j o u r n a l o f t r a f fi c a n d t r a n s p o r t a t i o n e n g i n e e r i n g ( e n g l i s h e d i t i o n ) x x x x ; x x x ( x x x ) : x x x

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.keaipublishing.com/jtte

Original Research Paper

Modeling the effect of a freeway road construction

project on link-level travel times

Venu Madhav Kukkapalli, Srinivas S. Pulugurtha*

Department of Civil and Environmental Engineering, The University of North Carolina at Charlotte, Charlotte, NC 28223, USA

highlights

Observed travel times are significantly different from estimates obtained using the BPR equation.

Travel times decreased on the freeway links during the construction project period.

Travel times increased on the connecting arterial street links during the construction project period.

Upstream and downstream link characteristics should be considered to examine the effects along with V/C and speed limit.

article info abstract

Article history: Drivers experience variations in travel time due to congestion and delays associated with
Received 22 March 2019 road construction projects on existing transportation facilities. The effect on the travel
Received in revised form time of a link decreases as its distance from the road construction project increases. In
11 November 2019 addition, the effect could extend to the connecting arterial street links within the vicinity of
Accepted 21 November 2019 the road construction project. Therefore, the focus of this research is to model the effect of
Available online xxx a road construction project on the freeway and connecting arterial street link-level travel
times. Data pertaining to a resurfacing construction project, which lasted for six months,
Keywords: was gathered from the traffic incident management systems (TIMS). The travel time data
Traffic engineering was collected and processed, for each selected freeway and connecting arterial street link,
Travel time for six months before the start of the resurfacing construction project and six months
Road construction project during the construction of the resurfacing project. Generalized linear models (GLM) were
Freeway developed and validated for before and during the construction project periods with the
Generalized linear model average travel time on a link as the dependent variable. Data such as time-of-the-day, day-
of-the-week, the distance of a link from the work zone, subject link characteristics, and
upstream and downstream link characteristics were collected and used as the predictor
variables. The results indicate that the subject link, upstream link and downstream link
characteristics have a significant effect on the freeway and connecting arterial street link-
level travel times due to the resurfacing construction project.
© 2020 Periodical Offices of Chang'an University. Publishing services by Elsevier B.V. on
behalf of Owner. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

