Development of Accident Modification Factors for Rural Frontage Road Segments in Texas

Similar documents
Safety Impact of Gateway Monuments

Report No. FHWA/LA.13/508. University of Louisiana at Lafayette. Department of Civil and Environmental Engineering

Crash Frequency and Severity Modeling Using Clustered Data from Washington State

Evaluation of a Center Pivot Variable Rate Irrigation System

Cost Effective Safety Improvements for Two-Lane Rural Roads

Pedestrian Crash Prediction Models and Validation of Effective Factors on Their Safety (Case Study: Tehran Signalized Intersections)

Risk analysis of natural gas pipeline

PERFORMANCE AND COMPENSATION ON THE EUROPEAN PGA TOUR: A STATISTICAL ANALYSIS

Driver s Decision Model at an Onset of Amber Period at Signalised Intersections

The impact of foreign players on international football performance

Engineering Analysis of Implementing Pedestrian Scramble Crossing at Traffic Junctions in Singapore

OWNERSHIP STRUCTURE IN U.S. CORPORATIONS. Mohammad Rahnamaei. A Thesis. in the. John Molson School of Business

Reduced drift, high accuracy stable carbon isotope ratio measurements using a reference gas with the Picarro 13 CO 2 G2101-i gas analyzer

VOLUME TRENDS NOVEMBER 1988 TRAVEL ON ALL ROADS AND STREETS IS FOR NOVEMBER 1988 AS COMPARED UP BY 3.4 PERCENT TO NOVEMBER 1987.

First digit of chosen number Frequency (f i ) Total 100

ADDITIONAL INSTRUCTIONS FOR ISU SYNCHRONIZED SKATING TECHNICAL CONTROLLERS AND TECHNICAL SPECIALISTS

Methodology for ACT WorkKeys as a Predictor of Worker Productivity

M.H.Ahn, K.J.Lee Korea Advance Institute of Science and Technology 335 Gwahak-ro, Yuseong-gu, Daejeon , Republic of Korea

11. Contract or Grant No. Lubbock, Texas

Coastal Engineering Technical Note

Planning of production and utility systems under unit performance degradation and alternative resource-constrained cleaning policies

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore.

GAS-LIQUID INTERFACIAL AREA IN OXYGEN ABSORPTION INTO OIL-IN-WATER EMULSIONS

Equilibrium or Simple Rule at Wimbledon? An Empirical Study

Major League Duopolists: When Baseball Clubs Play in Two-Team Cities. Phillip Miller. Department of Economics. Minnesota State University, Mankato

Pedestrian Facilities Planning on Tianjin New Area program

ITRS 2013 Silicon Platforms + Virtual Platforms = An explosion in SoC design by Gary Smith

Recreational trip timing and duration prediction: A research note

Comparisons of Means for Estimating Sea States from an Advancing Large Container Ship

The Initial Phases of a Consistent Pricing System that Reflects the Online Sale Value of a Horse

Muscle drain versus brain gain in association football: technology transfer through

Numerical Study of Occupants Evacuation from a Room for Requirements in Codes

ALASKA DEPARTMENT OF FISH AND GAME DIVISION OF COMMERCIAL FISHERIES NEWS RELEASE

2017 GIRLS DISTRICT-SPECIFIC PLAYER DEVELOPMENT GUIDE

JIMAR ANNUAL REPORT FOR FY 2001 (Project ) Project Title: Analyzing the Technical and Economic Structure of Hawaii s Pelagic Fishery

Analysis of Hold Time Models for Total Flooding Clean Extinguishing Agents

Keywords: Ordered regression model; Risk perception; Collision risk; Port navigation safety; Automatic Radar Plotting Aid; Harbor pilot.

'!' CORDOVA BRANDON GREEN

Decomposition guide Technical report on decomposition

Evaluating the Effectiveness of Price and Yield Risk Management Products in Reducing. Revenue Risk for Southeastern Crop Producers * Todd D.

COMPENSATING FOR WAVE NONRESPONSE IN THE 1979 ISDP RESEARCH PANEL

A Study on Parametric Wave Estimation Based on Measured Ship Motions

Aalborg Universitet. Published in: 9th ewtec Publication date: Document Version Publisher's PDF, also known as Version of record

Lake Clarity Model: Development of Updated Algorithms to Define Particle Aggregation and Settling in Lake Tahoe

1.1 Noise maps: initial situations. Rating environmental noise on the basis of noise maps. Written by Henk M.E. Miedema TNO Hieronymus C.

Experimental And Numerical Investigation Of The Flow Analysis Of The Water-Saving Safety Valve

2017 GIRLS CENTRAL DISTRICT PLAYER DEVELOPMENT GUIDE

Beating a Live Horse: Effort s Marginal Cost Revealed in a Tournament

Modeling the Performance of a Baseball Player's Offensive Production

BETHANY TAX INCREMENT FINANCING DISTRICT NO. 1 NOTICE OF TWO PUBLIC HEARINGS

Johnnie Johnson, Owen Jones and Leilei Tang. Exploring decision-makers use of price information in a speculative market

John Keho, AICP, Interim Director. Bianca Siegl, Long Range and Mo y nning Manager~ Walter Davis, Traffic Specialist ~

Dynamic Analysis of the Discharge Valve of the Rotary Compressor

A PROBABILITY BASED APPROACH FOR THE ALLOCATION OF PLAYER DRAFT SELECTIONS IN AUSTRALIAN RULES

A comparison study on the deck house shape of high speed planing crafts for air resistance reduction

