ITE Journal - June 2021 - 36

demonstrate the probability of each crash type occurring at each
traffic control device. This method has been proven effective at
predicting crashes based on collision type.7 As crashes are assumed
to be independent of one another, the necessary independence of
irrelevant alternatives (IIA) assumption holds true.
When MLR is applied, a log-linear model is fitted with
coefficients of zero for a specified response variable category. This
category acts as the baseline outcome. In this study, a crash location
designated with " no control " (or no traffic control device involved)
was set as the baseline outcome for the response variable of traffic
control. The crash type predictor variable baseline category was rear
end crashes as they occurred at the highest frequency and tend to
result in relatively low injury severity.8 All statistical analysis was
implemented by the mulitnom() function in the " nnet " package of
R version 3.4.3.9
Figure 1. Number of crashes by traffic control device in analysis.

Traffic Control Devices

No Control

519,142

Traffic Signal

418,947

Stop Sign

129,949

School Zone 13,156
Police/Human Control 2,823
Yield Sign 931
Other

725,953
0

500,000
1,000,000
Number of Crashes

Table 1. Summary of crash type variables in analysis.
Variable

Number
of Crashes

Percent

Rear end

567,841

31.4%

Angle

225,386

12.4%

Animal

179,144

9.9%

Sideswipe (meeting same direction)

167,077

9.2%

Sideswipe (meeting opposite direction)

32,598

1.8%

Backing

27,148

1.5%

Parked vehicle

17,578

1.0%

Head-on

13,948

0.77%

Pedestrian

6464

0.36%

Pedal cycle

4584

0.25%

Other

569,130

31.4%

36

J u ne 2021

i te j o urnal

Results and Discussion
The results are presented as exponential values of the logit coefficients. The relative risk ratios of each traffic control device are
presented in Table 2 based on crash type and Table 3 based on
the speed limit and weather condition. These relative risk ratios
are interpreted as odds. For example, the odds of an angle crash
occurring at a traffic control signal as compared to a rear-end
crash occurring at a location without a traffic control device would
be 53.7 percent higher (1.537 minus 1). Due to its continuous
nature, the speed variable unit is in relation to a 1-mph (1.61 km/
hr) increase in the speed limit. Results of the " other " categories
were not included in the results as they were a mixture of different
outcomes, providing no useful insight. Table 4 presents crash types
as probabilities at each traffic control device based on a separate
MLR model including the same variables at a common U.S. speed
limit of 45 mph (72.42 km/hr). The approximate p-values included
in both tables refer to the statistical significance of the result. A
p-value below 0.05 is commonly accepted as statistically significant,
as this indicates there is less than a five percent probability that the
results are random.
Traffic signal. Crash types of animal, sideswipe, meeting in
the opposite direction, meeting in the same direction, backing,
parked vehicle, pedestrian, and pedal cycle were each significantly
less likely to occur at a traffic signal compared to angle crashes,
which were significantly more likely to occur. A 1-mph speed
limit increase resulted in a 5.8 percent less likely chance of a crash
occurring at a traffic signal, as compared to no traffic control.
These results present that angle crashes at traffic signals should be
considered to a higher degree as they are more likely to occur and
are often more severe than other crash types.10,11
Stop sign. At stop sign controlled locations, all of the following
crash types were significantly more likely to occur compared to
rear end crashes at locations without a traffic control device: angle,
backing, head-on, pedestrian, and pedal cycle. Given the high
likelihood of head-on, pedestrian, and pedal cycle crashes, all high
severity crash types, future crash countermeasure selection at these
locations should be guided by these crash types.
School zone. Of the significant results of crash types in school
zone locations, pedal cycle crashes were less likely to occur, while
pedestrian, head-on, parked vehicle, and backing crashes were
more likely. As school zone locations often have high parking lot
movement at peak hours and high pedestrian movement, these
results seem reasonable. However, given pedestrian crashes tend to
be of higher severity simply due to the vulnerability that they face
with no outward protection, this crash type should be considered
to a higher degree in school zone locations. There is a high risk
of pedestrian crashes in these areas, both in occurrence and in
severity, than other crash types. Similarly, countermeasures for
head-on crashes would be beneficial to consider in these locations.
ITE Journal - June 2021

Table of Contents for the Digital Edition of ITE Journal - June 2021
