Appendixes to NCHRP Report 572: Roundabouts in the United States (2007)

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.

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-1 APPENDIX A LITERATURE REVIEW ON SAFETY PERFORMANCE This appendix presents a detailed literature review of the safety models used in this project. A reference list for this appendix is included at the end. Safety Prediction Models The following contains a comprehensive review of each source, by country of origin followed by a summary indicating how useful the insights from this review were in guiding the current research effort. United Kingdom In the mid 1980’s the Transportation Research Group of the University of Southampton conducted a study of accidents at four-arm roundabouts for the Transport and Road Research Laboratory on behalf of the UK Government Department of Transport (A1). The researchers selected a cross-sectional sample of a pre-defined target type of roundabout (4-arm, single-grade, approximately circular central island, no unusual features, etc.) with specific sub-samples having particular characteristics (e.g., small or large central island, speed limit 30–40 mph or 50–70 mph). An extensive reconnaissance survey of possible sites was undertaken and the samples were selected to give the widest range of vehicle and pedestrian flows and geometry within each sub-group, while being as similar as possible in those characteristics that were not being measured (e.g. environment, congestion). A sample of 84 four-arm roundabouts on main roads in the UK was used. At each site, traffic and pedestrian flow counts were obtained and detailed geometric measurements were made. Personal-injury accidents occurring over a six-year period (1974–1979) were also obtained. Each accident was classified by type and associated (by a convention) to a particular arm of the roundabout. The type of each road user involved was also linked to each of the vehicle or pedestrian movements defining the accident type. The resulting accident type groups were: • Entering-circulating accidents (between an entering vehicle and a circulating vehicle) • Approaching accidents (mostly rear-ends, but also changing lane accidents) • Single-vehicle accidents (a single vehicle colliding with some part of the intersection layout or furniture) • Other accidents (variety of non-pedestrian accidents) • Pedestrian accidents (any accident involving a pedestrian casualty). The main statistical analysis used generalized linear modeling to investigate the relationships between the accident frequency and the traffic and pedestrian flows and geometry at the roundabout sites. The analyses were undertaken in two main stages: • Analysis of total injury accidents at the roundabout as a whole, where each roundabout contributed one data unit to the analysis. • Analyses of arm-specific accidents by type, where each roundabout contributed four data units (i.e. arms) to each accident type analysis.

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-2 The analysis was conducted for three categories of roundabouts: • Small roundabouts. Small roundabouts were defined as those with central islands greater than 4 m (13.1 ft) in diameter with a relatively large ratio of inscribed circle diameter to central island size and often with widened entries and flared approaches. • Normal roundabouts with single carriageway (undivided roadway) arms. As compared to small roundabouts, normal roundabouts have relatively large central islands and un-flared entries. • Normal with one pair of dual carriageway (divided roadway) arms. The basic model in each case was of the form given in Equations A-1a or A-1b as follows: αkQA = , or (A-1a) βα ba QkQA = (A-1b) where A = injury accidents per year; Q or Qa, Qb = functions of the vehicle and pedestrian flow movements, respectively, at the roundabout (all 24-hour annual average flows in thousands); and k, α, β = parameters to be estimated. The analysis method used assumed that the dependent variable had a Poisson error distribution. For the analysis of total injury accidents at the whole roundabout, the study tried three basic flow functions: • Total inflow; • Cross-product flow (product of total entering flows on one pair of opposite arms with the total entering flow on the other pair of opposite arms); and • Entering-circulating flow (sum of the products of entering and circulating flow at each entry). All three flow-functions fitted very well with few distinctions for choosing between them. For comparison with other types of intersections the cross-product flow function was preferred, yielding the models for total injury accidents as shown in Table A-1. The table shows a common value for the flow exponent (α), since there was no significant difference in its value between the roundabout categories. The models indicate that injury accidents are expected to be higher at small roundabouts than normal roundabouts (due presumably to their wider flared entries) and that higher speeds are generally associated with a higher accident frequency. Using the simplifying assumption of averaging the constant parameter for the semi-urban (30–40 mph, or 48–64 km/h) and rural (50–70 mph, or 80–112 km/h) models, the specific models used for this study (with model designations noted) as are follows: A = 0.062 (Major AADT/1000 * Minor AADT/1000) 0.68 (A-1c, UK-INJ1) A = 0.0685 (Major AADT/1000 * Minor AADT/1000) 0.68, single carriageway (A-1d, UK-INJ2) A = 0.059 (Major AADT/1000 * Minor AADT/1000) 0.68, dual carriageway (A-1e, UK-INJ2) A = 0.04 (Entering AADT)1.256 (A-1f, UK-INJ3)

