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.

65 This chapter presents the design findings for this project. The following sections discuss an analysis of predicted versus observed speeds, pedestrian behavior, and bicyclist behavior. The chapter concludes with additional design findings drawn from the safety and operational analyses presented in previous chapters. Speed Analysis The speed of vehicles through a roundabout is widely considered to be one of the most important parameters in designing a roundabout. Therefore, the ability to predict the speeds that vehicles will take when traveling through a proposed design is fundamental. To address this need, the research team conducted a detailed analysis of predicted speeds versus actual field-measured speeds to determine how well existing techniques predict reality. Upon predicting values of entry speed (V1p), through- movement circulating speed (V2p), through-movement exit speed (V3p), and left-turn-movement circulating speed (V4p) and recording actual speeds, Va, at the same locations, the research team compared the predicted and actual values for V1 through V4. The datasets for the single-lane and multilane sites were evaluated separately in order to compare similar geometry. Definitions for each speed variable are provided in Appendix G. The following sections highlight the findings for each speed value evaluated as part of this report. They are organ- ized in the following order to reflect an increasing degree of uncertainty with each parameter: • Reproduction of FHWA speed prediction • Left-turn-movement circulating speed • Through-movement circulating speed • Exit speed • Entry speed Reproduction of FHWA Speed Prediction The speed prediction formula presented in FHWA’s Roundabouts: An Informational Guide (1) is based on the basic highway design principles found in the AASHTO’s A Policy on Geometric Design of Streets and Highways (33). The basic relationship among speed, vehicle path radius, superelevation, and side friction factor is as follows: where V  speed (mph) R  vehicle path radius (ft) e  superelevation (ft/ft) f  side friction factor where V  speed (km/h) R  vehicle path radius (m) e  superelevation (m/m) f  side friction factor The FHWA Roundabout Guide presents its speed method- ology using a series of graphs to demonstrate the relationship among these parameters, recognizing that side friction factor varies with speed. This process can be simplified by fitting an equation to the relationship between speed and path radius for the two most common superelevation values, e  0.02 and e  0.02. These fitted equations (with a coefficient of determination exceeding 0.997) are as follows: V R e= = −3 4614 0 020 3673. , . ( ). for 5-2b V R e= = +3 4415 0 020 3861. , . ( ). for 5-2a V R e f= +127 ( ) ( )5-1b, Metric V R e f= +15 ( ) ( )5-1a, U.S. Customary C H A P T E R 5 Design Findings

where V  predicted speed (mph) R  radius of vehicle path (ft) where V  predicted speed (km/h) R  radius of vehicle path (m) The original FHWA graphs and the associated fitted equa- tions are shown in Figures 38 and 39 for U.S. customary units and metric units, respectively. Left-Turn-Movement Circulating Speed Table 50 summarizes the characteristics of the V4 data used for this analysis. As the table shows, 1,007 of the 1,231 obser- vations, or 82%, occur at sites with at least 15 observations; only these observations were used in determining percentiles. Three percentiles—average (mean), 85th-percentile, and 95th-percentile—were examined to determine the best fit to V R e= = −8 6164 0 020 3673. , . ( ). for 5-3b V R e= = +8 7602 0 020 3861. , . ( ). for 5-3a predicted values. As shown in Table 51, the 85th-percentile val- ues for each site with 15 or more observations result in the lowest mean deviation and lowest RMSE of the three per- centiles considered. Figures 40, 41, and 42 present plots of predicted versus actual speeds for all sites, the subset of single-lane sites, and the subset of multilane sites, respectively, along with 85th-per- centile values for all sites with 15 or more observations. As can be seen from the figures, the current V4 method predicts 85th- percentile speeds remarkably well. Based on these findings, the remaining analysis presented in this report assumes the following: • The factors influencing the relationship between path radius and speed—side friction factor and supereleva- tion—are reasonable, and no adjustments to side friction factors appear to be necessary. • The V4 method is a reasonable predictor of 85th-percentile speed; thus, all subsequent analyses will use 85th-percentile speeds. Through-Movement Circulating Speed Table 52 summarizes the characteristics of the V2 data. As the table shows, 756 of the 990 observations, or 76%, occur at sites with 15 or more observations. These sites with 15 or more observations were used in determining percentiles. Figures 43, 44, and 45 present plots of predicted versus actual speeds for all sites, the subset of single-lane sites, and the subset of multilane sites, respectively, along with 85th- percentile values for all sites with 15 or more observations. Table 53 summarizes the goodness-of-fit analysis. As can be seen from Figure 43, the current method for predicting V2 speeds generally overestimates 85th-percentile speeds by an average of 2 to 3 mph (3 to 5 km/h). Review of the speed pre- dictions for individual sites suggests that the current method for drawing through-movement paths is somewhat conser- vative, with drivers not cutting as straight a path as the method suggests. In addition, circulating speeds may be influ- 66 y = 3.4415x0.3861 R2 = 0.9989 y = 3.4614x0.3673 R2 = 0.9977 0 5 10 15 20 25 30 35 40 0 50 100 150 200 250 300 350 450400 Radius (ft) Sp ee d (m ph ) e = +0.02 e = –0.02 Fitted Equation (e = +0.02) Fitted Equation (e = –0.02) y = 8.6164x0.3673 R2 = 0.9977 y = 8.7602x0.3861 R2 = 0.9989 0 10 20 30 40 50 60 0 20 40 60 80 100 120 140 Radius (m) Sp ee d (km /h) e = +0.02 e = –0.02 Fitted Equation (e = –0.02) Fitted Equation (e = +0.02) Figure 38. Fitted equation for FHWA speed-radius curves (U.S. customary). Figure 39. Fitted equation for FHWA speed-radius curves (metric). Characteristic Range of measured speeds (11 – 45 km/h) Total number of sites Total number of observations Number of sites with 15+ observations 43 Number of observations at sites with 15+ observations 1,007 Total 7 – 28 mph 69 1,231 Table 50. Summary of left-turn-movement circulating speed (V4) data.

67 Mean Deviation Root Mean Square Error Average (Mean) Speed 2.1 mph (3.4 km/h) 2.6 mph (4.2 km/h) 85th-Percentile Speed 0.2 mph (0.3 km/h) 1.7 mph (2.7 km/h) 95th-Percentile Speed –1.0 mph (–1.6 km/h) 2.4 mph (3.9 km/h) Table 51. Left-turn-movement circulating speed prediction error. Predicted Speed, V4p (mph), All Sites y = 1.1041x - 1.8409 R2 = 0.6483 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 A ct ua l S pe ed , V 4a (m ph ), A ll S ite s Data Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) y = 0.7653x + 2.8849 R2 = 0.30840 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 Predicted Speed, V4p (mph), Single-Lane Sites A ct ua l S pe ed , V 4a (m ph ), S ing le- La ne S ite s Data Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 40. Actual vs. predicted values for left-turn-movement circulating speeds, all sites. Figure 41. Actual vs. predicted values for left-turn-movement circulating speeds, single-lane sites.

enced by hesitation on entry, which, over time, could be rea- sonably expected to reduce as drivers become more comfort- able. Therefore, the current method for estimating V2 is generally conservative at the present time but reasonable, and no changes are proposed. Exit Speed Table 54 summarizes the characteristics of exit speed data, denoted by V3 for through movements and V6 for left-turn movements (these two were combined for analysis). As the exhibit shows, 1,480 of the 1,767 observations, or 84%, occur 68 y = 1.1291x - 1.2795 R 2 = 0.8867 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 Predicted Speed, V4p (mph), Multilane Sites A ct ua l S pe ed , V 4a (m ph ), M ult ila ne S ite s Data Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Characteristic Range of measured speeds (11 – 50 km/h) Total number of sites Total number of observations Number of sites with 15+ observations 28 Number of observations at sites with 15+ observations 756 7 – 31 mph 58 990 Total y = 0.8924x - 0.5511 R 2 = 0.7664 0 5 10 15 20 25 30 35 40 0 5 10 20 2515 30 35 40 Predicted Speed, V2p (mph), All Sites A ct ua l S pe ed , V 2a (m ph ), A ll S ite s V2 Data Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 42. Actual vs. predicted values for left-turn-movement circulating speeds, multilane sites. Table 52. Summary of through-movement circulating speed (V2) data. Figure 43. Actual vs. predicted values for through-movement circulating speeds, all sites.

at sites with 15 or more observations. Only 85th-percentile speeds from sites with 15 or more observations were used in the analysis. Figures 46, 47, and 48 present plots of predicted versus actual exit speeds differentiated by through movements and left-turn movements for all sites, the subset of single-lane sites, and the subset of multilane sites, respectively, along with 85th-percentile values for all sites with 15 or more observations. As can be seen from the figures, the current method for predicting V3 speeds generally overestimates 85th-percentile speeds, with the error increasing substantially with higher predicted speeds. Note that the cluster of sites with predicted speeds of around 45 mph is arbitrary, as tan- gential exits with a path radius of infinity were arbitrarily 69 y = 0.6836x + 3.1551 R2 = 0.2679 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 Predicted Speed, V2p (mph), Single-Lane Sites A ct ua l S pe ed , V 2a (m ph ), S ing le- La ne S ite s V2 Data Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) y = 0.9293x - 1.3597 R 2 = 0.6838 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 Predicted Speed, V2p (mph), Multilane Sites A ct ua l S pe ed , V 2a (m ph ), M ult ila ne S ite s V2 Data Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 44. Actual vs. predicted values for through-movement circulating speeds, single-lane sites. Figure 45. Actual vs. predicted values for through-movement circulating speeds, multilane sites. Mean Deviation 85th Percentile Speed Root Mean Square Error 3.2 mph (5.1 km/h)2.6 mph (4.2 km/h) Table 53. Through-movement circulating speed prediction error.

70 Characteristic Through-Movement Exit Speed, V 3 Left-Turn- Movement Exit Speed, V 6 Total Range of measured speeds 8 – 37 mph (13 – 60 km/h) 8 – 31 mph (13 – 50 km/h) 8 – 37 mph (13 – 60 km/h) Total number of sites 56 52 108 Total number of observations 1,084 683 1,767 Number of sites with 15+ observations 38 22 60 Number of observations at sites with 15+ observations 960 520 1,480 Table 54. Summary of exit speed data. y = 0.2513x + 13.834 R 2 = 0.2933 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Unadjusted Predicted Exit Speed, V3pbase or V6pbase (mph), All Sites A ct ua l E xi t S pe ed , V 3a o r V 6a (m ph ), A ll Si te s V3 Data Match Line 85th %ile (15+ obs.) V6 Data Linear (85th %ile (15+ obs.)) Figure 46. Actual vs. unadjusted predicted values for through- movement and left-turn-movement exit speeds, all sites. y = 0.1259x + 16.554 R2 = 0.135 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Unadjusted Predicted Exit Speed, V3pbase or V6pbase (mph), Single-Lane Sites A ct ua l E xi t S pe ed , V 3a o r V 6a (m ph ), Si ng le -L an e Si te s V3 Data Match Line 85th %ile (15+ obs.) V6 Data Linear (85th %ile (15+ obs.)) Figure 47. Actual vs. unadjusted predicted values for through- movement and left-turn-movement exit speeds, single-lane sites.

