Assessing Interactions Between Access Management Treatments and Multimodal Users (2018)

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.

227 A P P E N D I X F : D A T A S U M M A R Y Data Summary Introduction This appendix summarizes the data that were collected for the study of three access management (AM) techniques: (1) driveway design, (2) right-turn deceleration, and (3) TWLTL vs. non-traversable median. The final study design for each of the three techniques is documented in Appendix E. The final study design outlines the modal focus, study objectives, analysis scale, data sources, site characteristics, sample size, and data collection methods needed to develop quantitative relationships for predicting specific performance measures. The data are summarized in this appendix on a technique-by-technique basis. The first section to follow summarizes the data collected for the driveway design technique. The second section summarizes the data for the right-turn deceleration technique. The third section summarizes the data for the TWLTL vs. non- traversable median technique. The fourth section summarizes the data used for the VISSIM Simulation Model Calibration. The last section summarizes the data used for the SSAM Simulation Model Calibration. Driveway Design This section summarizes the data collected for the driveway design study. The objective and scope of the study are provided in Appendix E (in a section having the same title as this section). The data used for this study were obtained from field measurements and reported crash data. The summary includes a description of the sites at which data were collected, the database organization, data collection techniques, data reduction procedures, and statistics used to describe the collected data. Study Design Challenges To be successful, this study would need a sufficiently large number of events where pedestrians, bicycles, trucks, cars) interact such that it has a negative impact on the performance of one (or both) travel modes. Several hundred such interactions would likely need to be represented in a database to ensure that statistically valid conclusions could be reached regarding event cause and effect. Given the typically low volume of pedestrians, bicycles, trucks, and cars at driveways, this study was initially identified as having a relatively low probability of success (see the discussion associated with Table 72). Simulation was considered as a cost-effective means of obtaining the desired sample size, but this option was abandoned because further investigation indicated that existing simulation programs do not simulate multimodal interactions at driveways in a highly reliable manner. Based on panel guidance, the study design for this technique was revised such that (1) field data would be used for the operations study component, (2) field-measured conflict data and crash data would be used for the safety study component, and (3) the study would examine only the pedestrian and bicycle travel modes. These changes were made to improve the probability of success for the overall study (see the discussion associated with Table 83).

228 Operations and Safety Study Study Sites This section describes the study sites for the driveway design study. The data collected at these sites were intended to be used to develop relationships describing the effect of driveway design on pedestrian and bicycle performance. A study site is defined to be one street-driveway intersection. The study sites selected for study are identified in Table 119. Twenty sites are identified: five sites in each of four states. The sites are collectively located in suburban and urban areas. Sites in a central business district (CBD) were not included because the associated driveways tended to have unique design characteristics (e.g., substantially wider sidewalks), relative to driveways outside of the CBD. Table 119. Study site locations – driveway design. Nbr. Region City Location Nearest Activity Generator 1 Oregon Beaverton 10600 SW Canyon Road New Car Dealership 2 Troutdale 25101 SE Stark Street Big Box Shopping Center 3 Gresham 18415 SE Division Street Strip Commercial 4 Gresham 24090 SE Stark Street Public medical services 5 Gresham 2150 NE Division Street Public medical services 6 Florida Tampa 1520 W Kennedy Boulevard Fast-food restaurant 7 Tampa 6505 S Dale Mabry Highway Fast-food restaurant 8 Tampa 717 S Howard Avenue Strip Commercial 9 Ft. Lauderdale 700 W Broward Boulevard Pharmacy store 10 Tampa 1601 W Kennedy Boulevard Big Box Shopping Center 11 New Jersey Glen Ridge 985 Bloomfield Avenue Strip Commercial 12 Wayne 1059 Hamburg Pike Strip Commercial 13 Belleville 46 Washington Avenue Variety store 14 Belleville 674 Washington Avenue Big Box Shopping Center 15 Magnolia 181 S White Horse Pike Big Box Shopping Center 16 Wisconsin Middleton 6625 Century Avenue Sit-down restaurant 17 Madison 4500 University Avenue Fast-food restaurant 18 Middleton 2405 Allen Boulevard Strip Commercial 19 Novanta 8422 Old Sauk Road Strip Commercial 20 Middleton 4900 Century Avenue Apartment complex Table 120 provides some additional descriptive information about each of the study sites. These characteristics were used during the site selection process. In general, a desired goal of the site selection process was that the collective set of sites included a range of values for each characteristic. The values in this table indicate that this goal was achieved. An additional site selection criterion was based on volume. It specified the need for minimum pedestrian, bicycle, and motorized vehicle volumes (i.e., 20 pedestrians per hour, 20 bicycles per hour, and 250 vehicles per hour). The minimum vehicle volume pertained to the driveway volume. None of the agencies contacted had pedestrian, bicycle, and vehicle count data for driveways in their jurisdictions. As a result, the adequacy of a site’s ability to meet the minimum volumes was based on a review of the adjacent land use and the personal knowledge of local agency contacts.

229 Table 120. Study site characteristics – driveway design. Nbr. Bike Lane Bus Route On- Street Parking Left-Turn Prohibition Curb Transition (width or radius)1 Driveway Width, ft Major Street Lanes Throat Length, ft 1 No Yes No No PE 33 4 25 2 Yes Yes No No PE 48 4 172 3 Yes Yes No No PE 35 4 28 4 Yes Yes No No PE 36 4 27 5 Yes Yes Yes No PE 30 4 45 6 No Yes No No CR (10') 28 4 48 7 Yes Yes No Yes FT (3') 28 4 32 8 No Yes Yes No CR (5') 24 2 38 9 Yes Yes No Yes CR (15') 28 6 37 10 No Yes No No CR (45') 36 4 124 11 No Yes No No FT (20') 22 4 40 12 No Yes No Yes CR (15’) 50 4 34 13 No Yes Yes No PE 24 4 68 14 No Yes No No CR (10') 30 4 40 15 No Yes No Yes CR (35') 30 4 200 16 No Yes No No FT (5') 28 4 20 17 Yes Yes No Yes FT (10') 42 6 32 18 Yes Yes No No CR (25') 32 4 27 19 Yes Yes No No FT (10') 28 4 58 20 No No No No FT - 7' 30 4 35 Note: 1 - PE: perpendicular edge; FT – flare or taper (width in feet); CR – curved radius (radius in feet). Database Organization This section describes the individual data elements (i.e., variables) in the various databases needed to describe the study sites and their performance. The database categories needed for this study are listed in Table 121. The worksheet name wherein the data were recorded is also shown. Table 121. Databases – driveway design. Data Category Database Name Worksheet Name Road inventory data Site characteristics data Geometry Traffic characteristics data Volume Operations performance measure data Pedestrian delay data PedDelay Bicycle travel time data BikeTime Safety performance measure data Pedestrian conflict data PedConflict Bicycle conflict data BikeConflict Crash data Crash

230 Road Inventory Data This section describes the site characteristics and traffic characteristics data. With one exception, the site characteristics data were obtained from Google Earth. The traffic characteristics data were obtained from videotape recordings by the researchers. The one exception is AADT data for the major street. These data were obtained from the local transportation agency. The site characteristics data are listed in Table 122. These data were recorded in the “Geometry” worksheet. One row in this worksheet describes the data for one site. Table 122. Site characteristics data – driveway design. Category Data Element Variable Name General Site identification number. Used to link site-specific data in other databases link_ID Major street name street_name_sc Code to identify issues that make site unsuitable for analysis prob_flag_sc Text to describe observations or concerns notes_sc Latitude and longitude of intersection center lat_sc long_sc Design Driveway left-turn prohibition (i.e., prohibit only left turns out of driveway, or prohibit left turns into and out of driveway) left_prohib Number of driveway lanes (entering, exiting) lanes_ent lanes_ext Width of driveway lanes (entering, exiting) lane_wid_ent lane_wid_ext Number of major street lanes (both directions) lanes_maj Median width on major street med_wid_maj Distance from major street curb line to near edge of sidewalk (at driveway) and width walk_setback walk_width Driver sight distance (to/from driveway) sight_left sight_right Presence and width of outside shoulder shldr_wid Number and width of on-street parking stalls within 250 ft. of the driveway parking_stall parking_wid Presence and width of marked bicycle lane bike_wid Curb transition geometry transition_typ transition_wid Presence of a triangular channelizing island island_ent island_ext Width of a non-traversable median on the driveway med_wid_dwy Presence of a right-turn bay on the major street rt_bay Distance to nearest downstream signalized intersection dwn_signal Distance to nearest upstream driveway (or intersection) up_access Distance to nearest downstream driveway (or intersection) dwn_access Distance from major street curb line to marked (or effective) driveway stop line (i.e., stop line setback) stop_setback Presence of horizontal curve in major street horiz_curve Presence of changes in major street vertical alignment through street–driveway intersection (e.g., for drainage) vert_curve Pavement condition rating for major street in vicinity of street–driveway intersection (see Chapter 17 of 2010 HCM) pvmt_cond

231 Category Data Element Variable Name Driveway grade in vicinity of sidewalk dwy_grade Driveway throat length len_throat Traffic control Speed limit of major street speed_lmt Presence of crosswalk markings on driveway crosswk_mrk Driveway traffic control (none, yield, stop, signal) dwy_cntl_LT dwy_cntl_R Traffic volume Major street annual average daily traffic (AADT) volume AADT2012 AADT2013 AADT2014 AADT2015 AADT2016 The traffic characteristics data are listed in Table 123. These data were recorded in the “Volume” worksheet. One row in this worksheet describes the data for one 15-minute period at one site. These data were collected for all five hours of videotape recorded at each site. Table 123. Traffic characteristics data – driveway design. Category Data Element Variable Name General Site identification number. Used to link site-specific data in other databases link_ID Major street name street_name_tc Code to identify issues that make site unsuitable for analysis prob_flag_tc Text to describe observations or concerns notes_tc Traffic characteristics Date of count cnt_day_tc cnt_month_tc Time at start of 15-minute count period start_time_h_tc start_time_m_tc start_time_s_tc Count of light vehicles (passenger cars, pickup trucks, and motorcycles) nb_left_pc nb_thru_pc nb_right_pc sb_left_pc sb_thru_pc sb_right_pc eb_left_pc eb_thru_pc eb_right_pc wb_left_pc wb_thru_pc wb_right_pc Count of heavy vehicles (including trucks, vehicles pulling trailers, excluding pickup trucks) nb_left_tk nb_thru_tk nb_right_tk sb_left_tk sb_thru_tk sb_right_tk eb_left_tk eb_thru_tk

232 Category Data Element Variable Name eb_right_tk wb_left_tk wb_thru_tk wb_right_tk Cardinal direction associated with major street lane nearest to driveway maj_dir Pedestrian count crossing driveway (each direction counted separately) ped_vol_same ped_vol_opp Bicycle count on major street bike_vol_same bike_vol_swalk Count of parked vehicles on major street parking_cnt Operations Performance Measure Data This section describes the raw data used to compute pedestrian delay and bicycle travel time. These data are listed in Table 124. They were recorded in the “PedDelay” worksheet or “BikeTime” worksheet. One row in each worksheet describes the data for one pedestrian (or bicycle) at one site. These data were collected using three of the five hours of videotape recorded at each site. For a given site, performance measure data were not extracted from the videotape if the previously extracted traffic characteristics data indicated that the site did not meet the required minimum volume levels. Table 124. Operations performance measure data – driveway design. Category Data Element Variable Name General Site identification number. Used to link site-specific data in other databases link_ID Major street name street_name_pd Code to identify issues that make site unsuitable for analysis prob_flag_pd Text to describe observations or concerns notes_pd Distance between reference marks on sidewalk travel_dist_pd Pedestrian delay data Date of count cnt_day_pd cnt_month_pd Time at start of 15-minute count period start_time_h_pd start_time_m_pd start_time_s_pd Pedestrian number ped_nbr Time crossing reference mark 1 xing1_time_h xing1_time_m xing1_time_s Time crossing reference mark 2 xing2_time_h xing2_time_m xing2_time_s Bicycle travel time data Date of count cnt_day_bk cnt_month_bk Time at start of 15-minute count period start_time_h_bk start_time_m_bk start_time_s_bk

233 Category Data Element Variable Name Bicycle number bike_nbr Time crossing reference mark 1 bik1_time_h bik1_time_m bik1_time_s Time crossing reference mark 2 bik2_time_h bik2_time_m bik2_time_s Safety Performance Measure Data This section describes the pedestrian and bicycle safety performance data. Safety performance was quantified using both conflict data and crash data. The performance relationships were envisioned to be developed using conflict data. The crash data were used to quantify the relationship between conflict frequency and crash frequency. The raw data used to compute pedestrian and bicycle conflict frequency (and severity) are listed in Table 125. The conflict data were obtained from videotape recordings. Techniques for measuring and recording these data are the subject of discussion in the next section. These data were collected for five hours at each site. For a given site, performance measure data were not extracted from the videotape if the previously extracted traffic characteristics data indicated that the site did not meet the required minimum volume levels. Table 125. Pedestrian and bicycle conflict data – driveway design. Category Data Element Variable Name General Site identification number. Used to link site-specific data in other databases link_ID Major street name street_name_pv Code to identify issues that make site unsuitable for analysis prob_flag_pv Text to describe observations or concerns notes_pv Pedestrian conflict characteristics Date of count cnt_day_pv cnt_month_pv Time pedestrian begins crossing driveway entrance lane (or lanes) xing_beg_h xing_beg_m xing_beg_s Time pedestrian ends crossing driveway entrance lane (or lanes) xing_end_h xing_end_m xing_end_s Direction of pedestrian travel ped_dir Direction of conflicting vehicle travel veh_dir_pv Data to estimate time to collision TTC_dwy Severity of vehicle evasive action SEA_veh_pv Severity of pedestrian evasive action SEA_ped Complexity of vehicle evasive action CEA_veh_pv Complexity of pedestrian evasive action CEA_ped Distance to collision DTC_pv

