American Journal of Public Health | November 2002, Vol 92, No. 111822 | Research and Practice | Peer Reviewed | Retting and Kyrychenko RESEARCH AND PRACTICE
Objectives.This study estimated the impact of red light camera enforcement on motor vehicle crashes in one of the first US communities to employ such cameras—Oxnard, California.
Methods. Crash data were analyzed for Oxnard and for 3 comparison cities. Changes in crash frequencies were compared for Oxnard and control cities and for signalized and nonsignalized intersections by means of a generalized linear regression model.
Results. Overall, crashes at signalized intersections throughout Oxnard were reduced by 7% and injury crashes were reduced by 29%. Right-angle crashes, those most as- sociated with red light violations, were reduced by 32%; right-angle crashes involving in- juries were reduced by 68%.
Conclusions. Because red light cameras can be a permanent component of the trans- portation infrastructure, crash reductions attributed to camera enforcement should be sustainable. (Am J Public Health. 2002;92:1822–1825)
Reductions in Injury Crashes Associated With Red Light
Camera Enforcement in Oxnard, California
| Richard A. Retting, MS, and Sergey Y . Kyrychenko, MS some additional rear-end crashes might result
from nonuniform changes in driver behavior.
For example, drivers, if they stop more often
for red lights, may be struck from behind by
drivers not intending to stop.
Rocchi and Hemsing
7 identified numerous
anecdotal and unpublished reports on reduc-
tions in right-angle or “red light running re-
lated” crashes following implementation of red
light camera enforcement; reported reductions
ranged from 10% in New York City to 88% in
Essex, United Kingdom. In Australia, where
red light cameras have been in use for about
20 years, controlled studies in Melbourne,
Sydney, and Victoria have generally reported
reductions in right-angle crashes that range
from 32% to 50%.
8–1 0Moderate increases in
rear-end crashes have been reported in Syd-
ney and Victoria but not in Melbourne.
The purpose of this study was to estimate
the impact of red light camera enforcement
on intersection crashes in Oxnard, California,
one of the first US communities to employ
such cameras.
METHODS
Changes in the number of motor vehicle
crashes were evaluated in Oxnard, which had
as estimated population of 156,372 in 1999
and a land area of 24.4 square miles.
11,12
A statewide red light camera law took ef-
fect in California in January 1996, permitting
municipal governments to establish local red
light camera enforcement programs. Under
this California law, a vehicle driver who runs
a red light, if sufficiently identified, is charged
with a moving violation. Front photography is
used to capture a likeness of the driver and
the vehicle’s front license plate (if present). In
cases in which the sex and estimated age of
the photographed driver match those of the
registered vehicle owner, the owner is pre-
sumed to be the driver and is issued a ticket
by mail (the registered owner is permitted to
rebut this presumption in court). Under Cali-
fornia law, citations issued through red light
camera enforcement programs carry the same
monetary penalties and license sanctions as
those resulting from conventional police traf-
fic stops—currently $271 and 1 driver’s li-
cense demerit point.
A 30-day warning period, during which
red light cameras photographed violators but
no tickets were issued, preceded red light
camera enforcement in Oxnard. As required
by state law, signs advising motorists of photo
enforcement of traffic signal laws were posted
on major roadways at numerous locations, in-
cluding entrances to the city. In addition, city
officials attempted to publicize and generate
awareness of the new program by issuing a
Each year in the United States more than
800 people die and an estimated 200000
are injured in crashes that involve red light
running.
1 Red light cameras can help reduce
red light running by automatically photo-
graphing vehicles whose drivers run red
lights. A red light camera system is connected
to a traffic signal and to sensors that monitor
traffic flow. The system continuously moni-
tors the traffic signal, and the camera itself is
triggered by any vehicle passing over the sen-
sors beyond both a preset minimum speed
and a specified time after the signal has
turned red. The camera records the date,
time of day, time elapsed from the beginning
of the red signal, and the speed of the vehicle.
Tickets typically are issued to owners of vio-
lating vehicles, based on review of the photo-
graphic evidence. T o date, approximately 50
communities in the United States have imple-
mented red light camera enforcement.
Red light camera enforcement is very ef-
fective in reducing red light violations. For
example, in both Oxnard, California, and
Fairfax, Virginia, violations due to red light
running dropped by about 40% during the
first year of camera enforcement.