assigned a predicted exit speed of 45 mph. However, the clus- ter of data points with predicted speeds in the 30- to 40-mph range (for which such arbitrary assignments were not made) suggests a significant error in the current method for pre- dicting exit speeds. To improve the prediction fit for exit speeds, the following formulation is proposed: where V3  V3 speed (mph) V3pbase  V3 speed predicted based on path radius (mph) V2  V2 speed predicted based on path radius (mph) a23  acceleration along the length between the mid- point of V2 path and the point of interest along V3 path  6.9 ft/s2 (see text) d23  distance between midpoint of V2 path and point of interest along V3 path (ft) where V3  V3 speed (km/h) V3pbase  V3 speed predicted based on path radius (km/h) V V a d V pbase 3 2 2 23 23 3 3 6 3 6 2 = ⎛⎝⎜ ⎞⎠⎟ + ⎧ ⎨ ⎪⎪⎪ ⎩ ⎪ min . . ⎪⎪ ⎫ ⎬ ⎪⎪⎪ ⎭ ⎪⎪⎪ ( )5-4b, Metric V V V a d pbase 3 3 2 2 23 23 1 1 47 1 47 2 = + ⎧ ⎨ ⎪⎪ ⎩ ⎪⎪ ⎫ min . ( . ) ⎬ ⎪⎪ ⎭ ⎪⎪ ( )5-4a, U.S. Customary V2  V2 speed predicted based on path radius (km/h) a23  acceleration along the length between the mid- point of V2 path and the point of interest along V3 path  2.1 m/s2 (see text) d23  distance between midpoint of V2 path and point of interest along V3 path (m) Vehicle acceleration rates are documented in Exhibit 2-24 in the 2001 AASHTO Policy (34), which is based on the find- ings in NCHRP Report 270 (35). For vehicles in the speed range of 20 to 30 mph (32 to 48 km/h), the latter reference suggests a “design” car acceleration of 0.137 g, or 4.4 ft/s2 (1.3 m/s2), which is the equivalent of 0 to 60 mph in 20.0 s or 0 to 100 km/h in 20.7 s. Most cars today are capable of straight-line accelerations of at least twice that value, how- ever. Therefore, the data from this study were reviewed to estimate a reasonable acceleration rate exhibited in the field by drivers at roundabout exits. The average acceleration exhibited by exiting vehicles was estimated by determining the average acceleration between the 85th-percentile field-measured speed for either V2 and V3 for through vehicles or V4 and V6 for left-turning vehicles, using one half of the distance measured on the plans between the locations where the field measurements were taken (mid- point of the splitter island and the boundary between circu- latory roadway and exit, respectively). The use of one half of the distance is arbitrary but reasonable, as it approximates the drivers’ ability to accelerate along only approximately half of the vehicular path between (1) the point where they pass the last splitter island prior to exiting and (2) the exit point. This procedure estimated an average acceleration of 6.9 ft/s2 (2.1 m/s2), which is the equivalent of 0 to 60 mph in 12.7 s or 0 to 100 km/h in 13.2 s. This acceleration rate appears rea- sonable and conservative for design purposes. 71 y = 0.3638x + 11.549 R 2 = 0.209 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Unadjusted Predicted Exit Speed, V3pbase or V6pbase (mph), Multilane Sites A ct ua l E xi t S pe ed , V 3a o r V 6a (m ph ), M ul til an e Si te s V3 Data Match Line 85th %ile (15+ obs.) V6 Data Linear (85th %ile (15+ obs.)) Figure 48. Actual vs. unadjusted predicted values for through- movement and left-turn-movement exit speeds, multilane sites.

Using this new formulation, 85th-percentile speeds were plotted against adjusted predicted speeds for all sites, the sub- set of single-lane sites, and the subset of multilane sites, as shown in Figures 49, 50, and 51, respectively. Estimates for mean deviation and RMSE for both the unadjusted and adjusted predictions are shown in Table 55. The CO03 (Golden, Colorado) site was eliminated from this analysis because of the lack of circulating speed field data to calibrate predicted circulating speeds. As can be seen, the revised pre- diction is substantially better than the unadjusted prediction; it also eliminates the error associated with estimating a speed for a tangential or nearly tangential exit. The same data can be presented by site, as shown in Table 56. This table shows that, for most sites, exit speeds appear to be governed by circulating speed and acceleration rather than exit path radius. However, a few single-lane sites—MD02 (Leeds, Maryland), MD03 (Jarrettsville, Mary- land), ME01 (Gorham, Maine), OR01 (Bend, Oregon), and WA05 (Sammamish, Washington)—have one or more exit movements whose speeds appear to be governed principally by exit path radius. Based on this analysis, the proposed exit speed prediction method appears to be a substantial improvement on the current method. 72 y = 0.6694x + 5.9115 R2 = 0.5156 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Adjusted Predicted Speed, V3p2 or V6p2 (mph), All Sites A ct u a l S pe ed , V 3a o r V 6a (m ph ), A ll S ite s Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) y = 0.4702x + 9.8221 R2 = 0.2224 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Adjusted Predicted Speed, V3p2 or V6p2 (mph), Single-Lane Sites A ct u a l S pe ed , V 3a o r V 6a (m ph ), S ing le La n e Si te s Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 49. Actual vs. adjusted predicted values for exit speeds, all sites. Figure 50. Actual vs. adjusted predicted values for exit speeds, single-lane sites.

Entry Speed Table 57 summarizes the characteristics of the entry speed (V1) data. As the table shows, 1,140 of the 1,503 observations, or 76%, occur at sites with 15 or more observations. These sites with 15 or more observations were used in determining percentiles. Figures 52, 53, and 54 present plots of predicted versus actual entry speeds differentiated by through movements and left-turn movements for all sites, the subset of single-lane sites, and the subset of multilane sites, respectively, along with 85th-percentile values for all sites with 15 or more observations. As can be seen from the figures, the pattern is similar to that observed for exit speeds, with considerable overestimation of 85th-percentile speeds. As with the exit speeds, the cluster of sites with pre- dicted speeds of around 45 mph is arbitrary,as tangential entries with a path radius of infinity were arbitrarily assigned an entry speed of 45 mph. However, the cluster of data points with predicted speeds in the 30- to 40-mph range (for which such arbitrary assignments were not made) suggests a significant error in the current method for predicting exit speeds. To improve the prediction fit for entry speeds, the follow- ing formulation is proposed: where V1  V1 speed (mph) V1pbase  V1 speed predicted based on path radius (mph) V2  V2 speed predicted based on path radius (mph) a12  deceleration between the point of interest along V1 path and the midpoint of V2 path  4.2 ft/s2 (see text) d12  distance along the vehicle path between the point of interest along V1 path and the midpoint of V2 path (ft) V V V a d pbase 1 2 2 12 12 1 1 1 47 1 47 2 = ⎧ ⎨ ⎪⎪ ⎩ ⎪⎪ ⎫ min . ( . ) – ⎬ ⎪⎪ ⎭ ⎪⎪ ( )5-5a, U.S. Customary 73 y = 0.5495x + 10.335 R2 = 0.4275 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Adjusted Predicted Speed, V3p2 or V6p2 (mph), Multilane Sites A ct u a l S pe ed , V 3a o r V 6a (m ph ), M u lti la n e Si te s Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 51. Actual vs. adjusted predicted values for exit speeds, multilane sites. V3 Prediction Mean Deviation Root Mean Square Error Unadjusted All Sites Single-Lane Sites Multilane Sites 9.1 mph (14.6 km/h) 8.3 mph (13.4 km/h) 12.0 mph (19.3 km/h) 11.6 mph (18.7 km/h) 11.5 mph (18.5 km/h) 13.7 mph (22.0 km/h) Adjusted All Sites Single-Lane Sites Multilane Sites 1.8 mph (2.9 km/h) 1.8 mph (2.9 km/h) 2.1 mph (3.4 km/h) 3.6 mph (5.8 km/h) 3.5 mph (5.6 km/h) 4.2 mph (6.8 km/h) Table 55. Exit speed prediction error.

74 Site Movement Type Number of Observations 85th-Percentile Field Speed (mph) Raw Predicted Speed (mph) Revised Predicted Speed (mph) Controlling Factor MD01-N Through 75 24.0 45.0 29.4 Circ and accel WA01-E Through 41 20.0 27.2 18.3 Circ and accel VT01-S Through 37 17.0 23.8 20.7 Circ and accel WA07-N Left turn 35 18.0 27.3 20.9 Circ and accel WA01-N Left turn 34 23.1 28.5 22.7 Circ and accel NV02-N Through 33 30.2 42.9 32.2 Circ and accel CO01-W Left turn 32 23.0 28.1 22.4 Circ and accel MI01-W Left turn 32 24.0 34.8 18.8 Circ and accel CO02-W Through 31 20.5 34.4 27.8 Circ and accel WA07-S Through 31 23.0 22.5 22.1 Circ and accel MD06-N Left turn 31 20.0 25.0 21.0 Circ and accel WA04-N Left turn 31 24.5 34.1 27.5 Circ and accel CO02-E Through 30 23.7 32.7 29.2 Circ and accel NV02-E Through 30 28.0 37.6 30.1 Circ and accel WA03-E Through 30 18.0 26.5 22.9 Circ and accel VT01-N Left turn 30 13.7 29.5 19.9 Circ and accel MD01-S Through 29 20.0 26.0 21.4 Circ and accel OR01-N Through 29 20.8 20.5 20.5 Exit path radius MD07-N Left turn 29 22.0 33.4 20.1 Circ and accel OR01-W Left turn 29 20.0 18.8 18.8 Exit path radius WA03-E Left turn 29 19.0 23.2 20.0 Circ and accel OR01-W Through 28 22.0 21.2 21.2 Exit path radius VT03-N Through 28 23.0 31.8 28.0 Circ and accel NV02-W Through 27 32.1 43.8 32.1 Circ and accel VT03-S Through 25 25.0 31.8 27.9 Circ and accel MD03-W Through 24 22.0 22.5 22.5 Exit path radius VT03-W Through 24 26.0 28.5 27.9 Circ and accel MI01-E Through 23 23.0 43.8 25.9 Circ and accel VT01-N Through 23 15.3 28.9 22.6 Circ and accel WA04-E Left turn 23 27.0 45.0 27.1 Circ and accel VT03-E Through 22 25.0 33.6 28.3 Circ and accel WA05-W Left turn 21 17.0 18.3 18.3 Exit path radius WA05-W Through 20 19.0 21.2 20.0 Circ and accel MD01-E Left turn 20 21.0 45.0 21.2 Circ and accel MD04-E Left turn 19 19.0 36.8 17.7 Circ and accel ME01-E Through 18 19.5 43.8 20.4 Circ and accel WA02-E Left turn 18 14.9 25.0 19.3 Circ and accel MD02-E Through 17 19.0 21.8 21.8 Exit path radius WA02-W Through 17 20.0 27.6 27.4 Circ and accel MD03-N Through 16 23.5 22.5 21.2 Circ and accel OR01-S Through 16 20.0 18.8 18.8 Exit path radius MD01-N Left turn 16 19.0 21.8 20.8 Circ and accel WA01-E Left turn 16 25.0 29.5 22.5 Circ and accel VT02-N Left turn 15 19.0 22.5 18.8 Circ and accel MD02-N Through 15 21.8 36.1 21.2 Circ and accel MD02-S Through 15 20.0 36.1 24.4 Circ and accel MD02-W Through 15 17.9 23.2 21.8 Circ and accel MD07-N Through 15 20.0 43.8 28.8 Circ and accel MI01-W Through 15 25.0 43.8 33.6 Circ and accel NV02-S Through 15 33.0 43.8 32.2 Circ and accel WA02-E Through 15 21.9 43.8 25.0 Circ and accel WA03-S Through 15 19.0 27.2 22.4 Circ and accel WA03-W Through 15 17.0 25.0 21.6 Circ and accel WA05-E Through 15 17.0 19.7 19.7 Exit path radius CO02-E Left turn 15 22.0 36.8 22.7 Circ and accel CO02-S Left turn 15 26.8 32.7 24.0 Circ and accel ME01-N Left turn 15 20.7 18.8 18.8 Exit path radius VT02-E Left turn 15 19.0 27.6 20.3 Circ and accel Legend: Circ = circulating speed; accel = acceleration Table 56. Exit speed prediction by site.

