DOT HS 811 037
September 2008
The Influence of
Rear Turn Signal
Characteristics on Crash Risk
This document is available to the public from the National Technical Information Service, Springfield, Virginia 22161
This publication is distributed by the U.S. Department of Transportation, National Highway
Traffic Safety Administration, in the interest of information exchange. The opinions, findings,
and conclusions expressed in this publication are those of the author(s) and not necessarily those
of the Department of Transportation or the National Highway Traffic Safety Administration. The
United States Government assumes no liability for its content or use thereof. If trade or
manufacturers’ names or products are mentioned, it is because they are considered essential to
the object of the publication and should not be construed as an endorsement. The United States
Government does not endorse products or manufacturers.
Note: Agency staff will be performing additional analysis on the safety effectiveness of
amber/red turn signals. Once that work is complete, the agency will conclude whether any action
may be warranted.
Technical Report Documentation Page
1. Report No.
2. Government Accession No.
3. Recipient’s Catalog No.
DOT HS 811 037
4. Title and Subtitle
5. Report Date
The Influence of Rear Turn Signal Characteristics on Crash
Risk
September 2008
7. Author(s)
8. Performing Organization Report No.
6. Performing Organization Code
John M. Sullivan and Michael J. Flannagan
9. Performing Organization Name and Address
10. Work Unit no. (TRAIS)
The University of Michigan
Transportation Research Institute
2901 Baxter Road
Ann Arbor, MI 48109-2150 U.S.A
11. Contracts or Grant No.
DTNH22-05-D-01019
12. Sponsoring Agency Name and Address
13. Type of Report and Period Covered
National Highway Transportation Safety Administration
1200 New Jersey Avenue SE.
Washington, DC 20590
14. Sponsoring Agency Code
15. Supplementary Notes
16. Abstract
The relationship between the relative risk of a rear-end collision involving a turn, merge, or lane change
maneuver and the characteristics of the rear turn-signal configuration was examined using rear-end
collision crash data pooled from seven States. To perform the analysis, a detailed database of rear signal
characteristics was developed for the vehicles most frequently involved in crashes among five of the
States. The signal lamp characteristics were combined with other factors contained in the crash record in
a stepwise logistic regression that modeled the odds of a rear-end collision while performing a turnsignal-related maneuver as a function of all of these factors. Two contrast groups were used as the
denominator of the odds ratio in separate analyses. The first contrast group was comprised of the
striking vehicles involved in turn-signal-related rear-end collisions. The second contrast group was
comprised of rear-struck vehicles not engaged in turn-signal relevant maneuvers.
The first analysis suggests that there is an association between amber turn signals and a reduction of
between 3 and 28 percent in the odds of being the struck (versus striking) vehicle in a turn-signalrelevant maneuver. The second analysis found no association between turn signal color and the odds of
being struck in a turn-signal-relevant maneuver. The differences in the two analyses are discussed in
terms of how well the respective contrast groups are insulated from potential effects of turn signal
configurations.
Additional analyses substituted vehicle series name with body style as a predictor in the model, and
examined turn signal characteristics in vehicles that have been produced with both amber and red turn
signals at different times.
Although the analysis suggests that there may be a safety benefit associated with amber turn signals, it is
unclear that turn signal color itself is completely responsible for the benefit. Further investigation of
factors confounded with lamp color seems warranted before drawing a strong conclusion that turn signal
color, by itself, is responsible for the observed differences.
17. Key Words
18. Distribution Statement
Turn signals, crash risk, amber turn signals,
rear-end collisions
Unlimited
19. Security Classification (of this report)
20. Security Classification (of this page)
None
None
21. No. of Pages
37
22. Price
Contents
Contents ......................................................................................................................................... iv
Executive Summary .........................................................................................................................v
Method .......................................................................................................................................v
Results and Discussion ..............................................................................................................v
Conclusions.............................................................................................................................. vi
Part I: The Association of Turn Signal Characteristics With Rear-End Collisions........................1
Overview....................................................................................................................................1
Signal characteristics related to crash risk .................................................................................3
General analysis approach .........................................................................................................6
Method .............................................................................................................................................9
Vehicle Selection and Rear Signal Database Development ......................................................9
Crash Scenario Selection and Data Processing........................................................................15
Logistic Regression Models.....................................................................................................19
Results............................................................................................................................................22
Analysis 1: Log Odds of Struck/Striking Role ........................................................................22
Analysis 2: Log Odds of Relevant/Non-Relevant Collisions ..................................................27
Analysis 3: Vehicles grouped by body style............................................................................29
Analysis 4: Turn-signal color changes within models.............................................................33
Conclusions....................................................................................................................................36
References......................................................................................................................................37
iv
Executive Summary
Requirements for the color of rear turn signals differ between the standards set by
the Economic Commission for Europe (ECE) and the United States standards. In the
ECE standards, all rear turn signals are required to be amber, while in the United States,
they can be either red or amber. This difference has led to questions about whether there
are safety differences associated with each turn signal color. Consistent color coding of
functions might facilitate recognition of the meaning of the signal, allowing a driver to
respond more efficiently. It is also possible that, apart from color coding, an amber turn
signal is also more conspicuous to a following driver amid a field of red tail and stop
lamps. This might allow an amber turn signal to be recognized more quickly or confused
less often with another rear signal.
The possible effects of turn signal color have been explored in a limited number
of laboratory studies and crash analyses, with mixed results. Some laboratory studies
suggest no effect on driver responses, while others cite advantages with amber lamps.
Crash data analyses are similarly mixed. Examining rear-end collisions into turning
vehicles, some reports have found no evidence of an effect of turn signal color on
crash rates.
The following analysis revisits this issue with the benefit of a larger pool of crash
data than before, and a logistic regression model that include many factors—such
as driver age, gender, vehicle model, and vehicle age—that were not considered in
previous analyses.
Method
This analysis examines the crash records of seven States to determine if there is
an association between rear signal characteristics and the risk of a rear-end collision into
a turning vehicle, taking into account a variety of other characteristics associated with the
crash. To perform the analysis, detailed rear signal characteristics were determined
for the top 50 vehicle models involved in crashes in Kentucky, Maryland, New Jersey,
North Carolina, and Pennsylvania during 2003. For each model, the history of the rear
signal configuration was examined over a 15-year window beginning with 1990,
identifying rear signal color, lamp function configuration (i.e., whether a turn signal is
separated from the stop lamp or tail lamp), light source (LED, tungsten filament), and
optical characteristics.
The detailed lamp characteristics were combined with several other factors
identified in the crash data in a stepwise logistic regression that modeled the odds of a
rear-end collision while turning, merging, or changing lanes (i.e., a turn-signal-relevant
maneuver), as a function of all of these factors. In a slight departure from previous crash
analyses, the contrast group (used as the denominator of the odds ratio) in the first
analysis was made of the striking vehicles from the same collisions. In a second analysis,
the contrast group was made of rear-struck vehicles engaged in maneuvers that were not
turn-signal relevant.
Results and Discussion
The first analysis found a reduction of between 3 and 28 percent in the odds of
being the struck (versus striking) vehicle in a turn-signal-relevant maneuver when the
v
vehicle was equipped with amber versus red turn signals. In addition, an effect of rear
signal light source was also observed such that LED turn signals appeared to reduce crash
odds between 33 and 92 percent. Although interesting, the latter result is based on a
single vehicle model equipped with LED turn signals and cannot be readily generalized
to all vehicles equipped with LED turn signals.
The second analysis found no association between turn signal color and the odds
of being struck in a turn-signal-relevant maneuver, although turn-signal reflector optics
(versus lens optics) were found to reduce the odds by between 5 and 51 percent.
However, similar to the previously described LED result, this result is largely based on a
few vehicle models equipped with turn signals with reflector optics, and generalization to
all such lamps would be premature.
The differences between the two analyses indicate that selection of a contrast
group can influence the effects observed. Because the contrast group in the first analysis
is better insulated from the influence of rear signals, these results may be more accurate
than those from the second analysis, implying that amber turn signals may be associated
with lower odds of rear-end collision during turning, merging, or lane change maneuvers.
Conclusions
Although the analysis suggests that there may be a safety benefit associated with
amber turn signals, it is unclear that turn signal color itself is completely responsible for
the benefit. It is important to recognize that color is likely to be confounded with other
factors that could also contribute to this effect. For example, although separation of
functions was partly controlled for in this study, amber turn signals are usually separated
from red stop and tail lamps. Also, requirements for the minimum and maximum
candlepower of amber turn signals are 1.6 to 2.5 times greater than red turn signals.
Further investigation of these other factors seems warranted before drawing the strong
conclusion that turn signal color, by itself, is responsible for the observed differences.
vi
Part I: The Association of Turn Signal Characteristics With
Rear-End Collisions
Overview
As vehicle turn signals evolved and became standardized over the last 100 years,
differences arose in the European and United States standards (Moore & Rumar, 1999).
One particular difference is that European standards, governed by the Economic
Commission for Europe (ECE), require rear turn signals to be exclusively amber
(yellow). In the United States, Federal Motor Vehicle Safety Standard (FMVSS) 108
allows rear turn signals to be either amber or red. Thus, in Europe all rear turn signals
are consistently colored amber, while in the United States, they are colored either amber
or red.
One reason for the difference, ironically, stems from subjective tests of amber
turn signals and recommendations made in the United States by the Vehicle Lighting
Committee (VLC) of the Automobile Manufacturer’s Association to the European
organization of lighting experts, the Group de Travail-Bruxelles (GTB), at a meeting in
1960. The GTB, in turn, recommended the exclusive use of rear amber turn signals to the
Europeans, and in 1967 the ECE made it a requirement. Meanwhile, U.S. manufacturers
rejected their use based on unproven cost benefits (Maurer, 1980).
