TRANSPORTATION RESEARCH RECORD 1168

A Study of Driver Noncompliance with

Traffic Signals
STEWART R. GORDON AND H. DOUGLAS ROBERTSON

rate increased by a factor of five when sign configurations

There is a perception among traffic engineers that driver non-
compliance with traffic control devices is a significant problem. (a symbol only, instead of a symbol and message) were
Summarized in this paper are the results of a study of driver changed (4).
noncompliance at 12 signalized intersections in the Wash- The traffic control device violations may result from the
ington, D.C., metropolitan area to determine whether driver combined effects of human behavior characteristics and related
noncompliance ls a problem and to define Its relationship to traffic operational characteristics. Such human factors might
intersection operational characteristics and roadway features. include the driver age, vision, and perceived travel time. High-
The results Indicated that driver noncompliance is a problem
that requires attention. It was found that higher violation rates way geometrics and such traffic operation characteristics as
occurred at Intersections with low annual average daily traffic volume, type of regulatory control, and speed may also affect
volume levels. These high violation rates were predominant on driver noncompliance.
one-lane approaches during the off-peak hours. A correlation In order to develop solutions, however, the problem of driver
analysis indicated moderately high associations between high noncompliance must be defined in terms of where, when, how,
traffic signal violation rates and low traffic volumes. This how much, how serious, and why. The objective of this study
research study recommends that drivers, local police, and local
was to determine the magnitude of driver noncompliance with
traffic engineers be informed that driver noncompliance with
traffic signals is a problem with potential safety consequences traffic signals at intersections as it related to roadway features
and should be addressed through education, increased enforce- and traffic operational characteristics.
ment, and the application of sound engineering principles. Driver noncompliance with traffic signals was studied at 12
intersections in Virginia, Maryland, and the District of Colum-
Traffic engineers have expressed a growing concern over the bia during June and July of 1986. Drivers were observed under
lack of driver compliance with traffic control devices in recent various operational conditions at signalized intersections dur-
years. Driver noncompliance with traffic control devices, es- ing the peak and off-peak day and nighttime hours. Violation
pecially regulatory control devices, has been increasing signifi- frequencies, operational characteristics, and roadway features
cantly over the years. Drivers' apparent disregard for and were recorded and analyzed. The results of this study, which
perhaps lack of confidence in traffic control devices has been was sponsored by the Federal Highway Administration's Grad-
recognized by the American Association of State Highway and uate Research Fellowship Grant program, are documented fully
Transportation Officials' (AASHTO) Highway Subcommittee elsewhere (5).
on Traffic Engineering. The AASHTO Standing Committee on
Highways conducted a nationwide survey in October 1985 to METHODOLOGY
determine if a driver noncompliance problem exists and, if so,
The experimental plan for determining the magnitude of driver
what can be done to correct it. Surveys were sent to each state
noncompliance with traffic signals was made up of five parts:
and the District of Columbia traffic engineer. They were asked
to comment on motorist noncompliance with traffic control 1. Measures of effectiveness (MOEs),
devices. To the question "Is traffic control device non- 2. Sample size,
compliance a significant problem?" 34 of the 46 respondents 3. Site selection criteria,
said "yes" while 12 states answered "no" (1 ). 4. Data collection procedures, and
Other research studies in recent years have indicated that 5. An analysis plan.
specific traffic control devices are being violated more than
others. For example, studies have found that the violation rate Measures of Effectiveness
with stop signs has been increasing linearly since 1935 (2). In
another study it was found that the violation rate (i.e., not The principal MOEs were the four driver violation types de-
stopping when required) increased from 0.1 percent to 0.6 fined as follows. Each type was expressed in terms of hourly
percent when the signal configuration changed from regular frequency and rate-violations per 100 vehicles.
operation to flashing red (J). In another instance, the violation
• Running the red signal (RUNRED)-the number of
through and left-turning vehicles entering the intersection past
S. R. Gordon, Edwards and Kelcey, Inc., 70 South Orange Avenue, the near curb line after the onset of the red signal indication.
Livingston, N.J. 07039. Current affiliation: Garmen Associates, 150
River Road, Building E, Montville, N.J. 07045. H. D. Robertson, • Right-tum-on-red (NOSTOP)-the number of right turn-
University of North Carolina at Charlotte, Civil Engineering Depart- ing vehicles not corning to a complete stop during the red
ment, Charlotte, N.C. 28223. signal indication.
2 TRANSPORTATION RESEARCH RECORD 1168