75 Characteristic Through- Movement Entry Speed, V1 Left-Turn-Movement Entry Speed, V1L Total Range of measured speeds 8 – 35 mph (13 – 56 km/h) 8 – 30 mph (13 – 48 km/h) 8 – 35 mph (13 – 56 km/h) Total number of sites 61 63 124 Total number of observations 927 576 1,503 Number of sites with 15+ observations 34 21 55 Number of observations at sites with 15+ observations 738 402 1,140 Table 57. Summary of entry speed (V1) data. y = 0.3437x + 9.8447 R2 = 0.4772 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Unadjusted Predicted Speed, V1p (mph), All Sites A ct u a l S pe ed , V 1a (m ph ), A ll S ite s V1 Data V1L Data 85th %ile (15+ obs.) Match Line Linear (85th %ile (15+ obs.)) y = 0.1452x + 14.411 R2 = 0.0772 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Unadjusted Predicted Speed, V1p (mph), Single-Lane Sites A ct u a l S pe ed , V 1a (m ph ), S ing le -L a n e Si te s V1 Data V1L Data 85th %ile (15+ obs.) Match Line Linear (85th %ile (15+ obs.)) Figure 52. Actual vs. unadjusted predicted values for entry speeds, all sites. Figure 53. Actual vs. unadjusted predicted values for entry speeds, single-lane sites.

where V1  V1 speed (km/h) V1pbase  V1 speed predicted based on path radius (km/h) V2  V2 speed predicted based on path radius (km/h) a12  deceleration between the point of interest along V1 path and the midpoint of V2 path  1.3 m/s2 (see text) d12  distance along the vehicle path between the point of interest along V1 path and the midpoint of V2 path (m) The 2001 AASHTO Policy recommends a deceleration rate of 11.2 ft/s2 (3.4 m/s2) when calculating stopping sight distance (34). In addition, a deceleration rate of 10 ft/s2 (3.0 m/s2) is commonly assumed when calculating clearance intervals for signalized intersections (36). Deceleration under either of those conditions, however, is likely to be higher than at the approach to a roundabout, because the need to deceler- ate for an object in the roadway or for a change in signal indi- cation is less predictable than the need to decelerate upon entry into a roundabout. Therefore, it seems reasonable that the deceleration for drivers anticipating a slower circulating speed will be more gradual than that used for signalized intersections. The average deceleration exhibited by entering vehicles was estimated by determining the average deceleration between V V a d V pbase 1 2 2 12 12 1 3 6 3 6 2 = ⎛⎝⎜ ⎞⎠⎟ ⎧ ⎨ ⎪⎪⎪ ⎩ ⎪ min . . –⎪⎪ ⎫ ⎬ ⎪⎪⎪ ⎭ ⎪⎪⎪ ( )5-5b, Metric the 85th-percentile field-measured speed for either V1 and V2 for through vehicles or V1L and V4 for left-turning vehicles, using one half of the distance measured on the plans between the locations where the field measurements were taken (entry-circulatory roadway boundary and splitter island, respectively). As discussed previously for exit speeds, the use of one half of the distance is arbitrary but reasonable. This procedure estimated an average deceleration of 4.2 ft/s2 (1.3 m/s2). This deceleration rate appears reasonable for design purposes. Using this new formulation, 85th-percentile speeds were plotted against adjusted predicted speeds for all sites, the sub- set of single-lane sites, and the subset of multilane sites, as shown in Figures 55, 56, and 57, respectively. Estimates for RMSE for both the unadjusted and adjusted predictions are shown in Table 58. The CO03 (Golden, Colorado) and WA02 (Gig Harbor, Washington) sites were eliminated from this analysis because of the lack of circulating speed field data to calibrate predicted circulating speeds. As can be seen, the revised prediction is substantially better than the unadjusted prediction. The same data can be presented by site, as shown in Table 59. This table shows that, for most sites, entry speeds appear to be governed by deceleration into an anticipated circulating speed rather than an entry speed governed by entry path radius alone. However, a few sites—MD01 (Bel Air, Maryland), MD03 (Jarrettsville, Maryland), ME01 (Gorham, Maine), and OR01 (Bend, Oregon) among the single-lane sites and MI01 (Oke- mos, Michigan) among the multilane sites—have one or more entry movements whose speeds appear to be governed princi- pally by entry path radius. Based on this analysis, the proposed entry-speed predic- tion method appears to be a substantial improvement on the 76 y = 0.4266x + 7.3281 R2 = 0.4446 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Unadjusted Predicted Speed, V1p (mph), Multilane Sites A ct u a l S pe ed , V 1a (m ph ), M u lti la n e Si te s V1 Data V1L Data 85th %ile (15+ obs.) Match Line Linear (85th %ile (15+ obs.)) Figure 54. Actual vs. unadjusted predicted values for entry speeds, multilane sites.

77 y = 0.6237x + 6.0316 R2 = 0.3491 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Adjusted Predicted Speed, V1p1 (mph), All Sites A ct u a l S pe ed , V 1a (m ph ), A ll S ite s Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) y = 0.2051x + 13.762 R2 = 0.0275 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Adjusted Predicted Speed, V1p1 (mph), Single-Lane Sites A ct u a l S pe ed , V 1a (m ph ), S ing le -L a n e Si te s Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 55. Actual vs. adjusted predicted values for entry speeds, all sites. Figure 56. Actual vs. adjusted predicted values for entry speeds, single-lane sites. current method. However, given the hesitancy currently exhibited by drivers under capacity conditions, the observed entry speeds may increase over time after drivers acclimate further. Therefore, the research team believes that an analyst should be cautious when using deceleration as a limiting fac- tor when establishing entry speeds for design. Furthermore, the research team believes that a good design should rely more heavily on controlling the entry path radius as the pri- mary method for controlling entry speed, particularly for the fastest combination of entry and circulating path (typically the through movement). Another factor that may influence entry speeds is the amount of available intersection sight distance. The research team gave this hypothesis a cursory evaluation but was unable to pursue it in detail. Additional research on the effect of sight distance on entry speed is recommended. Conclusions and Recommendations Based on the analysis presented herein, the following con- clusions can be made. • Current speed prediction methods for predicting 85th- percentile circulating speeds appear to be reliable. Speed prediction is better for movements that track closely around the central island, such as left-turn paths, than for those that are influenced by the central island but do not precisely track around it, such as through-movement paths.

• Current speed prediction methods significantly overesti- mate entry and exit speeds, particularly for entry paths and exit paths that are tangential or nearly tangential. These prediction methods are significantly improved by incorpo- rating acceleration and deceleration effects as they relate to predicted circulating speeds. Pedestrian Analysis Data were collected for 769 pedestrian crossing events that occurred at 10 legs, which were distributed among seven roundabouts. The overwhelming majority of these crossings involved adults traversing the crosswalk at a normal pace. There were 6 youth-only crossings and 19 youth-with-adult crossings. There were five crossings on skates or skateboards and five that involved pedestrians walking bicycles. Finally, there were 13 crossings with strollers and 8 crossings in motorized wheelchairs. There were no crossings observed by pedestrians with other diminished capabilities, such as visual impairments. The analysis of these data was site-based, with a site defined as a leg at a roundabout. Thus, it was possible to have more than one site (leg) at a selected roundabout. Average values for the behaviors at each site were produced and then aggre- gated to produce overall means. The alternative analysis would have been a pedestrian-based approach, in which all pedestrian events for a given variable were first aggregated before developing means. As previously shown in Table 11, there was also a large range in the number of pedestrian events observed at the various legs. Because the goal of this analysis was to specifically look for geometric or operational features that may contribute to the observed behaviors, the site-based approach was the best choice. By developing site means, any weighting bias due to large sample sizes at one or more sites is removed. The analysis results are presented in terms of number of lanes on either the entry side or exit side of the leg at the loca- tion of the crosswalk. For the 10 legs in the analysis, 5 had one lane in each direction and 5 had two lanes in each direction. In some cases, the number of lanes at the yield line of the 78 Prediction Mean Deviation Root Mean Square Error Unadjusted prediction All Sites Single-Lane Sites Multilane Sites 9.1 mph (14.6 km/h) 6.6 mph (10.6 km/h) 13.2 mph (21.2 km/h) 11.0 mph (17.7 km/h) 8.6 mph (13.8 km/h) 14.6 mph (23.5 km/h) Adjusted prediction All Sites Single-Lane Sites Multilane Sites 2.0 mph (3.2 km/h) 2.6 mph (4.2 km/h) 1.5 mph (2.4 km/h) 3.8 mph (6.1 km/h) 4.2 mph (6.8 km/h) 4.0 mph (6.4 km/h) Table 58. Entry speed prediction error. y = 0.5169x + 10.035 R2 = 0.3596 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Adjusted Predicted Speed, V1p1 (mph), Multilane Sites A ct u a l S pe ed , V 1a (m ph ), M u lti la n e Si te s Match Line 85th %ile (15+ obs.) Linear (85th %ile (15+ obs.)) Figure 57. Actual vs. adjusted predicted values for entry speeds, multilane sites.

79 Site Movement Type Number of Observations 85th-Percentile Field Speed (mph) Raw Predicted Speed (mph) Revised Predicted Speed (mph) Controlling Factor VT01-N Through 40 16.0 22.5 20.2 Circ and decel OR01-S Through 35 15.0 19.1 19.1 Entry path radius VT01-S Through 35 17.1 27.2 19.3 Circ and decel CO03-E Through 32 22.0 37.6 — — CO03-W Through 31 25.0 33.6 — — WA05-W Through 31 19.0 22.5 21.3 Circ and decel CO02-E Through 30 18.7 34.4 26.1 Circ and decel MI01-E Through 30 19.0 23.2 23.2 Entry path radius CO01-W Left turn 30 22.0 45.0 21.6 Circ and decel CO02-W Through 29 21.8 36.1 26.5 Circ and decel MD01-S Through 29 16.8 20.0 20.0 Entry path radius MD03-E Through 29 22.0 22.5 20.8 Circ and decel WA01-E Left turn 29 19.0 26.0 19.0 Circ and decel WA07-N Left turn 29 18.0 21.2 19.2 Circ and decel WA07-S Through 28 21.0 24.8 19.9 Circ and decel MD04-E Left turn 25 22.0 32.7 16.3 Circ and decel MD07-N Left turn 25 20.0 30.8 18.8 Circ and decel MI01-W Left turn 23 18.0 28.9 16.8 Circ and decel VT03-N Through 22 19.9 27.6 27.3 Circ and decel WA03-E Through 21 14.0 25.0 21.3 Circ and decel MD01-N Through 20 21.2 19.1 19.1 Entry path radius MD02-N Through 20 20.0 26.0 20.8 Circ and decel VT03-E Left turn 19 21.3 29.8 23.9 Circ and decel VT03-S Left turn 19 22.3 37.6 22.9 Circ and decel WA04-S Left turn 18 20.5 30.5 24.1 Circ and decel WA01-E Through 17 18.2 26.0 18.3 Circ and decel NV03-W Left turn 17 29.6 43.8 29.0 Circ and decel MD03-S Through 16 22.5 23.2 21.5 Circ and decel NV02-E Through 16 24.8 42.9 28.8 Circ and decel NV02-S Through 16 25.0 37.6 30.9 Circ and decel OR01-N Through 16 17.0 19.7 19.7 Entry path radius MD07-W Left turn 16 14.0 25.5 18.6 Circ and decel NV02-N Left turn 16 26.5 42.9 22.8 Circ and decel WA05-E Left turn 16 19.8 21.8 18.1 Circ and decel KS01-E Through 15 25.5 42.8 23.2 Circ and decel KS01-W Through 15 23.9 37.0 22.8 Circ and decel MD02-S Through 15 20.9 24.8 23.8 Circ and decel MD03-N Through 15 22.0 20.5 20.5 Entry path radius MD07-S Through 15 15.0 26.5 24.6 Circ and decel ME01-W Through 15 15.9 18.8 18.8 Entry path radius NV02-W Through 15 27.0 37.6 31.0 Circ and decel OR01-E Through 15 16.0 18.8 18.8 Entry path radius VT03-S Through 15 23.0 37.6 25.7 Circ and decel WA01-W Through 15 17.0 25.5 21.1 Circ and decel WA02-E Through 15 15.9 23.8 23.8 Circ and decel WA02-W Through 15 17.0 29.8 26.4 Circ and decel WA03-N Through 15 13.9 22.5 20.8 Circ and decel CO02-S Left turn 15 14.9 36.8 20.2 Circ and decel KS01-S Left turn 15 23.8 43.8 22.4 Circ and decel MD04-S Left turn 15 16.9 27.6 16.4 Circ and decel NV03-S Left turn 15 26.9 36.8 27.5 Circ and decel OR01-W Left turn 15 16.9 19.7 19.7 Entry path radius VT01-N Left turn 15 16.9 22.5 17.6 Circ and decel WA03-E Left turn 15 13.0 25.0 17.3 Circ and decel WA04-E Left turn 15 19.8 32.7 23.7 Circ and decel Legend: Circ = circulating speed; decel = deceleration Table 59. Entry speed prediction by site.

roundabout did not match the number of lanes at the loca- tion of the crosswalk. For example, one leg had two lanes at the yield line for motor vehicles but only a single lane at the point where the crosswalk was located. This leg was classified as a one-lane site in the analysis. The results are also stratified by entry side and exit side, both in terms of where the pedes- trian initiated the crossing and in terms of the behaviors on each side. There have been a number of concerns raised over the safety of pedestrians within the exit lanes specifically. The results were stratified to determine if there are differences in the behaviors on each side of the crossing. The analysis results are presented in the following sections: • Pedestrian Crossing Behaviors • Motorist Behaviors • Pedestrian-Motor Vehicle Conflicts • Pedestrian Crossing/Wait Times • Comparison to Other Intersection Types For further breakdowns of the figures provided,Appendix K includes several tables with results on a site-by-site basis. Appendix L includes images to help describe some of the maneuvers observed. Pedestrian Crossing Behaviors The examination of pedestrian crossing behaviors revealed that the majority of crossings involved no interaction with a motor vehicle, where interaction is defined as the pedestrian either accepting or rejecting a gap when a vehicle was present. Figure 58 shows the percentages of crossings requiring inter- action with a vehicle on either the entry side or exit side as a function of where the crossing was initiated. For one-lane sites, there was virtually no difference in the level of interac- tion between the entry and exit sides if starting position is not considered: 27% on the entry side and 26% on the exit side. For crossings that started on the entry side, 32% of the cross- ings required interaction on the entry side (the first stage of a two-stage crossing), while only 27% of such crossings required interaction on the exit side (the second stage of a two-stage crossing). For crossings that started on the exit side, the numbers are reversed: 26% of the crossings required interaction on the exit side, while 22% required interaction on the entry side. This same pattern was true for exit starts for two-lane sites. Almost half (45%) of such starts required interaction on the exit side, as opposed to 22% requiring interaction on the entry side. For entry side starts on two-lane sites, the numbers were reversed: 26% of these starts required interaction on the entry side, while 33% required interaction on the exit side. Overall, the level of interaction was greater on the exit side (39%) compared to the entry side (24%) when the starting position is not considered. For those pedestrians who did interact with vehicles and ultimately crossed the leg, their behaviors were categorized as one of the following: • Normal: Pedestrian crossed the street at a normal pace (walking speed). None of the following behaviors were observed, and the vehicle yielded. • Hesitates: Pedestrian hesitated on the curb or splitter island because of an approaching vehicle. Most often, the hesitation occurred while the pedestrian made visual or other contact with the driver. Once this communication was made and the vehicle began slowing, the pedestrian would then proceed with the crossing. 80 32 22 27 27 26 2626 22 24 33 45 39 29 22 25 30 35 33 0 5 10 15 20 25 30 35 40 45 50 Entry Side Start Exit Side Start Either Side Entry Side Start Exit Side Start Either Side Interaction on Entry Side Interaction on Exit Side Pe rc en ta ge 1-Lane 2-Lane All Figure 58. Percentage of crossings in which the pedestrian is required to interact with one or more motor vehicles.

