Analysis of the SHRP 2 Naturalistic Driving Study Data (S08B) Evaluation of Offset Left-Turn Lanes

SHRP 2 Safety SymposiumJuly 10, 2014

Jessica M. Hutton & Karin M. Bauer

Research Team

• MRIGlobal– Jessica Hutton, PI and Traffic Engineer– Karin Bauer, Co-PI and Statistician– Chris Fees, Traffic Engineer– Jessica Brucker, Data Analyst– John Ronchetto, Technician

• Human Factors North– Alison Smiley– Tom Smahel

Overview• Objectives• Background• Intersection Selection• Video Data Reduction• Statistical Analysis• Results• Future Research

Research Objectives• Answer the study questions:

– How does left-turn lane offset affect turn behavior and gap acceptance?

– What effect does the presence of a vehicle in the opposing left-turn lane have on gap-acceptance behavior? Does it vary by offset?

– What other factors affect gap-acceptance behavior? (Driver age and gender, weather and lighting conditions, presence of following vehicle, time spent waiting for a gap, etc.)

• Develop design guidance for offset left-turn lanes• Document experience for future users of NDS data

4

Background

5

Background

6

Background

• AASHTO Green Book states:– Consider offset left-turn lanes where the median

is 18 ft or greater to avoid potential sight

7

is 18 ft or greater to avoid potential sight restriction from opposing left-turn vehicle

– Assume a critical gap length of 5.5 sec for a two-lane road and add 0.5 sec for each additional lane

Research Activities

IRB approval and Data Sharing Agreement

Identify intersections for 

research

Define and refine data request and receive data

Reduce video data Analyze dataDraw conclusions 

and develop recommendations

8

Intersection Selection• To answer our research questions, we needed

intersections with the following characteristics:– Dedicated left-turn lanes opposing each other– Unsignalized or permissive left-turn signal phasing– No sight limitations due to geometry– Drivers making left turns at those intersections

• Among the intersections chosen, we needed a variety of offset conditions to observe how changes in offset affect changes in gap acceptance behavior

9

Intersection Selection• 6-step intersection selection process

10

Development of Video Data Reduction Tool

• Our primary data source was the forward- and rear-facing videos.

• In Phase 1, video data reduction was manual with all observations recorded in Excel spreadsheet.

• In Phase 2, we developed our own viewer to watch the forward and rear videos in synch, define allowable values for data reduction, and collect reduced video data in database format.

11

User Interface for Data Reduction

12

Variables Recorded from Video DataVariable Definition/Options

Video‐level variablesVideo ID Automatically recorded when video is selectedIntersection ID Automatically recorded from video folder namelight condition Drop down list: light, dawn/dusk, dark with street light, dark without street 

lightsWeather condition Drop down list: dry, rain, snowArrives in Q Time stamp: recorded when vehicle reaches rear of queue or stop bar if no

13

Arrives in Q Time stamp: recorded when vehicle reaches rear of queue or stop bar if no queue exists

Number of vehicles in Q Drop down list: Integers 0‐7 and “8 or more”Time when rear not visible Time stamp: recorded when reviewer can no longer discern through 

vehicles passing through the intersection in the rear‐facing video

No visibility in rear view Check box: checked if this condition appliesTime light turns green Time stamp: recorded if signal is red at NDS time of arrival in queue at time 

when signal turns green

Final gap estimated long Check box: checked if this condition applies (no oncoming vehicles visible in field of view of driver)

Notes Text box

Variables Recorded from Video Data (cont.)Variable Definition/Options

Event‐level variablesEvent No. Automatic consecutive numbering for each observed event (gap)

Start gap time Time stamp: recorded at the time of each new start of gap or lag

End gap time Time stamp: recorded at the time of each gap or lag end

Turning vehicle No. Drop down list: Integers 1‐5 and “NDS” (non‐NDS vehicles numbered consecutively)

Vehicle position Drop down list: behind/at stop bar, ahead of stop bar, positioned to turn, driver never pauses (recorded at gap start time)

LT signal at start of gap Drop down list: green arrow, green ball, red, yellow ball, yellow flashing arrow

14

g g p p g , g , , y , y g

Time begin turn Time stamp: recorded at the time vehicle is both accelerating and orienting vehicle toward leftLT signal at turn initiation Drop down list: green arrow, green ball, red, yellow ball, yellow flashing arrow