• No right tum on red (NRTOR)-the number of vehicles violation rates or frequencies varied according to volume lev-
making a right tum on red when there is a "No Tum on Red" els. Although AADT cannot indicate volume variations
sign. throughout the day, it was recognized that most of the travel
• Total violations (TOTVIOL)-the total number of vehi- occurred during the 7 :00 a.m. to 11 :00 p.m. study time period.
cles that violate the red signal indication (i.e., the sum of the A balanced number of approach lanes was sought, with the
three previous MOEs). selected sites consisting of 21 one-lane approaches and 27 two-
lane approaches, for a total of 48 approaches.
Violation rates were calculated per 100 vehicles for each The other regulatory control of interest was the "No Tum On
MOE-RUNRED, NOSTOP, NRTOR, and TOTVIOL--based Red" sign posted at a signalized intersection. Posted speed
on its corresponding traffic volume-through left turn limits were controlled to reduce the variation of vehicle ap-
(THLTVOL), right tum (RTVOL), right tum (RTVOL), and proach speeds that might influence driver noncompliance. All
total volume (TOTVOL), respectively. intersections had approach speed limits posted at 25 mi/hr,
except for one intersection with a 30 mi/hr limit.
Sample Size With respect to jurisdiction, Arlington and Fairfax counties
in Virginia; Montgomery County, Maryland; and Washington,
Twelve signalized intersections were selected from a popula- D.C., were represented by three, three, two, and four intersec-
tion of 30 intersections. The sample intersection population tions, respectively.
was limitl".d hP~a11~P- of thP- availahle infonnation provided by
each transportation agency, a short data collection time period
Data Collection Procedures
of 2 months, and the manual data collection effort by only one
observer. Intersection inventory data were obtained by calling and visit-
The minimum sample of observations (380) for each inter- ing the transportation agencies in each jurisdiction and by
section was found using a standard statistical estimating pro- visiting the sites. Operational data were collected for 2-hr
cedure that determined the minimum number of observations periods during each of the morning peak, midday off peak,
needed to meet a desired level of confidence and permitted afternoon peak, and evening off-peak time frames, a total of 8
error (6). hr per intersection approach. Data collection for the 12 inter-
sections was performed between June 23 and July 25, 1986.
Site Selection Criteria Dry pavement conditions existed throughout each observation
time period except for one 2-hr off-peak period at one
The intersection location and characteristics used as criteria in intersection.
the selection of sites included the number of approaches (four Traffic volumes were counted on a rotation of 5-min (or
only), number of approach lanes (one or two lanes), average nearest multiple signal cycle length) time intervals per ap-
annual daily traffic (AADT) (high, medium, and low), right proach for 2 hr. This provided a 15-min sample count per hour
tum on red (permitted or prohibited), approach speed limits (25 per approach. While counting traffic, all approaches were
to 30 mi/hr), and jurisdiction in which the intersection was monitored for traffic violations. Thus, recorded violations rep-
located. resented total counts, while hourly volumes were obtained by
The results of applying the site selection criteria are given in expanding the 15 min of sample counts for each approach. Data
Table 1. The primary site selection criterion, AADT volume collection quality was maintained by having only one observer
level, was used to stratify the intersection approaches into low, collect all data.
medium, and high AADT. Low was fewer than 7,500 entering
vehicles per day, medium was 7 ,500 to 15,000, and high was
Analysis Plan
greater than 15,000. These AADT volume levels were selected
based on the volume ranges in the intersection population. To quantify driver noncompliance with traffic signals, the num-
Ranges of AADT volume levels were used to determine if ber of violations and violation rates for each MOE were

TABLE I CHARACTERISTICS OF SAMPLED INTERSECTIONS

Intersection
No. No. of Approach Lanes/AADT Jurisdiction Signal Location Land Use
I I/low AADT; 2/low AADT Washington, D.C. Post comer Office
2 2/medium AADT and high AADT Washington, D.C. Post comer Commercial
3 I/low AADT; 2/low AADT Washington, D.C. Post comer Commercial
4 2/medium AADT and high AADT Washington, D.C. Post comer Commercial
5 2/medium AADT Arlington Co., Va. Overhead Commercial
6 I/low AADT and medium AADT Arlington Co., Va. Overhead Residential
7 I/low AADT; 2/low AADT Arlington Co., Va. Cantilever arm CBna
8 I/low AADT and medium AADT Fairfax Co., Va. Overhead Office
9 I/low AADT; 2/high AADT Fairfax Co., Va. Cantilever arm Commercial
IO I/low AADT; 2/high AADT Fairfax Co., Va. Overhead Commercial
Il 2/medium AADT Montgomery Co., Md. Cantilever arm CBD
12 l/low AADT; 2/low AADT Montgomery Co., Md. Post corner Residential

aCBD = central business district.

