Koepsell T, McCloskey L, Wolf M, Moudon AV, Buchner D, Kraus J, Patterson M. Crosswalk Markings and the Risk of Pedestrian–Motor Vehicle Collisions in Older Pedestrians. JAMA. 2002;288(17):2136-2143. doi:10.1001/jama.288.17.2136
Author Affiliations: Department of Epidemiology (Drs Koepsell and Wolf and Mr Patterson), Harborview Injury Prevention and Research Center (Drs Koepsell, McCloskey, and Wolf and Mr Patterson), Department of Health Services (Dr Koepsell), and College of Architecture and Urban Planning (Dr Moudon), University of Washington, Seattle; Physical Activity Branch, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, Atlanta, Ga (Dr Buchner); and Southern California Injury Prevention Research Center, University of California at Los Angeles, Los Angeles (Dr Kraus).
Context Motor vehicles struck and killed 4739 pedestrians in the United States
in the year 2000. Older pedestrians are at especially high risk.
Objective To determine whether crosswalk markings at urban intersections influence
the risk of injury to older pedestrians.
Design Case-control study in which the units of study were crossing locations.
Setting Six cities in Washington and California, with case accrual from February
1995 through January 1999.
Participants A total of 282 case sites were street-crossing locations at an intersection
where a pedestrian aged 65 years or older had been struck by a motor vehicle
while crossing the street; 564 control sites were other nearby crossings that
were matched to case sites based on street classification. Trained observers
recorded environmental characteristics, vehicular traffic flow and speed,
and pedestrian use at each site on the same day of the week and time of day
as when the case event had occurred.
Main Outcome Measure Risk of pedestrian–motor vehicle collision involving an older
Results After adjusting for pedestrian flow, vehicle flow, crossing length,
and signalization, risk of a pedestrian–motor vehicle collision was
2.1-fold greater (95% confidence interval, 1.1-4.0) at sites with a marked
crosswalk. Almost all of the excess risk was due to 3.6-fold (95% confidence
interval, 1.7-7.9) higher risk associated with marked crosswalks at sites
with no traffic signal or stop sign.
Conclusions Crosswalk markings appear associated with increased risk of pedestrian–motor
vehicle collision to older pedestrians at sites where no signal or stop sign
is present to halt traffic.
On September 14, 1899, at the corner of 74th Street and Central Park
West in Manhattan, a man named Henry Bliss stepped off a streetcar and was
struck and killed by an electric taxicab. Mr Bliss, 68, became the first American
to be fatally injured in a pedestrian–motor vehicle collision.1,2 A century later, pedestrian–motor
vehicle collisions caused 4739 deaths in the United States in 2000, accounting
for 11% of all motor vehicle deaths and 78 000 reported injuries.
Adults aged 65 years and older accounted for 21% of pedestrian deaths
in 2000 and had a pedestrian–motor vehicle collision mortality rate
of 2.85 per 100 000 person-years—higher than in any other age group.3- 5 The excess of pedestrian
deaths among older adults has been found to be even more marked after accounting
for time spent as a pedestrian and the number of streets crossed.6 However, the most common mechanisms leading to pedestrian–motor
vehicle collisions appear to differ in older adults from those in younger
age groups. Collisions in which the victim is an older pedestrian are less
likely to involve an intoxicated pedestrian,7,8 are
more likely to occur at an intersection rather than mid block,9 and
are more likely to occur during daylight hours.8,10,11
Because these collisions result from the interplay of pedestrian, vehicle-driver,
and environment, any of these variables can be targeted for preventive interventions.10,12,13 Strategies aimed
at maximizing the safety of the street crossing environment may be especially
attractive because they may avoid the need to re-educate and to motivate millions
of drivers and pedestrians to make long-lasting behavioral change. One such
environmental factor is crosswalk markings, which are intended to guide pedestrians
in a safe path across the street so that pedestrians can be seen by drivers,
and to alert motorists that pedestrians may be encountered there. Whether
crosswalk markings actually prevent pedestrian–motor vehicle collisions
remains unclear. Some studies have suggested that sites with marked crosswalks
may be safer than sites without markings,14 but
others have found marked crosswalks to be associated with higher injury rates,15,16 even after accounting for differences
in pedestrian volume. After reviewing the evidence, the National Committee
for Injury Prevention and Control recommended in 1989 that crosswalk marking
be halted pending additional research into its safety effects.2
We wanted to determine whether marked crosswalks increase or decrease
the risk of pedestrian–motor vehicle collision to an older person crossing
the street at an urban intersection. We also investigated whether the risk
of these collisions varies by type of marking pattern, the condition of the
markings, and other environmental characteristics.
