Design Guidance for Intersection Auxiliary Lanes (2014)

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.

58 Double Left-Turn Lane Field Study Background The objective of NCHRP Project 03-102 was to recommend improvements to the guidance provided in the AASHTO Pol- icy on Geometric Design of Highways and Streets (commonly known as the Green Book) (2). Task 1 was to identify the state of design practice for auxiliary lanes at intersections by gathering and synthesiz- ing information on existing practices and research. Task 2 determined the issues that merit further study to validate, enhance, and expand current Green Book guidance. Task 3 produced the interim report and a Phase II work plan, which includes suggested research efforts for the remaining tasks. An area suggested for additional research was on multiple left-turn lanes, such as double left-turn lanes or triple left- turn lanes (TLTL). Green Book Review on Double or Triple Left-Turn Lanes (Subsection within 9.7.3) The Green Book section on double or triple left-turn lanes presents information about double and triple left-turn lanes, including advice on offtracking and swept path width. This section is new to the 2011 Green Book. Several general state- ments are made regarding multiple turn lanes. Table 5-1 lists Green Book statements along with potential research needs. Item 5 can be addressed with a reference to the Manual on Uniform Traffic Control Devices (72), Section 3B.08 (Exten- sions through Intersections or Interchanges). Figure 3B-13 (D) in the MUTCD includes an example of typical dotted line markings to extend lane line markings into the intersection for double left-turn lanes. A recent TxDOT study (52), the FHWA Signalized Inter- section: Informational Guide (7), and a Florida study (73) pro- vide guidance for multiple left-turn lanes and/or double right- turn lanes. These references could be used to develop guidance statements for the Green Book. Therefore, the research team suggested that efforts within NCHRP Project 03-102 focus on items 1, 2, and 3 in Table 5-1. The project panel instructed the research team, at the conclusion of Phase I, to focus on the operations of double left-turn lanes. Study Selection The goal of this study was to determine the effects of geomet- ric characteristics on operations, as measured using saturation flow rate and lane utilization, for double left-turn lanes. The geometric variables that were the focus of this study were receiv- ing leg width, left-turn lane width, and downstream friction location (type and distance). A friction location is a roadway feature that could affect the behavior of left-turning vehicles within the receiving leg. Examples of friction locations include bus stops, driveways, and exits from channelized right turns. Literature Several projects have documented research with multiple turn lanes, and this section summarizes findings. Wortman (48) conducted a state-of-the-art review on double (then known as dual) left-turn lanes for the Arizona DOT in the 1980s. In it, he cited work from Neuman (5) that double left-turn lanes should be considered at any signalized intersection with high left-turn design-hour demand vol- umes and a general rule-of-thumb of 300 vehicles per hour or more as the appropriate demand volume for consideration of the double left-turn lanes. Other guidelines from Neuman (5) include the following: • The throat width for the turning traffic is the most impor- tant design element. Drivers are most comfortable with extra space between the turning queues of traffic. Because of the offtracking characteristics of vehicles and the relative difficulty for acceptance of two abreast turns, a 36-ft throat width is desirable for acceptance of two lanes of turning C H A P T E R 5

59 traffic. In constrained situations, 30-ft throat widths are acceptable minimums. • Guiding pavement markings to separate the turning lanes are recommended. These channelization lines should be carefully laid out to reflect offtracking and driving characteristics. • Double left-turn lanes operate at approximately 1.8 times the capacity of a single left-turn lane. Wortman (48) also cited some studies showing that the capacity of either lane in a double left-turn lane approach was lower than a single left-turn lane (e.g., a lane in a double left- turn might have 80% of the capacity of a single left-turn lane) and that the inside (median-side) left-turn lane had a somewhat lower capacity than the outside (shoulder-side) left-turn lane. Ackeret (75) sought a better understanding of the opera- tional characteristics of single, double, and triple left-turn lanes through measurements of saturation flow rates. His study included 36 double left-turn lane approaches along with 23 single and 3 triple left-turn lanes within the Las Vegas metro- politan area. Evaluation of analysis plots for key intersection characteristics suggested the following: lane widths over 12 ft are associated with higher saturation flow rates, lower satura- tion flow rate is present in the absence of longitudinal lane line pavement markings that are extended through the inter- section area, and concurrent left-turn movements and sep- aration distances may influence left-turn saturation flow rates. His research found no significant difference in performance between the double left-turn inside lane and outside lane. He recommended a double “left-turn lane saturation flow rate # 2011 Green Book Statements Potential Additional Research Needs Potential Source for Information 1 “…with three-phase signal control, such an arrangement results in an increase in capacity of approximately 180% of that of a single median lane.” Does the 180% value need to be updated/verified? Is the saturation flow rate affected by the location of a downstream friction point such as a bus stop or driveway? Data are available from previous studies regarding saturation flow rates for triple and double left-turn lanes. These studies; however, are not sensitive to factors of interest such as turn lane width, receiving leg width, and downstream friction points. A study by Sando and Moses (74) identified some geometric features that influence saturation flow rate for triple left turns. Additional study that considers left-turn lane width, downstream friction points, and receiving leg width is needed. 2 “Occasionally, the two- abreast turning maneuvers may lead to sideswipe crashes.” Is more current information available? What design characteristics affect the crash rate (e.g., lane width on approach, receiving lane/leg width, presence of pavement markings, location of downstream friction point(s)) Study needed. 3 “The receiving leg of the intersection should have adequate width to accommodate two lanes of turning traffic. A width of 9 m [30 ft] is used by several highway agencies.” Is more current information available? Study needed. 4 “Triple left-turn lanes have also been used at locations with very high left-turn volumes.” Should more insight into triple left-turn lane operations be provided? Material to add to Green Book may be available in the recent TxDOT 0- 6112 project (52), the FHWA Signalized Intersections: Informational Guide (7), and a report from a Florida study (73). 5 “…longitudinal lane line markings of double or triple left lanes may be extended through the intersection area to provide positive guidance.” Should this statement be followed by a reference to the MUTCD? Reference could be added to MUTCD Section 3B.08 (Extensions Through Intersections or Interchanges. Table 5-1. Green Book material on double or triple left-turn lanes where additional research may be needed.