234 Category Data Element Variable Name Bicycle conflict characteristics Date of count cnt_day_bv cnt_month_bv Time bicycle begins approaching driveway bike_beg_h bike_beg_m bike_beg_s Time bicycle ends crossing driveway bike_end_h bike_end_m bike_end_s Direction of bicycle travel bike_dir Direction of conflicting vehicle travel veh_dir_bv Severity of vehicle evasive action SEA_veh_bv Severity of bicycle evasive action SEA_bike Complexity of vehicle evasive action CEA_veh_bv Complexity of bicycle evasive action CEA_bike Distance to collision DTC_bv The crash database attributes are listed in Table 126. These data were recorded in the “Crash” worksheet. One row in this worksheet describes the data for one bicycle-related or pedestrian-related crash at one site. These data correspond to the most recent five years at each site. Table 126. Crash database – driveway design. Category Data Element Variable Name General Site identification number. Used to link site-specific data in other databases link_ID Major street name street_name_cr Nearest intersecting street name cross_name_cr Distance to nearest intersecting street cross_dist_cr Crash location coordinates lat_cr long_cr Source of original crash data source_cr County of crash county_cr Date of crash minute_cr hour_cr day_cr month_cr year_cr Milepost of crash milepost_cr Crash description First harmful event harm1_cr Crash location relative to other junctions location_cr Work zone related workzone_cr Vehicle maneuver prior to crash maneuver_cr Crash severity severity_cr Number of pedestrians involved in crash nbr_ped_cr Number of bicyclists involved in crash nbr_bike_cr

235 Data Collection Techniques This section describes the techniques used to collect and process individual data elements. The focus of the discussion is on techniques used to obtain the raw data included in each database. Details of the data collection process are provided in Appendix E. The traffic characteristic data and performance measure data were collected using two video recorders on opposite sides of (and facing) the subject driveway. The two cameras were positioned as shown in Figure 6 of Appendix E. One camera was used to observe traffic on the subject driveway leg of the intersection. The second camera was used to observe traffic on the adjacent major street approach. Both cameras were used to observe pedestrians crossing the driveway and bicycles traveling along the major street. The desired data were then extracted from the videotape recordings during replay in the office. The video recorders were used to record traffic conditions at each site during each of five one-hour study periods. Each one-hour study period occurred during a time period for which the pedestrian and bicycle delay was at its highest level during the typical day. This time period coincided with the hours of peak pedestrian volume, peak bicycle volume, and moderate-to-high motorized vehicle volume. The study periods occurred during typical weekday daylight time periods. No data were collected during inclement weather, during nighttime hours, or when traffic flow was disrupted by incidents, work zone activity, or special events. Data Reduction Procedures This section describes the procedures used to reduce the raw data elements. The focus of the discussion is on procedures used to convert raw data into the desired database elements. Road Inventory and Traffic Characteristics Data The road inventory data were obtained from Google Earth aerial imagery. The Historical Imagery view in Google Earth was used to briefly review all available photos during the years 2012 to 2016 to determine if construction occurred at the subject driveway. Measurements were made using the Ruler tool. The AADT was obtained from the agency that operates and maintains the major street. All other traffic characteristics data were extracted from the videotape recorded at each site. Operations Performance Measure Data Pedestrian delay was computed using the reference mark crossing times recorded in the pedestrian delay database. The videotape recording was processed in consecutive 15-minute intervals. These intervals were the same as those used for the traffic characteristics data (described in the previous section). During each 15-minute count period, raw data were recorded for each pedestrian that crossed the subject driveway (in either travel direction) (see “subject pedestrian travel path” in Figure 5 of Appendix E). During a given 15-minute count period, data were recorded for the first ten pedestrians observed to cross both reference marks. Any additional pedestrians that crossed during the remainder of the count period were ignored. Once data were collected for ten pedestrians in a given count period, the video was advanced to the start of the next count period and the process repeated. In this manner, data for ten pedestrians were recorded for each 15-minute period. Bicycle travel time was computed using the reference mark crossing times recorded in the bicycle travel time database. The videotape recording was processed in consecutive 15-minute intervals. These intervals were the same as those used for the traffic characteristics data (described in a previous section). During each 15-minute count period, raw data were recorded for each bicycle that traveled on the major street through the street-driveway intersection (see “subject bicycle travel path” in Figure 5 of Appendix E). During a given 15-minute count period, data were recorded for all bicycles observed to cross both reference marks.

236 Safety Performance Measure Data Vehicle-Pedestrian Conflicts. The study of vehicle-pedestrian conflicts was based on the technique developed by Kaparias et al. (2010). Their technique used the following definition of a conflict, “[a conflict is] an observational situation in which two or more road users approach each other in space and time to such an extent that a collision is imminent if their movements remain unchanged.” Based on the aforementioned definition, a vehicle-pedestrian conflict was defined to occur when two events took place. First, a pedestrian begins to cross the driveway. Second, while the pedestrian is crossing, a vehicle enters the driveway by turning right from the major street, turning left from the major street, or crossing the major street as a through vehicle. One row in the worksheet was used to describe the data associated with these two events. Each pedestrian that is conflicted by a vehicle represents one conflict, regardless of whether the conflicting vehicle is the same for several pedestrians. The data for each conflicted pedestrian was allocated one row in the worksheet. Thus, if two pedestrians were conflicted by the same vehicle, then this situation represented two conflicts. Two rows in the worksheet are used to describe these two conflicts (one row for each involved pedestrian). On the other hand, if a given pedestrian is conflicted by one vehicle and then by a second vehicle before the crossing is completed, only data for the first conflict was entered into the worksheet. Conflicts between bicycles and pedestrians were not the focus of this study. The driveways selected for this study require that drivers on the driveway exit lanes have a legal obligation to yield the right-of-way to major street vehicles before entering the major street. With this type of control, vehicle-pedestrian conflicts were negligible on the driveway exit lanes. For this reason, the study did not focus on pedestrians crossing the driveway exit lanes. During videotape review, the tape was advanced to the point in time where a pedestrian began crossing the driveway entrance lane (or lanes). This pedestrian was referred to as the “subject pedestrian.” Then, the tape was advanced slowly while the subject pedestrian crossed the driveway entrance lane (or lanes). During this time period, the following three movements were monitored: 1. Vehicle approaches the driveway by turning right from the major street, 2. Vehicle approaches the driveway by turning left from the major street, and 3. Vehicle approaches the driveway by crossing the major street as a through vehicle. If one or more of the three movements was observed to approach the driveway with the intent to enter the driveway while the subject pedestrian was crossing, then a conflict was said to have occurred. The conflicting vehicle was referred to as the “subject vehicle.” If there were two or three movements observed to approach the driveway while the subject pedestrian was crossing, then the vehicle that was projected to reach the driveway first was considered the subject vehicle. The data listed in Table 125 were then recorded for the subject pedestrian and subject vehicle. Vehicle-Bicycle Conflicts. The study of vehicle-bicycle conflicts was based on the vehicle-pedestrian technique developed by Kaparias et al. (2010). This technique was extended to vehicle-bicycle conflicts using some of the bicycle behaviors and actions developed Hunter et al. (1998). Based on the aforementioned definitions, a vehicle-bicycle conflict was defined to occur when two events took place. First, a bicycle begins to approach the driveway while traveling in the major street. Second, while the bicycle is approaching and then crossing, a vehicle enters or exits the driveway. One row in the worksheet was used to describe the data associated with these two events. Each bicycle that was conflicted by a vehicle represents one conflict, regardless of whether the conflicting vehicle is the same for several bicyclists. The data for each conflicted bicycle was allocated one row in the worksheet. Thus, if two pedestrians were conflicted by the same vehicle, then this situation

237 represented two conflicts. Two rows in the worksheet were used to describe these two conflicts (one row for each involved bicycle). On the other hand, if a given bicycle was conflicted by one vehicle and then by a second vehicle before the crossing is completed, only data for the first conflict was entered into the worksheet. Conflicts between bicycles and pedestrians were not the focus of this study. Similarly, bicycle-bicycle conflicts and bicycles traveling on the sidewalk were not the focus of this study. The major street was divided in half by its centerline. Bicycles traveling on a path that is in the half nearest to the driveway were of interest to this study. Bicycles traveling on the half that is furthest from the driveway were not the focus of this study. Bicycles that turn left or right at the driveway were not the focus of this study. During videotape review, the tape was advanced to the point in time where a bicycle began to approach the driveway. This bicycle was referred to as the “subject bicycle.” Then, the tape was advanced slowly while the subject bicycle was approaching and then crossing the driveway. During this time period, the following movements were monitored: 1. Vehicle approaches the driveway by turning right from the major street, 2. Vehicle approaches the driveway by turning left from the major street, 3. Vehicle approaches the driveway by crossing the major street as a through vehicle, 4. Vehicle exits the driveway by turning right, 5. Vehicle exits the driveway by turning left, and 6. Vehicle exits the driveway by crossing the major street as a through vehicle. If one or more of the six movements was observed to approach/exit the driveway while the subject bicycle is approaching or crossing, then a conflict was said to have occurred. The conflicting vehicle was referred to as the “subject vehicle.” If there were two or more movements observed to approach/exit the driveway while the subject bicycle was crossing, then the vehicle that was projected to reach the driveway first was considered the subject vehicle. The data listed in Table 125 were then recorded for the subject bicycle and subject vehicle. Crash Data. Electronic copies of the crash reports (including fatal, injury, and property-damage-only crashes) for the major street associated with each study site were requested from the local agency. The crash data requested described crashes that occurred at or within 50 feet of the street-driveway intersection. The original electronic crash data obtained from an agency were kept in separate files. The crash data from these files was entered into the Crash worksheet using the variables described in Table 126. The variables in the original crash data sometimes had to be converted to be consistent with the variable definitions developed for this project. Upon receipt, the original agency data was screened to identify only crashes where the number of pedestrians involved (nbr_ped_cr) or the number of bicycles involved (nbr_bike_cr) equaled or exceeded one. All bicycle-related and pedestrian-related crashes identified using this screening technique were entered into the Crash worksheet. Database Summary Video recordings were made for each of the 20 sites identified in Table 119. As a first step in the data reduction process, traffic volumes were extracted from the video tapes. The volumes obtained in this manner are listed in Table 127. Inspection of these volumes indicated that none of the sites satisfied the minimum volume criteria (i.e., 20 pedestrians per hour, 20 bicycles per hour, and 250 vehicles per hour on the driveway). With this realization, it was determined that it would not be a cost-effective use of

238 resources to proceed with the extraction of performance measure data from the video recordings (as described in the previous sections). Table 127. Site volume levels – driveway design. Nbr. Driveway Volume Major Street Volume, veh/h Pedestrian, p/h Bicycle, b/h Vehicle, veh/h Approaching Driveway from Left Approaching Driveway from Right 1 2.7 0.0 26 1021 689 2 11.7 1.0 16 512 722 3 3.3 3.0 51 900 1065 4 7.3 1.7 25 686 579 5 6.3 3.7 28 666 582 6 3.3 2.0 51 1278 1311 7 0.3 0.3 34 445 917 8 17.3 1.3 29 536 478 9 5.7 1.7 12 1498 1921 10 3.0 1.7 64 1349 1377 11 8.7 1.7 49 750 882 12 2.7 0.0 20 850 1026 13 47.3 3.3 26 542 568 14 9.7 0.7 28 508 474 15 3.7 0.0 36 985 1131 16 3.3 0.0 8 716 563 17 1.0 0.0 60 1613 1699 18 1.0 1.3 40 1079 605 19 4.0 2.7 66 687 626 20 0.3 0.0 25 558 593 After some additional site review and volume checks, it was determined that only driveways in downtown central business districts or those near the center of an academic institution (e.g., University of Wisconsin - Madison) would have sufficient pedestrian and vehicle volumes to satisfy the minimums established for this project. However, it was generally found that driveways near academic institutions with a high volume of bicycles do not have a high number of pedestrians (and vice versa) due to due to differences in travel length and the availability of alternative modes. As such, the pool of candidate high- volume driveways tends to narrow to those associated with parking garages and other urban core primary access points. These driveways have unique design characteristics (e.g., substantially wider sidewalks) that are inconsistent with traditional suburban driveways. Based on this information, the researchers suspended activities on the driveway design study. Right-Turn Deceleration This section summarizes the data collected for the right-turn deceleration study. The objective and scope of the study are provided in Appendix E (in a section having the same title as this section). The data used for this study were obtained from a traffic simulation model (i.e., VISSIM). The summary includes a

239 description of the simulation test beds, the database organization, data collection techniques, data reduction procedures, and statistics used to describe the collected data. Operations Study Simulation data were used to evaluate the operational effects of right-turn lane presence at a signalized intersection. A simulation testbed was established using intersections without right-turn deceleration lanes to represent the “before” condition. A second testbed was based on the before-condition intersections with a right-turn deceleration lane to represent the “after” condition. The data describe the operational performance of bicycles, transit vehicles, and trucks, as influenced by right-turn lane presence. The VISSIM simulation model was used to generate the performance data. This model has the capability of simulating cars, pedestrians, bicycles, transit, and trucks. The data used to calibrate this model are summarized in a later section titled VISSIM Simulation Model Calibration. Test Bed Characteristics The test bed was used to develop a large number of realistic combinations of intersection design (i.e., scenarios). This test bed (RT6) for the right-turn deceleration (RTD) study was created for the VISSIM model. The simulation test beds were based on the following two sites:  FL4: Eastbound lanes of East Hillsborough Avenue @ North Florida Avenue, Tampa, Florida  OR5: Westbound lanes of SE Killingsworth Street, Portland, Oregon A test bed without the presence of a right-turn deceleration lane was used to establish base performance conditions. This test bed is called the “base” test bed. Its characteristics are identified in Table 111 (of Appendix E). The “right-turn lane” test bed data elements that were evaluated are shown in Table 112 (of Appendix E). A right-turn deceleration lane was added to the base test bed to create the right-turn lane test bed. All total, 5,184 scenarios (= 1,296 base + 3,888 right-turn lane) were evaluated using simulation. Database Organization A database was produced in an Excel worksheet to compile the outputs of the simulation runs. The database contains simulation information (e.g., simulation runs), input variables (scenario factors), and operations performance measures. Each observation (row) in the database represents the results of a one- hour simulation run. Input values are constant for each scenario. Operations performance measures represent an average value for the one-hour simulation run. Table 128 describes the data fields in the database. Table 128. Data fields in output database for operations study – right-turn deceleration. Category Data Field Description Code or Value (Unit) Input variables treat AM treatment indicator 0 – No right-turn deceleration (RTD) Lane 100 – RTD at 100 ft. 200 – RTD at 200 ft. 300 – RTD at 300 ft. signal_cycle Signal cycle length indicator 1 – 100; 2 – 150 (seconds) veh_vol Through vehicle inputs 450; 550; 650 (per hour per lane)