Degassing of deep groundwater in fractured rock

2018 GIRLS DISTRICT-SPECIFIC PLAYER DEVELOPMENT GUIDE

Transportation Research Forum

Wave Breaking Energy in Coastal Region

Mechanical Engineering Journal

ACCIDENT MODIFICATION FACTORS FOR MEDIANS ON FREEWAYS AND MULTILANE RURAL HIGHWAYS IN TEXAS

Sports Injuries in School Gaelic Football: A Study Over One Season

Evaluating Rent Dissipation in the Spanish Football Industry *

Aerator Performance in Reducing Phenomenon of Cavitation in Supercritical Flow in Steep Channel Bed

Incidence and Risk Factors for Concussion in High School Athletes, North Carolina,

Proceedings of the ASME nd International Conference on Ocean, Offshore and Arctic Engineering OMAE2013 June 9-14, 2013, Nantes, France

Impact of Intelligence on Target-Hardening Decisions

Aalborg Universitet. Published in: 9th ewtec Publication date: Document Version Accepted author manuscript, peer reviewed version

High Speed 128-bit BCD Adder Architecture Using CLA

Price Determinants of Show Quality Quarter Horses. Mykel R. Taylor. Kevin C. Dhuyvetter. Terry L. Kastens. Megan Douthit. and. Thomas L.

Free Ride, Take it Easy: An Empirical Analysis of Adverse Incentives Caused by Revenue Sharing

Monitoring Physical Activity from Active Transport. Dr Russell G. Thompson Institute of Transport Studies Monash University

Canadian Journal of Fisheries and Aquatic Sciences. Seasonal and Spatial Patterns of Growth of Rainbow Trout in the Colorado River in Grand Canyon, AZ

SECOND-ORDER CREST STATISTICS OF REALISTIC SEA STATES

LSSVM Model for Penetration Depth Detection in Underwater Arc Welding Process

Hedonic Price Analysis of Thoroughbred Broodmares in Foal

Study on coastal bridge under the action of extreme wave

Availability assessment of a raw gas re-injection plant for the production of oil and gas. Carlo Michelassi, Giacomo Monaci

Referee Bias and Stoppage Time in Major League Soccer: A Partially Adaptive Approach

Product Information. Radial gripper PRG 52

For models: 660 EF/EFO

CS 2750 Machine Learning. Lecture 4. Density estimation. CS 2750 Machine Learning. Announcements

What does it take to be a star?

OPTIMAL LINE-UPS FOR A YOUTH SOCCER LEAGUE TEAM. Robert M. Saltzman, San Francisco State University

English Premier League (EPL) Soccer Matches Prediction using An Adaptive Neuro-Fuzzy Inference System (ANFIS) for

Peak Field Approximation of Shock Wave Overpressure Based on Sparse Data

Evolutionary Sets of Safe Ship Trajectories: Evaluation of Individuals

Terminating Head

An intro to PCA: Edge Orientation Estimation. Lecture #09 February 15 th, 2013

Sectoral Business Cycle Synchronization in the European Union *

Product Information. Long-stroke gripper PFH-mini

PREDICTION OF POLYDISPERSE STEAM BUBBLE CONDENSATION IN SUB-COOLED WATER USING THE INHOMOGENEOUS MUSIG MODEL

IDENTIFICATION OF TRANSPORTATION IMPROVEMENT PROJECTS IN PHNOM PENH CONSIDERING TRAFFIC CONGESTION LEVEL

Product Information. Long-stroke gripper PSH 42

DETECTION AND REFACTORING OF BAD SMELL

Valuing Beach Quality with Hedonic Property Models

PREDICTIONS OF CIRCULATING CURRENT FIELD AROUND A SUBMERGED BREAKWATER INDUCED BY BREAKING WAVES AND SURFACE ROLLERS. Yoshimitsu Tajima 1

Heart rates during competitive orienteering

Cross-shore Structure of Longshore Currents during Duck94

Seabed type clustering using single-beam echo sounder time series data

Transcription:

Development of Accdent Modfcaton Factors for Rural Frontage Road Segments n Texas Domnque Lord* Zachry Department of Cvl Engneerng & Center for Transportaton Safety Texas Transportaton Insttute Texas A&M Unversty System 3135 TAMU, College Staton, TX, 77843-3135 Tel. (979) 458-3949 Fax: (979) 845-6481 d-lord@tamu.edu James A. Bonneson, P.E. Texas Transportaton Insttute Texas A&M Unversty System 3135 TAMU, College Staton, TX, 77843-3135 Tel. (979) 845-9906 Fax: (979) 845-6254 j-bonneson@tamu.edu Paper presented at the 86 th Annual Meetng of the Transportaton Research Record November 12 th, 2006 * Correspondng author Word Count: 5,339 + 4 Tables + 5 Fgures = 7,589 words

ABSTRACT Frontage roads are most frequently used n rural and urban envronments along freeway and full-controlled prncpal arteral corrdors. Ther prmary functon s to dstrbute and collect traffc between local streets and nterchanges. They have been the prmary desgn soluton for provdng access along Texas rural freeways and full-controlled prncpal arterals. Wth the growng publc demand for safer streets and hghways, state and natonal transportaton agences have developed safety programs that emphasze publc educaton, accelerated hghway renewal, communty-senstve street systems, and nnovatve technology to facltate safe hghway desgn practces. Unfortunately, there currently exst no relable tools, ncludng the ones proposed n the upcomng Hghway Safety Manual, that specfcally address the safety performance of rural frontage roads. The orgnal research on whch ths paper s based s amed at developng a safety performance functon (SPF) for rural one- and two-way frontage roads n Texas and, through the modelng effort, estmate accdent modfcaton factors (AMFs) for quantfyng the relatonshp between changes n hghway geometrc desgn characterstcs and frontage road safety. To accomplsh the objectves of ths study, an SPF was estmated usng data collected on frontage roads located along rural freeways n central Texas. The fndngs from ths research show that wder lane and shoulder wdths are assocated wth a reducton n segment-related collsons. In addton, the data suggest that edge markng presence has a sgnfcant mpact on the safety performance of rural two-way frontage roads. However, the magntude of the crash reducton due to markng presence was sgnfcant and beleved to overstate the true beneft of such markngs. The results also show that the SPF developed for ths research ndcates that rural frontage road segments experence about the same number of severe crashes as typcal rural twolane hghways for the same traffc volume. Dfferences n turnng volume and weavng actvty on these two faclty types may explan the subtle dfferences noted n the SPF estmates for the two faclty types.