60 on the order of 1870 pcphgpl” (75). Based on the suggested left-turn lane saturation flow rate and his findings for the neighboring through lane, he suggested that a higher left- turn factor on the order of 0.98 or 0.99 be used rather than the 0.95 value present in the 1994 Highway Capacity Manual (76). The 2010 Highway Capacity Manual (8) includes a left- turn adjustment factor of 1/1.05 or 0.952. Brich (49) examined positive guidance pavement mark- ings for double left-turn lanes in Virginia and concluded that the “prevailing opinion is that this type of pavement mark- ings for double left-turn lanes facilitates safe and efficient movement through the intersection.” A figure was provided to illustrate that in the cases where the receiving roadway has more than two lanes, the pavement markings should be installed to have the left-most turn lane enter the left-most travel lane on the receiving roadway. NCHRP Report 505 (51) discusses characteristics of inter- section geometry to accommodate trucks. The authors ref- erenced the existing Green Book guideline that the desirable turning radius for a double left-turn lane is 90 ft, but they concluded that there was little else on the design of multiple turn lanes to account for turning trucks. They stated that the primary factor to consider in designing double left-turn lanes is vehicle offtracking or swept path width. When vehicles nego- tiate the turn side by side, the vehicles should not encroach on the adjacent travel lane. Because many factors affect the con- trol turning radius of double left-turn lanes, they stated that it is necessary to provide guidance on the range of offtracking or swept path width of design vehicles for various turning radii. They determined offtracking and resultant swept path widths of several design vehicles for 90-degree turns with centerline turn- ing radii of 50, 75, 100, and 150 ft using AutoTURN software. Based on their analysis, they recommended that an exhibit be included in the Green Book that indicates the swept path width of several design vehicles for centerline turning radii of 75, 100, and 150 ft, to provide flexibility in designing adequate turning paths for double left-turn lanes by allowing for interpolation of swept path widths for a range of turning radii. Cooner et al. (52) conducted research in Texas on design and operations of triple left-turn and double right-turn lanes. The field studies in Texas collected both static (e.g., lane widths, grades, pavement markings, traffic signs, upstream and downstream conditions, and signal timing) and dynamic (e.g., volumes by lane, saturation flow rate, and critical events) data in order to evaluate design and operational performance. Researchers collected the data at five TLT and 20 DRT lane sites, primarily in the Dallas–Fort Worth and Houston urban areas. They reported the following as key findings for TLTL: • Lane utilization patterns were varied for each of the five sites studied. • All sites were T-intersections with peak-hour volumes from 646 to 2,846 vehicles. • Lighted pavement markers used to delineate the lane lines between the TLTL were effective at reducing violations and well received by the public at one site. • Saturation flow rates in Texas were consistent with earlier published national values. Sando and Moses (74) and Sando and Mussa (77) exam- ined the influence of intersection geometrics on the opera- tion of triple left-turn lanes. Their studies identified several geometric characteristics that did and did not influence oper- ations. Variables that did not influence saturation flow rate included intersection type (four-leg versus T-intersection), shadowing effect (shadow or no shadow), time of day (morn- ing, afternoon, or evening), and lane (outermost, middle, or inner most). Variables that significantly influenced saturation flow rate included the following: • Skewness (skewed intersections resulting in less than 90-degree left turns had higher saturation flow rates than right-angled intersections). • Street type (two-way streets had higher saturation flow rate than one-way streets, which was attributed to the tightness of the turning curve for the left-turning vehicles on the one-way streets). • Approach grade (downgrade had higher saturation flow rates than level grade). • Approach type (straight approaches had higher saturation flow rates than curved approaches). They noted that there is a need to investigate downstream attraction bias. The left-turn lane width was provided but not discussed, and receiving lane width was not discussed in these studies. Another variable known to affect left-turn saturation flow rate is U-turns. A 2005 study (78) measured vehicle headways in exclusive left-turn lanes at 14 signalized intersections—six sites had single left-turn lanes and eight had double left-turn lanes where only the inside turn lane was studied (because that was the lane assumed to be affected by U-turns). Regression analysis of saturation flow rate data showed a 1.8% satura- tion flow rate loss in the left-turn lane for every 10% increase in U-turn percentage. The safety analysis in that study used a set of 78 intersections. One of the findings was that sites with double left-turn lanes were found to have a significantly greater number of U-turn collisions. The 2010 version of the Highway Capacity Manual (8) pro- vides a lane width adjustment factor to “account for the nega- tive impact of narrow lanes on saturation flow rate and allows for an increased flow rate on wide lanes.” The values provided are 0.96 (average lane width of 8 to <10 ft), 1.00 (average lane width between 10 and 12.9 ft), and 1.04 (average lane width of more than 12.9 ft). These values indicate that saturation flow rate is affected by lane width.

61 Objective and Measures of Effectiveness The objective for this research study was to determine the effects of geometric characteristics on operations for dou- ble left-turn lanes. The following geometric variables were the focus of this study: receiving leg width, left-turn lane width, and downstream friction point (type and distance).The key measures of effectiveness for the field study were • Saturation flow rate measured as number of vehicles per hour of green per lane. • Lane utilization. These measures were used to identify the geometric variables that influence operations at DLTLs. Study Matrix The initial task for the field study was to refine the study matrix based on comments from the panel, information available in the literature, and availability of sites with pre- ferred site characteristics. The variables that might affect the operations were identified as being left-turn lane width, receiving leg width, and downstream friction point. To ensure sufficient variability in these study variables, the following ranges were used during site identification: • Receiving leg width, measured at extension of stop bar: – Narrow, less than 26 ft. – Moderate, between 26 and 30 ft. – Wide, more than 30 ft. • Left-turn lane width, average: – Less than 12 ft. – 12 ft or more. • Location of friction point: – Near, within 150 ft of end of turn. – Medium, between 150 and 350 ft of end of turn. – Long, more than 350 ft from end of turn. The left-turn lane width ranges were modified to have an 11.5-ft (rather than a 12-ft) breakpoint because of the limited number of sites identified with an average lane width of 12 ft or more. A previous study (74) indicated that the following can affect results: presence of an intersection skew, downgrade on the approach, horizontal curve on the approach, and approach street being one-way or two-way. Therefore, the research team attempted to control for these variables by minimizing the num- ber of sites with these characteristics. Table 5-2 lists the preferred site selection characteristics used during site identification. Variables that may affect operation, but were not the focus of the study, needed to be controlled by being held at a constant level. For example, the approach grade and horizontal align- ment can affect operations and safety; therefore, because this is known, this research focused on other variables. Sites with level grades and straight alignment were preferred. Study Sites The research team contacted colleagues to help identify dou- ble left-turn lane sites. Researchers also used aerial photographs in Google Earth to identify sites and to gather preliminary Variable Characteristics Number of left-turn lanes Two lanes (required) Area type Urban or suburban Surface condition Fair to good Intersection sight distance Adequate Grade Level (prefer between -2 and 2) Main road number of lanes Prefer four lanes Cross road number of lanes Four lanes (two receiving lanes for the left-turn movements) Left-turn lane assignment All left-turn lanes must be exclusive left-turn lanes Left-turn lane length Minimum of 140 ft Main road horizontal alignment Straight Cross road horizontal alignment Straight Adjacent parking lane Select sites without on-street parking Street type Prefer both streets to be two-way streets Shadowed* Prefer sites with shadowed turn lanes Presence of intersection lane line extensions Prefer sites with left-turn lane line markings extending into the intersection Median Prefer sites with raised median on both approaches Posted speed limit Prefer sites with 35, 40, or 45 mph posted speed limits Skew Prefer sites to be near 90 degrees * In shadowed left-turn configuration, the median prevents the left-turn lanes from being accessed directly by upstream through lanes. In unshadowed configuration, a through lane terminates in a left-turn lane at the intersection. A partially shadowed condition exists when a through lane vehicle could enter a portion of the left-turn lane when maintaining a straight path. Table 5-2. Preferred site selection characteristics.