• Retreats: Pedestrian began crossing and then retreated to the curb or splitter island because of an approaching vehicle. • Runs: Pedestrian ran across the leg because of an oncom- ing vehicle. Note that running did not indicate that a con- flict was imminent; it simply indicates a choice that was made by the pedestrian. Conflicts are covered in a later section of this report. Figures 59 and 60 show the distributions of these behaviors based on whether the crossing began on the entry side or exit side, respectively. The one behavior that did not occur for any of the observed crossings was a retreat to the curb or splitter island. For crossings that began on the entry side (see Figure 59), approximately 60% of the crossings were considered to be normal when considering all sites and either side of the cross- ing. The most observed non-normal behavior on the entry side was hesitation: 25% of the pedestrians hesitated when crossing one lane, while 40% of the pedestrians hesitated when crossing two lanes. The hesitation on the splitter island (captured under the exit side behavior) was much lower at 9% and 12% for crossing one and two lanes, respectively. The other behavior that was observed was running. For entry-side starts, the running behavior was much more prevalent on the exit side: 39% of the pedestrians completed their crossings by running across one-lane sites, while only 19% were observed to run across the exit side on two-lane sites. For both site types, the level of running was much lower on the entry side: 12% and 3% for one-lane and two-lane sites, respectively. 81 63 25 12 53 9 39 57 40 3 69 12 19 60 33 7 61 10 29 0 10 20 30 40 50 60 70 80 Normal Hesitates Runs Normal Hesitates Runs Behavior on Entry Side Behavior on Exit Side Pe rc en ta ge 1-Lane 2-Lane All 76 10 14 44 52 4 47 46 7 63 32 3 57 23 10 47 39 3 0 10 20 30 40 50 60 70 80 90 Normal Hesitates Runs Normal Hesitates Runs Behavior on Entry Side Behavior on Exit Side Pe rc en ta ge 1-Lane 2-Lane All Figure 59. Pedestrian crossing behaviors when a vehicle was present and the crossing began on the entry side. Figure 60. Pedestrian crossing behaviors when a vehicle was present and the crossing began on the exit side.

For crossings that were initiated on the exit side (see Fig- ure 60), the overall percentage of crossings that were coded as normal is lower than what was observed for entry side starts. For one-lane sites, 76% of the crossings were normal on the entry side, while only 44% of such crossings were normal on the exit side. For two-lane sites, 47% of the crossings were normal on the entry side, and 63% were considered normal on the exit side. Similar to entry-side starts, the most-observed non-normal behavior for exit-side starts was hesitation. For one-lane sites, 52% of the pedestrians hesitated on the exit side (which is the starting side in this case), while only 10% hesi- tated on the splitter island. For two-lane sites, 32% of the pedestrians hesitated on the exit side, and 46% hesitated on the splitter island. Crossings starting on the exit side of two- lane sites are the only instances in which the hesitation on the splitter island exceeded the hesitation on the initial curb. With respect to the behavior of running, the patterns for exit-side starts were similar to those for entry-side starts, i.e., pedestrians ran more often on the second stage of the cross- ing. At one-lane sites, 4% of the pedestrians ran across the exit side, while 14% ran across the entry side. At two-lane sites, 3% ran across the exit side, while 7% ran across the entry side. Another behavior that was observed for pedestrian cross- ings was whether the crossing was made within or outside the boundaries of the crosswalk. The overwhelming majority of crossings were made within the crosswalk boundaries (see Figure 61). However, 17% of the crossings at one-lane sites and 12% at two-lane sites occurred completely outside the crosswalk lines. For these crossings, 38% involved pedestrians who hesitated when crossing the entry side, while 27% did the same on the exit side. An additional 8% and 9% of these crossings involved pedestrians who ran across the entry and exit sides, respectively. At one of the one-lane sites (MD05, Towson, MD), half of the pedestrians observed to cross out of the crosswalk used a center turn lane that was located upstream of the splitter island as a refuge area. Motorist Behaviors For each pedestrian event captured, the behavior of the motorist was also recorded. These behaviors were collapsed into the following three categories for the analysis: • Active yield: The motorist slowed or stopped for a cross- ing pedestrian or a pedestrian waiting on the curb or split- ter island to cross. The pedestrian was the only reason the motorist stopped or slowed. • Passive yield: The motorist yielded to the pedestrian but was already stopped for another reason. This situation occurred most often when there was a queue of vehicles waiting to enter the roundabout or when the vehicle was already stopped for a prior pedestrian crossing event. • Did not yield: The motorist did not yield to a crossing pedestrian or a pedestrian waiting on the curb or splitter island to cross. Figure 62 shows the yielding behavior results when the pedestrian crossing is initiated on the entry side. For one-lane sites, 15% of the motorists did not yield to the pedestrian on either the entry or exit side. The remainder of the exit-side vehicles actively yielded. The remainder of the entry-side vehicles included 20% that were classified as passively yield- ing. For two-lane sites, the percentage of non-yielding vehi- cles increases to 33% on the entry side and 45% on the exit side. For those vehicles that did yield, 9% and 2% were classi- fied as passive yield for the entry and exit sides, respectively. 82 81 2 0 17 84 2 2 12 82 2 1 15 0 10 20 30 40 50 60 70 80 90 Crosswalk In-Entry/Out-Exit Out-Entry/In-Exit Off Crosswalk Pe rc en ta ge 1-Lane 2-Lane All Figure 61. Pedestrian behaviors related to position in the crosswalk when crossing.

The motorist yielding behavior results for pedestrian cross- ings that started on the exit side are shown in Figure 63. For one-lane sites, 29% of the motorists did not yield to the pedestrian on the exit side, and 10% did not yield on the entry side. Of the vehicles that did yield, 14% and 18% passively yielded on the exit and entry sides, respectively. For two-lane sites, the percentage of vehicles not yielding increased, just as it did for the entry-start crossings. The percentage of non- yielding vehicles increased to 62% on the exit side and 33% on the entry side. For those vehicles that did yield, 9% and 7% were classified as passive yield for the exit and entry sides, respectively. Overall, when looking at the entire two-stage crossing, approximately 27% of the motorists did not yield to crossing or waiting pedestrians that started crossing from the entry side. The percentage of non-yielding motorists increases to 34% for crossings initiated on the exit side. In addition, the lack of yielding on two-lane sites (43%) is substantially worse than on one-lane sites (17%). Yielding behavior was also observed for sites in different regions of the country, specifically east versus west. Sites from Florida, Maryland, and Vermont were included in the east group, while the west group included locations from Wash- ington, Nevada, and Utah. Each region was balanced to include two one-lane sites and two two-lane sites. Motorist non-yielding behavior was observed more often at the east- ern sites (35%) compared to the western sites (27%). The difference was most pronounced on the exit side, where the east and west non-yield percentages were 48% and 29%, respectively. These observations suggest that perhaps driver 83 66 20 15 84 15 58 9 33 52 2 45 62 14 24 68 0 1 30 0 10 20 30 40 50 60 70 80 90 Active Yield Passive Yield Did Not Yield Active Yield Passive Yield Did Not Yield Behavior on Entry Side Behavior on Exit Side Pe rc en ta ge 1-Lane 2-Lane All Figure 62. Yielding behavior of motorists when the pedestrian crossing begins on the entry side. 72 18 10 57 14 29 60 7 33 29 9 62 66 12 21 43 11 46 0 10 20 30 40 50 60 70 80 Active Yield Passive Yield Did Not Yield Active Yield Passive Yield Did Not Yield Behavior on Entry Side Behavior on Exit Side Pe rc en ta ge 1-Lane 2-Lane All Figure 63. Yielding behavior of motorists when the pedestrian crossing begins on the exit side.

pedestrians conflicting vehicle bus Figure 66. Pedestrian Conflict 3 at MD05SW-S (Towson, MD). behavior with respect to pedestrians may be influenced by more than just the design. Pedestrian-Motor Vehicle Conflicts One of the surrogate measures of safety for pedestrians is a conflict with a motor vehicle. In this study, as well as many others, a conflict was defined as an interaction between a pedestrian and motorist in which one of the parties had to suddenly change course and/or speed to avoid a crash. During the 769 pedestrian crossing events, only four conflicts were observed (0.5%). Two of these conflicts occurred at one one- lane site and two occurred at a different one-lane site. The con- flicts were also divided between the entry and exit sides (one each) at each site. Each conflict is further described below: • Conflict 1 – WA03-S (Bainbridge Island, Washington) entry side: Pedestrian emerges from a shadow; approach- ing vehicle brakes hard (Figure 64). • Conflict 2 – WA03-S (Bainbridge Island, Washington) exit side: Pedestrian crosses exit side; exiting vehicle brakes hard and swerves left (Figure 65). • Conflict 3 – MD05SW-S (Towson, Maryland) entry side: Pedestrians come from behind a stopped bus; approaching vehicle brakes hard (Figure 66). • Conflict 4 – MD05SW-S (Towson, Maryland) exit side: Pedestrian crosses exit side close to circulating lanes; exit- ing vehicle brakes hard (Figure 67). Rather than view conflicts in absolute numbers, another approach is to calculate a conflict rate based on opportuni- ties. In this study, an opportunity was defined as any time a pedestrian was either waiting to cross or crossing the leg and a motor vehicle was in the vicinity of the pedestrian. To avoid 84 pedestrian Figure 64. Pedestrian Conflict 1 at WA03-S (Bainbridge Island, WA). Figure 65. Pedestrian Conflict 2 at WA03-S (Bainbridge Island, WA). Figure 67. Pedestrian Conflict 4 at MD05SW-S (Towson, MD).