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-3 TABLE A-1: Models of Total Injury Accidents at UK Roundabouts Speed Limit Small Normal – Single Carriageway arms Normal - One pair of Dual-Carriageway arms 30-40 mph A = 0.101 Q 0.68 (25 sites) A = 0.057 Q 0.68 (11 sites) A = 0.057 Q 0.68 (14 sites) 50-70 mph A = 0.081 Q 0.68 (11 sites) A = 0.080 Q 0.68 (11 sites) A = 0.061 Q 0.68 (12 sites) All mph A = 0.095 Q 0.68 (36 sites) A = 0.062 Q 0.68 (48 sites) Note: Q = cross-product flow function = (major AADT)/1000 * (minor AADT)/1000 SOURCE: (A1) One of the objectives of the study was to try to relate the roundabout accidents to the geometry. In the first stage of the modeling described above, differences between types of layout are reflected only in the categories of the roundabout. However, each arm of a roundabout has a different geometry so in the second stage of the analysis, each arm (or more strictly each quadrant) of the roundabout was used as the basic unit of analysis. Full geometric data was available for 78 of the roundabouts, thus providing 312 data units for the analyses. The arm level models included the geometric and other site variables through using a model of the form given in Equation A-2 as follows: ( )∑ ∑+= iiijijba GDQkQA εγβα exp (A-2) where A = accident frequency, in accidents per year; Qa, Qb = functions of the vehicle and pedestrian flow movements; Dij (j=2,n) = dummy variables representing the 2nd to nth level of each discrete factor; Gi = continuous variables (e.g., flow proportions, geometric variables); and k, α, β, γij, εi = model parameters estimated from data. The geometric and other variables and factors were added to the models in a stepwise procedure. At each step the most useful explanatory variable or factor was selected from the whole range of available variables and factors. These choices were made on the grounds of plausibility (i.e., whether the variable was sensible or not), design usefulness (whether the variable was acceptable in a design sense) and statistical validity. At some steps there was a difficult choice between very similar variables so several possible combinations of variable were investigated to inform the most appropriate choice of model. The resulting full models are presented in linear form in Table A-2. Note that these models are found in the ARCADY software. These models were useful in informing the type and estimated magnitude of variables for consideration in this project’s research.

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-4 TABLE A-2: Full Approach Models by Accident Type at UK Roundabouts Accident Type Model term Parameter value Standard Error ENTERING-CIRCULATING ACCIDENTS L(Constant) Lk -3.09 0.47 L(entering flow) LQe 0.65 0.12 L(circulating flow) LQc 0.36 0.11 Entry path curvature Ce -40.3 9.6 Entry width e 0.16 0.025 Approach width correction ev -0.009 0.0038 Ratio factor RF -1.0 0.23 Percentage of motorcycles Pm 0.21 0.063 Angle between arms A -0.008 0.0025 Gradient category g 0.09 0.038 APPROACHING ACCIDENTS L(Constant) Lk -4.71 0.52 L(entering flow) LQe 1.76 0.15 Entry path curvature Ce 20.7 7.6 Reciprocal sight distance 1/Vr -43.9 13.4 Entry width E -0.093 0.038 Gradient category g -0.13 0.06 SINGLE VEHICLE ACCIDENTS L(Constant) Lk -4.71 0.40 L(entering flow) LQe 0.82 0.16 Approach width v 0.21 0.04 Entry path curvature Ce 23.7 6.4 Approach curvature category Ca -0.17 0.05 Reciprocal sight distance 1/Vr -33.0 13.1 OTHER (NON-PEDESTRIAN) ACCIDENTS L(constant) Lk -5.69 0.50 L(entering x circulating flow) LQec 0.73 0.10 Percentage of motorcycles Pm 0.21 0.08 PEDESTRIAN ACCIDENTS L(constant) Lk -3.59 0.27 L((entering + exiting vehicle flow) x Pedestrian flow) LQexp 0.53 0.13 • Entry path curvature - Equal to the inverse of the minimum radius of travel in the region of entry for a vehicle passing straight through the roundabout and taking the shortest possible path while staying on the curbside of the roadway. Signing convention is positive if deflection is to left and negative if the path deflects to the right. (m-1) • Entry width - Perpendicular roadway width at the point of entry (m). • Approach width - Width of the roadway on the approach (m) • Approach width correction - Equal to the product of entry and approach width (m2) • Approach curvature category - Categories of -3, -2, -1, 0, 1, 2, 3. Negative values represent left-hand bends and positive values right-hand bends on the approach. An absolute value of 3 indicates a severe bend, 0 would indicate a straight alignment. • Ratio factor - Ratio of inscribed circle diameter to central island diameter • Percentage of motorcycles - Percentage of the relevant traffic volumes consisting of motorcycles • Angle between arms - The angle in degrees between the approach arm and the next arm clockwise. • Gradient category - The gradient on the approach, categories of -3, -2, -1, 0, 1, 2, 3,. This is a subjective category with -3 indicating a severe downhill grade towards the roundabout and +3 indicating a severe uphill grade towards the roundabout. • Reciprocal sight distance - Reciprocal of the sight distance to the right, from 15m back from the give-way line (m-1) using an object height of 1.05m SOURCE: (A1)