Avoidance maneuver Drop down list: none, turning driver, opposing driver, both

Type oncoming vehicle Drop down list: passenger car, heavy vehicle

Vehicle behind? Yes/No (yes when vehicle present behind turning driver)

Driver move? Yes/No (yes when vehicle moves forward during gap)

Opposing queued? Yes/No (yes when opposing through vehicles were queued before proceeding through the intersection)

RT present? Yes/No (yes when opposing vehicle turns right during gap)

Opposing left present? Yes/No (yes when a vehicle is present in the opposing left turn lane)

Sight distance blocked? Yes/No (yes when driver’s view of oncoming traffic is restricted by opposing left‐turning vehicles)Gap accepted? Yes/No

Definitions of measured variablesTime T0: First opposing through vehicle reaches the stop bar after the study vehicle arrives

Time T2: Turn is initiated by study driver

Time T1: Next opposing through vehicle reaches the stop bar

Time T3: First opposing through vehicle reaches the stop bar after the study vehicle makes the left turn

15

Rejected gap length = T1 – T0

Accepted gap length = T3 – T1

T3 is estimated from forward camera or viewed in rear camera

T0, T1 and T2 are viewed in the forward-facing camera

PET = T3 – T2

Time spent waiting for a gap = T2 – Tarrived in queue

Time Series Data

• For each NDS left-turn maneuver for which we requested video data, we also requested time-series data, including:

• GPS • Time of trip (binned)

16

• Speed

• Acceleration

• Heading

• Alcohol detection

• Driver ID

• Driver age

• Driver gender

• Vehicle type

Analysis Approach• Basic questions to be answered with available data:

– What are the characteristics of the “critical gaps”? – Do they differ by:

• Left-turn offset?• Whether opposing through vehicle obstructs sight distance?• Weather and light condition?• Driver demographics (age, gender)?• Traffic factors (speed of oncoming vehicle, presence of vehicle

waiting behind, length of time waiting for gap)?

– Under what conditions to avoidance maneuvers take place?

– Were there any crashes or near misses?17

Surrogate Safety Measures

• Critical Gap—Gap length equally likely to be accepted or rejected by a driver

• Post-Encroachment Time—Time between when driver initiates the left turn and when the next opposing through vehicle arrives at the intersection. A measure of how much time separated the two vehicles from a collision.

18

Data Overview

From the video reviews, data were collected:• For 145 NDS and 204 non-NDS drivers• At 44 signalized intersection left-turn pairs (33 intersections) and

14 two-way stop-controlled intersections14 two-way stop-controlled intersections• In four states: Florida, Indiana, North Carolina, and Washington• For 770 left-turning maneuvers (i.e., trips or videos)• For 3,350 events, where an event is defined as either an accepted

or rejected gap, by either an NDS or non-NDS driver

19

Descriptive Statistics

Offset Category Total EventsNumber of Drivers Number of Trips

Number of Accepted Gaps

NDS Non‐NDS NDS All NDS AllSignalized Intersections(a) –16 ft or less 196 5 8 19 27 13 19(b) –11 to –15 ft 100 15 20 54 74 8 11(c) –6 to –10 ft 225 24 19 46 64 16 28(d) –1 to –5 ft 594 35 30 79 108 41 56

20

(d)  1 to  5 ft 594 35 30 79 108 41 56(e) 0 ft 618 39 62 149 211 57 95(f) 1 to 3 ft 234 21 50 76 126 20 35(g) 4 to 6 ft 98 21 20 29 49 14 25

All signalized 2,065 160 209 452 659 169 269Two‐way Stop‐Controlled Intersections(a) –16 ft or less 45 4 2 19 21 3 5(b) –11 to –15 ft 932 21 53 194 241 112 149(c) –6 to –10 ft 201 9 8 66 73 34 40(e) 0 ft 107 10 3 27 30 13 13All two‐way stop‐

controlled1,285 44 66 306 365 162 207

All intersections 3,350 204 275 758 1,024 331 476

Statistical Methodology– Logistic regression analysis was used to model the relationship

between (1) the probability of accepting or rejecting a gap of a given length and (2) the length of the gap and the left-turn offset distance

– From the regression model, the critical gap length, t50, and its 95% confidence limits were estimated by inverse regression

– t50 is the gap length (on X-axis) that corresponds to a probability of 0.5 (on Y-axis); that is, the gap length where the probability of accepting = probability of rejecting

