Institute of Transportation Studies
UC Berkeley Traffic Safety Center
( University of California, Berkeley)
Year 2007 Paper UCB - TSC - RR - 2007 - 2
Driver Behavior at Rail Crossings:
Cost- Effective Improvements to Increase
Driver Safety at Public At- Grade
Rail- Highway Crossings in California
Douglas L. Cooper  David R. Ragland†
 UC Berkeley Traffic Safety Center
† UC Berkeley Traffic Safety Center
This paper is posted at the eScholarship Repository, University of California.
http:// repositories. cdlib. org/ its/ tsc/ UCB- TSC- RR- 2007- 2
Copyright  c 2007 by the authors.
Driver Behavior at Rail Crossings:
Cost- Effective Improvements to Increase
Driver Safety at Public At- Grade
Rail- Highway Crossings in California
Abstract
This report examines conditions affecting vehicle- train collisions at rail cross-ings
in California, and recommends effective countermeasures and implementa-tion
strategies. In doing so, the report helps meet California’s goal of efficiently
utilizing state and federal funding available through SAFETEA- LU for increas-ing
the safety at public atgrade rail- highway crossings
FINAL REPORT ■ APRIL 2007
DRIVER BEHAVIOR
AT RAIL CROSSINGS
COST- EFFECTIVE IMPROVEMENTS
TO INCREASE DRIVER SAFETY
AT PUBLIC AT- GRADE RAIL- HIGHWAY
CROSSINGS IN CALIFORNIA
PREPARED FOR
COOPERATIVE AGREEMENT
T. O. 5208
PREPARED BY
DOUGLAS L. COOPER
and
DAVID R. RAGLAND
University of California Traffic Safety Center ■ Institute of Transportation Studies
University of California ■ Berkeley, California 94730- 7360
Tel: 510/ 642- 0655 ■ Fax: 510/ 643- 9922
ACKNOWLEDGEMENTS
The University of California Traffic Safety Center appreciates
and acknowledges the contributions of the following participants.
Theodore Cohn
University of California School of Optometry
Scott Johnston
PATH
Robert E. Brydia
Texas Transportation Institute
Jeff Hullquist
Napa Valley Railroad Police Chief
Joseph E. Barton
Kevin Schumacher
California Public Utilities Commission
LeeAnn Dickson
USDOT/ Federal Railroad Administration
Operation Lifesaver
Funding for this project was provided by Caltrans.
TABLE OF CONTENTS
1. EXECUTIVE SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2. INTRODUCTION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3. BACKGROUND. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4. FIVE AND TEN YEAR CALIFORNIA CRASH DATA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.1. Description of Data Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.2. Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3.1. California and the U. S. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3.2. Crash Characteristics: Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3.3. Crash Characteristics: Driver Behavior. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3.4. Crash Characteristics: Train Speed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.3.5. Crash Characteristics: Driver Age and Gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3.6. Crash Characteristics: Multiple Crash Sites. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3.7. Crash Characteristics: Crossing Angle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5. CROSSING IMPROVEMENTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.1. Potential Rail Crossing Upgrades. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.1.1. Long- Arm Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.1.2. Medians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.1.3. Four- Quadrant Gate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.1.4. Photo Enforcement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2. Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3. Benefit vs. Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6. DRIVER DECISIONS AT RAIL CROSSINGS: A CONCEPTUAL MODEL. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.1. Signal Detection Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2. Perception of Train Speed and Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.3. The Leibowitz Hypothesis: Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.4. Driver Decisions Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7. CROSSING OBSERVATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.1. College Station, Texas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.2. Berkeley, California . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.3. Napa, California . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8. CONCLUSIONS AND RECOMMENDATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
9. SUGGESTIONS FOR FURTHER RESEARCH. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
10. REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
11. APPENDICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
11.1. Appendix A: California PUC Sample Form A Crossing Inventory Entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
11.2. Appendix B: FRA Crossing Inventory Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
11.3. Appendix C: Sample Accident Report and Narrative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
11.4. Appendix D: Upgrade Effectiveness Calculation and Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
11.5. Appendix E: Crash Sites with Multiple Crashes 1995- 2004 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
11.6. Appendix F: FRA Website Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
11.7. Appendix G: Leibowitz Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
11.8. Appendix H: California Vehicle Code: Automated Enforcement: Photographic Records . . . . . . . . . . . . . 47
11.9. Appendix I: Crossing Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
LIST OF FIGURES
Figure 1: Ten Year U. S. and California Rail- Highway Crossing Incidents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Figure 2: California Motor Vehicle/ Train Crashes at Rail- Highway Crossings 1995- 2004 . . . . . . . . . . . . . . . . . . . . . . . . 6
Figure 3: Speed of Trains Involved in Crashes at Public Crossings in California ( 2000- 2004) . . . . . . . . . . . . . . . . . . . . 9
Figure 4: Crash Severity by Train Speed at Public Crossings in California ( 2002- 2004) . . . . . . . . . . . . . . . . . . . . . . . . 10
Figure 5: Long- Arm Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Figure 6: Street Mounted Channelization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Figure 7: Island Mounted Channelization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Figure 8: Four- Quad Gate System Picture and Diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Figure 9: Internal Response Probabilities for Noise with Signal and Noise Only . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Figure 10: Internal Response Probability Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Figure 11: Effects of Shifting Criterion Response Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Figure 12: Changes in Perceptual Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Figure 13: View of Approaching Train from Vehicle Stopped at Crossing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 14: Test of the Leibowitz Hypothesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Figure 15: Overlapping Signal and Signal- Plus- Noise Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Figure 16: College Station, Texas, Holleman Drive Camera View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Figure 17: Gilman Avenue Crossing, Berkeley, California . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Figure I- 1: College Station, Texas, Camera View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Figure I- 2: Holleman and Wellborn Intersection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Figure I- 3: Train Speed Graphic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure I- 4: PATH Quicktime Playback Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure I- 5: Gillman Avenue Crossing, Berkeley, California . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
FigureI- 6: Gilman Ave. First Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Figure I- 7: Gilman Ave. Second Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Figure I- 8: Observation Equipment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
LIST OF TABLES
Table 1: California Public At- Grade Crossing Warning Equipment ( 2005). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Table 2: Warning Equipment for California Public Crossings with Crashes 2000- 2004 . . . . . . . . . . . . . . . . . . . . . . . . . 8
Table 3: Action and Position of Motorist At Gated Crossing Crashes in California ( 2000- 2004) . . . . . . . . . . . . . . . . . . 9
Table 4: Age and Gender of Drivers Involved in Crashes at Public Crossings in California ( 2000- 2004) . . . . . . . . . . 10
Table 5: California Motor Vehicle/ Train Crash Counts per Public Crossing 1995- 2004. . . . . . . . . . . . . . . . . . . . . . . . . 11
Table 6: California Public Crossing Angle Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Table 7: California Public Crossings with Four or More Crashes 1995- 2004 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Table 8: Five Year California Public Highway- Rail At- Grade Crossing Statistics 2000- 2004 . . . . . . . . . . . . . . . . . . . . . 13
Table 9: Cost and Effectiveness of Highway- Rail Crossing Equipment Upgrades . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Table 10: Benefits and Costs to Upgrade California Multi- Crash Crossings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Table 11: Potential Outcome Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Table 12: Approach Speeds of the Large ( 10’) Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1. EXECUTIVE SUMMARY
In 1994, the U. S. Department of Transportation prepared a new national rail- highway crossing safety action plan. The
plan succeeded in decreasing vehicle- train collisions, and over the last ten years the number of national crossing
incidents fell 35 percent, while in California they decreased 23 percent. These decreases were due to a combination
of railroad crossing closures, upgrading of warning devices, and the efforts of grassroots organizations such as
Operation Lifesaver. However, despite decreasing numbers, crash counts remain undesirably high and ongoing
efforts to improve rail crossing safety are a priority.
This report examines conditions affecting vehicle- train collisions at rail crossings in California, and recommends
effective countermeasures and implementation strategies. In doing so, the report helps meet California’s goal of
efficiently utilizing state and federal funding available through SAFETEA- LU for increasing the safety at public at-grade
rail- highway crossings.
At the present time there are 7,719 public at- grade rail- highway crossings in California. During the 5- year period from
2000 to 2004, there were 593 train- vehicle crashes at these crossings. While the majority of crossings with collisions
had only one crash ( 72%) a significant number of crossings ( 28%) had multiple collisions, ranging from two to 12 in
number. The crashes resulted in a total of 99 deaths and 205 injuries.
The 593 crashes exhibited a number of characteristics, including:
■ 73% occurred at crossings equipped with gates.
■ 26.8% involved vehicles that had driven around or through lowered gates.
■ 59.2% involved vehicles that were still moving over the crossing.
■ 20.9% involved a vehicle running into the side of the train.
A large proportion of these collisions were caused by drivers deliberately circumventing warning equipment, with
devastating consequences. This behavior included ignoring flashing lights or other active warning devices, passing
through descending barrier gates, or even driving around stopped traffic and already- lowered gates. Although the
end- result of a collision is a relatively rare event, the behavior is widespread. Depending on the location, it appears
that between 20% and 60% of drivers who are in the position to ‘ run’ descending gates do so. The group of drivers
who are not deterred by lowered gates are primarily male and mostly under 40 years old, which is the same profile
seen for other risky driving behaviors. However, given the high proportion of drivers engaging in the behavior, it is
clearly not limited to any one demographic segment.
Among this group of drivers, active warning signals such as descending gates and flashing lights do not cue the driver
to stop. Rather, the active warning systems merely act as a signal that a decision must be made, and the driver uses
his/ her own judgment of train location and speed to decide whether or not to yield to the train. For those people,
the ‘ problem’ is determining the speed and proximity of the train, rather than establishing its presence. However, the
interplay of perception, expectation, and human information processing that is required can easily lead to failures in
judgment.
It has been shown that people’s ability to accurately judge the speed and distance of an oncoming train is quite
limited. In general, it is much more difficult to determine the speed of an object approaching the viewer than for an
object traveling across the field of vision. Additionally, the Leibowitz hypothesis suggests that drivers underestimate
the speed of trains because human vision underestimates the speed of large objects, such as locomotives.
1
Additionally, other disruptive factors— such as poor visibility, ‘ noisy’ signage, or in- car distractions— may impede the
driver’s ability to make a sound judgment. Signal detection theory tells us that the decision to proceed or stop at a
rail crossing is based on our ability to separate a meaningful signal from background noise. While measures exist that
could further increase the conspicuity of trains ( the ‘ signal’) or decrease the background noise, these measures might
actually encourage gate running by increasing driver confidence in his/ her ability to judge train speed and distance.