240 Category Data Field Description Code or Value (Unit) bike_vol Through bicycle inputs 0; 30; 60 (per hour) truck_per Truck percentage 3%; 6% rtor Right turn on red 1 – Yes; 2 – No bus_stop Bus stop frequency 0; 1; 2 bus_dwell Bus dwell time 15; 30 (seconds) Simulation information sim_run Simulation run index 1 – the first run; 2 – the second run Computed performance measures1 bike_delay Average bicycle delay Numeric (seconds per bicycle) truck_delay Average delay of through and right- turn trucks Numeric (seconds per truck) bus_delay Average bus delay Numeric (seconds per bus) Traffic outputs2 vol_tot Traffic volume (cars, trucks, and buses) = sum of left-turn, through and right-turn volume Numeric (vehicles per hour) vol_right Right-turn volume (cars, trucks, and buses) Numeric (vehicles per hour) pr_right Proportion right turns = right-turn volume ÷ traffic volume Numeric (dimensionless) vol_truck Truck volume (left, right, and through) Numeric (trucks per hour) pr_truck Proportion trucks = truck volume ÷ traffic volume Numeric (dimensionless) vol_bike Bicycle volume (through only) Numeric (bicycles per hour) Notes: 1 – Collected from travel time measurements. 2 – Collected from data collection points designated in simulation test bed. Data Collection Process The procedure for simulation and data collection is described by the following steps. Step 1. Create Scenarios Basic Characteristics. A simulation model was created in VISSIM using the basic characteristics shown in Table 111. Four different right-turn deceleration lane lengths (i.e., 0, 100, 200, and 300 ft) were configured on the eastbound approach of the test bed, respectively, to produce four second-level test beds. Evaluation Measures. Two travel time measurement sections (at a total length of 1,150 ft) were coded for the eastbound approach as the study boundaries: one for through movements and another for right-turn movements. The measurement sections are illustrated in Figure 10. Scenarios. In addition to the four right-turn deceleration lane lengths, two scenario factors were coded manually for signal cycle length and for right-turn-on-red (RTOR) operation. A total of 16 testbeds (4 lane length levels × 2 signal cycle lengths × 2 RTOR levels) were produced. A MATLAB program was developed to create simulation scenarios with the 16 testbeds based on the remaining five scenario factors. Finally, a total of 5,184 simulation scenarios were created. Step 2. Check Errors The research team randomly selected 10 scenarios to review the simulation scenarios for possible errors. The error checking included: (1) a review of the animations to identify visible errors (e.g., geometric errors, movements, signal sequence, etc.), (2) verification of vehicle inputs and turning movements, and (3) validation of performance measure outputs. All identified errors were corrected.

241 Figure 10. Plan view of intersection test bed for travel time measurement − right-turn deceleration. Step 3. Run Simulations The MATLAB program executed the 5,184 simulation scenarios in VISSIM. For each scenario, two simulation runs were conducted to account for the microsimulation randomness. A 15-minute warm-up period plus a 60-minute simulation period was applied in each simulation run. On a Dell Precision T7600 Workstation (Intel Xeon E5-2680 2.7GHz), one simulation run was required approximately 140 to 210 seconds. To reduce the simulation time, parallel computing technologies were used. Specifically, four simulation threads were launched simultaneously through the MATLAB codes. Step 4. Read Data from Simulation Results and Export to Database After completing the simulations, the MATLAB program retrieved operations performance measures from the simulation model and exported the results to a database using a CSV format. Traffic outputs were collected at the traffic data collection points shown in Figure 10. The database data fields are defined in Table 128. Database Summary Site Characteristics The basic geometric and traffic control characteristics of the test bed (e.g., location, through lanes, volume, etc.) are summarized in Table 111.

242 Operations Performance Measures A total of 10,368 observations were produced from the simulations (= 5,184 scenarios × 2 simulation runs per scenario). Of this total, 6912 observations had a non-zero bicycle volume. The key operations measures include:  Average bicycle delay (seconds per bicycle)  Average bus delay (seconds per bus)  Average truck delay (seconds per truck) The summary statistics for these operations measures are shown in Table 129 to Table 131. Table 129. Summary statistics for average bicycle delay – right-turn deceleration. Factors Obs. Average Bicycle Delay (seconds per bicycle) Mean Std. Minimum Maximum Overall 6912 21.43 4.85 14.44 29.37 Right-turn deceleration lane length (feet) 0 1728 21.71 4.90 14.45 29.37 100 1728 21.34 4.83 14.46 26.76 200 1728 21.34 4.83 14.46 26.75 300 1728 21.34 4.84 14.44 26.76 Signal cycle length (seconds) 100 3456 16.73 1.53 14.44 19.71 150 3456 26.14 0.66 25.14 29.37 Right-turn-on-red Yes 3456 21.43 4.85 14.45 29.37 No 3456 21.43 4.85 14.44 29.26 Traffic volume (vphpl) 450 2304 21.41 4.85 14.44 29.26 550 2304 21.43 4.84 14.44 29.00 650 2304 21.46 4.86 14.44 29.37 Right-turn percentage 2% 2304 21.34 4.82 14.44 26.85 10% 2304 21.44 4.87 14.44 28.88 20% 2304 21.52 4.87 14.44 29.37 Bicycle volume (bph) 0 - - - - - 30 3456 21.75 4.16 16.50 29.37 60 3456 21.11 5.44 14.44 28.08 Truck percentage 3% 3456 21.44 4.86 14.44 29.37 6% 3456 21.42 4.85 14.44 29.26 Bus stop frequency (buses per hour) 0 2304 21.43 4.85 14.44 29.04 1 2304 21.43 4.86 14.44 29.37 2 2304 21.43 4.85 14.44 28.91 Bus dwell time (seconds) 15 3456 21.43 4.85 14.44 29.37 30 3456 21.43 4.85 14.44 29.04 Note: “Std.” – standard deviation

243 Table 130. Summary statistics for average bus delay – right-turn deceleration. Factors Obs. Average Bus Delay (seconds per bus) Mean Std. Minimum Maximum Overall 6912 62.78 29.23 19.54 179.28 Right turn deceleration lane length (feet) 0 1728 77.89 26.79 19.59 179.28 100 1728 66.12 28.94 19.58 171.26 200 1728 54.59 27.31 19.54 153.22 300 1728 52.53 26.58 19.54 134.17 Signal cycle length (seconds) 100 3456 53.09 23.10 19.54 124.07 150 3456 72.48 31.43 19.54 179.28 Right-turn-on-red Yes 3456 61.42 29.49 19.54 176.79 No 3456 64.15 28.91 19.54 179.28 Traffic volume (vphpl) 450 2304 57.38 26.37 19.54 149.39 550 2304 62.95 28.86 19.54 176.79 650 2304 68.03 31.29 19.54 179.28 Right turn percentage 2% 2304 62.74 29.01 19.54 179.28 10% 2304 61.75 29.73 19.54 177.42 20% 2304 63.86 28.92 19.54 171.26 Bicycle volume (bph) 0 2304 63.12 28.97 19.55 170.29 30 2304 62.41 28.27 19.54 176.79 60 2304 62.82 30.42 19.54 179.28 Truck percentage 3% 3456 62.17 29.22 19.54 179.28 6% 3456 63.40 29.23 19.54 171.26 Bus stop frequency (buses per hour) 0 - - - - - 1 3456 65.02 33.17 19.54 179.28 2 3456 60.55 24.47 20.09 156.14 Bus dwell time (seconds) 15 3456 61.02 26.64 19.54 156.14 30 3456 64.55 31.52 19.54 179.28 Table 131. Summary statistics for average truck delay – right-turn deceleration. Factors Obs. Average Truck Delay (seconds per truck) Mean Std. Minimum Maximum Overall 10,368 36.40 9.03 18.81 73.85 Right-turn deceleration lane length (feet) 0 2592 39.50 9.81 25.21 73.85 100 2592 36.16 8.62 20.00 54.32 200 2592 35.21 8.55 18.81 55.25 300 2592 34.72 8.29 19.87 51.78 Signal cycle length (seconds) 100 5184 28.64 3.27 18.81 49.47 150 5184 44.16 5.65 29.67 73.85 Right-turn-on-red Yes 5184 35.87 9.08 18.81 73.85 No 5184 36.93 8.95 22.50 72.70 Traffic volume (vphpl) 450 3456 34.02 5.98 18.81 45.99

244 550 3456 37.20 10.07 20.62 58.79 650 3456 37.97 9.94 21.41 73.85 Right-turn percentage 2% 3456 37.07 8.16 25.49 57.40 10% 3456 36.43 8.78 22.50 66.67 20% 3456 35.69 10.01 18.81 73.85 Bicycle volume (bph) 0 3456 36.35 9.01 18.82 72.70 30 3456 36.37 9.01 18.81 72.60 60 3456 36.47 9.07 19.02 73.85 Truck percentage 3% 5184 36.47 8.89 20.62 73.85 6% 5184 36.32 9.17 18.81 72.70 Bus stop frequency (buses per hour) 0 3456 36.19 8.87 18.82 62.42 1 3456 36.43 9.03 19.01 72.60 2 3456 36.57 9.19 18.81 73.85 Bus dwell time (seconds) 15 5184 36.35 8.99 18.81 73.85 30 5184 36.44 9.07 18.82 72.37 Note: “Std.” – standard deviation Safety Study Simulated conflict data were used to evaluate the safety effects of right-turn deceleration lane presence at a signalized intersection. The simulation testbed established for the operations-focused study was used for this purpose, as described in the previous section. The data describe the safety performance of transit vehicles and trucks, as influenced by right-turn lane presence. The VISSIM simulation model was used to generate vehicle trajectory data. These data were then post- processed using the Surrogate Safety Assessment Model (SSAM) to extract conflict and surrogate safety data. The data used to calibrate this model are summarized in a later section titled SSAM Simulation Model Calibration. Test Bed Characteristics The test bed for the right-turn deceleration safety study is the same as that for the operations study, which was built based on two approaches (legs) at two signalized intersections. The evaluation section was set on the eastbound approach to assure the consistency in all unstudied factors across scenarios. The characteristics associated with each scenario are shown in Table 111 and Table 112. Database Organization A database was produced in an Excel worksheet to compile the conflict outputs of the simulation runs. The database contains simulation information (e.g., simulation runs), input variables (scenario factors), and safety performance measures (conflicts). Each observation (i.e., row) in the database represents the results of a one-hour simulation run. Input values are constant for each scenario. Safety performance measures represent an average value for the one-hour simulation run. Table 132 describes the data fields of the safety database.

245 Table 132. Data fields in output database for safety study – right-turn deceleration. Category Data Field Description Code or Value (Unit) Input variable treat AM treatment indicator 0 – No right-turn deceleration (RTD) 100 – RTD at 100 ft. 200 – RTD at 200 ft. 300 – RTD at 300 ft. signal_cycle Signal cycle length indicator 1 – 100; 2 – 150 (seconds) veh_vol Through vehicle inputs 450; 550; 650 (per hour per lane) bike_vol Through bicycle inputs 0; 30; 60 (per hour) truck_per Truck percentage 3%; 6% rtor Right turn on red 1 – Yes; 2 – No bus_stop Bus stop frequency 0; 1; 2 bus_dwell Bus dwell time 15; 30 (seconds) Simulation information sim_run Simulation run index 1 – the first run; 2 – the second run Truck performance measures truck_re Truck rear-end conflicts Numeric (number of conflicts per hour) truck_lc Truck lane-change conflicts Numeric (number of conflicts per hour) Transit performance measures bus_re Bus rear-end conflicts Numeric (number of conflicts per hour) bus_lc Bus lane-change conflicts Numeric (number of conflicts per hour) Traffic outputs1 veh_vol_lt Left turn car volume Numeric (number of cars per hour) veh_vol_th Through car volume Numeric (number of cars per hour) veh_vol_rt Right turn car volume Numeric (number of cars per hour) truck_vol_lt Left turn truck volume Numeric (number of trucks per hour) truck_vol_th Through truck volume Numeric (number of trucks per hour) truck_vol_rt Right turn truck volume Numeric (number of trucks per hour) bus_vol_th Bus volume Numeric (number of buses per hour) bike_vol_th Bike volume Numeric (number of bicycles per hour) Note: 1 - Collected from data collection points designated in simulation test bed. Data Collection Process The procedure for the safety simulation and data collection is described in the following steps. Step 1. Run VISSIM Simulation The safety study shared the same VISSIM simulation results from the operations study, including scenario creation, error checking, and running of the simulation. Details on the data collection process can be found in the section describing the operations study. For the safety assessment, two direct output files were generated from each simulation replication. These two files are listed below.  *.trj – SSAM formatted vehicle trajectory file  *.fhz – Vehicle input file indicating vehicle type and associated vehicle identification number These two files were used to generate traffic conflict statistics for all runs using SSAM. Step 2. Identify Traffic Conflicts A MATLAB program was developed to identify traffic conflicts from the two VISSIM output files. The identification process is described as below:

246 SSAM Analysis. The Surrogate Safety Assessment Model (SSAM, version 3) was used to process each trajectory file (*.trj) to produce a list of conflict events for each simulation replication. The events that were produced are described in the following list.  tMinTTC: the simulation time when the minimum TTC (time-to-collision) value for this conflict was observed  FirstVID (SecondVID) : the vehicle identification number of the first (second) vehicle  FirstLink (SecondLink): a number indicating which link the first (second) vehicle is traveling on at tMinTTC  xFirstCSP (xSecondCSP): the x-coordinate of the first (second) vehicle at the conflict starting point (CSP)  yFirstCSP (ySecondCSP): the y-coordinate of the first (second) vehicle at the CSP  ConflictType: whether the conflict is the result of a rear end (|Conflicting Angle| < 30○), lane change, or crossing maneuver Vehicle Type Identification. The focus of the study is on two primary conflict types. These types are listed below:  Bus conflict: a conflict event in which at least one bus is involved  Truck conflict: a conflict event in which at least one truck is involved In addition, the vehicle type was identified and a data field for vehicle type was added in the conflict data table. The MATLAB program was used for this purpose. It matched the vehicle types defined in the vehicle input files (*.fhz) to the two vehicles involved in the produced traffic conflicts. Two new data fields were produced in the conflict output file. These variables are described in the following list.  FirstVType: the vehicle type of the first involved vehicle  SecondVType: the vehicle type of the second involved vehicle Conflict Filtering. The MATLAB program applied the following criteria to filter the conflicts produced by the SSAM model.  The simulation time of the identified conflict should be greater than the ending time of the warm- up period (tMinTTC > 900s)  The first vehicle or the second vehicle should be target vehicle types, bus or truck (FirstVType or SecondVType = ‘bus’ for bus conflicts, FirstVType or SecondVType = ‘truck’ for truck conflicts)  The conflict location is within the study boundary, as shown in Figure 11.