– The confidence intervals of the critical gaps are then compared in a pairwise fashion to assess which offset category differs statistically from which other offset category with respect to critical gap

21

Research Results

22

Research Results (cont.)Comparison Between: Statistically 

SignificantlyDifferent?Offset Category 1 Offset Category 2

(a) ‐‐ 16 ft or less (b) – 15 ft to – 11ft No(c) ‐‐ 10 ft to – 6 ft No(d) – 5 ft to – 1 ft No(e) 0 ft No(f) 1 ft to 3 ft Yes(g) 4 ft to 6 ft Yes

(b) – 15 ft to – 11ft (c) ‐‐ 10 ft to – 6 ft No

23

(d) – 5 ft to – 1 ft No(e) 0 ft No(f) 1 ft to 3 ft No(g) 4 ft to 6 ft No

(c) ‐‐ 10 ft to – 6 ft (d) – 5 ft to – 1 ft No(e) 0 ft No(f) 1 ft to 3 ft Yes(g) 4 ft to 6 ft No

(d) – 5 ft to – 1 ft (e) 0 ft No(f) 1 ft to 3 ft Yes(g) 4 ft to 6 ft Yes

(e) 0 ft (f) 1 ft to 3 ft Yes(g) 4 ft to 6 ft No

(f) 1 ft to 3 ft (g) 4 ft to 6 ft No

Sight Obstruction

Offset Category

Available Gaps Accepted Gaps OnlyPercentage of Events when an 

Opposing Vehicle is Present

Percentage of events when 

Driver’s View is Blocked

Ratio of Driver’s Sight Blocked to 

Opposing Vehicle Present

Percentage of Events when an 

Opposing Vehicle is Present

Percentage of events when 

Driver’s View is Blocked

Ratio of Driver’s Sight Blocked to 

Opposing Vehicle Present

Signalized Intersections(a) –16 ft or less 34.7 30.1 86.8 7.4 7.4 100.0(b) –11 to –15 ft 25.0 12.0 48.0 23.0 8.1 35.3

24

(b)  11 to  15 ft 25.0 12.0 48.0 23.0 8.1 35.3(c) –6 to –10 ft 48.0 44.9 93.5 32.8 25.0 76.2(d) –1 to –5 ft 26.1 23.6 90.3 24.1 18.5 76.9(e) 0 ft 26.5 3.9 14.6 21.3 4.7 22.2(f) 1 to 3 ft 35.5 3.0 8.4 34.9 3.2 9.1(g) 4 to 6 ft 21.4 3.1 14.3 30.6 4.1 13.3Two‐Way Stop‐Controlled Intersections(a) –16 ft or less 4.4 0.0 0.0 9.5 0.0 0.0(b) –11 to –15 ft 7.8 6.4 82.2 8.7 7.5 85.7(c) –6 to –10 ft 23.9 18.9 79.2 9.6 8.2 85.7(e) 0 ft 9.3 0.0 0.0 3.3 0.0 0.0All Intersections CombinedNegative offset 20.9 17.9 85.6 15.8 11.2 70.8Zero offset 24.0 3.3 13.8 19.1 4.1 21.7Positive offset 31.3 3.0 9.6 33.7 3.4 10.2

Research Results (cont.)

Sig

naliz

ed

op-C

ontro

lled

25

Traffic Control Type

Is Sight Distance Obstructed?

Critical Gap Estimate (sec)

95% ConfidenceLimits (sec)

Significant Difference Between 

Obstruction and No Obstruction?Lower Upper)

Signalized Yes 7.5 6.6 8.5 NoNo 6.4 6.0 6.9Two‐Way Stop Yes 6.4 5.3 7.6 NoNo 5.1 4.8 5.4

Signalized intersections: significant sight obstruction effect (p-value = 0.02)Two-way stop-controlled intersections: significant sight obstruction effect (p−value = 0.03)

S

Sto

Safety Analyses

• Analysis of near-crashes– VTTI reported no near-crash or crash events related to

left-turn maneuvers at any of the intersections in our study.

– Only 6 events (of 3,350 observed by video reviewers) were found to include an avoidance maneuver by the turning driver, oncoming driver, or both. No pattern was observed among these events.

26

Safety Analyses (cont.)