Given the physiological limitations that virtually preclude the driver from accurately judging the time remaining
before an approaching train reaches the crossing, there appears to be no purpose served by giving the driver this
additional information.
The best solution to rail crossing crashes is to remove the need for the driver to engage in a potentially faulty
decision- making process by making it impossible, or at least very difficult, for the driver to bypass lowered gates.
There are two low- technology, low- cost, low- maintenance methods that, while not 100% effective, have been
deployed in many locations and shown to prevent deaths and injuries while remaining economically feasible. These
are long- arm gates and median separators. Adding either long- arm gates or median separators has been estimated
to have reduced collisions by 75%, compared to standard flashing lights and gates. The cost of long- arm gates is
approximately $ 5,000 per crossing, but long- arm gates may not be appropriate in locations with significant truck or
bus traffic, wide crossings, multiple rails, or high winds. Medians have a cost of $ 14,000 per crossing, and may be
suitable for different locations than long- arm gates.
Where these technologies cannot be deployed, photo enforcement should also be considered as an option.
Although the consequences of getting a traffic ticket are far less severe than being hit by a train, studies have shown
that the threat of a traffic violation ticket is as effective in changing driver behavior as long- arm gates or medians.
However, the cost for installation of cameras can be quite high.
2. INTRODUCTION
In response to a congressional directive, the U. S. Department of Transportation prepared a new national rail- highway
at- grade crossing safety action plan that was issued on June 13, 1994. Over the last ten years, the results of this plan can
be seen as the number of
grade crossing incidents
has fallen 35 percent,
from 4,633 at the end of
1995 to 3,026 at the end
of 2004. In California,
during this same period,
the number of inci-dents
has decreased 23
percent, from 201 to 154
( Figure 1).
For the most part, the
progress achieved
under the 1994 Action
Plan is attributable to
the closures of 41,070
public and private grade
crossings, upgrades at
3,985 public crossings
100
125
150
175
200
225
250
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
California Incidents
2000
2500
3000
3500
4000
4500
5000
National Incidents
California Incidents National Incidents
Figure 1
TEN YEAR U. S. AND CALIFORNIA
RAIL- HIGHWAY CROSSING INCIDENTS
SOURCE: FRA
2
with a high probability for incidents with active warning devices, such as automatic gates, flashing lights, and highway
traffic signals. The progress was also bolstered by annual education campaigns by Operation Lifesaver, a non- profit,
international continuing public education program established to end collisions, deaths and injuries at places where
roadways cross train tracks ( Federal Railroad Administration ( FRA, 2004).
While there is little doubt that upgrading crossings from passive to active significantly decreases the number of rail
crossing incidents, a 2004 Federal Railroad Administration ( FRA) report found that incidents continued to occur at
public grade crossings equipped with active warning devices. In California, for the five- year period 2000 to 2004, 508
or 85.7 percent of the public at- grade crossing incidents occurred at crossings already equipped with automatic or
active warning devices. Of these incidents, 434 occurred at public crossings with automatic gates, 69 had flashing
lights, and 5 were equipped with wig- wags.
There are over 250,000 public and private at- grade highway- rail crossings in the United States which provided the
backdrop for 3,026 reportable incidents in 2004 resulting in 368 deaths and 1,077 injuries in 2004. California’s 12,784
at- grade crossings had 154 incidents in that same year with 34 deaths and 53 injuries.
The focus of this report will be California’s 7,719 public at- grade crossings. During the five year from 2000 to 2004,
there were a total of 593 crashes between trains and motorized vehicles at these crossings that resulted in 99 deaths
and 205 injuries.
There are three primary sections of the California Vehicle Code that deal with motor vehicles at railway crossings:
PRIMA FACIE SPEED LIMITS
22352. ( a) The prima facie limits are as follows and shall be applicable unless changed as authorized in this
code and, if so changed, only when signs have been erected giving notice thereof:
( 1) Fifteen miles per hour:
( A) When traversing a railway grade crossing, if during the last 100 feet of the approach to the crossing the driver
does not have a clear and unobstructed view of the crossing and of any traffic on the railway for a distance of
400 feet in both directions along the railway. This subdivision does not apply in the case of any railway grade
crossing where a human flagman is on duty or a clearly visible electrical or mechanical railway crossing signal
device is installed but does not then indicate the immediate approach of a railway train or car.
RAILROAD OR RAIL TRANSIT GRADE CROSSINGS
22451.( a) The driver of any vehicle or pedestrian approaching a railroad or rail transit grade crossing shall
stop not less than 15 feet from the nearest rail and shall not proceed until he or she can do so safely,
whenever the following conditions exist:
( 1) A clearly visible electric or mechanical signal device or a flagman gives warning of the approach or
passage of a train or car.
( 2) An approaching train or [ rail] car is plainly visible or is emitting an audible signal and, by reason of its
speed or nearness, is an immediate hazard.
( b) No driver or pedestrian shall proceed through, around, or under any railroad or rail transit crossing gate
while the gate is closed.
PARKING UPON OR NEAR RAILROAD TRACK
22521. No person shall park a vehicle upon any railroad track or within 7 1/ 2 feet of the nearest rail.
3
3. BACKGROUND
Rail crossings provide different levels of warnings and/ or barriers to alert drivers to the potential dangers presented
by the at- grade crossing. These protective devices range from four- quadrant gates with medians to mere stop signs
or crossbucks. Since some type of warning device is always present, crashes are caused either by people violating the
signs/ signals/ gates or people not perceiving or mis- perceiving an approaching train’s distance and speed.
In a 1999 study, Carlson and Fitzpatrick found that 60 percent of drivers at 19 sites in Texas equipped with lights and
gates, crossed the track between the time the lights activated and two seconds after gate arms began to descend.
In addition, violations occurring after the arms had been in motion more than 2 seconds and until the arms were
horizontal, occurred during one- third of the gate- activations. Similarly, a 2004 FRA report found that accidents
continued to occur at public grade crossings equipped with active warning devices. For the period 1994 to 2003,
51 percent of the public grade crossing accidents occurred at crossings already equipped with automatic or active
warning devices1 ( FRA, 2004).
There is research to suggest that certain types of drivers may be more likely to ignore and violate such protective
systems. Survey results of 891 randomly selected residents in Michigan found that the stronger a person’s sensation
seeking tendencies, the more likely they are to inflate their ability to judge train distance, train speed, and the ease
with which they can get their car over the tracks before a train arrives. Additionally, the stronger the sensation seeking
tendencies, the more likely people are to experience frustration while having to wait for a train, which appears to
independently influence the judgment processes. Thus, the greater one’s frustration, the more likely he or she is to
make biased judgments which, in turn, can increase risky driving behavior ( Witte and Donohue, 2000).
A study based on the reports from 85 consecutive fatal crashes involving motor vehicles and trains at all types of
railway crossings in Victoria, Australia, on the other hand, concluded that, ‘‘... in most cases, the accident occurred
to a law- abiding citizen going about his or her daily work and was attributable to human overload unrelated to any
breach of regulation.’’ Additionally, at least 86% of those killed were persons who lived locally and were therefore
familiar with the existence of this crossing ( Wigglesworth, 1979).
An important finding in a study by Meeker and Barr ( 1989) was that two thirds of the 57 drivers who approached a
rural rail grade crossing in the presence of activated warning flashers crossed the tracks despite the warnings and the
approaching train. This would appear to indicate that crossing an activated warning device is a widespread activity
not limited to a small proportion of drivers. Clearly, the activated devices in their observations were not commonly
perceived as a signal that the risk was too great and that the driver should not cross. Rather, the results are consistent
with the view of Leibowitz ( 1985), who suggested that “ active” warning systems merely cue drivers as to the need to
make a decision whether or not to cross.
Meeker and Barr ( 1989) go on to say that “... it is not entirely satisfactory to conclude that two thirds of all drivers in our
sample were engaging in life- threatening behavior when they decided to cross. One might argue that pedestrians
regularly cross busy thoroughfares with a much smaller safety margin than the margin that drivers we observed
allowed themselves.”
Drivers crossing around barrier gates tended to stop or slow on approach significantly less than those crossing with
flashers only. It was suggested that the gates themselves provided an impediment to crossing which forced drivers
inclined to cross into making a hurried and sometimes perilous decision. Their behavior was seen as explaining the
surprisingly high number of accidents that occur at barrier- gate crossings. Perhaps the only way that drivers at these
1 Although no information is readily available on the role of warning equipment malfunctions in these incidents, a New York Times article from
December 30, 2004, stated that a “ computer analysis of government records found that from 1999 through 2003, there were at least 400 grade-crossing
accidents in which signals either did not activate or were alleged to have malfunctioned... Proving that a signal malfunctioned can be
difficult. In the more than 400 accidents in the Times analysis, 30 percent of the signal problems were listed as confirmed.” This works out to 2.5%
alleged and 0.7% confirmed.
4
barrier- gate crossings can achieve an acceptable safety margin is to make the decision to proceed through the
crossing without stopping or slowing their vehicles early on. The fact that a substantial number of accidents tend to
occur at these crossings is not surprising given this behavior. ( Meeker et al., 1997)
A common driver error is misjudgment of the time remaining until the train arrives at the crossing ( i. e., train speed
and distance). Speed estimation can be influenced by a number of factors, including driving experience, visual
cues available, light conditions, the presence of visual information in the background, and adaptation to previously
encountered train speed levels ( Dewar and Olson 2002). Additionally there are two perceptual problems associated
with rail crossing decisions. First, humans have difficulty judging the approach speed of a vehicle when it is seen
nearly head on, as their only indication of speed is the rate of change in the size of the object. Second, Leibowitz
( 1985) noted that there is the illusion that large object appear to move more slowly than small ones which are actually
traveling at the same speed.
To assist the state of California in efficient utilization of state and federal funding available through SAFETEA- LU for
increasing the safety at public at- grade rail- highway crossings, the results of this project aim to recommend effective
countermeasures and an implementation strategy such that drivers are provided a sufficient level of warning and
are motivated to comply with cues. This report first presents five and ten year crash data for California to assess the
magnitude of the problem. Next, driver and crossing factors that may be associated with vehicle- train collisions are
examined. This is followed by a conceptual model of why drivers may make poor judgments at crossings. Last, we
present a cost- benefit analysis of the most appropriate countermeasures for use in high- collision areas.
4. FIVE AND TEN YEAR
CALIFORNIA CRASH DATA
4.1. DESCRIPTION OF DATA SOURCES
The statistics used in this section were obtained from the FRA Office of Safety Analysis Web Site ( http:// safetydata.
fra. dot. gov/ officeofsafety/ Default. asp – see Appendix C) with supplementary data from the California Public Utilities
Commission ( CPUC) Crossing Inventory and California municipal and county personnel and websites.