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-5 Australia Two documents authored by Arndt were reviewed. The first (A2) discusses in detail the collection of accident, traffic flow and geometric data for roundabouts constructed in Australia. Models for determining the vehicle paths of drivers through roundabouts and the 85th-percentile speed were developed. The output of these models was used as explanatory variables in a linear regression model developed to predict single vehicle accidents. The second document (A3) follows upon the first by revisiting the single vehicle accident model and developing additional models for several more accident types. Both linear and non- linear Poisson-based regression models were developed, with the non-linear models recommended as being more accurate for predicting accident frequencies. The regression models were based on driver behavior. To predict an accident of a given type, the vehicle paths of the relevant vehicles and the 85th-percentile speeds are first predicted. This information is then fed into a regression model, along with other significant predictors of accidents, to estimate the annual accident frequency. Table A-3 summarizes the accident data used to develop the accident prediction models. All injury severities were included. Not all of the accidents that occurred were used for specific accident type models due to their scarcity. In this case, these crashes were combined into an “other” model. Over eighty percent of accidents involved more than one vehicle and, of these, just over one half involved an entering vehicle colliding with a circulating vehicle. The author noted that detailed information for accidents could not be found for some accidents and subsequently multiplied the model equations by 1.209 to account for the missing data. TABLE A-3 Summary of Accident Data used by Arndt Accident Categories and Frequencies Single Vehicle Accidents 90 (18.3%) Single Vehicle - Crashes involving only one vehicle. 90 (18.3%) Approaching – 13 Entering – 13 Circulating – 51 Departing – 13 Rear End One vehicle collides into the rear of another vehicle. 90 (18.3%) Approaching – 83 Circulating – 6 Departing – 1 Entering/Circulating - An entering vehicle fails to yield and collides with a circulating vehicle. 250 (50.8%) NA Exiting/Circulating A vehicle driving from the inner circulating lane to the departure leg collides with a vehicle circulating on the outer circulating lane. 32 (6.5%) NA Side Swipe - Two vehicles side swipe while travelling on different paths in the same direction. 18 (3.7%) Approaching – 2 Entering – 2 Circulating – 10 Exiting – 2 Departing – 2 Total Accidents 492 (100%) Multiple Vehicle Accidents 402 (81.7%) Low Frequency - Other infrequent crash types. 12 (2.4%) SOURCE: (A3)

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-6 Detailed instructions are provided for constructing vehicle paths through single and multi-lane roundabouts. The procedure is based on observations that vehicles traveling through roundabouts do so in a manner to obtain the largest possible radii, and thus speed, possible. Construction of the vehicle paths allows the distance traveled and the estimated 85th percentile speed on each geometric element to be determined. On the approach and departure legs, assuming a 2-m-wide vehicle, the following distances from the center of the vehicle to the centerline and edge geometric features are assumed: • 1.5 m from a road centerline • 1.5 m from a concrete curb • 1.0 m from a painted edge line or chevron For movements within the roundabout, curves are drawn tangent to the centerline and the approach-departure path line that allow for the largest radii of movement. A model for predicting the 85th-percentile speed as a function of the speed environment and the horizontal curve radius is presented in graphical form in Figure A-1. The predicted speed is plotted on the y-axis versus curve radius on the x-axis for a series of curves for various speed environments. This model has altered previous work on rural roads to allow the predicted 85th- percentile speed to equal the speed environment at large values of curve radius and to more accurately predict speeds at lower radii curves. Note: Radius in meters. SOURCE: (A3) Figure A-1: Australian Method for Speed Estimation

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-7 Accident Model for Single Vehicles. This model applies only to vehicles making right turns (equivalent to left turns in the US); it does not apply to vehicles making left turns (equivalent to right turns in the US) or U-turns through the roundabout. The model, divided into two components, is given in Equations A-3a and A-3b as follows: ( ) 91.1 12.417.1121064.1 R SSLQAsp Δ+× = − , and (A-3a) ( ) 65.0 93.191.091079.1 R SSLQAsa Δ+× = − ; (A-3b) where Asp = number of single vehicle accidents per year per leg for vehicle path segments prior to the giveway line; Asa = number of single vehicle accidents per year per leg for vehicle path segments after the giveway line; Q = AADT in direction considered; L = length of vehicle path on the horizontal geometric element (m); S = 85th-percentile speed on the horizontal geometric element (km/h); ΔS = decrease in 85th-percentile speed at the start of the horizontal geometric element (km/h); and R = vehicle path radius on the horizontal geometric element (m). Accident Model for Approaching Rear-End Vehicles. Only rear-end accidents occurring on the roundabout approach curves are considered. The model is given in Equation A- 4 as follows: ( ) 31.277.465.039.1181081.1 aaciar NSQQA ∑−×= (A-4) where Ar = number of approaching rear-end vehicle accidents per year per approach leg; Qa = AADT on the approach; Qci = circulating vehicle AADTs from the other approaches; Sa = 85th-percentile speed on the approach curve (km/h); and Na = number of lanes on the approach. Accident Model for Entering-Circulating Vehicles. This model includes only the entering vehicles from the approach of interest, the right-turn movement (equivalent to the left- turn movement in the US) from the opposite approach, and the through and right-turn movements (equivalent to the through and left-turn movements in the US) from the previous approach in the direction of traffic. The entering vehicles are assumed to be in the inside lane for multilane approaches. Left turns (equivalent to right turns in the US) and U-turn vehicles from the approach of interest are ignored. The model is given in Equation A-5 as follows:

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-8 ( ) 21.0 38.141.09.047.071031.7 Ga racica e t SQNQ A ∑ −× = (A-5) where Ae = number of entering-circulating vehicle accidents per year per approach leg; Qa = AADT on the approach; Nc = number of circulating lanes; ∑ ciQ = sum of the circulating vehicle AADTs from the other approaches; Sra = ∑ ∑ ci rici Q SQ ; TGa = ∑ ∑ ci Gici Q tQ ; Sri = the various relative 85th-percentile speeds between vehicles on the approach curve and circulating vehicles from each direction (km/h); tGi = the various travel times taken from the give way line of the approach to the intersection point between the entering and circulating vehicles; tGi = ciGi Sd6.3 ; dGi = distance from the give way line of the approach to the intersecting point between entering and circulating vehicles (m); and Sci = the various 85th-percentile speeds of the circulating vehicles adjacent to the approach (km/h). Accident Model for Exiting/Circulating Vehicles. The accidents modeled included only those occurring on multi-lane roundabouts that contained one of the following marking systems: (1) No lane lines between circulating lanes; (2) Broken lane lines between circulating lanes marked adjacent to each splitter island; (3) Broken lane lines between circulating lanes marked fully around the roundabout; or (4) The vehicle volumes modeled include the exiting vehicles from the inside lane and the circulating vehicles from the outside lane of the approach under consideration and the previous approach in the direction of traffic. The model is given in Equation A-6 as follows: ( ) ( ) 13.468.032.0111033.1 raeicid SQQA ∑∑−×= (A-6) where ∑ ciQ = sum of the circulating vehicle AADTs from the other approaches; ∑ eiQ = sum of the various AADT flows exiting the roundabout at the exit point of the departure leg; Sra = ∑ ∑ ei riei Q SQ ; and

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-9 Sri = the various relative 85th percentile speeds between vehicles exiting the roundabout and circulating vehicles at the particular departure leg. Accident Model for Sideswipe Vehicles. This model was developed for segments consisting of two lanes. Left-turn (right-turns where vehicles travel on the right side of the road) and u-turn vehicles from the approach of interest are ignored within the roundabout. The model is separately applied to segments prior to the approach curve, the approach curve, the circulating-through segment, the circulating-right turn segment, the departing-through segment and the departing-right turn segment. The model is given in Equation A-7 as follows: ( ) 59.072.081049.6 ltss fQQA Δ×= − (A-7) where Ass = number of sideswipe vehicle accidents per leg per vehicle path segment; Q = AADT for the particular movement on the particular geometric element; Qt = total AADT on the particular geometric element; and Δfl = difference in potential side friction (km/h2/m). Accident Model for Other Vehicles. The other vehicle accident model is simply an accident rate calculated by dividing the total number of accidents not falling into one of the five model categories by the total number of vehicles approaching all of the roundabouts and multiplied by 365. The model is given in Equation A-8 as follows: ∑ −×= aO QA 61029.4 (A-8) where AO = number of “other” accidents per year; and Qa = AADT on approach a. France A French model (A4) for predicting the total number of injury accidents at a roundabout does not contain any geometric variables and applies to roundabouts where the total incoming traffic ranges between 3,200 and 40,000 vehicles per day. The model is given in Equation A-9 as follows (designated in this project as FR-INJ1): Injury accidents/year = ( ) cTE FQ51015.0 − (A-9, FR-INJ1) where QTE = total daily incoming traffic; and Fc = adjustment coefficient for the period under consideration. Sweden A study in Sweden (A5) surveyed roughly 650 roundabouts in 1997, classifying them according to geometric design, speed level and other variables. Accident and vehicle, bicycle and pedestrian volumes were collected for certain subsets of the total database for 1994 to 1997. An analysis of cyclist and pedestrian accident data was undertaken for 72 roundabouts and an

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-10 analysis of vehicle accident data for 182 roundabouts. A speed analysis was conducted for 536 roundabouts. With regards to vehicle speeds the following conclusions were drawn: • Speeds are higher when the general speed limit is higher than the local limit; • Speed is, on average, higher at multilane roundabouts than single lane roundabouts; • Speed is lower if the radius of the central island is 10 to 20 m (33 to 66 ft) than if the radius is smaller or larger; • Provision of additional travel surface around the central island has no effect on speed; • Developing the approach to be as perpendicular as possible at the roundabout entry reduces speed into and through the roundabout; With regards to accidents involving cyclists and pedestrians the following conclusions were drawn: • Single lane roundabouts are much safer for cyclists than multilane roundabouts • Fewer cyclist accidents occur when the central island is greater than 10 m (33 ft) and when bicycle crossings are provided • It is safer for cyclists to bypass a roundabout on a bicycle crossing than to travel on the carriageway • For pedestrians, roundabouts are no less safe than conventional intersections and single lane roundabouts are more safe than multi-lane roundabouts For vehicle accidents, 456 accidents occurred at the 182 roundabouts from 1994–1997. Nineteen percent of these resulted in an injury and there were no fatalities. General observations with regards to vehicle accidents include: • Accident frequency is directly proportional to vehicle speeds. • Injury accident frequency has a more quadratic relationship with speed. • A central island radius of 10 to 25 m (33 to 82 ft) results in the lowest accident frequency. Two variations of a prediction model were developed to predict vehicle accident rates (accidents per million entering vehicles) at roundabouts (A5). The models predict vehicle accident rates (accidents per million entering vehicles) and are based on data from both urban and rural sites. The first of these was designated in this study as SWED-TOT1; both are given in Equations A-10a and A-10b as follows: Predicted accident rate = 0.1353 × 0.863leg × 1.88speed70 × 1.22lanes (A-10a, SWED-TOT1) where 3leg = 1 if 3-legged, 0 if 4-legged; speed70 = 1 if speed limit is 70 km/h (44 mph), 0 if 50 km/h (31 mph); and 2lanes = 1 if there are 2 lanes on roundabout, 0 if there is one. Predicted accident rate = 0.1130 × 0.923leg × 1.84speed70 × 1.40loclow × 1.172lanes (A-10b) where 3leg = 1 if 3-legged, 0 if 4-legged; speed70 = 1 if speed limit is 70 km/h (44 mph), 0 if 50 km/h (31 mph); loclow = 1 if general speed limit within 600 m (1970 ft) of roundabout is higher than the local limit; and

