Professional Documents
Culture Documents
Red-Light Running
David Shinar, Muki Bourla, and Liat Kaufman, Ben Gurion University of the Negev, Beer-
Sheva, Israel
The goal of this study was to determine if traffic signals that are synchronized along
a route are associated with fewer red-light violations than traffic signals that are not
synchronized. A total 3600 cycles of traffic signals at 12 intersections along 2
major urban arteries were observed. Synchronized intersections were effective in
reducing the likelihood of red-light running (RLR) by (a) providing fewer oppor-
tunities than nonsynchronized intersections for RLR and (b) having a lower rate
of RLR relative to the number of opportunities. After adjustment for the number of
opportunities, the odds of entering the intersections in red in synchronized inter-
sections were nearly 1/7 the odds of RLR in nonsynchronized intersections.
Congestion reduced the effectiveness of synchronized intersections relative to non-
synchronized intersections. Male drivers were slightly more likely to run red lights
than female drivers, and the effects of synchronization were fairly constant across
age, gender, and the presence or absence of passengers. Actual or potential appli-
cations of this research involve signal synchronization to reduce aggressive driving
in general and RLR in particular.
Address correspondence to David Shinar, Department of Industrial Engineering and Management, Ben Gurion University
of the Negev, Beer Sheva 84105, Israel; shinar@bgumail.bgu.ac.il. HUMAN FACTORS, Vol. 46, No. 2, Summer 2004, pp.
367–372. Copyright © 2004, Human Factors and Ergonomics Society. All rights reserved.
368 Summer 2004 – Human Factors
The particular engineering solution that was Aviv. The desire was to match the synchronized
evaluated was the use of synchronized traffic and unsynchronized intersections as much as
signal lights. The synchronization consists of possible in terms of traffic volume, location, and
the coordination of traffic signals along several general purpose. Thus all intersections were
blocks of selected urban streets, so that their along two main north-south urban thorough-
green phase is synchronized with traffic mov- fares with two through lanes in each direction
ing on that route at or slightly below the speed and separate drop-off lanes for right and left
limit. This measure is designed to increase traf- turns. Eight of the intersections were in the cen-
fic flow, and to reduce speed variability and con- tral sections of major streets with synchronized
flicts among vehicles along the synchronized signals, and four were with nonsynchronized sig-
route. The aggressive driving measure that was nals. Data were recorded for the two through
used to evaluate the effectiveness of synchro- lanes during the daylight hours of the dry sum-
nized intersections was the tendency of drivers mer months, when the sky is clear. No data
to run red lights (red-light running, or RLR). were collected in extreme or unusual situations,
Synchronized intersections smooth the flow such as when a police officer was present or
of traffic, thereby reducing frustration from when traffic was stopped in one of the lanes
having to stop repeatedly (Hulfine & Adams, because of an accident. Observation periods
1995; Khalaf, Ergun, Ratrout, Al-Senan, & Al- consisted of rush hours (Sunday–Thursday,
Ahmadi, 1995). There is even evidence for a 7:00–9:30 and 16:30–19:00; Sunday is a regular
small decrease in intersection-related accidents, workday in Israel), nonrush hours (weekdays
particularly rear-end collisions (Berg, Kaub, & 10:00–15:00), and weekend hours (Saturday
Belscamper, 1986; Hulfine & Adams, 1995; Kha- 9:00–19:00).
laf et al., 1995; Schlabbach, 1988). However, the
direct contribution of synchronized intersections Experimental Design
to safer driver behavior is less obvious and has We observed 100 signal cycles on each of
not been studied. The principal issue is whether three observation periods at each of the 12 inter-
synchronization, by causing vehicles to move sections, yielding a total of 3600 traffic signal
in platoons only, merely reduces the number of cycles. The main independent measure was the
opportunities for conflicts at intersections or if nature of the intersection: synchronized versus
it changes driver behavior as well. The present nonsynchronized. For drivers going through a
study was designed to test whether or not syn- red light, as well as for a representative sample
chronized signals also reduce drivers’ inclina- of all drivers going through the intersections,
tion for RLR. Because some studies have found the following data were observed or estimated:
that male drivers, young drivers, and drivers driver gender, perceived age (younger than 25,
without passengers are more aggressive than 25–50, and 50+ years of age), and presence or
female drivers, older drivers, and drivers with absence of other occupants in the car.