* Corresponding author. Tel.: þ1 704 687 1233; fax: þ1 704 687 0957.
E-mail addresses: kukkapalli.venu@gmail.com (V.M. Kukkapalli), sspulugurtha@uncc.edu (S.S. Pulugurtha).
Peer review under responsibility of Periodical Offices of Chang'an University.
https://doi.org/10.1016/j.jtte.2019.11.002
2095-7564/© 2020 Periodical Offices of Chang'an University. Publishing services by Elsevier B.V. on behalf of Owner. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
2 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx
1. Introduction
Travel time reliability of a transportation network depends on
non-recurring congestion events, such as a crash, mechanical
breakdown of a vehicle, special event, weather and the type of
ongoing road construction project (Dowling et al., 2015). The
types of road construction projects include the construction
of a new road, addition of a new lane, pavement repair,
resurfacing, installation of pavement markers, etc., and
often involve one or more lane closures.
Work zones associated with the road construction projects
on freeways have a significant influence on traffic flow and the
severity of congestion (Hohmann and Geistefeldt, 2015).
Fontaine et al. (2014) stated that the use of mobility
performance measures enables practitioners to better assess
the effect of work zones on traffic congestion, identify
specific projects that need enhanced temporary traffic
control actions, and penalize contractors creating excessive
impacts. However, accurate prediction of travel time for a
given route or a freeway is a challenge, as it is influenced by
many different traffic and road parameters. In addition,
traffic would queue up the upstream side of the construction
project location due to the staggering, queuing, and delay at
the work zone. Studies, such as by Tanvir et al. (2016), based
on a link-based performance observed worse queue
spillback effects on the upstream links of the work zone
than in reality.
Vehicular traffic could also migrate from the freeway links
to the connecting arterial street links to avoid congestion and
major delays at the work zone. These effects vary by the time-
of-the-day and the day-of-the-week. The traffic volume and
travel time also vary during the road construction project
period when compared to before the construction project
period. Therefore, this research focuses on modeling the ef-
fect of a road construction project on the freeway and con-
necting arterial street link-level travel times and comparing
outputs for before and during construction project periods.
The remainder of the paper comprises four sections. A
review of the literature on the effect of a road construction
project on traffic delays, queue lengths, and travel times, and
limitations of the past research is presented next. It is fol-
lowed by a discussion of the methodology adopted in this
research and the results from this research. The final section
includes conclusions from this research and the scope for
future work.
2. Literature review
In recent years, there has been an increase in the number of
road construction projects and work zones due to the
continued deterioration of highway infrastructure in the
United States (Abdelmohsen and El-Rayes, 2016; Qu et al.,
2015). Vehicles go through these work zones at reduced
speeds and with fluctuated traffic flow rates (Jiang, 2002).
These fluctuations in traffic flow result in inconsistent travel
times along a route. Evaluating and predicting the effect of a
road construction project on travel time variations by the
time-of-the-day and the day-of-the-week will, therefore,
Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
help better understand their effect on travel time variations
or travel time reliability, and proactively adopt enhanced
temporary traffic control actions. Such practices may
include the implementation of a real-time work zone
information system or dynamic lane merging system to
control the safe transition of vehicles within the work zone
area (Koilada et al., 2018).
Several researchers have evaluated the effect of work
zones on traffic delays, queue lengths, and travel times in the
past. Ghosh-Dastidar and Adeli (2006) developed a neural
network-wavelet micro-simulation model to track the travel
time of an individual vehicle, for estimating traffic delay and
queue length at freeway work zones. Ramezani and
Benekohal (2011) investigated the mechanism of queue
propagation and dissipation at two potential freeway work
zone bottlenecks. They showed that, when the traffic
volume exceeds the capacity of the transition area and the
workspace, both locations will be active bottlenecks. Imran
and Pulugurtha (2014) evaluated the effect of the reduced
speed limit, length of the work zone, the number of lanes
closed, and the traffic volume on delay and queue length at
a freeway work zone using VISSIM traffic simulation software.
Petersen et al. (2014) worked on queue warning and travel
time estimation near a concrete overlay work zone and
compared with the baseline data. Qu et al. (2015) developed
simplified methods that identify the effect of a work zone on
travel time and travel time reliability to address the work
zone scheduling problem. Abdelmohsen and El-Rayes (2016)
developed a novel multi-objective optimization model for
generating optimal tradeoffs between minimizing the traffic
delay and the construction cost. Fitzpatrick et al. (2016)
explained the operational implications of the reduced
shoulder and lane widths on freeways; wider the shoulder,
higher the speed. Tanvir et al. (2017) developed a method to
estimate the change in travel time and traffic volume due to
work zones. They observed that travel times increased in
work zone routes during evening peak hours and overnight
hours. Du et al. (2017) developed a hybrid machine-learning
model to predict spatio-temporal delays by considering road
geometry, number of lanes closed, and work zone duration
as the predictor variables.
Yesantarao and Pulugurtha (2017) and Kukkapalli and
Pulugurtha (2018) examined the travel time and travel time
variations before, during, and after the completion of
selected road construction projects by computing the ratios
of travel time performance measures before, during, and
after the completion of selected road construction projects.
Srivastava et al. (2018) observed a reduction in travel time in
a work zone during a peak period at some locations. They
also observed that traffic volume influences travel time
reliability of the work zone.
It is beneficial, though challenging, to use travel time
estimation as an effective index to proactively identify
measures for reducing traffic congestion and improving
reliability (FHWA, 2005a, b). As travelers rely on travel time
reliability measured across days or different times of the day
for decision-making, researchers in the past have proposed
and quantified various travel time reliability measures (Al-
Deek and Emam, 2006; Bogers et al., 2006; Chen et al., 2003;
Dowling et al., 2015; FHWA, 2005a, b; Lomax et al., 2004;
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx
Lyman and Bertini, 2008; Pulugurtha and Duddu, 2014;
Sisiopiku and Islam, 2012; van Lint et al., 2004; van Lint and
van Zuylen, 2005; Wakabayashi and Matsumoto, 2012; Wil-
liams et al., 2013), examined the relationships between travel
time measures (Pu, 2011; Pulugurtha et al., 2016, 2017;
Wakabayashi and Matsumoto, 2012), and developed various
models for predicting and estimating the travel time reli-
ability on freeways (Mane and Pulugurtha, 2018; Yang et al.,
2010). However, from the literature review, it can be stated
that only a few researchers focused on travel time variations
at the work zones. There is little to no research on examining
the effect of a road construction project on travel time at the
link-level. Those who have worked on work zone travel times
have not explored the role of a comprehensive list of con-
struction project location characteristics, such as the number
of lanes open and closed during the construction, the speed
limit, the shoulder width, and the lane width of the subject
link, the upstream link and downstream link characteristics,
the time-of-the-day, the day-of-the-week and the distance of
the subject link from the road construction project on travel
time. Furthermore, the effect of the construction project on
the connecting arterial street links was not widely explored
in the past. This research contributes by considering these
factors and modeling the effect of a road construction project
on the freeway and connecting arterial street link-level travel
times.
3. Methodology
The methodology adopted in this research includes the
following steps.
1. Selecting the road construction project.
2. Identifying the data elements.
3. Defining the study area and collecting data.
4. Processing the travel time data.
5. Selecting the predictor variables for modeling.
6. Developing and validating the models.
3.1. Selecting the road construction project
The city of Charlotte, North Carolina was selected as the study
area for modeling the effect of a road construction project. The
list of road construction projects on the freeways was
collected from the city of Charlotte Department of Trans-
portation (CDoT) and the Traffic Incident Management System
(TIMS) maintained by the North Carolina Department of
Transportation (NCDoT). The collected road construction
projects were started and completed from the year 2013 to the
year 2015. Construction projects prior to the year 2013 were
not considered, as travel time data is not available for most of
the links in the study area. The year of the construction
project completion was set as 2015, as research efforts were
initiated during the spring of the year 2017.
The target is to collect real-world and most recent travel
time data which lasted for at least six months during the road
construction project period, in and around Charlotte city
limits. There could be significant changes within the vicinity
Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
3
of a road construction project, such as new developments that
affect the travel time at the link-level. It is hard to assess the
change in travel times and travel time variations due to a road
construction project in such cases. Therefore, multi-year road
construction projects were not considered in this research.
Further, the effect of other road construction projects or
developmental activities other than the selected road con-
struction project was considered to be minimal or consistent
in, both, before and during the road construction project pe-
riods. From the list of road construction projects, a resurfacing
construction project which lasted for six months on I-485 in
Charlotte, North Carolina was considered for analysis and
modeling.
3.2. Identifying the data elements
Travel time, generally, increases with an increase in the traffic
volume. Similarly, a section would attract a relatively higher
number of vehicles if the number of lanes is higher. In addi-
tion, travel time decreases if the speed limit is higher and all
other factors remain the same. Contrarily, travel time in-
creases if the lane width and the shoulder width are lower
(due to a decrease in motorist comfort level when driving).
Therefore, traffic volume, link capacity, speed limit, the
number of lanes, the shoulder width, and the lane width were
considered for analysis and modeling.
One or more lanes may be closed for construction. The
speed limit, the lane width, and the shoulder width may be
reduced along the construction section, which in turn in-
creases the travel time. Therefore, the number of lanes closed
during the construction project period, the reduced work zone
speed limit, and the shoulder width and the lane width during
the construction project period were considered for analysis
and modeling.
Upstream and downstream traffic volume, the speed limit,
the number of lanes, and the link length could have an in-
fluence on the travel time on a link in the construction section.
Therefore, the characteristics of upstream and downstream
links were identified and considered for analysis and
modeling.
Travelers tend to change their paths and migrate to the
connecting arterial streets to avoid congested sections and
minimize their total travel delay during the construction
project period when one or more lanes are closed on the
freeway work zones. Therefore, connecting arterial street link
characteristics, such as traffic volume, link capacity, the
number of lanes, if the road is divided or undivided, the
shoulder width and the lane width were also considered for
modeling and analysis. In addition, the distance of a link from
the work zone was also collected since the effect on travel
time decreases with an increase in the distance from the work
zone.
During the morning or evening peak hours, the traffic
volume is typically higher than when compared with the off-
peak hours. In other words, traffic volume varies by the time-
of-the-day. Similarly, the traffic volume is higher during the
weekdays when compared with the weekend days. Therefore,
the time-of-the-day and the day-of-week were also consid-
ered for analysis and modeling.
4 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx

Fig. 1 e Resurfacing construction project study area.

3.3. Defining the study area and collecting the data
The selected corridor (I-485) was operating at a 65 mph posted
speed limit before the construction period and at a 70 mph
posted speed limit during and after the construction period.
The number of lanes closed during the construction period is
one-lane in both directions, while two lanes were open for
traffic in both directions. The resurfacing construction project
was started in June 2015 and was completed in six months.
The length of the resurfacing construction project is
approximately 8 mi. However, data related to upstream and
downstream links for up to 3 mi were also collected. Simi-
larly, links on the connecting arterial streets were also
selected to account for the variations in travel times over
space and time.
An aerial view of the selected resurfacing construction
project location is shown in Fig. 1. The actual extent of the
construction project, upstream and downstream links, and

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx 5

Table 1 e Various characteristics considered for modeling the effect of the resurfacing construction project.
Parameter Characteristic
Freeway link parameters Traffic volume (veh/h)
Capacity (veh/h)
Number of lanes
Speed limit (mph)
Lane width (ft)
Shoulder width (ft)
Upstream and downstream link parameters Number of lanes closed
Upstream link length (mi)
Upstream link traffic volume (veh/h)
Upstream link speed limit (mph)
Upstream link number of lanes
Upstream link capacity (veh/h)
Downstream link length (mi)
Downstream link traffic volume (veh/h)
Downstream link speed limit (mph)
Downstream link number of lanes
Downstream link capacity (veh/h)
Connecting arterial street link parameters Traffic volume (veh/h)
Capacity (veh/h)
Number of lanes
Speed limit (mph)
Lane width (ft)
Shoulder width (ft)
Divided or undivided road
Upstream link length (mi)
Upstream link traffic volume (veh/h)
Upstream link speed limit (mph)
Upstream link number of lanes
Upstream link capacity (veh/h)
Downstream link length (mi)
Downstream link traffic volume (veh/h)
Downstream link speed limit (mph)
Downstream link number of lanes
Downstream link capacity (veh/h)
Other parameters Time-of-the-day
Day-of-the-week
Distance of the link from the study corridor (D) (mi)