• Analysis of short post-encroachment timesOffset Category

All Accepted Gaps

Number of Observations

Percentile Percent of Observations with Post‐encroachment Time Less Than:

1st 5th 10th 15th 1 sec 2 sec 3 sec 4 sec

27

Signalized Intersections

Negative 114 ‐1.33 0.34 2.28 2.71 6 9 18 36

Zero 95 0.02 2.15 2.97 3.58 1 3 11 21

Positive 60 ‐1.50 1.17 3.00 3.53 3 7 10 20

Two‐Way Stop‐Controlled Intersections 

Negative 196 2.14 2.42 2.85 3.38 0 1 11 19

Zero 13 1.76 1.76 3.97 3.97 0 8 8 15

Secondary Analyses• When we looked at gender and age by offset category, we saw no effect. After

combining all offset categories, we found that at signalized intersections, older drivers accept longer gaps. At stop-controlled intersections, males accepted longer gaps.

• For many of the offset categories, it did appear that the presence of a following vehicle resulted in drivers accepting shorter gaps, although a statistical evaluation could not confirm this.W did fi d h h i i i f ff d h d• We did not find that the time spent waiting for a gap affected the accepted gap length or post-encroachment time.

• We did not have enough variability in lighting and weather conditions to conduct an analysis.

• We did not have enough variability in NDS vehicle type to conduct an analysis.• We did not find a relationship between posted speed limit and post-encroachment

time.• We did not find an effect of left-turn signal phasing (permissive only vs. permissive/

protected) on post-encroachment time.

28

Benefits of NDS Data

• Field data collection has already been done (cost savings, more intersections available)

• Truly naturalistic behaviors (unlike simulator ystudies)

• Can view oncoming traffic from driver’s perspective (don’t have to make assumptions about sight restrictions)

29

Summary• Critical gap times ranged from 4.7 sec (at positive offset) to 7.5 sec (at the most

negatively offset)

• The critical gaps observed in the research are closely aligned with the critical gap lengths reported in other studies

• Sight distance restrictions due to the presence of opposing left-turn vehicles tend to increase critical gap lengths

• Opposing left-turning vehicles are much more likely to obstruct the view of a left-turning driver at a negative offset than at a zero or positive offset.

• Improving an offset from more negative to less negative will likely not adequately address the restricted view of the left-turning driver.

• Improving an offset that is minimally negative to zero or positive will likely improve visibility for left-turning drivers.

30

Summary (cont.)• Improving offset conditions will improve intersection operations, since gaps with

shorter lengths will be accepted by more drivers.

• While on average, drivers tend to wait for longer gaps when their view of oncoming traffic is restricted, the shortest post-encroachment times are more likely to be taken by drivers with on obstructed view.

• Safety issues resulting in crashes or near misses are rare for this specific scenario, but that does not mean intersections with negative offsets are not a safety concern. Proactive safety strategies seek to identify locations with conditions that may lead to crashes even if none have occurred. Restricted sight distance for left-turning drivers creates a potential safety concern.

• The limited sample size of this study led to inconclusive results for many of the secondary analyses; however, we showed that several different analyses are possible with the NDS data.

31

Recommendations• Update Green Book text to indicate that any amount of negative offset can allow

a left-turning vehicle to restrict the view of oncoming traffic for the opposing left-turning vehicle and that zero or positive-offset left-turn lanes are desirable to minimize this possibility and allow turning drivers to adequately judge available gaps.

• Retrofitting negative offset left-turn lanes to positive or zero offset should be id d h f iblconsidered where feasible.

• New intersection designs should consider positively offsetting left-turn lanes, especially at signalized intersections with permissive left-turns.

• Where negative-offset left-turn lanes create potential sight restrictions, protected-only left-turn signal phasing should be considered.

32

Future Research• More data! With more videos, we would have been able to conduct more secondary

analyses and potentially show stronger results in the primary analysis.• Additional data could be collected about the geometrics of an intersection to determine if

vehicles are able to position themselves to avoid the sight obstruction, and this could be considered in a similar analysis.

• The NDS data provide an opportunity to follow a single driver through many left-turn maneuvers to identify factors that change behavior.

• The NDS data include information about vehicle dynamics which could be evaluated for their influence on turning behavior and safety.

• Crash history could be used in the study site selection process to compare intersections with low left-turn crash rates to those with high crash rates.

• Left-turning behavior could be evaluated in specific conditions, such as hours of darkness, or during rain events. The NDS data allow for such queries so researchers can ensure desired sample sizes of more rare events.

• The effect of left-turn lane offset on intersection operational performance (in terms of delay or level of service measures) could be evaluated to provide a benefit/cost ratio.

33