a conflict, both parties had to respond correctly. The pedes- trian had to reject gaps when the motorist did not yield, and the motorist had to yield when the pedestrian was crossing. Table 60 shows the rates of pedestrian-vehicle conflicts across all study sites. The rate for one-lane sites (2.8 conflicts/ 1000 opportunities) was slightly greater than the rate for two-lane sites (2.0 conflicts/1000 opportunities). Similar rates were calculated for the two sites where the con- flicts were observed. The one-lane site, WA03-S (Bainbridge Island, Washington), had a rate of 7.1 conflicts/1000 opportu- nities. The two-lane site, MD05SW-S (Towson, Maryland), had a rate of 15.0 conflicts/1000 opportunities. Both of these values are much greater than the mean rates and may provide an indication of a potential safety concern at these sites. Pedestrian Crossing/Waiting Times Times were recorded for each pedestrian event from the point at which the pedestrian arrived until s/he completed the crossing. These data allowed for the derivation of the follow- ing time-based measures: • Initial waiting time: The difference in time between the point of arrival and the time at which the pedestrian began crossing the street. • Splitter time: The difference in time between the arrival time and the departure time at the splitter island. This time included both the time to traverse the splitter island and any time spent waiting on the island. • Crossing time: The difference in the time at the end of the crossing and the time at which the pedestrian began cross- ing the street. The values for each of these measures are shown in Figure 68 for one-lane, two-lane, and all sites. For one-lane sites, the average waiting time was about 1.3 s, irrespective of starting location for the crossing. For two-lane sites, the same value was derived for crossings starting on the entry side. However, crossings initiated on the exit side on two- lane sites required an average waiting time of 2.9 s. Splitter times and crossing times varied little on the basis of starting position. The longer splitter times on two-lane sites (4.7 s mean) compared to one-lane sites (2.0 s mean) was more likely a function of the size of the islands and the time required to traverse it than a function of actual waiting time on the island. Almost all pedestrians traversed the splitter island without stopping; the average waiting time (across all sites) on the splitter island was 0.4 s. The mean crossing time for a one-lane site was 9.0 s and included time to cross the entry lane, splitter island, and exit lane. For two-lane sites, the mean crossing time was 14.4 s. The crossing times are discussed further in the next section of the report as it relates to the pace of the crossing. Comparison to Other Intersection Types The analysis of the pedestrian data collected at round- abouts included an examination of crossing behaviors, 85 Number of Lanes Opportunities Conflicts Conflict Rate/ 1000 Opportunities 1 707 2 2.8 2 1,011 2 2.0 All 1,718 4 2.3 Table 60. Pedestrian-vehicle conflicts per 1000 opportunities. 1.3 1.2 1.3 2.9 1.3 2.11.7 2.3 4.9 4.5 3.3 3.4 9.1 8.9 14.2 14.5 11.7 11.7 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 1-Lane (entry start) 1-Lane (exit start) 2-Lane (entry start) 2-lane (exit start) All (entry start) All (exit start) Ti m e (se co n ds ) Initial Wait Time Splitter Time Crossing Time Figure 68. Average pedestrian waiting and crossing times.

motorist yielding behaviors, and pedestrian waiting and crossing times. In an effort to provide more insight into these results, the roundabout findings were compared to results from intersections with other types of traffic control that are more commonly found in the United States, such as signal- ization and stop signs. In an ongoing FHWA-sponsored proj- ect titled “Safety Index for Assessing Pedestrian and Bicyclist Safety at Intersections” (3), an effort was undertaken to acquire similar types of pedestrian and motorist behavior data at 68 signalized and stop-controlled intersections. Twelve of those sites were on one-way streets and were not included in this comparative analysis; two other sites were also not included in the analysis. For the 54 sites included, a total of 2,881 pedestrian crossing events were observed. This section of the report compares the results for roundabouts from this study and the results for other intersection types from the FHWA study in the following areas: • Pedestrian crossing behavior • Motorist behavior • Pedestrian waiting time • Pedestrian crossing pace Comparison of Pedestrian Crossing Behaviors The FHWA study observed pedestrian crossings at cross- walks and classified their behavior in one of the following categories: • Went around a vehicle that was blocking crosswalk • Ran to avoid approaching vehicle • Stopped while crossing to let vehicle pass • Aborted crossing; stepped into roadway and then stepped back onto the curb to let vehicle pass • Proceeded normally across; did not take one of the above actions As shown in the above list, the behavior categories are similar to the categories used in this research. However, the FHWA study did not collect data on every rejected gap, only those rejected gaps when the pedestrian did something other than stand and wait (i.e., stopped or aborted crossing). Therefore, the results that follow are based on events where there was a vehicle present, which allows for a more accurate comparison. Similar to the analysis for the roundabout research, the FHWA data were summarized on a per-site basis, where mean performance measures were calculated for each site and then averaged together. As with the roundabout analysis, a site is defined as a crossing location or crosswalk; thus, a four-leg intersection could have four sites included in the analysis. At signalized and all-way-stop–controlled intersections, all sites are subject to the same type of traffic control. However, at two-way-stop–controlled intersections, there are two stop-controlled sites and two sites with no control. The behavior distributions from each study are shown in Tables 61 and 62. Because these were two independent stud- ies, the behavior categories were slightly different. However, there is enough similarity to draw reasonable conclusions. The dominant behaviors for crossings at a roundabout were “normal” (proceeded without stopping) at 58% and “hesi- tated on curb” at 27%. The dominant behavior for all other intersections was “proceeded normally”at 88%. However, the FHWA study only recorded pedestrian behavior if the pedes- trian had started crossing, which means that any hesitation on the curb at these locations was captured within the normal crossing category. Combining the normal crossings and curb hesitations for the roundabout crossings produces a value of 85%, which is essentially equivalent to 88% for normal cross- ings at standard intersections. As shown in Table 62, there were differences in the percentage of crossings considered to be normal when considering the type of traffic control: sig- nalized (90%), stop-controlled (100%), and no control (70%). The value of 85% for roundabout crossings falls between the values for no control and signalized control and is expected, given that the yield control present at round- abouts falls between these extremes. Aborted crossings, which were coded as “retreated” in the roundabout effort, were non-existent across all levels of traf- fic control. Crossings in which the pedestrian stopped after starting did not occur at roundabouts or at stop-controlled 86 Pedestrian Behavior Occurrence Normal 58% Hesitated on curb 27% Hesitated after start 1% Stopped after start 0% Retreated 0% Ran 14% Total 100% Occurrence by Traffic Control on Leg with Crosswalk Pedestrian Behavior Signal Stop Sign None All Types of Control Proceeded normally 90% 100% 70% 88% Aborted crossing 0% 0% 0% 0% Went around blocking vehicle 5% 0% 1% 3% Ran to avoid 1% 0% 2% 1% Stopped to let vehicle pass 4% 0% 27% 8% Total 100% 100% 100% 100% Table 62. Pedestrian behavior at common intersection-type crosswalks. Table 61. Pedestrian behavior at roundabout crosswalks.

intersections. It did occur for 4% of the crossings at signal- ized intersections and for 27% of the crossings at uncon- trolled intersections. Running behavior was observed much more often at roundabout crossings than at any other type of crossing. However, running at roundabout crossings was observed to occur mainly during the second half of the crossing and was usually done out of courtesy to waiting motorists, as opposed to a behavior that was required to avoid a conflict. Comparison of Motorist Behaviors The FHWA study observed motorist behavior during pedes- trian crossings and classified it as yielding or not yielding. Data on the yielding behavior were insufficient to categorize the yields as active or passive, as was done in the analysis for this roundabout research. Table 63 shows how the motorist behav- ior at roundabouts compares to behavior at the three other types of traffic control. Almost half (48%) of the vehicles on uncontrolled legs did not yield to pedestrians. The crossings subject to yield control (roundabouts) resulted in 32% of the motorists not yielding. Finally, the stop-controlled and signal- ized intersections produced non-yielding vehicles percentages of 4% and 15%, respectively. These results correlate with the pedestrian behaviors previously discussed and reflect the level of traffic control’s influence on the yielding behavior of motorists. Comparison of Pedestrian Waiting Time Pedestrian waiting time for roundabout crossings was cal- culated as the amount of time between the arrival at the curb and start of crossing plus the waiting time on the splitter island. Waiting time at signalized, stop-controlled, and uncontrolled sites was calculated as the amount of time between arrival and start of crossing. The traffic control type that resulted in the longest pedestrian waiting time (10.7 s) was signalization (see Table 64). This result is expected, as most pedestrians complied with the signals as opposed to selecting their own gaps. Crossings at sites with no traffic con- trol produced an average waiting time of 3.0 s, while round- about crossings caused pedestrians to wait for an average of 2.1 s. Crossings at stop-controlled sites resulted in virtually no waiting time (0.3 s) for pedestrians. Comparison of Pedestrian Crossing Pace Pedestrian crossing pace was calculated as the crossing width divided by the average crossing time for each site. This comparison was based on crossing pace rather than total crossing time so that sites with different lane widths and different splitter island and median configurations could be compared. Figure 69 shows the pace comparisons for one-lane and two-lane sites (in each direction), as well as all sites. Overall, the crossing paces were very similar, ranging only from 4.4 to 5.0 ft/s (1.3 to 1.5 m/s). The type of traffic control that is present at a crossing does not appear to produce any practical differences in the walking pace of crossing pedestrians. Analysis of Findings This study was undertaken to develop a better picture of pedestrian operations at roundabouts and to gain insight on the interactions between this mode and motor vehicles. Data were collected from 10 sites located at seven roundabouts to answer the questions that were posed in the introduction section of this report. Provided below are answers to these questions on the basis of the analysis conducted in this effort. What is the yielding behavior of motorists when they encounter a pedestrian who is crossing or waiting to cross? On average across all sites, approximately 30% of the motorists did not yield to pedestrians who were crossing or waiting to cross. In all but one case, the pedestrians were wait- ing to cross, so there was no imminent risk. There was a difference in this behavior with respect to the entry side ver- sus exit side of the leg being crossed. Motorists did not yield to pedestrians on the entry side 23% of the time, compared to 38% of the time on the exit side. There was also a difference in the yielding behavior depending on where the crossing was initiated. If the pedestrian started crossing from the entry side of the leg, 27% of the motorists did not yield. However, if the crossing began on the exit side, the percentage of motorists not yielding increased to 34%. Yield behavior also varied with the number of lanes at the crosswalk. The lack of yielding on two-lane sites (43%) was substantially worse than on one-lane sites (17%). The results for two-lane sites showed that 54% of the motorists did not 87 Type of Traffic Control Yielded Did Not Yield Roundabout (yield control) 68% 32% Signal 85% 15% Stop Sign 96% 4% None 52% 48% Type of Traffic Control Average Waiting Time (s) Roundabout (yield control) 2.1 Signal 10.7 Stop Sign 0.3 None 3.0 Table 63. Motorist behaviors by traffic control type. Table 64. Waiting time by traffic control type.

yield on the exit side compared to 33% not yielding on the entry side. For one-lane sites, the exit and entry non-yield percentages were 22% and 13%, respectively. How do pedestrians respond to vehicles when preparing to cross or crossing the street? Just over half (58%) of the pedestrian crossings that occurred in the presence of vehicles were considered to be normal crossings, implying no unusual behaviors on the part of the pedestrian. The non-normal behavior that was observed most often was hesitation, which could occur on the entry-side or exit-side curb before start- ing to cross or on the splitter island when preparing to com- plete the crossing. Most often, the hesitation occurred while the pedestrian made visual or other contact with the driver. Once this communication was made and the vehicle began slowing, the pedestrian would then proceed with the crossing. Across all sites, the average number of crossing events in which a pedestrian hesitated on the curb or after starting was 28%. There was a difference in the percentage of hesitation crossings associated with the number of lanes at the cross- walk. When averaged across both sides of the crossing and from both starting positions, single-lane sites resulted in hes- itation crossings 24% of the time, while two-lane sites pro- duced such crossings 33% of the time. When averaged across both sides of the crossing, approxi- mately 22% of the pedestrians that started crossing from the entry side hesitated, compared to 31% of those that started from the exit side. Irrespective of the starting location, the majority of hesitations occurred at the curb prior to initiating the crossing, rather than halfway through the crossing while on the splitter island. For example, pedestrians hesitated 33% of the time on the entry-side curb when starting from the entry side and only 10% of the time at the splitter island. Similarly, pedestrians starting from the exit-side curb hesitated 39% of the time at the exit-side curb and 23% at the splitter island. The other crossing behavior that was observed quite often was running. The running behavior in this case was not a “sudden” behavior on the part of the pedestrian to avoid a conflict (see next question). It appears to have been simply a choice made by the pedestrian to speed up the crossing and was most often done as a courtesy to the yielding motorist. The average number of pedestrian crossings across all sites in which running was an observed behavior was 14%. For one- lane sites, 17% of the crossings involved a running pedestrian, compared to 8% of the crossings at two-lane sites. Running was most prevalent for crossings that began on the entry side (18%) as opposed to the exit side (7%). The running behavior was most often observed on the second half of the crossing. For example, of the pedestrians starting on the entry side, 7% ran across the entry side, and 29% ran across the exit side. The same was true for exit-side starts, but less pronounced; 3% ran across the exit side, while 10% ran across the entry side. Did the behaviors of motorists and pedestrians create unsafe situations? The measure of safety that is most often applied to the roadway environment is a crash. There were no crashes in the pedestrian-vehicle interactions observed in this study. As a surrogate, conflicts between the two modes that required one or both parties to suddenly change course and/or speed to avoid a crash were studied. Out of 769 pedestrian crossings across the 10 sites, there were only four conflicts. The resulting conflict rate was 2.3 conflicts/1000 opportunities.An opportunity was defined as any time a pedestrian was either waiting to cross or crossing the leg and a motor vehicle was in the vicinity of the pedestrian. This rate was slightly greater for one-lane sites (2.8) than for two-lane sites (2.0). What are the geometric or operational characteristics that tend to cause problems for pedestrians or tend to result in safer and more accessible designs? Based on the answers 88 4.6 4.5 4.5 5.0 4.7 5.0 4.8 4.9 4.7 4.5 4.4 4.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 1-Lane 2-Lane All Pa ce (f t/s ) Signal Stop Sign None Roundabout Figure 69. Pedestrian crossing pace by traffic control and number of lanes.