247 Figure 11. Measurement boundary for safety evaluation - right-turn deceleration. Step 3. Export to Database The MATLAB program matched the identified bus- and truck-related conflicts to the scenario factors associated with each simulation run. The conflict and scenario data were then exported to a database using the CSV format (using the data fields defined in Table 132). Database Summary Site Characteristics The characteristics associated with each scenario are shown in Table 111 and Table 112. Conflict Characteristics A total of 10,368 observations were produced from the simulations (= 5,184 scenarios × 2 simulation runs per scenario). Of this total, 6912 observations had a non-zero bus stop frequency. The key safety measures are identified in the following list.  Bus rear-end conflicts (per hour)  Bus lane-change conflicts (per hour)  Bus crossing conflicts (per hour)  Truck rear-end conflicts (per hour)  Truck lane-change conflicts (per hour)  Truck crossing conflicts (per hour) The summary statistics for these safety measures are shown in Table 133 and Table 134 for buses and trucks, respectively.

248 Table 133. Summary statistics for bus conflict frequency – right-turn deceleration. Factors Obs. Bus Conflict Frequency by Conflict Type (conflicts per hour) Rear End Lane Change Crossing Mean Std. Min. Max. Mean Std. Min. Max. Mean Std. Min. Max. Overall 6912 3.62 2.62 0 22 0.42 0.68 0 6 0.34 0.67 0 5 Right -turn deceleration lane length (feet) 0 1728 1.76 1.95 0 22 0.67 0.88 0 6 0.02 0.16 0 2 100 1728 4.77 2.73 0 15 0.47 0.67 0 4 0.55 0.78 0 5 200 1728 4.13 2.41 0 13 0.29 0.52 0 3 0.38 0.71 0 4 300 1728 3.81 2.30 0 13 0.24 0.48 0 2 0.41 0.71 0 4 Signal cycle length (seconds) 100 3456 3.64 2.59 0 22 0.41 0.66 0 5 0.28 0.56 0 5 150 3456 3.59 2.65 0 15 0.42 0.69 0 6 0.41 0.75 0 4 Right-turn-on-red Yes 3456 3.63 2.62 0 22 0.41 0.67 0 5 0.36 0.70 0 5 No 3456 3.60 2.62 0 14 0.42 0.68 0 6 0.32 0.63 0 4 Traffic volume (vphpl) 450 2304 3.34 2.60 0 22 0.38 0.65 0 6 0.32 0.63 0 4 550 2304 3.75 2.72 0 15 0.42 0.71 0 4 0.36 0.70 0 4 650 2304 3.76 2.51 0 15 0.44 0.66 0 5 0.34 0.67 0 5 Right turn percentage 2% 2304 3.67 2.57 0 22 0.49 0.75 0 6 0.36 0.67 0 4 10% 2304 3.52 2.57 0 14 0.40 0.66 0 4 0.34 0.68 0 5 20% 2304 3.67 2.71 0 15 0.36 0.60 0 4 0.33 0.65 0 4 Bicycle volume (bph) 0 2304 3.17 2.20 0 13 0.43 0.68 0 6 0.23 0.48 0 3 30 2304 3.46 2.46 0 22 0.40 0.66 0 5 0.31 0.64 0 4 60 2304 4.22 3.02 0 15 0.42 0.68 0 5 0.49 0.82 0 5 Truck percentage 3% 3456 3.60 2.62 0 15 0.41 0.67 0 5 0.33 0.66 0 5 6% 3456 3.63 2.62 0 22 0.42 0.68 0 6 0.35 0.68 0 4 Bus stop frequency (buses per hour) 1 3456 2.88 2.45 0 15 0.25 0.49 0 3 0.37 0.74 0 4 2 3456 4.35 2.58 0 22 0.58 0.79 0 6 0.32 0.59 0 5 Bus dwell time (seconds) 15 3456 3.55 2.60 0 15 0.40 0.64 0 4 0.32 0.64 0 5 30 3456 3.69 2.64 0 22 0.44 0.71 0 6 0.36 0.69 0 4 Note: “Std.” – standard deviation

249 Table 134. Summary statistics for truck conflict frequency – right-turn deceleration. Factors Obs. Truck Conflict Frequency by Conflict Type (conflicts per hour) Rear End Lane Change Crossing Mean Std. Min. Max. Mean Std. Min. Max. Mean Std. Min. Max. Overall 6912 24.57 11.17 4 66 4.28 2.67 0 15 0.11 0.34 0 3 Right -turn deceleration lane length (feet) 0 1728 24.46 10.33 7 53 5.56 2.77 0 15 0.06 0.25 0 2 100 1728 25.89 11.30 7 66 4.29 2.68 0 14 0.15 0.41 0 2 200 1728 24.49 11.18 8 63 3.75 2.25 0 11 0.08 0.31 0 3 300 1728 23.43 11.70 4 58 3.54 2.50 0 14 0.13 0.35 0 2 Signal cycle length (seconds) 100 3456 23.61 10.62 4 56 3.86 2.37 0 14 0.08 0.28 0 2 150 3456 25.52 11.61 7 66 4.71 2.89 0 15 0.13 0.38 0 3 Right-turn-on-red Yes 3456 24.71 11.10 4 65 4.27 2.67 0 15 0.09 0.32 0 2 No 3456 24.42 11.23 7 66 4.29 2.68 0 15 0.12 0.35 0 3 Traffic volume (vphpl) 450 2304 17.30 6.75 4 38 2.88 1.97 0 11 0.18 0.42 0 3 550 2304 25.03 9.25 9 55 4.92 2.17 0 13 0.08 0.31 0 2 650 2304 31.37 11.99 10 66 5.05 3.14 0 15 0.05 0.24 0 2 Right turn percentage 2% 2304 23.74 10.23 7 50 4.02 2.44 0 15 0.14 0.38 0 3 10% 2304 25.07 11.50 7 63 4.57 2.69 0 15 0.05 0.23 0 2 20% 2304 24.89 11.67 4 66 4.26 2.85 0 15 0.12 0.37 0 2 Bicycle volume (bph) 0 2304 24.38 11.07 5 65 4.28 2.65 0 15 0.10 0.32 0 2 30 2304 24.57 11.14 5 65 4.29 2.69 0 15 0.10 0.32 0 3 60 2304 24.76 11.29 4 66 4.28 2.68 0 15 0.12 0.36 0 3 Truck percentage 3% 3456 17.20 5.45 4 33 3.12 1.93 0 11 0.09 0.30 0 3 6% 3456 31.93 10.54 10 66 5.45 2.80 0 15 0.12 0.37 0 3 Bus stop frequency (buses per hour) 0 2304 24.55 11.07 4 63 4.30 2.68 0 15 0.11 0.35 0 3 1 2304 24.58 11.26 6 66 4.27 2.67 0 15 0.10 0.32 0 2 2 2304 24.54 11.17 5 66 4.30 2.67 0 15 0.10 0.32 0 2 Bus dwell time (seconds) 15 3456 24.59 11.17 4 65 4.27 2.68 0 15 0.11 0.35 0 3 30 3456 24.46 10.33 7 53 5.56 2.77 0 15 0.06 0.25 0 2 Note: “Std.” – standard deviation TWLTL vs. Non-Traversable Median This section summarizes the data collected for the TWLTL vs. non-traversable median study. The study is intended to produce one set of performance relationships that describe the safety and operation of the TWLTL, and a second set of relationships that describe the safety and operation of the non-traversable median (NTM). When both sets of relationships are used together, the results can be compared to obtain information about the relative performance of the two techniques. The objective and scope of the study are provided in Appendix E (in a section having the same title as this section). The data used to develop the operations relationships were obtained from a traffic simulation model (i.e., VISSIM). The summary of these data includes a description of the simulation test beds, the database organization, data collection techniques, data reduction procedures, and statistics used to describe the collected data. The data used to develop the safety relationships were obtained from field measurements and reported crash data. The summary of these data includes a description of the sites at which data were collected, the database organization, data collection process, data reduction procedures, and statistics used to describe the collected data.

250 Operations Study Simulation data were used to evaluate the operational effects of the TWLTL and the NTM. A simulation testbed was established using several different prototype street segments. One-half of the segments included a TWLTL, the other one-half of the segments included a NTM. The database collectively included simulation results for a range of traffic characteristics, geometric design elements, and traffic control features. The data describe the association between operational performance and various street design factors for both the TWLTL and the NTM. The VISSIM simulation model was used to generate the performance data. This model has the capability of simulating cars, pedestrians, bicycles, transit, and trucks. The data used to calibrate this model are summarized in a later section titled VISSIM Simulation Model Calibration. Test Bed Characteristics Two urban street segments were used to establish four sites, where a “site” is one direction of travel on a segment. The four sites are identified in the following list.  Site 1. E. Hillsborough Ave, Tampa, Florida, from N. Florida Ave. to N Central Ave., eastbound direction (code: 1 - FL1 east).  Site 2. E. Hillsborough Ave, Tampa, Florida, from N. Florida Ave. to N. Central Ave., westbound direction (code: 2 - FL1 west).  Site 3. N. 54th Ave, St. Petersburg, Florida, from N. 71st Ave. to N. 66th St., eastbound direction (code: 3 - FL2 east).  Site 4. N. 54th Ave, St. Petersburg, Florida, from N. 71st Ave. to N. 66th St., westbound direction (code: 4 - FL2 west). The characteristics of the TWLTL and NTM test beds are summarized in Table 115 and Table 116, respectively. Most of the characteristics were held constant across all scenarios for analysis. Some characteristics (e.g., through volume) were changed according to the study design. The combination of characteristics that changed produced 432 unique combinations (i.e., scenarios) for the TWLTL and 288 scenarios for the NTM. Thus, a total of 720 scenarios were created for two median types combined. Database Organization A database was produced in an Excel worksheet to store all output data from the simulation. The database contains input variables (i.e., characteristics), simulation information, and outputs (measures of effectiveness, MOEs) for two one-hour simulation periods for each scenario. Table 135 describes the data fields of the database.

251 Table 135. Data fields in output database for operations study – TWLTL vs. non-traversable median. Category Data Field Description Code or Value (Unit) Input variable site Site ID 1 – FL1 East, 2 – FL1 West 3 – FL2 East, 4 – FL2 West base_model Base test bed indicator 1 – FL1; 2 – FL2 direction Approach 1 – East Bound (EB) 2 – West Bound (WB) treat AM treatment indicator 1 – TWLTL; 2 – NTM; signal_cycle Signal cycle length indicator 1 – 100; 2 – 150 (seconds) access_density Number of access points per mile TWLTL: 9.1, 13.6, 22.7 for site 1 (0.22 mi) 9.1, 13.6, 18.2 for site 2 (0.22 mi) 8.3, 12.5, 16.7 for site 3 (0.48 mi) 8.3, 12.5, 20.8 for site 4 (0.48 mi) NTM (includes full- and partial-access): 9.1, 18.2 for site 1 (0.22 mi) 9.1, 9.1 for site 2 (0.22 mi) 8.3, 12.5 for sites 3 and 4 (0.48 mi) veh_vol Through vehicle inputs 450; 550; 650 (per hour per lane) bike_vol Through bicycle inputs 0; 30; 60 (per hour) truck_per Truck percentage 3%; 6% Simulation information sim_run Simulation run index 1 – the first run; 2 – the second run Performance measures bike_traveltime Average travel time of bicycles Numeric (seconds per bicycle) bike_speed Average space mean speed of bicycles Numeric (miles per hour per bicycle) truck_traveltime Average travel time of trucks Numeric (seconds per truck) truck_speed Average space mean speed of trucks Numeric (miles per hour per truck) Traffic outputs vol_traffic Traffic volume (passenger cars and trucks) entering site during study Numeric (vehicles per hour per lane) vol_truck Truck volume entering during site during study Numeric (trucks per hour) vol_bike Bicycle volume entering site during study Numeric (bicycles per hour) Data Collection Process The procedure for simulation and data collection is described by the following steps. Step 1. Create Scenarios Basic Characteristics. The basic characteristics of the two test beds were updated, based on the characteristics shown in Table 115 and Table 116. Evaluation Measures. Two evaluation test beds were coded. Each test bed consisted of an urban street segment bounded by a signalized intersection. For each test bed, an eastbound site and a westbound site were defined, corresponding to the two travel directions on the segment. A plan view sketch of each test bed is shown in Figure 12.

252 a. E. Hillsborough Avenue test bed (FL1). b. 54th Avenue test bed (FL2). Figure 12. Evaluation segment test beds – TWLTL vs. non-traversable median. Scenarios. The test beds were updated manually to produce the desired simulation scenarios. The three factors that were updated in this manner include: (1) median treatment (i.e., TWLTL/NTM); (2) access density levels (as shown in Table 135) based on the combination of test bed and median treatment; and (3) signal cycle length (i.e., 100s and 150s). As the result, a total of eight VISSIM base models (= 2 test beds × 2 access density levels × 2 signal cycle lengths) were produced. Based on the eight models, a MATLAB application coded the remaining scenarios (i.e., the vehicle inputs, bicycle inputs, and truck percentages) through the VISSIM COM APIs. Through this process, 720 simulation scenarios were produced.

