Enhancing Safety and Operations at Complex Interchanges With Improved Signing, Markings, and Integrated Geometry

Chapter 6. Simulator Study

To prepare for the simulator study, the project team conducted the tasks outlined in earlier sections of this report. The literature review identified existing research, results, and limitations. The practices evaluation provided insights into the existing practices of agencies throughout the United States and Canada. The identification of attributes contributing to complexity and the development of topic areas provided focus for the simulator study and the selection of sites for the field study. The field study also served to identify issues with merit for further examination in the simulator environment. All of these tasks generated information used to prioritize conditions that make an interchange complex and identify challenges associated with treatments at these locations. For the purposes of this project, complexity is defined as follows:

Complexity occurs when the choice of more than one movement is available from a lane or group of lanes where the decision points occur successively in close proximity.

As part of task 4, a simulator study was conducted to experimentally evaluate driver lane selection in complex interchange situations. Complex interchanges typical of the existing field applications were designed, and multiple alternative approaches to guide signing were developed for each interchange layout. The effectiveness of driver decisionmaking was evaluated in terms of whether drivers made accurate lane choices (i.e., those that led to arriving at the correct location) and in terms of the potential impacts to safety and efficiency associated with the timing of these decisions and with making ULCs because of poor comprehension or inadequate information.

Objective

The simulator study was conducted to understand driver behavior at complex interchanges and identify signing characteristics that are related to more effective lane selection. The simulator study results contribute to the literature describing driver behavior in the presence of different signing information and can be used to develop guidance that helps practitioners make effective choices for common complex interchange scenarios.

Research Questions

The simulator study addresses the following research questions:

• Which signing alternatives contribute to the fewest number of lane changes?
• Which signing alternatives exhibit the highest number of inaccurate lane changes?
• Do drivers in the option lane leave the option lane as through traffic? As exiting traffic?
• Is option lane use consistent with the downstream exit direction on the exit ramp?
• Do drivers initially react to information upstream of the mainline exit, only to change lanes into an incorrect lane for the instructed destination, either prior to or beyond the exit?

Interchange Design and Signing Approaches

Four interchange layouts were developed to be representative of existing configurations of interchanges exhibiting attributes related to complexity (see figure 24). While two layouts contained similar geometric designs, each consisted of a different exiting lane configuration downstream of the primary exit.

Source: FHWA.

A. Interchange layout A.

Source: FHWA.

B. Interchange layout L.

Source: FHWA.

C. Interchange layout C.

Source: FHWA.

D. Interchange layout E.

Figure 24. Graphic. Interchange layouts considered in this study.

Table 42 shows how each simulator study layout provided various geometric design features that are present in real-world complex interchanges.

X indicates attributes present at the interchange.

—These attributes were not present at the interchange.

TH 62 = State Trunk Highway 62.

Each interchange layout addressed a specific configuration of exiting lanes in one discrete geometric design. A variety of signing alternatives were applied to each layout. These alternatives generally consisted of a single approach to signing for an interchange. In the present simulator study, participants encountered a unique combination of an interchange layout, signing alternative, destination, and starting lane position. An overview of these is provided here, and more detail follows in subsequent sections.

Appendix B provides a complete catalog of the signing alternatives advanced for testing in the simulator in conjunction with diagrams of the geometric layouts associated with each. Each signing alternative was designed to accommodate the three possible destinations for each of the alternatives in a given layout. These movements are considered THRU (T), LEFT (L), and RIGHT (R). Participants were told that their task was to follow the signs toward Greenville; Greenville was always the destination to which they were instructed to drive. For example, a participant might be trying to navigate to Greenville on Route 28 without being told a cardinal direction for Route 28. Using the information provided on overhead guide signs, the participant would either continue THRU to Greenville or would exit the interchange to the RIGHT or the LEFT toward Greenville based on the experimental scenario. As there is no LEFT movement in layout A, a destination of “L” for this layout represents the second RIGHT movement.

Each layout and signing alternative was tested as a discrete exercise with a single starting lane assignment (lanes number 1 to 4 from left to right). As shown in table 43, there was a total of 12 signing alternatives, each of which allowed for between 2 and 9 total possible experiences based on starting lane and destination combinations, for a total of 87 possible discrete simulator experiences.

L = left; R = right; T = thru.

—Not applicable.

Each layout is described in more detail in the subsections that follow, along with the signing alternatives that were evaluated for each.

Layout A

Layout A consists of a limited-access roadway segment with three lanes in one direction and an auxiliary lane on the right. The upstream portion of layout A in advance of the first exit is 13,200 ft long. The auxiliary lane terminates at an exit ramp with an adjacent option lane. The two-lane exit ramp is approximately 1,980 ft in length and terminates in a downstream split, where the left lane continues as the left-hand movement and the right lane exits as a right-hand movement.

Layout A’s characteristics may be challenging to the driver because there are two movements available from the right-hand lane. The first movement, a non-mandatory exiting movement, occurs upstream but in close proximity to the second movement, the mandatory exiting movement. Drivers may move out of the right lane anticipating the first exit, depending on signing, reducing segment capacity and increasing conflicts due to lane changes between the two exits. To evaluate what signing would best convey lane selection information to drivers, three signing alternatives were developed for layout A.