Lord & Bonneson 1 INTRODUCTION Frontage roads are most frequently used n rural and urban envronments along freeway and full-controlled prncpal arteral corrdors. Ther prmary functon s to dstrbute and collect traffc between local streets and nterchanges. They also serve other purposes dependng on the type of facltes they serve and the characterstcs of the surroundng areas. For nstance, frontage roads can be used to control access, provde access to adjacent propertes, and mantan crculaton on each sde of the arteral (1). In general, they usually run parallel to the man traveled way, and may be provded on both sdes of the prncpal arteral. Dependng on the characterstcs of the adjacent land, frontage roads can operate as a one-way or two-way faclty. Frontage roads have been the prmary desgn soluton for provdng access along Texas rural freeways and full-controlled prncpal arterals. As of 2000, the State of Texas had more than 4,510 mles of urban and rural frontage roads (2). The Texas Department of Transportaton (TxDOT) has developed several gudelnes for desgnng rural frontage roads (3). The gudelnes address varous mportant desgn elements, such as selectng the desgn speed, lane and shoulder wdths, and the type of operaton (.e., one-way or two-way). Wth the growng publc demand for safer streets and hghways, state and natonal transportaton agences have developed safety programs that emphasze publc educaton, accelerated hghway renewal, communty-senstve street systems, and nnovatve technology to facltate safe hghway desgn practces. Recent effort has been devoted to ncorporate safety n a quanttatve manner wthn the hghway desgn process (4, 5). In fact, the forthcomng Hghway Safety Manual (HSM) (6) wll specfcally provde tools for estmatng the safety performance of several types of hghway facltes. Unfortunately, the frst edton of the HSM wll not be addressng the safety performance of rural frontage roads. Ths paper descrbes the development of quanttatve tools for estmatng the safety of rural frontage road segments on Texas hghways. These tools nclude a safety performance functon (SPF) and several accdent modfcaton factors (AMFs). The SPF quantfes the relatonshp between the crash frequency on a frontage road segment and ts length and average daly traffc volume (ADT). An AMF quantfes the relatonshp between a change n a specfc hghway geometrc desgn element (e.g., lane wdth) and safety. The focus of ths research s the geometrc desgn factors that nfluence the safety of twoway and one-way frontage road segments n rural areas. The tools descrbed heren are ntended to help the desgner evaluate alternatve geometrc desgn element dmensons, such as a lane wdth of 11 ft versus 12 ft, n terms of ther mpact on safety. These tools do not address the safety of ramp/frontage-road termnals or the safety of frontage-road/crossroad ntersectons. Moreover, they are not ntended to be used to drectly address questons related to the safety mpact of one-way frontage-road operaton versus two-way frontage road operaton. The remander of ths paper s dvded nto sx parts. The frst part summarzes the lterature on the safety performance of frontage roads. The second part descrbes the segment selecton and data collecton actvtes. The thrd part explans the statstcal analyss procedure used to develop the SPF. The fourth part descrbes the frontage road AMFs derved from the

Lord & Bonneson 2 data. The ffth part descrbes the frontage road SPF that was calbrated usng the data. A summary of the work s provded n the last part. BACKGROUND There has been lttle research nto the safety performance of rural or urban frontage roads. Only two research projects could be dentfed that specfcally evaluated the safety performance of frontage roads. The frst project was conducted by Woods and Chang (7). They evaluated the change n safety at nne frontage road stes followng ther converson from twoway to one-way operaton. These authors found a reducton of about 20 percent n crash frequency followng the frontage road converson. Jacobson et al. (8) analyzed crash data occurrng near frontage-road entrance ramps at several locatons n Texas. They used ths data to determne the dstance necessary to mnmze the speed dfferental between vehcles enterng the entrance ramp and those contnung on the frontage road. Ther objectve was to develop gudelnes for determnng the mnmum dstance between ramp termnals and commercal drveways when located on one-way frontage roads. Jacobson et al. collected data at fve stes n Austn and San Antono. They reported that segments wth drveways located wthn 100 ft of the ramp termnal had sgnfcantly hgher crash rates than segments wth drveways located further away from the termnal. They recommended an absolute mnmum of 100 ft upstream (200 ft s more desrable) of the ramp termnal where no drveways should be located. They also recommended mantanng a drveway-free secton equal to 50 ft downstream from the ramp termnal. No frontage-road-based AMFs were specfcally dentfed n the lterature. AMFs that have been developed for rural two-lane hghways could arguably be used to for rural frontageroads gven ther smlar area-type, hgh-speed character, and two-lane cross secton. AMFs for rural two-lane hghways have been documented n several reports (5, 6). However, the frontage road s dfferent from the two-lane hghway because t has restrcted access along at least one sde of the road, a hgher percentage of turnng traffc, and perodc ramp-frontage-road termnals wth yeld control. As a result of these dfferences, a gven desgn element s lkely to have a dfferent effect on frontage-road safety than on two-lane hghway safety. For ths reason, there s a need to develop AMFs for rural frontage roads. DATA COLLECTION ACTIVITIES Ths secton descrbes the data collecton actvtes undertaken to assemble a database sutable for developng a frontage road SPF and assocated AMFs. The frst sub-secton outlnes the crtera used n the segment selecton process. The second sub-secton descrbes the characterstcs of the crash data. The thrd sub-secton descrbes the process used to collect traffc flow and geometry data. The last sub-secton explans how the data were assembled and analyzed.

