Beruflich Dokumente
Kultur Dokumente
1632 | Research and Practice | Peer Reviewed | Pérez et al. American Journal of Public Health | September 2007, Vol 97, No. 9
RESEARCH AND PRACTICE
campaign. The cameras take a photo of a ve- of daylight or darkness but to assess com- 486, respectively; the corresponding means
hicle when its speed exceeds by at least 5% muter traffic flows during these periods) and for the arterial roads were 2567 and 2394.
the limit allowed, and a ticket is mailed to the (2) weekdays (early Monday through late Mean numbers of people injured yearly on the
driver within 2 to 3 weeks. The amount of the Friday) with weekends (9 PM Friday to 4 AM beltway in the preintervention and postinter-
fine is €300 (approximately US $400) when Monday). Month was the unit of analysis. vention periods were 946 and 696, respec-
the speed exceeds by 50% the limit allowed. tively; means for the arterial roads were 3254
We conducted a natural “quasi-experiment” Statistical Analysis and 3074. Finally, mean numbers of vehicles
(a time-series study with a comparison group) Initially, we calculated ratios of road colli- involved in beltway collisions yearly during
designed to assess the effects of the beltway sions per 1000 vehicle-kilometers traveled per the preintervention and postintervention peri-
speed cameras. The “intervention group” was year for the beltway and the arterial roads. ods were 1466 and 1108, respectively, and
the beltway, and the comparison group con- Subsequently, we analyzed our outcome mea- the corresponding means for the arterial roads
sisted of arterial roads on which no fixed sures using Poisson regression models.7 After were 5068 and 4658.
speed cameras had been installed. The pre- adjusting for potential confounding by linear Ratios of collisions per 100 000 vehicle-kilo-
intervention period of the study took place trends and seasonal patterns, we conducted an meters traveled on the beltway increased from
from January 1, 2001, through March 31, intervention analysis in which we fit a dummy 21.1 in 2001 to 21.6 in 2002 and then de-
2003. The postintervention period spanned variable to compare the preintervention and creased to 17.2 in 2003 and 13.2 in 2004.
April 1, 2003, through March 31, 2005. postintervention periods. The model for each On arterial roads, ratios of collisions per
outcome can be summarized as follows: 100 000 vehicle-kilometers traveled decreased
Data Source from 100.2 to 93.2, 87.4, and 86.1 during
(1) ln(E[Yt]) = B + B1 × t + B2 × sin(2π t/T) +
We derived our data from the local police 2001, 2002, 2003, and 2004 respectively.
B3 × cos(2π t/T) + B4 × Xt ,
accident database. In Barcelona, a special Figure 1 shows numbers of road collisions
division of the police department is responsi- where T is the number of periods described by according to monthly subseries. Preinterven-
ble for the reporting of road collisions in the each sinusoidal function (e.g., T = 12 months), tion and postintervention years are depicted
city. Trained police personnel report road t is the time period (e.g., t=1 for January, t=2 for all monthly series. Numbers of beltway
collisions according to a standard protocol. for February), and Xt =1 represents the postin- road collisions were consistently lower in the
All collisions that involve damage to property tervention period (Xt =0 otherwise). postintervention period than in the preinter-
or injuries are reported. We considered only We used Stata statistical software to con- vention period. Trends were inconsistent on
collisions that occurred on the beltway or on duct the statistical analyses.8 We analyzed arterial roads.