Gordon and Robertson 3

examined to assess the overall noncompliance problem. Inter- TABLE 3 OVERALL VIOLATIONS
section characteristics such as the number of approach lanes,
Measures of Effectiveness
grades, stop bar distance to the near curb, signal location (i.e.,
post comer, overhead), land use, cycle lengths, curb parking, NOS TOP NRTOR RUNRED TOTVIOL
and traffic volume levels were examined in conjunction with Violation
violation occurrence. The resulting relationships indicated sum 767 64 469 1,300
which independent variables (i.e., time, volumes, number of Range of
lanes, etc.) were related to the dependent variables (MOEs). violations
per hour 0 to 23 0 to 8 0 to 11 0 to 23
These relationships were examined for each signalized inter- Percent of at
section and for all 12 intersections combined. Pearson's cor- least one
relation coefficients were also calculated to determine if the violation
MOEs were associated with traffic volumes. These analyses per total
observed
established where, when, and how, as well as the magnitude of
hours 74 48.7 54.3 85.9
driver noncompliance with traffic signals. No. of
violations
per
FINDINGS intersection
approach
The MOEs will be discussed first with respect to violation per hour 2.56 0.84 1.25 3.46
frequency and rate and then by examining the intersection
characteristics at intersections with similar violation rates.
Characteristics of the sample of intersections used in the anal- On Red" sign more often than drivers who properly executed
ysis included AADT volume levels, number of approach lanes, RTOR by coming to a complete stop before turning right.
intersection jurisdiction, traffic signal location, and primary The violation rates by MOE for each intersection are given
land use and are summarized by intersection in Table 1. The in Table 4. As expected, NOSTOP violation rates were signifi-
distribution of these characteristics by number of approaches cantly higher at all sites than the other MOEs. The propensity
and numbe.r of intersections is shown in Table 2. for many drivers to not come to a full stop before turning right
on red has been recognized since the implementation of RTOR.
TABLE 2 DISTRIBUTION OF INTERSECTION The most hazardous violation, RUNRED, was observed for
CHARACTERISTICS about one of every 200 through vehicles entering the intersec-
tion. Overall 1.25 violations occurred for every 100 vehicles
No. of No. of
Approaches Intersections entering the intersection.
AADT volume levels
Low 24 TABLE 4 TRAFFIC VIOLATION RATES BY MEASURES OF
Medium 16 EFFECTIVENESS AND INTERSECTION
High 8
No. of lanes Overall Rate
Inter-
One 21
Two 27 section NOS TOP NRTOR RUNRED TOTVlOL
Jurisdiction 1 12.20 2.28 0.36 0.64
Montgomery Co. 2 2 3.14 0.23 0.44
Arlington Co. 3 3 7.13 0.99 0.61 1.18
Fairfax Co. 3 4 4.46 1.75 0.32 0.95
District of Columbia 4 5 5.68 0.59 1.64
Signal location 6 3.91 1.09 1.40
Post corner 5 7 9.25 0.88 2.64
Overhead spanwire 4 8 17.22 0.65 4.06
Cantilever arm 3 9 11.22 0.27 1.16
Primary land use IO 7.00 0.42 0.74
CBD 2 11 2.56 0.95 1.22
Commercial 6 12 17.87 0.93 3.07
Residential 2 Mean
Office 2 rate 7.59 2.02 0.52 1.25
NoTEs: All rates per 100 vehicles. Dashes indicate not applicable.

Frequency Violation and Rate

Intersections with similar violation rates were grouped and
Presented in Table 3 for each MOE are the total violations for their characteristics examined. Intersections were considered to
all observation hours, the range of violations per hour, the have similar violation rates if their rates were within plus or
percentage of time that at least one violation per hour was minus 1.5 of one another. This range was chosen arbitrarily
observed, and the average number of hourly violations per based on the researchers' judgment.
intersection approach. The number of violations per hour per The intersection groupings by overall violation rate for each
approach was the greatest for NOSTOP, while NRTOR was the MOE are displayed in Table 5. As an example, for the
lowest. This result implies that drivers obeyed the "No Turn NOSTOP MOE, 9 of the 12 intersections were combined into
4 TRANSPORTATION RESEARCH RECORD 1168