Summary of Hypotheses for Layout A

Table 44 summarizes the description and hypotheses for layout A. Alternative A1 is expected to perform poorly for ULCs upstream of the first exit, until downstream of the first exit, destination LEFT for drivers who are taking the second exit. Alternative A2 is expected to perform more poorly than alternative A1 for lane changes into lane 4 in advance of the first exit for vehicles assigned to destination RIGHT. Alternative A3 is anticipated to result in a larger number of vehicles using lane 4 upstream of the first exit for both destination RIGHT and destination LEFT vehicles.

Layout L

Layout L consists of a limited-access roadway segment with three lanes in one direction and an auxiliary lane for a downstream left-hand exit. The upstream portion of layout L in advance of the first exit is 10,560 ft long. The first exit ramp is a single-lane exit with a standard tapered departure. The exit gore areas are separated by 1,320 ft. The left-hand auxiliary lane terminates as a single-lane, left-hand exit ramp as the second exit.

The characteristics of layout L are challenging to the driver because left exits are less common, and a disruption in freeway flow characteristics is more likely to occur with left exits because slower traffic will move into the generally higher-speed left lanes. Compounding the issue of left exits, in situations where a continuing lane terminates as the left-hand mandatory exiting movement, significant lane change events will occur as through traffic moves into right-hand lanes. To evaluate what signing would best convey lane selection information to drivers, two signing alternatives were developed for layout L.

Summary of Hypotheses for Layout L

Table 45 summarizes the description and hypotheses for layout L. It is anticipated that drivers assigned destination “LEFT” will perform equally well in both scenarios, because overhead advance guide signing is provided in advance of the exit, depicting both the exiting movement and the distance to the exit. However, for drivers assigned destination “THRU,” it is expected that drivers in alternative L2 will perform better because of the presence of positive guidance directing them into the through lanes ahead of the left-hand exit.

Layout C

Layout C consists of a limited-access roadway segment with three lanes in one direction and an auxiliary lane. The upstream portion of layout C in advance of the primary exit (sections 1 and 2) is 10,560 ft long. The auxiliary lane terminates at an exit ramp with an adjacent option lane. The two-lane exit ramp is approximately 1,980 ft in length and terminates in a downstream split, where the left lane continues on as the left-hand movement and the right lane exits as a right-hand movement.

Layout C’s characteristics are challenging to the driver because drivers must first make an upstream lane selection (prior to the mainline exit) that may be predicated on their downstream lane selection, depending on the driver’s driving style. For example, drivers who will correctly exit destination “LEFT” would most expeditiously choose to use the option lane on the upstream segment. However, if upstream information is absent or unclear, drivers may choose to use the right-most lane (in this case, the mandatory movement lane) to obtain some assurance that they are indeed taking the exit. To evaluate what signing would best convey lane selection information to drivers, four signing alternatives were developed for layout C.

Summary of Hypotheses for Layout C

Table 46 summarizes the description and hypotheses for layout C. It is anticipated that drivers will make the highest number of upstream lane changes for the signing in alternative C3 because it makes a clear lane assignment upstream of the exit from the mainline roadway. Alternatives C1, C2, and C3 are expected to perform equally in terms of upstream lane choice, although option lane use for exiting traffic may be higher for alternatives C1 and C2 because the signing in those two alternatives does not indicate a multiple movement from the option lane.

Layout E

Layout E consists of a limited-access roadway segment with three lanes in one direction and an auxiliary lane. The upstream portion of layout E in advance of the primary exit (sections 1 and 2) is 10,560 ft long. The auxiliary lane terminates at an exit ramp with an adjacent option lane. The two-lane exit ramp is approximately 1,980 ft in length and terminates in a downstream split, where the left lane continues as the left-hand movement and the right lane exits as a right-hand movement.

Layout E’s characteristics are challenging to the driver because, as in the scenarios for layout C, drivers must first make an upstream lane selection (prior to the mainline exit) that may be predicated on their downstream lane selection, depending on the driver’s driving style. In the case of layout E, drivers will be able to access destination “LEFT” from either exiting lane (which could only be done from the left-most exiting lane for layout C), but the destination “RIGHT” movement will require a lane change to the right-most lane. Three signing alternatives were developed for layout E.

Summary of Hypotheses for Layout E

Table 47 summarizes the description and hypotheses for layout E. In general, drivers driving in scenario 3 are expected to exhibit better performance as additional positive guidance elements are added, such as the upstream supplemental guide sign in alternative E2. Despite the incorrect signing of the right-hand lane as solely used for the right-hand movement on the C/D roadway in alternative E3, that signing is expected to produce more upstream lane changes into the right lane for destination “RIGHT” drivers while reducing the utility of the right-hand lane for destination “LEFT” drivers.

Research Design

This section describes the key elements of the research design that involved the recruitment of drivers to perform a driving task in a partial cab driving simulator. The details of this study are described in greater detail in the subsections that follow.

Independent and Dependent Variables

Three factors were of particular interest to this study and served in statistical models as dependent variables: accuracy of lane selection, ULCs, and lane selection distance (LSD). These measures were further assessed within different segments of the interchange, separated by decision points, or locations where the participant is presented with options for how to proceed.