connecting arterial streets that connect to the freeway are also The time-of-the-day considered for analysis and modeling
shown in the figure. are morning peak (7 AMe9 AM), morning off-peak (10 AMe12
For this research, 39 freeway links and 60 connecting PM), evening peak (5 PMe7 PM), and evening off-peak (10
arterial street links were selected for analysis and modeling. PMe12 AM) hours. Monday through Friday was considered as
The travel time data was collected, for each selected link, for weekdays, while Saturday and Sunday were considered as
six months before the start of the resurfacing construction weekend days.
project and six months during the construction of the resur-
facing project.
3.4. Processing the travel time data
A summary of various characteristics considered for
modeling the effect of the resurfacing construction project on
Travel time data was downloaded from the Regional Integrated
link-level travel time is presented in Table 1. The
Transportation Information Systems (RITIS) website in a raw
characteristics, such as the link capacity, the speed limit,
unprocessed format. The raw data file usually has traffic mes-
and the number of lanes for each link, summarized in Table
sage channel (TMC) code (tmc_code), time-stamp (measur-
1 were gathered from the CDoT regional travel demand
ement_tstamp), speed (speed), average speed (average_speed),
model and aerial images. The characteristics related to the
reference speed (reference_speed), travel time (trav-
road construction project period were obtained from the el_time_minutes), and score (confidence_score; 30 indicates
TIMS database. Traffic volume before and during the
real-time data; 20 indicates real-time data across multiple seg-
construction project period was collected from NCDoT and
ments; 10 indicates historical data). Only real time data records
CDoT. The average width of all lanes, for each selected
were considered for analysis and modeling in this research.
freeway and connecting arterial street link, was captured
The data requested has the travel time at the 1 min inter-
using the Google Earth Pro software. All the data used in this
val, for different study periods (before and during). The travel
research are for the year 2015.
time data was processed using Microsoft SQL Server 2012 to

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
6 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx

Table 2 e Validation results from BPR.

Link Validation result RMSE MAPE (%)
Freeway links (BPR) Before the construction project period 0.27 19.67
During the construction project period 0.27 20.93
Connecting arterial street links (BPR) Before the construction project period 0.62 74.23
During the construction project period 0.70 83.01

compute the average travel time for each selected freeway
and connecting arterial street link, by time-of-the-day and
day-of-the-week of each study period. The missing and blank
travel time records were removed prior to analysis and
modeling using SQL query.
All other data summarized in Table 1 were also added for
each link. Reasonable assumptions based on information
gathered from the online sources (Google Earth Pro software,
aerial images, etc.) and field visits were made for any
unclear or missing data details. A data dictionary was
developed to explain all data elements in the database.
Overall, 312 samples (39 freeway links  4 time periods  2
days of the week) on the freeway links and 480 samples (60
freeway links  4 time periods  2 days of the week) on the
connecting arterial street links were initially considered. From
these samples, outliers (links that are less than 0.3 mi), very
high travel time samples and null values were removed to
avoid bias that might arise due to missing travel time data,
considering very high values or dividing with very small
values. From the final database, 80% of the samples were
randomly selected and used for modeling the travel time
while the remaining 20% were used for validating the devel-
oped travel time model.
3.5. Selecting the predictor variables for modeling
The correlation between the average travel time and all the
predictor variables pertaining to work zone characteristics,
subject link, upstream link, and downstream link character-
istics, connecting arterial street link characteristics, the time-
of-the-day, the day-of-the-week and the distance of a link
from the study corridor were examined using SPSS software
(SPSS, 2008). The computed Pearson correlation coefficients lie
between ±1. Two variables are considered as strongly
correlated with each other at a 95% confidence level if the P-
value is less than or equal to 0.05.
The correlation between the dependent variable (average
travel time) and each predictor variable was first examined.
The correlation between the predictor variables was then
examined to select predictor variables that are not correlated
to each other for modeling. This was done to minimize the
effect of multicollinearity and improve the accuracy of the
average travel time estimates. The generalized linear models
(GLM) were then developed to model the effect of a road
construction project on link-level travel time before and dur-
ing the construction project period.
arterial street links, travel times were first estimated by using
the Bureau of Public Roads (BPR) travel time equation. It is
mathematically represented as in Eq. (1).
 