arterial roads. data for 24 months after the intervention. We Once we controlled for trends and seasonal-
derived relative risks (RRs) from the adjusted ity by fitting Poisson distribution models, we
Variables and Indicators models and calculated attributable fractions observed a decrease of 27% in number of col-
Our outcome measures were numbers of from these relative risks ([RR − 1] ÷ RR) to esti- lisions and number of people injured and a de-
road collisions, numbers of people injured, mate the numbers of prevented collisions and crease of 26% in number of vehicles involved
and numbers of vehicles involved in colli- the reductions in numbers of people injured in collisions. The RR of a road collision occur-
sions. All types of collisions were included, and vehicles involved in collisions. ring on the beltway in the postintervention (vs
rather than only collisions involving injuries preintervention) period was 0.73 (95% confi-
(the percentage of noninjurious collisions is RESULTS dence interval [CI]=0.63, 0.85). That is, the
fairly constant, at approximately 9%). We did speed cameras appeared to have a protective
not analyze numbers of deaths separately After 2 years of speed camera operation effect. The RR for the first year of intervention
from numbers of injuries because the number on the Barcelona beltway, a significant de- was 0.71 (95% CI=0.62, 0.82), and the effect
of deaths resulting from beltway collisions is crease in the number of road traffic colli- increased in the second year (RR=0.61; 95%
very low (approximately 1 or 2 per year). sions was observed, not only on the beltway CI=0.5, 0.75). As can be seen in Table 1, RRs
There is often heavy daytime traffic on the but also on the arterial roads. Before the in- of injury were similar, as were values for vehi-
beltway, particularly on weekdays; during stallation of the speed cameras, probably as cles involved in collisions. By contrast, on arte-
these periods, it is impossible for motorists to a result of other road safety media cam- rial roads, RRs of all 3 outcomes assessed
exceed speed limits. Thus, speed cameras paigns and actions, a decreasing general were close to 1, indicating that there were no
should be more effective in reducing injuries trend in the number of accidents had been differences between the 2 periods.
during nighttime hours and on weekends, evident in the city since the year 2000. The protective effect of speed cameras in
when traffic flows are lighter. We conducted This trend was taken into account in the reducing collisions was evident during both
separate analyses comparing (1) daytime (7 AM time-series analysis. daytime and nighttime hours and on both
to 8:59 PM; this could perhaps have been Mean annual numbers of collisions occur- weekdays and weekends; effects were greater
more appropriately labeled “working hours” ring on the beltway during the preintervention in magnitude on weekends. RRs were similar
because our aim was not to assess the effects and postintervention periods were 638 and for daytime (RR = 0.73; 95% CI = 0.61, 0.88)
September 2007, Vol 97, No. 9 | American Journal of Public Health Pérez et al. | Peer Reviewed | Research and Practice | 1633
RESEARCH AND PRACTICE
FIGURE 1—Monthly subseries time trends in numbers of road traffic collisions occurring on the beltway (a) and arterial (b) roads: Barcelona,
Spain, January 2001–March 2005.
and nighttime (RR = 0.72; 95% CI = 0.52, 507 fewer people were injured, and 789 the first in which the number of vehicles in-
0.99) hours but were lower on weekends fewer vehicles were involved in collisions as volved in collisions has been included as an
(RR = 0.66; 95% CI = 0.46, 0.94) than on a result of implementation of speed cameras outcome measure.
weekdays (RR = 0.75; 95% CI = 0.62, 0.91). (Table 2). No significant reductions were ob- The study design we used and the statisti-
Figure 2 shows monthly observed distribu- served for arterial roads. cal analyses we conducted allowed us to con-
tions of number of road collisions and ex- trol for major factors that might have affected
pected distributions had speed cameras not DISCUSSION the results and that usually have an impact
been installed. There was a clear decrease in in evaluation studies of road safety interven-
the number of beltway road collisions during Our results show that speed cameras were tions. Using time-series analyses, we con-
the postintervention period, although the effective in reducing the numbers of road trolled for regression to the mean and long-
slope of the trend did not change (i.e., was collisions—and consequently the numbers of term effects, in that adjustment was made for
not statistically significant). There was also a people injured and vehicles involved in seasonality and trends. However, the study
steady decrease in the number of arterial collisions—in the context of a general overall period was not particularly long (52 months),
road collisions between 2001 and 2005, but decline in the number of collisions occurring and further research is necessary to deter-
again there was no change in the slope. Ex- in Barcelona. Effects were similar during day- mine trends over longer periods.