Presumably, the VLC demonstrations suggested that rear amber signals were
more conspicuous than red, but detailed accounts of the demonstrations are not available.
Based on human factors design principles, a case can easily be made that exclusive use of
an amber rear turn signal would be preferred to a condition in which a turn signal could
be either red or amber. At the perceptual level, an amber turn signal would be more
easily distinguished from other red-colored rear lighting by the difference in color.
Moreover, current lamp technology also necessitates the use of separate lamp
compartments for differently colored lamps, ensuring that an amber turn signal is likely
to be spatially offset from the red tail and brake lamps. Physical lamp separation also
results in likely differences in the contrast between the off and on states of a lamp—there
is likely to be less contrast between the off and on states of red turn signals that share
compartments with red tail or brake lamps than for amber lamps in separate
compartments. Photometric differences between amber and red signal intensity might
also exist—FMVSS 108, for example, stipulates higher minimum and maximum
intensities for yellow (amber) turn signals than for red (see Table 1). Actual intensity
differences in the vehicle fleet could affect the relative conspicuity of each lamp
(although one reason for the intensity difference was to offset the greater conspicuity of
red signals). Unfortunately, there are no surveys available that accurately quantify such
differences, but it is likely that differences exist. Finally, consistency in the meaning of
the signal’s color across the vehicle fleet could also speed recognition. Europeans, for
example, might recognize rear-turn signals more quickly than Americans because they
may benefit from the redundancy in color-coding the turn function (i.e., amber means
turn, red means stop); in contrast, American drivers must recognize that a red signal
could indicate a turning or braking vehicle and base their recognition on other signal cues
(e.g., flashing lamps, asymmetric lamp illumination, non-energized center high-mounted
stop lamp).
1
Table 1.
Minimum and maximum candlepower values stipulated in FMVSS 108.
Lighted
sections
Lamp
1
2
3
Red turn signal
80/300
95/360
110/420
Yellow (amber) turn signal
130/750
150/900
175/1050
There have been a few laboratory studies examining the potential consequence of
allowing two differently colored turn signals. Luoma, Flannagan, Sivak, Aoki, and
Traube (1995) found that reaction time to a braking signal is shorter in the presence of
amber turn signals compared to red. However, because this study blocked trials by signal
color, it is important to recognize that the result may not be directly applicable to the
American driving environment where a mixture of amber and red is always present. In
the study, participants may have adopted a strategy whereby a brake detection response
was solely dependent on the appearance of a red light within the amber turn-signal block.
In the red turn-signal blocks, an alternative strategy that does not rely on color-coding
would be necessary. The result suggests that European drivers might enjoy an advantage
of color-coded function while American drivers do not. However, it does not suggest that
that amber or red signals have a particular advantage in the context of a roadway
environment where mixtures of both colors are present.
Post (1975) examined driver reaction time to respond to turn, brake, and hazard
signals with a variety of lamp configurations. Although reaction time to amber hazard
signals was generally shorter than to red, no difference was found in reaction time to turn
signals or brake signals when a configuration employing amber signals was compared to
one with red. Similarly, Mortimer and Sturgis (1974) found no significant effect of turn
signal color on reaction time, although many of their results suggest shorter reaction
times were observed with amber turn signals. Recognizing that as a red stimulus shifts
toward the visual periphery it appears less saturated and more yellow, Sivak, Flannagan,
Miyokawa, and Traube (2000) questioned the general utility of signal color coding
in normal driving, where objects are presumably first detected in the periphery. They
found that in daylight conditions, color discrimination declined at viewing angles of 10
degrees or more, suggesting that the usefulness of color might be somewhat limited in
hastening detection.
As mentioned earlier, use of amber rear turn signals necessitates separation from
the red stop lamps, resulting in both a spatial offset from the other rear signal lamps, and
a likely boost in contrast between the off and on state of the signal. Relevant to the issue
of lamp separation, a recent study reported by (Luoma, Sivak, & Flannagan, 2006)
examined whether rear-end collisions were influenced by separation of the brake lamp
function from the other rear signal functions. They reasoned that braking activity for
2
vehicles equipped with separated (dedicated) brake lamps would be easier to distinguish
from other signal functions, especially at night when tail lamps are normally energized,
reducing the contrast between the off and on states for combined lamps. The results
suggest that dedicated stop lamps reduce rear-end collisions, although the authors caution
that further examination of the data are warranted.
Beyond laboratory investigations, there has been just one analysis of crash data
that investigated whether any differences in turn signal characteristics might be reflected
in the crash record. Taylor and Ng (1981) examined insurance claim records involving
rear-end collisions. They compared the proportion of turning crashes involving struck
lead vehicles equipped with red turn signals to those equipped with amber turn signals.
To account for exposure differences, the proportions were compared to rear-end
collisions that did not involve turning vehicles. The analysis found no difference in rearend crash rates between vehicles equipped with amber and red turn signals. Although the
study attempted to analyze factors like driver age, gender, vehicle size, model year, light
conditions, and at-fault driver status, the sample size of 1440 vehicles (386 amber; 1053
red) may have been insufficient to observe a clear effect. A power analysis that assumes
a 6-percent difference in proportions of red lamps involved in turning versus non-turning
crashes (with proportionally similar crash distribution) suggests that nearly 3,000 crashes
would be needed to detect such a difference.
All of the vehicles compared in the preceding crash study pre-date the
introduction of center high-mounted stop lamps (CHMSL), introduced in the 1986 model
year. One effect of the introduction of CHMSL was an estimated reduction of rear-end
collisions by about 4.3 percent (Kahane & Hertz, 1998). One might argue that, in the
years prior to the introduction of CHMSL, red rear turn signals might have been more
confusable with rear brake signals and possibly led to more rear-end collisions during
turning. Once the CHMSL was introduced, the difference between a turning and a
braking vehicle may have become clearer to following drivers. It seems that now, there
is perhaps even less chance that a difference between turn signal color might lead to a
rear-end collision.
It is important to keep in mind that the experimental studies cited above do not
present subjects with conditions that are directly comparable to the circumstances in
which a crash occurs. For example, studies that address questions about unique colorcoding of rear signal function (e.g., Luoma, Flannagan, Sivak, Aoki, & Traube, 1995;
Mortimer & Sturgis, 1974) do not resemble the existing U.S. and Canadian crash
environments where there is a mixture of both red and amber turn signals—red could
signal either a turn or braking, a turn could be signaled with either an amber or red
flashing lamp, however amber would mean a turn. While there may be some benefit for
all vehicles to use amber rear turn signals, domestic crash data cannot directly assess this
benefit. Instead, any differences found in crash rates between the two colors are most
likely attributable to differences that make one more conspicuous than another.
Signal characteristics related to crash risk
In the few crash analyses that attempt to address the potential differences between
red and amber turn signals there is little discussion about what mechanisms might lead a
driver to be more likely to strike a vehicle equipped with a red turn signal and less likely
to strike a vehicle equipped with an amber turn signal. Indeed, it is often implied that
3
amber turn signals are more conspicuous than red. Perhaps amber turn signals are
noticed earlier than red turn signals and provide drivers with more time to anticipate the
movements of a forward vehicle.
On the other hand, perhaps drivers mistake a red turn signal for a brake signal.
Unfortunately, this would not explain how confusion between a turn and a brake signal
would lead to a rear-end collision. That is, if a driver sees a turn signal and mistakes it
for a brake signal, it is reasonable to assume that the normal response would be to brake,
perhaps inappropriately. If a driver sees a brake signal, and mistakes it for a turn signal,
it is conceivable that the driver might fail to brake and strike the forward vehicle.
However this failure could easily occur for either amber or red turn signal equipped
vehicles. Unless there is a scenario in which (from the following driver’s perspective) a
turn signal indicates that a rapid deceleration is imminent, and in which a brake signal
does not indicate an imminent deceleration, it is difficult to see how the confusion of one
signal for another would lead to a rear-end collision.
There are a few scenarios in which something like this is plausible. As vehicles
approach a lane closure on a limited access highway, there may be both braking and
signaling of a merge into an adjacent lane. If a driver following behind another vehicle in
an adjacent lane, mistakes the forward vehicle’s signal as braking instead of merging, a
rear-end collision could result when the forward vehicle encroaches into the following
driver’s lane (shown in Figure 1). In another scenario, it is plausible that as a vehicle
transitions from a high speed to a lower speed, braking may occur over an extended
duration in order to decelerate in a smooth fashion. From the perspective of a following
vehicle, in this context the forward vehicle’s brake lamp does not signal an imminent
deceleration. However, if a turn signal is energized, a following vehicle may well
anticipate that a stronger deceleration is about to happen (in order to make a turn at a
comfortable speed). In this scenario, a failure to detect the turn signal may impede the
following driver’s ability to anticipate the deceleration, resulting in a rear-end collision
(shown in Figure 2).
4
Figure 1. A crash scenario in which a turn signal is mistaken for a brake signal and leads
to a rear-end collision.
5
Right turn signal; rapid
deceleration to make turn.
Following driver mistook
turn signal for continued
braking.
Slow deceleration; 45 mph; stop
lamps on.
Slow deceleration from 55 mph;
stop lamps on.
Looks like driver is
slowing down for
something.
Figure 2. Crash scenario in which a forward vehicle initiates a gentle deceleration before
turning. In this example, the turn signal could help a following driver identify where the
forward vehicle will initiate a sharp deceleration in order to execute a turn.