V
TT ¼ TTfreeflow 1 þ a b
(1)
ðCÞ
where TTfreeflow is the free flow travel time on the selected
link, TT is travel time, a ¼ 0.15 and b ¼ 4 were considered as
the default values, and V/C is volume over the capacity on the
selected link.
Travel time was estimated for each selected link using the
BPR equation and compared with the actual average travel
time for the same link. The root mean square error (RMSE) and
the mean absolute percentage error (MAPE) were computed to
assess the effectiveness of BPR equation in estimating the
travel time. Table 2 summarizes the RMSE and MAPE before
and during the construction project period on the freeway
and connecting arterial street links.
Higher variations were observed when travel time is esti-
mated using the BPR equation compared to the actual average
travel time for freeway and connecting arterial street links
before and during the construction project period. This in-
dicates that factors other than V/C influence the average
travel time before and during the road construction project
period. Therefore, new models (instead of the BPR equation)
were developed to estimate average travel times before and
during the road construction project.
The average travel time was selected as the dependent
variable to model the effect of a road construction project
since the practitioners and researchers are interested in esti-
mating the expected travel time. It was observed to better
explain the effect of a road construction project.
The RMSE and MAPE were computed and used for vali-
dating the model. The model can be considered as the best-
fitted model if the values of RMSE and MAPE are closer to zero.
The data were segregated into two different databases;
before and during the construction project period. For each
study period, data were separated for freeway links and con-
necting arterial street links.
The details of the predictor variables considered for
developing the average travel time model for freeway links
and the connecting arterial street links, before and during the
construction project period (dependent variables), are pre-
sented in Table 3.

3.6. Developing and validating the models 4. Results

Prior to developing average travel time models for before and The model development and validation for before and during
during the construction period on freeways and connecting construction project periods are discussed in this section.

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx 7

subject link and downstream link increases. This, in turn,

Table 3 e Predictor variables.
increases the traffic volume and the average travel time. In
No. Predictor variable addition, the average travel time per unit distance is expected
1 Volume/capacity (V/C) to decrease as the distance from the study corridor increases.
2 Number of lanes The predictor variables such as the V/C, the upstream V/C,
3 Speed limit (mph) the upstream number of lanes, the upstream speed limit, the
4 Shoulder width (ft)
downstream V/C, the downstream number of lanes, and the
5 Lane width (ft)
downstream speed limit were correlated with the remaining
6 Divided/undivided (1 or 0)
7 Upstream link length (mi) predictor variables but were not correlated to each other at a
8 Upstream V/C 95% confidence level.
9 Upstream number of lanes Generalized linear estimating equations analysis in SPSS
10 Upstream speed limit (mph) software (SPSS, 2008) was used for developing the average
11 Downstream link length (mi) travel time model for freeway links, using the
12 Downstream V/C
aforementioned predictor variables that are not correlated to
13 Downstream number of lanes
14 Downstream speed limit (mph)
each other. Of 297 freeway samples, 226 randomly selected
15 Distance to the study corridor (D) samples were used for modeling the effect of the resurfacing
16 Weekday (0 or 1) construction project, while the remaining 71 randomly
17 Weekend day (0 or 1) selected samples were used for validating the developed
18 Morning peak (0 or 1) model. The maximum significance level considered was 0.05
19 Evening peak (0 or 1)
(at a 95% confidence level). The predictor variables with a
20 Morning off-peak (0 or 1)
significance value greater than 0.05 were eliminated, except
21 Evening off-peak (0 or 1)
V/C, one after another when developing the models. The
elimination process was repeated until all other predictor
4.1. Average travel time before the construction project variables in the model have a significance value less than or
period for freeway links equal to 0.05.
Linear, gamma log-link distribution, negative-binomial
A Pearson correlation coefficient analysis was conducted log-link distribution, and Poisson log-link distribution based
using before the construction project period data for 297 models were developed for the freeway links before the con-
samples on the freeway. From the computed Pearson corre- struction project using the selected predictor variables. Table
lation coefficients, the predictor variables were selected to 4 summarizes the coefficients (C0 ), standard errors (SE),
minimize the effect of multicollinearity between the predictor significance values (P), quasi-likelihood under independence
variables. model criterion (QIC) and corrected quasi-likelihood under
The predictor variables such as the number of lanes, the independence model criterion (QICC) for the various freeway
upstream link length, and the downstream number of lanes links models for before the construction project period.
are positively correlated with the average travel time before The linear model has lower QIC and QICC, and most of the
the construction project period on the freeway links. The selected predictor variables are significant at a 95% confidence
distance from the study corridor is negatively correlated with level when compared with other distributions. The QIC and
the average travel time before the construction project period QICC are also reasonably close to each other for the linear model.
on the freeway links. A relatively higher number of vehicles Therefore, the linear model was selected as the best model for
may use the freeway section if the number of lanes on the the freeway links before the construction project period.