Lord & Bonneson 3 Selecton Process Fgure 1 shows the frontage road segment that formed the bass for the analyss. All frontage road segments consdered for ncluson n the database were located between successve nterchanges. Also, each segment selected was requred to have at least one ramp termnal along ts length (however, the ramp termnal area was not consdered to be part of the segment). Interchange Interchange Frontage Road Segment (excludng ramp termnals) Fgure 1. Frontage Road Analyss Segment. Four Texas hghway corrdors were consdered for segment selecton. In all cases, only those frontage-road segments n rural areas were consdered for ncluson n the database. The frst corrdor was located along I-35 between the Cty of Georgetown and the locaton where I-35 splts between I-35E and I-35W north of Waco (segments n the vcnty of the Ctes of Temple and Waco were excluded). The second corrdor was located along S.H. 6/S.H. 190 near the Ctes of Bryan/College Staton. The thrd study corrdor was located on I-10 between the Ctes of Gldden and Brookshre. The last study corrdor was located on I-45 between the Ctes of Wlls and Centervlle. The study corrdors collectvely contaned a mx of one-way and two-way frontage roads. A total of 141 segments were ultmately dentfed from a revew of varous maps and aeral photos. Each segment was subsequently vsted to collect addtonal data not avalable from other sources. A dstance measurng nstrument (DMI) was used to measure segment length and the locaton of drveways and ramp termnals. After screenng the ntal 141 segments for data avalablty and constructon actvty, the sample sze was reduced to 123 segments. The characterstcs of these segments are summarzed n Table 1. Crash Data Crash data for each frontage road segment were extracted from the Department of Publc Safety (DPS) electronc database. A total of fve years of crash data (1997 to 2001) were used. Only crashes that were segment-related were ncluded n the database assembled for ths research. Crashes that were related to the ramp/frontage-road termnals or that were related to the frontage-road/crossroad ntersectons are referred to as non-segment-related crashes and were excluded. These crashes were ratonalzed to be strongly nfluenced by the desgn of the

Lord & Bonneson 4 termnal, or ntersecton, and would not be helpful n dentfyng correlaton between segment desgn and segment crash frequency. Non-segment-related crashes were defned to be those crashes dentfed as At-Intersecton or Intersecton Related n the DPS database. Table 1. Frontage-Road Segment Physcal Characterstcs. ADT, veh/day Hghway Operaton Number of Segments Percent Segments wth Edge Delneaton Segment Length, mles Ave. Mn. Max. Left Rght Ave. Mn. Max. S.H. 6/ Two-way 11 2,360 110 6,168 73 73 2.00 1.06 2.66 S.H. 190 One-way 20 2,550 140 5,270 100 10 1.12 0.78 1.89 I-10 Two-way 16 675 168 1,585 25 25 2.74 1.36 5.34 Two-way 57 575 125 2,199 33 33 1.97 0.69 3.76 I-35 I-45 Summary One-way 6 790 361 1,046 100 100 1.93 1.13 2.64 Two-way 10 1,990 218 1,988 50 40 2.40 0.79 4.21 One-way 3 4,470 3,093 5,766 100 100 1.26 1.00 1.77 Two-way 94 2,385 110 6,168 38 37 2.15 0.69 5.34 One-way 29 940 140 5,766 100 38 1.30 0.78 2.64 Overall 123 1,230 110 6,168 53 37 1.92 0.69 4.21 Crashes for all types of severty were extracted from the Department of Publc Safety (DPS) database. The severty classes consdered nclude: fatal (K), ncapactatng-njury (A), non-ncapactatng njury (B) and mnor njury (C) and property damage only (O). A summary of the crash data characterstcs s presented n Table 2. The data n ths table correspond to crashes that were reported to have occurred on the frontage-road segment and that were determned to be segment-related crashes. The numbers lsted n Table 2 are based on crashes that occurred durng a fve-year perod.