The FRA web site allows access to railroad safety information including accidents and incidents, inspections and
highway- rail crossing data. Users can run dynamic queries, download a variety of safety database files, publications
and forms, and view current statistical information on railroad safety. The data are organized into the following nine
categories ( the complete list of headings and sub- headings can be seen in Appendix F):
1 Overview
2 Query Accident/ Incident Trends
3 Train Accidents
4 Casualties
5 Highway- Rail Crossing Accidents
6 FRA Inspections
7 Downloads
8 Highway- Rail Crossing Inventory
9 FRA Safety Reporting
5
While these sources provide the best available and most complete information on railroad- related issues, there are
a number of significant problems that undermine the reliability of the data. As noted in a number of reports ( e. g.,
FRA, 2004, U. S. Government Accountability Office, 1996), both the inventory and accident/ incident databases contain
inaccurate as well as incomplete information. As an example, highway traffic information for the 7,719 open, at- grade
public crossings in California is often out of date with 16% of the vehicular traffic counts dating from the 1970s, 67%
from the 1980s, and 17% from the 1990s. Among the 593 public at- grade crashes that occurred between 2000 and
2004 examined for this report, 100 had either a crossing number with a location that did not match the information in
the rest of the incident report or else the latitude and longitude listed for the crossing in the FRA inventory yielded a
location that did not match the rest of the information in the inventory or incident report. As noted by the FRA ( 2004),
its Inventory Data File, a record of grade crossing location, physical, and operational characteristics, is dependent on
voluntary state reporting.
Unlike aircraft accidents, which are investigated by the National Transportation Safety Board ( NTSB) or the Federal
Aviation Administration ( FAA) unless only minor injury or property damage is involved, the FRA depends on the
railroad involved in the incident to submit the report ( the exceptions being if there are multiple deaths or a great
deal of publicity). As will be seen later in this section, this leads to a general dearth of detailed information. Quoting
from the FRA’s Railroad Safety Statistics 2004 Annual Report:
The completeness and accuracy of the information presented in this bulletin are primarily dependent upon
the data collection and reporting processes of the nation’s railroads. The FRA conducts routine audits
of these procedures, but does not have sufficient resources to perform comprehensive reviews of each
railroad’s reporting procedures. We extensively review and edit the reports we receive and make inquiry
when information is incomplete or inconsistent.
It is not possible to identify reportable events that were omitted from a railroad’s submission. Likewise,
there may be instances where incorrectly reported information passes all reviews and is accepted. Although
we attempt to be as vigilant as possible in both the editing and presentation of the accident/ incident data
reported, errors do occasionally occur.
The California Public Utility Commission maintains its own incident and inventory database. Lack of funding
has prevented the
CPUC from keeping
its inventory up to
date, although some
crossing information is
more recent than that
of the FRA database.
The CPUC database
was especially useful
for analyzing the angle
at which the highway
crossed the railroad
tracks for the crashes
under review. The last
time the CPUC issued
its “ Annual Report of
Railroad Accidents
Occurring in California”
was 1999.
80
100
120
140
160
180
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Year
Public Crossing Crashes
10
15
20
25
30
35
Private Crossing Crashes
Public Private
Figure 2
CALIFORNIA MOTOR VEHICLE/ TRAIN CRASHES
AT RAIL- HIGHWAY CROSSINGS 1995- 2004
SOURCE: FRA
6
4.2. METHODS
Raw data for California was downloaded from the FRA site and categorized by vehicular and crossing factors. When
possible, data was compared to information from other sources such as the CPUC. Because of the previously noted
problems with the FRA data inventory, there was no way to insure that the crossing number listed in the accident
report was actually where the crash occurred. Therefore, warning equipment at the crash site information was taken
from the accident report rather than from the crossing inventory database.
4.3. RESULTS
4.3.1. CALIFORNIA AND THE U. S.
FRA data show that rail accidents increased 14% from 2002 to 2004 ( Figure 2) and while many states have seen
a decrease in rail related accidents, California is one of six states ( along with Texas, Illinois, Indiana, Ohio, and
Louisiana) that continue to rank as the worst in rail safety based on the raw number of accidents and fatalities at
public grade crossings. Together, these six states account for 37% of the nation’s reported public grade crossing
accidents. By taking exposure ( based on the number of public at- grade rail crossings in each state) into account,
however, California’s ranking improve from fourth worst to 22nd for total collisions and from second to seventh in
fatalities.
4.3.2. CRASH CHARACTERISTICS: EQUIPMENT
At the present
time there are
7,719 public at-grade
crossings
in California of
which 43% are
passive and 57%
are active ( Table 1).
Most of the active
crossings ( 71%)
are equipped with
gates and flashing
lights. Equipment
at public crossings
where train- vehicle
crashes occurred
during 2000 through
2004 is shown in
Table 2. Perhaps
the most significant statistic from this table is that 434 crashes ( 73%) occurred at crossings equipped with gates,
which would seem to indicate that, for some drivers, standard two- quadrant gates are not a deterrent.
4.3.3. CRASH CHARACTERISTICS: DRIVER BEHAVIOR
In California during the five years 2000 - 2004, there were 789 rail- highway crossing crashes, of which 675 were at
public crossings. Eighty- two of the crashes involved pedestrians, leaving 593 train- vehicle crashes at public highway-rail
crossings. Table 8 shows these crashes broken out by year as well as type, and includes the number of people
killed or injured. Three noteworthy statistics from this table are:
Traffic Control Device Type Number Percentage
No Signs or Signals 172 2.2%
Other Signs or Signals 17 0.2%
Crossbucks 2805 36.3%
Stop Signs 307 4.0%
Special Signs or Warning 42 0.5%
Hwy Traffic Sig, Wigwags, or other Activated 270 3.5%
Flashing Lights 982 12.7%
All Other Gates 3124 40.5%
4 Quad 0 0.0%
Total Public At Grade 7719 100%
Table 1
CALIFORNIA PUBLIC AT- GRADE CROSSING
WARNING EQUIPMENT ( 2005) 1
1 The devices listed are the highest level of warning at a particular crossing.
SOURCE: FRA
7
■ 20.9% involved a vehicle running into a train.
■ 59.2% involved vehicles that were moving over the crossing.
■ 26.8% involved vehicles that had driven around or through lowered gates.
Of special interest are the 434 crashes that occurred at crossings equipped with gates. The motorist’s actions prior to
the crash and vehicle positions for each action at the time of the crash are shown in Table 3.
The crash records in the FRA database are often lacking in detail ( See example record in Appendix C). While there
is a narrative section that should describe the circumstances of the crash, this section appears to be constructed
from checked boxes or short statements recorded elsewhere in the record. This makes interpreting the data difficult.
For example, in Table 3 there are 40 crashes involving a vehicle that failed to stop and was hit as it moved over the
crossing. Given that these are all gated crossings and that the gates must be down at least five seconds before the
train arrives, how could these vehicles not have gone around or through the gates before being struck? The narratives
shed no light on this question.
4.3.4. CRASH CHARACTERISTICS: TRAIN SPEED
Figure 3 shows the cumulative distribution of train speeds for the 593 train- vehicle crashes at public rail- highway
crossings. The bars shows the actual number of crashes for each 10 MPH category, while the line shows the cumulative
percentage of crashes at that speed or slower. As an example, 63 crashes occurred with trains traveling between ten
and 19 MPH and nearly 33% of the total ( 192 out of 593) crashes involved trains moving at less than 20 MPH.
In Figure 4, the relationship between train speed and crash severity is shown. Within each speed grouping, the
percentages for all three crash types sum to 100%. Thus, for example, for those crashes that occur with a train speed
between 40 and 49 MPH ( 13.3% of all crashes), 65.7% are Property Damage Only ( PDO), 22.9% involve injuries,
Control Device
# Train/ Ve hicle
Crashes
Percentage of All
Train/ Ve hicle Crashes
# Train/ Pedestrian
Crashes
Percentage of All
Train/ Pedestrian Crashes
Gates 434 73.2% 78 95.1%
Cantilever Flashing Lights 23 3.9% 0 0.0%
Std Flashing Lights 46 7.8% 42 4.9%
Wig Wags 5 0.8% 0 0.0%
Hwy Traffic Sig 2 0.3% 0 0.0%
Audible 2 0.3% 0 0.0%
Cross Bucks 57 9.6% 0 0.0%
Stop Signs 20 3.4% 0 0.0%
Watchman 0 0% 0 0.0%
Flagged by Crew 0 0% 0 0.0%
Other 1 0.2% 0 0.0%
None 3 0.5% 0 0.0%
Total 593 100% 82 100%
Table 2
WARNING EQUIPMENT FOR CALIFORNIA PUBLIC
CROSSINGS WITH CRASHES 2000- 20041
1 The devices listed are the highest level of warning at a particular crossing. Thus a crossing with gates and flashing lights would be
listed only under the “ Gates” category.
2 The type of flashing lights was not given so all four crashes were arbitrarily placed in this category.
SOURCE: FRA
8
and 11.4% involve fatalities. The injury and fatality categories are mutually exclusive in that a crash that has both
injuries and at least one fatality is counted as a fatal crash. As can be seen, train speed plays a role in the number of
fatalities.
4.3.5. CRASH
CHARACTERISTICS:
DRIVER AGE AND
GENDER
Male drivers are over-represented
in all but
one of the 13 age
categories shown in
Table 4, with an overall
average of nearly 75%.
4.3.6.
CRASH
CHARACTERISTICS:
MULTIPLE CRASH
SITES
Table 5 shows that most crashes ( 72%) occurred at sites with only one crash during the ten year period 1995- 2004.
The other 28% occurred at sites with 2 to 12 crashes. Table 7 is a listing of crossings with four or more crashes during
this period, and includes information on the crash dates, crossing equipment, Average Annual Daily Traffic ( AADT),
collection year for AADT, average daily train counts, the angle at which the road and track intersect, the sightlines
at each of the four corners of the intersection, and the crossing location. Of the 36 crossings listed, 25 had gates
installed at the time the crashes occurred.