We used a case-control study design in which the units of study were
street-crossing locations ("crossings"). Case sites were crossings at which
a collision involving an older pedestrian had occurred. Control sites were
other crossings in the same neighborhood, matched to case sites on street
The study was conducted in 4 cities in western Washington and 2 in southern
California, with case accrual from February 1995 through January 1999 (Table 1). Data collection began in Seattle
and Tacoma, Wash, and in Long Beach, Calif, and later expanded to Everett
and Bellevue, Wash, to boost case accrual. After the first 3 months, an area
encompassing 4 police jurisdictions in West Los Angeles, Calif, was substituted
for Long Beach to reduce travel time and costs from the field center at the
University of California at Los Angeles. Throughout the study, case sites
were matched to control sites from the same city.
A case site was defined as a crossing location at which a pedestrian
aged 65 years or older had been crossing the street when he/she was struck
by a motor vehicle (with or without injury), resulting in a police report.
Collisions involving an older pedestrian who was not in the process of crossing
the street—those struck along the roadside or on the sidewalk or who
fainted in the roadway—were excluded.
The street being crossed was termed the index street. To qualify for inclusion, it had to be a public thoroughfare other
than an alleyway, driveway, or a parking lot and it could not be on private
property. Copies of police reports of pedestrian–motor vehicle collisions
were forwarded by traffic authorities in each city when a collision involving
an older adult had occurred. Each report was then reviewed to verify eligibility
under the study's case definition. Periodic checks of police report files
were conducted to verify ascertainment of all qualifying cases. This article
concerns crossings located within 30 feet of an intersection. (For an additional
70 mid block pedestrian–motor vehicle collision sites, the control-selection
and data-collection protocols were quite different, so they are excluded from
Two control sites were matched to each case site on the basis of neighborhood
and the classifications of roadways meeting at the intersection. On maps provided
by city traffic engineers, each street had been designated as a principal
arterial, minor arterial, collector-arterial, or nonarterial, based chiefly
on number of lanes, daily traffic volume, and speed limit. A crossing was
deemed a potential control for a certain case if the street being crossed
was in the same category as the index street at the case site and if the 2
next busiest streets radiating out from the intersection were also in the
same classifications as the 2 next busiest streets at the case intersection.
Overall, this produced an approximate match on traffic volume for the index
street, as well as for the 2 next busiest intersecting streets.
From other maps we determined the US Census block group that contained
the case site. A typical block group is an area of about 12 city blocks. Within
it, we identified all potentially eligible control crossings, numbered them,
and chose 2 at random using a random-number table. If the block group contained
fewer than 2 potential control crossings, control sites were sought in the
surrounding block groups using a random-selection scheme, working outward
from the case's block group in concentric rings until 2 control sites had
Environmental conditions, traffic, and pedestrian flow at a given location
can change over time. Using an incidence-density sampling scheme,17 a site that had already been studied as a case or
control could qualify again to serve as a case or control site at a different
date and time. Among all 282 case and 564 control sites, 5 were studied twice
as case sites, 17 were studied once as a case and once as a control site,
and 18 were studied twice as control sites.