Figure 25 shows an example of the two decision points on layout E. A participant’s accuracy is calculated both in terms of overall accuracy (i.e., getting to his or her designated destination) and in terms of accuracy at individual decision points during the simulation. Consider an example where the designated exit was to the right of the downstream split (ramp E3), and the participant successfully exited the mainline, but stayed to the left (ramp E2). In this example, accuracy for the participant would have been recorded as correct at decision point 1 (DP1), incorrect at decision point 2 (DP2), and incorrect overall. A similar approach was used for ULCs and LSD. However, in the case of LSD, no value was recorded for overall LSD.

Source: FHWA.

Figure 25. Graphic. Layout E decision points.

Accuracy

Lane accuracy was measured on a bivariate scale (correct or incorrect). Accurate drives were those in which participants ultimately navigated to the given destination, regardless of intermediate maneuvers. All lanes that allow the participant access to the correct destination were considered correct, and no priority or preferences were assigned (i.e., optimal).

ULCs

For this study, a lane change was defined as the moment when more than 50 percent of the participant’s simulated vehicle moved into an adjacent lane. (Directional signals (turn signals) were not enabled during the study.) For a given interchange and destination, a certain number of lane changes were required to reach the correct destination (minimum number of lane changes); those in excess of this minimum were considered ULCs. ULCs were calculated in two ways: (1) across the entire interchange and (2) per the segment of road leading up to each decision point. In both cases, the initial starting lane position is assigned based on the scenario being run. In the latter example, the participant’s starting lane position approaching DP2 is not assigned; rather, it is based on the lane the participant selected at DP1. Because of this subtle difference between the starting lane assignments in both approaches, the sum of ULCs in the latter approach (counted per segment) does not necessary equal those in the first approach (counted across the interchange). While the first approach is useful to get an idea of potential navigation issues, in general, with a given layout and signing alternative, the second approach allows further insight into where these ULCs are occurring.

LSD

The location of the participants’ final lane change tells us how far in advance of the exit their selection was finalized. This variable was measured in feet from the exit location (the point at which they could not change their mind). This measure will reflect the degree to which participants waited until the last minute to change lanes. However, it cannot precisely describe when participants decided what their lane choice should be.

Accuracy reflects the outcome of each drive; whereas, the ULC and LSD variables reflect the decision process. Signing options that produce more accurate outcomes may improve driving reliability, and those that do so with the least confusion (fewer ULCs, better LSD) could improve safety.

What Signing Alternative Results in the Best Driver Performance for Each of the Four Interchange Layouts?

Driver performance on each of the four interchange layouts was not compared against one another. Instead, the focus of this research was to identify which signing alternative results in the best performance for a particular interchange layout. For instance, in layout A, under what signing guidance do drivers make the most accurate and most efficient lane changes (A1, A2, or A3)?

Table 48 shows the 12 interchange signing alternatives seen by all drivers that create a repeated measures factor.

—Not applicable (no additional signing alternatives for this layout).

How Do the Characteristics of the Required Lane Maneuver Affect Performance?

In the real world, drivers approach an interchange from various starting positions based on their origin, driving experience to that point, and personal preferences. To account for this, each participant was exposed to a variety of starting position (1, 2, 3, and 4) and destination (left, right, through/straight) combinations when approaching the interchanges. Based on the interchange layouts and signing alternatives in this study, and as shown in table 48, this resulted in 87 combinations, or 87 possible discrete simulator experiences.

To minimize the amount of time participants used the driving simulator, not every driver could see every one of these 87 discrete simulator experiences. Thus, the 87 combinations were placed into 9 different groups, or scenes, as shown in table 49. For example, A1-3L refers to interchange layout A, signing alternative A1, starting lane 3, and destination LEFT. Not all drivers saw every combination, making each combination a between-drivers variable. Instead, each participant was assigned to one of the nine groups, or scenes, and these groups were designed such that each participant saw every interchange layout and signing alternative, but only saw a subset of starting lane and destination combinations. This process was semi-random so that no one participant would encounter similar combinations of starting lane and destination.

Note: Each participant was assigned to one of these scenes for the experimental session.

Controlling for Order Effects

Two different interchange layout orders were used to control for potential order or learning effects. An order was generated randomly to produce order A; order B was produced by reversing order A. Table 50 shows the two different orders. Each scene above was then ordered accordingly, creating 18 scenes (scenes 1A/1B, 2A/2B, 3A/3B, 4A/4B, 5A/5B, 6A/6B, 7A/7B, 8A/8B, and 9A/9B). Each order contains the 12 interchange layouts that the study participants encountered.

Detecting Differences in Driving Performance

The statistical power of the proposed experiment was estimated before data collection began using several assumptions. The standard value of power, (1 – β) = 0.80, was used, but the familywise error rate, α = 0.05/6 = 0.0083, was adjusted for the six pairwise comparisons possible with four signing alternatives. Various samples sizes per interchange-signing combination were calculated separately for the two main variables of interest, with accuracy measured as a proportion and number of ULCs as independent group means.