62 characteristics of the sites. The goal was to collect data from sites that varied in the following characteristics: • Left-turn lane widths (range between 9 and 12 or more ft). • Receiving leg width (range between 24 ft and 40 or more ft). • Distance along receiving leg to friction point (range between 10 ft and more than 450 ft). These preliminary characteristics were used to select sites while also considering researchers’ ability to efficiently collect data at these locations. More than 200 sites were considered during site selection. Additional considerations during site selection included the following: • The double left-turn bay needed to be a minimum of 140 ft so that a queue length of seven vehicles or more was possible; otherwise, saturation flow rate was not calculable. • Researchers preferred a left-turn bay length of 200 ft, which could accommodate a queue length of 10 vehicles. The goal was to collect data in a minimum of three states— data were collected in Texas (College Station, Bryan, and Houston), Arizona (Flagstaff, Phoenix, and Tucson), and California (San Leandro and Palo Alto). Obtaining the desired range of receiving leg width was the most difficult of the study variables, although finding sites with an average double left- turn lane width greater than 12 ft also proved challenging. Only two sites met the greater than-12-ft average lane width criterion. Several locations had the double left-turn lanes turn- ing into a receiving leg that had three lanes. The presence of the third lane enabled drivers to adjust their travel path. There- fore, only sites with a two-lane receiving leg (or three lanes when the third lane is added from a channelized right-turn lane) were considered. All of the sites had longitudinal lane line markings extended through the intersection to provide posi- tive guidance to the left-turning drivers. Some of the sites had to be eliminated after site selection and initial data collection because an insufficient number of queues were recorded or the signal operated in both pro- tected and permissive modes. Data reduction was completed for 26 sites. The variables and their descriptions collected for each site used in the analysis are provided in Table 5-3. Table 5-4 and 5-5 list the study site characteristics for the double left-turn lane approach and the receiving leg, respectively. Variable Description Site Site name A-PSL DLTL approach: posted speed limit (mph) A-LT_W DLTL approach: average width of left-turn lanes (ft) A-Med_W DLTL approach: median width (ft) A-Dis_Sig DLTL approach: distance to nearest signal (ft) A-Sig0-5 DLTL approach: number of signals within 0.5 mile A-DwRt DLTL approach: number of driveways within 1000 ft on right edge A-DwLf DLTL approach: number of median openings/driveways within 1000 ft on left edge A-Bar-Nose DLTL approach: stop bar to median nose (ft) A-DLTL_Len DLTL approach: length of DLTL, a minimum value was required to include as a study site (ft) A-Tap_Len DLTL approach: length of the taper (ft) A-Tap_Des DLTL approach: type of taper design; either curved, straight, or none A-Shadow DLTL approach: type of shadow design; either shadow, partial shadow, or two-way, left-turn lane A-Bar-Curb DLTL approach: distance between stop bar and extension of curb line (ft) Ri Turn: Radius at start of turn (ft) Rf Turn: Radius at end of turn (ft) R-Lg_W-bar Receiving leg: width of the receiving leg at extension of stop bar (ft) R-Lg_W-100 Receiving leg: width of the receiving leg 100 ft beyond extension of stop bar (ft) R-TH_W-100 Receiving leg: average width of the through lanes on receiving leg, 100 ft beyond extension of stop bar (ft) R-Bike Receiving leg: presence of bike lane (yes/no) R-Dis_Sig Receiving leg: distance to downstream signal (ft) R-Sig0-5 Receiving leg: number of signals within 0.5 mile R-DwRt Receiving leg: number of driveways within 1000 ft on right edge R-DwLf Receiving leg: number of median openings/driveways within 1000 ft on left edge R-Friction Receiving leg: type of friction point, either bus stop, channelized right-turn lane, driveway/intersection, or friction is >500 ft downstream R-D_Beg_Fric Receiving leg: distance to start of friction point (499 used when >500 ft) (ft) R-SP_Add100 Receiving leg: lane added on the receiving leg within 100 ft of intersection (yes/no) Table 5-3. Descriptions for the site variables.

63 Figure 5-1 depicts the site characteristics gathered by techni- cians for each site. The descriptions of the per-queue variables are provided in Table 5-6. Data Collection The data collection method used was video recording. Video provided a permanent recording of the conditions at the site. Prior to beginning data collection, data collectors visited the site to determine the best locations to set up equip- ment, and they coordinated as needed with local authorities and adjacent landowners. The length of recording at each site was generally between 3 and 6 hr, with a few sites being recorded across multiple days to ensure that a sufficient sample size was obtained. Researchers typically used camcorders to collect data at the sites. These recorders created a time-stamped video recording of traffic conditions at each intersection during the study period. The video was used to obtain the time each double left-turning vehicle departed the stop bar. These times were used to calculate the headway data and, from that data, the saturation flow rate. The video also pro- vided the opportunity to validate conditions (e.g., presence of pedestrians or if the driver made a U-turn maneuver). The recording from another camera provided a record of lane changing just downstream of the turn due to the presence of a friction point (e.g., bus stop or driveway), any erratic maneuvers (e.g., panic braking, swerving), or conflicts. Typically, a three-camera setup was used at each site, as shown in Figure 5-2. Camcorders were installed so that • Camcorder #1 captured vehicles crossing the stop bar in the double left-turn lane. This view was the view of high- est importance for the study because the wheels crossing the stop bars must be easily seen in the video. The view of Camera 1 had to be wide enough (but not too wide) to catch at least the first two vehicles in the queue (before the stop bar) and also the vehicles crossing the stop bar and entering the intersection. • Camcorder #2 captured the entire double left-turn lane queue, especially vehicles 7, 8, 9, and 10 in the queue. Site A -P SL A -L T_ W A -M ed _W A -D is_ Si g A -S ig 0. 5 A -D w R t A -D w Lf A -B ar -N os e A -D LT L_ Le n A -T ap _L en A -T ap _D es A -S ha do w * A -B ar -C ur b AZ-FS-03 35 12.0 3 1382 1 4 0 7 438 101 curved SR 15 AZ-FS-04 35 12.0 4 455 1 1 1 2 150 94 curved PS 23 AZ-FS-05 35 12.0 5 1485 1 0 0 4 315 88 curved PS 25 AZ-FS-06 40 12.0 1 3000 0 0 1 3 225 N/A straight TW 27 AZ-FS-07 40 11.0 1 1097 1 6 0 0 218 N/A straight TW 42 AZ-PH-02 40 10.0 4 885 3 2 1 -5 170 120 curved SR 19 AZ-PH-06 30 11.0 6 800 2 3 6 10 200 136 curved SR 25 AZ-PH-07 40 12.0 1 2700 0 1 1 0 268 N/A straight TW 25 AZ-PH-08 35 9.5 1 553 2 8 4 5 150 183 curved SR 22 AZ-PH-09 40 10.5 4 355 2 2 2 0 260 N/A none SR 48 AZ-PH-12 45 10.5 4 363 3 3 2 3 271 N/A none SR 50 AZ-PH-13 45 10.0 3 367 2 2 2 2 265 N/A none SR 59 AZ-PH-15 45 11.5 4 368 2 6 7 1 271 N/A none SR 53 AZ-PH-16 45 10.5 3 364 3 1 1 0 266 N/A none SR 55 AZ-TU-09 45 12.0 3 2810 1 4 2 9 250 425 curved SR 29 AZ-TU-10 45 13.0 6 5000 0 2 3 14 410 300 curved SR 45 CA-BA-04 35 11.0 4 446 3 10 1 0 205 214 curved SR 21 CA-ST-01 35 11.0 4 1266 2 5 3 -6 344 113 curved SR 23 CA-ST-02 35 10.5 4 1200 3 6 4 -5 284 128 curved SR 28 CA-ST-04 35 11.0 5 1336 2 0 3 4 280 112 curved SR 46 TX-CS-01 40 11.5 3.5 1900 1 3 1 7.5 140 121 curved SR 32 TX-CS-02 40 13.0 3 1900 1 6 0 5 309 137 straight SR 10 TX-CS-03 40 11.0 3 2635 1 4 5 4.5 151 148 curved SR 20 TX-CS-04 45 12.0 2.5 1219 1 0 0 0 422 223 straight SR 20 TX-HO-02 35 9.5 1.5 1584 1 4 2 3 124 121 curved SR 18 TX-HO-03 45 12.0 7 975 1 4 0 5 223 150 curved SR 27 *A-Shadow, SR = shadow with raised median (no direct entry to turning lanes, i.e., driver must change lanes to enter turn lane), PS = partial shadow, TW = two-way, left-turn on approach. Table 5-4. Site characteristics on double left-turn lane approach.