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-11 2lanes = 1 if there are 2 lanes on roundabout, 0 if there is one. Due to a low number of injury accidents, injury accident models were developed by fitting observed injury accident rates to a function of the prediction for the total accident rate. Two variations were developed (with the first designated in this study as SWED-INJ1) and are given in Equations A-11a and A-11b as follows: Predicted injury rate = 0.8178 (Predicted accident rate)1.6871 (A-11a, SWED-INJ1) Predicted injury rate = 0.7215 (Predicted accident rate)1.6119 (A-11b) The authors compared predictions from the roundabout models to previously calibrated models for conventional signalized and unsignalized intersections and concluded that roundabouts are generally slightly safer than conventional intersections, particularly when the local speed limit is assumed to be 50 km/h (31 mph). However, evidence is not provided in the paper to support the notion that the two sets of models indeed make valid comparisons. For example, the data used to develop the conventional intersection models may have come from locations with a much different distribution of traffic volumes, different location types and/or other factors that may impact safety but are not necessarily included in any of the models. Thus, these conclusions on the relative safety of roundabouts compared to other intersection types may not be reliable. An earlier study also developed injury models for several classes of roundabouts (A6). The aggregate models for predicting the number of motor vehicle crashes with personal injury per year for urban junctions are as follows (designated for this project as SWED-INJ2 and SWED-INJ3): 50 km/h (31 mph): Injury Crashes /year = 0.00000308 (Total entering AADT)1.20 for 4 legs (A-12a, SWED-INJ2) Injury Crashes /year = 0.00000232 (Total entering AADT)1.20 for 3 legs (A-12b, SWED-INJ2) 70 km/h (44 mph): Injury Crashes /year = 0.00000440 (Total entering AADT)1.20 for 4 legs (A-12c, SWED-INJ3) Injury Crashes /year = 0.00000332 (Total entering AADT)1.20 for 3 legs (A-12d, SWED-INJ3) Review of Before-After Safety Studies Studies on the safety effect of converting conventional intersections to roundabouts were also reviewed. The results are usually reported without reservations. This does not mean that there are not any. For example, the reported safety benefits of roundabout installation may be exaggerated due to regression-to-the-mean in cases where this bias is not accounted for. In most cases, it is difficult to make a determination if this bias exists or, if so, if it was accounted for. Thus, the reader is cautioned to accept the results summarized here in the spirit in which this section is provided – to provide a flavor for the safety benefits of roundabouts. The decision to report these results in spite of possible reservations was based on a belief that, with the very large reductions that were consistently observed, the benefits of roundabouts would remain substantial if regression-to-the-mean effects were removed and any other methodological limitations were to be overcome. United States A before-after study of roundabout conversions in the United States used the empirical Bayes methodology to control for regression-to-the-mean and other trends in accident occurrence