Statistical power was assessed to determine the optimal number of participants to complete this study and show statistically reliable and valid results. A power analysis showed that, if 120 participants completed this study, this would allow for the detection of accuracy differences as small as 14.4 percentage points. The farther the two groups are from 100 percent, the larger the minimum detectable difference becomes (the less powerful the test becomes). Accuracy is expected to be high overall, but if the best group in a pairwise comparison is 75 percent accurate, the smallest detectable difference with 100 drivers is 23 percentage points.

The power to detect differences in discrete variables (such as the number of ULCs) is calculated differently than with proportions. Participants may only commit a small number of ULCs, perhaps zero ULCs or one to two ULCs. Expressed statistically, these represent two Poisson random variables with means of 0 and 1.5. The common standard deviation between the two groups is 1.1. If 120 participants complete this study, this would allow for the detection of ULC differences as small as 0.51 ULCs (an improvement of 0.05).

The above calculations are at the interchange layout and signing alternative level. Starting lane and destination will be equally represented in each interchange-signage combination; therefore, aggregating over them (for comparisons of signing alternatives within interchange) is appropriate and valid. Examining for effects due to starting lane or destination is not likely to yield strong statistical conclusions regarding accuracy, but substantially small differences in ULCs may still be detectable. Table 51 uses the previously stated assumptions and adjusts the familywise error rates according to the number of potential pairwise comparisons to calculate the minimum detectable difference for each comparison with different total numbers of participants.

Method

This section describes the participants, apparatus and materials, stimuli, and procedures used for conducting the study.

Participants

This study included a sample of 121 research participants (60 male and 61 female) in 3 different geographic areas: Orlando, FL; Myrtle Beach, SC; and Gainesville, VA. Participants ranged in age from 18 to 83 yr (mean = 44.9). Each participant possessed a valid U.S. driver’s license and passed a vision screening with at least 20/40 vision in at least one eye (corrected if necessary). Participants were paid 70 dollars for their participation.

Of the 121 participants who completed the study, half were in the younger age group (18 to 45 yr, mean = 29.7 yr) and half were in the older age group (46+ yr, mean = 60.4 yr). Each age group (younger and older) was evenly distributed between males and females. Of the 133 participants who began the experiment, 5 were stopped due to issues with the laboratory and/or simulator, 4 were stopped due to simulator sickness, and 3 participants did not complete for other reasons.

Participants were randomly assigned to an experimental condition representing 1 of the 18 scenes described above. However, the project team sought to achieve a balance across gender, age, and location within each condition.

The recruitment process used a variety of advertising methods, including flyers in community centers and at local businesses, online ads, and word-of-mouth. The entire experiment (including instructions, informed consent, questionnaires, and debriefing) took approximately 90 min to complete. Each participant was paid 60 dollars for completing the study as well as a 10-dollar bonus for attempting to make as few lane changes as necessary to complete the driving task accurately.

Apparatus

A Mobile Human Factors Laboratory (MHFL), shown in figure 26, was used to collect data. The mobile laboratory is a cutaway van with dimensions of approximately 7 x 20 ft and includes a comfortable, climate-controlled laboratory space; a high-end, business-grade computer capable of advanced graphics generation; a 65-inch display; specialized software for sign display and testing; and a driving simulator platform that can be added or removed to the mobile device, as required. The interior has been configured to limit the view of the researcher’s workstation from the participant space, permitting unobtrusive monitoring. The MHFL is equipped for cross-country travel, enabling the testing of different populations of road users in multiple regions. This vehicle platform has a low operational cost, and the lead time for its deployment is short compared to mobilization of testing at laboratory facilities with large workloads. The facility is comfortable for visiting participants, being outfitted with a climate-controlled waiting area and workspace, including windows in the waiting area for natural light.

Source: FHWA.

Figure 26. Photo. MHFL.

MiniSim™ Driving Simulator

The University of Iowa’s National Advanced Driving Simulator MiniSim™ suite is used within the MHFL. The MiniSim™ repackages the framework and technology of the National Advanced Driving Simulator-1 driving simulator into a mobile platform. The MiniSim™ suite includes all of the tools required to completely customize and build a driving simulator study, as well as test and analyze the findings, including:

• Tile Mosaic Tool to develop detailed roadway network databases.
• Interactive Scenario Authoring Tool to author scenarios and program behavior of deterministic and autonomous dynamic objects (i.e., other traffic and pedestrians) as well as TCDs.
• MiniSim™ to run the driving simulation and collect data.
• ndaqTools to assist in reducing the raw MiniSim™ output into measures.

MiniSim™ Development Approach

The MiniSim™ works from a tiled approach, so each segment of the test drive was developed as separate tiles. These tiles were then combined into the appropriate sequences per the direction of the experimental team. Signing was developed by a traffic engineer and placed within the sequences per the direction of the engineer. Special attention was given to ensure data accuracy in conjunction with visual accuracy to maintain data integrity in preparation for data reduction.

Materials

The following materials were developed in paper and pencil format.

Motion Sickness History Screening Form

This screening was administered verbally prior to scheduling a participant for the study to identify people who might be likely to experience simulator sickness. The scoring criteria were used to discourage participation as appropriate.

Record of Informed Consent

The informed consent document describes the study, participant and researcher responsibilities, risks, risk mitigation plan, and participant consent.