64 • Camcorder #3 captured vehicles turning in the inter- section and erratic vehicle maneuvers caused by the friction location. Camcorder #3 was placed about 20 ft after the friction location on the receiving leg (see Fig - ure 5-3) in order to have a good view of the left-turn vehicles traveling from the intersection toward the cam- era and passing the friction location. If there was no friction location within 450 ft of the intersection (on the receiving leg), then Camcorder #3 was mounted approximately 450 ft downstream of the intersection (on the receiving leg) where there was a good view of all the vehicles traveling from the intersection toward the camera. The research team also collected roadway geometry data for each study site. Typically, the geometric data were obtained using aerial photographs. These measure- ments were confirmed in the field or the geometric data were recorded on pre-printed worksheets created for this study. Data Reduction Saturation Flow Rate Technicians reviewed the video and documented the time each left-turning vehicle crossed the stop bar. These times were used to determine the headway between following vehi- cles. Technicians also recorded whether the vehicle was a truck or whether the vehicle was in the queue at the start of the cycle. If either case was true, then the queue was eliminated from the study. Both Camera 1 and Camera 2 views were used to gather the vehicle type data and the last vehicle in queue data. According to the Highway Capacity Manual (8) proce- dure, each cycle must have more than eight vehicles to be considered in the determination of saturation flow rate. The ITE Manual of Transportation Engineering Studies (79) rec- ommends using the seventh, eighth, ninth, and tenth vehicle in the queue. Both procedures require the first four vehicles to be dropped in the analysis to eliminate headways with startup lost times and that only the passenger cars in the traffic stream are to be included. During early stages of data Si te R i R f R -L g_ W @ ba r R -L g_ W @ 10 0 R -T H _W @ 10 0 R -B ik e R -D is_ Si g R -# Si g0 .5 R -# D w R t R -# D w Lf R -F ri ct io n* R -D _B eg _F ri c R -S p_ A dd 10 0 AZ-FS-03 107 78 29 42 13 no 1030 1 0 0 RTL 33 Yes AZ-FS-04 67 94 31 25 12 no 575 1 2 1 D/I 234 No AZ-FS-05 85 101 46 29 11 yes 1605 1 1 1 >500 999 No AZ-FS-06 58 97 34 31 11 yes 1056 1 1 0 D/I 289 No AZ-FS-07 77 109 33 30 12 yes 1056 1 0 1 >500 999 No AZ-PH-02 80 82 25 23 11 no >0.5 0 1 1 D/I 400 No AZ-PH-06 72 96 26 25 11 no 695 2 1 3 BS 110 No AZ-PH-07 62 83 42 37 12 no 2600 1 8 2 BS 30 Yes AZ-PH-08 67 84 25 24 10.5 yes 773 3 3 5 BS 57 No AZ-PH-09 94 65 46 24 12 no >0.5 1 0 0 >500 999 No AZ-PH-12 93 69 48 24 12 no >0.5 1 0 0 >500 999 No AZ-PH-13 85 76 52 24 12 no >0.5 1 0 0 >500 999 No AZ-PH-15 88 71 52 24 12 no >0.5 1 0 0 >500 999 No AZ-PH-16 88 77 54 24 12 no >0.5 1 0 0 >500 999 No AZ-TU-09 93 111 48 35 12 yes >0.5 1 3 5 D/I 365 No AZ-TU-10 128 135 28 27 12 yes 1839 1 3 0 D/I 135 No CA-BA-04 80 86 26 25 11.5 no 466 3 5 4 D/I 90 No CA-ST-01 78 95 34 34 11.5 yes >0.5 1 10 3 D/I 130 No CA-ST-02 82 101 35 35 11.5 yes 734 2 3 4 BS 150 Yes CA-ST-04 83 129 31 28 11.5 yes >0.5 1 0 0 RTL 42 No TX-CS-01 82 92 24 32 12 yes 1970 1 0 1 RTL 14 No TX-CS-02 75 94 36 38 12 no 1130 1 3 2 D/I 414 No TX-CS-03 83 97 25 37 11 no 1115 2 0 1 RTL 13 Yes TX-CS-04 60 88 40 41 12 no >0.5 0 1 1 RTL 20 Yes TX-HO-02 73 102 30 21 12 no 2100 1 4 2 D/I 64 No TX-HO-03 103 100 31 23 11 no >0.5 0 1 2 D/I 300 No *R-Friction: RLT = channelized right-turn lane, D/I = driveway/intersection, BS = bus stop, >500 = friction point is more than 500 ft from intersection. Table 5-5. Site characteristics on receiving leg.

65 Figure 5-1. Graphic used to assist with gathering site characteristics. reduction, few sites had sufficient number vehicles within the queue. The research team directed the field crew to record for 4 hr, or longer, to record at least 35 queues of eight passenger cars. Even with the longer recording times, several sites had less than 35 queues of eight passenger cars. Because this study focused on how geometric design variables (e.g., lane width) affected operations, the research team decided to retain the data for queues that had fewer than eight vehicles. The number of vehicles in the queue was included in the analysis to control for the effects that the queue length may have had on operations, and passenger cars in positions 4 through 10 of a queue were used in this study. Saturation flow rate was calculated for each passenger car using the following equation: ( )=SFR 3600 H VQ-4 where SFR = saturation flow rate in passenger cars per hour of green per lane (pcphgpl). H = time headway between subject vehicle and the fourth vehicle in the queue (sec). VQ = subject vehicle position in the queue (e.g., 5, 6, 7, 8, 9, or 10). Variable Description SFR or SatFlowRate Calculated saturation flow rate using the average headway, in passenger cars per hour of green per lane (pcphgpl) Site Site name Lane Lane number, where 1 = inside lane and 2 = outside lane Vehicle Queue Number of vehicles within the queue used to calculate the saturation flow rate Same Queue The same value in this column (for a given site) indicates that the SFR was based on vehicles used in another SFR calculation U-turns-w/in-queue The number of vehicles within a given queue that performed a U-turn Table 5-6. Descriptions for the per-queue variables.

66 Figure 5-2. Example of video camera setup for double left-turn lanes study. Figure 5-3. Subareas used while reducing Camera 3 driver behavior.