253 Step 2. Check Errors The research team randomly selected 10 scenarios to review the simulation scenarios for possible errors. The error checking included: (1) a review of the animations to identify visible errors (e.g., geometric errors, movements, signal sequence, etc.), (2) verification of the vehicle inputs and turning movements, and (3) validation of the performance measure outputs. All identified errors were corrected. Step 3. Run Simulations The MATLAB program executed the 720 simulation scenarios in VISSIM. For each scenario, four simulation runs were conducted to account for microsimulation randomness. A 15-minute warm-up period plus a 60-minute simulation period was applied in each simulation run. On a Dell Precision T7600 Workstation (Intel Xeon E5-2680 2.7GHz), one simulation run was executed for approximately 70 to 130 seconds. To reduce the simulation time, parallel computing technologies were used. Specifically, four simulation threads were launched simultaneously through the MATLAB codes. Step 4. Read Data from Simulation Results and Export to Database After completing the simulations, MATLAB program retrieved operations performance measures from the simulation model and exported the results to a database using a CSV format. Traffic outputs were collected at the traffic data collection points shown in Figure 12. The database data fields are defined in Table 135. Travel time measures were converted into average speed using the following equation: ∑ , where is average speed of bicycles or trucks along jth test bed; , is travel time of ith object (bicycle or truck) along jth testbed; is the length of testbed (i.e., 0.22 miles or 0.48 miles); is the number of objects (bicycles or trucks) counted. Traffic volume, which is defined as the total vehicles (passenger cars and trucks) entering the test beds, were counted the upstream traffic data collection points for one hour. The following equation was used to calculate traffic volume: , , where is traffic volume (per hour per lane) entering jth test bed; , and , are counted passenger cars and trucks within one hour on jth test bed, respectively; is the number of through lanes in the subject direction of travel (N equals 2 in this study). Truck volume and car volume was counted at the same data collection point. The proportion of trucks was calculated as truck volume divided by traffic volume. This proportion was multiplied by 100 to obtain “truck percentage.” Traffic volume was also collected at downstream traffic data collection points ( ). If the difference between the upstream traffic volume and downstream traffic volume ( ) was greater than 130 vehicles per lane per hour, it was typically due to congestion in the simulation. The simulation runs with 130 were considered invalid and their results were removed from the database. Database Summary Site Characteristics A study site is defined as one travel direction on a street segment, where the segment is bounded on each end by a signalized intersection and there are no signalized intersections along the segment length. The four study sites are shown in Figure 12. The basic geometric and traffic control characteristics of the four sites (e.g., location, through lanes, volume, etc.) are summarized in Table 115 and Table 116.

254 Operations Performance Measures A total of 2,880 observations were produced from simulations (= 720 scenarios × 4 simulation runs per scenario). After removing some invalid observations, the number of observations was reduced to 2801 observations (= 1152 NTM + 1649 TWLTL). Of this total, those having no bicycle volume were removed for the bicycle-speed analysis, leaving 1489 observations to develop the bicycle-speed predictive relationship (= 768 NTM + 721 TWLTL). The key operations measures include:  Average bicycle speed (mph)  Average truck speed (mph) The summary statistics for these operations measures are shown in Table 136. Table 136. Summary statistics for bicycle and truck speed – TWLTL vs. non-traversable median. Factor Bicycle Speed by Median Type (mph) Truck Speed by Median Type (mph) Non-Traversable TWLTL Non-Traversable TWLTL Obs. Mean Std. Obs. Mean Std. Obs. Mean Std. Obs. Mean Std. Overall 768 9.26 1.45 721 8.78 1.83 1152 15.77 2.89 1649 15.18 3.29 Cycle length (s) 100 384 9.74 0.63 384 9.55 0.89 576 16.56 2.95 862 16.19 3.50 150 384 8.78 1.83 337 7.90 2.20 576 14.97 2.60 787 14.07 2.62 Segment length (mi) 0.22 384 8.23 1.33 382 7.44 1.49 576 14.10 2.54 862 13.09 2.35 0.48 384 10.29 0.57 339 10.29 0.57 576 17.44 2.17 787 17.47 2.56 Access density for Sites 1 & 2 (0.22 mi) (a.p./mile) 9.1 288 8.43 1.16 - - - 432 14.29 2.59 288 13.65 2.17 13.6 - - - 192 7.44 1.49 - - - 288 12.92 2.28 18.2 96 7.64 1.63 95 7.21 1.28 144 13.50 2.29 143 12.71 2.84 22.7 - - - 95 7.66 1.64 - - - 143 12.68 2.10 Access density for Sites 3 & 4 (0.48 mi) (a.p./mile) 8.3 192 10.29 0.57 - - - 288 17.36 2.10 278 17.27 2.55 12.5 192 10.28 0.57 166 10.29 0.57 288 17.52 2.24 252 17.62 2.49 16.7 - - - 87 10.31 0.62 - - - 129 17.53 2.32 20.8 - - - 86 10.26 0.52 - - - 128 17.53 2.94 Traffic volume (vphpl) 450 256 9.26 1.45 256 8.85 1.85 384 17.04 2.29 576 16.55 2.72 550 256 9.26 1.46 252 8.83 1.82 384 15.90 2.66 572 15.04 3.09 650 256 9.25 1.45 213 8.62 1.82 384 14.36 3.04 501 13.76 3.46 Bike volume (bph) 0 - - - - - - 384 15.81 2.88 554 15.18 3.26 30 384 9.26 1.48 356 8.84 1.84 384 15.74 2.89 548 15.25 3.34 60 384 9.26 1.42 365 8.72 1.82 384 15.75 2.92 547 15.10 3.26 Truck percent 3% 384 9.26 1.45 360 8.78 1.84 576 15.90 3.01 824 15.36 3.45 6% 384 9.26 1.45 361 8.77 1.83 576 15.64 2.77 825 15.00 3.11 Note: 1 – Refer to the definition of access level in Table 135. “Std.” – standard deviation Safety Study A cross-sectional database was assembled to evaluate the safety effects of the TWLTL and NTM on transit vehicles and trucks. The set of sites represented in the database collectively include a range of traffic characteristics, geometric design elements, and traffic control features. Data for sites having a flush-painted median were also included in the database. These sites were grouped with those having a TWLTL because both median types can be characterized as “traversable”. Hereafter, the two classes of median type addressed are the NTM and the traversable median (TM). The

255 data describe the association between safety performance and various street design factors for both the TM and the NTM. Study Sites This section describes the 214 study sites represented in the database. Each site represents a two-way street segment located between two signalized intersections. The segments selected for study are identified in Table 137. Table 137. Study site description – TWLTL vs. non-traversable median. Category Description Non-Traversable Median Traversable Median Total Segments Length, mi Segments Length, mi Segments Length, mi State FL 24 9.7 29 13.6 53 23.3 NJ 32 9.9 1 0.2 33 10.1 OH 20 4.5 35 8.6 55 13.1 OR 10 2.0 48 18.1 58 20.1 WA 1 0.1 5 1.5 6 1.6 WI 9 1.8 0 0.0 9 1.8 Through lanes 4 53 16.7 111 39.2 164 55.9 6 43 11.4 7 2.9 50 14.3 Speed limit, mph 30 14 2.4 3 0.7 17 3.1 35 8 1.5 71 19.4 79 20.9 40 35 10.5 30 14.6 65 25.1 45 38 13.1 14 7.5 52 20.6 50 1 0.5 0 0.0 1 0.5 On-street parking presence Neither side 81 25.3 97 35.8 178 61.1 One side 6 1.2 3 0.5 9 1.7 Both sides 9 1.6 18 5.9 27 7.5 Transit stops per segment 0 23 6.3 10 3.9 33 10.2 1 18 4.2 19 3.9 37 8.1 2 35 10.2 24 5.4 59 15.6 3 11 3.8 23 6.3 34 10.1 4 5 2.0 15 5.5 20 7.5 5 4 1.5 13 6.2 17 7.7 6+ 0 0.0 14 11.0 14 11.0 Bike lane presence Neither side 67 21.1 68 24.7 135 45.8 One side 1 0.5 1 0.1 2 0.6 Both sides 28 6.5 49 17.3 77 23.8 Total: 96 28.0 118 42.2 214 70.2

256 Of the 214 segments in the database, 96 segments have a non-traversable median and total 28 miles in length. There are 118 segments with a traversable median; they total 42.2 miles in length. The 214 segments are collectively located in urban or suburban areas of six states. The characteristics listed in Table 119 were used during the segment selection process. Some characteristics were used to screen the segments. Some other characteristics were monitored during segment selection to ensure that the collective set of segments offered a wide and balanced range of values. In this regard, a desired goal of the segment selection process was that the collective set of segments included a range of values for each characteristic. The values in this table indicate that this goal was achieved for most characteristics. Database Organization This section describes the individual data elements (i.e., variables) in the various databases needed to describe the study segments and their performance. The database categories needed for this study are listed in Table 138. The worksheet name wherein the data were recorded is also shown. Table 138. Databases – TWLTL vs. non-traversable median. Data Category Database Name Worksheet Name Road inventory data Site characteristics data Geometry Safety performance measure data Crash data Crash Road Inventory Data This section describes the segment characteristics and traffic characteristics data elements. These data are listed in Table 139. They were recorded in the “Geometry” worksheet. One row in this worksheet described the data for one segment. Table 139. Site characteristics data – TWLTL vs. non-traversable median. Category Data Element Variable Name General Site identification number. link_ID Major street name street_name_sc Code to identify issues that make site unsuitable for analysis prob_flag_sc Text to describe observations or concerns notes_sc Latitude and longitude of south or west intersection center lat1_sc, long1_sc Latitude and longitude of north or east intersection center lat2_sc, long2_sc Design Segment length seg_len Number of through lanes lane1_nbr, lane2_nbr Width of through lanes lane1_wid, lane2_wid Median type (T-TWLTL, F-flush painted, R-raised curb, B-barrier) med_type Median width med_wid_maj Outside shoulder width lane1_shldr, lane2_shldr On-street parking width lane1_park, lane2_park Bicycle lane width lane1_bike, lane2_bike Number of stop locations for transit vehicle 1 bus1_stop_loc Number of stop locations for transit vehicle 2 bus2_stop_loc

257 Category Data Element Variable Name Number of stop locations for transit vehicle 3 bus3_stop_loc Number of stop locations for transit vehicle 4 bus4_stop_loc Number of full-access driveways associated with each land use category lane1_res_f, lane1_ind_f, lane1_bus_f, lane1_off_f, lane2_res_f, lane2_ind_f, lane2_bus_f, lane2_off_f Number of partial-access driveways associated with each land use category lane1_res_p, lane1_ind_p, lane1_bus_p, lane1_off_p, lane2_res_p, lane2_ind_p, lane2_bus_p, lane2_off_p Number of public street approaches lane1_xstrt, lane2_xstrt Presence of a continuous right-turn lane on the major street lane1_rt, lane2_rt Presence of horizontal curve in major street horiz_curve Traffic control Speed limit lane1_speed, lane2_speed Mid-segment pedestrian crosswalk (N-none, G-signal, Y-unsignalized) crosswk_mrk Traffic volume Major street annual average daily traffic (AADT) volume AADT2012, AADT2013, AADT2014, AADT2015, AADT2016 Major street annual average truck percentage TK2012, TK2013, TK2014, TK2015, TK2016 Transit volume Number of times per day that local transit vehicle 1 traverses segment bus1_traversals Number of times per day that local transit vehicle 2 traverses segment bus2_traversals Number of times per day that local transit vehicle 3 traverses segment bus3_traversals Number of times per day that local transit vehicle 4 traverses segment bus4_traversals Safety Performance Measure Data This section describes the crash database assembled to develop a performance relationship for transit vehicles and for trucks. These data are listed in Table 140. They were recorded in the “Crash” worksheet. One row in this worksheet describes the data for one transit-related or truck-related crash at one segment. Table 140. Crash database – TWLTL vs. non-traversable median. Category Data Element Variable Name General Site identification number. link_ID Major street name street_name_cr Nearest intersecting street name cross_name_cr Distance to nearest intersecting street cross_dist_cr Crash location coordinates lat_cr, long_cr Source of original crash data source_cr County of crash county_cr Date of crash minute_cr, hour_cr, day_cr, month_cr, year_cr Milepost of crash milepost_cr Crash description First harmful event harm1_cr Manner of collision manner_cr Crash location relative to other junctions location_cr Work zone related workzone_cr Vehicle maneuver prior to crash maneuver_cr

258 Unsignalized access points Signal Signal - Segment boundary N Segment length Category Data Element Variable Name Crash severity severity_cr Number of transit vehicles involved in crash nbr_bus_cr Number of non-transit trucks involved in crash nbr_trk_cr Data Collection and Reduction Techniques This section describes the techniques used to collect and process individual data elements. The focus of the discussion is on techniques used to obtain the raw data included in each database. Each segment represents the two-way street segment located between two signalized intersections. There are no signalized intersections along the length of the subject segment. There can be one or more unsignalized access points (i.e., a public street approach or a driveway is an access point) along the segment’s length. A segment is defined to include both directions of travel along a street segment. One- way streets were not included in this study. The segment boundaries are shown in Figure 13. The boundaries extend to the center of the signalized intersection conflict area. Figure 13. Segment boundaries – TWLTL vs. non-traversable median. Road Inventory Data The road inventory data were obtained from Google Earth aerial imagery and Street View. The Historical Imagery view in Google Earth was used to briefly review all available photos during the years 2012 to 2016 to determine if construction occurred at the subject driveway. Measurements were made using the Ruler tool. The AADT data were obtained from the local agency that operates and maintains the major street. The local public transit volume data were obtained from the corresponding transit agency website. Only through lanes were counted. Through lanes are continuous lanes for the length of the segment and serve traffic traveling the segment as through vehicles and exiting the segment at the downstream signalized intersection as a through movement. Median width was measured as the distance between the near edges of the traveled way for the two opposing travel directions. In this manner, the inside shoulder width (if present) is included in the measured median width. Through lane width is measured for each travel direction and divided by the number of through lanes. It is an average for all through lanes. For divided roadways, the through lane width is measured between the far edges of traveled way for the inside and outside lanes of each travel direction. The shoulder width is measured from the face of curb (or edge of pavement if uncurbed) to the near edge of the traveled way for the outside lane. If a curb-and-gutter section is present, its width was