Vision Screening Form

The Vision Screening form was used to track participants’ visual acuity as determined by a Snellen chart.

Simulator Health Screening

This was used as a secondary screening, after participants arrived for their appointment, to help identify participants who might be likely to experience simulator sickness.

Simulator Sickness Questionnaire

The Simulator Sickness Questionnaire (SSQ) is given before the participant drives in the simulator, during certain breaks between driving sessions, and at the end of the experiment. This questionnaire is also administered whenever a research participant becomes ill and periodically during the recovery period thereafter. The SSQ is designed to detect and monitor simulator sickness.

Instructions

The experimenters used a written guide to provide verbal instructions to participants.

Receipt for Payment

This was completed and signed by participants upon payment to track study funds.

Caution Acknowledgement Release

If a sick participant refused to take the SSQ and/or left the research facility without recovering, she/he was requested to sign a caution acknowledgement waiver.

Debriefing Statement

Participants read a brief debriefing statement that described the goals of the study.

Stimuli

As described in previous sections, multiple signing alternatives were developed for each interchange layout. Each layout consisted of four main sections. In general, the layout lengths within a simulator tile were multiples of 660 ft, and the length of each section was set by those multiples, as ¼-mi intervals are typically observed for signing in practice. The expected legibility distance for the guide signs in this study was based on an anticipated in-simulator legibility distance of approximately 500 ft, consistent with a 30-ft legibility distance for every 1 inch of letter height on the sign. The section coverage was previously described in this report. Based on practice evaluations, field reviews, and field data collection at similar sites, the project team developed eight potential geometric layouts, each representing a segment of motorway-grade facility approximately 3 mi in length. While some layouts were related, each consisted of a different exiting lane configuration. From those initial eight layouts, four were advanced for development in the simulator.

All traffic signing and pavement markings to be used in the simulation scenarios were designed using the principles identified in the MUTCD and the SHS. Specific design details were adapted from the policies of MnDOT, WSDOT, and Florida Department of Transportation.

Appendix B provides a complete catalog of the signing alternatives advanced for testing in the simulator in conjunction with diagrams of the geometric layouts associated with each.

Roadway Segmentation

Section 1 includes the mainline of the roadway upstream of any guide signing. For all layouts, this distance is set at a maximum of 5,280 ft. Within section 1, participants will drive 1,320 ft prior to seeing the overhead sign that assigns them to the starting lane position for that scenario. Subsequent to that, participants will observe a pair of speed limit signs no more than 2,640 ft upstream of the first guide signs for that signing alternative.

Section 2 accommodates the signing upstream of the first exit, beginning with the first guide sign, and this distance varies from 5,280 to 7,920 ft (1½ mi). For layout A, where the maximum distance upstream of the first exit is 1½ mi, the overall length of sections 1 and 2 is 11,880 ft.

Section 3 covers the distance between the first and second exits. For layouts A and L, that distance is measured along the mainline and is 1,320 ft. For layouts C and E, that distance is measured along the C/D roadway and is 1,980 ft, which includes a 400-ft exit taper, a 600-ft horizontal curve, and approximately 900 ft of distance prior to the split.

Section 4 covers the distance from the final decision point to the end of the tile, where all lanes have rejoined the mainline for a four-lane configuration to match the starting configuration of all of the tiles. This section is typically 3,960 ft long and consists of horizontal curves, tapered lane additions, and lane reductions that provide for participant driving into the four-lane section that will connect to the next tile, for a seamless participant experience.

Starting Lane Indication

All participants encountered each of the 12 layout-signing combinations once, and each participant was assigned to 1 starting lane–destination combination per layout-signing combination (as illustrated in table 49). The participants were directed to a destination and informed of the starting lane position using a sign consistent with figure 27.

The sign was purposely designed to not mimic a guide sign and to avoid providing information with conventional symbols, such as a route marker. A down arrow was provided over each lane so that participants could count lanes and determine which lane to choose based on their position from the right or left edgeline.

Source: FHWA.

Figure 27. Graphic. Overhead sign that directed participants to a starting lane using the asterisk symbol location in one of four lanes.

Procedures

Prior to participation, potential participants were screened for susceptibility to motion sickness. If willing and eligible, participants were then scheduled for participation. Table 52 provides an overview of the participant experience.

Intake

When participants arrived for their appointment, they were first asked to complete a basic visual screening to ensure a minimum of 20/40 acuity in at least one eye (corrected if necessary). Participants were instructed to stand the appropriate distance from a Snellen eye chart. After receiving instructions and completing the eye chart, the experimenter recorded their visual acuity on the vision screening form. Next, participants were asked to read and sign the Record of Informed Consent. After obtaining informed consent, participants were given a brief health survey; the goal of this questionnaire was to identify participants who might be likely to experience simulator sickness.

Participants were informed that they are participating in a research study to evaluate driving behavior in a driving simulator study of roadway signs. They were given a brief overview of the study process (i.e., they were told there would be a practice drive, two main drives, and a follow-up questionnaire).