2,3 In both
cities, reductions in red light violations were
nearly identical at intersections equipped and
not equipped with red light cameras, suggest-
ing that camera enforcement produces gen-
eral changes in motorists’ behavior rather
than simply encouraging drivers to obey traf-
fic signals at specific locations. Similar reduc-
tions in red light violations following imple-
mentation of camera enforcement were
reported in Australia,
4 Singapore,5 and the
United Kingdom.6
Less is known, however, about the impact
of red light camera enforcement on crashes,
the outcome of primary interest. Such en-
forcement would be expected to reduce the
frequency of right-angle collisions—the princi-
pal type of crash associated with red light
running—at signalized intersections. Also,
EXHIBIT 1
November 2002, Vol 92, No. 11 | American Journal of Public Health Retting and Kyrychenko | Peer Reviewed | Research and Practice | 1823 RESEARCH AND PRACTICE
press release and providing information to
local media. Also, postcards were mailed to
Oxnard residents during the warning period
to alert them to the new program.
Actual enforcement began on July 1, 1997.
Of approximately 125 intersections in Oxnard
with traffic signals, 11 intersections were
equipped with red light cameras. At each
camera location, only 1 of the typical 4 ap-
proaches to an intersection was monitored.
Controls
Three California cities that did not imple-
ment red light camera enforcement during the
study period were used as controls to elimi-
nate potentially confounding external factors
that might affect the frequency of motor vehi-
cle crashes, such as economic conditions, fuel
prices, and weather. The cities of Bakersfield
and San Bernardino were selected because
they each have approximately the same num-
ber of annual crashes as Oxnard, and their lo-
cations (more than 100 miles from Oxnard)
made it unlikely that camera enforcement in
Oxnard affected driver behavior in these
cities. The city of Santa Barbara, approxi-
mately 40 miles north of Oxnard, was also se-
lected because of its earlier use as a control
for Oxnard in evaluating changes in red light
violations following implementation of camera
enforcement.
2 Red light violation rates did not
change in Santa Barbara during the evaluation
of the Oxnard red light enforcement program.
Crash data for the 4 cities were obtained
from the California Statewide Integrated Traf-
fic Records System (SWITRS). Crashes were
analyzed for 29 months preceding camera
enforcement (January 1995–May 1997) and
for 29 months of enforcement (August
1997–December 1999). The first month of
enforcement (July 1997) was excluded from
analysis to ensure that drivers were aware of
the red light camera program, as was the
month before the start of enforcement (June
1997), when a warning period announcing
the enforcement was in effect.
Intersections of each city were divided into 2
groups, signalized and nonsignalized. Because
of the relatively small numbers of crashes asso-
ciated with the 11 camera enforcement sites in
Oxnard and because prior research has docu-
mented a large spillover effect of camera en-
forcement to intersections in the same commu-
nity not equipped with cameras,
2,3 crashes at
the 11 camera-equipped intersections were not
analyzed separately. As the “before” and “after”
periods would include different months, the
numbers of crashes for these periods would be
somewhat affected by seasonal variations; how-
ever, these variations were not expected to bias
estimates of the effect of red light cameras, be-
cause the statistical model used identical peri-
ods for all 4 cities and for nonsignalized as well
as signalized intersections.
SWITRS does not contain a variable that
indicates whether an intersection is signal-
ized; therefore, it was necessary to identify
manually all intersections with traffic signals
in the 4 cities by means of lists of signalized
intersections provided by the cities and the
California Department of Transportation
(CAL TRANS). Any intersection not identified
as having a traffic signal was considered non-
signalized. Information provided by the 4
cities and CAL TRANS regarding installation
of any new signals during the analysis period
was used to eliminate those intersections
from the study (a total of 82 new signals
were installed during the analysis period—10
in Oxnard, 41 in Bakersfield, 18 in San
Bernardino, and 13 in Santa Barbara).
Changes in crash rates after enforcement
were compared for Oxnard and control cities,
as well as for signalized and nonsignalized in-
tersections. Injury crashes were limited to
cases with a SWITRS variable for collision
severity equal to “fatal,” “severe injury,” or
“other visible injury,” the latter 2 classifications
being based on police-reported information.