TABLE 5 INTERSECTION GROUPINGS BY OVERALL tions. No patterns were evident except that Group I had the
VIOLATION RATE FOR EACH MEASURE OF lowest violation rates with low AADTs in common, while
EFFECTIVENESS Group IV had the highest violation rates with mostly medium
Group AADTs. These results suggest that low violation rates are
Measure of associated with low AADTs, while higher violation rates are
Effectiveness II III IV associated with medium AADT volume levels.
NOSTOP 1, 9 2, 4, 6 3, 10 8, 12
NRTOR 1, 11
Number of Approach Lanes
RUNRED 1, 4, 10 11, 12 3, 5, 8 2, 9
TOTVIOL 1, 10 3, 9, 11 5,6 8, 12
The violation rates by the number of approach lanes were
calculated for an 8-hr period per intersection and are shown in
four groups, with each group having NOSTOP violation rates Table 6. Some intersections had four approaches with two
within ±1.5 of one another. As is readily apparent, not all lanes, four approaches with one lane, or a combination of one-
intersections were included in the groupings. This was par- and two-lane approaches.
ticularly the case with the NRTOR MOE. A comparison of the The mean rate for all MOEs indicated that one-lane ap-
information in Tables 1, 4, and 5 yields the following proaches had twice as many violations as two-lane approaches,
observations: except for the NRTOR violation rate. This trend was not con-
sistent, however, for inter.<iectiorn; inciivici1rn lly. A further exam-
• Of the four groups of the NOSTOP MOE overall rates, ination of the NOSTOP rates revealed that one extreme value
three groups (I, III, IV) had low AADT volume levels in skewed the results, and when it was removed, the mean rate for
common with higher violation rates (rates greater than or equal one-lane approaches became 9.84 compared to the previous
to 7.00), and all groups within themselves had a similar number value of 5.89 for two-lane approaches.
of approach lanes. Only one group, Group II, had medium and The RUNRED mean violation rate indicated twice as many
high AADT volume levels associated with lower violation violations for one-lane versus two-lane approaches. Upon fur-
rates. These results suggest that low AADTs and the number of ther examination of each intersection individually, this trend
approach lanes could be correlated with high NOSTOP was not consistent.
violations. Looking at the overall TOTVIOL violation rates for one- and
• Although similar NRTOR violation rates occurred for In- two-lane intersection approaches, the one-lane approaches had
tersections 1 (2.28) and 11 (2.56), these intersections did not a higher violation rate with the exception of two intersections.
have similar characteristics. One had an approach with a downhill grade of 7 percent. Thus,
• Intersections with similar RUNRED violation rates were vehicles tended not to stop completely before they turned right
combined into four groups; however, there were no intersection (NOSTOP violation). This approach also had low AADT vol-
characteristics that dominated every group. It should be noted ume levels. The other intersection also had low AADT volume
that Group IV had the lowest RUNRED violation rate with levels, good sight distance, and the majority of its violations
high AADTs on two-lane approaches for each intersection. were NOSTOP. Thus, both intersections had geometric condi-
These results suggest that no single characteristic with all tions conducive to high NOSTOP violations and low AADT
approaches combined for an 8-hr period have an effect on volume levels, thus less chance for conflicts.
RUNRED violations.
• Looking at all four groups for the TOTVIOL violation Intersection Jurisdiction
rate, the nine intersections were characterized by low, medium,
and high AADT volume levels, one- and two-lane approaches, The violation rates calculated by intersection jurisdiction in-
all three signal locations, all land use types, and all jurisdic- cluded all four approaches for an 8-hr period. The results

TABLE 6 VIOLATION RATES BY NUMBER OF APPROACH LANES

One Lane Two Lanes
Inter-
section NO STOP NRTOR RUNRED TOTVIOL NOS TOP NRTOR RUNRED TOTVIOL
1 10.45 1.91 0.38 0.68 16.51 2.86 0.34 0.56
2 3.14 0.23 0.44
3 87.42 1.66 4.49 0.49 0.99 0.31 0.36
4 4.46 1.75 0.32 0.95
5 5.68 0.59 1.64
6 3.91 1.09 1.40
7 6.11 0.83 1.78 19.69 1.61 10.94
8 17.22 0.65 4.06
9 11.95 0.24 2.12 11.12 0.27 1.08
10 9.96 0.00 2.67 4.44 0.45 0.56
11 2.56 0.95 1.22
12 19.31 1.01 2.66 16.68 0.77 3.73
Mean
rate 11.56 1.91 0.86 2.21 5.89 2.03 0.41 0.95
Norns : All rates per 100 vehicles. Dashes indicate not applicable.
Gordon and Robertson 5