67 To handle extreme outliers, researchers decided to use a maximum selected saturation-flow-rate value based on an assumed headway of 1 sec, which resulted in a maximum saturation flow rate of 3600 pcphgpl value. A total of 18 of 10,041 data points were removed (less than 0.2%), resulting in 10,023 data points available for analysis. While a smaller saturation-flow-rate value could be used as the maximum value, the research team decided to start with a value that would eliminate obvious data reduction errors. Table 5-7 lists the average saturation flow rates by site for each lane. Lane Distribution At some DLTL intersections, drivers may choose one left- turn lane over another left-turn lane in anticipation of a turn at a downstream intersection or because of familiarity with a downstream friction point. The data available in this study were used to investigate if geometric elements are associated with how drivers distribute within the double left-turn lanes. The Highway Capacity Manual (8) has a lane utiliza- tion factor to account for the unequal distribution of traf- fic among the lanes in those movement groups with more than one exclusive lane. A lane utilization factor of 1.0 indi- cates traffic is evenly distributed across different lanes in the movement group. Values less than 1.0 indicate traffic is not evenly distributed. The lane utilization factor, fLU, can be computed using the following: ( )=f V V NLU g g1 where fLU = lane utilization factor. Vg = unadjusted demand flow for the lane group (veh/hr). Vg1 = unadjusted demand flow on a single lane in the lane group with the highest volume (veh/hr). N = number of lanes in the lane group. Unadjusted demand flow refers to traffic flow not adjusted to take into consideration uneven traffic distribution within a lane group. Because of the nature of the equation, which lane has the higher volume is not associated with the factor. Therefore, another measure was needed in this study so that the propor- tion of vehicles in a given lane can be compared to the condi- tions at the site. Site SFR Average Lane 1 Count Lane 1 SFR Average Lane 2 Count Lane 2 SFR Average Both Lanes Count for Site AZ-FS-03 1777 147 1839 164 1810 311 AZ-FS-04 1611 172 1629 180 1620 352 AZ-FS-05 1688 226 1730 280 1711 506 AZ-FS-06 1630 191 1645 498 1641 689 AZ-FS-07 1905 1 1776 39 1779 40 AZ-PH-02 1607 72 1782 140 1722 212 AZ-PH-06 1864 36 1798 38 1830 74 AZ-PH-07 1972 175 1858 178 1915 353 AZ-PH-08 1789 18 1818 30 1807 48 AZ-PH-09 1633 17 1791 6 1674 23 AZ-PH-12 1757 343 1749 132 1755 475 AZ-PH-13 1846 362 1866 258 1854 620 AZ-PH-15 1783 381 1771 283 1778 664 AZ-PH-16 1845 315 1799 289 1823 604 AZ-TU-09 1931 450 1888 394 1911 844 AZ-TU-10 1699 423 1680 322 1690 745 CA-BA-04 1942 3 1721 14 1760 17 CA-ST-01 1735 339 1655 366 1693 705 CA-ST-02 1820 235 1766 268 1792 503 CA-ST-04 1783 58 1769 44 1777 102 TX-CS-01 1647 101 1725 99 1685 200 TX-CS-02 1652 136 1741 131 1695 267 TX-CS-03 1780 144 1845 122 1810 266 TX-CS-04 1842 244 1954 296 1903 540 TX-HO-02 1754 319 1860 421 1814 740 TX-HO-03 1631 84 1757 39 1671 123 Total 1774 4992 1776 5031 1775 10023 Table 5-7. Average saturation flow rate (pcphgpl) for each lane and site.

68 A variable was created to calculate the percent of the volume present within a cycle to each lane. This lane share variable is aimed at determining the proportion of vehicles that use a lane out of all left-turn vehicles recorded during a cycle. It is to provide a good measure of the distribution of left-turn demand of passenger cars across the double left-turn lanes. The lane share variable, L-share, was calculated as ( )=L -share V V Ni i t where Li-share = proportion of vehicles in lane i relative to total left-turn volume. Vi = volume for lane i (veh/hr). Vt = total volume for both lanes. N = number of lanes in the lane group, 2 for double left-turn lanes. A previous study on triple left-turn lanes (74) found that some intersections had higher inside lane usage factors than the outer lane while other intersections were observed to have the opposite results. The authors made the following observations from their data: • One intersection with an on-street bus stop less than 500 ft after the left turn had lower lane usage of the outer lane. • Innermost lanes of the shadowed intersections were less used while the innermost lanes of the unshadowed inter- sections were highly used. They noted that drivers’ choice of lane results from many factors, including upstream and downstream conditions; however, they were not able to quantify the relationships. Driver Behaviors Driver behavior may be related to the geometric design characteristics of the double left-turn lanes, which could affect the saturation flow rate (i.e., operations) or the safety of the intersection. For example, slow-to-start behaviors may decrease the saturation flow rate and increase the potential for rear-end crashes. The video recordings from Cameras 1 and 3 were used to identify driver behaviors of interest. Camera 1 was used to record driver behaviors near the stop bar. Table 5-8 lists the definitions of each behavior recorded using Camera 1. Camera 3 was used to record driver behaviors after the drivers passed the stop bar. The Camera 3 view was sub- divided into three areas (see Figure 5-3): T-area (after turn- ing left and entering the cross street), B-area (after crosswalk, before the friction location), and F-area (at the friction point). The main function of Camera 3 was to record the movements of left-turn vehicles within the turn and near the friction point along the receiving leg. The driver behaviors listed in Table 5-9 were noted from watching the Camera 3 view. Driver behavior data were reduced for approximately 2 hr of video for each site that had a friction point within 450 ft of the intersection and that had an acceptable view of the left-turning vehicles. Code Description U-Turn Did the vehicle make a U-turn? Over Bar Did the vehicle stop beyond the stop bar? A suggested criterion is, was any part of the vehicle’s tire on the stop bar? Backing Up Did the vehicle back up? Pedestrian/Bike Conflict Was there a conflict between a pedestrian or bicyclist and the left-turning vehicle? Slow to Start Was the driver distracted and slow to start the turn? A suggested criterion is, was the vehicle more than one car length behind vehicles in the neighboring lane? Table 5-8. Driver behavior recorded from Camera 1. Area Code Description T RT Conflict with Right-Turning Traffic (Opposing Traffic) LT Conflict with opposing Left-Turning Traffic PT Conflict with Ped/Bike at the Crosswalk of the Cross Street DT Stopping/Sudden Deceleration after Turning Left B BT Lane Change before fiction location IB Left-Turn into Improper Lane between the Intersection & Friction Location SB Low Speed between the Intersection & Friction Location F DF Stopping/Sudden Deceleration at Friction Location CF Lane Change at Friction Location PF Conflict with Ped/Bike/Traffic Exiting or Entering the Driveway (Back of Queue) DQ Stopping/Sudden Deceleration after turning left due to Queue downstream on cross street Table 5-9. Driver behavior recorded from Camera 3.

69 Analysis/Results Saturation Flow Rate Saturation flow rate represents the number of vehicles served by one lane over 1 hr of green time. It is calculated using the headway between following vehicles when all vehicles being considered were passenger cars and were present at the start of the green phase. The headways for the first four vehicles are dropped from the calculation. The saturation flow rate data were analyzed using the analysis of covariance (ANACOVA) mixed model including several site variables in Table 5-10 as well as the number of vehicles within the queue used to calculate the saturation flow rate (vehicle queue) and the number of vehicles within a given queue that performed a U-turn (U-turns-w/in-queue) as fixed factors/ covariates. This model considered each unique saturation flow-rate value at the intersections as opposed to averaging the saturation-flow-rate value for each intersection. Because individual saturation flow-rate values within the same queue from the same site are likely to be correlated, the variables of site and same queue were included as random factors to account for correlation. Parameters were estimated by the restricted maximum likelihood method implemented in JMP (SAS product). Several models had a relatively good fit and provided insights into how the geometric features at the intersection affect the double left-turn lanes operations. The model selected as the most informative is shown in Table 5-10. It includes two of the key variables of interest: lane width and receiving leg width. Distance to friction point variable was removed from the model for several reasons, as will be discussed below. To verify that the results would be similar without the non- significant variables, another run was made using only the significant variables (see Table 5-11). Lane A 1987 study (48) stated that the inside turn lane had lower capacity than the outside left-turn lane, a finding not sup- ported by more recent studies (74, 75, 77). The results from this dataset of 26 sites also support the finding that there are no differences in performance between the two double left- turn lanes. The results in Table 5-10 for the Lane[1] variable demonstrate that there is no significant difference in saturation flow rate between the inside and the outside lane. U-Turns As part of the data reduction efforts, whether the turning vehicle made a U-turn rather than a left turn was recorded. The number of U-turns made within a cycle was summed and considered within the analysis. Because U-turns require drivers to slow more than they would for a left turn, it is rea- sonable to assume that it will also take more time and there- fore negatively affect saturation flow rate. The model found that for each additional U-turning vehicle within the queue, saturation flow rate would decrease by 56.45 pcphgpl. Stated in another manner, one U-turning vehicle is associated with Response SFR Summary of Fit RSquare 0.697342 RSquare Adj 0.69716 Root Mean Square Error 229.6468 Mean of Response 1775.154 Observations (or Sum Wgts) 10023 Parameter Estimates Term Estimate Std Error DFDen t Ratio Prob>|t| Intercept 1897.3516 169.7568 22.12 11.18 <.0001* Lane[1] 1.918438 6.176225 2873 0.31 0.7561 Vehicle queue -2.491214 1.753325 7535 -1.42 0.1554 U-turns-w/in-queue -54.82995 10.50856 5650 -5.22 <.0001* A-LT_W -17.98789 14.09065 21.11 -1.28 0.2156 R-Lg_W-bar 3.2048467 1.413015 22.37 2.27 0.0333* SP_Add100[No] -52.31953 16.56282 21.42 -3.16 0.0047* REML Variance Component Estimates Random Effect Var Ratio Var Component Std Error 95% Lower 95% Upper Pct of Total Site 0.0613196 3233.8501 1292.8548 699.90123 5767.7989 2.292 Same queue 1.6143177 85135.325 2911.7246 79428.449 90842.2 60.334 Residual 52737.65 906.7706 51004.58 54561.01 37.374 Total 141106.83 3150.3364 135130.27 147490.81 100.000 -2 LogLikelihood = 142648.28513 Note: Total is the sum of the positive variance components. Total including negative estimates = 141106.83 Table 5-10. Evaluation of saturation flow rate.