Training

Prior to beginning the experimental drives, participants were exposed to a brief 3- to 5-min practice scenario. The experimenters explained the simulator to participants and then had them complete a practice drive to familiarize themselves with the driving simulator. The practice scenario consisted of a four-lane roadway, on which participants practiced accelerating, changing lanes, exiting a roadway, and stopping. The roadway segment used in the practice drive looked similar to those that might be seen in the experimental drives; however, no guide signs were present in the practice drive. A starting lane sign was present at the beginning of the practice drive; this gave the experimenter an opportunity to show to participants what this sign looked like before beginning the experimental drives. Although the practice drive only lasted about 3 to 5 min, this drive was repeated as many times as necessary until the experimenter and participant both felt comfortable moving forward to the experimental drives.

Test Scenarios

Although specific distances may have varied slightly between layouts, each interchange layout should have taken approximately 3 to 3.5 min to traverse. Therefore, the entire experiment (12 runs per participant) consisted of approximately 40 min of driving. The 40 min were divided into two separate drives, each of which consisted of half (six) of the runs assigned to that scene and presented in the orders as discussed in the previous sections of this report. Therefore, each experimental drive lasted about 20 min with a break in between.

For the experimental drives, participants were told that their task for both drives was to follow the signs to continue toward Greenville; Greenville was always the destination that they were to drive toward. In other words, participants’ target destination was always Greenville (i.e., they will be instructed to always follow the signs to continue toward Greenville) on Route 28 without being told a cardinal direction for Route 28, which varied between scenarios. The use of a single target destination was undertaken so that participants were not confused by the need to remember a new target destination for each interchange. Using the information provided on overhead guide signs, participants would either continue through to their target destination, or they would exit the interchange to the right or the left toward their target destination. Participants were instructed to maintain the posted speed limit (65 mi/h), drive as they normally would, and determine what to do to reach their destination most efficiently.

Participants were reminded of the starting lane sign and told that they would see these signs occasionally throughout the experimental drives. They were instructed to, whenever they saw one of these signs, enter the lane over which there was an asterisk. Once in the appropriate starting lane, they could then make any lane changes necessary to complete the driving task.

To prevent participants from changing lanes too frequently or too early (such as moving into the right lane out of habit or comfort, rather than necessity), drivers were instructed to avoid making any ULCs (i.e., to only make the lane changes needed to complete the task of driving toward Greenville). To reinforce this, participants had the opportunity to earn the 10-dollar bonus (in addition to the stipend that they were already receiving to complete the study) by using the fewest lane movements possible to complete the driving task accurately and by doing their best to maintain the posted speed limit.

The instructions to participants are located in appendix C.

Close-out

Following the completion of the test scenarios, each participant was debriefed. They were paid their stipend for participating and were excused from the study.

Data Reduction and Analysis Approach

Data captured from the MiniSim™ include 69 variables at 60 Hz and another 66 variables at each change of state (e.g., cruise control: on, off). All variables were captured and recorded for all participants. For this study and the resulting analysis, the set of variables shown in table 53 was extracted from the MiniSim™ data acquisition (DAQ) files for analysis. In some cases, a single variable, as defined by the MiniSim™, contains several arrays of information. As an example, the variable SCC_Lane_Deviation contains information on (1) whether the vehicle is on a road or off-road, (2) the lane or corridor the vehicle is on, (3) the vehicle’s deviation from the center of the lane, and (4) the width of the corridor or lane.

The Python package undaqTools (version 0.2.3) was used to extract the variables from the MiniSim™ DAQ files into comma-separated values (CSV) files for each participant drive (i.e., string of six interchanges).⁽³⁹⁾ Quality assurance testing was completed on the raw CSV files before data reduction to confirm that each file was complete without data loss. Data reduction scripts developed by the project team were then used to reduce the raw CSV files into three datasets for analysis: lane selection per decision point, ULCs per interchange, and lane change information. After data reduction, a combination of quality assurance testing and visual inspection was completed to confirm accuracy of the reduced data.

Table 54 shows the scenario details. These data were developed when building each of the driving scenarios and not extracted from the simulation output, but they were critical in developing and analyzing the reduced data. This dataset includes one row of data for each possible simulation configuration.

Table 55 shows the variables captured in the first dataset, lane selection per decision point. This dataset included 2 rows of data for each participant, for each interchange, making 24 rows of data for all participants that successfully completed the full procedure.

Table 56 shows the variables captured in the second dataset, ULCs across the interchange. As discussed under Research Design, the number of ULCs calculated across the interchange does not necessarily equal the sum of ULCs per decision point. This dataset included 1 row of data for each participant, for each interchange, making 12 rows of data for all participants that successfully completed the procedure.

Table 57 shows the variables captured in the third dataset, lane changes. This dataset includes a row of data for each lane change made within the study area (i.e., once the participant enters their starting lane until the DP2).

Results

A statistical analysis of the study results is presented in the following subsections, organized by results for accuracy, ULCs, and LSD.

Accuracy

Participants completed their 12 drives each, and their accuracy at each decision point (2 for each layout) was recorded. There were two cases, shown in table 58, where an incorrect maneuver at the DP1 prevented a correct maneuver at the DP2; the accuracy of those DP2 maneuvers was not analyzed.