pected numbers of beltway collisions during time and nighttime hours and were greater Because of the characteristics and function
the postintervention period were higher than on weekends than on weekdays. To our of the beltway as the quickest route connect-
the actual numbers observed, but no differ- knowledge, ours is the first study in which ing the city with the outlying metropolitan
ences were apparent for arterial roads. time-series analyses with Poisson distribution area, we do not believe that there would have
From the attributable fractions, it can be models have been used to evaluate the effec- been a significant number of vehicles using
estimated that 364 collisions were prevented, tiveness of speed enforcement cameras and other routes in an attempt to avoid the speed
1634 | Research and Practice | Peer Reviewed | Pérez et al. American Journal of Public Health | September 2007, Vol 97, No. 9
RESEARCH AND PRACTICE
TABLE 1—Relative Risks (RRs) for Outcome Variables Relative Risks During the Postintervention
Period Relative to the Preintervention Period: Barcelona, Spain, January 2001–March 2005
Overall
April 2003–March 2004 0.71* (0.62, 0.82) 1.00 (0.90, 1.11) 0.72* (0.60, 0.87) 1.00 (0.89, 1.13) 0.73* (0.63, 0.84) 0.99 (0.89, 1.10)
April 2004–March 2005 0.61* (0.50, 0.75) 1.02 (0.88, 1.19) 0.62* (0.48, 0.81) 0.99 (0.83, 1.17) 0.66* (0.54, 0.81) 1.04 (0.89, 1.20)
Daytimea
April 2003–March 2004 0.72* (0.60, 0.87) 1.00 (0.89, 1.12) 0.70* (0.56, 0.88) 1.01 (0.89, 1.15) 0.73* (0.62, 0.87) 1.00 (0.90, 1.12)
April 2004–March 2005 0.66* (0.51, 0.86) 1.06 (0.90, 1.24) 0.62* (0.45, 0.87) 1.01 (0.84, 1.21) 0.69* (0.54, 0.89) 1.08 (0.93, 1.27)
Nighttimeb
April 2003–March 2004 0.71* (0.52, 0.96) 0.99 (0.85, 1.16) 0.79 (0.54, 1.15) 1.02 (0.85, 1.23) 0.70* (0.50, 0.99) 0.96 (0.82, 1.14)
April 2004–March 2005 0.50* (0.33, 0.78) 0.93 (0.74, 1.17) 0.57* (0.33, 0.98) 0.96 (0.74, 1.24) 0.53* (0.33, 0.87) 0.88 (0.70, 1.12)
Weekdays
April 2003–March 2004 0.73* (0.61, 0.89) 0.99 (0.88, 1.11) 0.79* (0.66, 0.96) 0.98 (0.86, 1.13) 0.76* (0.64, 0.91) 0.99 (0.87, 1.11)
April 2004–March 2005 0.65* (0.49, 0.85) 1.05 (0.89, 1.23) 0.70* (0.53, 0.92) 1.01 (0.83, 1.23) 0.69* (0.54, 0.89) 1.06 (0.90, 1.26)
Weekends
April 2003–March 2004 0.65* (0.46, 0.93) 1.00 (0.83, 1.21) 0.57* (0.35, 0.93) 1.05 (0.86, 1.29) 0.63* (0.43, 0.93) 1.00 (0.82, 1.21)
April 2004–March 2005 0.50* (0.31, 0.83) 0.92 (0.70, 1.20) 0.46* (0.23, 0.90) 0.91 (0.68, 1.23) 0.56* (0.33, 0.97) 0.93 (0.70, 1.24)
Note. CI = confidence interval. Values were derived from time-series analyses using Poisson regression models in which we controlled for trends and seasonality.
a
Daytime was defined as 7 AM to 8:59 PM.
b
Nighttime was defined as 9 PM to 6:59 AM.
*P < .05.