Two types of multiple-vehicle crashes—right
angle and rear end—were defined by means
of SWITRS variables. Right-angle crashes (ex-
pected to be reduced at signalized intersec-
tions by red light camera enforcement) were
defined by means of SWITRS codes as
“broadside” collisions involving 2 motor vehi-
cles that were traveling at right angles to each
other before the crash (based on recorded
compass directions). Because SWITRS does
not contain a specific code to identify red
light–running crashes, it was not possible to
categorize crashes specifically as red light
running events. The closest category, right-
angle collisions, also could include some left-
turn crashes—SWITRS does not have a sepa-
rate category for left-turn crashes—as well as
crashes in which drivers may have been exit-
ing from a driveway located close to an inter-
section. Left-turn crashes, however, do not
typically involve red light running.
Rear-end crashes, which might increase with
red light camera enforcement due to changes
in driver behavior with regard to stopping for
red lights, were defined by means of SWITRS
codes as “rear-end” collisions involving 2
motor vehicles traveling in the same direction.
Model
A generalized linear regression model was
developed to evaluate changes in total crashes,
injury crashes, and specific crash types. The
model used the natural logarithm of crash
counts as the response variable. Independent
variables were city, intersection type (signal-
ized and nonsignalized), and period (before
and after enforcement). Two-factor interactions
of City×Period and City×Intersection Type
also were included, because crash trends were
different in different cities. Analysis of variance
was used to test for statistical significance.
RESULTS
Table 1 summarizes changes in the num-
bers of crashes from the baseline period
through the enforcement period, for signal-
ized and nonsignalized intersections. For the
3 control cities, the frequency of crashes
changed in a roughly similar manner at both
signalized and nonsignalized intersections. In
Bakersfield and Santa Barbara, the number of
crashes declined at both types of intersec-
tions; in San Bernardino, it increased. Table 1
also summarizes the effect of red light camera
enforcement as evaluated by the model. We
estimated that red light camera enforcement
would reduce the number of crashes at sig-
nalized intersections in Oxnard by 7% (95%
confidence interval [CI]=1.3, 12.5).
Table 2 summarizes changes in the number
of injury crashes for signalized and nonsignal-
ized intersections in all 4 cities. As was found
for total crashes, the number of injury crashes
in control cities changed in a roughly similar
manner at signalized and nonsignalized inter-
sections from the baseline period through the
enforcement period. Results of the statistical
model used to evaluate changes in injury
crashes also are summarized in Table 2. We
EXHIBIT 1
American Journal of Public Health | November 2002, Vol 92, No. 111824 | Research and Practice | Peer Reviewed | Retting and Kyrychenko RESEARCH AND PRACTICE
TABLE 1—Total Crashes Before and After Enforcement and Its Estimated Effects City and type of Intersection Before After Change (%)
Bakersfield
Nonsignalized 760 753 –0.9
Signalized 771 739 –4.2
San Bernardino
Nonsignalized 1220 1283 5.2
Signalized 1324 1400 5.7
Santa Barbara
Nonsignalized 712 622 –12.6
Signalized 488 438 –10.2
Oxnard
Nonsignalized 994 1011 1.7
Signalized 1322 1250 –5.4
Estimated Effects
Degrees
Effect of Freedom Mean Square F value P value Estimate Change (%) Camera 1 0.0013308 11.33 0.0281 –0.07296 –7.0
Error 4 0.00011741 . . . . . . . . . . . .
TABLE 2—Injury Crashes Before and After Enforcement and Its Estimated Effects City and type of Intersection Before After Change (%)
Bakersfield
Nonsignalized 245 241 –1.6
Signalized 243 233 –4.1
San Bernardino
Nonsignalized 204 225 10.3
Signalized 239 246 2.9
Santa Barbara
Nonsignalized 113 115 1.8
Signalized 89 84 –5.6
Oxnard
Nonsignalized 173 194 12.1
Signalized 299 239 –20.1
Estimated Effects
Degrees
Effect of Freedom Mean Square F value P value Estimate Change (%) Camera 1 0.02865345 35.62 0.004 –0.33855 28.7
Error 4 0.00080437 . . . . . . . . . . . .
estimated that red light camera enforcement
would reduce the number of injury crashes at
signalized intersections in Oxnard by 29%
(95% CI=16.6, 39.1).