Otherwise, participants were highly accurate across the board. Accuracy was analyzed for each layout separately to determine which signing alternative yielded the best (most accurate) results. Generalized estimating equations—the preferred analysis technique for this setup—are impossible to estimate due to low or zero observations in some experimental conditions. Instead, binomial proportions and exact confidence intervals, adjusted for simultaneous hypothesis testing, were computed and used to detect differences in accuracy among the various experimental conditions.^(40,41)

Overall, there was no statistically significant difference detected in accuracy among the signing alternatives of a given interchange layout, as indicated by the overlapping confidence intervals in figure 28.

Source: FHWA.

A. Accuracy for layout A.

Source: FHWA

B. Accuracy for layout C.

Source: FHWA.

C. Accuracy for layout E.

Source: FHWA.

D. Accuracy for layout L.

Figure 28. Graphics. Participant accuracy for each combination of interchange layout and signing alternative.

Starting lane and destination were also analyzed. Again, no statistically significant differences were detected.

ULCs

The minimum number of lane changes was calculated for each interchange layout, signing alternative, starting lane, and destination combination. All lane changes in excess of this minimum were considered an ULC. Note that this calculation can produce negative values, representative of participants making fewer lane changes than necessary. Ten such cases were observed and are presented in table 59.

In addition, in two cases (see table 60), an incorrect maneuver at DP1 prevented a correct maneuver at DP2; the number of ULCs during those DP2 maneuvers was not analyzed.

ULCs were analyzed for each layout separately to determine which signing alternative yielded the best (fewest ULCs) results. Generalized estimating equations—the preferred analysis technique for this setup—are impossible to estimate due to low or zero observations in some experimental conditions. Instead, Poisson means and confidence intervals, adjusted for simultaneous hypothesis testing, were computed and used to detect differences in ULCs among the various experimental conditions.⁽⁴²⁾ Figure 29 plots the count of ULCs for each combination of interchange layout and signing alternative to show that ULCs follow a Poisson distribution.

–Not applicable.

Source: FHWA.

Figure 29. Graphic. Histograms of ULCs for each combination of interchange layout and signing alternative.

As shown in figure 30, there were two statistically significant differences in ULCs due to signing alternatives within a given interchange layout: in layout A, SA3 (mean = 0.49, confidence interval = [0.36, 0.63]) was associated with significantly more ULCs than SA1 (mean = 0.24, confidence interval = [0.15, 0.34]) and SA2 (mean = 0.23, confidence interval = [0.14, 0.32]).

Source: FHWA

A. ULCs for layout A.

Source: FHWA

B. ULCs for Layout C.

Source: FHWA

C. ULCs for Layout E.

Source: FHWA

D. ULCs for Layout L.

Figure 30. Graphics. Mean and 95-percent (familywise) confidence intervals for ULCs associated with each signing alternative within interchange layout.

LSD

Participants completed 12 drives each, and their lane changes within each decision point (2 for each layout) were recorded. Final lane changes were considered lane selections. LSD (in miles) begins at the legibility point of the first sign in a signing alternative and terminates where the participant makes the final lane change.

Data are formatted such that one row represents one observation, which captures the LSD and number of signs passed up to that point for a given decision point (along with other experimental conditions and demographics). There are up to 2 observations per drive per participant, or 24 observations total per participant; drives involving no lane changes are not represented here. The total number of data points should equal 121 × 12 × 2 = 2904, but one participant (75) failed to complete the sixth drive, and another (111) failed to complete the second set of six drives; therefore, the dataset contains 2904 – 2(1 + 6) = 2890 observations. Of those, 69.6 percent did not change lanes at all, and 1.8 percent did so before encountering any signs. The following analyses apply to the 828 cases in which valid lane changes were made.

Each layout and decision point was considered a distinct survival analysis. Whereas survival analysis is traditionally applied to medical data, the research team use it here to model LSD and use final lane changes as “deaths.” The homogeneity of survival curves for each signing alternative was tested using PROC LIFETEST in SAS 9.2. Median LSD and complete survival curves are presented. All reported p-values have been adjusted for multiple comparisons.

Layout A, DP1

Signing alternatives 1 and 3 were found to differ significantly from signing alternative 2 (Wilcoxon p_1,2 = p_2,3 < 0.01) but not from one another (p_1,3 = 0.42). The Wilcoxon test is used because the Likelihood Ratio test “assumes that the data in the various samples are exponentially distributed and tests that the scale parameters are equal.”⁽⁴¹⁾ Median LSD and simultaneous confidence intervals are shown in table 61.

Survival curves for each signing alternative are plotted in figure 31 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 31. Graphic. Survival analysis with 95-percent confidence intervals: layout A, DP1.

Layout A, DP2

Different signing alternatives did not produce significantly different LSDs in DP2 (all p > 0.05). Median LSD and simultaneous confidence intervals are shown in table 62.

Survival curves for each signing alternative are plotted in figure 32 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 32. Graphic. Survival analysis with 95-percent confidence intervals: layout A, DP2.

Layout C, DP1

Signing alternatives 1, 2, and 4 were found to differ significantly from signing alternative 3 (p_1,3 = p_2,3 = p_3,4 < 0.01) but not from one another (p_1,2 = p_1,4 = p_2,4 = 1.00). Median LSD and simultaneous confidence intervals are shown in table 63.