General analysis approach
Throughout this report, data were analyzed using stepwise logistic regression
procedures (the LOGISTIC procedure, in SAS). In this approach, the odds of a crash are
modeled using various characteristics of a driver-vehicle configuration (independent
factors). In this analysis, there is particular interest in turn-signal lamp characteristics,
but, as will be seen, other factors may also influence the odds of a crash. The stepwise
analysis proceeds by adding factors to a regression model one-by-one. Each factor is
drawn from the pool of candidate factors until no factors remain in the pool that can
improve the predictive power of the model. The resulting model contains only those
factors that significantly improve prediction of the dependent variable.
6
For this analysis, the dependent variable is the odds of a crash likely to be
associated to a driver’s response to a rear turn signal—that is, a relevant crash. To obtain
the odds of a relevant crash, non-relevant crashes are also required. The resulting odds
are given by:
⎛ frequency of relevant crashes ⎞
⎟⎟
odds = ⎜⎜
⎝ frequency of non - relevant crashes ⎠
Logistic regressions actually model the natural log of the odds of an event as a
function of multiple independent factors, providing estimates of the influence each factor
has on the resulting odds. The “event” in this analysis is the odds of a rear-end collision
into a vehicle that is either turning or changing lanes (and, presumably, influenced by
turn signal characteristics). The question addressed in the analysis is whether these odds
are smaller for lead vehicles equipped with amber turn signals than they are for lead
vehicles equipped with red turn signals.
As mentioned above, calculation of odds also requires counting the target
vehicle’s involvement in non-relevant crashes. Non-relevant crashes are crashes that are
not affected by the variable of interest (i.e., turn signal color) that can serve as a kind of
measure of general vehicle exposure. The crash analyses cited earlier determined nonrelevant crash frequency using rear-end crashes between vehicles in which turning or lane
change maneuvers were not involved (shown in Figure 3). More importantly, these nonrelevant crashes were classified as either red or amber, based on the rear signal
configuration of the lead (i.e., struck) vehicle. One critique of using the lead vehicle is
that it is possible that a rear signal configuration that reduces rear-end collisions in lane
change, merge, and turning scenarios might also produce side-benefits that reduce rearend collisions in other circumstances (i.e., non-relevant) as well. If lamp characteristics
influence both relevant and non-relevant crash characteristics in the same way, the
resulting odds ratio (relevant/non-relevant) may not show any influence. In this report,
an alternative calculation of the non-relevant crash is provided in which non-relevant
crashes are based on the rear-signal configuration of the striking vehicle. Presumably,
the drivers in striking vehicles are not influenced by the rear signal characteristics of their
own vehicles.
7
Figure 3. Crash scenarios in which the turn signal characteristics are relevant are
illustrated in A; crash scenarios in which turn signal characteristics are non-relevant are
illustrated in B.
Relevant vehicle:
struck while
turning.
Contrast vehicle:
striking a turning
vehicle.
Figure 4. As an alternative to B in Figure 3, odds ratios were based on the signal
characteristics of the struck versus striking vehicle in rear-end collisions involving
turning vehicles.
8
Method
Vehicle Selection and Rear Signal Database Development
As described in the analysis approach, an important component of the logistic
regression analysis is the association of rear signal lamp characteristics to the odds of a
crash. In the ideal situation, a complete set of rear signal characteristics would be
determined for each vehicle involved in a rear-end collision and factored into the
regression. Unfortunately, there are no available reference sources that describe a
vehicle’s rear signal configuration throughout its production history.
Alternative vehicle selection strategies were considered with the aim of producing
a sufficiently large vehicle sample to ensure sufficient power in the crash analysis so that
even modest influences of signal lamp characteristics could be determined. In prior crash
analyses (for example, Luoma et al., 2006; Taylor & Ng, 1981), researchers selected
companion vehicle pairs with known differences in signal color (or other attributes) and
made direct comparisons between them. These vehicle selections appear to have been
made in an ad hoc fashion—no reference sources exist that provide sufficient description
of vehicle rear signal configurations to support selection. Without such a reference,
determination of the rear signal characteristics of a vehicle requires individually
researching the signal characteristics of each vehicle model that might be included in a
crash analysis. Because rear signal characteristics are also an element of vehicle styling,
they change as a vehicle’s body style evolves over time. Thus it is necessary to trace the
rear signal production history of each vehicle. To compile this information, it was
necessary to gather data from several sources. These included dealerships, parts catalogs,
promotional brochures, owners groups, and contacts from within the auto companies.
Since it was not feasible to conduct an exhaustive survey of all vehicle makes and
models, limits were placed on the models and model-years included in the survey.
Model years were selected to span the years 1990 to 2005, and models were
selected to include only the top 50 models found in an initial survey of five State crash
datasets from the calendar year (CY) 2003. Each dataset included the vehicle
identification number (VIN) for each vehicle involved in a collision. The VIN was used
to determine the make, series name (model), and model year of each vehicle by decoding
it using VINDICATOR 2005 software, developed by the Highway Loss Data Institute.
Differences were found among the selected State crash datasets in the proportion of
involved vehicles that were successfully decoded (shown in Table 2). The North
Carolina datasets have the fewest decoding errors (around 8%); while Florida, Maryland,
and New Jersey have the most (around 40%).
9
Table 2.
Proportion of the total number of vehicles in each State dataset that could not be
successfully decoded by VINDICATOR software.
State
Florida
Kentucky
Maryland
Michigan
New Jersey
New Jersey
North Carolina
North Carolina
Pennsylvania
Year
2003
2003
2003
2004
2002
2003
2002
2003
2003
Total
Crashes
243,294
154,075
109,098
374,446
324,053
319,980
285,135
270,224
139,402
Total
Vehicles
478,182
278,531
202,808
637,539
606,502
609,439
448,162
470,561
230,413
Total
Decoding
Errors
192,236
48,316
78,200
194,044
252,981
246,656
33,702
41,197
25,742
Percent
40.2%
17.4%
38.6%
30.4%
41.7%
40.5%
7.5%
8.8%
11.2%
State crash datasets were selected based on the volume of crashes reported,
geographical distribution, inclusion of VIN data, and use of coding conventions that
would allow sufficient distinction of crash scenario details to enable determination of
relevant and non-relevant crashes, vehicle roles (striking/struck), and other factors
detailed in the crash section of this report. The resulting compilation of crash frequency
of vehicle models is shown in Table 3, sorted in descending order by frequency.
10
Table 3.
Counts of the most frequently occurring vehicle models in five State crash datasets from
CY 2003 collapsed over model years.
KY
4,127
7,981
4,677
5,882
1,698
2,603
2,852
3,106
1,860
3,655
2,310
1,733
1,711
2,722
2,569
1,182
3,510
2,614
2,550
3,730
1,545
1,886
1,648
2,170
2,644
1,556
2,868
1,218
1,467
1,757
740
1,667
1,808
1,280
1,356
1,897
697
1,739
1,291
1,289
910
2,316
1,818
1,406
1,448
388
1,521
906
825
1,801
1,078
MD
6,251
6,980
5,515
3,453
2,514
2,898
2,505
3,521
2,218
2,566
1,988
2,174
1,826
1,255
2,610
1,894
1,090
1,112
1,022
834
1,336
1,556
1,073
909
1,174
1,626
1,343
1,517
999
819
1,076
1,244
1,581
1,208
934
877
1,156
805
777
477
1,117
430
540
566
732
187
715
628
896
503
861
NC
17,864
12,850
7,311
9,234
6,185
5,975
5,291
5,213
4,771
3,695
4,612
4,821
4,071
5,530
3,721
4,173
5,069
4,727
4,032
4,569
3,666
3,461
3,389
3,973
3,373
2,940
2,950
4,381
2,685
3,245
3,581
2,906
2,776
2,554
2,264
3,302
3,121
3,020
2,154
3,215
2,346
3,557
3,187
3,164
2,471
3,275
2,354
2,372
1,215
2,116
2,319
NJ
12,027
11,053
7,817
7,340
6,567
5,058
5,972
2,738
5,242
2,594
5,835
3,650
3,717
1,925
2,161
4,179
1,460
1,772
2,133
1,088
2,865
2,053
3,464
2,141
1,797
2,255
2,240
2,119
3,428
2,226
3,209
2,994
2,227
2,491
2,648
1,910
2,662
1,821
2,605
978
1,967
541
914
805
1,334
268
1,416
1,589
2,423
1,240
1,237
PA
3,793
3,423
2,506
4,462
2,189
1,859
3,240
1,474
1,678
3,788
2,595
1,231
2,184
1,186
2,586
1,443
1,093
1,020
2,062
1,011
1,583
1,810
1,602
1,481
1,829
1,997
2,150
866
732
1,673
855
2,000
1,371
1,606
867
796
955
1,428
1,304
308
1,185
515
668
409
1,395
87
952
767
1,210
1,022
1,114
Make
HONDA
TOYOTA
TOYOTA
FORD
HONDA
NISSAN
FORD TRUCK
FORD
NISSAN
CHEVROLET
JEEP
NISSAN
SATURN
FORD
CHEVROLET
HONDA
FORD TRUCK
FORD TRUCK
PONTIAC
CHEVY TRUCK
DODGE TRUCK
FORD
FORD TRUCK
BUICK
CHEVROLET
DODGE
CHEVY TRUCK
MAZDA
LINCOLN
BUICK
HONDA
JEEP
CHEVROLET
DODGE TRUCK
MERCURY
CADILLAC
MITSUBISHI
DODGE
MERCURY
FORD TRUCK
FORD
CHEVY TRUCK
CHEVROLET
FORD TRUCK
FORD
FORD TRUCK
FORD
MITSUBISHI
HYUNDAI
PONTIAC
DODGE
11
Series
ACCORD 4D
CAMRY 4D 2WD
COROLLA SEDAN 2WD
TAURUS 4D
CIVIC 4D
ALTIMA 4D
EXPLORER 4D 4X4
LTD/CROWN VICTORIA 4D
810/MAXIMA SEDAN
CAVALIER 2D
GRAND CHEROKEE 4D 4X4
SENTRA 4D
SL 4D
MUSTANG 2D
CAVALIER 4D
CIVIC 2D COUPE
RANGER PICKUP 4X2
F150 PICKUP 4X2
GRAND AM 4D
S10 PICKUP 4X2
GRAND CARAVAN 2WD
ESCORT 4D
WINDSTAR VAN
LESABRE 4D
LUMINA 4D
NEON 4D
T10 BLAZER 4D 4X4
626 SEDAN
TOWN CAR/CONT. 4D
CENTURY 4D
ACCORD 2D
CHEROKEE 4D 4X4
MALIBU 4D
CARAVAN VAN 2WD
MARQUIS/G. MARQ. 4D
DEVILLE 4D FWD
GALANT 4D 2WD
INTREPID 4D
SABLE 4D
F150 SUPER PU 4X2
FOCUS 4D
10/1500 PU 1/2T
CAMARO 2D
RANGER SUPER PU 4X2
ESCORT 2D
EXPLORER 4D 4X2
CONTOUR 4D
ECLIPSE 2D 2WD
ELANTRA 4D
GRAND PRIX 4D
STRATUS 4D
Totals
51,958
51,173
35,649
35,206
23,096
22,477
20,770
19,954
18,392
18,354
18,191
16,895
16,504
16,187
16,095
15,827
15,380
14,457
13,484
13,285
12,985
12,952
12,823
12,646
12,464
12,181
12,029
11,950
11,822
11,550
11,481
11,398
11,389
11,313
11,282
10,870
10,730
10,266
9,536
9,439
9,094
8,852
8,826
8,638
8,631
8,024
7,936
7,871
7,853
7,720
7,669
Using this list as a basis, data on each vehicle’s rear signal lighting configuration were
compiled in a supplemental database that could be cross-referenced using decoded VIN
information from the crash tables. For each vehicle make and series name, the rear signal
lamp configuration was described using the following principal data fields (summarized
in Table 4:
Start/End Model Years (1990-2005). Within a model’s lifetime, styling changes
occur that frequently result in changes in a rear signal lamp’s configuration. This
data field identifies the spanning years for a given lamp configuration.