Lord & Bonneson 5 Table 2. Frontage-Road Segment Crash Characterstcs. 1 Hghway Operaton Severe (KABC) Crash Freq., 2 Total (KABCO) Crash Freq., 2 Total Crashes Per Segment 2 Ave. Mn. Max. Total Crash Rate, 3 cr/mvm S.H. 6/ Two-way 11 19 1.73 0 3 0.20 S.H. 190 One-way 23 33 1.65 0 7 0.31 I-10 Two-way 8 15 0.94 0 3 0.34 Two-way 44 72 1.26 0 5 0.60 I-35 I-45 Summary One-way 8 9 1.50 0 3 0.54 Two-way 17 21 2.10 0 6 0.59 One-way 13 17 5.67 4 8 0.59 Two-way 80 127 1.35 0 6 0.43 One-way 44 59 2.03 0 8 0.39 Overall 124 186 1.51 0 8 0.42 Notes: 1 Crash data apply to frontage-road segments and do not nclude crashes that may have occurred at ramp/frontageroad termnals or at frontage-road/crossroad ntersectons. 2 Crash frequences lsted n columns 3 through 7 are based on a fve-year perod. 3 Crash rate has unts of total crashes per mllon vehcle mles (cr/mvm). As ndcated n Table 2, there were 186 total crashes that occurred on the 123 frontageroad segments durng the fve years for whch crash data were avalable. Many of the segments experenced no crashes durng ths tme perod. One segment experenced eght crashes durng ths perod. Severe crashes accounted for 124 of the 186 crashes, or about 67 percent of all crashes (note: PDO collsons are probably under reported, but the magntude s unknown). Although not shown n the table, the severe crash rate for frontage road segments s 0.28 cr/mvm. Ths rate s slghtly hgher than the severe crash rate of 0.20 cr/mvm found for the typcal rural two-lane hghway (5). The slght ncrease may be due to the more complcated envronment of the rural frontage road (.e., ramp termnals, concentrated weavng). Supplemental Data Collecton Supplemental data were collected to facltate the statstcal examnaton of factors that may nfluence segment crashes. The data collected nclude: traffc counts, lane wdth, paved shoulder wdth, presence of pavement edge lne markngs (no rased pavement markers), presence of curb, and the number of prvate and commercal drveways. The characterstcs for some of the data are summarzed n Table 3. The physcal characterstcs of the hghway sectons were shown n Table 1.

Lord & Bonneson 6 Statstcs Paved Left- Shoulder Wdth, ft Table 3. Frontage-Road Segment Supplementary Data. Prvate Commercal Lane Wdth, ft Paved Rght- Drveways Drveways Shoulder Wdth, (dr/mle) (dr/mle) ft Two- One- Two- One- Two- One- Two- One- Operaton Two- One- Way Way Way Way Way Way Way Way Way Way Ave. 2.10 1.44 2.55 5.50 10.5 11.7 1.3 1.3 1.1 2.4 Mn 0 0 0 0.53 9 10 0 0 0 0 Max. 9.34 12.33 10.63 21.04 13 13 9 8 9 7 Sum 421 54 466 207 Traffc counts were extracted from TxDOT s Road-Hghway Inventory Network (RHNo) database for the years 1997 to 2001. RHNo counts were avalable for most of the 123 segments. Automatc traffc counters were used to obtan counts for those frontage roads not n the RHNo database. All traffc counts were collected for a 24-hour tme perod. They were then adjusted to yeld an estmate of the annual average daly traffc for the mdpont year of the 5- year perod correspondng to 1997 to 2001. Database Assembly Once the crash and the supplemental data were collected for each ste located along the study corrdors, the data were combned nto one database. The database was formatted to be mported n Genstat (9), the statstcal software program used to develop the statstcal models. STATISTICAL MODEL DEVELOPMENT The statstcal analyss of the database conssted of developng a safety performance functon relatng the reported crash frequency to the measured ste characterstcs. Several alternatve model forms were tested. However, only the model that provded the best combnaton of good ft to the data and a logcal relatonshp between the ndependent and dependent varables s descrbed n ths secton. mean The number of crashes at the -th rural frontage road segment, Y, when condtonal on ts µ, s assumed to be Posson dstrbuted and ndependent over all segments as: Y µ ~ Po( µ ) = 1, 2,, I and t = 1, 2,, T (1) t t t The mean of the Posson s structured as: where, µ = f ( X; β)exp( e ) (2) t t f (.) s the predctve model represented as a functon of the varables (X); β s a vector of unknown coeffcents; and, e t s the random error.

Lord & Bonneson 7 It s usually assumed that exp( e t ) s ndependent and Gamma dstrbuted wth a mean equal to 1 and a varance equal to 1 / φ for all (wth φ > 0). Wth ths characterstc, t can be shown that Y, condtonal on f (.) and φ, s dstrbuted as a Negatve Bnomal (NB) (or Posson-gamma) random varable wth a mean f (.) and a varance f (.)( 1+ f (.) / φ), respectvely. The term φ s usually defned as the "nverse dsperson parameter" for the NB dstrbuton. If φ, then the dstrbuton converges to a full Posson dstrbuton and a Posson regresson model was used for estmatng the predctve model. An mportant characterstc assocated wth the development of statstcal relatonshps s the choce of the functonal form lnkng the crashes to the varables. For ths work, the selected functonal form for the predctve model s: where, n x β β1 = 2 µ = β L F e (3) 0 µ = the estmated number of crashes per year for segment ; F = vehcles per day (both ways for two-way operatons) (ADT) for segment ; L = length of segment n mles; x = a seres of varables; and, β 0, β,, β n = coeffcents to be estmated. It was decded to use the segment length varable as an offset varable n Equaton 3, as opposed to a varable assocated wth an exponental regresson coeffcent (e.g., as s the case for the flow varable F). Ths approach was taken because segment length s consdered to be drectly related to segment crash frequency (.e., the number of crashes on a segment ncreases n drect proporton to the ncrease n ts length). Hence, no emprcal adjustment s beleved to be needed for the length varable. The coeffcents n the model were estmated usng Genstat (9). At the start of the modelng effort, several NB regresson models were attempted usng one-way and two-way frontage roads together n one model, and then n separate models. Due to the low sample mean values and small sample sze, some models dd not provde reasonable results. As explaned by Lord (10), models developed usng datasets subjected to these characterstcs can show sgnfcant sgns of nstablty durng the model estmaton process. In fact, the data may exhbt over-dsperson, but ths characterstc cannot be captured by a NB regresson model (see [10] for addtonal nformaton). Based on these consderatons, t was decded to use a Posson regresson model for fttng the frontage road data, snce the model output of the NB regresson model showed an nverse dsperson parameter equal to nfnty. Durng the model development, each varable was added to the model one at the tme. Intally, predctve models were developed separately for the one-way and two-way frontage road types. However, t was found that the regresson coeffcents for common varables n these