70 a 3.4% decrease in saturation flow rate (-56/1680), while two U-turning vehicles are associated with a 6.7% decrease in saturation flow rate (-113/1680). The findings from this study show a larger impact of U-turning vehicles on saturation flow rate than the findings from a 2005 study (78). Number of Vehicles in Queue The number of vehicles in the queue was also not significant (see Table 5-10). In other words, whether the queue length was five vehicles or ten vehicles, similar saturation flow rates were measured after controlling for variations caused by other variables within the model. Friction Point on Receiving Leg The location and type of friction that would first be encoun- tered by a left-turning driver was identified for each site. The type of friction was categorized as • Channelized right-turn lane exit. • Bus stop. • Driveway or minor intersection. • No friction within 450 ft of the intersection. The distance between the friction point and the stop bar extension was measured. Because the friction points are actually not “points” but have a dimension to reflect the width of the driveway or the length of the bus stop, the distance was measured to the leading edge of the friction point. Analyses were conducted that considered the type of friction, the distance to the leading edge of the friction, and/or grouping the distances into reasonable ranges. One of the reasons the distances were grouped was to combine all the sites where the friction was more than 450 ft from the intersection. At several locations the next roadside fric- tion was more than 1000 ft, which is beyond a reasonable distance that should be influencing the operations for the double left-turn lanes. The type of friction was found to not be significant. Whether the friction was a driveway or a channelized right- turn lane exit did not influence the performance of the double left-turn lane operations. According to the model- ing results, the distance to the leading edge of the friction did influence operations; however, it was in a manner not expected. As the distance to the friction increased, saturation flow rate decreased. Expected was that the closer a friction point is located to the intersection, the more operations of the double left-turn lane would be compromised. Addi- tional evaluations into the site characteristics and the data revealed another variable that should be included in the model—a variable that reflects when a lane is added to the receiving leg. At several locations, the channelized right-turn lane added a downstream lane, in some cases within only a few feet of the intersection. While the turning vehicles were constrained at the start of the receiving leg, a review of the video data revealed that drivers in the outside lane would angle their vehicle to make a smooth entry into the new lane. This behavior resulted in higher saturation flow rates, as dem- onstrated with the variable SP_ADD100[No] being signifi- Response SFR Summary of Fit RSquare 0.697047 RSquare Adj 0.696957 Root Mean Square Error 229.7227 Mean of Response 1775.154 Observations (or Sum Wgts) 10023 Parameter Estimates Term Estimate Std Error DFDen t Ratio Prob>|t| Intercept 1679.5098 53.54089 24.08 31.37 <.0001* U-turns-w/in-queue -56.45924 10.13527 5069 -5.57 <.0001* R-Lg_W-bar 3.1498985 1.423279 23.11 2.21 0.0370* SP_Add100[No] -49.65104 16.62346 22.51 -2.99 0.0067* REML Variance Component Estimates Random Effect Var Ratio Var Component Std Error 95% Lower 95% Upper Pct of Total Site 0.0629358 3321.2783 1292.685 787.66217 5854.8944 2.354 Same Queue 1.6100791 84967.917 2905.5898 79273.066 90662.768 60.235 Residual 52772.51 907.13018 51038.741 54596.58 37.411 Total 141061.71 3144.6954 135095.57 147433.94 100.000 -2 LogLikelihood = 142667.5979 Note: Total is the sum of the positive variance components. Total including negative estimates = 141061.71 Table 5-11. Evaluation of saturation flow rate using only significant variables.

71 cant. The model results indicate that the addition of this new lane results in an increase in saturation flow rate of about 50 pcphgpl. To determine the impacts of friction type and location would require a more detailed study that considers the type of friction (e.g., bus stop and driveway) and the specific action at the friction point (e.g., driver waiting on driveway, driver existing driveway, and bus stopped at bus stop) when the left- turning vehicle arrives. Also needed would be the action of the left-turning vehicle (e.g., turning into driveway or chang- ing lanes to avoid activity at the friction point). Left-Turn Lane Width The width of the double left-turn lanes was thought to affect overall operations, especially as the Highway Capac- ity Manual (8) includes an adjustment factor for lane width. When lanes are wider, drivers may feel more comfortable and drive faster, which would be reflected in higher saturation flow rates. This analysis actually found the opposite with lower saturation flow rates with the wider left-turn lane widths; however, the result was not significant. Figure 5-4 shows the average saturation flow rate for each site along with the aver- age for the left-turn lane width. While one can see a down- ward trend in saturation flow rate compared to left-turn lane width, the graph also shows large ranges of saturation flow rates. For example, sites with 12-ft lanes had both the highest and the lowest saturation flow rates. The graph indicates that more than just the left-turn lane width is affecting satura- tion flow rate. In summary, within this dataset, there is not enough evidence to suggest the width of the left-turn lanes at these sites influenced the saturation flow rate. The saturation flow rates used in this evaluation only included passenger cars. If a truck or a bus was present within the queue, the data were eliminated. A future study could investigate the effects of larger vehicles on double left-turn lane operations. Receiving Leg Width The width of the receiving leg was defined as the distance between the median and the curb. It represents the visual tar- get for the left-turning drivers. A narrow receiving leg width may result in drivers turning more slowly as drivers have to take more care in positioning their vehicles to ensure that they do not hit the curb or the neighboring vehicle. A consistent location for the measurement was needed, and the research team decided to use an imagined extension of the stop bar present for the opposing direction approach. When the receiving leg was a one-way road, the width of the receiving leg along the crosswalk marking was used for the measurement. Measuring the receiving leg width prior to the end of the corner radius makes it possible to account for the extra pavement available to turning drivers when a larger corner radius is present, when a tapered nose design is used for the raised median, or when, because of the angle of inter- section, additional pavement is available to the turning driver. Because the receiving leg width was one of the key study variables, sites were selected to represent a range of receiving leg widths. All sites had two lanes at the start of the receiv- ing leg. A few sites had a third lane added to the leg from a channelized right turn; however; all of these sites had a raised channelized right turn. In other words, all turning drivers had to turn into only two lanes. Preliminary reviews of double left-turn lane operations revealed that when drivers can turn into three, rather than two, lanes, the turning behavior and, therefore, the saturation flow rate are affected. For the sites included in this study, the width of the receiving leg ranged from 24 ft to 54 ft. Figure 5-5 shows the average saturation flow rate per site by receiving leg width. The Green Book states that “the receiving leg of the inter- section should have adequate width to accommodate two lanes of turning traffic. A width of 9 m [30 ft] is used by sev- eral highway agencies.” The analysis conducted as part of this study found that the width of the receiving leg affected the Figure 5-4. Saturation flow rate by left-turn lane width.