TABLE 7 JURISDICTION VIOLATION RATES TABLE 8 SIGNAL LOCATION VIOLATION RATES

Mean Violation Rates Mean Violation Rates

Jurisdiction 1a 2b 3c 4d Signal Location 2b

Fairfax Co. 11.94 0.38 1.33 Post comer 4.83 1.72 0.51 0.94
Montgomery Co. 17.87 2.56 0.94 1.78 Overhead spanwire 7.44 0.63 1.50
D.C. 4.96 1.72 0.36 0.79 Cantilever arm 10.34 2.56 0.55 1.43
Arlington Co. 6.06 0.82 1.74
NoTEs: All rates per 100 vehicles. Dashes indicate not applicable.
NoTES; All rates per 100 vehicles. Dashes indicate not applicable. ONQSTOP.
ONOSTOP. bNRTOR.
bNRTOR. CRUNRED.
CRUNRED. dTOTVIOL.
drOTVIOL.
RUNRED, and TOTVIOL violation rates, respectively. These
showed that Montgomery County, Maryland, had the highest rates were all greater than the post corner signal locations. This
NOSTOP, NRTOR, RUNRED, and TOTVIOL mean violation could be due to the overhead spanwire locations having better
rates, whereas the District of Columbia had the lowest rates visibility than post corner locations, thus giving drivers a better
(Table 7). Fairfax County had the second highest NOSTOP view of the signal and an ability to react to conflicting traffic
mean violation rate, and Arlington County had the second appropriately.
highest mean violation rates for RUNRED and TOTVIOL.
Each jurisdiction has a different law concerning the violation of Time-of-Day Comparisons
a red signal indication as summarized below:
The NOSTOP mean violation rate was highest during the
• Maryland-Drivers are required to exercise "caution" nighttime hours, second highest during the off-peak hours, and
when entering an intersection when the traffic signal is yellow. third and fourth highest during the morning and evening peak
However, they are not legally bound to stop. hours, respectively (Table 9). These mean violation rates indi-
• Virginia-Motorists are required to stop at an intersection cate that the NOSTOP violations occurred more often during
if the signal has turned yellow and if they have what police call the off-peak hours. The NRTOR mean violation rate was also
ample time to halt. highest during the nighttime hours, with the off-peak hours
• District of Columbia-Drivers are legally in violation and second, evening peak hours third, and morning peak hours
subject to a $50 fine if they enter the intersection when the fourth. The RUNRED mean violation rate indicated that the
signal is yellow and if they had ample time to stop. The fine for evening peak hours had the highest violations. The morning
running a red signal is $75, the stiffest penalty in the region. peak hours were second, with off-peak and nighttime hours
third and fourth, respectively.
Traffic officials of all three jurisdictions indicated that the With all the violations combined (TOTVIOL), the nighttime
yellow interval is generally the same (4 sec) in all jurisdic- hours produced the highest overall mean violation rate. The off
tions (7). peak was second, with evening and morning peak hours fairly
The jurisdictions' different laws may have affected the viola- even. To examine these results further, each individual intersec-
tion rates. The District of Columbia had the lowest violation tion MOE violation rate was compared to the different 2-hr
rates and also the stiffest penalties in the region. On the other time periods.
hand, Montgomery County had the highest violation rates with At 7 of 11 intersections, both the nighttime and off-peak
a lesser red signal violation penalty (i.e., to only exercise hours had the greatest NOS TOP violation rates compared to the
"caution" when entering an intersection when the traffic signal peak hours. The other four intersections had the second highest
is yellow). Arlington County and Fairfax County, Virginia, had NOSTOP violation rates during the peak hours.
greater violation rates than D.C., but lower than those of Only four intersections prohibited turning on red at any time
Montgomery County. This could be a result of Virginia's lesser during the day. Two intersections had the greatest violation rate
violation penalty compared with D.C's. The collection of en- during the nighttime hours, while the other two intersections
forcement data was not within the scope of this study, thus no were greatest during the peak periods. Overall (all four inter-
conclusions can be drawn concerning the role of existing en- sections combined), the highest violation rates occurred at
forcement levels. nighttime. However, no time period dominated the NRTOR
violation rates.
Traffic Signal Location An examination of each intersection RUNRED violation rate
indicated that seven intersections had the highest violation rates
Of the three traffic signal locations, the cantilever arm signals during at least one peak period (morning or evening). Three
had the highest mean violation rates for the NOSTOP and intersections had both peak periods with higher rates than the
NRTOR (Table 8). The overhead spanwire signal location had off-peak periods. These results strongly suggest that most of
the highest RUNRED and TOTVIOL mean violation rates. The the intersections had the highest RUNRED violation rates dur-
post corner signal intersections had the lowest mean violation ing the peak periods.
rates for all MOEs. When overhead spanwire and cantilever The TOTVIOL mean violation rates suggested that all viola-
arm intersections were combined, they had mean violation rates tions combined occurred more often in the nighttime and off-
of 8.29, 2.56, 0.60, and 1.47 for the NOSTOP, NRTOR, peak hours. Eight out of the eleven intersections with nighttime
6 TRANSPORTATION RESEARCH RECORD 1168

TABLE 9 TIME-OF-DAY VIOLATION RATES

NO STOP NRTOR
Inter-
section Mominga Eveningb Off Peak" Nightd Mominga Eveningb Off Peake Nightd
1 12.54 1.97 2.46 2.21
2 2.75 2.75 1.78 5.48
3 3.35 3.17 17.55 9.43 1.92 0.00 0.89
4 4.08 2.99 4.60 14.09 0.56 1.05 2.09 7.23
5 1.65 4.11 5.79 18.01
6 3.49 3.50 2.83
7 8.98 8.58 10.12
8 15.42 14.38 15.61 39.71
9 18.92 6.30 10.18 20.39
10 6.80 4.00 11.03 7.45
11 2.84 2.12 1.72 4.79
12 26.25 10.28 21.95 16.47
Mean
rate 6.51 5.43 8.54 13.14 1.34 1.61 1.79 5.89
NoTES: All rates per 100 vehicles. Dashes indicate not applicable.
aMomine 2-hr p(lak.
b£ vcning 2-hr penk.
CQff peak 2 hr.
dNighttime 2 hr.