to the previous questions, the two design elements that cor- relate to differences in behaviors are the number of lanes and the directional side of the site (entry lanes versus exit lanes). More lanes resulted in a higher number of vehicles not yielding to crossing/waiting pedestrians (43% on two-lane sites versus 17% on one-lane sites). Pedestrian behaviors also differed at these two types of sites, but the results were not consistent. Crossings in which the pedestrian hesitated were 33% on two-lane sites compared to 24% on one-lane sites. Crossings in which the pedestrian ran were 8% for two-lane sites compared to 17% for one-lane sites. The behaviors of both motorists and pedestrians also differed depending on the directional side of the site: entry versus exit. Motorists were less likely to yield to pedestrians on the exit side (38% of the time, based on an average of entry-side and exit-side starts from Figures 62 and 63) com- pared to the entry side (23% of the time, based on a similar average). Pedestrian behaviors on the two sides differed according to the starting position of the crossing. Pedestrians were more likely to hesitate when starting from the exit side (31% of the crossings) than when starting from the entry side (22% of the time). Pedestrians were also more likely to run across the exit side when it was the second stage of the cross- ing than the entry side when it was the second stage (29% versus 10%, respectively). Further review of the geometric and operational character- istics associated with individual sites was undertaken to deter- mine if specific elements—such as lane widths, splitter island designs, and other factors—were associated with the behaviors observed. These reviews did not produce any additional insights into the observed differences between locations. How do the behaviors of pedestrians and motorists at roundabouts compare to the behaviors of pedestrians and motorists at other types of intersections? The pedestrian and motorist behavior results for the roundabout sites in this study were compared to the results for two-way- stop–controlled, all-way-stop–controlled, and signalized intersections obtained from the FHWA study titled “Safety Index for Assessing Pedestrian and Bicyclist Safety at Inter- sections” (3). The FHWA results showed 70% of the cross- ings at intersections with no traffic control were considered normal (i.e., no running, hesitation, or stopping). The next highest percentage (85%) of normal crossings occurred at roundabouts, which are yield-controlled intersections. The crossings at signalized and stop-controlled intersections exhibited the highest normal crossing percentages at 90% and 100%, respectively. Running was the single pedestrian behavior that was substantially more common at round- abouts than at the other types of controlled intersections. As noted previously, however, the running behavior observed at roundabouts appears to have been most often done as a courtesy on the part of the pedestrian. Pedestrian waiting times and crossing pace (walking speed) were also compared among the four levels of traffic control. The traffic control type that resulted in the longest pedestrian waiting time of 10.7 s was signalization, which is expected because most pedestrians will comply with the signal. Crossings with no traffic control produced an aver- age waiting time of 3.0 s, while crossings at stop-controlled locations resulted in virtually no waiting time (0.3 s). For roundabouts, the average waiting time was 2.1 s, which is between the time for the stop-controlled locations where pedestrians are confident that the vehicle will stop and the no-control locations where there may be an increase in uncertainty about the intent of the vehicle. With respect to pace, there were no practical differences on the basis of traffic control type; all locations ranged between 4.4 and 5.0 ft/s (1.3 to 1.5 m/s). With respect to motorist behavior and the type of traffic control, almost half (48%) of the vehicles on uncontrolled legs did not yield to pedestrians. The stop-controlled and signalized intersections produced non-yielding vehicle per- centages of 4% and 15%, respectively. Roundabout crossings, which are subject to yield control, were in between these extremes with 32% of the motorists not yielding. Bicyclist Analysis Data were collected for 690 bicyclist events at 19 legs dis- tributed among seven roundabouts. Only two of these sites had two lanes, so the analysis did not include any one- lane/two-lane site comparisons. Bicyclists were observed as they entered, exited, or circulated in the roundabout or crossed at the crosswalk. Data were collected for all bicyclists who entered the study area shown in dashed lines in Figure 70. The study area included the part of the circulating lanes near the entry/exit of the leg and the leg as far as the end of the splitter island. 89 Figure 70. Bicyclist study area.

The analysis of bicyclist events covers the following topics: • Bicyclist position • Bicyclist behaviors • Motorist behaviors • Bicycle-motor vehicle conflicts • Other bicyclist behaviors Refer to Appendix L for images to help describe some of the behaviors observed and described in the subsequent sections. Bicyclist Position Bicyclist position refers to the location of the bicyclist’s path as the bicyclist enters, exits, or traverses the roundabout. This study classified bicyclist position as on the sidewalk; on the shoulder, bike lane, or edge of travel lane; or possessing the vehicle lane. There were not enough sites with and with- out bike lanes and paved shoulders to allow for comparison between these configurations. Bicyclist positions classified as edge of lane in the following analyses include positions on bike lanes and paved shoulders. Possessing the lane was defined as a bicyclist riding close to the center of the lane such that motorists would not attempt to pass. Bicyclists were most commonly found to ride on the edge of the lane. Just over half (54%) rode in this position (see Table 65). The remainder was split between possessing the lane and riding on the sidewalk, with the latter being the least common position. Position by Event Type Event type refers to the type of movement that the bicyclist made at the roundabout—entering, exiting, or circulating the roundabout or crossing at the crosswalk. Figure 71 shows that the majority of bicyclists entering or exiting were positioned on the edge of lane, whereas circulating bicyclists more often possessed the lane. A higher percentage of bicyclists entering the roundabout were positioned at the edge of the lane when compared to bicyclists exiting the roundabout (73% entering versus 61% exiting). The most likely explanation for this difference is that entering bicyclists more often had to com- pete with vehicles for roadway space. Because of queuing at the entry, many entering bicyclists were observed to enter simultaneously with a vehicle or follow very closely behind a vehicle as they entered. Vehicle Presence and Position of Bicyclists Bicyclists that ride in the roadway, as opposed to the side- walk, frequently interact with vehicles and must make deci- sions about where to position themselves on the basis of factors such as their own comfort level in traffic, the amount of space available, and speed and volume of motor vehicle traffic. For each bicyclist that entered, exited, or circulated in the roundabout, data were collected on the presence and proximity of motor vehicles to the bicyclist. Motor vehicle presence was classified as one of the following: • Leading bicyclist within two car lengths • Trailing bicyclist within two car lengths • Passing bicyclist • Queued in front of bicyclist (entering bicyclists only) • Queued behind bicyclist (entering bicyclists only) • No vehicle leading, trailing, passing, or queued near bicyclist Out of 450 events in which bicyclists were entering, exit- ing, or circulating, only 6 events were observed to involve queued vehicle presence. Five bicyclists had vehicles queued in front of them, and one bicyclist had a vehicle queued behind. In all but one case, the bicyclist was positioned on the shoulder. The other categories for vehicle presence occurred more frequently and are shown in Table 66. When there were no vehicles in the vicinity, 42% of bicyclists possessed the vehicle lane. When a vehicle was leading the bicyclist (within two car lengths), the percentage decreased to 35%. When a vehicle was trailing the bicyclist, even fewer bicyclists (23%) possessed the lane. The 12 percentage point difference from leading vehicles to trailing vehicles may indicate that bicy- clists were not as comfortable possessing the lane when a vehicle was approaching them from the rear. The category denoted as “more than one type” indicates that the bicyclist had at least two vehicles in proximity. Most often, the bicyclist had one vehicle leading and another one trailing. As would be expected, the position percentages asso- ciated with this occurrence falls in between the values for “leading” and “trailing.” In the case of passing vehicles, all but 1 event out of the 37 observed had the bicyclist positioned on the shoulder. In that one event, the bicyclist was on the shoul- der for the part of the observation when the passing occurred. Bicyclist Behaviors Bicyclist behavior was captured for the two event types where a bicyclist had to accept and enter a gap—entering the 90 Position Category Bicyclists Edge of Lane/Shoulder/Bike Lane 54% Possessing Lane 28% Sidewalk 18% Total 100% Table 65. Distribution of bicyclists by position.

circulating lane of the roundabout and crossing at the cross- walk. Behaviors were categorized as one of the following: • Normal (passes through without stopping because there is no vehicle in the vicinity or vehicle yields) • Hesitates or waits before starting because of approaching vehicle • Hesitates after starting because of approaching vehicle • Retreats after starting because of approaching vehicle • Swerves to avoid approaching vehicle For each entering or crossing event, the data coding process also included a safety code. Each gap was subjectively deemed to be safe or unsafe, depending on the proximity and speed of nearby vehicles. For example, if a bicyclist approached the crosswalk at a fast speed and crossed without looking at or yielding to oncoming traffic, the event was coded as unsafe. An “unsafe” code could be given even if the interaction did not result in a conflict or crash. Bicyclist Behaviors on Entering the Roundabout Out of 238 bicyclists that entered a roundabout, 70% pro- ceeded into the circulating lane without stopping, 11% waited before entering the circulating lane, and 19% entered on the sidewalk. There was one case of a bicyclist swerving to avoid a vehicle. The swerving case was a conflict caused by the bicy- clist entering from the exit side and is described in more detail in the conflicts section. The difference in percentages between proceeding- without-stopping and waiting-before-entering is most likely correlated with the amount of vehicular traffic in the circu- lating lane, but these data were not available to confirm this hypothesis. From subjective observations, sites with heavier traffic caused bicyclists to yield more frequently to circulat- ing vehicles. Only two cases of entering bicyclists were deemed unsafe. One was the swerving conflict, and the other was a case where the circulating vehicle did not yield but passed on the bicyclist’s right. The safe entrance on the road- way or the sidewalk of almost all bicyclists indicates that 91 73 61 15 16 83 12 23 17 0 0 10 20 30 40 50 60 70 80 90 Entering roundabout Exiting roundabout Circulating Maneuver Pe rc en ta ge Edge of Lane/ Shoulder/Bike Lane Possessing Lane Sidewalk Edge of Lane/ Shoulder/Bike Lane Possessing Lane Total Motor Vehicle Presence Number % Number % Number % None 155 58% 113 42% 268 100% Leading 31 65% 17 35% 48 100% Trailing 48 77% 14 23% 62 100% More than one type 21 66% 11 34% 32 100% Passing 36 97% 1 3% 37 100% Total 291 65% 156 35% 447 100% Figure 71. Position of bicyclist on the basis of event type or maneuver at the roundabout. Table 66. Distribution of bicyclist positions by vehicle presence.