72 saturation flow rate of the double left-turning traffic. While significant, the incremental difference is small. Each addi- tional foot of receiving leg width is predicted to increase satu- ration flow rate by only 3.2 pcphgpl (see Table 5-12). Within an analysis, both statistical and practical differences need to be considered. How to judge a practical difference is not as clearly defined as how to judge a statistical difference. One approach is to use the accuracy of the measurement technique (for example, 1 mph when measuring speed using radar). Another approach is to use a value that drivers could perceive, say 2 or 3 mph for a change in free-flow speed. A third approach could be to use a percentage of the average of the measurement. The average saturation flow rate for these sites is 1781 pcphgpl. If the variable contributes more than, say, 1% (18 pcphgpl) or 5% (89 pcphgpl), then one could argue that there is a practi- cal difference. The receiving leg width variable represents a change of 3.2 pcphgpl for 1 ft of change. So a 5.6-ft change in the receiving leg width could be considered to have both a sta- tistical and practical effect on saturation flow rate when using a 1% change in the average saturation flow rate. The pattern of increasing saturation flow rate for increas- ing receiving leg width was examined to try to identify if there were dimensions where a sizable increase in saturation flow rate occurs. A change point detection method was used to detect a shift in the mean vector (and the covariance matrix) when the data set consists of multivariate individual observations (R-Lg_W-bar and average [SFR] for each site). The method concluded that the change point appears at R-Lg_W-bar=36. The average saturation flow rate by group is shown in Table 5-12. These findings indicate that there is a difference in SFR when the receiving leg width is 24-36 ft and when it is 40 ft or more. The sites are different not only in receiving leg width (R-Lg_W-bar) but also in other characteristics, so there could be some confounding if just the relationship between receiv- ing leg width and saturation flow rate is examined without considering other factors. Therefore, an analysis of covariance was done based on the average SFR data to incorporate the effects of other variables in assessing the relationship between receiving leg width and average saturation flow rate. The least squares means for R-Lg_W-bar can be considered as the pre- dicted saturation flow rates that have been adjusted for the effects of other factors in the model. The change point detec- tion analysis based on those predicted saturation rates again identified receiving leg width of 36 ft as the change point. Lane Distribution As indicated earlier, the average lane utilization factor, fLU, accounts for the unequal lane utilization of the double left-turn lanes as a lane group. Before investigating whether unique site characteristics were affecting lane distribution, the measures were examined to identify if traffic demands are also affecting lane distribution. As demand increases, the selection of which left-turn lane to enter may be more of a reflection of drivers selecting the lane with the shorter queue rather than being concerned with downstream conditions. The lane utilization factor along with the lane share factors were plotted using the volume in the outside lane (Lane 2 in Fig- ure 5-6a) and the volume in the inside lane (Lane 1 in Fig- ure 5-6b). As shown in both plots and as expected, there is a strong relationship between fLU and the volume in Lane 1 or Lane 2. When more than approximately 7 vehicles for Lane 1 (see Figure 5-6b) or more than approximately 11 vehicles for Lane 2 (see Figure 5-6a) are present within a queue, the lane utilization factor is greater than 90%. Therefore, to be able to Figure 5-5. Saturation flow rate (site average with bars showing one standard deviation) by receiving leg width. Receiving Leg Width (ft) Average Saturation Flow Rate (pcphgpl) 24 to 36 1725 40 to 54 1833 Table 5-12. Average saturation flow rate by receiving leg width groups.

73 better focus on data when lane selection may be more influ- enced by geometric conditions, two datasets were created: • Dataset 11 contained the data when 11 or fewer vehicles are in each lane. • Dataset 7 included the cycles when at least one of the two lanes had seven or fewer vehicles. A preliminary plot of the Lane 2 share results for Data- set 11 and Dataset 7 (see Figure 5-7) shows that the percent of vehicles in Lane 2 was less than 50 for some sites and more than 50 for other sites. The characteristics of the sites may be influencing which lane is being selected by drivers. When using the dataset based on a maximum of 11 vehicles in a queue, the average fLU was 90% (see Table 5-13). The overall L2 share average for the 26 sites was 0.51. In other words, the split between the inside and outside lanes was nearly even when averaging the data across all sites. When reviewing the findings per site (see Table 5-13), variations between sites are seen with a range of 0.44 to 0.58 for the L2 share variable. The L2 share data were analyzed using the ANACOVA mixed model, including several site variables in Table 5-3 as well as Lane 1 volume (L1_Vol) and Lane 2 volume (L2_Vol) as fixed factors/covariates. Because individual L2 share values from the same site may be correlated, the variable site was included as a random factor to account for potential correla- tion. Parameters were estimated by the restricted maximum likelihood method implemented in JMP. The results for the model selected as the best and most informative is shown in Table 5-14. The variables found to be significant using Dataset 11 (within 0.10) included • Lane 1 volume • Lane 2 volume • Median width on the approach • Number of signals within 0.5 mi of the double left-turn lanes on the approach • The radius at the end of the left turn • The distance to the beginning of the friction point Examining the results of the statistical evaluation reveals the following: • Almost all of the L2 share prediction is accomplished by the Lane 1 and the Lane 2 volumes, indicating that drivers are making lane selection based on the length of queues present when the driver approaches the intersection. • The expectation is that the proportion of drivers select- ing Lane 2 would be smaller when the distance to the beginning of the friction point is small. In other words, drivers would avoid Lane 2 to evade potential conflicts at the nearby friction point. As expected, as the distance to the beginning of the friction point increases, the L2 share increases; however, even at the maximum distance avail- able in the dataset (999 ft), the increase in L2 share is only 0.004. Therefore, practically speaking, the location of the friction point overall does not influence the decision of the driver regarding which lane to select within the double (b) Lane Utilization by Lane 1 (Inside Lane) Volume (a) Lane Utilization by Lane 2 (Outside Lane) Volume Figure 5-6. Lane utilization by left-turn lane volume. Figure 5-7. Lane 2 share data by site.

