METHODOLOGY: Empirical Bayes Before-After REFERENCE: NCHRP Project 17-25 research results Iowa Predominately U.S. and state routes within small urban areas (average population of 17,000) Total Crashes 15 15 miles 0.53 (0.02) California/Washington Predominately corridors within suburban areas surrounding larger cities (average population of 269,000) Total Crashes 30 30 mi 0.81 (0.03) All Sites TotalCrashes 45 40 mi 0.71 (0.02) COMMENTS: • The study conducted was a reanalysis of data from two prior studies. A,B • The reanalysis of the Washington/California data indicated a 19% decrease in total crashes. The reanalysis of the Iowa data showed a reduction of 47% in total crashes. If the characteristics of the treated site can be defined on the basis of road and area type (as shown above), the AMFs of 0.53 and 0.81 should be used. Otherwise, it is recommended that the aggregate AMF of 0.71 be applied. FOOTNOTES: A Huang, Stewart, and Zegeer - 2002 (43 ). B Pawlovich et al. - 2006 (44 ). STUDY SITES: • 15 urban locations in Iowa with a mean length of 1.02 miles, a minimum and maximum length of 0.24 and 1.72 miles. AADT after conversion ranged from 3,718 to 13,908. • 30 urban locations from Washington and California studied previously with a mean length of 0.84 miles, a minimum and maximum length of 0.08 and 2.54 miles. AADT after conversion ranged from 6,194 to 26,376. AMF Level of Predictive Certainty: High CRASH TYPE STUDIED AND ESTIMATED EFFECT No. of Treated Sites AMF Accident TypeState/Site Characteristics

TREATMENT: Increase Pavement Friction on Roadway Se g me n t METHODOLOGY: Empirical Bayes Before-After 0.76 (0.02 ) 0.43 (0.02 ) 0.83 (0.04 ) 0.58 (0.06 ) 0.7 (0.04) FOOTNOTES: A Bray - 2001 ( 80 ). All Crashes Wet-Road Crashes Rear-End Crashes Single Vehicle Crashes AMF Level of Predictive Certainty: High REFERENCE: NCHRP Project 17-25 research results STUDY SITES: • The treatment data for this analysis were from the Skid Accident Reduction Program (SKARP) developed by NY State DOT in 1995. A • Data were collected from New York State for 36.3 miles of treated segments and 1,242.4 miles of reference segments. Locations were in both urban and rural locations. • The segments are in close proximity to treated intersections, which are the primary targets of the treatment. • Sites are selected for treatment based on both a high proportion of wet- road accidents and low friction numbers. CRASH TYPE STUDIED AND ESTIMATED EFFECTS 36 miles Rear-End Wet-Road Crashes COMMENTS: • The treatment generally involved a 1.5-in. resurfacing or a 0.5-in. microsurfacing using non-carbonate aggregates. • Table 10 in Chapter 3 of this report provides additional AMFs by categories for number of lanes and urban versus rural locations, although not all are statistically significant. Accident Type AMF No. of Treated Sites

TREATMENT: Change Median Width METHODOLOGY: Cross-Sectional Model COMMENTS: • NCHRP Projects 17-25/17-26 expert panel reviewed several studies of the effects of median width on crashes and reached a recommendation to either reanalyze data from one of those efforts or conduct a more robust analysis. Full Access Control Partial or No Access Control AMF Level of Predictive Certainty: Medium-High REFERENCES: NCHRP Project 17-25 research results ACCIDENT MODIFICATION FACTORS STUDY SITES: • Ten years of data from 1993 to 2002 on divided roadway sections in California were obtained from the Highway Safety Information System (HSIS). • The dataset included 500 miles of rural and urban roadways with parital or no access control and 1,400 miles with full access control. There were no median barriers on any of these roadway segments, meaning the barriers were traversable. Median widths for the segments included in the analysis ranged from 4 ft to 100 ft. • Over the 10-year period, the partial/no-access control sections experienced approximately 41,000 total crashes and 5,000 cross-median crashes. The full-access control sections experienced approximately 125,000 total crashes and 5,000 cross-median crashes.

TREATMENT: Change Roadside Sideslope METHODOLOGY: Expert Panel FOOTNOTES: A Zegeer et al. - 1988 ( 84 ). AMF Level of Predictive Certainty: Medium-High REFERENCES: NCHRP Project 17-25 research results ACCIDENT MODIFICATION FACTORS COMMENTS: • Original study conducted by Zegeer et al. A used log linear regression models to develop estimates of the effects of sideslope on single-vehicle crashes and total crashes on rural two-lane roads. The AMFs shown were derived from these models. • The NCHRP Projects17-25/17-29 expert panel on rural multilane highways concluded that the AMFs derived were valid and the best available for both rural two-lane roads and rural multilane highways.