259 measured and interpreted to represent an effective shoulder width. If a marked bicycle lane is present, its width is not included in the shoulder width. That is, shoulder width and bicycle lane width are mutually exclusive. Access points were counted for each travel direction separately. An access point was a driveway, an unsignalized public street approach, or an alley. Full-access driveways allow left turns and right turns in and out of the property. Partial-access driveways use a median or island channelization to prohibit one or more turn movements in and out of the property. Driveways that are unused were not counted. Similarly, driveways leading into fields, small utility installations (e.g., cellular phone tower), and abandoned buildings were not counted. A circular driveway at a residence was counted as one driveway even though both ends of the driveway intersect the subject segment. Similarly, a small business (e.g., gas station) that has two curb cuts separated by only 10 or 20 feet was considered to have effectively one driveway. Similarly, a business that has a circular one-way driveway was considered to have effectively one driveway even though both ends of the driveway intersect the subject segment at different locations. The land use served by a driveway was categorized as residential (or undeveloped), industrial, commercial, or office. Table 141 was used to determine the land use associated with each driveway along the subject arterial segment. Table 141. Adjacent land use characteristics – TWLTL vs. non-traversable median. Land Use Characteristics Examples Residential or undeveloped  Buildings are small  A small percentage of the land is paved  If driveways exist, they have very low volume  Ratio of land-use acreage to parking stalls is large  Single-family home  Undeveloped property, farmland  Graveyard  Park or green-space recreation area Industrial  Buildings are large and production oriented  Driveways and parking may be designed to accommodate large trucks  Driveway volume is moderate at shift change times and is low throughout the day  Ratio of land-use acreage to parking stalls is moderate  Factory  Warehouse  Storage tanks  Farmyard with barns and machinery Commercial  Buildings are larger and separated by convenient parking between building and roadway  Driveway volume is moderate from mid-morning to early evening  Ratio of land-use acreage to parking stalls is small  Strip commercial, shopping mall  Apartment complex, trailer park  Airport  Gas station  Restaurant Office  Buildings typically have two or more stories  Most parking is distant from the building or behind it  Driveway volume is high at morning and evening peak traffic hours; otherwise, it is low  Ratio of land-use acreage to parking stalls is small  Office tower  Public building, school  Church  Clubhouse (buildings at a park)  Parking lot for “8 to 5” workers Safety Performance Measure Data Crash data were obtained for the most recent five-year period that records were available. This time period varied among the agencies contacted. However, in most cases, the crash data obtained correspond to the years 2012 to 2016. Electronic copies of the crash reports (including fatal, injury, and property-damage-only crashes) for the major street associated with each study segment were requested from the local agency. The crash data

260 requested described crashes that occurred on the subject segment. Upon receipt, the original agency data was screened to identify those crashes that were transit-related or truck-related. Crashes associated with other vehicle types were not included in the crash database. Transit vehicle-involved crashes are not included in the count of truck crashes. With one exception, all crashes that occurred between the mileposts associated with the begin- and end- points of the segment were included in the database. The exception was crashes that were located at (or related to the operation of) the boundary signalized intersections. These crashes were not included in the database. Also, crashes that occurred on the side-street approaches to the segment were not included in the database. Database Summary This section summarizes the data in the assembled database. The first section summarizes the segment characteristics and the second section summarizes the crash characteristics. Site Characteristics The average and range for selected data elements are listed in Table 142. These statistics are listed separately by median type. The segments with traversable median have an average lane with of 10.8 feet, while those with non-traversable median have an average of 11.5 feet. In contrast, the traversable median segments have an average 12.2-foot median while the non-traversable median segments have an average 20.1-foot median. The segments with traversable medians tend to have twice as many driveways per mile, relative to the non-traversable median segments. Table 142. Site characteristics summary – TWLTL vs. non-traversable median. Data Element Non-Traversable Median Traversable Median Average Minimum Maximum Average Minimum Maximum Median width, ft 20.1 6.0 62.0 12.2 10.0 26.0 Lane width, ft 11.5 10.5 12.5 10.8 9.5 12.7 Shoulder width, ft 2.2 0.0 10.0 1.5 0.0 7.0 Shoulder+bike lane width, ft 3.8 0.0 11.0 3.6 0.0 12.0 Residential access point density, ap/mi 4.7 0.0 63.8 8.2 0.0 137.9 Industrial access point density, ap/mi 0.7 0.0 17.9 0.7 0.0 19.8 Commercial access point density, ap/mi 16.2 0.0 61.2 34.1 0.0 79.6 Office access point density, ap/mi 1.8 0.0 13.0 2.6 0.0 16.2 Cross street approach density, ap/mi 5.4 0.0 26.3 9.9 0.0 25.1 AADT (overall) 36,228 12,600 72,800 25,647 9640 48,600 Truck AADT 2407 139 10,294 1264 326 4075 Transit AADT 110 0 492 131 0 356

261 The AADT shown in the third row from the bottom of Table 142 represents the “traditional” AADT that describes the overall traffic stream. Thus, this AADT includes all vehicles traveling on the roadway (including cars, trucks, and transit vehicles). The segments with traversable medians tend to have lower overall AADT and truck AADT than non-traversable median segments. In contrast, the traversable median segments tend to have higher transit AADT than non-traversable median segments. Crash Characteristics The count of crashes reported for the collective set of study segments is shown in Table 143. The counts are separated by median type. There are a total of 172 transit-related crashes and 369 truck-related crashes included in the database. Table 143. Crash characteristics summary – TWLTL vs. non-traversable median. Severity Category Crash Count and Crash Rate by Vehicle Class and Median Type Transit Truck Non-Traversable Median Traversable Median Non-Traversable Median Traversable Median Count Rate, cr/mvm Count Rate, cr/mvm Count Rate, cr/mvm Count Rate, cr/mvm K 0 0.000 0 0.000 0 0.000 0 0.000 A 0 0.000 3 0.002 2 0.001 3 0.002 B 6 0.003 4 0.002 10 0.005 10 0.005 C 14 0.008 12 0.006 73 0.040 19 0.010 KABC 20 0.011 19 0.010 85 0.046 32 0.016 PDO 56 0.030 77 0.039 184 0.100 68 0.034 Total 76 0.041 96 0.048 269 0.146 100 0.050 mvm 1841   1995   1841   1995   Note: mvm – million vehicle miles. K – fatal, A – incapacitating injury, B – non-incapacitating injury, C – possible injury, PDO – property-damage-only. Table 143 also shows the crash rate, which is computed using the overall AADT and segment length to compute the total million vehicle miles (mvm) for each location. This metric is intended to remove differences among segments due to segment length and AADT. It does not fully remove differences in truck volume or transit vehicle volume because these volumes typically represent a small proportion of the overall AADT. The crash rates shown in the second-to-last row of Table 143 indicate that the crash rate for transit- related and truck-related crashes. For transit-related crashes, the rates are 0.041 and 0.048 crashes per mvm (cr/mvm) for segments with non-traversable and traversable median, respectively. A comparison of these two rates suggests that the non-traversable segments are safer. However, when the respective transit AADT is divided into the corresponding crash rate (e.g., 0.041/110; 0.048/131)), the resulting value is smaller for the traversable median which suggests that the traversable median segments have a smaller transit vehicle-related crash risk than those segments with a non-traversable median. The crash rate for truck-related crashes is 0.146 and 0.050 cr/mvm for segments with non-traversable and traversable median, respectively. A comparison of these two rates suggests that the traversable median segments have a smaller truck-related crash risk than the non-traversable median segments. This

262 trend remains when the respective truck AADT is divided into the corresponding crash rate. It is consistent with the finding for transit-related crashes. This simple analysis of crash rates provides insight into the trends present in the data. It suggests that traversable medians may provide safer operation for trucks and transit vehicles than would a non-traversable median. The insights obtained from this examination were used to guide the development for the safety performance relationship for trucks and transit vehicles as influenced by median type. The discussion in this section is not intended to indicate conclusive results or recommendations. The recommended performance relationships (and associated trends) are documented in Appendix G. VISSIM Simulation Model Calibration This section summarizes the data collected to calibrate the VISSIM traffic simulation model such that it accurately describes driver behavior at sites in Oregon and Florida. This model was used to create the operations data needed for the right-turn deceleration study and the TWLTL vs. non-traversable median study. The objective and scope of the study are provided in Appendix E (in a section titled, Simulation Model Calibration). Data used to calibrate the SSAM software tool are summarized in the next section. Field Data This section summarizes the field data that were collected and used to calibrate the VISSIM model. The summary of these data includes a description of the sites at which data were collected, the database organization, data collection techniques, data reduction procedures, and statistics used to describe the collected data. Study Sites For the TWLTL vs. non-traversable median study, three urban street segments were used to establish six sites, where a “site” is one direction of travel on a segment. The six sites are identified in the following list.  Site 1. E. Hillsborough Ave, Tampa, Florida, from N. Florida Ave. to N Central Ave., eastbound direction (code: 1 - FL1 east).  Site 2. E. Hillsborough Ave, Tampa, Florida, from N. Florida Ave. to N. Central Ave., westbound direction (code: 2 - FL1 west).  Site 3. N. 54th Ave, St. Petersburg, Florida, from N. 71st Ave. to N. 66th St., eastbound direction (code: 3 - FL2 east).  Site 4. N. 54th Ave, St. Petersburg, Florida, from N. 71st Ave. to N. 66th St., westbound direction (code: 4 - FL2 west).  Site 5. Killingsworth St., Portland, Oregon, from NE 72nd Ave. to NE 82nd Ave., eastbound direction (code: 5 – OR3 east).  Site 6. Killingsworth St., Portland, Oregon, from NE 72nd Ave. to NE 82nd Ave., westbound direction (code: 6 – OR3 west). The basic characteristics of the six segment sites are summarized in Table 144.

263 Category Characteristic Characteristic Value by Site Number 1 2 3 4 5 6 Site information Test bed FL1 FL2 OR3 Facility type Arterial Segment Arterial Segment Arterial Segment City, State Tampa, FL St. Petersburg, FL Portland, OR Street name Hillsborough Ave 54th Ave Killingsworth St Boundary streets Florida Ave – Central Ave 71 st St – 66th St NE 72 nd Ave – NE 82nd Ave Approach (side) EB WB EB WB EB WB Geometric design Length, mi 0.22 0.48 0.51 Median type NTM TWLTL TWLTL Segment through lanes 2 2 2 2 2 2 Segment through lane width, ft 11 11 11 11 11 11 Lanes at downstream signal Left-turn lanes 1 1 1 1 0 1 Left-turn lane width, ft 10 11 10 10 12 12 Right-turn lanes 0 1 0 0 1 0 Right-turn lane width, ft 10 11 10 10 12 12 Marked bicycle lane width, ft 4 4 5 5 4 4 Bus-only lane No No No No No No Right-turn bay to access points No No No No No No Full-access points 0 0 9 12 20 17 Partial-access points 3 0 1 1 0 0 Midblock bus stops 1 1 1 3 3 3 On-street parking No No No No No No Horizontal curve No No No No No No Grade, % 0 0 0 0 0 0 Speed limit, mph 40 45 40 40 35 35 Traffic character- istics Volume at upstream signal Left-turn volume, vph 28 20 210 145 6 265 Through volume, vph 1344 1683 393 491 1791 704 Right-turn volume, vph 58 24 48 139 46 0 U-turn volume, vph 8 5 0 0 0 0 Volume at downstream signal Left-turn volume, vph 43 81 143 73 0 34 Through volume, vph 1479 1520 371 533 1509 927 Right-turn volume, vph 12 114 171 149 370 6 U-turn volume, vph 5 8 0 0 0 0 Inbound access point vol., vph 3 0 7 10 10 5 Outbound access point vol., vph 19 0 7 12 8 6 Truck percentage 3.95% 3.05% 2.6% 1.4% 6.35% 9.30% Bus frequency, vph 2 2 1 1 12 9 Bus dwell time, s 14 18 11 7 20 18 Bicycle volume, bph 5 3 1 1 2 1 Pedestrian volume, pph 9 5 2 1 14 13 Average desired speed Automobile and truck, mph 44 45 43 42 40 40 Bus, mph 31 37 38 37 30 30 Bicycle, mph 15 15 15 15 10 10 Table 144. Segment site characteristics – VISSIM calibration.

264 Category Characteristic Characteristic Value by Site Number 1 2 3 4 5 6 Pedestrian, mph 3 3 3 3 3 3 Downstream signal timing Coordination Yes Yes Yes No No Yes Cycle length, s 210 210 200 (free) (free) 100 Offset, s 56 176 26 (free) (free) 52 Left-turn phase settings Left-turn operational mode Perm. Prot. + Perm. Prot. + Perm. Prot. + Perm. (No left) Prot. + Perm. Minimum green, s 5 5 5 3 Maximum green, s 16 26 20 19 Yellow change interval, s 4.9 4.4 4.4 3 Red-clearance interval, s 2.0 2.1 2.1 0 Through phase settings Minimum green, s 10 15 5 20 25 10 Maximum green, s 165 135 49 50 45 35 Yellow change interval, s 4.6 4.9 4.4 4.4 4.4 4.4 Red-clearance interval, s 2.0 2.3 2.8 2.1 1.0 1.0 For the right-turn deceleration study, two urban signalized intersections were used to establish two sites, where a “site” is one approach to a signalized intersection. The two sites are identified in the following list:  Site 7. E. Hillsborough Ave at N. Florida Ave., Tampa, Florida, eastbound approach (code: 7 – FL4 east).  Site 8. SE Division St. at SE 162nd Ave., Portland, Oregon, westbound approach (code: 8 – OR5 west). The basic characteristics of these two intersection sites are summarized in Table 145. Table 145. Intersection site characteristics – VISSIM calibration. Category Characteristics Characteristic Value by Site Number 7 8 Site information Test bed FL4 OR5 Facility type Signalized Intersection Signalized Intersection City, State Tampa, FL Portland, OR Intersection Hillsborough Ave. at N. Florida Ave. SE Division St. at SE 162nd Ave. Approach EB (Hillsborough Ave) WB (SE Division Street) Geometric design Number of lanes and lane width Left-turn lanes 1 1 Left-turn lane width, ft 10 10 Through lanes 2 2 Through lane width, ft 11 10 Right-turn lanes 0 1 Right-turn lane width, ft 10 Length of right-turn deceleration, ft 0 200

265 Category Characteristics Characteristic Value by Site Number 7 8 Width of marked bike lane, ft 4 5 Distance of bus stop to intersection, ft Yes Yes Number of driveways within 250 ft 2 0 Speed limit, mph 40 35 Traffic character- istics On-street parking activities No No Intersection volume Left-turn volume, vph 137 34 Through volume, vph 1253 440 Right-turn volume, vph 13 95 U-turn volume, vph 7 0 Truck percentage 1.1% 4.2% Inbound access point volume, vph 5 0 Outbound access point volume, vph 4 0 Bus frequency, vph 2 4 Bicycle volume, bph 3 1 Pedestrian volume, pph 3 2 Average desired speed Automobile and truck, mph 44 40 Bus, mph 31 30 Bicycle, mph 15 10 Pedestrian, mph 3 3 Signal timing Cycle length, s 210 Free Offset, s 176 Free Left-turn phase settings Left-turn operational mode Prot. + Perm. Prot. only Minimum green, s 5 3 Maximum green, s 16 30 Yellow change interval, s 4.9 3.0 Red-clearance interval, s 2.0 1.0 Through phase settings Minimum green, s 15 20 Maximum green, s 135 40 Yellow change interval, s 4.9 4.0 Red-clearance interval, s 2.3 1.0 Database Organization Field data were collected at all eight sites, including road characteristics and operations measures. These data were used to develop simulation models and calibrate simulation parameters. Table 146 describes the data fields in the database. Table 146. Data fields in database for calibration study – VISSIM calibration. Category Database Data Fields Evaluation Period Source Road inventory Geometric design  Number of left-turn, through and right-turn lanes at segment and downstream signal Recent five calendar years (2013 – 2017)  FDOT Aerial Photography Archive (FL sites)