• Turn Signal Color (Red, Amber). This field identifies the color of the
energized turn signal.
• Turn Signal Lens Color (Clear/Tinted). Clear signal lenses admit more light
than tinted lenses potentially affecting daytime visibility.
• Turn Signal Source (Tungsten/LED). Although most vehicles are equipped
with tungsten-filament bulbs, some newer vehicles are beginning to appear
equipped with LED turn signal sources. There is some evidence that the rapid
rise time of an LED lamp enhances a driver’s response (Sivak, Flannagan, Sato,
Traube, & Aoki, 1994).
• Turn Signal Optics (Lens/Reflector). A signal lamp can distribute light using a
faceted lens or using a smooth lens and faceted reflector combination.
• Rear Signal Separation Code. This code identified how the stop, turn, and tail
(presence) lamps were distributed among the separate lamp compartments on the
rear of the vehicle. In the code, separate compartments were indicated by comma
separations; combined functions within a compartment were indicated using
slashes to separate the codes.
The lens color, source, and optics of brake and tail lamps were also identified
during this process, although these attributes were not specifically investigated in the
turn-signal analyses. Additional fields were used to flag exceptions, record notes,
identify vehicles equipped with rear fog lamps, and reference photographs of sample
lamps. An example data-entry screen is shown in Figure 5.
•
12
Table 4.
Definition of supplemental database fields used to describe rear signal configuration.
Grouping
Field Name
Values
Description
Model Year
Span
Start Year
End Year
4-digit year
4-digit year > start year
Many vehicles change signal lamp
characteristics along with other
styling changes.
Turn Signal
Color
Lens Color
Source
Optics
Red, Amber
Clear, Tinted
Tungsten, LED
Lens, Reflector
Brake Signal
Lens Color
Source
Optics
Clear, Tinted
Tungsten, LED
Lens, Reflector
Tail Signal
Lens Color
Source
Optics
Clear, Tinted
Tungsten, LED
Lens, Reflector
Rear Signal
Separation
Configuration
e.g., S/T/TS, T
S,T,TS
S/T/TS
S/T, TS
S/T, T, TS
S/T, T/TS
S/T/TS, T
S/TS, T
S/T, S/T, TS
S/T/TS, S/T
13
In the example, S = Stop Lamp, T =
tail lamp (presence), and TS = Turn
Signal. Separate lamp compartments
are separated by commas. The
example code identifies a rear-signal
in which one compartment has
combined stop, tail, and turn signal
functions, and a separate
compartment containing only the tail
lamp.
All lamp functions separate.
All lamp functions combined.
Stop and Tail combined, Turn Signal
separate.
Stop and Tail combined, Tail and
Turn Signal separate.
Tail combined with separated Stop
and Turn Signal.
All functions combined in one
compartment, separate compartment
for Tail.
Stop and Turn Signal combined, Tail
lamp separate.
Two compartments with Stop and
Tail combined; Turn Signal separate.
Stop, Tail, and Turn Signal
combined, Stop and Tail combined.
Figure 5. Microsoft Access data entry form used to compile lamp characteristics of rear
signal lamps.
14
Crash Scenario Selection and Data Processing
Crash records were initially obtained for Florida, Kentucky, Maryland, North
Carolina, New Jersey, and Pennsylvania for CY2003 and Michigan CY2004. The
analysis was later expanded to include data from CY2002 for New Jersey and North
Carolina based on a power analysis that estimated the number of cases required to
observe risk differences as small as 9 percent.
The VIN data from each State dataset was initially linked to the VINDICATOR
2005 dataset to produce a standard labeling for vehicle makes and series names. Each of
the resulting State datasets was then examined to determine the extent of data loss that
occurred as a consequence of VIN-decoding errors. Datasets varied in the accuracy with
which the VIN is transcribed. Common reasons for a decoding error include: the
recorded VIN is missing, is invalid, or has failed to match a series/model key in the
VINDICATOR dataset. As shown in Table 2, the Florida and New Jersey datasets had
the largest proportion of vehicle identification errors (about 40%); the North Carolina
datasets had the fewest identification errors (8 to 9%).
Relevant and Non-Relevant Crashes. As described earlier, the key dependent
measure in the logistic regression is an odds ratio that relates the odds of a relevant crash
with respect to a non-relevant crash. In this analysis, relevant crashes used the following
selection criteria:
• The crash was identified as a rear-end collision.
• Only two vehicles were involved in the collision. This was done to simplify
the crash configuration so that each vehicle’s role could be
unambiguously determined.
• The lead (i.e., struck) vehicle was either described as changing lanes, merging,
or turning. It is plausible to assume that the lead vehicle may have been using a
turn signal prior to the maneuver, although it is clear that drivers often omit use of
turn signals.
• One vehicle’s impact location was identified as in the rear (struck); and the other
vehicle’s impact location was identified as in the front (striking). This restriction
served to exclude crashes where both crash participants are identified as occurring
in the same location. For this analysis, the characteristics of the lead vehicle’s
rear signals served as the basis of the relevant dataset.
In the logistic regression, the non-relevant dataset serves as a kind of exposure
control helping to normalize the relevant crash data to the varying concentrations of
vehicles on the roadway. Selection of a suitable basis for this control sample can
introduce artifacts into an analysis that can obscure or even bias an outcome. For
example, suppose the relationship between driver age (young/old) and risk of
involvement in rear-end collisions were evaluated using another collision type—for
example, single-vehicle road departure (SVRD)—as an exposure control. Systematic
differences in the SVRD crash sample—especially related to a factor of interest, driver
age—could lead to an erroneous conclusion.
In this analysis, two non-relevant datasets were developed. The first dataset used
the same crash criteria employed in the relevant crash selections with the exception that
the rear-signal characteristics were determined for the striking vehicle. It is thus assumed
15
that the rear signal characteristics of a striking vehicle play little role in these crashes and
can serve as a reasonable measure of exposure. While there may be significant
demographic differences between striking and struck drivers in this scenario, it
is assumed that such differences are not systematically related to a driver’s signal
lamp characteristics.
The second non-relevant dataset was developed based on the analyses reported by
Taylor and Ng (1981). In their analyses, non-relevant crashes were identified as rear-end
collisions that did not involve turns, merges, or lane change maneuvers. Importantly, the
authors identified the characteristics of the lead (struck) vehicle in these crashes. One
potential difficulty with this approach is that if the turn signal characteristic of interest—
amber versus red—somehow influences the salience of other rear signals, we might find
that other rear-end collision types (i.e., those not necessarily involving turns, merges, or
lane changes) are also affected. In this case, there is a chance that an amber turn signal
might reduce both the relevant and non-relevant crashes. This would diminish the
likelihood of observing an effect. The selection criteria for this dataset, referred to as
non-turning crashes, applied the same selection criteria as described for the Relevant
crash selection except that the lead vehicle maneuver was not described as turning,
merging, or changing lanes.
Table 5 provides a breakdown of the rear-end crash scenarios and their
distribution within each State dataset. For most of the States, non-turning crashes make
up between 13 and 18 percent of all reported crashes; the relevant crashes make up
between 1 and 2 percent of all crashes. Notably, Florida appears to be an outlier with
proportionally less than half of the crash percentages found in the other State datasets. It
is currently unclear what the basis of this difference is.
Table 5.
Frequency of all crashes and rear-end crashes involving two vehicles in which the struck
vehicle was not turning, merging, or changing lanes, and two vehicles in which the struck
vehicle was turning, merging, or changing lanes.