Driver Action/ Driver Position Action Action Percentage Position Position Percentage
Drove Around Or Through Gates/ 159 36.7%
Moving Over Crossing 159 36.7%
Ve hicle Stopped And Then Proceeded/ 15 3.5%
Moving Over Crossing 15 3.5%
Failed To Stop/ 40 9.2%
Moving Over Crossing 40 9.2%
Stopped On Crossing/ 130 30.0%
Stalled 29 6.7%
Stopped 87 20.0%
Trapped 14 3.2%
Other/ 90 20.7%
Stalled 19 4.4%
Stopped 57 13.1%
Moving Over Crossing 9 2.1%
Trapped 5 1.2%
Total 434 100.0% 434 100.0%
Table 3
ACTION AND POSITION OF MOTORIST AT GATED
CROSSING CRASHES IN CALIFORNIA ( 2000- 2004)
SOURCE: FRA
0
20
40
60
80
100
120
140
0- 9 10- 19 20- 29 30- 39 40- 49 50- 59 60- 69 70- 79 80+
Speed ( MPH)
Number of Crashes
0
20
40
60
80
100
Cumulative Percentage of Crashes
Crashes Cumulative Percent
Figure 3
SPEED OF TRAINS INVOLVED IN CRASHES
AT PUBLIC CROSSINGS IN CALIFORNIA ( 2000- 2004)
SOURCE: FRA
9
4.3.7. CRASH
CHARACTERISTICS:
CROSSING ANGLE
It is plausible that
crossing angle could
play a significant role
in crossing crashes,
perhaps because this
could require the driver
to look back over his/ her
shoulder. To examine this
hypothesis, crash records
were examined for
information on crossing
angle. For the 5- year
period 2000- 2004, 508
of the 593 train- vehicle
crashes had records that
included crossing angle
information. Table 6
describes the number of crashes in each ten degree crossing angle group. Column 1 describes the angle at which the
road crosses the tracks, grouped into ten degree categories. Columns 2 and 3 list the total number and percentage
of public railroad crossings in California in each crossing angle category, regardless of whether crashes occurred
at the site or not. The data for Column 2 was taken from the CPUC Crossing Inventory database. Columns 4 and 5
present the total number and percentage of vehicle- rail crashes for each angle category. Columns 6 and 7 present
the number and percentage of unique railroad crossings at which at least one crash occurred. In these two columns,
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
0- 9
( 24.1%)
10- 19
( 13.6%)
20- 29
( 9.0%)
30- 39
( 10.5%)
40- 49
( 13.3%)
50- 59
( 9.4%)
60- 69
( 8.8%)
70- 79
( 10.6%)
80+
( 0.5%)
Speed ( MPH)
Percent
% Fatal Crashes % Injury Crashes % PDO Crashes
Figure 4
CRASH SEVERITY BY TRAIN SPEED AT PUBLIC
CROSSINGS IN CALIFORNIA ( 2000- 2004)
SOURCE: FRA
Age Group Number % of Total Male
% of Age
Group Female
% of Age
Group
20 and younger 27 6.9% 20 74.1% 7 25.9%
21- 25 36 9.2% 27 75.0% 9 25.0%
26- 30 69 17.6% 62 89.9% 7 10.1%
31- 35 55 14.0% 38 69.1% 17 30.9%
36- 40 45 11.5% 33 73.3% 12 26.7%
41- 45 35 8.9% 25 71.4% 10 28.6%
46- 50 30 7.6% 24 80.0% 6 20.0%
51- 55 27 6.9% 19 70.4% 7 25.9%
56- 60 15 3.8% 9 60.0% 6 40.0%
61- 65 19 4.8% 16 84.2% 3 15.8%
66- 70 9 2.3% 7 77.8% 2 22.2%
71- 75 8 2.0% 3 37.5% 5 62.5%
76 and Older 18 4.6% 11 61.1% 7 38.9%
Total 3931 100% 2942 74.8% 982 25.2%
Table 4
AGE AND GENDER OF DRIVERS INVOLVED IN CRASHES
AT PUBLIC CROSSINGS IN CALIFORNIA ( 2000- 2004)
1 200 crossing crash records did not have drivers age
2 One of the 393 crash records with driver age did not have driver gender
SOURCE: FRA
10
only unique crossings are counted, regardless of the
number of crashes that occurred at the site. Column
8 describes the percentage of all public California
crossings in each angle category that had any crashes
occur ( Column 6 divided by Column 2).
A quick scan of the percentages in Columns 3, 5 and
7 shows that the distribution of total crashes and
of unique crash sites both conform fairly closely to
the distribution of all California crossings. Column 8
confirms that there does not appear to be any trend
in crossing angle and crash rate. Overall, 6.6% of
California crossings experienced a crash, and no single
angle category deviates largely from this percentage.
It would appear, then, that crossing angle is unlikely
to play a large role in vehicle- train crashes. This
was confirmed by the use of chi- square tests on the
crash data, which indicated no significant differences.
However, these tests rely on an assumption of uniform
vehicle exposure to crossing angles, that is, each angle
category receives a proportionate amount of traffic.
Additionally, the combination of the approach direction of both the train and the driver in relation to the intersection
play a role in the viewing angle of the driver. In a non- perpendicular crossing, the tracks on one side of the driver will
be difficult to see, and will require the driver to look back over his/ her shoulder. However, the tracks on the other side
of the driver will be very easily viewed. It may be that the increased visibility in one direction offsets poor visibility in
the other direction. On the other hand, better visibility could lead to increased risk- taking if the driver feels overly
confident about gauging the train’s position and speed. This subject should be investigated further using viewing
angle rather than intersection crossing angle.
Number of Crashes
At Crossing Number of Crossings
1 657
2 167
3 51
4 25
5 6
6 1
7 1
8 0
9 0
10 1
11 1
12 1
10 Year T otal 911
Table 5
CALIFORNIA MOTOR VEHICLE/ TRAIN
CRASH COUNTS PER PUBLIC CROSSING
1995- 2004
SOURCE: FRA
1 2 3 4 5 6 7 8
Crossing
Angle
# of CA
crossings at
this angle *
% of
crossings in
CA at this
angle*
Total # of
crashes at
this angle
% of total
crashes at
this angle
# of unique
crossings
with one or
more crashes
% of unique
crossings
with one or
more crashes
% of public CA
crossings at this
angle that had
one or more
crashes
81- 90o 3284 54.7% 261 51.4% 214 54.2% 6.5%
71- 80o 803 13.4% 58 11.4% 46 11.6% 5.7%
61- 70o 331 5.5% 34 6.7% 28 7.1% 8.5%
51- 60o 503 8.4% 54 10.6% 40 10.1% 8.0%
41- 50o 667 11.1% 64 12.6% 41 10.4% 6.1%
31- 40o 86 1.4% 5 1.0% 4 1.0% 4.7%
<= 30o 325 5.4% 32 6.3% 22 5.6% 6.8%
Totals 5999 100.0% 508 100% 395 100.0% 6.6%
Table 6
CALIFORNIA PUBLIC CROSSING ANGLE DATA
SOURCE: California Public Utility Commission database
11
Table 7
CALIFORNIA PUBLIC CROSSINGS WITH FOUR OR MORE CRASHES 1995- 2004
1 Entries in this column are in the form: Number of Crashes- Equipment. FL- Flashing Lights, G- Gates, Stop- Stop sign
2 G- Good, F- Fair, R- Restricted, O- Obstructed, B- Bad, Op- Open, C- Clear, P- Poor
SOURCE:
12
2000 2001 2002 2003 2004 Total Total
2000 2001 2002 2003 2004 Total % of Total Killed Injured Killed Injured Killed Injured Killed Injured Killed Injured Killed Injured
Drove Behind Or
In Front Of
Passing
Train, And Struck
By Second Train
2 6 9 6 8 31 0 0 6 3 4 6 1 0 2 2 13 11
Passed Standing
Vehicle
9 20 5 8 13 55 3 8 9 7 3 1 1 1 3 3 19 20
Train Hit Car 106 107 86 85 85 469 79.1% 11 32 28 25 14 26 18 45 14 30 85 158
Car Hit Train 20 34 19 24 27 124 20.9% 1 5 3 15 6 10 1 7 3 10 14 47
Total 126 141 105 109 112 593 12 37 31 40 20 36 19 52 17 40 99 205
Vehicle Stalled
On Crossing
12 13 15 6 7 53 8.9% 1 4 0 2 0 3 0 0 0 2 1 11
Stopped On
Crossing
37 27 31 38 36 169 28.5% 0 8 0 2 5 9 6 6 2 8 13 33
Moving Over
Crossing
70 95 57 62 67 351 59.2% 10 24 31 35 15 24 12 46 15 29 83 158
Vehicle Trapped
On Crossing
7 6 2 3 2 20 3.4% 1 1 0 1 0 0 1 0 0 1 2 3
Tot al 126 141 105 109 112 593 100 12 37 31 40 20 36 19 52 17 40 99 205
Drove Around Or
Through Gates
32 39 26 27 35 159 26.8% 7 16 16 14 10 13 8 34 12 16 53 93
Vehicle Stopped
And Then
Proceeded
7 9 5 11 4 36 6.1% 1 2 9 4 1 0 1 2 0 2 12 10
Failed To Stop 29 42 24 26 28 149 25.1% 2 6 6 15 4 10 3 12 3 11 18 54
Stopped On
Crossing
31 27 25 38 31 152 25.6% 1 6 0 5 5 8 6 4 2 8 14 31
Other 27 24 25 7 14 97 16.4% 1 7 0 2 0 5 1 0 0 3 2 17
Total 126 141 105 109 112 593 12 37 31 40 20 36 19 52 17 40 99 205
Crossings With
Gates
95 99 77 80 83 434 73.2% 12 31 19 26 17 29 16 46 14 31 78 163
Table 8
FIVE YEAR CALIFORNIA PUBLIC HIGHWAY- RAIL AT- GRADE CROSSING STATISTICS 2000- 2004
Note: Killed and Injured includes highway users, railroad employees, and railroad passengers
SOURCE: FRA
13
5. CROSSING IMPROVEMENTS
Based on a review of the literature as well as our own observations of driver behavior at rail crossings, there exists
a subset of drivers who will go around lowered gates if they think it is “ safe” to do so. As will be demonstrated in
Section 6 of this report, humans, in general, have an innate inability to judge the speed and distance of an oncoming
train. No amount of sight- line improvements, train conspicuity improvements, or warning system upgrades, will
improve this situation.
The only way to absolutely prevent drivers from going around or through crossing gates is to make it physically
impossible to do so. This can be accomplished by constructing a separation of grade, closing the crossing, or by
deploying an impenetrable barrier, all of which carry a high monetary or social ( e. g., such as loss of convenience,
slower response times for emergency vehicles, or loss of potential customers driving by a business) cost. There
are a number of other approaches that, while not being 100% effective, can be used to find a middle ground
that can prevent deaths and injuries while remaining economically feasible. These will be briefly described in this
section along with their associated costs and potential ability to reduce crashes when added to a 2- quad gate
system.