Limitations
Comparison groups can allow for control-
ling of changes in traffic volume and general
changes; however, in our study, the interven-
tion group and the comparison group were
not equivalent, as a result of engineering char-
acteristics and because mobile speed cameras
operated randomly on the arterial roads. The
engineering features of the beltway and arte-
rial roads are clearly different (the latter hav-
ing intersections and involving lower mean
speeds), but the function of the roads is the
same: to connect the center city with the out-
lying metropolitan area. Furthermore, the belt-
way and arterial roads account for similar pro-
Note. n = number of road collisions each year. Confidence intervals (CIs) refer to expected road collisions. We controlled for portions of vehicle-kilometers traveled.
trends and seasonality using time-series analysis with Poisson regression models. Because of our time-series study design, we
a
2005 includes January 1 through March 31.
did not need a strict comparison group, but
FIGURE 2—Monthly numbers of observed and expected road collisions on the beltway (a) and the use of this group bolstered our findings.
arterial (b) roads: Barcelona, Spain, January 2001–March 2005. Other road safety interventions that we did
not control for, such as screening for drinking
September 2007, Vol 97, No. 9 | American Journal of Public Health Pérez et al. | Peer Reviewed | Research and Practice | 1635
RESEARCH AND PRACTICE
TABLE 2—Relative Risks, (RR) Attributable Fractions, and Observed, Expected, and Prevented beltway during certain hours, especially on
Numbers of Collisions, Injuries, and Vehicles Involved in Collisions After 24 Months of weekdays, traffic jams make it impossible to
Speed Camera Operation: Barcelona, Spain, January 2001–March 2005 drive at speeds higher than those permitted.
To take into account times during which there
Attributable are typically higher traffic flows, we defined
RR Fraction No. No. No.
(95% CI) (95% CI) Observed Expected Preventeda “daytime” as 7 AM to 8:59 PM. Our aim was
not to assess effects of daylight or darkness
Beltway but, rather, to examine commuter traffic
Collisions 0.73* (0.63, 0.85) –0.37 (–0.59, –0.18) 971 1335 –364 flows on workdays. In Spain, most people
Injuries 0.73* (0.61, 0.89) –0.37 (–0.64, –0.12) 1391 1898 –507
start work at 8 AM, and shops primarily close
Vehicles involved 0.74* (0.64, 0.85) –0.35 (–0.56, –0.18) 2215 3004 –789
around 8:30 PM.
in collisions
As expected, speed cameras had a signifi-
Arterial roads
cant effect in terms of reducing road injuries
Collisions 0.99 (0.90, 1.10) –0.01 (–0.11, 0.09) 4788 4821 –33
on weekends, although confidence intervals
Injuries 1.10 (0.90, 1.14) 0.09 (–0.11, 0.12) 6147 6094 53
overlapped with weekday results. Contrary
Vehicles involved 0.99 (0.89, 1.09) –0.01 (–0.12, 0.08) 9315 9451 –136
in collisions to expectations, however, the protective ef-
fects observed were similar for daytime and
Note. CI = confidence interval. Values were derived from time-series analyses using Poisson regression models in which we nighttime hours. To our knowledge, only 1
controlled for trends and seasonality.
a
Observed minus expected. study has provided RRs of injurious accidents
*P < .05. according to time of day,10 and that study
showed daytime and nighttime RRs of 0.46
and 0.55, respectively. Although in that inves-
and driving or general safety campaigns, A more recent systematic review re- tigation daytime and nighttime referred to
could also have influenced the trends ob- ported preintervention–postintervention daylight and lack of daylight, the results were
served here, although they would have af- reductions in the vicinity of camera sites similar to those revealed in our study. We
fected both the beltway and the arterial from 14% to 72% for all collisions, 8% to were unable to find any studies reporting
roads. There also could have been effects as- 46% for collisions involving injuries, and weekday–weekend comparisons of injury
sociated with motorists driving at reduced 40% to 45% for collisions resulting in fa- outcome measures.
speeds on other city streets. Such factors talities or serious injuries.9 Hess, in a study Although our study was set in a very local-
would have tended to produce reductions in conducted in the United Kingdom, esti- ized area, the characteristics of the roads
collisions, leading to underestimates of the ef- mated a 46% global reduction in the num- studied (i.e., an urban beltway and arterial
fects produced by the introduction of speed ber of accidents occurring in the vicinity roads) are probably similar to those of many
cameras. of cameras and a 29% reduction in the other cities, not only in Spain but elsewhere.
number of accidents occurring on urban Thus, we believe that the results described
Speed Camera Effectiveness roads.10 Christie et al. reported a 51% re- here would be generalizable to other cities.