266 Category Database Data Fields Evaluation Period Source  Number of bicycle lanes  Lane width (ft)  Median type (TWLTL, NTM)  Number of median openings (segment sites)  Number of access points  Turning bay length  Google Earth Pro (OR sites) Traffic pattern  Vehicle turning volume at intersections and driveways (vph)  Bicycle volume on major roads (bph)  Truck turning volume (tph)  Bus stop location  Bus frequency (per hour)  Bus dwell time (seconds)  Speed distribution (mph)  One peak hour for segment sites  One non-peak hour for intersection sites  At 15-min intervals  Road tube (FL sites only)  Field videos Signal timing  Mode (coordination, free)  Phases  Min/Max green time  Yellow/all red time  Recalls  Cycle length (coordination mode)  Offset (coordination) 2017 version Signal timing sheets from local traffic agencies (districts or cities) Operations performance measures Vehicle travel time (seconds) Average vehicle travel time at a given measure range for each 15-s interval during the evaluation period One hour of typical weekday Field video Maximum queue length (ft) The maximum queue length in feet during each signal cycle One hour of typical weekday For intersection sites only Field video Data Collection Techniques For the Florida sites, geometric data were collected from high-quality aerial photography obtained from local agencies. For the Oregon sites, aerial photographs from Google Earth were used to obtain the geometric data. The research staff reviewed the photographs to ensure that no major facility upgrades or changes to geometric designs had taken place at the selected sites. For each turning movement, though movement, and bike lane, the research team measured the lane width at several points. The length of the left-turn bay and the right-turn bay (if applicable) at signals was measured from the end of the taper to the stop line. Research staff manually counted turning movements from video recordings, distinguishing between vehicles, heavy vehicles, pedestrians, and bicycles at signalized intersections (all approaches) and selected driveways during midday (2:00 PM – 3:00 PM), and evening (5:00 PM – 6:00 PM) time periods during two weekdays. The data were aggregated in 15-minute intervals for analysis. In addition, road tube counters were installed across the Florida segments to record traffic count, spot speed data, and headways at the segment mid-point. At the same time, several video recorders were set up along the segment and intersection sites to monitor traffic conditions in the study area. The data were collected for two days at each site. Figure 14 shows the location of the tube counters and the video camera

267 fields-of-view for the segment test beds. Similarly, Figure 15 shows the video camera fields-of-view for the intersection test beds. The research staff requested signal timing sheets for the selected sites from local agencies (city or county). They also requested information describing the signal control mode, phase configurations, coordination patterns, and timing plan. a. Test Bed FL1: Hillsborough Avenue, Tampa, Florida. b. Test Bed FL2: 54th Avenue, St. Petersburg, Florida. c. Test Bed OR3: NE Killingsworth Street, Portland, Oregon. Figure 14. Plan view sketch of cameras and counters at three segment test beds – VISSIM calibration. (Aerial Photography – Google Earth) Data Reduction Procedures Geometric Data Most geometric data were directly collected from the aerial photos. These data include: number of lanes, lane width, turn bay length, and median type. The access points (side streets and major driveways)

268 with significant traffic were identified and counted. If a driveway connected to a single-family house or an undeveloped land, it was not counted because it had very little traffic volume. a. Test Bed FL4: Hillsborough Avenue at N. Florida Avenue, Tampa, Florida. b. Test Bed OR5: SE Division Street at SE 162nd Avenue, Portland, Oregon. Figure 15. Plan view sketch of cameras at two intersection test beds – VISSIM calibration. (Aerial Photography – Google Earth) Traffic Data The weekday evening peak hour (5:00 PM – 6:00 PM) was selected as the study period for segment sites. Turning movements at signalized intersections and major driveways during this period, including vehicles, trucks, bicycles, and pedestrians, were obtained from field traffic counts. For the major driveways that were not covered by the field traffic counts, research staff reviewed field videos to count turns into and out of the driveways for the same study period. The frequency and stop dwell time of buses at the study sites were also collected from field videos. At the intersection sites, one weekend hour (11:00 AM – 12:00 PM) was selected as the study period to avoid the queue end exceeding the upstream intersection. Turning movements at signalized intersections,

269 including vehicles, trucks, bicycles, and pedestrians, were obtained from the field traffic counting database. Driveway traffic and bus frequency/dwell time were retrieved from field videos. Speed Data Spot speed data were collected at the Florida sites through road tubes. Based on the spot speeds, free flow speeds (desired speed in VISSIM) were calculated as the average speed of vehicles with a headway that equals to 10 seconds or more. For Oregon sites, the desired speed was assumed as speed limit plus five miles per hour, consistent with guidance in the Highway Capacity Manual. Operations Performance Measures Research staff reviewed the field videos to measure vehicle travel times and the maximum queue length for the sites. The researchers recorded the time of vehicles entering and leaving the study boundary at each site during the study period. The travel time of a vehicle was calculated as the time difference between its entering and exiting time. The signal status when vehicles arrive at downstream intersection was also recorded. The recorded vehicles were randomly selected to ensure that the sampled vehicles represented the average vehicle traveling along the segment. The travel times of the selected vehicles were averaged to obtain the average travel time for the site. The research team reviewed the field videos at the two intersection sites to record the location of maximum queue end in each signal cycle. From these observations, the researchers measured the distance from stop bar to the recorded locations in Google Earth Pro and recorded this length as the maximum queue length. Database Summary The two operations performance measures that were collected in the field are summarized in Table 147. Average vehicle travel time within study boundaries was measured at both the segment and the intersection sites. Maximum queue length was measured only at the intersection sites. Table 147. Summary of observed operations performance measures – VISSIM calibration. Site Test Bed Study Average travel time, s Maximum Queue Length, ft Obs. Mean Std. Obs. Mean Std. 1 – EB, Hillsborough Avenue FL1 Segment 75 42.91 23.75 2 – WB, Hillsborough Avenue FL1 Segment 75 48.59 31.25 3 – EB, 54th Avenue FL2 Segment 100 93.85 43.00 4 – WB, 54th Avenue FL2 Segment 92 82.38 31.25 5 – EB, Killingsworth Street OR3 Segment 76 75.03 19.00 6 – WB, Killingsworth Street OR3 Segment 80 55.25 9.00 7 – EB, Hillsborough Ave at Florida Ave FL4 Intersection 96 46.09 17.93 17 400.53 184.34 8 – WB, SE Division St at SE 162nd Ave OR5 Intersection 88 31.76 24.46 42 126.95 62.51 Note: Std. – standard deviation. Simulation Data This section summarizes the simulation data that were compared with the field data to calibrate the VISSIM model. The summary of these data includes a description of the simulation test beds, database

270 organization, data collection process, model calibration process, and statistics used to describe the collected data. Test Bed Characteristics One simulation model was created in VISSIM (version 9) for each of the eight test beds at which field data were collected. These test beds were described in the previous section. One simulation scenario was created for each simulation model based on the field data described in Table 144 and Table 145. Each simulation scenario consisted of a simulation period of 4,500 seconds, including a 15-minute warm-up time and one-hour evaluation time. A comprehensive error check and QA/QC was conducted for each model. These checks are identified in the following list.  Animation review: (1) connectivity of links, connectors, and static routes; (2) signal head timing sequence; (3) conflict area and priority rules; and (4) abnormal congestion compared to field videos.  Signal timing review: signal records in output files (.lsa) and average green-to-cycle-length ratios compared to signal timing sheets.  Traffic volume review: comparison of vehicle flow rate at critical points (e.g., entry, middle, and exit sections of segments, driveways, and intersections) with observed vehicle inputs to verify vehicle inputs and static routes.  Simulation warning review: warning information from simulation. Database Organization The organization of simulated data for the one-hour simulation period is given in Table 148. Table 148. Data fields in output database for calibration study – VISSIM calibration. Category Data Field Description Code or Value (Unit) Site site Site ID 1 to 8 Simulation information run Simulation replication index 1 to 10 p1 Car following parameter - W74axm, Average Standstill Distance Numeric p2 Car following parameter - W74bxAdd, Additive Part of Safety Distance Numeric p3 Car following parameter - W74bxMult, Multiple Part of Safety Distance Numeric p4 Lane change parameter – SafDistFactLnChg, Safety distance reduction factor Numeric seed Simulation seed Integer Travel time measures tt_mean Average travel time Numeric (seconds/vehicle) tt_std Standard deviation of travel time Numeric (seconds/vehicle) tt_obs Number of observations for travel time Numeric (seconds/vehicle) Maximum queue length mql_mean Average maximum queue length during a simulation period (one hour) Numeric (feet) mql_std Standard deviation of maximum queue length Numeric (feet) mql_obs Number of observations for of maximum queues Numeric (feet)

Data Collection Process A MATLAB program was developed to conduct simulations using VISSIM. The simulation models and study area boundaries for travel time and queue length are shown in Figure 16 and Figure 17. The queue length was measured along the approach, from the location of the queue counter to the queue end. a. Test Bed FL1: Hillsborough Avenue, Tampa, Florida. b. Test Bed FL2: 54th Avenue, St. Petersburg, Florida. c. Test Bed OR3: NE Killingsworth Street, Portland, Oregon. Figure 16. Plan view of simulated street system at three segment test beds – VISSIM calibration.

272 a. Test Bed FL4: Hillsborough Avenue at N. Florida Avenue, Tampa, Florida. b. Test Bed OR5: SE Division Street at SE 162nd Avenue, Portland, Oregon. Figure 17. Plan view of simulated street system at two intersection test beds – VISSIM calibration.

273 VISSIM Calibration Process The objective of the simulation model calibration was to develop one set of calibration parameters that were suitable for all sites included in the study. It was rationalized that these parameters would then provide acceptable results for sites in other regions. To achieve this objective, the following steps were conducted. Sensitivity Analysis A sensitivity analysis was used to determine the VISSIM parameters that have significant impact on selected operations measures (e.g., average travel time and maximum queue length). Seven candidate parameters were selected for the sensitive analysis, as listed in Table 149. Table 149. Simulation parameters for sensitivity analysis – VISSIM calibration. Model Parameter Description Default Value Range Arterial car following W74ax Average standstill distance 6.56 ft 3.28 ~ 10.0 ft W74bxAdd Additive part of safety distance 2.0 1 ~ 3.5 W74bxMult Multiplicative part of safety distance 3.0 2 ~ 4.5 ObsrvdVehs Overserved vehicles 4 2 ~ 8 Lane change DiffusTm Waiting time before diffusion 60 sec 40 ~ 80 sec MinHdwy Min. headway (front/rear) 1.64 ft 1.5 ~ 6 ft SafDistFactLnChg Safety distance reduction factor 0.6 0.1 ~ 0.9 For each parameter, three values were specified (minimum, default, and maximum). These values are shown in Table 149. They were then varied, while keeping other parameters constant at their default values, in the five simulation models to create 15 scenarios. Ten simulation runs were conducted for each scenario. Using the simulation results, three operations measures, average travel time and/or maximum queue length (for intersection sites only) were computed for each site. An ANOVA test was used to test for a significant difference among the three measures for each site. The test results are given in Table 150. Table 150. ANOVA Test for simulation parameters – VISSIM calibration. Site p-Values by Simulation Parameter Average Travel Time W74ax W74bxAdd W74bxMult ObsrvdVehs DiffusTm MinHdwy SafDistFactLnChg 1 0.5779 0.0017 0.1087 0.9351 0.9665 0.9842 0.5306 2 0.0289 <0.0 0.0066 0.0604 0.9819 0.5658 <0.0001 3 0.0028 0.7421 0.0005 0.1033 0.2678 0.3735 0.0236 4 0.2209 0.4034 0.0213 0.3605 0.3514 0.1435 0.8085 5 0.0149 0.2453 0.9253 <0.0001 0.5599 0.5852 <0.0001 6 0.4302 0.0467 0.6605 0.5580 0.6813 0.7524 0.7202 7 0.0407 0.0001 0.0202 0.5314 0.9890 0.8544 0.7382 8 0.8890 0.8560 0.9917 0.8390 1.0 0.9983 0.9879 Maximum Queue Length 7 <0.0001 0.0005 0.0119 0.9308 0.9999 0.9473 0.7659 8 <0.0001 0.5397 0.7511 0.9241 1.0 0.9999 0.9974 Selected? Yes Yes Yes Yes No No Yes

274 The simulation parameters which significantly impact operations measures at any site were selected as the calibration parameters. The observed vehicles parameter (“ObsrvdVehs”) in the car-following model was set at “8” because the segment and intersection sites on arterial corridors have complex traffic conditions that require drivers to monitor several vehicles. Finally, four simulation parameters were selected for optimization. These parameters are identified in the following list.  W74ax – Average standstill distance  W74bxAdd – Additive part of safety distance  W74bxMult – Multiplicative part of safety distance  SafDistFactLnChg – Safety distance reduction factor Parameter Optimization A genetic-algorithm-based optimization process was used to search for optimal VISSIM parameter values. The objective of the process was to minimize the absolute difference between simulated operations measures and observed measures. The objective function is described using the following equation: 1 , , , 100% where , = simulated average measure (travel time or maximum queue length) for one simulation hour for site ; , = observed average measure (travel time or maximum queue length) over one-observation hour for site ; and = number of sites. The optimization was constrained such that the parameter value should be within their ranges, as shown in Table 149. The global optimization toolbox in MATLAB was used to conduct the optimization. During each iteration, the MATLAB program called the VISSIM to run the five test beds, with different parameter values, for ten replications and calculate the average measures cross the replications. The MATLAB genetic-algorithm function updated the parameter values until the optimization objective was reached. For segment sites, the operations measure was the average travel time within evaluation sections. For intersection sites, two optimization strategies were used: (1) minimize the difference of average maximum queue length, and (2) minimize the difference of average travel time. The “best-fit” calibration parameters obtained through this process are described in a subsequent section. Database Summary The simulated operations performance measures from the simulation models with optimized parameters are given in Table 151. In the next section, these values are compared with the observed values given in Table 147.