Two trained field workers conducted a standardized environmental assessment
at each case and control site. To control for cyclical variation in such factors
as pedestrian and vehicular traffic, signal phases and timing, and lighting
conditions, all such time-critical observations were made on the same day
of the week and at the same time of day as when the victim had been struck
at the case site. This was called the index time.
Field workers were kept blinded to the case and control status of sites they
visited, and the ordering of visits to case and control sites within a set
At each case and control site, the study crossing consisted of the zone
in which a pedestrian would walk from one side of the index street to the
other, whether or not any crosswalk markings were present. Some crossings
consisted of multiple segments if the path across
the street contained 1 or more refuges (such as a raised median) where a pedestrian
could stop safely and wait for the next signal cycle or for traffic to clear
before continuing. A segment was defined as a portion of a crossing between
Field workers recorded data on the geometry and dimensions of the intersection;
system of traffic flow regulation (such as a traffic signal or stop sign),
if any; presence or absence of crosswalk markings; and other environmental
factors. If a marked crosswalk was present, they recorded its dimensions,
the marking pattern, the number and width of marking stripes, and the extent
to which markings had been worn away by traffic and weather. As a measure
of the "dose" of crosswalk markings, a summary index of pigment density was
calculated. It estimated the proportion of the area within a marked crosswalk's
bounding rectangle that was covered by pigment. The value of this index depended
on the length and width of the marked area, the number and dimensions of pigmented
stripes used to form the marking pattern, and the extent of wear, as judged
by field workers as the approximate percentage of originally pigmented surface
area still covered.
To characterize vehicular traffic at each site, a portable radar gun
was used to measure the speed of 50 vehicles (or all vehicles for 10 minutes
at low-volume sites) as they passed over the target crossing in either direction.
Vehicular traffic was also videotaped for 10 minutes, centered around the
index time. These videotapes were later viewed by research assistants, blinded
to case and control status, who tabulated the number of vehicles of various
types and the path each vehicle had taken through the intersection.
Pedestrian flow at each crossing was videotaped for 30 minutes, divided
into two 15-minute periods, 1 before and 1 after the index time. In Washington,
pilot testing indicated that an openly visible video camera and tripod stimulated
pedestrians' curiosity and altered their behavior, so the video camera was
hidden in a plastic trash can with a hole cut in its side. In California,
pedestrians seemed generally more oblivious to a video camera, which was usually
left operating in an open location. These videotapes were later viewed by
2 research assistants who recorded the sex and estimated age of each pedestrian.
A videotape showing people of known ages crossing the street was used as a
training guide to estimate ages. Pilot testing had shown high correlation
between estimated age and age as determined by asking the pedestrian (Pearson r = 0.92, n = 44). During the study, the intraclass correlation
between observers on estimated pedestrian age was 0.91.
The main analyses concerned the extent to which crosswalk marking characteristics
were associated with a site's case-control status, controlling for other site
characteristics. The odds ratio (OR) was used as the measure of association,
and it is known to provide a good estimate of relative risk in case-control
studies of rare outcomes.18 The OR can be interpreted
herein as the risk of pedestrian–motor vehicle collision at a site with
a certain characteristic, divided by the risk at a site without that characteristic.
To account for case-control matching, we used conditional logistic regression.19 For simplicity of presentation, tables that compare
case and control sites show means and standard deviations or percentages for
all cases and all controls combined. All ORs, confidence intervals (CIs),
and P values were derived from conditional logistic
regression models that accounted for matching. P values
less than .05 were regarded as statistically significant.
Potential confounding factors included other environmental characteristics
and the amount of pedestrian and vehicular traffic flow at each site. In addition
to the number of older pedestrians observed on the videotape at each site,
we also included the number of younger pedestrians (age <65 years) seen
at each site as a covariate. This was because it provided useful information
about average pedestrian flow at the site even if few older pedestrians were
The relation between collision risk and pedestrian and vehicle flow
was modeled using the fractional-polynomial approach described by Roystan
and Altman.20 In general, a single logarithmic
term fitted the data well for sites with any pedestrians seen. Because of
the low number of pedestrians at some sites, dummy variables were also added
to indicate whether any older or younger pedestrians were observed at the
All analyses were performed using STATA 6.0 (STATA Corp, College Station,
Tex). The study protocol was approved by the University of Washington and
University of California at Los Angeles institutional review boards.