there are not significant safety problems with bicyclists entering roundabouts. Bicyclist Behaviors at Crosswalks There were 81 events where a bicyclist crossed at the cross- walk. Eighty-five percent of these crossings were entirely on the painted crosswalk. Only a few bicyclists rode on the exit- side crosswalk and moved off it for the entry side (7%) or rode on the entry-side crosswalk and moved off it for the exit side (7%). The observed behaviors included normal (proceeding through without stopping), waiting or hesitating before start- ing to cross, and hesitating after starting. Table 67 shows that normal behavior was the most common behavior, although there was a difference on the basis of which side was being crossed. More bicyclists waited or hesitated before crossing the exit side (27%) than did at the entry side (20%). Only three bicyclists hesitated after starting—two on the exit side and one on the entry side. Overall, there was more hesitation at the exit side. Not only do vehicles travel faster when exiting the roundabout than when entering, but also bicyclists are uncertain whether a vehicle approaching in the circulating lanes will exit. These conditions may explain the higher bicy- clist hesitation rates on the exit side. Motorist Behaviors Motorist behavior was recorded when a bicyclist entered the roundabout or crossed at the crosswalk. The motorist behavior was categorized as one of the following: • Slows or stops for waiting bicyclist: Motorist yields to bicyclist waiting on curb or splitter island. • Slows or stops for bicyclist in transit: Motorist yields to bicyclist in motion. • Already stopped for other reason: Motorist is queued at entry or already yielding to other bicyclist or pedestrian and remains stopped to yield to bicyclist. • Swerves: Motorist changes direction to avoid bicyclist. • Does not yield: Motorist does not yield to bicyclist in tran- sit or waiting on curb or splitter. • No vehicle present: No vehicle was in the vicinity at the time of the event. Motorist Behavior When Bicyclists Entered the Roundabout There were 238 events involving a bicyclist entering a roundabout. However, 45 of these bicyclists entered on the sidewalk; thus, there were only 193 opportunities for interac- tion with vehicles. Of these 193 opportunities, 188 (97%) occurred with no vehicle present. That is to say, when the bicyclist accepted the gap and entered the roundabout, there was no vehicle in the circulating lanes to immediately follow the bicycle and “close” the gap. For the remaining five events, a vehicle was present and the following behaviors occurred: • One motorist slowed for the waiting bicyclist. • One motorist stopped for the bicyclist entering the round- about. • One motorist swerved to avoid the bicyclist. • One motorist passed the entering bicyclist on the left. • One motorist passed the entering bicyclist on the right. The case of the swerving motorist resulted in a conflict and is described in more detail in the section on bicycle-motor vehicle conflicts. The two cases where the motorist passed the entering bicyclist did not result in conflicts but were cases where the bicyclist did not select an appropriate gap. Overall, there were few opportunities for motorist–bicyclist interaction on entry. For the most part, bicyclists chose to enter the roundabout when there were no vehicles immedi- ately approaching. The selected gaps were large enough that there was no need for approaching vehicles to yield. Motorist Behavior at Crosswalks The majority of the 81 crossing events also occurred with- out a vehicle present. Table 68 shows the distribution of motorist behaviors at the crosswalk, specific to the entry and exit sides of the leg. It was more common that the bicyclist would encounter no vehicles on the exit side (77%) than on the entry side (62%). For the other cases when a vehicle was present, the motorist was always observed to yield to the bicy- clist, whether waiting on the curb or in motion on the cross- walk. There were 17 cases where the motorist was already stopped and remained stopped for the bicyclist to pass; all but one of these cases occurred on the entry side. 92 Entry Side Exit Side Bicyclist Behavior Number % Number % Normal 64 79% 57 70% Waits/hesitates before starting 16 20% 22 27% Hesitates after starting 1 1% 2 2% Total 81 100% 81 100% Entry Side Exit Side Motorist Behavior Number % Number % No vehicle present 50 62% 62 77% Slows or stops for waiting bicyclist 4 5% 9 11% Slows or stops for bicyclist in transit 11 14% 9 11% Already stopped 16 20% 1 1% Total 81 100% 81 100% Table 67. Bicyclist behavior at crosswalks by side. Table 68. Motorist behavior at crosswalks by side.

Unlike the pedestrian study, behaviors of the motorist were only recorded for interactions when there was an accepted gap by the bicyclist. For this reason, Table 68 does not provide information on the percentage of motorists not yielding to bicyclists. For this information, the results found from the pedestrian-motor vehicle interactions, which included a sig- nificantly larger number of crossing events at a wider range of geometric conditions, are deemed to suffice. Bicycle-Motor Vehicle Conflicts One of the surrogate measures of safety for bicyclists is a conflict with a motor vehicle. In this study, as well as many others, a conflict was defined as an interaction between a bicyclist and motorist in which one of the parties had to sud- denly change course and/or speed to avoid a crash. During the 690 bicyclist events, only four conflicts were observed (0.6%). Two of these conflicts occurred while a bicyclist was crossing at the crosswalk; one occurred while a bicyclist was circulating in the roundabout; and one occurred while a bicyclist was entering the roundabout. Each conflict is fur- ther described below: • Conflict 1 – OR01-N1 (Bend,Oregon) at crosswalk: Bicyclist begins crossing crosswalk from exit side and continues across entry side without slowing. Vehicle in queue to enter round- about is in forward motion very close to crosswalk and has to slam on brakes to avoid hitting the bicyclist (Figure 72). • Conflict 2 – OR01-S1 (Bend, Oregon) at crosswalk: Bicy- clist begins to cross crosswalk from exit side with little slowing. Vehicle exiting roundabout brakes suddenly to avoid crash (Figure 73). • Conflict 3 – WA03-E3 (Bainbridge Island, Washington) circulating: Bicyclist circulating roundabout on the outside of the circulatory roadway attempts to continue circulating past the leg. A vehicle to the left of the bicyclist attempts to exit the roundabout. Both parties swerve and brake suddenly (Figure 74). • Conflict 4 – WA03-S2 (Bainbridge Island, Washington) entering: Bicyclist approaches roundabout going the wrong way on the exit lane. He begins to enter the round- about from the exit lane and swerves to avoid an exiting vehicle. Exiting vehicle also swerves to avoid crash. Bicyclist appears to be young (Figure 75). Other Bicyclist Behaviors Events involving bicyclist-pedestrian interaction or wrong- way riding are less common behaviors but were encountered in the course of the data collection. There were three cases 93 Figure 72. Bicycle Conflict 1 at OR01-N (Bend, OR). Figure 73. Bicycle Conflict 2 at OR01-S (Bend, OR). Figure 74. Bicycle Conflict 3 at WA03-E (Bainbridge Island, WA).

where a bicyclist interacted with a pedestrian. Each interac- tion occurred when the bicyclist was on the roadway and yielded to a pedestrian on the crosswalk. Nothing significant resulted from these interactions. Wrong-way riding was defined as a bicyclist riding on the paved roadway contrary to traffic flow. If the bicyclist was on the sidewalk, the event was not coded as wrong-way, even if the motion of the bicyclist was contrary to the flow of the roundabout (i.e., bicyclist entering the roundabout on the exit-side sidewalk). Seven cases of wrong-way riding were recorded. All seven involved a bicyclist entering the round- about from the exit lane. Of the seven cases, one resulted in a conflict (see Conflict 4 in the previous section). Another point to consider is that five of the wrong-way cases occurred at one roundabout where the camera view did not allow for observation of the bicyclists’ final positions once they passed through the study area and appeared to enter the roundabout. It is possible that these bicyclists proceeded to get on the side- walk to enter the roundabout. Overall, wrong-way riding was a rare event and only once resulted in a safety problem. Analysis of Findings How do bicyclists and motorists interact on the entry lanes, exit lanes, and circulating lanes of the roundabout? Where does the bicyclist tend to be positioned? The major- ity (73%) of bicyclists approaching a roundabout positioned themselves at the edge of the travel lane or on a bike lane or paved shoulder if available. Only 15% of the approaching bicyclists possessed the lane. The remaining 12% used the sidewalk. For exiting bicyclists, 23% used the sidewalk, while 16% possessed the lane. Those bicyclists in the circulating lane tended to take the lane (83%) rather than ride on the edge of the circle. With respect to other types of interactions between the two modes, no problems were observed. Bicyclists that entered the circulating roadway from the entry lane almost always selected gaps in which no vehicle was approaching on the cir- culating roadway. Bicyclists and motorists traversed the cir- culating lane with very little interaction. Are there conflicts or avoidance maneuvers due to the interactions of motorists and bicyclists? As a surrogate measure for safety, a conflict was defined in this study as an interaction between a bicyclist and motorist in which one of the parties had to suddenly change course and/or speed to avoid a crash. Only four conflicts were observed during the 690 bicyclist events, or 0.6%. Two of these conflicts occurred while a bicyclist was crossing at the crosswalk; one occurred while a bicyclist was circulating in the roundabout; and one occurred while a bicyclist was entering the roundabout. The one involving the bicyclist circulating the roundabout involved an exiting vehicle that almost struck a bicyclist who was traversing the circulating lane on the outside of the lane, which is one of the most vulnerable positions to be in as a bicyclist. Do bicyclists exhibit any behaviors that raise safety concerns? The one behavior observed that did raise some concern was wrong-way riding, particularly when entering the roundabout from the exit lane of the leg. This scenario did result in one of the four conflicts observed. While the number of observed wrong-way events was small (seven), this type of event can produce crashes as a result of expectancy violations. Other Design Findings This portion of the design analysis contains the evaluation of the safety and capacity modeling efforts and their respec- tive observations as they relate to specific design elements of the roundabout. The general approach to this analysis is twofold: • The sensitivity of various geometric parameters was tested in the development of prediction models for safety and capacity. These tests are documented in the model devel- opment for safety and capacity, respectively. • Sites with abnormal safety and/or capacity performance were examined using expert judgment to identify geo- metric elements that could be contributing factors to the abnormal safety and/or capacity performance. This analysis was not conducted to the same level of statistical rigor as the analysis associated with model development, but it contributes to an anecdotal understanding of the geometric elements that appear to have an influence on safety and capacity. This anecdotal understanding is presented here. 94 Figure 75. Bicycle Conflict 4 at WA03-S (Bainbridge Island, WA).

The safety and geometric data identify several trends related to the early roundabout experience in the United States. Overall, the crash experience has been positive (showed an overall reduction in crash frequency); however, there were several intersections where this was not the case. In some cases, either there was no change in crash frequency or there was actually an increase; although, in almost all cases, the crash counts are too small for the increase to be statisti- cally significant. Many of the roundabouts in the dataset were constructed before the publication of the FHWA Roundabout Guide (1). In general, the evaluation presented here focuses on crash frequency as the primary measure for flagging sites with high or low crash experience. However, this evaluation also com- pares sites sorted by crash frequency to the sites sorted by crash rate. This analysis confirmed the conclusion that mul- tilane roundabouts represent a greater risk for crashes than single-lane roundabouts. Table 69 presents an analysis of the relationship of overall roundabout geometry to crash frequency; Table 70 presents a similar analysis of overall roundabout geometry to crash rates. The analysis compares the roundabout geometry across a range of crash frequencies and crash rates: the full dataset, the 10 lowest, the 30 lowest, the 30 highest, and the 10 high- est. From this analysis, the following conclusions related to the number of lanes in the roundabout were reached: • The site sorting from best to worst generally stayed the same whether sorting by crash frequency or crash rate. In general, the use of crash rates for comparisons is not pre- ferred because of the known non-linear relationship between traffic volume and crash frequency. Therefore, the remaining analysis has been conducted using crash frequency. • Eight of the ten sites with the lowest crash frequencies were single-lane roundabouts. • Twenty-six of the thirty sites with the lowest crash fre- quencies were single-lane roundabouts. • Two of the ten sites with the highest crash frequencies were single-lane roundabouts. • Nine of the thirty sites with the highest crash frequencies were single-lane roundabouts. • Crash frequency increases as the inscribed circle diameter increases. • Crash frequency increases as the amount of vehicles enter- ing the roundabout increases. • Crash frequency increases slightly as the number of legs to the roundabout increases. Single-Lane Roundabout Evaluation Review of the plans for these sites clearly indicates that a designer can get by with making more design errors with single-lane roundabouts. Not all of the single-lane round- about designs were “perfect,” but the designs contained enough geometric changes to indicate the change in inter- section form to the drivers, to slow the drivers, and therefore to increase the safety of the intersection. Many of the single- lane roundabouts were also clustered in several states, so driver familiarity may have played a role. For the single-lane roundabouts that did result in a higher crash frequency, there was little deflection or speed reduction on the entry paths to the roundabout. For example, the single-lane roundabout with the highest crash frequency in the dataset is shown in Figure 76. The speeds estimated from fastest paths (using the current FHWA Roundabout Guide methodology) exceed the thresholds recommended in the FHWA Roundabout Guide. There is little or no deflection on the approaches to the 95 Crash Frequency (crashes/yr) Crash Rate (crashes/MEV) Average Number of Lanes in Group Average Inscribed Circle Diameter Average Daily Traffic (veh/day) Average Number of Legs in Group Total Dataset 4.95 0.75 1.39 133 ft (41 m) 16,606 3.89 First Ten 0.01 0.00 1.20 97 ft (30 m) 8,604 3.60 First Thirty 0.44 0.16 1.13 114 ft (35 m) 9,585 3.67 Bottom Thirty 12.13 1.59 1.83 162 ft (49 m) 23,935 4.13 Bottom Ten 22.89 2.64 2.20 215 ft (66 m) 28,300 4.30 Legend: MEV = million entering vehicles; veh = vehicles Table 69. Relationship between crashes and geometry, sorted on crash frequency.