Table 4 e Comparison of model parameters for freeway links before the construction project period.
Parameter Model
Linear model Gamma log-link Negative-binomial Poisson log-link
log-link
C0 SE P C0 SE P C0 SE P C0 SE P
Intercept 1.26 0.07 0.01 0.23 0.07 0.01 3.87 0.07 0.01 3.87 0.07 0.01
V/C 0.07 0.13 0.50 0.01 0.13 0.99 0.01 0.13 0.99 0.02 0.13 0.85
Upstream V/C 0.65 0.17 <0.01 0.76 0.17 <0.01 0.74 0.17 <0.01 0.73 0.17 <0.01
Upstream no. of lanes 0.25 0.04 <0.01 0.28 0.04 <0.01 0.28 0.04 <0.01 0.29 0.04 <0.01
Upstream speed limit 0.01 <0.01 <0.01 0.01 <0.01 <0.01 0.01 <0.01 <0.01 0.01 <0.01 <0.01
Downstream V/C 0.51 0.15 <0.01 0.54 0.15 <0.01 0.53 0.15 <0.01 0.59 0.15 <0.01
Downstream no. of lanes 0.11 0.06 0.05 0.10 0.06 0.07 0.10 0.06 0.07 0.13 0.06 0.02
Downstream speed limit 0.01 <0.01 0.04 0.01 <0.01 0.07 0.01 <0.01 0.06 0.01 <0.01 0.01
QIC 30.24 30.24 30.24 39.87 39.87 39.87 29.77 29.77 29.77 1341.6 1341.6 1341.6
QICC 33.97 33.97 33.97 46.33 46.33 46.33 44.57 44.57 44.57 1290.68 1290.68 1290.68

Note: For negative-binomial and Poisson log-link, the average travel time was converted into seconds. In the case of linear and gamma log-link
distributions, the average travel time is in minutes.

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
8 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx

Table 5 e Comparison of model parameters for connecting arterial street links before the construction project period.
Parameter Model
Linear model Gamma log-link Negative-binomial Poisson log-link
log-link
C0 SE P C0 SE P C0 SE P C0 SE P
Intercept 2.80 0.20 <0.01 1.97 0.20 <0.01 6.05 0.20 <0.01 5.85 0.20 <0.01
Divided/undivided 0.15 0.05 <0.01 0.10 0.05 0.04 0.10 0.05 0.05 0.09 0.05 0.10
V/C 0.60 0.10 <0.01 0.41 0.10 0.03 0.41 0.10 0.39 0.41 0.10 0.01
Speed limit 0.04 <0.01 <0.01 0.03 <0.01 <0.01 0.04 <0.01 <0.01 0.03 <0.01 <0.01
Upstream V/C 0.24 0.13 0.06 0.25 0.13 0.40 0.24 0.13 0.50 0.16 0.13 0.58
Upstream no. of lanes 0.04 0.03 0.07 0.04 0.03 0.27 0.04 0.03 0.32 0.03 0.03 0.49
D (mi) 0.04 0.02 0.02 0.02 0.02 0.60 0.02 0.02 0.66 0.02 0.02 0.46
QIC 34.66 34.66 34.66 86.25 86.25 86.25 23.41 23.41 23.41 1336.42 1336.42 1336.42
QICC 30.30 30.30 30.30 31.94 31.94 31.94 31.18 31.18 31.18 985.21 985.21 985.21

Note: For negative-binomial and Poisson log-link, the average travel time was converted into seconds. In the case of linear and gamma log-link
distributions, the average travel time is in minutes.

The average travel time model before the construction
project period on freeway links shows that upstream link
characteristics such as upstream V/C and the upstream num-
ber of lanes have a smaller effect on the average travel time at
a 95% confidence level. However, the upstream speed limit has
a higher effect on the average travel time. Downstream char-
acteristics such as downstream V/C and the downstream
number of lanes have a higher effect on the average travel time
at a 95% confidence level. However, the downstream speed
limit has a smaller effect on the average travel time before the
construction project period on freeway links.
The average travel time would increase if the upstream V/C
decreases. While the traffic is entering a work zone from the
upstream link, there could be a possibility that the traffic
would have staggered or queued due to the construction.
Similarly, the average travel time would increase as the up-
stream number of lanes decreases. Further, an increase in the
upstream speed limit would influence the average travel time
significantly. Vehicle queueing could be building on the up-
stream links due to the construction irrespective of a higher
speed limit. Similarly, the average travel times are expected to
increase as the downstream V/C increases. In addition, more
traffic would be attracted to the freeway links if the down-
stream number of lanes increases. This, in turn, increases the
average travel time. Moreover, the average travel time would
also increase significantly if the speed limit is reduced on the
downstream links.
The developed linear model was then validated using data
for 71 samples selected from the same construction project
database. The average travel times were computed using the
developed model and compared with the actual average travel
times. The RMSE and MAPE were computed and used for
validating the model. The computed RMSE is 0.11, while the
computed MAPE is 7.75%. Around 6 s variation was observed
from the developed model when compared with the actual
average travel times.
4.2. Average travel time before the construction project
period for connecting arterial street links
To understand the effect of a road construction project on the
connecting arterial street links before the construction
project period, Pearson correlation coefficients were
computed, and the average travel time model was developed
and validated using data captured for the connecting arterial
street links.
A Pearson correlation analysis was conducted using before
the construction project period data for 328 samples on the
connecting arterial street links.
From the computed Pearson correlation coefficient anal-
ysis, most of the predictor variables were correlated with the
average travel time before the construction project period at a
95% confidence level. The V/C, the number of lanes, the up-
stream link length, the upstream number of lanes, the
downstream link length, the downstream V/C, the down-
stream number of lanes, the downstream speed limit, and
morning and evening peak hours are positively correlated
with the average travel time before the construction project
period. The average travel time would increase as the V/C
increases. Likewise, an increase in the number of lanes would
attract more vehicular traffic, which in turn increases the
average travel time.
Similarly, other predictor variables such as the upstream
link length, the upstream number of lanes, the downstream
link length, and the downstream number of lanes would in-
crease the average travel time by attracting additional traffic
volume. Further, the average travel time is expected to in-
crease as the speed limit reduces. A similar trend is expected
in the case of lane and shoulder width.
The average travel time is expected to increase as the
downstream speed limit reduces. It is expected that the
average travel time would decrease as the distance from the
study corridor increases. During weekdays, traffic volume will
be higher increasing average travel time. However, traffic
volume and the travel time will be considerably low on
weekends when compared to weekdays.
The predictor variables such as, either the link is divided or
undivided, the V/C, the speed limit, the upstream V/C, the
upstream number of lanes, and the distance from the con-
struction corridor were selected for modeling as they were
correlated with all other predictor variables but are not
correlated to each other at a 95% confidence level. The speed
limit and the number of lanes are correlated with each other.
Therefore, either the speed limit or the number of lanes was