Table 3 summarizes the effects of red light
camera enforcement at intersections on 2 pri-
mary types of multiple-vehicle crashes—right
angle and rear end. Overall, right-angle
crashes accounted for approximately 36% of
all crashes at signalized intersections and 42%
of all crashes at nonsignalized intersections;
rear-end crashes accounted for approximately
9% of all crashes at both signalized and
nonsignalized intersections. Based on trends in
right-angle crashes in the 3 comparison cities
and relative to changes in the frequency of
these types of crashes at nonsignalized inter-
sections in Oxnard, the model estimated a sig-
nificant 32% (95% CI=3.2, 53.0) reduction
in right-angle crashes at all signalized intersec-
tions in Oxnard due to the camera enforce-
ment (with 95% confidence limits of 3.2 and
53.0) and a significant 68% (95% CI=56.7,
76.5) reduction in right-angle injury crashes at
all signalized intersections (with 95% confi-
dence limits of 56.7 and 76.5). Based on
trends in rear-end crashes in the 3 comparison
cities and relative to changes in the frequency
of these types of crashes at nonsignalized in-
tersections in Oxnard, we estimated a non-
significant 3% increase in rear-end crashes at
signalized intersections.
DISCUSSION
Despite the large numbers of communities
using red light camera enforcement and a
long history of international use, relatively lit-
tle is known, even in Australia, about the ef-
fect of camera enforcement on motor vehicle
crashes. This study provides evidence that red
light cameras in the United States can reduce
the risk of motor vehicle crashes, especially
injury crashes, at intersections with traffic sig-
nals. During the time frame of this study, no
other comprehensive traffic safety programs
were implemented in Oxnard that could ac-
count for these crash reductions.
Crash reductions at signalized intersections
were found on a citywide basis, even though
cameras were installed at only 11 of 125 sig-
nalized intersections in Oxnard. Intersections
typically have 4 approach legs; red light cam-
eras monitored only 1 approach leg at each
of the 11 enforcement sites in Oxnard, so
only about 2% of all approaches to signalized
intersections were camera enforced.
The finding that crash reductions were ob-
served at traffic signals on a citywide basis is
consistent with prior behavioral research find-
ings that red light camera enforcement can
EXHIBIT 1
November 2002, Vol 92, No. 11 | American Journal of Public Health Retting and Kyrychenko | Peer Reviewed | Research and Practice | 1825 RESEARCH AND PRACTICE
TABLE 3—Estimated Effects on Right-Angle, Right-Angle Injury, and Rear-End Crashes Degrees
Effect of Freedom Mean Square F value P value Estimate Change (%) Right-angle crashes
Camera 1 0.03871492 9.17 0.0388 –0.39352 –32.5
Error 4 0.00422139
Right-angle injury crashes
Camera 1 0.32634352 107.72 0.0005 –1.14253 –68.1
Error 4 0.00302947
Rear-end crashes
Camera 1 0.00022718 0.00418 0.9515
a 0.030145 3.1
Error 4 0.05430999
aNonsignificant.
provide general deterrence against red light vi-
olations, with effects not limited to the specific
intersections with cameras. This result is im-
portant because the goal of highly conspicu-
ous traffic enforcement is to produce general-
ized changes in driver behavior with respect
to traffic safety laws, not simply to penalize
identified violators. And because red light
cameras can be a permanent component of
the transportation infrastructure, crash reduc-
tions attributed to camera enforcement should
be sustainable.
Injury crashes may decline to a greater de-
gree than less severe crashes because of the
nature of red light running crashes, which are
characterized by more severe side impacts
and relatively high impact speeds. Crash re-
ductions estimated in this study may be con-
servative, because the SWITRS data did not
contain sufficient detail to identify crashes
that were specifically red light running events.
Although the findings of an overall 29% re-
duction in injury crashes at signalized inter-
sections and a 32% reduction in right-angle
crashes are very positive, it is possible that
crashes specifically related to red light run-
ning declined to an even greater extent.
Exposure data were not available to exam-
ine the possibility that traffic volumes in Ox-
nard shifted from signalized to nonsignalized
intersections during the course of the study;
however, such a shift in driving patterns on a
citywide basis is highly improbable. In addi-
tion, prior research provides evidence that no
such shift of exposure occurred. At 12 Ox-
nard signalized intersections studied by Ret-
ting et al.
