ITE Journal – December 2019 - 30

Despite the indisputable role of RLC in driver behavior, RLC
programs have always been under scrutiny. One common argument
against the program is that RLCs exacerbate the intersection safety
by referring to the increase in the number of crashes in some
cities or comparing the crash frequency before and after RLCs
installation. An evaluation of RLC in seven cities revealed that
right-angle crashes were reduced by 25 percent while the rear-end
crashes increased by 15 percent.4 Given that right-angle crashes
are inherently high injury risk crashes comparing to rear-end
collisions, the RLC program may still improve safety even if the
total number of crashes increases. In addition, the crash risk at
intersections is dependent on many factors, including the service
volume, local land use developments, intersection geometry and
control, user characteristics, etc., all which need to be considered
in RLC efficiency evaluations. This implies the need for a thorough
engineering analysis of the RLC impacts to include the type of
crashes and the impacts of risk factors into the discussion.
In addition, many RLC programs have been widely accused of
being used to generate revenues for cities rather than improving
safety, given fining regulations. Yet the number of communities
using RLCs has been dramatically decreasing since 2012 (from 533
to 430 communities), mainly because of the difficulties in sustaining
the financial viability of the program.2 Since the system costs are
mainly attributed to the RLC operation costs, efficient system design
can help to overcome the financial barriers associated with the RLC
program. The optimal distribution of the RLCs across the intersections can help the system to operate more cost-efficiently. Typically,
cameras installation is recommended at intersections with a higher
number of red-light running crashes or violations; however, given
the potential impacts of RLCs on the intersections in the vicinity,
this approach may not lead to optimum RLC allocation.2,5
In this study, we first proposed a methodological framework
with the aims of capturing the effect of RLCs on injury crash
frequency by: 1) incorporating observable factors (e.g. land use,
intersection geometry and service volume) and unobservable spatial
factors (e.g. immeasurable land use, social activities, and special
events), 2) considering the unobserved heterogeneity at an intersection such as variation in drivers characteristics, and 3) investigating the spatial effect of the RLC on adjacent intersections crash
frequency. Second, we defined an optimization problem using the
results of the model to seek for the optimal location of RLCs across
the system. The data from the City of Chicago RLC program in
Illinois, USA used in this study for evaluating the proposed model.

models have been used for examining the RLC effectiveness in
the form of linear, logistic, and generalized linear models.6,8-9 A
set of risk factors was used in these studies to better capture the
RLC impacts comprising intersection control as well as geometry
or functional characteristics. The red and yellow light durations
and cycle lengths can be used to reduce traffic violations.7,15-16 The
number of approaches at the intersection, speed limit, right/left
turn restriction, and the number of lanes are variables that have
been used in previous studies to characterize the intersection
geometry.5-6 The positive association between the intersection traffic
flow and RLR was shown in the literature. 5,6-9
The unobservable factors were not considered in previous
models for intersection crash analysis. The fact remains, however,
that the unobserved heterogeneity needs to be considered for
efficient and consistent model parameter estimates.17-18 In the
context of crash prediction models, the random parameter models
were mainly used to address the unobserved heterogeneity and
spatial dependencies were captured by spatial conditional autoregressive (CAR) models, and spatial weighting techniques.19-22 Spatial
dependency can be driven from the unobserved similarity in
interacted traffic flow as well as land-use and intersection characteristics.23 Dealing with autocorrelation (i.e., spatial correlation)
is considered as an instrument to boost the model prediction
by econometricians.24 Moreover, the spillover effect of RLC has
initially been discussed by Retting et al. and then further discussed
by other researchers.25-28 Ahmed and Abdel-Aty indicated that in
addition to significant crash reduction at intersections equipped
with an RLC, the safety of nearby intersections was improved in
lesser magnitude but still significantly.28
We proposed a Bayesian hierarchical spatial model in this
study that is capable of encountering the unobserved heterogeneity
and the spatial dependency between intersections crashes. Also,
the spillover effect of the RLC is captured in the form of a spatial
weighted variable. We examined land-use characteristics as well as
intersection function, geometry and control characteristics in the
proposed model.

Literature Review

In the first level, the data model describes the distribution of the
observed crashes (z) given the true process λ(.). We assumed crashes
are independent conditional on λ(.), which is a valid assumption in
the crash measurement context. In other words, independence in
the data model implies that the measurements of crashes frequency

The RLR violation is known as a safety issue,6-8 and enforcement
by RLC is introduced as an efficient solution to reduce RLR
violations.9-11 The effectiveness of the RLC has mainly been
investigated using before-after studies.5,12-14 In addition, statistical
30

D ecem ber 2019

i t e jo u rn al

Methodology
The proposed Bayesian hierarchical spatial model consists of three
levels:29
1. Data model: [data | process, parameters],
2. Process model: [process | parameters], and
3. Parameter model: [parameters].
ITE Journal – December 2019

Table of Contents for the Digital Edition of ITE Journal – December 2019