State
Florida
Kentucky
Maryland
Michigan
North Carolina
North Carolina
New Jersey
New Jersey
Pennsylvania
Year
2003
2003
2003
2004
2002
2003
2002
2003
2003
Total
Crashes
243,294
154,075
109,098
374,446
285,135
270,224
324,053
319,980
139,402
Two-Vehicle
Rear-End
Non-Turning
(Non-Relevant
Crashes)
17,991
27,517
17,148
62,433
50,346
47,674
59,017
58,872
18,279
16
Percent
7.4%
17.9%
15.7%
16.7%
17.7%
17.6%
18.2%
18.4%
13.1%
Two-Vehicle
Rear-End Turn,
Merge, Lane
Change
(Relevant
Crashes)
1,287
2,219
1,131
4,552
4,408
4,427
6,243
6,514
1,144
Percent
0.5%
1.4%
1.0%
1.2%
1.6%
1.6%
1.9%
2.0%
0.8%
Once the crash records of the two rear-end scenarios were selected (relevant and
non-relevant), the VIN data of both the striking and struck vehicles in the
turning/merging/lane changing rear-end collisions were matched to vehicles contained in
the Rear Signal Database. Note that the Rear Signal Database contains only the most
frequently occurring 50 vehicles among five CY 2003 State crash datasets (shown in
Table 3). With this restriction, the overall vehicle sample size becomes smaller. The
resulting breakdown by State is shown in the first two data columns of Table 6. The total
vehicle count used in this analysis is approximately 13 times greater than included in the
Taylor and Ng study (1981). The first two data columns of Table 7 shows the same data,
collapsing over States and showing the breakdown by turn signal color and driver role in
the collision.
A similar selection procedure was used to create a dataset comprised of rear-end
crashes not involving turning, merging, or lane change maneuvers. This dataset served as
a second non-relevant crash reference, similar to the striking drivers previously
described. The resulting crash breakdown by state is shown in the third data column of
Table 6, and the crash breakdown by turn signal color is shown in the third column of
Table 7. An overview of the data processing steps is shown in Figure 6.
Table 6.
Crash counts by State and driver role in collision for each rear-end collision type.
Rear-end collisions while turning,
merging, or changing lanes
State
Florida
Kentucky
Maryland
Michigan
North Carolina
New Jersey
Pennsylvania
Total
Struck
(Rear Impact)
361
812
496
1,478
3,163
3,398
382
10,090
Striking
(Front Impact)
285
676
421
1,276
2,756
2,839
336
8,589
Rear-end collisions not involving
turning, merging, or changing lanes
Struck
(Rear Impact)
4,904
10,525
7,047
19,900
36,289
31,098
5,972
115,735
Table 7.
Breakdown of crash frequencies by signal lamp color, role in crash, and for each rear-end
collision type.
Rear-end collisions while turning,
merging, or changing lanes
Rear-end collisions not involving
turning, merging, or changing lanes
Signal
Lamp
Color
Struck
(Rear Impact)
Striking
(Front Impact)
Struck
(Rear Impact)
Amber
4,975
4,417
58,964
Red
5,115
4,172
56,771
Total
10,090
8,589
115,735
17
State
State
State
State
Crash
Crash
Crash
Crash
Datasets
Datasets
Datasets
Datasets
(Vehicle)
(Vehicle)
(VIN
(Vehicle)
codes)
Vindicator
Dataset
Vindicator
processing
Enhanced State
Enhanced State
Enhanced
State
Tables
with
Enhanced
State
Tables
with
Tables
with
Make/Series/MY
Tables
with
Make/Series/MY
Make/Series/MY
Make/Series/MY
Scenario Selection
Rear-end collisions into
turning/lanechange/
merging vehicles
Struck vehicles
Striking vehicles
Rear Signal
Database
(MS Access
database)
Scenario Selection
Rear-end collisions into
Non-turning
Struck vehicles Only
Relevant Crash:
Struck vehicle
Non-Relevant Crash:
Striking vehicle
Turn signal lamp: Amber / Red
Relevant Crash:
Struck vehicle making a turn
Non-Relevant Crash:
Struck vehicle not making a turn
Turn signal lamp: Amber / Red
Regression Model
Regression Model
log (Struck/Striking) =
x
log (Relevant/NonRelevant) =
x
Figure 6. Overview of the data-processing steps that produced the datasets used in the
regression analysis. Processes are drawn as rectangles; datasets are drawn as
parallelograms. Orange datasets were supplied from external sources; green datasets
were generated as part of this project.
18
Logistic Regression Models
Two response variables are defined in separate regression models. The first is the
odds of being the struck vehicle versus the striking vehicle in a rear-end collision
involving a vehicle turning, merging, or changing lanes (a relevant collision). The
second response variable is the odds of being the struck vehicle in a relevant collision (as
before) versus a non-relevant collision—a struck vehicle in a rear-end collision involving
maneuvers that do not include turning, merging, or changing lanes. For each model,
several variables in the datasets were identified as a potential influence on the odds of a
crash. Each of these variables was identified and included in a logistic regression in
which a stepwise selection procedure was employed which evaluated each variable with
respect to its contribution to the predictive power of the model. Each variable was added
to the model only if it produced a significantly better predictive model. As variables
were added to the model, those already contained in the model were reevaluated and
removed if they no longer contributed to the model’s predictive power. Table 8 describes
individual candidate variables included in the stepwise regression analysis and provides a
rationale for their inclusion. Table 9 describes candidate interactions between variables
that were also added to the same regression models. Interactions considered in each
model included driver age with gender, light condition, and State; light conditions with
each of the turn-signal lamp characteristics, and vehicle age with series name.
One reason so many variables were included in this analysis was that, as with any
correlational analysis, many variables are likely to be indirectly related to each other. For
example, younger drivers are more likely to drive older or less expensive vehicles; older
drivers are more likely to drive luxury models. Amber turn signals are always physically
separated from red stop lamps, but red turn signals may or may not be separated. In this
latter example, to assess the importance of stop and turn signal separation, the model will
only evaluate this variable using the red turn signal cases because there are no differences
among the amber turn signals.
19
Table 8.
Main effect variables included in the logistic regression model.
Variable
State
Light Condition
Gender
Driver Age Group
Vehicle Age
Series Name
Body Style
Turn Signal Color
Turn Signal/Stop Signal
Separation
Turn Signal/Tail Lamp
Separation
Turn Signal Separation from
All
Turn Signal Source
Turn Signal Optics
Turn Signal Lens Color
Description
This is the State in which the crash occurred.
Because the conspicuity of signal lighting may interact with lighting conditions,
light condition (light, dark) was included in the model. Conditions were
classified as dark if the reported light conditions were “dark” or “dark with
lights.” Dawn, dusk, and unknown conditions were excluded from the analysis.
Reported gender of the driver.
This is a classification of the driver’s age as either young (<30 years), middle
(30-64 years), and old (>64 years).
This is a continuous variable computed as the calendar year of the crash minus
the model year of the vehicle plus 1 (to avoid negative numbers from new
vehicles with model numbers greater than the calendar year). Vehicle age at
time of crash has been reported to have an inverse relationship to risk of rear end
collision (e.g., Kahane & Hertz, 1998).
This is the model name of the vehicle. It is used to account for factors that may
influence crash risk within select populations of vehicle owners (e.g., Camaro
owners are likely different from Civic owners).
This variable is correlated with Series Name. It groups vehicles into the
following broad body style categories: Luxury, Sports, Utility, 4-Door, 2-Door,
Passenger Van, and Pickup. It was used as an alternate analysis level.
Amber or Red. This is the signal lamp characteristic that identifies the color
of the illuminated turn signal. It is the primary variable of interest in most
prior studies.
Yes / No. This variable identifies whether the turn signal shared a compartment
with a brake signal. It is plausible that turn signal color may be less important
than signal separation and only appears to be important because amber turn
signals are by necessity separate from the stop signal.
Yes / No. This variable identifies whether the turn signal shared a compartment
with a tail lamp. The rationale for this variable is similar to the rationale
presented for the preceding variable.
Yes / No. This variable identifies whether a turn signal share any compartments
with other lamps. If either of the two preceding variables is ‘yes,’ this variable
will be ‘yes.’
LED / Tungsten. This variable identifies the illumination source of the lamp.
Differences in the rise time of different signal sources could influence the
conspicuity of a signal lamp.
Lens / Reflector. This variable identifies whether the light distribution from a
signal lamp is controlled by a faceted lens, or by a silver reflector. A signal
lamp that employs reflector optics passes light through a smooth lens.
Clear / Tinted. This variable identifies whether the turn signal lens is clear or
tinted. In the case of a clear lens, the lamp color is produced by a tinted bulb. In
daylight, the non-energized clear-lens lamp appears silver and can reflect
ambient sunlight. This can reduce the contrast between off and on states.
20
Table 9
Interaction terms included in the logistic regression model.
Interactions
Between
Driver Age Group
Secondary variables
Description
Gender
Young male drivers are often identified as an
especially aggressive group. The rationale for
including this interaction is that crash risk may be
affected by gender differences more among younger
drivers than older drivers.
Crash risk may interact with driver age and light
conditions if, for example, older drivers
disproportionately avoid driving in darkness.
Driver age distribution among states may not be
homogeneous. For example, there may be an
observed higher crash risk among older drivers in
Florida than in New Jersey.
Light Condition
State
Light Condition
Turn Signal Color
Turn Signal/Stop Signal Separation
Turn Signal/Tail Lamp Separation
Turn Signal Separation from All
Turn Signal Source
Turn Signal Optics
Turn Signal Lens Color
The rationale for modeling the interaction between
light conditions and each of the lamp characteristics
is that it is plausible that the influence of any lamp
attribute on crash odds could differ under different
ambient light conditions.