Lord & Bonneson 8 two models were not sgnfcantly dfferent from one another. For ths reason, the data were combned and one predctve model was developed. Indcator varables were used n ths model whenever the effect of a specfc varable was found to be correlated wth frontage road type. Table 4 summarzes the coeffcent values assocated wth the calbrated model. These coeffcents correspond to a model that predcts total (.e., KABCO) crashes on rural frontage road segments. An addtonal model was ft to the severe crash data; however, the elmnaton of property-damage-only crashes from the database contrbuted further to the low sample mean and small sample sze problems noted prevously. For ths reason, further efforts to calbrate a model usng severe crash frequency were abandoned. The coeffcent values lsted n Table 4 ndcate the nature of the correlaton between the correspondng varable and crash frequency. Specfcally, postve coeffcent values ndcate that an ncrease n the varable value correlates wth an ncrease n crash frequency (and negatve values correlate wth a decrease n crashes). For example, the coeffcent of -0.188 assocated wth the Lane Wdth varable ndcates that an ncrease n lane wdth s assocated wth a decrease n the number of crashes. The nteracton varable (Edge Markng Presence x Two- Way Operaton) shows that the presence of a pavement edge lne s negatvely assocated wth the number of crashes for two-way frontage roads. It suggests that the addton of edge lnes to a two-lane frontage road reduces crash frequency. A smlar effect was not found for one-way frontage road segments n the assembled database; however, ntuton would suggest t s lkely to exst. Table 4. Frontage-Road Segment Safety Performance Functon for Total Crashes. (Posson Regresson) Model Varables Coeffcent Value (standard devaton) ln β ) -3.85 (1.11) Intercept ( [ ] 0 Log(ADT) ( β 1 ) 0.641 (0.0857) Lane Wdth ( β 2 ), ft -0.188 (0.104) Combned Shoulder Wdth a ( β 3 ), ft -0.035 (0.028) Edge Markng Presence x Two-Way Operaton b ( β 4 ) -0.518 (0.203) Summary Statstcs Scaled Devance = 143.0 (F=15.26) a Combned Shoulder Wdth = paved left shoulder wdth + paved rght shoulder wdth b Edge Markng Presence = Left edge markng (0.5 = yes, 0 = no) + rght edge markng (0.5= yes, 0 = no) b Two-Way Operaton (1 = yes, 0 = no) The coeffcent values lsted n Table 4 can be substtuted nto Equaton 3 to yeld the followng calbrated model:

Lord & Bonneson 9 0.641 ( 0.188LW 0.035SW 0.518EM I2 ) e µ = 0.021 L F (4) where, LW = average lane wdth, ft; SW = combned paved shoulder wdth (left + rght shoulder), ft; EM = proporton of segment wth pavement edge markngs (both drectons); and I 2 = ndcator varable (= 1.0 for two-way operaton, 0.0 for one-way operaton). The varable EM used n Equaton 4 s a proporton that vares from 0 to 1.0. It represents the length of edge lne on the rght sde of the roadway plus the length of edge lne on the left sde of the roadway, all of whch s dvded by twce the segment length. Thus, a 1.0 mle frontage road wth edge lnes only on the rght sde would have EM = 0.5 (= [1.0 + 0.0]/[2 1.0]). ACCIDENT MODIFICATION FACTORS Three AMFs were derved from the frontage-road segment database. Alternatve approaches for developng AMFs from cross secton data, such as that proposed by Washngton et al. (11), were examned and ruled out due to the small number of crashes n the database. Consequently, the AMFs were estmated drectly from the coeffcents of the model, as lsted n Table 4. Ths approach for AMF development assumes that each model varable s ndependent and, thus, not nfluenced by the value of any other varable. It also assumes that the relatonshp between the change n the varable value and the change n crash frequency s exponental (as suggested by Equatons 3 or 4). A more rgorous study desgn and a larger database (.e., one wth more segments) would be needed to test the valdty of these assumptons. However, experence n dervng AMFs n ths manner ndcates that the assumptons are reasonable and, wth thoughtful model development, the resultng AMFs can yeld useful nformaton about the frst-order effect of a gven varable on safety. AMF Lane Wdth The recommended AMF for lane wdth s: AMF LW ( 0.188 [ LW 12.0] ) = e (5) where, LW = average lane wdth, ft. The average lane wdth used n Equaton 5 represents the total wdth of all through traffc lanes on the frontage road dvded by the number of through lanes. The value of 12.0 n Equaton 5 reflects the base, or typcal, lane wdth condton. By defnton, t s assocated wth an AMF value of 1.0. The graphcal representaton of AMF LW s shown n Fgure 2. The relatonshp between lane wdth and AMF value shown n ths fgure suggests that crash frequency s reduced about 17