roundabouts, and the research team believes that this is the primary cause for the high crash frequency. Multilane Roundabout Evaluation Approximately one-third of the sites in the safety database are multilane roundabouts. However, 8 of the 10 sites with the highest crash frequency were multilane roundabouts. There- fore, it is apparent that at least some multilane roundabouts have abnormally high crash experiences that warrant further investigation. Review of these sites led the research team to believe that most were not designed using the natural vehicle path con- cept. This scenario is entirely likely because the majority of these sites were designed and constructed before the publica- tion of the FHWA Roundabout Guide (1), which was the first document to publish this concept. The natural vehicle path concept, shown in Figure 77, was refined in later guidelines such as the Kansas Department of Transportation’s Kansas Roundabout Guide (37). Lane widths also appear to have an effect on safety. For example, one roundabout appeared to be designed to 96 Crash Frequency (crashes/yr) Crash Rate (crashes/MEV) Average Number of Lanes in Group Average Inscribed Circle Diameter Average Daily Traffic (veh/day) Average Number of Legs in Group Total Dataset 4.95 0.75 1.39 133 ft (41 m) 16,606 3.89 First Ten 0.02 0.00 1.20 95 ft (29 m) 9,295 3.70 First Thirty 0.59 0.10 1.23 123 ft (37 m) 14,961 3.73 Bottom Thirty 11.75 1.69 1.70 165 ft (50 m) 20,186 4.07 Bottom Ten 18.51 3.03 1.90 150 ft (46 m) 16,734 4.20 Legend: MEV = million entering vehicles; veh = vehicles Table 70. Relationship between crashes and geometry, sorted on crash rates. Figure 76. Example of a single-lane roundabout with poor deflection characteristics.

accommodate the natural vehicle path yet still exhibited a higher than anticipated crash frequency. This site exhibits narrower lane widths than other sites reviewed and than recommended by the FHWA Roundabout Guide. The inter- section geometry for the site is shown in Figure 78. The FHWA and other guidebooks recommend lane widths in the range of 13 to 16 ft (4.0 to 4.9 m) at the entries and exits to the roundabouts, and circulatory roadway widths of 16 to 20 ft (4.9 to 6.1 m) for single-lane roundabouts and 26 to 30 ft (7.9 to 9.1 m) for multilane roundabouts. For this particu- lar site, the entry lane widths of 10 to 11 ft (3.0 to 3.4 m) were maintained on the approaches to the yield line. The circulatory roadway width was also approximately 22 ft (6.7 m) for two circulatory lanes. The natural vehicle paths and the FHWA speed paths for this site are illustrated in Figures 79 and 80, respectively. Note that the speed paths stay in their respective lanes on the legs; a speed path using the UK method (curb to curb without con- sideration of lanes) would result in slightly higher predicted speeds. Based on this analysis, the natural vehicle paths and the speed paths through the roundabout did not reveal major problems that would suggest a crash problem based on con- ventional design wisdom. Therefore, the entry lane widths are potentially the prime design feature contributing to the observed crash experience. Summary of Design Findings Overall, the data suggest that roundabouts can improve the safety performance at intersections. However, the research team believes that the performance of the roundabouts discussed here could be improved by relying on the guidance put forth in the FHWA guide and state supplements. Entry Width The conventional wisdom in roundabout design, as in intersection design in general, is that as the width of an entry increases, the capacity of the entry increases, while the safety of the entry decreases. In most countries, the safety and oper- ational effect of entry width is related primarily to the num- ber of lanes provided by the entry in question, with wide entries typically having more lanes than narrow entries. Using linear regression, Maycock and Hall established an empirical relationship in the UK between entry width and entering- circulating crashes (4); this model uses entry width as a direct input, rather than the number of lanes on the entry. Likewise, Kimber determined an empirical relationship between entry width and capacity, also using entry width as a direct input rather than the number of lanes (19). Most other known safety and capacity models are based on the number of lanes rather than the actual entry width. Analysis of U.S. data suggests that this principle generally holds true for U.S. conditions. Entry width was found to have a direct relationship with entering-circulating crashes and is part of the candidate model for estimating such crashes (see safety analysis for more detail). In addition, entry width in the aggregate sense—number of lanes—appears to have a direct relationship on capacity, as evidenced by the development of single-lane and multilane capacity models. However, the extension of the principle beyond number of entry lanes to the actual width of the entry does not appear to 97 Figure 77. Design technique to minimize entry path overlap.

have as strong a relationship in the United States. As demon- strated in the operational analysis work, while there appears to be a relationship between the additional width added as part of entry flare (see Chapter 4), there appears to be no significant effect on capacity for variations of entry width within a single-lane entry. This evidence suggests that, while the overall relationship between capacity and entry width appears to hold true in terms of the aggregate number of lanes on the approach, changes in entry width within a single-lane entry has a much lower-order effect on capacity. The number of sites with multilane entries is too limited to make similar conclusions about the influence of small changes in entry width on the capacity of multilane entries. Angle Between Legs The angle between legs of a roundabout appears to have a direct influence on entering-circulating crashes and is part of the candidate model for estimating such crashes (see Chapter 3 for more detail). As the angle to the next leg decreases, the number of entering-circulating crashes increases. This result is consistent with the experience in the UK, whose entering-circulating crash model also includes this geometric parameter. From a design perspective, this evidence suggests that roundabouts with more than four legs or with skewed approaches tend to have more entering-circulating crashes. In many of these cases the higher speeds enabled by these designs may be contributing to these higher crash frequen- cies. An example of this was discussed earlier (“Single-Lane Roundabout Evaluation”). Splitter Island Width and Effect of Exiting Vehicles The width of the splitter island and its effect on safety and capacity was investigated, because other countries found a relationship. Wider splitter islands were speculated to result in improved entry capacity due to drivers being more able to differentiate between circulating and exiting vehicles. The analysis of U.S. data, however, did not find a significant relationship between the capacity of the entry and the width of the splitter island, nor with the percentage 98 Figure 78. Example of a multilane roundabout with narrow entry and circulatory roadway widths.

of exiting vehicles. As a result, this factor has not been included in the recommended capacity models. As noted in the operational analysis, U.S. drivers appear to be navigat- ing roundabouts very cautiously at the present time. The research team believes that as drivers become more com- fortable and efficient, the effect of the width of the splitter island and/or percentage of exiting vehicles may become more noticeable and should be studied in the future. Intersection Sight Distance The FHWA Roundabout Guide (1) presents an intersec- tion sight distance methodology based on critical headway (critical gap) values; this methodology is consistent with the AASHTO Policy (33). The FHWA Roundabout Guide rec- ommends a critical headway value of 6.5 s, which was based on an adaptation of the AASHTO Policy values for yield- controlled intersections. As noted in the operational analysis, drivers exhibit a range of critical headway values based on the type of round- about (single lane versus multilane). Table 71 summarizes the critical headway analysis conducted for this project for single-lane and multilane approaches for those observations where a driver accepts a gap after rejecting a gap (Method 2, as described in Chapter 4). Note that all of these observations have been made with the waiting driver positioned at the yield line (entrance line). It is reasonable for the value of critical headway that is used for intersection sight distance calculations to be more conservative than that used for capacity estimation; this philosophy is the same that was employed in revising the intersection sight distance methodology in the 2001 AASHTO Policy, as documented in NCHRP Report 383 (38). Based on these findings, the critical headway estimate of 6.5 s in the FHWA Roundabout Guide appears to be some- what conservative for design purposes for both single-lane and multilane entries. A lower value of 6.2 s is recom- mended for design purposes, which represents approxi- mately one standard deviation above the mean observed critical headway. For comparison purposes, the critical headway value recommended by AASHTO for minor-street right turns at a stop-controlled intersection is 7.5 s, which is 99 Figure 79. Natural vehicle paths in a multilane roundabout with narrow entry and circulatory roadway widths.

greater than the 6.2 s value measured by Kyte et al. (39) as part of the procedure to estimate capacity for two-way- stop–controlled intersections (38). Although this study does not recommend a major change in the critical headway estimate, this study has previously pre- sented changes to the methodology for estimating vehicle speeds through a roundabout. These changes influence inter- section sight distance, as they dictate the distance over which conflicting vehicles will travel during the elapsed headway time. While the proposed estimation method for circulating vehicles remains unchanged from current practice, the speed estimates for vehicles approaching from the immediate upstream entry are likely to be lower with this revised methodology. In particular, the actual speed estimates for V1, which forms half of the estimated speed for the intersection sight distance methodology, may be substantially lower than previously estimated using the methodology from the FHWA Roundabout Guide. This lower speed may result in shorter sight triangles to the left (toward the immediate upstream entry); these sight distances are often the most challenging to provide in practice. The issue of the balance between providing adequate sight distance and providing too much sight distance has not been addressed in this study. As a result, the recommendations from this study should be viewed as interim until a more comprehensive study of sight distance requirements at roundabouts can be completed. Multilane Entry and Exit Design The concept of natural path overlap at multilane round- abouts was first introduced into a design guide with the 100 Figure 80. FHWA speed paths in a multilane roundabout with narrow entry and circulatory roadway widths. Number of Observations Mean Standard Deviation Single Lane 3,322 5.0 1.2 Multilane (both lanes) 3,350 4.5 1.6 Table 71. Critical headway summaries for intersection sight distance.

FHWA Roundabout Guide in 2000 (1) and continued in a number of state guides (e.g., Kansas [37]). The general concept is that for optimal safety and opera- tional performance of roundabout entries and exits, the entry lanes at the entrance line to the roundabout should align with their receiving lanes within the circulatory road- way; likewise, the exit lanes should align with their feeding lanes within the circulatory roadway. The most common case where entry lanes do not line up with their receiving lanes in the circulatory roadway is one where the outermost entry lane lines up with the inside portion of the circulatory road- way. The inner entry lane then lines up with the central island. In these cases, the natural tendency (the natural path) for vehicles in the outer lane is to pass close to the central island, thus effectively impeding vehicles in the adjacent (inner) lane. From a safety perspective, this situation could result in sideswipe crashes; from an operational perspective, this situation could result in poor lane utilization and thus reduced effective capacity. On the exit side, path overlap can occur where the exit path radius is small relative to the path radius of the circu- latory roadway. As vehicles traverse the circulatory roadway in the inside lane, their natural tendency is to proceed along a path of similar radius (and thus similar speed) through the exit. A small exit path radius can cause exiting vehicles in the inner lane to overlap with exiting vehicles in the outer lane. As noted in the safety analysis, the inscribed circle diameter and the width of the circulatory roadway appear to have a direct relationship with exiting-circulating crashes. As both parameters increase in value, the number of exiting-circulating crashes increases. This result is to be expected, as multilane roundabouts are the most likely to experience these types of crashes. A general analysis of the safety data for the multilane roundabouts within the database suggests that, of the sites with the highest crash frequencies and/or crash rates, the majority exhibit some form of path overlap on entries and/or exits. Likewise, sites without noticeable path overlap tend to have lower frequencies and crash rates. Anecdotal evidence suggests that corrections to path over- lap problems can have noticeable effects on safety perform- ance. These corrections can be geometric changes to entry and exit curvature, striping changes to better define lane positioning, or a combination of the two. Striping changes that have been anecdotally found to be successful consist of exit striping patterns that guide circulating vehicles to the proper exit without the need to change lanes within the cir- culatory roadway. In January 2006, the National Committee on Uniform Traffic Control Devices approved striping rec- ommendations consistent with this philosophy. After the Gateway Roundabout in Clearwater Beach, Florida, experienced an unacceptably high number of crashes, it was modified through a combination of geometric and striping changes to reduce the number of these crashes caused by path overlap (40). Anecdotal evidence suggests a large reduction in the number of crashes (41). While the research team does not dispute the overall improvement made in vehicular crash experience with the implemented changes, it notes that the vast majority of crashes in the before period consisted of “subreportable” PDO crashes for which a special reporting effort was made (42); whether these crashes were similarly reported after the improvements were made cannot be confirmed. Therefore, the overall magnitude of improvement to the intersection appears to be substantial but cannot be accurately quantified. Conclusion In general, the majority of the roundabouts in the United States appear to operate without any significant operational or reported safety deficiencies. However, the findings from this project suggest a number of areas where special attention is needed to ensure the safe and efficient operation of the roundabout for all users. These areas include the following: • Multilane roundabouts need to be carefully designed to avoid entry and exit path overlap. The majority of the mul- tilane roundabouts with high crash frequencies and high crash rates relative to the other sites in the database exhib- ited some degree of vehicle path overlap. In addition, some of these sites also experienced reduced operational per- formance in terms of unbalanced lane utilization on the approach. • Roundabout exits tend to have a higher percentage of vehi- cles that do not yield to pedestrians than roundabout entries. As a result, the design of the exit should be carefully considered to ensure that vehicle speeds are reasonable and that good sight lines exist between drivers and pedestrians. The recommended speed methodologies presented in this report may be used to estimate exit speeds based on the configuration of the roundabout. • Multilane roundabouts tend to have a higher percentage of vehicles that do not yield to pedestrians on either entry or exit. While no quantifiable crash experience has resulted from this behavior, it may reduce the usability of the roundabout crosswalk for pedestrians. 101