Vehicle Age
Series Name
The rationale for this interaction is that as a vehicle
ages, changes in vehicle function and ownership
demographics may influence crash odds.
21
Results
Analysis 1: Log Odds of Struck/Striking Role
The results of each logistic regression will be presented in a tabular form in which
the selected effects will be presented along with parameter estimates of each variable. In
the interest of clarity, estimates are not reported for variables in which many levels are
identified—e.g., vehicle series name (50 levels), vehicle body style (8 levels), and State
(7 levels). Such variables have been included in this analysis to effectively account for
the influence these factors have on the resulting odds ratio so that the effects of interest—
turn signal lamp characteristics—can be clearly observed. The results of the analysis are
shown in Table 10. The part of the regression model related to the lighting configuration
fits into the regression as follows:
Predicted logit of Struck =
0.0496
+(0.2976)(FEMALE)
+(.5962)(MIDDLE) or (0.6733)(OLD)
+(0.1105)(DARK)
+(0.0475)(Middle-aged and Female) or (-0.3399)(Old and Female)
+(-0.0434)(Vehicle Age)
+(0.3308)(Accord 4D) or (-0.2248)(Altima 4D) or
(β estimate for Series Name)(presence of Series Name)
+(-0.0259)(VehicleAge)(Accord 4D) or (0.0370)(Vehicle Age)(Altima 4D) or
(β estimate for Series Name x Age interaction)( Vehicle Age)( Series Name)
+(-0.1786)(AMBER)
+(-1.4431)(LED)
Parameter
(Intercept)
(Gender)
(Age)
(Light Condition)
(Age x Gender)
(Vehicle Age)
(Series Name)
(Vehicle Age x
Series Name)
(Signal Color)
(Signal Source)
Thus it appears that amber rear turn signals are associated with a smaller odds
of a being struck in a rear-end turning crash than red turn signals; likewise it also appears
the LED-based turn signals are associated with a even greater reduction in odds of being
struck compared with the odds found with tungsten light sources. The 95 percent
confidence interval on the odds ratio associated with turn signal color is 0.72 to 0.97.
Reinterpreted as the estimated percent crash reduction effect of amber versus red
(reported in previous analyses), this is equivalent to an estimated reduction of between
3 and 28 percent.
The observed effect of turn signal light source is substantially larger than
observed for turn signal color. The 95 percent confidence interval on the odds ratio
associated with source is 0.083 and 0.673. Interpreted as an estimated percent crash
reduction associated with using LED versus tungsten turn signals, this is equivalent to an
estimated reduction of between 33 and 92 percent. While the size of the effect appears to
be dramatically large, the confidence interval is also wide, suggesting that the result is
based on a small portion of the sample data. On further examination of the sample of
vehicles contributing to this analysis, it was found that virtually all the samples of LED
turn signals were from the 2000-2005 Cadillac DeVille. Essentially, the analysis
compared the rear-end collision odds of the 2000-2005 Cadillac DeVille (equipped with
LED turn signals) to the same odds for 1991 to 1999 version of the DeVille (equipped
with tungsten turn signals). Despite the general trend for older model vehicles to decline
22
in their rate of being struck in rear-end collisions, it appears that the newer model
DeVilles, equipped with LEDs, buck this trend. Given that the LED effect is based on
the implementation in one vehicle model line, it would be inappropriate to generalize this
result to all LED turn signal implementations.
Also found was an association between gender and crash odds—female drivers
have a greater odds of being struck than male drivers (conversely, male drivers have a
greater odds of playing a striking role). Similarly, older and middle aged drivers have
greater odds of being struck than younger drivers. There is also an interaction between
age and gender—middle-aged female drivers demonstrated an especially greater risk of
being struck (or conversely, they demonstrate an especially low risk of striking).
23
Table 10
Logistic regression analysis of the odds of struck role in a rear-end collision involving a
lead vehicle turning, merging, or changing lanes. Series name and series name by vehicle
age interactions were omitted from the table.
Predictor
Constant (intercept)
Gender (Male = 0, Female = 1)
Age Group (Young = 0)
Middle
Old
Light Condition (Light = 0, Dark = 1)
Gender x Age Group (Young, Male = 0)
Middle, Female
Old, Female
Vehicle Age
Turn Signals
Color (Red = 0, Amber = 1)
Source (Tungsten = 0, Led = 1)
eβ
Odds
Ratio
SE β
Wald’s
χ2
df
0.0496
0.2976
0.34
0.05
0.02
34.77
1
1
0.88
< .0001
1.05
1.35
0.5962
0.6733
0.1105
0.05
0.08
0.04
167.77
70.90
7.25
1
1
1
< .0001
< .0001
0.01
1.82
1.96
1.12
0.0475
-0.3399
-0.0435
0.07
0.12
0.05
0.51
8.13
0.83
1
1
1
0.48
0.00
0.36
1.05
0.71
0.96
-0.1786
-1.4431
0.08
0.53
5.61
7.30
1
1
0.02
0.01
0.84
0.24
β
Model evaluation
Likelihood ratio test
Score test
Wald test
Goodness of fit test
Hosmer and Lemeshow
24
χ2
df
792.12
778.93
749.78
117
117
117
4.25
8
p
p
< .0001
< .0001
< .0001
0.834
Beta (Estimate)
1.00
0.90
Male
0.80
Female
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
Young
Middle
Old
Age of Driver
Figure 7. Interaction effect between gender and driver age. The estimate reflects the sum
of the estimates for each condition; the larger the estimate, the greater the odds of being
the struck driver in a rear-end collision; the lower the estimate, the greater to odds of
being the striking driver.
The Influence of Other Turn Signal Characteristics. While an effect of lamp
color was observed in the preceding analysis, this does not necessarily preclude the
influence of lamp separation on crash odds, although it suggests that lamp color is a
better predictor of the observed odds ratio. If lamp color is excluded from the analysis,
then separation of the turn signal from the tail lamp becomes a significant predictor,
albeit weaker (see Table 11).
The color/separation issue was further explored in a separate analysis of the effect
of turn signal separation among vehicles equipped with red turn signals; and in an
analysis of the effect of color among vehicles equipped with separated signal lamps.
Lamp separation was a poor predictor of crash odds among vehicles equipped with red
turn signals (Wald χ2 = 0.93, df=1, p = .33); and turn-signal color was a poor predictor
among vehicles equipped with separated lamps (Wald χ2 = 2.07, df=1, p = .15). The
odds ratio estimates, although unreliable, suggest that color might be more influential on
the crash odds than lamp separation—the observed reduction for amber is about
18 percent, for separated lamps it is 11 percent.
25
Table 11
Logistic regression analysis of the odds of struck role in a rear-end collision involving a
lead vehicle turning, merging, or changing lanes, excluding lamp color as a predictor.
Note that turn signal separation, a factor correlated with lamp color, becomes a predictor
of the odds ratio. (As in Table 10, series name and series name by vehicle age
interactions are omitted from this table).
Predictor
Constant (intercept)
Gender (Male = 0, Female = 1)
Age Group (Young = 0)
Middle
Old
Light Condition (Light = 0, Dark = 1)
Gender x Age Group (Young, Male = 0)
Middle, Female
Old, Female
Vehicle Age
Turn Signals Separation
Tail Lamp (No = 0, Yes = 1)
Source (Tungsten = 0, Led = 1)
eβ
Odds
Ratio
β
SE β
Wald’s
χ2
df
0.0696
0.2980
0.34
0.05
0.04
34.85
1
1
0.84
< .0001
1.07
1.35
0.5951
0.6733
0.1107
0.05
0.08
0.04
167.24
70.90
7.28
1
1
1
< .0001
< .0001
0.01
1.81
1.96
1.12
0.0491
-0.3402
-0.0479
0.07
0.12
0.05
0.54
8.14
1.06
1
1
1
0.46
0.00
0.32
1.05
0.71
0.95
-0.1480
-1.4431
0.08
0.53
3.89
7.30
1
1
0.05
0.01
0.86
0.24
Model evaluation
Likelihood ratio test
Score test
Wald test
Goodness of fit test
Hosmer and Lemeshow
χ2
df
790.41
777.20
748.04
117
117
117
3.76
8
p
p
< .0001
< .0001
< .0001
0.878
Table 12
Cases identifying turn signal separation from stop lamp and tail lamps. Note that
combined signals are not possible with amber turn signals.
Tail Separate
From Turn Signal
Lamp Color
Stop Lamp Separate
From Turn Signal
Yes
No
Yes
No
Amber
9,392
0
9,392
0
Red
2,389
6,898
2,633
6,654
26
Analysis 2: Log Odds of Relevant/Non-Relevant Collisions
In this model, instead of employing striking vehicles as the contrast group to
computing odds, vehicles struck in non-turning (non-relevant) collisions were employed.
That is, the odds are the number of vehicles stuck while attempting to turn, merge, or
change lanes (a relevant crash) divided by the number vehicles involved in non-relevant
collisions. As mentioned earlier, this duplicates the analysis approach taken in previously
published reports. For this measure, no relationship between the odds of a relevant crash
and turn signal color or source was found. Perhaps turn-signal color and source influence
many kinds of rear-end collision types—both relevant and non-relevant crashes alike—
effectively obscuring the influence. Indeed, in the case of the 2000-2005 Cadillac
DeVille, both the brake lamp and turn-signal sources are LEDs. It is probable that on any
given vehicle, the characteristics of a rear turn-signal are related to the characteristics of
other rear signals. Use of the relevant/non-relevant odds ratio seems to assume the effect
of turn signal is independent of the effect of brake signal. Consequently, the remaining
analyses in this report will resume use of the logit of the odds of playing the struck role in
a relevant rear-end collision as the response variable for each model.