A total of 282 qualifying collisions involving an older pedestrian were
identified in study cities during the surveillance period (Table 1). Seattle and West Los Angeles contributed the largest share
of sites. According to the police reports, all but 5 collisions resulted in
injury to the pedestrian. In 20 cases the injury was fatal.
Table 2a compares case and
control sites by factors other than crosswalk markings. Although the number
of streets radiating from the intersection was not a matching factor, case
and control sites proved to be similar in this respect. Most intersections
had 4 radiating streets. The index street's width and number of traffic lanes
were also similar between case and control sites. Because only about 6% of
sites had more than 1 crossing segment, the remaining analyses were restricted
to case and control crossings with only 1 segment.
Single-segment case and control crossings were also generally similar
on segment length (along the pedestrian path), number of lanes crossed, type
and condition of surface material, and width (perpendicular to the pedestrian
path) of a marked crosswalk, if present (Table 2). Compared with control sites, traffic regulation on the
index street at case sites more often involved a phased traffic signal, usually
accompanied by a pedestrian signal.
Despite the matching on street classification, measured traffic flow
was about 4% greater at case sites than at control sites. This difference
was statistically significant for all vehicles combined, for automobiles,
for sport-utility vehicles (SUVs), and for vans, but more vehicles of every
type were observed at case sites than at control sites. However, mean vehicle
speeds over the study crossing were slightly slower (0.4 miles per hour) at
A total of 17 105 pedestrians were videotaped crossing the street
at single-segment sites, including 1154 whose estimated age was 65 years and
older. Pedestrian flow was observed to be about 50% greater at case sites
than at control sites (27.2 pedestrians per half hour at case sites vs 18.5
at control sites, P<.001). A similar ratio applied
to older pedestrians (1.9 at case sites vs 1.2 at control sites). At least
1 older pedestrian was observed at 48% of case sites vs 32% of control sites.
Crosswalk markings were more often present at case sites than at control
sites (68% vs 49%, Table 3). Overall,
presence of a marked crosswalk was associated with a 4.0-fold increase in
risk of a collision involving an older pedestrian (95% CI, 2.5-6.2). However,
among sites with a marked crosswalk, no significant differences were found
by marking pattern or condition of markings. Using a summary index of pigment
density as a measure of the "dose" of markings, the excess risk associated
with marked crosswalks was greatest with relatively faint markings, although
elevated risk was also found for marked crosswalks with medium or dense markings.
A test for trend showed no significant difference in the effect of crosswalk
markings in relation to pigment density.
An artifactual association between pedestrian–motor vehicle collision
occurrence and presence of a marked crosswalk could arise if markings had
been applied or refreshed after a pedestrian had been hit but before the study
team visited the site. To check on this possibility, records of the Seattle
Transportation Department were reviewed for all 39 case sites at which crosswalk
markings in "good" or "excellent" condition were found by the field team.
Crosswalk markings had not been newly applied or reapplied at any of the sites
between the collision date and the data collection date.
The association between pedestrian–motor vehicle collision risk
and presence of a marked crosswalk changed somewhat after adjusting for other
characteristics that differed between case and control sites. Table 4 summarizes the adjusted OR for crosswalk marking in several
multivariate models that adjusted for various combinations of potential confounding
factors. The adjusted OR for presence of a marked crosswalk declined from
4.0 in the unadjusted base model to 2.1 in the model that controlled for 5
covariates, with similar levels of precision. The last model suggested about
a 2.1-fold increase in risk associated with presence of a marked crosswalk,
after controlling for pedestrian flow, vehicular traffic flow, crossing segment
length, and type of traffic regulation.