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx 9

Table 6 e Comparison of model parameters for freeway links during the construction project period related to average
travel time before the construction project period.
Parameter Model
Linear model Gamma log-link Negative-binomial Poisson log-link Linear model (no
log-link intercept)
C0 SE P C0 SE P C0 SE P C0 SE P C0 SE P
Intercept 0.12 0.04 <0.01 1.07 0.08 <0.01 3.13 0.05 <0.01 3.02 0.08 <0.01 e e e
Average travel time before 0.86 0.04 <0.01 1.02 0.08 <0.01 0.01 >0.01 <0.01 0.01 <0.01 <0.01 0.98 <0.01 <0.01
QIC 10.32 10.32 10.32 19.96 19.96 19.96 394.77 394.77 394.77 11.30 11.30 11.30 6.14 6.14 6.14
QICC 8.07 8.07 8.07 15.51 15.51 15.51 387.34 387.34 387.34 14.90 14.90 14.90 6.40 6.40 6.40

Note: For negative-binomial and Poisson log-link, the average travel time was converted into seconds. In the case of linear and gamma log-link
distributions, the average travel time is in minutes.

used for model development. A similar logic was applied with
the upstream and downstream predictor variables.
Generalized linear estimating equations analysis in SPSS
software (SPSS, 2008) was used for developing the average
travel time model for connecting arterial street links before
the construction project period, using the aforementioned
predictor variables that are not correlated to each other.
For the connecting arterial street links before the con-
struction project period, 260 samples were used for developing
the models while 68 samples were used for validating the
developed models. The maximum significance level consid-
ered for developing the models was 0.07 (93% confidence
level). The predictor variables with a significance value greater
than 0.07 were eliminated, excluding V/C, one after another
when developing the models. The elimination process was
repeated until all the predictor variables in the model have a
significance value of less than or equal to 0.07. The QIC and
QICC were used to evaluate the strength of the predictor
variables and the model's goodness-of-fit.
Linear, gamma log-link, negative binomial log-link, and
Poisson log-link distribution based models were developed for
the connecting arterial street links before the construction
project period. Table 5 summarizes the coefficients (C0 ),
standard errors (SE), significance values (P), QIC, and QICC
for the various connecting arterial street links models for
before the construction project period.
The linear model has lower QIC and QICC when compared
with the gamma log-link, negative binomial log-link, and
Poisson log-link distribution based models. The QIC and QICC
are also reasonably close to each other. Therefore, the linear
model was selected as the best-fit model for estimating the
average travel time before the construction project period for
connecting arterial street links.
The developed average travel time model for before the
construction project period on the connecting arterial street
links shows that an undivided link and V/C has a higher effect
on the average travel time at a 95% confidence level. The up-
stream number of lanes has a higher effect on the average
travel time at a 93% confidence level. In addition, predictor
variables such as the speed limit, the upstream V/C, and the
distance of a link from the study corridor have a smaller effect
on the average travel time at a 95% confidence level.
The average travel time would increase significantly as the
traffic volume increases and the capacity decreases. Before
the construction project period, the average travel time is
expected to increase if the speed limit is reduced. As the up-
stream V/C decreases, the average travel time would increase
following similar trends as the freeway links. An increase in
the upstream number of lanes attracts high traffic volume and
increases the average travel time. The downstream link
characteristics do not have a significant effect on the average
travel time before the construction project period on con-
necting arterial street links.
The developed linear model was then validated using data
for 68 samples randomly selected from the same construction
project database. The average travel times were computed
using the developed model and compared with the actual
travel times. The computed RMSE is 0.45 and MAPE is 24.28%.
Around 20 s variation was observed from the developed model
when compared with the actual average travel times.
4.3. Average travel time during the construction project
period for freeway links
The average travel time during the road construction project
period could be related to the average travel time before the
road construction project period. Therefore, linear model, a
linear model with no intercept, Gamma log-link, Negative-
Binomial log-link, and Poisson log-link distribution models
were developed for freeway links with the average travel time
during the road construction project period as the dependent
variable and the average travel time before the road con-
struction project period as the predictor variable.
Data for 226 samples were used for developing the model
while data for 71 samples were used for validating the model.
A linear model with no intercept was selected since the QIC
and QICC are lower and close to each other. The computed
RMSE and MAPE for this model are 0.05 and 3.53%. The results
indicate that the average travel time during the construction
project period is lower than the average travel time before the
construction project period on a freeway link. Table 6
summarizes the coefficients (C0 ), standard errors (SE),
significance values (P), QIC, and QICC for the various freeway
link models for during the construction project period.
The average travel time during the construction project
period will depend on the V/C during the construction project
period on freeway links. However, it is strenuous and not
possible to collect the volume and capacity of the freeway
links during the construction project period before the start of
a future construction project. Therefore, V/C during the