observations had the highest violation rates. One intersection TABLE 10 OVERALL VIOLATION RATES RELATED
was not monitored at night, but it had the highest violation rate TO AADT VOLUME LEVELS FOR EACH MEASURE OF
during the off-peak hours. EFFECTIVENESS

AADT Volume Level Categories

Similarities Among Intersection Characteristics Measure of
Effectiveness Low Medium High
As previously discussed, intersections were grouped according NOS TOP High rate Low rate Low rate
to similar violation rates. Once these groups were established, NRTOR a a a
other intersection characteristics such as number of approach RUNRED High rate Low rate Low rate
lanes, land use, signal locations, and AADT volume levels TOTVIOL High rate Low rate Low rate
were examined to determine if any relationships existed. The aNo association noticed.
only similar intersection characteristic that emerged was the
intersection approach AADT volume level. higher rates (rates greater than 0.50). The low violation rate
The results presented here are based on intersections intersections had mostly medium and high AADT volume
grouped by their similar overall violation rates (rates calculated levels (60 percent of the approaches), whereas the high viola-
for an 8-hr period with all four approaches). It should be noted tion rate intersections had mostly low AADT volume levels.
that these violation rates are not directly related to AADT Thus, these results imply that low AADTs can be associated
volume levels (i.e., the violation rates were calculated from the with higher violation rates, and medium and high AADT's can
expanded traffic volume counts, not AADT volume levels). be associated with low violation rates.
Illustrated in Table 10 is the relationship between each The TOTVIOL MOE resulted in nine intersections with
MOE's overall violation rates (high or low) and the AADT similar violation rates that were placed into four groups. Seven
volume levels. The NOSTOP MOE resulted in 9 out of 11 of the intersections had violation rates greater than 1.00 (high
intersections extracted into four groups. Six of the nine inter- rate), and two intersections had violation rates less than 1.00
sections had high violation rates (greater than or equal to 7.00) (low rate). The seven intersections with high violation rates had
while three intersections had violation rates less than 7 .00. The 50 percent of the approaches with low AADTs. The two low-
combined six intersections with high violation rates had 75 rate intersections had 75 percent of the approaches with low
percent of the approaches with low AADT volume levels. The AADTs. Since the majority of intersections have high rates
three intersections with low violation rates had 83 percent of with low AADTs, this suggests that high rates are more often
the approaches with medium or high AADT volume kvds. associated with low AADTs.
Thus, the high NOSTOP violation rates were associated with
low AADTs and low violation rates with medium and high Violations Versus Volume Correlations
AADT volume levels.
The NRTOR MOE resulted in one group with two of four A useful technique for studying the volume versus violation
intersections. These two intersections did not have similar rate relationship is Lo establish a correlation between these
AADT volume levels. Therefore, no association existed be- variables. Descriptive statistics were calculated for the viola-
tween NRTOR and AADT volume levels. tion rate and volume distributions and included means, stan-
The RUNRED MOE resulted in four groups. There were 10 dard deviations, and a measure of skewness. The data were
of 12 intersections within these four groups. Five intersections aggregated from hourly observations to an 8-hr count by ap-
had low rates (rates less than 0.50), while five intersections had proach. Thus, there were 48 observations or cases. This was
Gordon and Robertson 7