Measurement error in a covariate can limit the ability to remove confounding
from that source.21- 23 To
help gauge the possible impact of residual confounding by pedestrian and vehicular
traffic flow, a sensitivity analysis was carried out, based on methods described
by Armstrong et al24 Measurement errors in
these factors were assumed to be independent of each other and of crosswalk-marking
status. The model suggested that the observed association between crosswalk
markings and pedestrian–motor vehicle collision risk could be explained
fully by residual confounding from older-pedestrian flow only if the test-retest
reliability of this measure were less than about 0.38. For vehicular traffic
flow, the corresponding threshold reliability was about 0.50. In actuality,
only 1 measurement of each variable was obtained per site, but a lower-bound
estimate of reliability was obtained by treating the measurements at the 2
paired control sites for each case as replicate observations. By this method,
the reliability (intraclass correlation coefficient) of older-pedestrian flow
was estimated to be 0.54, and that of vehicle flow was estimated to be 0.89.
Thus, neither source of imprecision appeared sufficient by itself to explain
away the observed association. Using these lower-bound estimates of reliability,
the adjusted OR for crosswalk markings, controlled for older-pedestrian and
vehicle flow and corrected for measurement error, would still be about 1.50.
Lastly, we examined variation in the effect of crosswalk markings in
relation to other characteristics of sites (Table 5). Most of the overall increase in risk was due to a 3.6-fold
elevation at crossing sites where the flow of vehicles on the index street
was unimpeded by a traffic signal or stop sign. In contrast, there was almost
no association between risk and presence of crosswalk markings at locations
with a traffic signal or stop sign. No significant variation in the adjusted
OR for crosswalk markings was found among the 3 cities with the most cases,
or according to whether any older pedestrians were observed during videotaping
at the site.
The results of this study suggest that, contrary to the good intentions
of traffic engineers, crosswalk markings alone may do little to protect older
pedestrians from being struck by a motor vehicle as they cross the street
at an urban intersection. In fact, we found that the presence of crosswalk
markings was associated with increased risk overall, even after controlling
for the amount of pedestrian traffic, vehicular traffic, and other site characteristics.
However, this association varied significantly according to the system of
traffic regulation on the street being crossed. When no traffic signal or
stop sign was present to control traffic flow, marked crosswalks were associated
with a 3.6-fold increase in risk. At intersections with a stop sign or traffic
signal, there was virtually no association between presence of markings and
pedestrian–motor vehicle collision risk.
Several study limitations should be borne in mind. First, this was an
observational study, not a controlled experiment. We attempted to measure
and control for several relevant factors, but confounding by other unmeasured
site characteristics cannot be ruled out. Moreover, despite the matching on
neighborhood and street classifications, pedestrian flow and vehicular traffic
flow emerged in the analysis as important confounding factors, and both characteristics
are subject to measurement error. Pedestrian and vehicle counts observed during
a limited time interval on the same weekday and time of day as the pedestrian–motor
vehicle collision may be imprecise as indicators of long-term use levels.
However, a sensitivity analysis suggested that measurement error in pedestrian
and vehicular traffic flow is unlikely to explain fully the observed association
between pedestrian–motor vehicle collision risk and presence of crosswalk
A second limitation is that the study was restricted to pedestrian–motor
vehicle collisions involving a pedestrian aged 65 years or older, and the
findings may not necessarily apply to other age groups. Nonetheless, older
pedestrians are a known high-risk group, and it is plausible that similar
mechanisms may apply to other vulnerable groups. Third, the study had limited
statistical power to detect differences in the effects of specific crosswalk
marking patterns or effects confined to subgroups of sites. Fourth, the study
involved 6 cities in Washington and California, and generalizability to other
urban areas is uncertain.