passengers, respectively (Hyde, 1984; Retting &
Williams, 1996; Shinar, 1998), we also observed Procedure
these aspects of individual differences. Two trained observers conducted all obser-
vations. At each intersection, they observed the
METHOD
traffic moving in one direction on the two
Participants through lanes. The observers positioned them-
The participants in the study were regular selves in such a way that they were obscured
drivers who drove through the selected signalized from the drivers’ forward field of view. For each
intersections during the times of observation. cycle, they made separate counts of (a) the num-
Drivers were observed going through 3600 sig- ber of vehicles that entered the intersection after
nal cycles. the light had turned yellow but cleared it after the
light had turned red and (b) the number of ve-
Observation Sites and Times hicles that entered the intersection after the
We chose 12 intersections after several con- light had turned red. The reference for entering
sultations with the city traffic engineer of Tel and clearing the intersection was the car’s front
RED-LIGHT RUNNING AND SYNCHRONIZATION 369
end crossing the intersection’s start and end, as chronized intersections is that synchronization
defined by the imaginary line that is tangential forces traffic to move in “platoons,” with most
to the edges of the curbs. A car was considered cars moving through the synchronized green
RLR only when both observers agreed. For phase. Thus, with proper synchronization, traf-
each such vehicle the driver’s gender, estimated fic waiting at the red light consists mostly of
age, and presence or absence of passengers cars that had just entered the synchronized route
was also noted. In addition, to obtain exposure from another route. The results, summarized in
data on driver gender, estimated age, and pres- Table 1, show that the number of opportuni-
ence or absence of passengers, these data were ties for RLR in the synchronized intersections
noted for the first 100 cars going through the was much lower than in the nonsynchronized
intersection at each observation period (before intersections.
recording RLR). They are assumed to be a repre- To adjust for the differences in opportunities
sentative sample of the drivers traveling though for RLR, we conducted a chi-square analysis on
that intersection at that time. the frequencies of entering the intersection
after the light had turned red in the synchro-
RESULTS AND DISCUSSION nized and nonsynchronized intersections, rela-
tive to the number of opportunities observed in
Synchronization and RLR these two types of intersections. The difference
The principal hypothesis concerned the im- was highly significant, χ2(1) =145.06, p < .0001,
pact of synchronization on RLR. In all 3600 cy- yielding an odds ratio of 6.7. Thus the odds of
cles of observations, 185 drivers were observed RLR were nearly seven times higher in the non-
entering the intersection after the light had synchronized intersections than in the synchro-
changed to red. Of those, 139 were in the four nized intersections.
nonsynchronized intersections and 46 were in There were also 1174 observations of cars
the eight synchronized intersections. that entered the intersection after the light had
This difference by itself is insufficient to indi- changed to yellow and that cleared the intersec-
cate that RLR is more common in nonsynchro- tion only after the light had changed to red. For
nized intersections, given that the opportunity the remaining analyses, these vehicles were
for RLR is lower in these intersections. An op- combined with the 185 vehicles that entered the
portunity for RLR was defined as a situation in intersection during the red phase, yielding a
which one or more cars approached the signal- sample of 1359 RLR observations. This was
ized intersection just as the light turned yellow, done for two reasons. The first, a practical one,
thus giving the driver the opportunity to cross is the relatively small number of cars that en-
the intersection on red. When two cars, one be- tered the intersection after the light had turned
hind the other, approach an intersection after the red. The second, a behavioral one, is that all these
light has turned yellow or red and the lead driver drivers had enough time to stop in response to
stops for the light, there is only one opportunity the yellow signal before entering the intersec-
for RLR because the second driver is not afford- tion but chose not to do so. Instead, they risked
ed the opportunity to cross the intersection. crossing the intersection against the light: Some
The rationale for fewer opportunities for RLR cleared the intersection under a red signal, and
in synchronized intersections than in nonsyn- others actually entered it under a red signal. It
Rush Nonrush
Signal Type Hours Hours Weekends Total
Synchronized 78 53 44 175
Nonsynchronized 89 66 58 213
Note. Data based on two lanes and 100 cycles at each time period. An opportunity is defined as
situation in which one or more cars reached the intersection just as or immediately after the light
turned red.
370 Summer 2004 – Human Factors
is important to note, however, that although all and in both cases the number of opportunities
of these drivers did in fact block the intersection increased with congestion. However, there was
during the red phase, only those who entered no differentially significant effect of the period of
the intersection after the light had turned red observation (rush hours vs. nonrush hours vs.
would be legally considered as RLR. When the weekends) on the frequency of RLRs, χ2(2) =
two types of behaviors were combined, chi- 1.92, p = .38.