done be{;ause of the small amount of variance using hourly correlations indicate that as traffic volumes increased, the vio-
observations. Based on standard deviation values and skew- lation rate de{;reased. A t-test tested the null hypothesis. Using
ness, the 48 observations were not normally distributed. For the t-test, all correlated variables' null hypotheses (i.e., that the
nonnormal data, either a nonparametric test, which is distribu- population correlation coefficient is zero) were reje{;ted at the
tion free, may be used or the data may be transformed so that 90 percent or greater confidence level, except that of NRTOR
parametric tests may be applied. The use of parametric tests is versus RTVOL.
generally more desirable because they are more powerful than If the correlation coefficient is squared, inferences can be
nonparametric tests. Consequently, the data were transformed. made about the total variation explained. For example, right
The transformation of data raised a contentious issue. On one turn volume explained 10.4 percent of the variation in
hand, some statisticians argue that transforming data is nothing NOSTOP. The THLTVOL explained 30.8 percent of the varia-
more than "fudging" the data to fit the model and that the tion in RUNRED, and TOTVOL explained 48.2 percent of the
implications of transforming data are not fully understood. On variation in TOTVIOL. These coefficients of determination
the other hand, other statisticians argue that all measurement indicated that volume is not significant in explaining the
systems are arbitrary; hence transformed data are just as valid amount of variation in the violation rate for NOSTOP but are
as untransformed data. This latter group has no reservation in fairly significant for RUNRED and TOTVIOL.
using a transformation to normalize data if normally distributed Additional correlations for violation rates versus volume
data are required (8). Thus, since statisticians have used trans- were calculated by the number of approach lanes (see Table
formation processes to normalize their data, and a normal 12). The NOSTOP versus RTVOL had a 5.2 percent de{;rease
distribution is required for parametric testing, it was applied in the amount of variance explained by one-lane approaches
here. versus the one- and two-lane approaches combined coefficient
For a transformation to be applied, the skewness measure of determination (10.4 percent). However, there was an in-
must be examined. If the data are positively skewed, a transfor- crease of 6.3 percent variance explained between one-lane and
mation is needed that will reduce values in the upper tail by a two-lane approaches. The correlation coefficient on two-Jane
greater amount than those located in the lower tail. This is approaches is fairly high with a 90 percent confidence level,
accomplished by taking either the square root (fairly moderate indicating a stronger association between NOSTOP versus
transformation) or the logarithm (more radical transformation) RTVOL on two-lane approaches, as opposed to one-lane
for each observation. Both of these transformation processes approaches.
were applied for the 48 hr observations. The logarithm transfor- There was no correlation coefficient for NRTOR versus
mation process gave the better result. RTVOL on one-lane approaches since there were only two
Given in Table 11 are the logarithmic transformations for an observations. On two-lane approaches, the correlation coeffi-
8-hr period by approach. These results indicate low standard cient was -0.2313. This correlation was not significant.
deviations versus the mean values for most variables, and the The RUNRED versus THLTVOL correlation was the least
skewness test for normality resulted in all variables falling for one-lane approaches and was not significant. The two-lane
within the Beta 1 critical limits with 98 percent confidence approach correlation (-0.6068) was much higher than that for
limits. The NRTOR MOE did not have B 1 critical limits, but as the one-lane approach. The difference in the amount of vari-
the sample size decreases, the B 1 critical limits increase. ance explained between one- and two-lane approaches was
Therefore, the NRTOR skewness value of 0.08 fell within the 33.8 percent. The two-lane approaches had higher correlations
B 1 critical limit. Since logarithmic transformations produced than with both one- and two-lane approaches combined and
normal data, Pearson's correlations were calculated. explained 6.0 percent more variation.
The total violation versus total volume correlations indicated
that two-lane approaches had a higher value compared with
Correlation Analysis Results those of one-lane approaches. These values were -0.6898 and
-0.4242, respectively. The one-lane approach correlations were
The correlation coefficients are presented in Table 12. They significant at the 95 percent confidence limit while the two-lane
ranged from -0.0800 for NRTOR versus RTVOL to -0.6941 approaches were significant at 99.5 percent. It can be con-
for TOTVIOL versus TOTVOL for all 48 observations. These cluded that two-lane approaches had a greater degree of asso-
ciation between TOTVIOL and TOTVOL than one-lane ap-
TABLE 11 LOGARITHM TRANSFORMATION FOR AN 8-HR
proaches, but not with both one- and two-lane approaches
COUNT BY APPROACH combined.

8-Hr
CONCLUSIONS
Observa- Standard Skew- Bl
Variable tions Mean Deviation ness (2%)
The violation frequencies and rates in the pre{;eding sections
NO STOP 43 0.838 0.361 --0.118 0.71 described quantitatively the magnitude of driver non-
NRTOR 10 0.280 0.258 0.121
compliance with traffic signals at signalized intersections as it
RUNRED 46 -0.266 0.340 --0.137 0.64
TOTVIOL 48 0.128 0.389 0.006 0.64 related to traffic operational characteristics and roadway
RTVOL 48 0.325 0.333 --0.058 0.64 features.
THLTVOL 48 1.128 0.381 --0.078 0.64 Considering each MOE separately, the NOSTOP violation
TOTVOL 48 1.214 0.344 --0.062 0.64 ra.te was greatest on low AADT volume intersection ap-
NoTE: Dash indicates no table value. proaches. This was also supported by its rates being highest
8 TRANSPORTATION RESEARCH RECORD 1168

TABLE 12 PEARSON'S CORRELATION TABLE FOR VIOLATIONS VERSUS VOLUMES

48 Observations One-Lane Approach Two-Lane Approach

Correlated NOS TOP NRTOR RUNRED TOTVIOL NOSTOP NRTOR RUNRED TOTVIOL NOSTOP NRTOR RUNRED TOTVIOL
Variables RTVOL RTVOL TIILTVOL TOTVOL RTVOL RTVOL THLTVOL TOTVOL RTVOL RTVOL TIILTVOL TOTVOL
No. of
observa-
tions 43 10 46 48 21 2 19 21 22 8 27 27
Pearson ' s
correla-
tions --0.3222 --0.0800 --0.5551 --0.6941 --0.2279 --0.1731 --0.4242 --0.3391 -0.2313 --0.6068 --0.6898
R"'*2 0.1038 0.0064 0.3081 0.4818 0.0519 0.0300 0.1800 0.1150 0.0535 0.3682 0.4758
I-test Reject N.reject Reject Reject N.reject N.reject Reject *Reject N.reject Reject Reject
NoTES: R "'*2 =coefficient of determination. Reject= rejected at 5%, 2.5%, and 0.5% levels. N.reject =not rejected. *Reject= rejected at 10% level. Dashes
indicate sample size too small.