Lord & Bonneson 10 percent (.e., 1 e 0.188 ) for a 1-ft ncrease n lane wdth. Based on the range of lane wdths n the database, the lane wdth AMF s applcable to lane wdths rangng from 9 to 13 ft. 1.80 1.60 Rural Frontage-Road Segment Accdent Modfcaton Factor 1.40 1.20 Rural Two-Lane Hghway 1.00 0.80 9 9.5 10 10.5 11 11.5 12 12.5 13 Lane Wdth, ft Fgure 2. AMF for Lane Wdth. Also shown n Fgure 2 s the lane wdth AMF for rural two-lane hghways (5). A comparson of ths AMF wth the lane wdth AMF for frontage roads suggests that lane wdth on a frontage road has a greater mpact on crash frequency than t does on a two-lane hghway. It s possble that ths trend stems from the relatvely hgh percentage of turnng traffc and the consderable weavng actvty that occurs on frontage roads (between the ramp termnals and the crossroad ntersecton), relatve to a two-lane hghway. Wder lanes on frontage road segments may provde some addtonal room for recovery when these turnng and weavng-related conflcts occur. AMF Shoulder Wdth The recommended AMF for shoulder wdth s: ( 0.070 [ ASW 1.5]) AMF SW = e (6) where, ASW = average paved shoulder wdth ([left shoulder wdth + rght shoulder wdth]/2), ft. Ths AMF s derved from Equaton 4; however, the assocated regresson coeffcent (.e., -0.035) has been doubled such that the resultng AMF s based on the average paved shoulder wdth. Ths average s computed as the sum of the left and rght shoulder wdths dvded by 2.0. The value of 1.5 n Equaton 6 reflects the base, or typcal, average paved shoulder wdth condton.

Lord & Bonneson 11 The graphcal representaton of AMF SW s shown n Fgure 3. The relatonshp between shoulder wdth and AMF value shown n ths fgure suggests that crash frequency s reduced 7.0 percent for a 1-ft ncrease n shoulder wdth. Based on the range of shoulder wdths n the database, the shoulder wdth AMF s applcable to average shoulder wdths rangng from 0 to 9 ft. 1.40 1.20 Accdent Modfcaton Factor 1.00 Rural Two-Lane Hghway Rural Frontage-Road Segment 0.80 0.60 0 1 2 3 4 5 6 7 8 9 Average Shoulder Wdth, ft Fgure 3. AMF for Paved Shoulder Wdth. Also shown n Fgure 3 s the shoulder wdth AMF for rural two-lane hghways developed by Bonneson et al. (5). The base condton for ths AMF has been changed to a shoulder wdth of 1.5 ft to facltate comparson wth Equaton 6. As suggested by the trend lnes n ths fgure, shoulder wdth has a slghtly lower mpact on frontage road safety than on rural two-lane hghways. However, the dfference s somewhat subtle and may only be a result of random varaton n the database. AMF Edge Markng Presence on Two-Way Frontage Roads The AMF that was derved from varables assocated wth the presence of edge lne delneaton s: AMF EM e ( 0.518EM ) = (7) where, EM = proporton of segment wth pavement edge markngs (both drectons). Ths AMF was derved usng data for two-way frontage roads. An equvalent AMF for one-way frontage roads could not be derved. The AMF lkely explans the effect of edge lne

Lord & Bonneson 12 delneaton and other traffc control devces that are used to hghlght the two-way operaton (relatve to the more common one-way operaton) and to ensure correct drvng behavor. Equaton 7 reflects a base, or typcal, frontage road condton where there are no pavement edge lnes. A graphcal representaton of the AMF EM s shown n Fgure 4. Accdent Modfcaton Factor 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Proporton of Segment Wth Edge Lnes Fgure 4. AMF for Presence of Edge Lne Delneaton on Two-Way Rural Frontage Roads. The trend shown n Fgure 4 suggests that edge markngs can reduce crashes on two-way frontage roads by 40 percent f placed fully along both sdes of the frontage road. Ths reducton s relatvely large and suggests that ths AMF s explanng more than just the effect of pavement edge markng presence on crash frequency. The presence of pavement edge markngs s also lkely to be accompaned by addtonal warnng sgns that denote two-way operaton. Hence, the 40 percent reducton noted prevously s lkely a reflecton of the effectveness of the full complement of traffc control devces often deployed on two-way frontage roads to mtgate the ncreased potental for wrong way drvng. Gven ths speculaton and the lack of corroboratng evdence from other research projects, ths AMF cannot be recommended for evaluatng edge markng presence. However, t does provde some valdaton to the belef that a full complement of traffc control devces on two-way frontage roads wll reduce crashes, although the amount of reducton s uncertan at ths tme. SAFETY PERFORMANCE FUNCTION The SPF developed from the analyss s based on the regresson model n Equaton 4, after adjustment such that t can be used to obtan an estmate of severe crash frequency. Ths adjustment entaled multplyng the constant 0.021 n Equaton 4 by 0.67, where the multpler

Lord & Bonneson 13 0.67 reflects the fact that 67 percent of the crashes n the database correspond to severe crashes (as noted prevously n the dscusson of Table 2). 0.641 µ = 0.00134LF (8) where, µ = the estmated number of severe crashes per year for segment ; F = vehcles per day (both ways for two-way operatons) (ADT) for segment ; and L = length of segment n mles. Equaton 8 predcts the severe crash frequency that would be estmated for a frontage road segment wth 12 ft lanes and a combned paved shoulder wdth of 3.0 ft. In applcaton, the crash frequency predcted by Equaton 8 would be multpled by the AMFs for lane wdth and shoulder wdth (provded n the prevous secton) to estmate the severe crash frequency for a gven segment wth a specfed lane and shoulder wdth. The estmate obtaned from Equaton 8 does not nclude the crashes that would be attrbuted to the ramp/frontage-road termnal or the frontage-road/crossroad ntersecton. It also does not nclude any crashes that may occur on the man lanes that may ndrectly be attrbuted to the frontage road operaton or ts ramp desgn. Equaton 8 s compared n Fgure 5 wth the rural two-lane hghway SPF ncluded n the Interm Roadway Safety Desgn Workbook (5). The Workbook SPF s based on a severe crash rate of 0.20 cr/mvm. The trend lnes n the fgure ndcate that a frontage road experences slghtly more severe crashes than a rural two-lane hghway for ADTs that are less than 3500 veh/d. The reverse trend apples for ADTs greater than 3500 veh/d. It s possble that the ncreased turnng and weavng actvty assocated wth the frontage road (relatve to the two-lane hghway) may explan the slghtly hgher severe crash frequency on frontage roads for ADTs less than 3500 veh/d. As ADT exceeds 3500 veh/d, there may be less opportunty for turnng (.e., fewer gaps) and the weavng actvty may be more constraned (.e., lower speed) on the frontage road, such that frontage road crash frequency s lower than that found on two-lane hghways. The predcted values from the two SPFs shown n Fgure 5 are not statstcally dfferent from each other when a 95 th percentle confdence nterval s used. Hence, there s a small chance that the trend shown n Fgure 5 s a result only of random varaton n the data and that the two faclty types actually have a smlar severe crash frequency for segments.