The odds of a relevant crash were associated with turn signal optics. Vehicles
with reflector optics appear less likely to be involved in relevant collisions with lens
optics. This translates to an approximately 32-percent reduction in the odds of
involvement in relevant crashes with reflector optics compared to lens optics (95% CI = 5
to 51%). It is unclear why this happens, although the sample of vehicles equipped with
reflector optics is small (0.8% of the sample) and seems to be dominated by late-model
(1999-2004) Ford Mustangs, driven by young drivers (58%—18% female, 40% male).
As discussed earlier with respect to the LED finding in the first analysis, it seems
inappropriate to generalize this particular result to all rear turn signals equipped with
reflector optics.
The difference between this analysis and the first one suggests that selection of a
contrast group can influence the observed effects. In this case, an influence of turn signal
color is present in the first analysis and absent in the second. One reason for its absence
in the second analysis might be that both the crash-relevant group (rear-end collisions
into vehicles making turn-signal-relevant maneuvers) and the contrast group (rear-end
collisions into vehicles not making turn-signal relevant maneuvers) are similarly affected
by the rear turn signal configuration. Rear signals are visible for both crash groups, and
this allows the possibility that rear signal characteristics might both enhance turn-signal
conspicuity as well as stop-lamp conspicuity. Thus, if both the relevant crash group and
the contrast group are similarly affected by rear signal characteristics, no effect would be
observed in the logistic regression. On the other hand, the first analysis using the striking
driver’s vehicle as a contrast group seems to remove the possibility that the striking crash
could be influenced by the rear signals—they are completely out of the striking driver’s
direct sight. Consequently, the first analysis is preferred to the second analysis.
27
Table 13.
Logistic regression analysis of the odds that a vehicle is making a tuning, merging, or
lane change maneuver in a rear-end collision. Series name effects (50 total) were omitted
from the table.
Predictor
Constant (intercept)
State (PA = 1)
Florida
Kentucky
Maryland
Michigan
North Carolina
New Jersey
Gender (Male = 0, Female = 1)
Age Group (Young = 0)
Middle
Old
Vehicle Age
Turn Signals
Optics (Lens = 0, Reflector = 1)
eβ
Odds
Ratio
β
SE β
Wald’s
χ2
df
-2.9646
0.102
847.99
1
< .0001
0.052
0.1447
0.1856
0.1288
0.2019
0.336
0.5947
0.1338
0.078
0.066
0.072
0.061
0.058
0.058
0.022
3.41
7.95
3.23
10.95
33.43
107.13
34.45
1
1
1
1
1
1
1
0.065
0.005
0.072
0.001
< .0001
< .0001
< .0001
1.156
1.204
1.137
1.224
1.399
1.812
1.143
-0.0406
0.2129
0.0335
0.025
0.043
0.004
2.71
24.75
94.32
1
1
1
0.0998
< .0001
< .0001
0.96
1.237
1.034
-0.3868
0.17
5.13
1
0.024
0.679
Model evaluation
Likelihood ratio test
Score test
Wald test
Goodness of fit test
Hosmer and Lemeshow
28
p
χ2
df
639.86
641.90
634.22
65
65
65
< .0001
< .0001
< .0001
13.73
8
0.089
p
Analysis 3: Vehicles grouped by body style
In this analysis, vehicle body style was substituted for vehicle series name using
VINDICATOR’s vehicle body style field. The grouping generally collected together
similar vehicles (e.g., passenger vans: Dodge Caravan and Ford Windstar; utility
vehicles: Ford Explorer, Jeep Grand Cherokee, Chevy Blazer), although the 2- and 4door vehicle categories did not distinguish differences in vehicle size (e.g., a Ford Escort
and Pontiac Grand Am were classified together as 2-door vehicles). Vehicle groupings
are shown in Table 14 along with turn signal color breakdowns.
The analysis (shown in Table 15) suggests that the 4-door vehicles and passenger
vans are not particularly different from the 2-door comparison group with respect to the
odds of being the struck vehicle in relevant rear-end collisions. Luxury vehicles appear
to be more likely to play the struck role; while pickup trucks, sports cars, and utility
vehicles are less likely to be the struck vehicle.
The pattern of association between the rear turn-signal characteristics and the
odds of being struck in relevant rear-end collisions is similar to the pattern reported for
Model 1. That is, there is an effect of lamp color such that amber turn signals appear to
be associated with reduced odds of rear-end collision. The 95-percent confidence
interval on the odds ratio associated with turn signal color is 0.87 to 0.99—a weaker
influence than previously seen. The equivalent percent crash reduction associated with
amber lamps would be between 1 and 13 percent.
The observed effect of turn signal light source is, like before, larger than turn
signal color. The 95-percent confidence interval on the odds ratio associated with LED
(versus tungsten) light sources is 0.25 to 0.81. Interpreted as an estimated percent crash
reduction associated with LED turn signal sources, this would be equivalent to an
estimated reduction of between 19 to 75 percent.
Overall, the body style variable produced a less powerful model (indicated by the
higher goodness-of-fit score χ2 in the Hosmer and Lemeshow test), and possibly absorbed
some of the predictive power formerly attributed to turn signal characteristics.
29
Table 14
Breakdown of the vehicle sample by body style and turn signal color.
Body style
2-Door
2-Door Total
4-Door
4-Door Total
Turn Signal Color
Amber
Red
93
127
369
295
120
171
69
108
40
105
136
497
1,136
Series Name
ACCORD 2D
CAVALIER 2D
CIVIC 2D COUPE
ECLIPSE 2D 2WD
ESCORT 2D
GRAND AM 2D
THUNDERBIRD 2D
626 SEDAN
810/MAXIMA SEDAN
ACCORD 4D
ALTIMA 4D
CAMRY 4D 2WD
CAVALIER 4D
CENTURY 4D
CIERA 4D
CIVIC 4D
CONTOUR 4D
COROLLA SEDAN 2WD
ELANTRA 4D
ESCORT 4D
FOCUS 4D
GALANT 4D 2WD
GRAND AM 4D
GRAND PRIX 4D
INTREPID 4D
JETTA SEDAN
LESABRE 4D
LTD/CROWN VICTORIA 4D
LUMINA 4D
MALIBU 4D
MARQUIS/G. MARQ. 4D
NEON 4D
SABLE 4D
SENTRA 4D
SL 4D
STRATUS 4D
TAURUS 4D
30
265
356
1,308
444
997
350
727
164
103
82
83
64
133
130
97
295
4
374
446
132
131
6,685
5
102
372
325
127
155
238
224
252
146
325
182
136
258
161
331
4
185
316
291
3
81
783
5,002
Total
220
369
415
171
177
145
136
1,633
265
361
1,308
546
997
372
325
127
505
238
727
164
327
252
228
408
246
269
130
258
258
331
299
185
316
295
377
446
213
914
11,687
Table 14. (continued)
Breakdown of the vehicle sample by body style and turn signal color.
Body Style
LUXURY
LUXURY Total
Series Name
DEVILLE 4D FWD
TOWN CAR/CONT. 4D
CARAVAN VAN 2WD
GRAND CARAVAN 2WD
WINDSTAR VAN
PASSENGER VAN Total
Turn Signal Color
Amber
Red
237
203
440
PASSENGER VAN
PICKUP
PICKUP Total
SPORTS
SPORTS Total
UTILITY
UTILITY Total
10/1500 PU 1/2T
F150 PICKUP 4X2
F150 SUPER PU 4X2
RANGER PICKUP 4X2
RANGER SUPER PU 4X2
S10 PICKUP 4X2
CAMARO 2D
MUSTANG 2D
CHEROKEE 4D 4X4
EXPLORER 4D 4X2
EXPLORER 4D 4X4
GRAND CHEROKEE 4D 4X4
T10 BLAZER 4D 4X4
Grand Total
31
237
203
440
719
322
397
351
1,070
381
127
252
52
77
70
290
868
129
252
52
310
216
290
1,249
84
84
97
264
361
97
348
445
351
351
2
233
146
332
92
328
642
322
397
Total
1,394
368
744
332
162
634
642
368
2,138
9,392
9,287
18,662
70
306
Table 15
Logistic regression analysis of the odds of struck role in a rear-end collision involving a
lead vehicle turning, merging, or changing lanes.
eβ
Odds
Ratio
β
SE β
Wald’s
χ2
df
p
Constant (intercept)
-0.0702
0.068
1.08
1
0.30
0.93
Gender (Male = 0, Female = 1)
Age Group (Young = 0)
Middle
Old
Light Condition (Light = 0, Dark = 1)
Gender x Age Group (Young, Male = 0)
Middle, Female
Old, Female
Vehicle Age
Vehicle Body style (2-Door = 0)
4-Door
Luxury
Passenger Van
Pickup
Sports (versus 2-Door)
Utility (versus 2-Door)
Turn Signals
Color (Red = 0, Amber = 1)
Source (Tungsten = 0, Led = 1)
0.3074
0.05
38.02
1
< .0001
1.36
0.5781
0.6544
0.1059
0.05
0.08
0.04
164.50
71.87
6.78
1
1
1
< .0001
< .0001
0.009
1.78
1.92
1.11
0.06
-0.3305
-0.0289
0.07
0.12
0.004
0.82
7.83
39.87
1
1
1
0.364
0.005
< 0.0001
1.06
0.72
0.97
-0.0194
0.3803
-0.0436
-0.2629
-0.2701
-0.1968
0.058
0.126
0.086
0.081
0.115
0.071
0.1133
9.0717
0.2593
10.4646
5.5095
7.5903
1
1
1
1
1
1
0.7365
0.0026
0.6106
0.0012
0.0189
0.0059
0.981
1.463
0.957
0.769
0.763
0.821
-0.0734
-0.7979
0.033
0.302
5.0078
6.9642
1
1
0.0252
0.0083
0.929
0.45
χ2
df
604.430
597.663
581.789
15
15
15
< .0001
< .0001
< .0001
7.359
8
0.4984
Predictor
Model evaluation
Likelihood ratio test
Score test
Wald test
Goodness of fit test
Hosmer and Lemeshow
32
p
Analysis 4: Turn-signal color changes within models
This analysis included vehicle models in which amber and red turn signals
appeared on the same model in different model years spanning 1990 to 2005. Models
that have had exclusively amber turn signals or exclusively red turn signals throughout
their model history were excluded from this analysis. For example, in 1996, the styling
on the Ford Taurus was redesigned to use amber turn signals in anticipation of marketing
the vehicle to European customers. The styling was given a minor revision in 1998—the
amber lens was replaced by a red lens. The next major styling change occurred in the
2000 model year. This evolution is shown in Figure 8. An analysis of within-model
signal lamp change may help control factors related to driver demographics that may
differ between models. That is, there are likely to be greater similarities between two
drivers of the same vehicle model with differently colored turn signals, than there are
between two drivers of different models with differently colored turn signals. Use of the
series name variable helps account for such differences, allowing the variables associated
with lamp characteristics to shine through.