5.1. POTENTIAL RAIL CROSSING UPGRADES
5.1.1. LONG- ARM GATES
Gate- arms at gated crossings typically extend to the centerline of the road and are currently prohibited from
extending further by the California Public Utility Commission’s General Order 75- C. Where they are legal and have
been deployed, longer gate arm systems, which cover at least 3/ 4 of the roadway, have been shown to be an effective
means of discouraging gate “ drive- arounds” ( Caird et al. 2002; FRA, 2001).
Long- arm gates have been deployed successfully in the North Carolina sealed corridor between Charlotte and
Raleigh, NC. Lessons learned from that deployment include:
1 At least 6’ of shoulder are needed on
each side of the road so that cars that
go under a descending gate can go
around the lowered arm after crossing the
tracks.
2 Long- arm gates should not be installed
where there is significant level of truck
traffic since even trucks that cross legally
( i. e., before the gates start down) can clip
the gate as it starts down on the far side
of the crossing.
3 Long- arm gates should not be installed
where there is significant level of bus
traffic for the same reason as with trucks.
4 Long- arm gates should not be installed in
locations with more than two tracks.
Figure 5
LONG- ARM GATES
14
The Norfolk Southern Railway, which is responsible for maintaining warning equipment along the corridor, has set a
maximum length of 38’ for the gate arms. Longer than this, the arms become vulnerable to breakage due to high winds.
Long- Arm Gate Estimated Efficacy: 75% ( FRA, 2001)
Estimated Cost Per Crossing: $ 5,000 ( FRA, 2001)
5.1.2. MEDIANS
For this report, medians will be taken to
mean mountable centerline medians with
channelization devices. These can be
applied directly to the existing roadway, as
shown in Figure 6, or can be part of a more
complex structure consisting of an island
with reflectors mounted on the top, as shown
in Figure 7. Such systems present drivers with
a visual cue intended to impede crossing
to the opposing traffic lane. The curbs are
no more than six inches in height, usually
less than twelve inches in width, and built
with a rounded design to create minimal
deflection upon impact. The reflectorized
paddle delineators or tubes, typically 24- 36
inches high, are built to be able to bounce
back up after being hit or run over. These
systems are designed to allow emergency
vehicles to cross over into opposing lanes
to go back in the opposite direction but not
for the purpose of circumventing the traffic
control devices at the crossing. Usually, such
a system can be placed on existing roads
without the need to widen them.
Medians are currently being used in a large
number of locations including the North
Carolina sealed corridor and in Washington
state. The durability and maintenance
experience in these locations has been good.
In Puyallup, WA, seven sites, with average
AADTs of 9,800, require replacement of three
to four upright tubes per site per year.
In North Carolina, with average AADTs of
12,000, approximately 16 uprights must be
replaced per site per year
Median Separators Estimated Efficacy:
75% ( FRA, 2005) – 80% ( FRA, 2001)
Estimated Cost:
$ 13,000 - $ 15,000 ( FRA, 2005)
Figure 6
STREET MOUNTED CHANNELIZATION
Figure 7
ISLAND MOUNTED CHANNELIZATION
15
5.1.3.
FOUR- QUADRANT
GATE SYSTEMS
Four- Quadrant Gate
Systems consist of a series
of automatic flashing- light
signals and gates where the
gates extend across both
the approach and departure
side of roadway lanes. Unlike
two- quadrant gate systems,
four- quadrant gates provide
additional visual constraint
and inhibit nearly all traffic
movements over the crossing
after the gates have been
lowered. At this time, only
a small number of four-quadrant
gate systems have
been installed in California
and incorporate different
types of designs to prevent
vehicles from being trapped
between the gates.
Four- Quad Gates Estimated
Efficacy: 82% ( FRA, 2001)
Estimated Cost: $ 125,000
( FRA, 2001) to $ 350,000
Costs for the installation of
4- quad gates vary widely. For
a single track crossing, the cost to upgrade from a passive crossing or 2- quad gate to a four- quad gate was given by
Burlington Northern Santa Fe Railroad ( BNSF) as “ well over $ 300,000.” In general, the upgrades from a 2- quad gate
are complete upgrades due to the age of existing equipment and circuitry ( Crakes, S., BNSF, unpublished data).
5.1.4. PHOTO ENFORCEMENT
The California Vehicle Code, Section 21455.5: Traffic Signal Automated Enforcement ( see Appendix H) authorizes
governments and law enforcement agencies to operate automated- enforcement systems at both traffic- light
intersections and railroad grade crossings. In the event of a signal or gate violation, such systems are can be designed
to obtain a clear photograph of the violation, the vehicle’s license plate, and the driver of the vehicle.
Photo enforcement, while not erecting a physical barrier, can still provide a very strong deterrent against
inappropriate railway crossings. In Los Angeles, a 6- month demonstration project resulted in an 84% reduction in the
number of violations ( Meadow, 1994). Considering what should already be a powerful incentive to stop at lowered
gates, it is somewhat surprising that the threat of a fine would be an effective motivator of behavior. However, the
past experience of a traffic ticket seems to carry more weight than the vague possibility of a crash, even though the
consequences of a crash could be catastrophic.
Figure 8
FOUR- QUAD GATE SYSTEM PICTURE AND DIAGRAM
16
Carroll and Warren, 2003, note that capital costs for photo enforcement can vary greatly depending on the
requirements of the community served. These requirements can include the need for a picture of front and/ or rear
license plates, pictures of the driver’s face, number of lanes, and location. One way to reduce the cost of photo
enforcement is to move one camera among several sites without drivers knowing which ones are active at any given
time. The authors list the following cost examples:
■ The Insurance Institute for Highway Safety lists equipment costs of about $ 50,000 for a red- light camera
and $ 5,000 for installation and sensors.
■ In North Carolina, the cost for a prototype system at one intersection was $ 100,000 which included four
cameras, two towers, loop detectors, infrared lighting units, software, controller and cabinet, printers and
connections, and two advance- warning signs.
■ In Florida, passive video monitoring at four sites with varying volume and numbers of tracks ( including
detection of vehicles, trains, and the status of gate arms and signal- crossing lights), using multiple
cameras, is costing nearly $ 400,000, with $ 200,000 attributed to equipment costs. The larger sum
provides for site analysis and selection, all equipment, construction and installation, and reporting.
■ In Illinois, the cost to install and maintain one installation ( site) for 1 year averages $ 300,000, with the
lower end at $ 263,000 and the high end at $ 344,000. Local police departments are also incurring costs in
conjunction with this program. Both Naperville and Wood Dale indicate that they devote approximately 1
full day per week to process citations and appear in court. Naperville has one officer responsible, assisted
by one technician, while Wood Dale has trained five officers to use the system.
Photo Enforcement Estimated Efficacy - 72% ( FRA, 2001)
Estimated Cost - $ 55,000 - $ 100,000 ( Caird et al., 2002; FRA, 2001; Carroll and Warren, 2003)
5.2. SUMMARY
In Table 9, these methods are listed along with their estimated costs and relative effectiveness. The first column lists
crossing equipment currently in use as listed in the FRA crossing inventory for California. While there may be some
state crossings that have other equipment ( e. g., four- quad gates), they are not listed in the inventory. The second
column gives:
■ Inventory: the number of state crossings with this type of equipment ( crossings are listed by their highest
level of warning device)
■ Inc/ K/ Inj: the number of incidents/ number killed/ number injured at crossings of this type in California
from 2000 to 2004
■ Cost per Inc: the average cost of each crash incident at this type of crossing.
■ Total Cost: the five- year total cost of all crashes at this type crossing
The next nine columns list the potential upgrades to the equipment listed in the first column. For each combination
of old and new equipment, three numbers are given:
■ “ E” is the effectiveness of this upgrade. A rating of E- 81% means that incidents would be reduced by
81% by upgrading to this type equipment.
■ “ C” is the cost to upgrade one crossing.
■ “ TC” is the total cost to upgrade all crossings of this type in the current inventory.
17
Table 9
COST AND EFFECTIVENESS OF HIGHWAY- RAIL CROSSING EQUIPMENT UPGRADES
18
These numbers are estimates and should be used as general indicators only in that each crossing may have unique
characteristics and conditions. In constructing this matrix, two basic assumptions were made: ( 1) multiple treatments
are multiplicative in effectiveness and ( 2) multiple treatment costs are additive.
The values and sources used for determining crash costs are:
Vehicle Damage: $ 4,680 ( Lee 2004)
Death: $ 3,052,000 ( California Highway Patrol [ CHP], 2003)
Injury: $ 104,255 ( Lee, 2004)
Calculations for the effectiveness of crossing equipment upgrades are given in Appendix D. To date, there have
been no studies showing the effectiveness of upgrading from wigwags/ audible warnings to 2- quad gates. In lieu of
this information, the cost and effectiveness of upgrading from flashing lights to 2- quad gates will be used. The costs
should be similar and the given effectiveness will be a conservative estimate for this type of upgrade.
5.3. BENEFIT VS. COST
Since the cost to upgrade all at- grade crossings would be prohibitive, this study attempts to determine which
crossings would yield the greatest benefit from an upgrade. First, sites with multiple crashes were examined using
ten- year crash data. Out of a total of 911 crossings which had crashes between 1995 and 2004, 252 had two or more,
and 87 had at least three ( Table 5). The complete list of the 252 multiple crash crossings is presented in Appendix E.
The warning equipment components at these sites are:
Gates: 69%
Flashing Lights: 17%
Other Active Devices: 2%
Passive Warning: 12%
Next, the cost and potential benefit of upgrading the 252 sites with multiple crashes was calculated. The minimum
upgrades considered for both passive and active sites were to include 2- quad gates plus one of the following: photo
enforcement, long- arm gates, or median separators. Four- quad gates were not included due to their substantially
higher cost. The formula used to calculate the potential annual benefit for each site was:
Benefit = ( AvgCrash x Eff) x AvgCrashCost
Where:
AvgCrash = the average annual number of crashes at this site
Eff = the effectiveness of the upgrade
AvgCrashCost = the average cost of a crash at this type of crossing
As an example, to upgrade from a 2- quad gate to 2- quad + median separators at crossing number 026476Y in
Riverside, which had four crashes in the ten years from 1995 to 2004:
Annual Benefit = ( 0.4 x 0.8) x $ 592,352 = $ 189,553
The cost to add median separators is $ 14,000. The potential annual benefit benefit/ cost ratio is: $ 189,553/$ 14,000
= 13.5. The same ratio for a similar site with two crashes in the ten year period rather than four, would be:
$ 94,776/$ 14,000 = 6.8.
These methods were applied to all multi- crash sites. Although it is unlikely that all sites would have the same upgrade,
there are too many possible combinations to list here. As such, it was assumed that all sites will receive the same final
equipment. The results are shown in Table 10.