275 Table 151. Summary of simulated operations performance measures – VISSIM calibration. Site Average Travel Time, s Maximum Queue Length, ft Runs Obs. per Run Mean Std. Runs Obs. per Run Mean Std. 1  10  36  40.76  12.21  2  10  36  44.46  22.79  3  10  36  90.42  24.58  4  10  36  79.24  21.19  5  10  36  76.18  15.98  6  10  36  56.26 5.79  7  10  36  40.43  22.93  10  17  644.21  240.27  8  10  36  30.10  8.39  10  42  117.25  53.77  Note: Std. – standard deviation. Model Calibration Findings and Recommendations Operations Performance Measures A comparison was conducted between simulated operations measures and observed operations measures to validate the model calibration. As shown in Figure 18, the difference between simulated travel time and observed travel time is less than 10 percent at each site. Figure 18. Relative difference between observed and simulated travel time – VISSIM calibration. Table 152 compares the differences between the observed and simulated performance measures that were shown in Table 147 and Table 151, respectively. The p-values shown can be compared with the commonly used threshold value of 0.05 (corresponding to a 95 percent confidence level). The p-value indicates the probability of error when concluding that a difference is significantly different from zero. With one exception, the p-values are all much larger than 0.05 so one cannot confidently conclude the

276 observed and simulated means are different. The one exception corresponds to the maximum queue length at site 7. This exception is discussed in the next few paragraphs. The maximum queue length was examined at two intersection sites (i.e., sites 7 and 8). The difference of the observed and simulated maximum queue length at site 7 is quite large. In fact, it is significantly different from zero (p-value ≤ 0.05, as shown in Table 152). The large difference is caused by the impact of the traffic signal upstream at the intersection of Hillsborough Avenue and Central Avenue. The upstream signal is coordinated with the signal at site 7. The short segment length combined with a near- optimal signal offset result in this segment having very favorable traffic progression. As a result, the observed arrival rate during the red indication at site 7 is lower than that obtained from the simulation model and the observed queue length is significantly lower than the simulated queue length. For site 8, the difference of the observed and simulated maximum queue length is relatively small. In fact, the t-test indicates that the difference is not significantly different from zero (p-value = 0.620). Thus, the calibrated simulation model is rationalized to provide a reasonable estimate of queue length when segment lengths or signal coordination settings produce more typical traffic progression. Table 152. Comparison of observed and simulated operational measures – VISSIM calibration. Performance Measure Site Test Bed Mean of Observation Mean of Simulation t-statistics p-value Travel time, seconds 1 FL1 – EB 42.91 40.76 0.454 0.650 2 FL1 - WB 48.59 44.46 0.512 0.608 3 FL2 – EB 93.85 90.42 0.386 0.699 4 FL2 – WB 82.38 79.24 0.421 0.673 5 FL3 – EB 75.03 76.18 0.209 0.834 6 OR4 – EB 55.25 56.26 0.483 0.629 7 OR4 – WB 46.09 40.43 0.757 0.449 8 OR5 – WB 31.76 30.10 0.446 0.655 Maximum queue length, ft 7 FL4 – EB 400.53 644.21 2.764 0.006 8 OR5 – WB 126.95 117.25 0.496 0.620 In addition to travel times and queue lengths, the saturation flow rates obtained from VISSIM were compared with those obtained from the Highway Capacity Manual (HCM). Figure 19 shows the maximum simulated volume obtained from VISSIM for the test beds with two through lanes. Two trend lines are shown in the figure. One trend line shows the maximum simulated volume obtained from VISSIM using the default model parameters. The other trend line shows the maximum volume obtained when using the calibrated model parameters. Both trend lines indicate that the maximum volume equals the input volume up to the point where the signal is oversaturated and the maximum volume reaches the saturation flow rate. The saturation flow rate produced by the calibrated model equates about 1,750 vphgpl (= 3500/2). It is significantly lower than that produced by the model with default parameters (≈ 2,250 vphgpl). The reduced, calibrated saturation flow rate is believed to better reflect traffic conditions at the study sites, and it is consistent with value obtained from the Signalized Intersections chapter of the HCM.

277 Figure 19. Simulated saturation flow rate – VISSIM calibration. Summary of Recommended Calibration Parameters Based on the discussion above, it can be concluded that the simulation models with calibrated parameter values adequately replicate the field observations for the segment and intersection study sites in the two study regions (i.e., northwest and southeast). The recommended simulation parameter values are identified in Table 153. Table 153. Recommended values for simulation parameters – VISSIM calibration. Category Default and Recommended Values for Four Calibration Parameters W74ax W74bxAdd W74bxMult SafDistFactLnChg Default 6.56 2.0 3.0 0.60 Recommended 6.56 2.5 2.9 0.68 SSAM Simulation Model Calibration This section summarizes the data collected to calibrate the SSAM traffic simulation model such that it accurately predicts crash-related safety surrogate measures at sites in Oregon and Florida. This model was used to create the surrogate measure data needed for the TWLTL vs. non-traversable median study. The objective and scope of the study are provided in Appendix E (in a section titled, Simulation Model Calibration). Data used to calibrate the VISSIM software tool are summarized in the previous section. Field Data This section summarizes the field data that were collected and used to calibrate the VISSIM model. The summary of these data includes a description of the sites at which data were collected, the database

278 organization, data collection techniques, data reduction procedures, and statistics used to describe the collected data. Study Sites The two intersection sites (i.e., sites 7 and 8) described in the section titled, VISSIM Simulation Model Calibration were used to provide the data for the SSAM calibration. Both sites are signalized intersection approaches in urban areas and have moderate to high levels of bicycle, transit, and truck traffic volume. The basic characteristics of the two sites were summarized in Table 145. Database Organization Field data were collected at the two sites, including road characteristics inventory, traffic volume, signal timing, and safety measures. The inventory, volume, and timing data that were collected are listed in Table 146. The safety measures in the database are described in Table 154. Table 154. Data fields in database for calibration study – SSAM calibration. Category Database Data Fields Evaluation Period Source Safety performance measures Bus conflicts  Bus rear-end conflicts  Bus lane-change conflicts Eight hours Field video Truck conflicts  Truck rear-end conflicts  Truck lane-change conflicts Eight hours Field video Data Collection Techniques and Reduction Procedures The data collection techniques and data reduction procedures for the SSAM calibration are similar to those used for the VISSIM calibration. For the SSAM calibration, additional data items include the counts of bus- and truck-related conflicts. To this end, a researcher reviewed the field videos to identify and categorize the observed conflicts. Four hours of video were reviewed for the Florida site and four hours of video were reviewed for the Oregon site. These hours occurred during the weekend to avoid saturation status in weekday peak hours. The researcher identified bus- and truck-related conflicts using the criteria in the following list.  Any traffic event involving a bus or a truck;  The event has an observable break, deceleration, or swerving maneuver to avoid a potential collision between a bus (or truck) and another vehicle; or  Objective judgment on conflict type: rear-end or lane-change. Database Summary The collected bus- and truck-related conflicts are summarized in Table 155.

279 Table 155. Summary of observed traffic conflicts – SSAM calibration. Site Test Bed Scenario Conflict Frequency by Vehicle Type Bus Truck 7 – EB, Hillsborough Ave. at Florida Ave, FL4 13:00 – 14:00 Saturday 6 8 14:00 – 15:00 Saturday 2 8 16:00 – 17:00 Saturday 4 15 17:00 – 18:00 Saturday 1 10 8 – WB, SE Division St. at SE 162nd Ave. OR5 8:00 – 9:00 Saturday 4 2 9:00 – 10:00 Saturday 2 5 10:00 – 11:00 Saturday 2 4 11:00 – 12:00 Saturday 4 5 Total: 25 57 Simulation Data This section summarizes the simulation data that were compared with the field data to calibrate the SSAM model. The summary of these data includes a description of the simulation test beds, data collection process, model calibration process, and statistics used to describe the collected data. Test Bed Characteristics The two intersection test beds prepared for the VISSIM calibration were also used in the SSAM calibration. The test bed characteristics are described in Table 145. Data Collection Process Eight simulation scenarios were created based on the two intersection test beds. The VISSIM calibration parameters identified in the previous section were used with each test bed. The traffic inputs and signal timing were updated according to the collected data in the field and from the video recordings. Each scenario represented a one-hour simulation period plus a 15-minute warm-up time. A MATLAB program executed 10 simulation runs for each scenario in VISSIM. Two output files were generated from each simulation replication. These files are identified in the following list.  *.trj – SSAM formatted vehicle trajectory file  *.fhz – Vehicle input file with vehicle type and vehicle ID information In total, 80 trajectory files, representing various traffic conditions in eight hours, were produced for the SSAM calibration. SSAM Calibration Process The objective of the simulation model calibration was to develop one set of “global” calibration parameters for all sites used in the study. It was rationalized that these parameters would then provide acceptable results sites in other regions. To achieve this objective, the following steps were conducted.

280 Conflict Identification The MATLAB program called the SSAM 3.0 software to identify bus- and truck-related conflicts from the trajectories produced by VISSIM. The SSAM parameters that were used to identify these conflicts are identified in the following list.  Time-to-collision (TTC);  Post-encroachment (PET); and  Angle threshold for conflict type. As the result, a conflict list was produced from the SSAM model for each trajectory. A filtering algorithm was coded in MATLAB to select the target conflicts. This algorithm used the rules described in the following list.  The simulation time of a conflict event should be greater than the ending time of the warm-up period (> 900 s);  The first vehicle or the second vehicle should be bus or truck; and  The location of conflict event should be within the evaluation section, as shown in Figure 17. The MATLAB program retrieved the identified conflicts (i.e., bus- or truck-related rear-end and lane- change conflicts) for each simulation replication and matched them to each scenario. The final data were exported to a database using a CSV format for the SSAM calibration. Parameter Optimization A genetic-algorithm-based optimization process was used to search for optimal SSAM parameter values. The objective of the process was to minimize the absolute difference between simulated conflict counts and observed counts. The objective function is described using the following equation 1 , , , , , , 100% where , , = simulated conflict frequency (transit- or truck-related) for one simulation hour for site ; , , = observed conflict frequency (transit- or truck-related) over one-observation hour for site ; = number of sites; and = number of study hours. Optimization constraints are identified in the following list.  0 ≤ TTC ≤ 5.0, and  0 ≤ PET ≤ 10.0, and  TTC < PET. The conflict type was not considered in this calibration because it was difficult to accurately identify from the videos recorded in the field (due partly to the camera location and picture quality). The default parameters for conflict types in the SSAM model were used. The global optimization toolbox in MATLAB was used to conduct the optimization. During each iteration, the MATALB program called the VISSIM to run the two test beds, with different parameter values, for ten replications and calculate the average measures cross the replications. The MATLAB genetic-algorithm function updated the parameter values until the optimization objective was reached.

281 The “best-fit” calibration parameters obtained through this process are described in a subsequent section. Database Summary The simulated conflict frequencies from the simulation models with optimized parameters are given in Table 156. Table 156. Summary of simulated traffic conflicts – SSAM calibration. Site Test Bed Scenario Conflict Frequency by Vehicle Type Bus Truck 7 – EB, Hillsborough Ave at Florida Ave FL4 13:00 – 14:00 Saturday 2 5 14:00 – 15:00 Saturday 5 13 16:00 – 17:00 Saturday 0 8 17:00 – 18:00 Saturday 4 9 8 – WB, SE Division St at SE 162nd Ave OR5 8:00 – 9:00 Saturday 4 0 9:00 – 10:00 Saturday 1 5 10:00 – 11:00 Saturday 2 4 11:00 – 12:00 Saturday 5 15 Total: 23 59 Model Calibration Findings and Recommendations Safety Performance Measures A comparison was conducted between the simulated safety measures and observed safety measures to validate the model calibration. Table 157 compares the differences between the observed and simulated performance measures that were shown in Table 155 and Table 156, respectively. The p-values shown can be compared with the commonly used threshold value of 0.05 (corresponding to a 95 percent confidence level). The p-value indicates the probability of error when concluding that a difference is significantly different from zero. The p-values are all much larger than 0.05 so we cannot confidently conclude the observed and simulated means are different.

282 Table 157. Comparison of observed and simulated safety measures – SSAM calibration. Site Scenario Conflict Frequency by Vehicle Type Bus Conflict Frequency Truck Conflict Frequency Simulated Observed Simulated Observed 7 13:00 – 14:00 Saturday 2 6 5 8 14:00 – 15:00 Saturday 5 2 13 8 16:00 – 17:00 Saturday 0 4 8 15 17:00 – 18:00 Saturday 4 1 9 10 8 8:00 – 9:00 Saturday 4 4 0 2 9:00 – 10:00 Saturday 1 2 5 5 10:00 – 11:00 Saturday 2 2 4 4 11:00 – 12:00 Saturday 5 4 15 5 Total: 23 25 59 57 Mean: 2.875 3.125 7.375 7.125 Standard Deviation: 0.667 0.581 1.742 1.445 Difference between Simulated and Observed: 8% 3.5% t statistics: -0.261 0.137 p – value: 0.802 0.895 Summary of Recommended Calibration Parameters Based on the discussion above, it can be concluded that the calibrated simulation models (VISSIM and SSAM) adequately replicate the observed bus- and truck-related conflicts for the two intersection study sites in the two study regions (i.e., northwest and southeast). The recommended SSAM parameter values are identified in Table 158. Table 158. Recommended values for simulation parameters – SSAM calibration. Category Default and Recommended Values for Four Calibration Parameters TTC, s PET, s Angle Threshold for Rear End, degrees Angle Threshold for Crossing, degrees Default 1.5 5.0 30 80 Recommended 1.3 5.0 30 80 References Kaparias, I., M. Bell, J. Greensted, S. Cheng, A. Miri, C. Taylor, and B. Mount. (2010). “Development and Implementation of a Vehicle-Pedestrian Conflict Analysis Method.” Transportation Research Record: Journal of the Transportation Research Board, No. 2198. Transportation Research Board of the National Academies, Washington, D.C. pp. 75-82. Hunter, W., J. Stewart, J. Stutts, H. Huang, and W. Pein. (1998). A Comparative Analysis of Bicycle Lanes Versus Wide Curb Lanes: Final Report. Report No. FHWA-RD-99-034. Federal Highway Administration, Washington, D.C.