As before, the results suggest that the odds of being the struck vehicle in a
relevant rear-end collision are smaller with amber turn signals than with red turn signals.
The odds ratio of amber versus red turn signals is 0.785. Reinterpreting this effect
estimate as a percent crash reduction, the use of amber turn signals may reduce the risk of
rear end collision by about 22 percent. A 95-percent confidence interval places the lower
and upper bound of this estimate between 12 and 30 percent. Driver age and gender were
also strongly associated with the odds of being struck. The odds ratio of female to male
drivers was 1.35; the odds ratio of middle to younger-aged drivers is 1.83, and older to
younger-aged drivers is 1.84. Finally, a relationship was also observed between the
vehicle series name and the odds of being struck, suggesting that vehicle series
contributed some predictive power to the model.
33
1990-1995
1996-1997
1998-1999
2000-2003
Figure 8. Changes in turn signal color within the Ford Taurus from 1990 to 2003.
34
Table 16.
Logistic regression analysis of the odds of struck role in a rear-end collision involving a
lead vehicle turning, merging, or changing lanes examining only vehicle models in which
the turn signal color was changed within the model’s lifespan.
Predictor
Constant (intercept)
Gender (Male = 0, Female = 1)
Age Group (Young = 0)
Middle
Old
Vehicle Series (Escort 4D = 0)
10/1500 PU 1/2T
810/MAXIMA SEDAN
ACCORD 2D
ALTIMA 4D
CIVIC 2D COUPE
CIVIC 4D
ESCORT 2D
EXPLORER 4D 4X2
EXPLORER 4D 4X4
GALANT 4D 2WD
GRAND AM 2D
GRAND AM 4D
GRAND PRIX 4D
INTREPID 4D
LTD/CROWN VICTORIA 4D
MALIBU 4D
MUSTANG 2D
RANGER PICKUP 4X2
RANGER SUPER PU 4X2
SABLE 4D
SENTRA 4D
STRATUS 4D
TAURUS 4D
Turn Signals
Color (Red = 0, Amber = 1)
eβ
Odds
Ratio
SE β
Wald’s
χ2
Df
-0.2058
0.123
2.802
1
0.094
0.814
0.2968
0.050
35.791
1
< .0001
1.350
0.6042
0.6111
0.051
0.102
140.037
36.080
1
1
< .0001
< .0001
1.830
1.842
-0.3948
0.0107
0.1836
0.0729
0.1474
0.3636
-0.0767
-0.0673
-0.4114
0.1635
-0.0957
0.0040
-0.1015
-0.062
-0.336
0.3181
-0.3669
-0.2173
0.1513
0.0379
0.138
0.0079
0.032
0.221
0.166
0.186
0.151
0.159
0.153
0.195
0.203
0.145
0.185
0.210
0.157
0.178
0.175
0.178
0.175
0.165
0.171
0.187
0.172
0.165
0.189
0.137
3.1944
0.0042
0.9747
0.235
0.856
5.6567
0.1551
0.1102
8.0957
0.7843
0.2081
0.0007
0.3244
0.1259
3.5507
3.2936
4.9639
1.6231
0.6526
0.0486
0.6961
0.0017
0.0548
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0.0739
0.9484
0.3235
0.6279
0.3548
0.0174
0.6937
0.7399
0.0044
0.3758
0.6483
0.9795
0.5689
0.7228
0.0595
0.0695
0.0259
0.2027
0.4192
0.8256
0.4041
0.9668
0.8149
0.674
1.011
1.202
1.076
1.159
1.439
0.926
0.935
0.663
1.178
0.909
1.004
0.904
0.94
0.715
1.375
0.693
0.805
1.163
1.039
1.148
1.008
1.032
-0.2425
0.059
16.961
1
< .0001
0.785
χ2
Df
p
290.278
286.235
277.546
27
27
27
<.0001
<.0001
<.0001
9.9706
8
0.2671
β
Model evaluation
Likelihood ratio test
Score test
Wald test
Goodness of fit test
Hosmer and Lemeshow
35
p
Conclusions
With any regression analysis, one should remember that merely finding a
relationship between a variable and a response measure does not demonstrate that the
variable caused the response. In the preceding analysis, a relationship seems to exist
between turn signal color and the odds of involvement as the struck vehicle in rear-end
collisions. Changing the color of a turn signal from red to amber appears to reduce the
odds of being struck by 3 to 28 percent. The exact mechanism responsible for this,
however, is unclear.
Although rear turn-signal color is implicated, it is important to recognize that
signal color is also confounded with other factors that may contribute to this relationship.
For example, if signal lamp color is dropped from the regression model, turn-signal/taillamp separation becomes a predictor, albeit weaker. It is also important to recognize that
the lamp characteristics included in the logistic regressions are incomplete. It is plausible
that characteristics of the light output of amber and red signal lamps differ in systematic
ways. Perhaps an amber turn signal appears brighter than a red one. Although prior
evidence (Mortimer & Sturgis, 1974) suggests that a red lamp is more conspicuous than
an amber one, this evidence was collected under static viewing conditions that are quite
different from the conditions drivers typically face on a roadway where the signal lamps
are likely first detected in the peripheral visual field. Does a red lamp still look more
conspicuous than an amber lamp when offset 10 degrees from the direction of gaze? If
this is true, then the differences observed between red and amber may not be so much
related to differences in color as it is to differences in lamp brightness.
The results also show that the choice of comparative data can also influence the
observed results. Using lead vehicles involved in non-relevant rear-end collisions (with
respect to turn-signal operations) as comparative data, Analysis 2 did not find any
relationship between color and the odds of involvement in a relevant crash. This suggests
that the non-relevant crash data may have not been as non-relevant as previous
researchers believed. Whatever influences a specific turn signal characteristic may have
had on a driver’s ability to detect and react to a turning or merging lead vehicle, that same
characteristic could have also influenced the detectability of other (perhaps non-relevant)
maneuvers of a lead vehicle. This influence crosstalk between the target and reference
datasets could diminish the chance of observing an influence of a rear signal
characteristic on the odds of a particular crash. Instead, the reference group should share
as many similarities as possible with the target group except that the variables of specific
interest—the rear signal lamp characteristics—should play no conceivable role in the
reference group crashes.
36
References
Kahane, C. J., & Hertz, E. (1998). The long-term effectiveness of Center High Mounted
Stop Lamps in passenger cars and light trucks (Technical Report No. DOT HS
808 696). Washington, DC: National Highway Traffic Safety Administration.
Luoma, J., Flannagan, M. J., Sivak, M., Aoki, M., & Traube, E. C. (1995). Effects of
turn-signal color on reaction times to brake signals (Technical Report No.
UMTRI-95-5). Ann Arbor, MI: The University of Michigan Transportation
Research Institute.
Luoma, J., Sivak, M., & Flannagan, M. J. (2006). Effects of dedicated stop lamps on
nighttime rear-end collisions (Technical Report No. UMTRI-2006-15). Ann
Arbor, MI: University of Michigan Transportation Research Institute.
Maurer, P. W. (1980). History of automotive lighting in Europe (Technical Paper No.
800339). Warrendale, PA: Society of Automotive Engineers.
Moore, D. W., & Rumar, K. (1999). Historical development and current effectiveness of
rear lighting systems (Technical Report No. UMTRI-99-31). Ann Arbor, MI:
University of Michigan Transportation Research Institute.
Mortimer, R. G., & Sturgis, S. P. (1974). Evaluations of automotive rear lighting and
signaling systems in driving simulator and road tests (Final Report No. UMHSRI-HF-74-24).
Washington,
DC:
National
Highway
Traffic
Safety Administration.
Post, D. V. (1975). Performance requirements for turn and hazard warning signals (Final
Report No. UM-HSRI-HF-75-5/DOT-HS-4-01001). Washington, DC: National
Highway Traffic Safety Administration.
Sivak, M., Flannagan, M. J., Miyokawa, T., & Traube, E. C. (2000). Color identification
in the visual periphery: consequences for color coding of vehicle signals.
Transportation Human Factors, 2(2), 135-150.
Sivak, M., Flannagan, M. J., Sato, T., Traube, E. C., & Aoki, M. (1994). Reaction times
to neon, LED, and fast incandescent brake lamps. Ergonomics, 37(6), 989-994.
Taylor, G. W., & Ng, W. K. (1981). Measurement of effectiveness of rear-turn-signal
systems in reducing vehicle accidents from an analysis of actual accident
data (SAE Technical Paper No. 810192). Warrendale, PA: Society of
Automotive Engineers.
37
DOT HS 811 037
September 2008