It is difficult to compare our results with duction of injurious collisions in a study Our findings show that speed cameras are
those of other published studies, because of focusing on 101 mobile speed cameras im- effective in reducing the numbers of road
the different methodologies used. It must be plemented in South Wales.11 collisions and consequently the numbers of
borne in mind that our study focused on the Chen et al. reported a reduction of 16% in people injured and numbers of vehicles in-
beltway of a large city of which the speed the number of accidents on a British Columbia volved in collisions in an urban setting. In-
limit is 80 km per hour. Few published stud- highway on which the maximum speed was 80 stallation of speed cameras in urban areas
ies have reported maximum speeds, and to 90 km per hour.12 Mountain et al.,3 in a should be considered in efforts to increase
many have focused on highways. After we study focusing on speed cameras installed in road safety.
controlled for time-dependent confounders, 48 km per hour areas of the United Kingdom,
we estimated a reduction of 27% in the num- reported a 24% reduction in personal injury
ber of collisions and the number of people in- accidents, of which 19% was attributable to the About the Authors
jured and a reduction of 26% in the number effects of the cameras. Overall, these compar- Katherine Pérez and Marc Marí-Dell’Olmo are with the
of vehicles involved in collisions. Reported re- isons show that our results fall into the range of Agència de Salut Pública de Barcelona, CIBER Epidemi-
ología y Salud Pública, Barcelona, Spain. Aurelio Tobias
ductions in outcomes across other studies those from previously reported studies. is with the Escuela Nacional de Sanidad, Instituto de Salud
have ranged from 5% to 69% for collisions, We explored differences in the effective- Carlos III, Madrid, Spain. Carme Borrell is with the Agèn-
12% to 65% for injuries, and 17% to 71% ness of speed cameras during daytime versus cia de Salut Pública de Barcelona, CIBER Epidemiología
y Salud Pública, Barcelona, and the Departament de Cièn-
for deaths in the immediate vicinity of cam- nighttime hours and on weekdays versus cies Experimentals i de la Salut, Universitat Pompeu
era sites.4 weekends because on some stretches of the Fabra, Barcelona.
1636 | Research and Practice | Peer Reviewed | Pérez et al. American Journal of Public Health | September 2007, Vol 97, No. 9
RESEARCH AND PRACTICE
Requests for reprints should be sent to Katherine Pérez, and catchment area. Available at: http://www.cts.cv.ic.
PhD, Agència de Salut Pública de Barcelona, Pl Lesseps, 1, E ac.uk/StaffPages/StephaneHess/papers/Hess_slec_
08023 Barcelona, Spain (e-mail: cperez@aspb.es). 2004.pdf. Accessed May 18, 2007.
This article was accepted July 27, 2006.
11. Christie SM, Lyons RA, Dunstan FD, Jones SJ. Are
mobile speed cameras effective? A controlled before
Contributors and after study. Inj Prev. 2003;9:302–306.
K. Pérez originated and designed the study and wrote 12. Chen G, Meckle W, Wilson J. Speed and safety
the original draft of the article. M. Marí-Dell’Olmo per- effect of photo radar enforcement on a highway corri-
formed the statistical analyses. A. Tobias assisted with dor in British Columbia. Accid Anal Prev. 2002;34:
the statistical analyses. C. Borrell assisted in conceptual- 129–138.
izing ideas.
Acknowledgments
The work described in this article was partially sup-
ported by the Red Española de Centros de Epidemiolo-
gia y Salud Pública (grant C03/09).
We thank Mercé Navarro, Diego Navarro, Angel
López, Joan Mañosa, and Alex Culubret, as well as
Manuel Haro, and Josep Royuela (Guàrdia Urbana de
September 2007, Vol 97, No. 9 | American Journal of Public Health Pérez et al. | Peer Reviewed | Research and Practice | 1637