2 before red light camera enforce-
ment and several months after enforcement
began, there was no significant change be-
tween the baseline and enforcement periods
in the number of vehicles per hour.
Changes in motor vehicle crashes associ-
ated with red light camera enforcement may
differ in other communities because of factors
that could influence program effectiveness, in-
cluding number of cameras, penalty structure,
and extent to which camera enforcement is
publicized. Camera enforcement publicity will
be important because driver awareness of
traffic enforcement is the principal mecha-
nism against red light violations.
Although red light cameras are effective in
reducing red light violations and associated
crashes, such enforcement should be viewed
as a supplement to, and not a substitute for,
good engineering design practices that can re-
duce red light running and enhance intersec-
tion safety. These practices include ade-
quately timed yellow signal change intervals,
use of all-red clearance intervals, conspicuous
traffic signal housings, adequate signal bright-
ness, coordinated signal timing, and use of ad-
vance warning signs on high-speed roads or
at locations with limited sight distances.
About the Authors
Richard A. Retting and Sergey Y. Kyrychenko are with the Insurance Institute for Highway Safety, Arlington, Va.
Requests for reprints should be sent to Richard A. Ret- ting, MS, Insurance Institute for Highway Safety, 1005
North Glebe Rd, Arlington, VA 22201-4751 (e-mail:
[email protected]).
This article was accepted December 21, 2001.
Contributors
R. Retting managed the project and participated in the
study design and analyses. S. Kyrychenko developed the
regression model and performed the statistical analyses.
Acknowledgments
This work was supported by the Insurance Institute for
Highway Safety.
The authors acknowledge contributions to this re-
search made by the cities of Bakersfield, Oxnard, San
Bernardino, and Santa Barbara and by the California
Department of Transportation. The authors also ac-
knowledge technical contributions by Charles M.
Farmer of the Insurance Institute for Highway Safety.
Human Participant Protection
No protocol approval was needed for this study.
References
1. R etting RA, Ulmer RG, Williams AF. Prevalence
and characteristics of red light running crashes in the United States. Accid Anal Prev. 1999;31:687–694.
2. Retting RA, Williams AF, Farmer CM, Feldman
AF. Evaluation of red light camera enforcement in Ox-
nard, California. Accid Anal Prev. 1999;31:169–174.
3. Retting RA, Williams AF, Farmer CM, Feldman
AF. Evaluation of red light camera enforcement in Fair- fax, Virginia. ITE Journal. 1999;69:30–34.
4. Zaal D. Traffic law enforcement: a review of the
literature. Victoria, Australia: Monash University Acci- dent Research Centre; 1994. Report no. 53.
5. Chin HC. Effect of automatic red-light cameras on
red-light running. Traffic Engineering and Control.
1989;30:175–179.
6. County Surveyors’ Society. Traffic signal cameras.
Derbyshire, England: Special Activity Group on Acci-
dent Reduction; 1990. Report no. 1/12.
7. R occhi S, Hemsing S. A review of the road safety
benefits of red light cameras. In: Enhancing Transporta- tion Safety in the 21st Century, ITE International Confer- ence [conference report on CD-ROM]. Washington, DC:
Institute of Transportation Engineers; 1999.
8. Hillier W, Ronczka J, Schnerring F. An evaluation
of red light cameras in Sydney. New South Wales, Aus-
tralia: Roads and Traffic Authority; 1993.
9. Ogden KW. Safer Roads: A Guide to Road Safety
Engineering. Hampshire, England: Avebury T echnical;
1996.
10. South D, Harrison W, Portans I, King M. Evalua-
tion of the red light camera program and the owner
onus legislation. Victoria, Australia: Victoria Road Traf- fic Authority; 1988.
11.U S Census Bureau. Table 3. Land area, population,
and density for places in California: 1990. Washington, DC; May 1996. Available at: http://www.census.gov/
population/censusdata/places/06ca.txt. Accessed April
13, 2001.
12.U S Census Bureau. 1990 to 1996 cities and
places population estimates, California. Washington, DC; October 2001. Available at: http://www.census.gov/
population/estimates/metro-city/scful/SC99F_CA.txt.
Accessed April 13, 2001.
EXHIBIT 1