Lord & Bonneson 14 0.40 Severe Crash Frequency, cr/m/yr 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 Rural Frontage Road Segments (Two-Way and One-Way) Rural Two-Lane Hghway Segments 0 1000 2000 3000 4000 5000 6000 Average Daly Traffc Volume, veh/d Fgure 5. Comparson between Frontage Road Segment SPF and Rural Two-Lane Hghway Segment SPF. SUMMARY AND CONCLUSIONS Ths paper examned the safety performance of rural frontage road segments. A SPF and three AMFs were derved from a statstcal model that was estmated usng data collected on rural frontage road segments. The varables that were found to have sgnfcant correlaton wth crash frequency nclude lane wdth, paved shoulder wdth, and for two-lane frontage roads, edge markng delneaton. The SPF and AMFs do not consder crashes that would be attrbuted to the ramp/frontage-road termnal or the frontage-road/crossroad ntersecton. Moreover, they do not consder crashes that occur on the man lanes that may ndrectly be related to wrong-way travel down an ext ramp. The fndngs from ths research have shown that wder lane and shoulder wdths are assocated wth a reducton n segment-related collsons. In addton, the data suggest that edge markng presence has a sgnfcant mpact on the safety performance of rural two-way frontage roads. However, the magntude of the crash reducton due to markng presence was sgnfcant and beleved to overstate the true beneft of such markngs. Addtonal research s needed to confrm the safety beneft assocated wth the pavement markngs and related control devces used on frontage roads. The SPF developed for ths research ndcates that rural frontage road segments experence about the same number of severe crashes as typcal rural two-lane hghways for the same traffc volume. Dfferences n turnng volume and weavng actvty on these two faclty types may explan the subtle dfferences noted n the SPF estmates for the two faclty types.

Lord & Bonneson 15 ACKNOWLEDGEMENTS The authors would lke to thank Ms. Elzabeth Hlton from TxDOT for provdng comments on an earler draft of the paper. REFERENCES 1. AASHTO. A Polcy on Geometrc Desgn of Hghways and Streets, Amercan Assocaton of State Hghway and Transportaton Offcals, 2004, Washngton, DC. 2. Kockelman, K.M., R. Machemehl, A.W. Overman, J. Sesker, M. Mand, J. Peterman, and S. Handy. Frontage Roads: Assessment of Legal Issues, Desgn Decsons, Costs, Operatons, and Land-Development Dfferences. Journal of Transportaton Engneerng, Vol. 129, No. 3, 2003, pp. 242-252. 3. TxDOT. Texas Roadway Desgn Manual, 2002, Austn, TX http://manuals.dot.state.tx.us/docs/coldesg/forms/rdw.pdf 4. Lord, D., and B.N. Persaud. Estmatng the Safety Performance of Urban Transportaton Networks. Accdent Analyss & Preventon. Vol. 36, No. 2, 2004, pp. 609-620. 5. Bonneson, J.A., K. Zmmerman, and K. Ftzpatrck. Interm Roadway Safety Desgn Workbook. FHWA/TX-06/0-4703-P4. Texas Department of Transportaton, Austn, Texas, Aprl 2006. 6. Hughes, W., K. Eccles, D. Harwood, I. Potts, and E. Hauer. Development of a Hghway Safety Manual. Appendx C: Hghway Safety Manual Prototype Chapter: Two-Lane Hghways. NCHRP Web Document 62 (Project 17-18(4)). Washngton, D.C. 2005. (avalable at http://www.hghwaysafetymanual.org/. accessed May 2006) 7. Woods, D.L., and M. Chang. Accdent Analyss of the Converson from Two-Way to One- Way Frontage Road Operaton. TTI Research Report 288-3. Texas Transportaton Insttute, College Staton, TX., 1983. 8. Jacobson, M.S., R. Arrendondo, and R.H. Henk. Development of Improved Gudelnes for Frontage Road Drveway Access at Entrance Ramp Locatons. TTI Research Report 2927-2. Texas Transportaton Insttute, College Staton, TX., 1999. 9. Payne, R.W. (ed.) The Gude to Genstat. Lawes Agrcultural Trust, Rothamsted Expermental Staton, Oxford, UK, 2000. 10. Lord, D. Modelng Motor Vehcle Crashes usng Posson-gamma Models: Examnng the Effects of the Low Mean Problem and Small Sample Sze on the Estmaton of the Fxed Dsperson Parameter. Accdent Analyss & Preventon, Vol. 38, No. 4, 2006, pp. 751-766.

Lord & Bonneson 16 11. Washngton, S.P., B.N. Persaud, C.Lyon, and J. Oh. Valdaton of Accdent Models for Intersectons. Report No. FHWA-RD-03-037. Federal Hghway Admnstraton, Washngton, D.C., 2005.