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-12 (A7). The analyses used data from seven states—Colorado, Florida, Kansas, Maine, Maryland, South Carolina, and Vermont—where a total of 23 intersections were converted to modern roundabouts between 1992 and 1997. Of the 23 intersections studied, 19 were previously controlled by stop signs, and four were controlled by traffic signals. Fourteen of the roundabouts were single-lane circulation designs, and nine, all in Colorado, were multilane. For all of the study intersections the empirical Bayes procedure estimated a highly significant 40-percent reduction for all crash severities combined. Because injury data were not available for the period before construction of the four roundabouts in Vail, overall estimates for changes in injury crashes are based on the other 19 intersections. The empirical Bayes procedure estimated a highly significant 80-percent reduction for injury crashes for these 19 converted intersections. Because of major operational differences between various roundabout designs and settings, results were analyzed and reported for several groups of conversions for which there were sufficient crash data to provide meaningful results. These include eight urban single-lane roundabouts that prior to construction were stop-controlled, five rural single-lane roundabouts that prior to construction were stop-controlled, six urban multilane roundabouts that prior to construction were stop-controlled, and four urban intersections converted to roundabouts from traffic signal control. The results are summarized in Table A-4 which is taken directly from the aforementioned study (A7). In summary: • For the group of eight urban single-lane roundabouts converted from stop control, the empirical Bayes procedure estimated highly significant reductions of 72 percent for all crash severities combined and 88 percent for injury crashes. • For the group five rural single-lane roundabouts converted from stop control, similar effects were estimated: a 58-percent reduction for all crash severities combined and an 82-percent reduction for injury crashes. • For the group of six urban multilane roundabouts, however, the estimated effect on all crash severities combined was much smaller: a 5-percent reduction. Because injury data were not available for the period before construction of four of these roundabouts, overall estimates for changes in injury crashes were not computed for this group of intersections. • For the four roundabouts converted from traffic signal control, estimated reductions were 35 percent for all crash severities combined and 74 percent for injury crashes. Three of these roundabouts had multilane circulation designs. A recent project undertaken by some members of the NCHRP 3-65 project team for the New York State Department of Transportation (NYSDOT) (A8) increased the sample size of the Persaud et al. study (A7) with more sites and additional years of data. The preliminary results were generally supportive of those in Persaud et al. (A7) as well as those of an earlier and limited effort by Flannery and Elefteriadou (A9) who used a number of the same sites in their study. Finer levels of disaggregation than was possible for the Persaud et al. study were, however, not feasible despite the larger sample size.

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-13 TABLE A-4: Estimates of Safety Effects for Groups of Conversions Group Characteristic Before Count of Crashes During Period After Conversion Crashes Expected During After Period Without Conversion (Standard Deviation) Index of Effectiveness (Standard Deviation) Percent Reduction in Crashes Conversion/Jurisdiction All Injury All Injury All Injury All Injury Single Lane, Urban, Stop Controlled Bradenton Beach, FL 1 0 9.9 (3.6) 0 (0) Fort Walton Beach, FL 4 0 16.9 (3.9) 2.7 (1.1) Gorham, ME 4 0 6.8 (1.4) 0.9 (0.4) Hilton Head, SC 9 0 42.8 (6.0) 8.2 (1.9) Manchester, VT 1 1 1.7 (0.7) 0 (0) Manhattan, KS 0 0 4.2 (1.2) 1.2 (0.5) Montpelier, VT 1 1 4.3 (1.8) 1.1 (0.6) West Boca Raton, FL 7 0 8.1 (3.0) 2.6 (1.3) Entire group (8) 27 2 94.6 (9.0) 16.6 (2.6) 0.28 (0.06) 0.12 (0.08) 72% 88% Single Lane, Rural, Stop Controlled Anne Arundel County, MD 14 2 24.6 (4.0) 6.2 (1.7) Carroll County, MD 4 1 15.2 (2.6) 3.2 (0.9) Cecil County, MD 10 1 14.3 (2.9) 5.6 (1.4) Howard County, MD 14 1 36.7 (5.5) 7.7 (2.1) Washington County, MD 2 0 14.4 (3.1) 4.2 (1.3) Entire group (5) 44 5 105.2 (8.4) 26.9 (3.4) 0.42 (0.07) 0.18 (0.09) 58% 82% Multilane, Urban, Stop Controlled Avon, CO 3 0 19.9 (4.9) 0 (0) Avon, CO 17 1 12.2 (3.1) 0 (0) Vail, CO 14 — 19.1 (4.4) — Vail, CO 61 — 50.9 (7.6) — Vail, CO 8 — 9.8 (2.1) — Vail, CO 15 — 11.8 (2.3) — Entire group (6) 118 123.7 (11.0) n/a 0.95 (0.10) n/a 5% n/a Urban, Signalized Avon, CO 44 1 49.8 (7.0) 5.4 (1.7) Avon, CO 13 0 30.1 (5.7) 2.3 (1.0) Avon, CO 18 0 52.1 (7.0) 5.3 (1.7) Gainesville, FL 11 3 4.8 (1.5) 1.3 (0.5) Entire group (4) 86 4 131.7 (10.9) 15.0 (2.7) 0.65 (0.09) 0.26 (0.14) 35% 74% All conversions (23) 275 12 454.6 (19.8) 58.5 (5.1) 0.60 (0.04) 0.20 (0.06) 40% 80% SOURCE: (A7) Non-US Studies International experience with roundabout conversion was summarized in a report for the Insurance Institute for Highway Safety (A10). A summary of this information is provided here. A before and after evaluation of 230 sites where roundabouts were built (and 60 control sites) in New South Wales, Australia, found a 41-percent reduction in total crashes, 45-percent reduction in injury crashes, and 63-percent reduction in fatal crashes (A11). A Danish study found that the reconstruction of urban give-way (yield) intersections into roundabouts leads to a considerable reduction in injury crashes for occupants of cars, in fact, about 85 percent. For bicyclists there was no safety benefit from the reconstruction from give- way intersections to roundabouts. The number of cyclist injury crashes was unchanged (and now make up roughly 70-percent of all injury crashes). The effect on the number of injury crashes at reconstructed rural intersections is approximately the same as that for urban ones (A12). The