Survival curves for each signing alternative are plotted in figure 33 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 33. Graphic. Survival analysis with 95-percent confidence intervals: layout C, DP1.

Layout C, DP2

Different signing alternatives did not produce significantly different LSDs (p > 0.05) in DP2. Median LSD and simultaneous confidence intervals are shown in table 64.

Survival curves for each signing alternative are plotted in figure 34 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 34. Graphic. Survival analysis with 95-percent confidence intervals: layout C, DP2.

Layout E, DP1

Signing alternatives 1 and 2 were found to differ significantly from Signing alternative 3 (p_1,3 = p_2,3 < 0.01) but not from one another (p_1,2 = 0.21). Median LSD and simultaneous confidence intervals are shown in table 65.

Survival curves for each signing alternative are plotted in figure 35 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 35. Graphic. Survival analysis with 95-percent confidence intervals: layout E, DP1.

Layout E, DP2

Different signing alternatives did not produce significantly different LSDs (p > 0.05) in DP2. Median LSD and simultaneous confidence intervals are shown in table 66.

Survival curves for each signing alternative are plotted in figure 36 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 36. Graphic. Survival analysis with 95-percent confidence intervals: layout E, DP2.

Layout L, DP1

Different signing alternatives did not produce significantly different LSDs (p > 0.05). Median LSD and simultaneous confidence intervals are shown in table 67.

Survival curves for each signing alternative are plotted in figure 37 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 37. Graphic. Survival analysis with 95-percent confidence intervals: layout L, DP1.

Layout L, DP2

Different signing alternatives did not produce significantly different LSDs (p > 0.05). in DP2. Median LSD and simultaneous confidence intervals are shown in table 68.

Survival curves for each signing alternative are plotted in figure 38 (where “survival” corresponds to not selecting the final lane yet).

Source: FHWA.

Figure 38. Graphic. Survival analysis with 95-percent confidence intervals: layout L, DP2.

Discussion

The combination of the three analyses (accuracy, ULCs, and LSD) provides better insight into the different signing alternatives. Because the layouts were not compared, the findings that follow focus on comparisons between signing alternatives within a single layout. The analysis found that, when considering signing alternatives within a single layout, no signing alternatives had a statistically significant difference in accuracy; in all cases, participants were accurate in getting to their destination. Other findings include the following:

• Layout A. Signing alternative 3 was shown to produce significantly more ULCs than signing alternative 1 and signing alternative 2, and participants made their final lane change earlier in signing alternative 1 and signing alternative 3 than in signing alternative 2. The three signing alternatives for layout A are shown in figure 39.
• Layout C. When considering ULCs, no differences were found across the four signing alternatives in layout C. For LSD, the analysis found that participants entered their final lane significantly earlier in signing alternative 3 when compared with signing alternative 1, signing alternative 2, and signing alternative 4. The four signing alternatives for layout C are shown in figure 40.
• Layout E. When considering ULCs, no differences were found across the three signing alternatives in layout E. For LSD, the analysis found that participants entered their final lane significantly earlier in signing alternative 3 when compared with signing alternative 1 and signing alternative 2. The three signing alternatives for layout E are shown in figure 41.
• Layout L. No significant differences were found for ULCs or LSD in either of the signing alternatives for layout L. The two signing alternatives for layout L are shown in figure 42.

To summarize, the signing alternatives that produced the best (i.e., fewest ULCs, earliest) movement into the final lane are shown in table 69.

A typical driver in the United States has seen many guide signs in various environments and, generally, is able to follow guide signs to his or her final destination. In this study, participants navigated interchanges signed using a variety of approaches, and participants were found to be accurate regardless of the approach used. Similarly, participants seemed to understand the signing alternative as, in general, there was an average of less than one ULC per interchange. Together, the high accuracy presented by drivers and few ULCs indicate that drivers tend to understand a series of guide signs leading up to complex interchanges as long as they are designed consistently and with good signing practices.

Another finding from this study is that the best signing alternative for both layouts C and E was found to be designed where the signs present the driver one destination per lane, even in cases where some lanes may provide access to multiple locations (e.g., layout E). This characteristic is also present to an extent in the best signing alternative for layout A (signing alternative 1). In layout A, signing alternative 1, two destinations sharing a single lane are listed on a single sign, but the sign has a full-width horizontal separator and clearly lists the distance to each exit.

While accuracy, ULCs, and LSD are important measures, it is also important to consider other factors not discussed in this study when designing signs for complex interchanges. For instance, while in layout E, signing alternative 3 was found to perform best; this approach could cause issues with lane use. In this signing alternative, drivers making a left at the downstream split are guided into the option lane on the mainline and the left lane on the C/D roadway, but both the exit-only lane and the right lane on the C/D roadway would lead the driver to the same direction (left at the downstream split). In effect, drivers making a left at the downstream split would be bunched in the left lane on the C/D roadway, potentially leaving unused capacity in the right lane.

Source: FHWA.

Figure 39. Graphic. Layout A signing alternative examples.

Source: FHWA.

Figure 40. Graphic. Layout C signing alternative examples.

Source: FHWA.

Figure 41. Graphic. Layout E signing alternative examples.

Source: FHWA.

Figure 42. Graphic. Layout L signing alternative examples.