TREATMENT: Add/Remove On-Street Parking METHODOLOGY: Analysis-Driven Expert Panel where: ACCIDENT MODIFICATION FUNCTION AMF Level of Predictive Certainty: Medium-High REFERENCE: Bonneson, Zimmerman, and Fitzpatrick - 2005 ( 91 ) FOOTNOTES: A Bonneson and McCoy - 1997 ( 92 ). B McCoy et al. - 1990 ( 93 ). C Bonneson, Zimmerman, and Fitzpatrick - 2005 ( 91 ). D Box - 2002 ( 94 ). AMF = 1 + P pk (B pk - 1) P pk = proportion of curb length with on-street parking (= 0.5 L pk /L) L pk = curb length with on-street parking (mi) L = roadway segment length (mi) B pk = (1.10 + 0.365 I u2 + 0.609 P b/ o )[(f ap/pp 1.0) P ap + 1.0] where: I u2 = cross-section indicator variable (two-lane street = 1; otherwise = 0) P b/ o = proportion of street with parking that has business or office as adjacent land use f ap/pp = ratio of crashes on streets with angle parking to crashes on streets with parallel parking (= 2.34; see comment) P ap = for that part of the street with parking, the proportion of the street with angle parkin g COMMENTS: • Expert panel on urban/suburban arterials considered this AMF function to be the best estimate for the addition or removal of on-street parking. • AMF was derived from a negative binomial regression model (Bonneson and McCoy) A and from other prior study data (McCoy et al.). B • Value for the ratio of crashes on streets with angle parking to crashes on streets with parallel parking (f ap/pp ) derived by Bonneson et al. C to be 2.34 on the basis of data from McCoy et al. B and Box. D

TREATMENT: Install Raised Medians at Crosswalks METHODOLOGY: Matched Comparison Marked Crosswalks* Unmarked Crosswalks* STUDY SITES: • 2,000 sites were included in the study to evaluate the effect of marked vs. unmarked crosswalks (1,000 matched pairs of each type), • 260 of these sites were on multilane roads and had raised medians. • On average, 5 years of crash data were collected for each site, as well as traffic data and pedestrian volume estimates. REFERENCE: Zegeer et al. - 2001 ( 38 ) COMMENTS: • The AMFs were computed from the pedestrian crash rates (pedestrian crashes per million crossings) for sites with medians versus the sites without medians. 0.61 * Applicable to urban and suburban multilane roads (up to eight lanes) with traffic volumes greater than 15,000 vpd. AMF Level of Predictive Certainty: Medium-High AMF 0.54 Total Pedestrian Accidents (all severities) No. of Median Sites 173 CRASH TYPE STUDIED AND ESTIMATED EFFECTS TREATMENT: Add Roadway Segment Lighting METHODOLOGY: Meta-analysis/Expert Panel Total Crashes All Injury Crashes Total Crashes All Injury Crashes CRASH TYPE STUDIED AND ESTIMATED EFFECTS AMF Level of Predictive Certainty: Medium-High REFERENCE: Elvik and Vaa - 2004 (60 ); NCHRP 17-25 Final Report; NCHRP 17-26 Final Report COMMENTS: • The meta-analysis results produced AMF estimates for reductions in fatal, injury and property-damage-only accidents of 0.36, 0.72, and 0.83, respectively.A • The NCHRP 17-25/17-26 expert panel on urban/suburban arterials recommended that the meta-analysis results be applied to roadway segments and that the fatal and injury results be combined into a single AMF for all levels of injury. • The NCHRP 17-26 Final Report includes a distribution of crashes by time of day and injury severity for several roadway classes. • AMFs shown represent the mean estimates for all roadway classes and were derived on the basis of these distributions and the meta-analysis AMFs. FMAsehsarC emitthgiN 0.80 0.71 STUDY SITES: • 38 studies were evaluated as part of the meta-analysis, including 14 U.S. studies.A • Distributions of crashes by injury severity and time of day were obtained from the HSIS data for the states of Minnesota and Michigan. FOOTNOTES: A Elvik and Vaa (60 ) FMAsehsarC llA 0.94 0.92

TREATMENT: Reduce Mean Travel Speed METHODOLOGY: Reanalysis of Existing Data AMF Level of Predictive Certainty: Medium-High REFERENCES: NCHRP Project 17-25 research results ACCIDENT MODIFICATION FACTORS COMMENTS: • Original study conducted by Elvik and colleagues used the Power Model to develop estimates of the effectiveness of changes in mean travel speeds. Data included mean speed change and the related crash-frequency change from 97 published international studies containing 460 results. Each result contained information on mean speed and crash frequency before treatment and mean speed and crash frequency after treatment. A • The NCHRP Projects 17-25/17-26 expert panel reviewed the original study and requested supplemental analysis to explore the validity of the results and to develop AMFs. NOTE: This series of AMFs is related to any treatment that is associated with a changed mean speed (e.g., changes in enforcement or installation of traffic calming measures). FOOTNOTES: A Elvik, Christensen, and Amundsen - 2004 ( 74 ).