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-14 study also found that injury crashes became less serious, decreasing from an average of 1.3 to 1.0 injured persons per injury crash for urban junctions and from 2.1 to 1.25 for rural junctions. The injury crashes at the rural roundabouts were mainly bicyclist crashes. This study also looked at the severity of crashes and concluded that small roundabouts do well with respect to this parameter. The percent of serious and fatal crashes was substantially lower at small roundabouts (3.3 percent) than at conventional ones (8.4 percent). The average percent of serious and fatal crashes at roundabouts on dual carriageways was 4.2 percent. At signalized dual carriageway intersections this percentage was 9.2 percent for two-phase signals and 7.8 percent for three- phase signals (separate right-turn phase—left hand driving). A study of 83 roundabouts in France concluded that conversion to roundabouts reduces injury crashes by 78 percent and fatality crashes by 82 percent (A13). A German study by Brilon (A14) compared the crash experience of 32 newly constructed single-lane roundabouts to their safety prior to reconstruction. Two of the intersections had previously been signalized; the others had had two-way stop control or yield. The total effect was a 40-percent reduction in crash frequency. The effect was particularly impressive for rural intersections, where the number of serious injuries was reduced from 18 to 2, the number of light injuries went from 25 to 3, and the number of crashes causing “heavy” property damage went from 24 to 3. There was a small reduction in the number of pedestrian crashes as a result of the construction of roundabouts, whereas the number of bicycle crashes went up somewhat. This increase took place at roundabouts that had marked bicycle lanes on the outer edge of the circulating roadway (where crash numbers went from one to eight). The number of bicycle crashes was more or less unchanged at locations with mixed traffic as well as at locations with separate bicycle/pedestrian paths. The Dutch research institute SWOV did a study on the safety of 201 single-lane roundabouts built in 1990 (A15). These all have yield at entry and access roads with radial orientation. This study found that “the substitution of an intersection by a roundabout has a particularly favorable effect on road safety: a reduction of 47 percent in the number of crashes and 71 percent in the number of road crash victims (after trend correction). However, the various categories of road user did not all profit from the change to the same degree: a large reduction in road crash victims was noted amongst occupants of cars and pedestrians (95 percent and 89 percent, respectively) and a slight reduction amongst cyclists (‘only’ 30 percent).” According to Ourston (A16) another study from the Netherlands investigated the effect of conversion of nine traffic signals to roundabouts. They found a 27-percent reduction in crashes and a 33-percent reduction in casualties (A16). A recent paper by Elvik performed a meta-analysis study on 28 non-U.S. studies evaluating the safety effects of converting conventional intersections to roundabouts (A17). Table A-5 lists the year and origin of publication and the study design used for the studies reviewed. Some of these studies were reviewed in the IIHS report summarized above. The analysis found that roundabouts reduce the number and severity of injury crashes. For all severities of injury the best estimate is a 30- to 50-percent reduction. Fatal crashes are estimated to reduce from 50 to 70 percent. For property-damage-only crashes the effect is uncertain but may in fact increase after roundabout conversion, particularly at 3-legged intersections. There is some evidence that roundabout conversion has a greater effect on injury crashes at 4-legged intersections than 3-legged. A greater effect is apparent at yield-controlled than at signalized intersections.

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-15 TABLE A-5: Year and origin of Studies Reviewed by Elvik Year Country Study design 1975 Great Britain Simple before-and-after 1977 Great Britain Before-and-after, controlling for general trends 1981 Denmark Comparative study and before-and-after, controlling for general trends 1983 Sweden Comparative study of crash rates in various types of intersections 1983 Norway Before-and-after, data on traffic volume 1985 Sweden Before-and-after, controlling for trends and regression-to-the-mean 1985 Norway Comparative study of crash rates in various types of intersections 1988 Great Britain Comparative study of crash rates in various types of intersections 1988 Norway Before-and-after, controlling for general trends 1990 Norway Comparative study of crash rates in various types of intersections 1990 Australia Before-and-after, controlling for general trends 1990 Netherlands Simple before-and-after 1991 Denmark Comparative study and simple before-and-after 1992 Sweden Comparative study of crash rates in various types of intersections 1992 Switzerland Simple before-and-after 1992 Germany Simple before-and-after 1992 Sweden Simple before-and-after 1992 Denmark Before-and-after, controlling for general trends 1992 Norway Before-and-after, controlling for general trends 1992 Germany Before-and-after, data on traffic volume 1993 Germany Before-and-after, data on traffic volume 1993 Netherlands Simple before-and-after 1994 Germany Comparative study and before-and-after, data on traffic volume 1994 Denmark Before-and-after, controlling for trends and regression-to-the-mean 1994 Norway Before-and-after, controlling for general trends 1994 Switzerland Before-and-after, controlling for general trends 1995 Norway Before-and-after, controlling for general trends 1997 Norway Before-and-after, controlling for trends and regression-to-the-mean SOURCE: (A17) As for the safety effect of specific design features, few studies provided adequate information to reach conclusions. The 1983 Swedish study looked at AADT, proportion of traffic entering from the minor road, number of legs, speed limit and diameter of the central traffic island but found no significant relationship between the diameter of the central traffic island and crash rate. It is unclear if this lack of effect is “true” or due to inadequate sample sizes or data ranges. Most of the roundabouts in the study had a central traffic island with a diameter of more than 40 m (131 ft). The 1992 Swedish study and the 1991 and 1992 Danish studies found an increase in the crash rate with a larger central traffic island. The effects of other design parameters were found to be small. The four factors with the strongest effect on crash rates in roundabouts are cited as total traffic volume, proportion of vehicles entering from the minor road, speed limit, and number of legs. The direction of this relationship is not specified in Elvik's paper.