19
It should be remembered that the values of this section are based on property damage, injury, and death cost
estimates. The results, therefore, show an unrealistic degree of precision that should be, at the least, rounded to the
nearest thousand. These results could change greatly if the assumptions underlying the cost estimates are altered.
6. DRIVER DECISIONS AT RAIL CROSSINGS:
A CONCEPTUAL MODEL
What failures in perception or judgment would cause 503 drivers ( 2000- 2004) to ignore active warnings ( gates and/ or
flashing lights) and become involved in crashes with trains and, even more incredibly, would cause 84 of them to
drive around or through gates INTO the side of a train? This section aims to provide insight into the interplay of
perception, expectation, and human information processing which can assist in the development of strategies for
grade crossing crash prevention.
6.1. SIGNAL DETECTION THEORY
Signal detection theory ( SDT) has been used by a number of researchers as a means of analyzing and predicting
railroad crashes ( e. g., Raslear, 1995, Rapoza and Fleming, 2002). “ The starting point for signal detection theory is
that nearly all reasoning and decision making takes place in the presence of some uncertainty” ( Heeger, 1997). Thus,
someone at a party trying to determine if they have previously met someone, a radiologist looking for evidence
of a tumor, and a motorist at a rail highway crossing are all in the same situation of trying to detect a signal in a
background of noise. In all of these situations, it is often difficult to distinguish signal from noise, and a decision will
be made which is not solely dependent upon the sensory information alone.
In the SDT model, both the signal and the noise are represented as a single internal response continuum which
varies in magnitude. Even if all of the sensory inputs to an individual are identical, signals, such as the locomotive,
are capable of producing perceptual magnitudes which vary between encounters. This produces a
“... probability distribution of internal response which is associated with a particular locomotive configuration
( e. g., size, loudness, color, brightness, etc.). This distribution of perceptual magnitudes has a mean and
variance which can be used to specify the perceptual magnitude of the locomotive as a signal. Similarly, the
background noise also has a distribution of perceptual magnitudes which can also be specified by a mean
and a variance. For the sake of simplicity it is often assumed that the distribution of perceptual magnitudes
2- Quad Gates
+ Photo
2 Quad
+ Long- Arm
Gates
2 Quad
+ Long- Arm
Gates + Photo
2- Quad Gates
+ Median
Separators
2- Quad Gates +
Median Separators
+ Photo
Costs To Upgrade to These Levels
Upgrade Sites with 3 to 12 Crashes $ 8,030,000 $ 3,730,000 $ 8,460,000 $ 4,504,000 $ 9,234,000
Upgrade Sites with 2 or More Crashes $ 25,710,000 $ 13,110,000 $ 26,970,000 $ 15,378,000 $ 29,238,000
Expected Annual Upgrade Savings
Upgrade Sites with 3 to 12 Crashes $ 13,959,844 $ 14,459,172 $ 17,415,505 $ 15,291,108 $ 17,591,717
Upgrade Sites with 2 or More Crashes $ 28,492,914 $ 29,460,869 $ 35,185,348 $ 31,079,117 $ 35,531,307
Expected Benefit/ Cost Ratio 1.1 2.2 1.3 2.0 1.2
Table 10
BENEFITS AND COSTS TO UPGRADE CALIFORNIA MULTI- CRASH CROSSINGS
20
for noise and signal are normal.
Additionally, the basic SDT model
assumes that the variances of signal
and noise distributions are equal,
although this assumption is not
critical to the theory” ( Raslear, 1995).
A typical representation of noise and
signal plus noise only distributions are
shown in Figure 9.
A key point to note is that the distributions
overlap. Thus there are times when it is
not possible to distinguish between signal
and noise, necessitating the adoption of
some other means to decide which it
is and what action to take. This is the
criterion and the point on the internal response axis at which this criterion is set is the criterion line ( see Figure 10).
In the case of the motorist at a crossing, the criterion line provides the basis for the decision to stop ( all points to
the right of the line) or continue crossing ( all points to the left of the line). There are four potential outcomes for the
decision as shown in Table 11. There are two response categories: “ Stop ( the train is too close)” and “ Don’t Stop
( the train is not too close).” And there are two possible events: a train is close to the crossing and a train is not too
close to the crossing ( or not present).
These outcomes can be seen in Figure 10 where the train is close in diagram ( a) and not close or absent in diagram
( b). For our purposes, the
more important question is
not whether or not the train
is perceived as present but
rather is it perceived as close
enough and moving fast
enough to represent a threat
to the driver’s crossing the
tracks ahead of it.
0
0.1
0.2
0 2 4 6 8 1 0 1 2 1 4 1 6
Internal Response
Probability
Distribution When
Signal Is Not Present
Distribution When
Signal Is Present
Figure 9
INTERNAL RESPONSE PROBABILITIES FOR NOISE
WITH SIGNAL AND NOISE ONLY
Stop Don’t Stop
Train Is Close Valid Stop Crash
Train Is Not Close, or
No Train
False Stop
( driver stops unnecessarily)
Correct Crossing
( driver crosses safely)
Table 11
POTENTIAL OUTCOME MATRIX
0
0.1
0.2
0 2 4 6 8 1 0 1 2 1 4 1 6
Internal Response
Probability
Criterion Response
Don't Stop Stop
Incorrect
" Train"
Decision
Correct
" No Train"
Decision
Figure 10
INTERNAL RESPONSE PROBABILITY CURVES
( a) Signal ( Train) Present ( b) Signal ( Train) Not Present
21
0
0.1
0.2
0 2 4 6 8 1 0 1 2 1 4 1 6
Internal Response
Probability
Correct
" Train"
Decision
Incorrect
" No Train"
Decision
Criterion Response
Don't Stop Stop
In diagram ( a), where the train is close, the striped area to the right of the criterion represents the correct decision
to stop. The shaded area to the left of the line is the incorrect decision to proceed, resulting in a crash. In diagram
( b), the striped area represents the correct decision to proceed, while the shaded area is the decision to stop
unnecessarily.
For any given level of detectability of the signal, moving the criterion response line will change the probabilities of the
potential outcomes. By choosing a low criterion, the driver could be assured a very low probability of crashes but at
the cost of a large number of unnecessary stops. The effects of shifting the criterion response line are shown in Figure
11. It is important to note that the criterion for detection is not consciously set, but rather corresponds to the amount
of visual “ evidence” required for detection, which itself can be heavily influenced “ by the observer’s expectations
( probability of signal, probability of noise), motivation ( values of each of the decision outcomes), and other cognitive
functions ( e. g., memory, attention, decision strategy). For instance, a driver who is familiar with a particular grade
crossing has an expectation regarding the frequency of trains at that crossing” ( Raslear, 1995).
Note that changes in the criterion do not change the distribution of the detectability of the proximity of the train.
The only means in this model of altering detectability is to move the signal and noise distributions further apart,
thus lessening the area of overlap. There are three ways to achieve this: ( 1) decrease the level of background noise
( Figure 12a), ( 2) increase the level of the signal ( Figure 12b), and ( 3) change the variance of one or both distributions.
Mathematically, how detectable the signal is from no- signal can be expressed as:
0
0.1
0.2
0 2 4 6 8 10 12 14 16
0
0.1
0.2
0 2 4 6 8 10 12 14 16
0
0.1
0.2
0 2 4 6 8 10 12 14 16
Valid Stops = 50%
False Stops = 16.7%
Valid Stops = 84.1%
False Stops = 50%
Valid Stops = 97.7%
False Stops = 69.2%
Figure 11
EFFECTS OF SHIFTING CRITERION RESPONSE LINE
0
0.05
0.1
0.15
0.2
0.25
0 2 4 6 8 10 12 14 16
Internal Response
Probability
0
0.05
0.1
0.15
0.2
0.25
0 2 4 6 8 10 12 14 16
Internal Response
17 18
Probability
Figure 12
CHANGES IN PERCEPTUAL DISTRIBUTION
( a) Lower Noise ( b) Stronger Signal
22
Again, changes in the criterion only affect the probabilities of the outcomes, while changes in the distributions can
effect a change in both detectability and the probabilities of the outcomes ( Raslear, 1995).
Given that over 86% of the 593 crashes that occurred between 2000 and 2004 took place at crossings with active
warning devices, it would appear that knowledge of the presence of a train is not sufficient reason to stop for some
people. For them, the problem is determining the speed and proximity of the train, rather than its presence.
SDT indicates that there are two classes of variables which can be manipulated to prevent crashes: ( 1) variables
which increase the Signal/ Noise Ratio and ( 2) variables which increase the bias to stop. An approaching train gives
off a large signal, with visual, auditory, and physical characteristics. While there are several signal boosting strategies
available to further the detectability of trains ( e. g., enhancing locomotive conspicuity, reflectorization of freight cars,
and altering the train horn), this strategy does not appear to be especially promising given that determining train
speed and proximity are the problem, rather than just train presence.
A more promising strategy might be to increase the S/ N ration by decreasing noise, thus allowing more effort to
be spent on speed and distance judgments. Raslear ( 1996) noted that grade crossings with active devices actually
have lower train detectability values than crossings with passive or no devices. This could be due to the fact that the
warning equipment is not part of the train, so the increases in light and sound at the crossing acts as a distraction,
decreasing the S/ N ratio. Interestingly, SDT predicts that automated horns and illumination of grade crossings should
increase the accident rates at grade crossings for the same reason ( Raslear, 1996). Following this line of reasoning,
one possible crossing enhancement might be to change the flashing lights to steady red and stop the bells once the
gates are fully down. The motorist at this point is aware of the presence of the train and can concentrate on speed
and location.
Another method to increase S/ N, is to improve the line of sight of the motorist at the crossing and reduce visual
clutter ( e. g., other traffic, traffic signs and signals, street lights, etc.). Obviously, visual information is extremely
important when compared to other sensory information for determining speed and proximity, so any improvements
could have a large effect on reducing noise and strengthening the signal.
Raslear ( 1996) quotes a recent FRA study of 56 grade crossings with an average of more than one accident per year
that found that 97% of these crossings had visual obstructions, 95% had a large number of driveways and intersecting
roadways, and 80% had visual clutter on the approach.
Finally, directing a driver’s attention toward the train may serve to enhance the S/ N ratio. Signs which indicate where
motorists should look could function to enhance both detectability and bias to stop. Signals and other changes in
the sensory stimulation provided by grade crossing devices should be more focused on causing motorists to orient
toward the train rather than just indicating the train’s presence ( Raslear, 1996). Care must be taken, however, that the
indicator cannot be misinterpreted. A lighted arrow, for example, could be interpreted as pointing to where the train
is OR the direction it is traveling.