Our findings agree with those of Herms,15 who
studied 400 unsignalized San Diego intersections at which 1 marked and 1 unmarked
crosswalk extended across the same street. During a 5-year period, nearly
6 times as many pedestrian–motor vehicle collisions occurred at marked
crosswalks. Pedestrian volume was measured at a 10% sample of sites and was
about 3-fold greater at marked crosswalks, suggesting that differences in
pedestrian use could not account fully for the difference in risk. Zegeer
et al16 used a study design involving 2000
unsignalized sites and found that pedestrian–motor vehicle collision
rates, while generally low, were higher at marked than at unmarked crosswalks,
particularly on wider and busier streets.
In contrast, Knoblauch et al14,25 compared
762 intersections and mid block locations at which a pedestrian–motor-vehicle
collision had occurred with a stratified sample of 495 control locations in
the same 5 cities. They found that unmarked crosswalks were associated with
increased risk of pedestrian–motor vehicle collision when site comparisons
were based on a "hazard score" calculated from pedestrian and vehicle flow.
That study combined intersection and mid block locations as well as signalized
and unsignalized intersections, it was not restricted to older pedestrians,
it used entire intersections or mid block locations (rather than specific
crossings) as the units of analysis, and it used no matching or multivariate
analysis to control for other confounding factors. These methodological differences
may have contributed to the discrepant results.
A possible explanation for the association we found is that marked crosswalks
may give older pedestrians a false sense of security, based on their questionable
assumptions about driver behavior. Tidwell and Doyle26 found
that nearly 40% of pedestrians incorrectly believed that traffic must stop
for a pedestrian who is on the curb waiting to cross at a marked crosswalk.
Washington and California laws require vehicles to stop when a pedestrian
is actually present in a marked crosswalk, but even then many drivers fail
to comply.27,28 Baker et al29 found that drivers involved in a sample of Maryland
pedestrian–motor vehicle collisions had worse-than-average driving records.
Saibel and colleagues28 found that driver compliance
was actually somewhat worse when the pedestrian was an older adult, and there
were more "near misses" at marked than at unmarked crossings. The limited
information available suggests that pedestrians are no less vigilant at marked
crossings than at unmarked locations.14,30 Oxley
et al31 found that older pedestrians tended
to be more cautious than younger ones by waiting for longer gaps in traffic,
but that this safety advantage was more than offset by their slower walking
We found that the association between presence of a marked crosswalk
and increased pedestrian–motor vehicle collision risk was essentially
confined to sites where no traffic-control device was present to restrict
the flow of vehicles. This result emerged from subgroup analyses and was not
hypothesized in advance, so it should be interpreted with caution. Nonetheless,
older pedestrians, who have slower walking speeds,32 may
be more vulnerable at crossing locations where vehicles can normally proceed
unimpeded. The pedestrian may venture into the street at a sanctioned location
and have the legal right of way, but drivers may be accustomed to proceeding
through such an intersection without stopping. The pedestrian's safety depends
heavily on driver alertness and compliance. In contrast, when a traffic signal
or stop sign is present, pedestrians have a much stronger guarantee that traffic
will stop and allow them safe passage.
Traffic engineers have been found to hold widely varying opinions about
the effectiveness of crosswalk markings, resulting in considerable variation
in policies and practices among jurisdictions.33,34 Additional
research may be needed to eliminate uncertainty about the safety and effectiveness
of crosswalk markings as a preventive measure. Findings from our study suggest
that measuring and controlling for pedestrian flow, vehicle flow, and signalization
pattern are important in future studies. Ultimately, controlled intervention
studies may be needed to establish causes beyond reasonable doubt.
In sum, we found that marked crosswalks at urban street crossings without
a traffic signal or stop sign were associated with elevated risk of pedestrian–motor
vehicle collision to older pedestrians. This information may be useful to
traffic engineers for setting policies on the placement and maintenance of
crosswalk markings. Older pedestrians may wish to be especially cautious when
crossing the street at high-risk locations.