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-16 Summary of the Review on the Safety Effect of Converting Conventional Intersections to Roundabouts Prior to NCHRP 3-65, the one definitive study of US conversions, and its subsequent update for the NYSDOT, was based on a rather small sample size. As such, only limited disaggregate analysis could be done to try to isolate the geometric factors associated with the greatest safety benefits of roundabout construction. While some of these factors have been isolated in evaluations outside of the US it is not clear that that knowledge is directly transferable. In addition, several of those studies had methodological limitations. The review of the previous studies did provide useful insights for guiding the disaggregated before-after analysis done for this study. Useful lessons were learned from the pitfalls and limitations of many of those studies (e.g., small sample sizes, ignoring regression to the mean, and improperly accounting for traffic volume and secular changes over time). These lessons emphasized the need for, and use to be made of, recent advances in safety estimation methodology aimed at overcoming these limitations. References A1. Maycock, G., and R. D. Hall. Accidents at 4-arm roundabouts. Report LR 1120. Transport and Road Research Laboratory, Crowthorne, Berkshire, United Kingdom, 1984. A2. Arndt, O. K. Relationship Between Roundabout Geometry and Accident Rates. M. E. thesis. Queensland University of Technology, Brisbane, Queensland, Australia, June 1994. A3. Arndt, O. K. Relationship Between Roundabout Geometry and Accident Rates - Final Report. Infrastructure Design, Transport Technology Division, Department of Main Roads, Brisbane, Queensland, Australia, 1998. A4. Service d’Etudes Techniques des Routes et Autoroutes (SETRA). Accidents at intersections: the use of models to predict average accidents rates. Memorandum. Bagneux Cedex, France, 1998. A5. Brüde, U., and J. Larsson. What Roundabout Design Provides the Highest Possible Safety? In Nordic Road and Transport Research, No. 2. Swedish National Road and Transport Research Institute, 2000. A6. Brüde, U., K. Hedman, J. Larsson, and L. Thuresson. Design of Major Urban Junctions—Comprehensive Report. VTI EC Research 2, 1998. A7. Persaud, B. N., R. A. Retting, P. E. Garder, and D. Lord. Observational Before-After Study of the Safety Effect of U.S. Roundabout Conversions Using the Empirical Bayes Method. In Transportation Research Record: Journal of the Transportation Research Board, No. 1751. TRB, National Research Council, Washington, DC, 2001, pp. 1–8. A8. Eisenman, S., J. Josselyn, G. List, B. Persaud, C. Lyon, B. Robinson, M. Blogg, E. Waltman, and R. Troutbeck. Operational and Safety Performance of Modern

NCHRP Web-Only Document 94: Appendixes to NCHRP Report 572: Roundabouts in the United States A-17 Roundabouts and Other Intersection Types. Final Report, SPR Project C-01-47. New York State Department of Transportation, Albany, NY, April 7, 2004. A9. Flannery, A., and L. Elefteriadou. “A Review of Roundabout Safety Performance in the United States.” In Proceedings of the 69th Annual Meeting of the Institute of Transportation Engineers. Institute of Transportation Engineers, Washington, DC, 1999. A10. Persaud, B., and P. Garder. Roundabout Safety in the U.S.—Volume 1: Background Paper on Roundabout Safety. Draft Final Report. Insurance Institute for Highway Safety, 1999. A11. Tudge, R. T. Crashes at Roundabouts in New South Wales. Proc., Part 5 of the 15th AARB Conference. August 1990, pp. 341–349. A12. Jørgensen, E., and N. O. Jørgensen. Trafiksikkerhed i 82 danske rundkørsler—anlagt efter 1985. Rapport 4. IVTB, DTU, 1994. A13. Centre D’Etudes Techniques de l’Equipment de l’Ouest. Evolution de la Securite Sur Les Carrefours Giratoires, Nantes, France, 1986. As referenced in Jacquemart, G. NCHRP Synthesis of Highway Practice 264: Modern Roundabout Practice in the United States, TRB, National Research Council, Washington, DC, 1998. A14. Brilon, W. Sicherheit von Kreisverkehrsplätzen. Ruhr-University Bochum, Bochum, Germany, 1996, p. 3. A15. Schoon, C., and J. van Minnen. Ongevallen op rotondes II. Report No. R-93-16. SWOV, Leidschendam, Netherlands, 1993. A16. Ourston, L. Comparative Safety of Modern Roundabouts and Signalized Intersections. March 2, 1996. A17. Elvik, R. Effects on road safety of converting intersections to roundabouts: Review of evidence from non-U.S. studies. Transportation Research Record: Journal of the Transportation Research Board, No. 1847. Transportation Research Board of the National Academies, Washington, DC, 2003, pp. 1–10.