74 Site Dataset 11 Dataset 7 fLU L2 share Count fLU L2 share Count AZ-FS-03 84% 0.58 76 81% 0.60 64 AZ-FS-04 88% 0.51 171 87% 0.51 146 AZ-FS-05 90% 0.52 92 88% 0.52 69 AZ-FS-06 86% 0.58 213 85% 0.58 192 AZ-FS-07 85% 0.57 23 85% 0.57 23 AZ-PH-02 87% 0.55 93 87% 0.55 92 AZ-PH-06 92% 0.50 64 92% 0.50 64 AZ-PH-07 93% 0.51 36 92% 0.51 25 AZ-PH-08 90% 0.55 35 90% 0.55 35 AZ-PH-09 93% 0.48 32 93% 0.48 32 AZ-PH-12 87% 0.44 127 86% 0.43 116 AZ-PH-13 93% 0.47 69 91% 0.46 25 AZ-PH-15 93% 0.48 101 90% 0.47 62 AZ-PH-16 95% 0.48 23 91% 0.46 7 AZ-TU-09 91% 0.49 81 91% 0.49 68 AZ-TU-10 91% 0.47 85 89% 0.46 58 CA-BA-04 92% 0.53 14 92% 0.53 14 CA-ST-01 91% 0.54 24 86% 0.58 10 CA-ST-02 91% 0.52 73 88% 0.53 52 CA-ST-04 91% 0.52 40 91% 0.52 37 TX-CS-01 91% 0.50 86 91% 0.51 82 TX-CS-02 89% 0.51 77 87% 0.51 63 TX-CS-03 93% 0.49 62 92% 0.49 54 TX-CS-04 92% 0.53 102 90% 0.54 69 TX-HO-02 90% 0.54 150 89% 0.55 115 TX-HO-03 89% 0.47 84 89% 0.47 83 Total 90% 0.51 2033 88% 0.52 1657 Table 5-13. Average lane utilization factor and L2 share per site. Response L2 share Summary of Fit RSquare 0.948108 RSquare Adj 0.947955 Root Mean Square Error 0.019131 Mean of Response 0.513478 Observations (or Sum Wgts) 2033 Parameter Estimates Term Estimate Std Error DFDen t Ratio Prob>|t| Intercept 0.5041318 0.006638 25.54 75.94 <.0001* L1_Vol -0.040686 0.00026 1521 -156.6 0.0000* L2_Vol 0.0388758 0.000267 1753 145.56 0.0000* A-Med_W -0.002238 0.000557 21.63 -4.02 0.0006* A-Sig0-5 0.0023339 0.001 24.41 2.33 0.0282* Rf 0.0001337 0.000064 24.17 2.09 0.0473* R-D_Beg_Fric 4.3374e-6 2.403e-6 24.19 1.80 0.0836 REML Variance Component Estimates Random Effect Var Ratio Var Component Std Error 95% Lower 95% Upper Pct of Total site 0.0291733 1.0677e-5 5.1257e-6 6.3056e-7 2.0723e-5 2.835 Residual 0.000366 1.1572e-5 0.0003443 0.0003898 97.165 Total 0.0003767 100.000 -2 LogLikelihood = −10184.46056 Fixed Effect Tests Source Nparm DF DFDen F Ratio Prob > F L1_Vol 1 1 1521 24508.63 0.0000* L2_Vol 1 1 1753 21187.57 0.0000* A-Med_W 1 1 21.63 16.1674 0.0006* A-Sig0-5 1 1 24.41 5.4448 0.0282* Rf 1 1 24.17 4.3688 0.0473* R-D_Beg_Fric 1 1 24.19 3.2577 0.0836 Table 5-14. Evaluation of L2 share.

75 left-turn lane group. Note, however, that the dataset does not include information on whether there was activity at the friction point during a cycle. Considering the situation when activity is present may result in other conclusions. • More drivers selected the outside lane at the sites with longer Rf and sites with a higher number of signals within ½ mi. Fewer drivers selected the outside lane at sites with wider medians. All of these differences are very small and not of practical value. Driver Behaviors The driver behaviors recorded with the camera used to determine saturation flow rate (i.e., Camera 1) included • Did the vehicle make a U-turn? • Did the vehicle stop beyond the stop bar? • Did the vehicle back up? • Was there a conflict between pedestrian or bicyclist and a turning vehicle? • Was the driver distracted and slow to start the turn? The number of vehicles within a queue making a U-turn was considered as part of the saturation-flow-rate evaluation. The saturation-flow-rate analyses clearly showed that the number of vehicles making a U-turn affects the operations at the signalized intersection (see Table 5-11). For this dataset, no conflicts between a pedestrian or bi- cyclist and a left-turning vehicle were observed. A vehicle passing the stop bar and then backing up to be behind the stop bar was also not observed. A few vehicles were observed stopping beyond the stop bar or being slow to start the turn. Only two of the 10,023 lead vehicles were observed stopping over the stop bar and less than 1% of the fifth to tenth vehicles in a queue were slow to start. Once the queue started to flow, very few left-turn drivers were distracted, in the opinion of the technicians reducing the data. Driver behaviors after the turn had started and around the friction point had more interesting data. Figure 5-3 shows the subareas used when reducing driver behavior data from Camera 3. Table 5-15 summarizes the number of driver behaviors observed. For the 18 sites included in this review, the most common driver behavior recorded was lane changes before the friction location, followed by lane changes at the friction location. Table 5-16 lists the sites by number of driver behaviors for those sites where more than 60 driver behaviors were observed. The three sites with the highest number of driver behaviors all had a driveway or intersection within 450 ft of Site RT LT PT DT BT IB SB DF CF PF DQ Total AZ-FS-03 7 2 94 1 104 AZ-FS-04 64 7 16 18 105 AZ-FS-05 21 21 AZ-FS-06 43 1 4 9 57 AZ-PH-02 52 3 55 AZ-PH-06 6 13 19 AZ-PH-07 3 1 4 AZ-PH-08 8 18 26 AZ-TU-09 1 66 70 72 2 211 AZ-TU-10 1 11 7 19 CA-BA-04 7 1 17 3 5 2 35 CA-ST-01 2 1 7 2 14 13 7 3 49 CA-ST-02 5 8 4 1 1 9 28 CA-ST-04 1 1 1 3 TX-CS-02 109 1 1 9 8 128 TX-CS-03 46 46 TX-CS-04 4 1 5 34 44 TX-HO-03 6 5 2 4 17 Total 8 0 1 20 399 17 115 120 267 7 18 972 Note: RT = conflict with right-turning traffic (opposing traffic); LT = conflict with opposing left-turning traffic; PT = conflict with ped/bike at the crosswalk of the cross street; DT = stopping/sudden deceleration after turning left; BT = lane change before friction location; IB = left-turn into improper lane between the intersection & friction location; SB = low speed between the intersection & friction location; DF = stopping/sudden deceleration at friction location; CF = lane change at friction location; PF = conflict with ped/bike/traffic exiting or entering the driveway (back of queue); DQ = stopping/sudden deceleration after turning left due to queue downstream on cross street. Blank cells = 0. Table 5-15. Number of driver behaviors observed within 2 hr at each site.

76 the intersection. Several other study sites also had a drive- way or intersection near the intersection but did not have as many driver behaviors. While the presence of the downstream driveway or intersection can be associated with multiple drive behaviors of interest, it is not always the case. There were some lane changes before the friction point at all study sites that had a downstream driveway/intersection. This finding offers a caution that downstream driveways and intersections are associated with greater activity that can affect the operations of the roadway. The addition of a lane on the major street from a cross-street right-turn lane is also associated with increased lane chang- ing behaviors. Lane changes into and out of this additional lane were observed at several sites. While this additional lane is generally intended to provide a free-flow opportunity for the cross-street right-turning vehicles, at several of the study sites, left-turning drivers were observed to move directly into this lane, in some cases, over a solid white pavement marking. The behaviors that reflected conflicts (RT, LT, PT, DT) along with potential conflicts—DT (stopping/sudden decel- eration after turning left) and IB (left-turn into improper lane between the intersection and friction location)—were reviewed to identify if there were common elements. The three sites with the largest number of these types of behaviors— CA-ST-01, CA-ST-02, and TX-CS-04—all had a friction point within 150 ft of the intersection. Site Total Comments R:Friction Distance to Friction (ft) AZ- TU-09 211 Many of the driver behaviors were right turns into the driveway. One vehicle could have multiple driver behaviors (BT, SB, DF). Driveway/ Intersection 365 TX- CS-02 128 Most of the lane changes were into a left-turn lane for an entrance across from the friction point. Driveway/ Intersection 414 AZ-FS- 04 105 Similar to TX-CS-02, many lane changes into a left- turn lane that is near the friction point. Driveway/ Intersection 234 AZ-FS- 03 104 Lane changes into and out of the lane added by the right-turn channelization were most of the driver behaviors observed. Channelized RTL 33 Table 5-16. Sites with the highest number of driver behaviors.