square analysis of the number of RLRs in the For a more detailed analysis of the source of
two types of intersections relative to the num- differences among the different observation
ber of opportunities yielded a highly significant periods, we looked at the number of RLRs as a
effect, χ2 = 111.06, p < .0001, and an odds function of the average number of cars queued
ratio of 2.1. in each lane at the end of the red-light phase
(just before it turned green). This provides a
Synchronization and Congestion Effects on quantitative, direct, and continuous measure
RLR of congestion (unlike the indirect assessment
At all three times of observation, the rates through period of observation). The results
of RLRs (number of RLRs/signal/observation with the best-fitting functions are presented in
period) were significantly lower in the synchro- Figure 1. Best-fitting quadratic equations yield-
nized intersections, as illustrated in Table 2. ed R2 = .93 for the synchronized intersections
These results reflect 40% fewer RLRs in syn- and R2 = .84 for the nonsynchronized intersec-
chronized intersections than in nonsynchro- tions. A linear fit to the two sets of data points
nized intersections. Furthermore, although the yielded R2 = .83 for both synchronized and
absolute number of RLRs was greater in rush nonsynchronized intersections.
hours than in nonrush hours and weekends, the These results are consistent with those of
relative difference between synchronized and Table 2, but they show more clearly that as con-
nonsynchronized intersections increased from gestion increases the absolute number of RLR
rush hours to nonrush hours and to weekends, increases in both types of intersections, whereas
χ2(2) = 49.29, p < .0001, χ2 for trend (1) = the difference in RLR between the synchronized
49.02, p < .0001. Thus there were 83% more and nonsynchronized intersections decreases
RLRs in nonsynchronized intersections than in until it finally disappears completely. Thus as
synchronized intersections during rush hours, congestion increases, the effectiveness of syn-
303% more in nonrush hours, and 386% more chronization decreases until it becomes totally
on weekends. irrelevant to driver behavior at the intersections.
As already stated, because of the nature of
signal synchronization, one possible source for Synchronization and Individual
the effect of synchronization could be simply Differences in RLR
that synchronization yields fewer opportunities Gender. The percentage of men in the RLRs
for RLR. Table 1 presents the number of oppor- (82%) was slightly (but significantly, p = .0005)
tunities for RLR in each condition. As can be higher than their percentage in the driving pop-
seen from Table 1, in each of the three observa- ulation at the same intersections (72%). The
tion periods the number of opportunities for relationship between gender and synchroniza-
RLR was in fact greater in the nonsynchronized tion (see Table 3) was not statistically signifi-
intersections than in the synchronized ones, cant, χ2 = 2.75, p = .10, indicating that the small
Rush Nonrush
Signal Type Hours Hours Weekends Total
Synchronized 56 13 6 75
Nonsynchronized 103 53 34 190
Note. Data based on two lanes and 100 cycles at each time period.
RED-LIGHT RUNNING AND SYNCHRONIZATION 371
Figure 1. The increase in RLR as a function of congestion in synchronized and nonsynchronized intersections.
difference between male and female drivers was tween age and synchronization (see Table 4) was
unaffected by synchronization. not statistically significant, χ2(2) = 0.25, p = .88,
Age. The percentage of older drivers (50+ indicating that synchronization did not have a
years) in the RLRs (23%) was slightly (but sig- differential effect on younger and older drivers.
nificantly, p = .027) higher than their percentage Presence of passengers. The percentage of
in the driving population at the same intersec- drivers without passengers in the RLRs (64%)
tions (17%). In contrast, the percentage of young was almost identical to their percentage in the
drivers (<25 years) in the RLRs (9%) was actu- driving population at the same intersections
ally slightly lower than their percentage in the (63%). The relationship between presence or
driving population at the same intersections absence of passengers and synchronization (see
(12%), but this difference was not significant Table 5) was not statistically significant, χ2 = 0.12,
(p = .15). Overall, it seems that age – at least p =.73, indicating that the effect of synchroniza-
as it was estimated – was unrelated to the ten- tion on drivers without passengers was not dif-
dency to run a red light. The relationship be- ferent from that on drivers with passengers.
TABLE 3: Number of RLRs Made by Male and Female Drivers Per Intersection
Synchronized 56 19 75
Nonsynchronized 152 38 190
Note. Data based on two lanes and 100 cycles across all observation periods.
TABLE 4: Number of RLRs Per Intersection for Each Estimated Age Group
Synchronized 6 54 15 75
Nonsynchronized 16 141 33 190
Note. Data based on two lanes and 100 cycles across all observation periods.
372 Summer 2004 – Human Factors
TABLE 5: Number of RLRs Per Intersection for Drivers With and Without
Passengers
Without With
Signal Type Passengers Passengers Total
Synchronized 57 18 75
Nonsynchronized 1310 59 190
Note. Data based on two lanes and 100 cycles across all observation periods.