during nighttime and off-peak hours (low volumes during these professional judgment. It is also recommended that signalized
time periods). When correlated with right tum volumes, the intersections be monitored periodically for driver non-
NOSTOP violation rate had significant correlations for all compliance levels. As traffic volume conditions change, further
approaches combined and for two-lane approaches. These re- intersection improvements may be justified.
sults may be attributed to a lesser risk of conflict events (low
traffic volume levels), better sight distances on two-lane ap- ACKNOWLEDGMENTS
proaches, and possibly lower enforcement levels at off-peak
and nighttime hours. Special appreciation is extended to the people and organiza-
The NRTOR violation rate constituted a relatively small tions who provided valuable input, support, and advice through
sample size, and therefore, no associations were found between the completion of this paper. Grateful acknowledgment is also
it and right-tum volumes. This MOE was found to occur most expressed for the knowledge and advice given by the following
often at nighttime. thesis committee at the University of North Carolina at
The RUNRED violation rate was highest during the evening Charlotte: Ellis King, Tom Faulkner, and Wayne Walcott. A
peak hours. It was also found that these high rates occurred for special thanks is offered to academic advisor Doug Robertson
mostly low AADT approaches. Relatively high negative cor- for his dedicated support, guidance, and time spent in review of
relations indicated that as the traffic volumes increased, the the thesis. Thanks are also extended to the Turner Fairbank
RUNRED violation rate decreased. These results are expected Highway Research Center, Federal Highway Administration,
since more conflict opportunities exist during high levels of where this study was initiated, for their support and sponsor-
traffic volumes. ship, and to Edwards and Kelcey, Inc., for the support of that
With all violations combined, TOTVIOL resulted in the organization.
highest significant correlations, most violations occurred at
nighttime, and the highest rates occurred on low-AADT- REFERENCES
volume intersection approaches.
These results suggest that driver noncompliance with traffic 1. T. Hicks. Traffic Control Device Compliance-Summary of
AASHTO Actions. AASHTO Highway Subcommittee on Traffic
signals exists at low-volume intersections during off-peak and Engineering, Seattle, Wash., Oct. 1985.
nighttime hours on one-lane approaches. Does this suggest that 2. W. R. Stockton, R. Q. Brackett, and J. M. Mounce. Stop, Yield and
driver noncompliance with traffic signals is a significant prob- No Control at Intersections. Report FHWA-RD-81-084. Texas
lem? With all violations combined, the answer is yes. An Transportation Institute; FHWA, U.S. Department of Transporta-
tion, June 1981.
overall TOTVIOL violation rate of 1.25 vehicles per 100 vehi-
3. B. Berriott and T. Rorabaugh. A Study of Clearance Intervals,
cles is fairly significant with the highest MOE violation rate Flashing Operation, and Left Turn Phasing al Traffic Signals.
being NOSTOP. Does this violation type pose potential safety Summary Report FHWA-RD-78-46, Vol. 1. FHWA, U.S. Depart-
hazards? What can be done to correct it? The answers are the 3 ment o f Transportation, May 1980.
Es, Education, Enforcement, and Engineering. 4. M. A. Kraft Effectiveness of lnlerna1ional Symbol Sign. Bureau of
Traffic Engineering and Operations, Traffic Planning and Design
Efforts should be made to inform the driver, local police, and Division, District of Columbia Government, May 1971.
local traffic engineers that driver noncompliance is a problem 5. S. R. Gordon. Driver Compliance with Traffic Signals. M.S. thesis.
and should be addressed. Improvements such as higher en- University of North Carolina at Charlotte, May 1987.
forcement levels at low traffic volume intersections, stiffer 6. P. C. Box and J. C. Oppenlander. Manual of Traffic Engineering
violation penalties, and educating the public of what constitutes Studies, 4th ed. Institute of Transportation Engineers, Inc.,
Arl ington, Va., 1978.
a traffic signal violation should be considered. Engineering 7. K. Barker and S. He.ilbronner. Running on Red: Violations on the
improvements might include removing unnecessary informa- Rise. Washington Post, June 3, 1985, Sec. A, p. AS.
tional, or regulatory control, devices adjacent to intersection 8. G. B. Norcliff. lnferential Statistics for Geographers: An Introduc-
approaches , better signal timing and network progression, tion, 2nd ed. Hutchinson & Co., Ltd., London, England, 1982.
lighted intersections, and intersection geometric improvements.
Not all of these improvements are applicable for every sig- Publication of this paper sponsored by Committee on User Information
nalized intersection; therefore, the traffic engineer must use Systems.