In addition to changing the S/ N ratio, increasing a motorist’s bias to stop should also reduce rail- highway grade
crossings. This bias has been shown to be strongly influenced by expectation and motivation. The first of these is best
illustrated by the fact that accident rates vary inversely with train frequency. While this at first seems counterintuitive,
the key word here is “ rates.” As Lerner et al. ( 1990) reported, “ If the driver assigns a low probability to the presence
of a train... he will adopt a higher criterion for detecting the train, and this will increase his chances of [ not seeing it].
It is important to note that the criterion for detection is not consciously set, but rather corresponds to the amount of
visual ‘ evidence’ required for detection.”
One method of increasing the bias to stop is through the use of enforcement. In Los Angeles, a photo enforcement
demonstration project was conducted in 1992 that began with the un- announced installation of cameras at two
locations where counts were made over a two month period to serve as a baseline for evaluation of the system.
23
Following this, a press
conference was held and
signs were installed at
the crossings. After two
months of sending out
warnings only to violators,
ticketing began and
continued for four months.
The demonstration pro-ject
resulted in an 84%
reduction in the number
of violations ( Meadow,
1994).
Considering what should
be an already powerful
incentive to stop at
lowered gates, it is some-what
surprising that the
threat of a $ 50 or $ 100
fine would be an effective
motivator of behavior. As
Raslear ( 1996) points out,
however, there are other
costs associated with fines including inconvenience and loss of time, embarrassment caused by publicly receiving a
fine and the possibility of losing one’s license due to the points that might be added to the driver’s record. Another
possible reason for the effectiveness of photo enforcement is that most people have firsthand knowledge of receiving
a ticket whereas very few have been hit by a train. Thus, the certainty and past experience of a ticket seem to carry
more weight than the vague possibility of a crash, even though the consequences of a crash could be catastrophic.
6.2. PERCEPTION OF TRAIN SPEED AND DISTANCE
Between 2000 and 2004, 73% of drivers involved in crashes had been made aware of the approaching train by the
presence of lowered gates. If we assume that a driver ignores this warning and decides to proceed across the tracks
because he or she believes there is enough time to do so safely, there must be some perceptual problems that affect
an individual’s ability to make this judgment correctly.
Detecting speed or time to collision from changes in an object’s size has been shown to be relatively difficult
( Leibowitz, 1985). In addition to problems associated with judging speeds of large objects ( discussed in greater
detail in the next section), as an object approaches, the growth in size is not linear but hyperbolic, with the apparent
rate of growth of a distant object being quite slow and then accelerating as the object gets closer ( See Figure G3
in Appendix G). The result is that drivers tend to be effective at estimating the speed of the train when it is closest
because the change in visual angle is rapid, but when the train is at greater distances, at the time when drivers tend
to decide on the safety of proceeding across the tracks, the change in visual angle is slow and they are more likely
to underestimate the train’s speed ( NTSB, 1998).
This phenomenon can be seen in Figure 13, taken from an NTSB simulation of a train approaching a stationary car
at 40 MPH from a distance of 1,000 feet. Each frame represents the movement of the train covering one quarter of
the original distance. Half of the distance is covered before any appreciable difference in the size of the train can be
noted and the remaining time to collision is only 8.5 seconds.
Figure 13
VIEW OF APPROACHING TRAIN FROM
VEHICLE STOPPED AT CROSSING
24
6.3. THE LEIBOWITZ HYPOTHESIS: EXPERIMENTAL RESULTS
In 1985, H. W. Leibowitz suggested that drivers underestimate
the speed of trains because human vision underestimates
the speed of large objects. The author of this theory
introduced only anecdotal evidence in its favor ( a 747 seems
to land more slowly than a Piper Cub, though the opposite
is true). Cohn and Nguyen ( 2003) found indirect evidence
that he may have been correct. If so, at least some of the
collisions at rail crossings might be due to a simple driver
misperception and specific countermeasures might then be
examined.
According to Barton et al. ( See appendix G), the Leibowitz’
hypothesis has never been tested, and so the authors set
out to do this using a 3D visual simulator. They constructed a
two alternative, forced choice ( 2AFC) experiment consisting
of two sequential time epochs. In one of the epochs, chosen at random, a five foot diameter sphere approached the
observer at eye level, traveling at 35 mph. In the other epoch, a ten foot diameter sphere approached at one of the
speeds given in Table 12. The observer’s task was to indicate by pressing a button which epoch contained the faster
approaching sphere. An experiment consisted of 270 such trials.
The authors tested the ability of five
males, ranging in age from the early
20s to the mid 50s, with corrected
normal eyesight to identify the faster
of two different sized approaching
spheres. The results of these tests
are summarized in Figure 14, which
plots, for each subject, the proportion
of times the 5 ft diameter sphere was
judged to be faster ( P5) as a function
of 10 ft sphere speed ( V10). This
shows a strong tendency to judge
the smaller sphere as the faster,
even when the actual approach
speed of the larger sphere is 20 mph
greater ( V10= 55 mph). Only when
V10 reaches speeds of 65- 75 mph
( twice that of the smaller sphere) does the observer become unsure as to which is approaching faster ( P5≈ 0.5).
The experimental data, then, show a strong tendency to judge the smaller ball to be the faster, even when the
opposite is the case, and often by a considerable margin. The plots in Figure 14 suggest that experimenters would
have to include trials in which the large ball approaches in excess of 95 mph ( 2.7 times faster than the small ball)
before subjects would unambiguously pick the large ball as the faster approaching.
6.4. DRIVER DECISIONS CONCLUSION
From both signal detection theory and the tests of the Leibowitz hypothesis, it is apparent that, in general, humans
have a great deal of difficulty in judging the speed and distance of an oncoming train as depicted in the nearly
Speed ( mph) # Trials ( Out of 270)
25 40
35 40
45 40
55 50
65 50
75 50
Table 12
APPROACH SPEEDS OF THE
LARGE ( 10’) SPHERE
0.0
0.2
0.4
0.6
0.8
1.0
25 35 45 55 65 75
Large ( 10 ft.) Ball Speed
Proportion Small Ball Chosen As Faster
Small Ball
Faster
Small Ball
Slower
Figure 14
TEST OF THE LEIBOWITZ HYPOTHESIS
25
overlapping signal and signal- plus- noise curves in Figure 15. Since no amount of sight- line improvements, train
conspicuity improvements, or warning system upgrades will improve this situation, the solution to rail crossing
crashes must be found by removing the need to make such a decision ( i. e., driving the criterion response point all the
way to the left) by making it impossible, or at least very difficult, for the driver to bypass the lowered gates.
7. CROSSING OBSERVATIONS
Observation of drivers at rail crossings provides a valuable tool for understanding their behavior under different
combinations of grade crossing equipment and train frequencies and speeds. Three different methods were
examined: a crossing camera in College Station, Texas, a crossing camera in Berkeley, California, and a train engine
based camera in Napa, California. This section presents the results of these observations. A complete description of
the sites, procedures, setups, and results can be found in Appendix I.
7.1. COLLEGE STATION, TEXAS
College Station, Texas, population of 70,000,
is located 90 miles northwest of Houston. It
has a rail monitoring system, The College
Station ITS Integration Project ( CSIP), set
up along the Wellborn Road Corridor which
is a major north- south arterial in College
Station. The system was set up to provide
the City’s Fire Station # 4 with grade crossing
status and travel time prediction information
for trains traveling in both directions in the
project corridor to aid station personnel in
making route decisions when servicing an
emergency call.
Adjacent to Wellborn Road lies the Union
Pacific Railroad’s Fort Worth Subdivision
mainline which carries approximately 20 to
25 trains per day, varying from 1⁄ 2 mile to
11⁄ 2 miles in length. Train speed through
the corridor can be as low as 15 to 20 mph
0
0.05
0. 1
0.15
0. 2
0.25
0 2 4 6 8 10 12 14 16
Internal Response
17 18
Probability
Sufficient Time To Cross Insufficient Time To Cross
Criterion Response
Don't Stop Stop
Figure 15
OVERLAPPING SIGNAL AND SIGNAL- PLUS- NOISE CURVES
0
0.05
0.1
0.15
0.2
0.25
0 2 4 6 8 10 12 14 16
Internal Response
17 18
Probability
Sufficient Time To Cross Insufficient Time To Cross
Criterion Response
Don't Stop Stop
Figure 16
COLLEGE STATION, TEXAS,
HOLLEMAN DRIVE CAMERA VIEW
26
in the northern end of the corridor and as high as 50 mph at the southern end. Trains in the corridor do not travel
on a fixed time schedule, but arrive randomly throughout the day, depending on train traffic ( Texas Transportation
Institute, 2005).
PROCEDURE
Approximately 300 hours of live video feed from the College Station Holleman Avenue camera was downloaded
from the internet and stored over a total of 24 weekdays between June 22, 2005 and September 2, 2005. Train speed
information was also recorded during this period
RESULTS
During the observation period, 116 gate cycles during which cars were present, were recorded. During 45 of those,
cars were present in the storage area beyond the tracks, preventing approaching traffic on Holleman from crossing
the tracks. In the remaining 71 cycles, 48 cars had the opportunity ( defined as arriving at the crossing before the road
was blocked by the gate) to go under the descending gate and 28 cars ( 58%) did so. One of the 28 cars went around
stopped traffic and one car was hit by the gate.
Also during the 71 unblocked cycles, nine cars went around a lowered gate. Six of these took place after the train had
passed and the gate did not go up. Two of the remaining three occurred in front of a train traveling at seven miles-per-
hour and the last one in front of a train traveling at 26 miles- per- hour. In the case of the slow train, 35 seconds
passed from the time the second car cleared the tracks until the train arrived. In the third case, the train arrived at the
crossing nine seconds after the car had cleared.
7.2. BERKELEY, CALIFORNIA
The Gilman street crossing in Berkeley,
California, has two lanes of traffic crossing
three sets of tracks, of which only two are
used ( Figure 17). The crossing is equipped
with two quadrant gates, bells and flashing
lights. There are up to 70 trains per day
including 24 operated by Amtrak’s Capitol
Corridor, consisting of an engine and four
passenger cars traveling at speeds up to 60
MPH.
Observations at this location were recorded
using two cameras, each located in the back
of a van parked along Gilman Avenue. Each
camera was set up so as to shoot traffic
coming at it diagonally across the tracks.
RESULTS
Over a period of four days, there were a total
114 gate cycles with vehicles present ( eastern
and western gate cycles counted separately). There were 86 opportunities for a vehicle to go under a descending
gate — 17 vehicles ( 19.8%) did so. No cars went around fully descended gates.
Figure 17
GILMAN AVENUE CROSSING
BERKELEY, CALIFORNIA
27
7.3. NAPA, CALIFORNIA
The Napa Valley Wine Train provides a 3- hour round- trip covering the 36- miles beginning in the town of Napa,
through the village of St. Helena, and back. The train consists of nine rail cars and a double- sided Alco Diesel Engine.
The data collected from this train comes from a camera mounted in the engine and operated by the engineers.
The resulting tapes were obtained from the Napa Valley Railroad Police Department. While the data are anecdotal
in nature they provide valuable insight into the public’s general lack of knowledge of both the law regarding rail
crossings and the basic laws of physics. One person, for example, a passenger in a car that had stalled on the tracks,
got out of her car and stood between the car and the oncoming train, waving for the engineer to stop. Fortunately,
a woman in another car got out and dragged the first woman to safety just before the train hit her car.
8. CONCLUSIONS & RECOMMENDATIONS
Rail- Highway grade crossing collisions fall under the category of bilateral accidents in that the probability of their
occurrence is affected by both the railroad and the other involved party ( Savage, 1998). Between 2000 and 2004, there
were 99 people killed and 205 injured due to collisions between motor vehicles and trains at rail highway crossings
in California, virtually all the fault of the highway user.
There is a group of drivers, more than half less than 40 years old, and male by a ratio of three to one, who are not
deterred by lowered gates and have a misplaced confidence in their ability to judge train location and speed. Signal
detection theory tells us that the decision to proceed or stop at a rail crossing is a function of our ability to separate
signal from noise ( both external and internal), and the criterion point, which is itself a function of expectation, prior
experience, and personality.
It would seem, then, that to cut the crash rate at grade crossings, we could begin by finding a means to increase
the S/ N ratio. This might consist of increasing signal strength by increasing train conspicuity ( although this would be
difficult to accomplish during daylight hours), installing some form of indicator of where to look for the train, and/ or
decreasing noise by improving viewing angles and switching to a steady red light instead of flashing red light and
quieting the bells once the arms are fully down.
But at a fully functioning gated crossing, where 73% of California’s crashes occurred, the driver has been fully
informed, by means of lowered gates, that a train is near. Should we be concerned about providing better information
to the driver in order to facilitate a more informed decision to run the gates? In fact, could every effort we make
to increase the SDT signal ( train conspicuity, louder horns, etc.) and decrease noise ( better sight lines, turning off
flashing lights once the gate is down) actually encourage gate running by increasing driver confidence in his/ her
ability to judge train speed and distance?
From both signal detection theory and the tests of the Leibowitz hypothesis, it is apparent in general, that humans
have difficulty judging the speed and distance of an oncoming train. Since no amount of sight- line improvements,
train conspicuity improvements, or warning system upgrades will improve this situation, the solution to rail crossing
crashes must be found by removing the need to make such a decision. This translates to making it impossible, or at
least very difficult, for the driver to bypass the lowered gates.
While making it impossible to violate a crossing can be accomplished in a number of ways, including constructing a
separation of grade, closing the crossing, or by deploying an impenetrable barrier, this solution tends to be relatively
expensive. There are, however, two low technology, low cost, and low maintenance methods that while not being
100% effective, have been deployed in many locations and shown to prevent deaths and injuries while remaining
economically feasible. These are long- arm gates and median separators.
28
9. SUGGESTIONS FOR FURTHER RESEARCH
There appears to be widely held belief among public agency decision makers that implementation of safety related
measures can, unless universally applied, expose the agency to liability lawsuits. The feeling is that public plaintiffs
will argue that the addition of a safety device ( e. g., upgrading a rail- highway crossing) is a tacit admission of the
existence of a dangerous condition and putting it one place and not another constitutes negligence on the part of
the agency. The question to be answered is whether or not lawsuits of this type actually occur and, if so, are they
being won by the plaintiffs?
The second area for future study involves those sites with multiple crashes. Specifically, do these sites differ in some
significant way from other rail- highway crossings?
Finally, as previously discussed in the section on crossing angles ( Section 4.3.7), while crossing angle appears to play
no part in crash rates, it may very well be that viewing angle does. This needs to be investigated further.
9. REFERENCES
Caird, J. K., Creaser, J. I. , Edwards, C. J., Dewar ( 2002), A Human
Factors Analysis Of Highway- Railway Grade Crossing Accidents
In Canada, Transportation Development Centre Transport
Canada
California Highway Patrol ( 2003), 2003 Annual Report Of Fatal
And Injury Motor Vehicle Traffic Collisions, Sacramento CA
Carlson, Paul J., Fitzpatrick, Kay ( 1999), Violations at Gated
Highway- Railroad Grade Crossings, Transportation Research
Record 1692
Carroll, Anya A. and Warren, Judith D. ( 2002), Photo Enforcement
at Highway– Rail Grade Crossings in the United States, July
2000– July 2001, Transportation Research Record 1801 Paper
No. 02- 2517
Cohn, T. E., Nguyen , T. ( 2003), A Sensory Cause of Railroad
Grade- Crossing Collisions: Test of the Leibowitz Hypothesis,
Transportation research record. No. 1843 ( 2003)
Dewar , Robert E., Olson , Paul L., Editors ( 2002), Human Factors
In Traffic Safety, Lawyers & Judges Pub Co., Tucson, AZ
Farr, E. H. and Hitz, J. S. ( 1985), Effectiveness of Motorist Warning
Devices at Rail- Highway Crossings, Publication No. FHWA- RD-
85- 015. Federal Highway Administration, Washington, D. C.
Federal Railroad Administration ( 2001), North Carolina “ Sealed
Corridor” Phase I, U. S. DOT Assessment Report: Report to
Congress
Federal Railroad Administration, ( 2004), Audit Of The Highway-
Rail Grade Crossing Safety Program, Federal Highway
Administration, Federal Transit Administration, Report Number:
MH- 2004- 065 Date Issued: June 16, 2004
Federal Railroad Administration ( 2005), Use of Locomotive
Horns at Highway- Rail Grade Crossings: Final Rule, 49 CFR
Parts 222 and 229, Federal Register, Vol. 70, No. 80
Lee, Douglass Jr., Gay, Kevin, Carroll, Anya, Hellman. Adrian,
Sposato. Suzanne ( 2004), Benefit- Cost Evaluation of a Highway-
Railroad Intermodal Control System ( ICS) Final Report, U.
S. DOT, Volpe National Transportation Systems Center,
Cambridge, MA
Leibowitz, H. W., ( 1985), Grade Crossing Accidents and Human
Factors Engineering, American Scientist. Vol. 73, pp. 558- 562.
Lerner, N., Ratte, D., and Walker, J., ( 1990), Driver Behavior
At Rail- Highway Crossings ( Report No. FHWA- SA- 90- 008).
Washington, DC: Federal Highway Administration, U. S.
Department of Transportation.
Meadow, Linda ( 1994), Los Angeles Metro Blue Line Light Rail
Safety Issues, Transportation Research Record. No. 1433
Meeker, Frank L., Barr, Robin A. ( 1989) An Observational Study
Of Driver Behavior At A Protected Railroad Grade Crossing As
Trains Approach, Accid. Anal & Prey. Vol. 21, No. 3, pp. 255-
262, 1989
Meeker, Frank, Fox, Daniel, and Weber, Christopher. ( 1997), A
Comparison Of Driver Behavior At Railroad Grade Crossings
With Two Different Protection Systems, Accident Analysis &
Prevention., Vol. 29, No. I, pp. 16, 1997
National Transportation Safety Board, ( 1998) Safety At Passive
Grade Crossings. Volume 1: Analysis, Safety Study NTSB/ SS-
98/ 02. Washington, DC.
Rapoza, Amanda S., Fleming, Gregg G. ( 2002), Determination
of a Sound Level for Railroad Horn Regulatory Compliance, U. S.
Department of Transportation, Federal Railroad Administration,
Washington, D. C.
Raslear, T. G. ( 1995). “ Driver Behavior at Rail- Highway Grade
Crossings: A Signal Detection Theory Analysis,” In: Safety of
Highway- Railroad Grade Crossings, Research Needs Workshop,
Volume II-- Appendices, Caroll, A. A. and Helser, J. L. ( Eds.).
Report No. DOT/ FRA/ ORD/ 14.2, DOT- VNTSC- FRA- 95- 12.2,
pp. F9- F56. U. S. Department of Transportation, Washington,
D. C
Texas Transportation Institute ( 2005), Evaluation and Lessons
Learned From The College Station Integration Project, College
Station, TX
Wigglesworth, E. C. ( 1979), Epidemiology of road- rail crossings
in Victoria. Journal of Safety Research 11, 162.
Witte , Kim and Donohue , William ( 2000), Preventing Vehicle
Crashes With Trains At Grade Crossings: The Risk Seeker
Challenge, Accident Analysis and Prevention 32 ( 2000) 127–
139
29
11.1. APPENDIX A:
CALIFORNIA PUC SAMPLE FORM A CROSSING INVENTORY ENTRY
30
31
32
11.2. APPENDIX B:
FRA CROSSING INVENTORY EXAMPLE
33
34
11.3. APPENDIX C:
SAMPLE ACCIDENT REPORT AND NARRATIVE
35
11.4. APPENDIX D:
UPGRADE EFFECTIVENESS CALCULATION AND SOURCES
36
11.5. APPENDIX E:
CRASH SITES WITH MULTIPLE CRASHES 1995- 2004
37
38
39
40
41
42
11.6. APPENDIX F:
FRA WEBSITE CONTENTS
43
11.7. APPENDIX G:
LEIBOWITZ HYPOTHESIS
44
45
0.0
0.2
0.4
0.6
0.8
1.0
25 35 45 55 65 75
10 ft Sphere Speed ( mph)
P S
S1
S2
S3
S4
S5
Overall
46
0
5
10
15
20
25
0 1 2 3 4 5 6
Time ( sec)
θ
( deg)
25 mph
35 mph
45 mph
55 mph
65 mph
75 mph
5' Dia Sphere
95 mph
135 mph
0
5
10
15
0 1 2 3 4 5 6
Time ( sec)
d
θ
/ dt ( deg/ sec)
25 mph
35 mph
45 mph
55 mph
65 mph
75 mph
5' Dia Sphere
95 mph
135 mph
11.8. APPENDIX H:
CALIFORNIA VEHICLE CODE:
AUTOMATED ENFORCEMENT: PHOTOGRAPHIC RECORDS
47
48
49
11.9. APPENDIX I:
CROSSING OBSERVATIONS
50
51
52
53
54
1
1
Gilman Ave
55
2
Gilman Ave
2
56