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
10 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx

Table 7 e Model during the construction project period for estimating V/C on freeway links.
Parameter Model
Linear model Linear model (no intercept) Gamma log-link
0 0 0
C SE P C SE P C SE P
Intercept 0.01 <0.01 <0.01 e e e 2.91 0.08 <0.01
V/C before 1.19 0.05 <0.01 1.16 0.03 <0.01 5.95 0.10 <0.01
QIC 8.69 8.69 8.69 6.78 6.78 6.78 30.42 30.42 30.42
QICC 4.75 4.75 4.75 2.76 2.76 2.76 29.58 29.58 29.58

construction project period was estimated using the V/C
before the construction project period as the predictor
variable.
A model was developed for estimating V/C during the
construction project period on the freeway links. The sample
size used for estimating the V/C during the construction
project period was 226 samples. Data for 71 samples were
used for validating the developed model. A linear model, a
linear model with no intercept and Gamma log-link distribu-
tion model were developed. Table 7 presents the coefficients
(C0 ), standard errors (SE), significance values (P), QIC, and
QICC for the freeway link V/C model, for during the
construction project period.
The QIC and QICC were lower and closer to each other for
the linear model with no intercept when compared with the
other developed models. Therefore, to avoid the negative
intercept, the linear model with no intercept was selected for
estimating the V/C during the construction project period on
freeway links. From the developed model, V/C before the
construction positively influences V/C during the construc-
tion project period on the freeway links at a 95% confidence
level.
The developed model was then validated with data for 71
links. The computed RMSE and MAPE for this model are 0.04
and 15.04%, respectively. The estimated V/C during the con-
struction project period was then used as one of the predictor
variables for computing the Pearson correlation coefficients
and developing the average travel time model for during the
construction project period on the freeway links.
During the construction project period, 297 samples were
used for computing the Pearson correlation coefficients. From
the Pearson correlation coefficient analysis, the downstream
number of lanes is positively correlated with the average
travel time on the freeway links during the construction
project period. As the downstream number of lanes increases,
the traffic volume and, hence, the average travel time
increases.
The predictor variables such as the V/C, the upstream link
length, the upstream V/C, the downstream link length, the
downstream V/C, the downstream number of lanes, and the
downstream speed limit were considered for the model
development. The upstream speed limit and the upstream
number of lanes are correlated to each other. Therefore, either
the upstream number of lanes or the upstream speed limit
could be used for the model development. However, the up-
stream number of lanes were eliminated since the signifi-
cance value is greater than 0.05. The downstream link
characteristics were not correlated to each other and are sig-
nificant at a 95% confidence level.
Of 297 freeway samples, data for 226 samples were used for
modeling while data for the remaining 71 randomly selected
samples were used for validating the model. The maximum
significance level considered was 0.05 (at a 95% confidence
level). The predictor variables with a significance value greater
than 0.05 were eliminated, except the V/C, one after another
when developing the models. The elimination process was
repeated until all the predictor variables in the models have a
significance value less than or equal to 0.05.

Table 8 e Comparison of model parameters for freeway links during the construction project period.
Parameter Model
Linear model Gamma log-link Negative binomial Poisson log-link
log-link
C0 SE P C0 SE P C0 SE P C0 SE P
Intercept 1.14 0.08 <0.01 0.21 0.08 0.03 4.30 0.08 <0.01 4.26 0.08 <0.01
V/C 0.04 0.10 0.35 0.20 0.10 0.88 0.02 0.10 0.87 0.04 0.10 0.74
Upstream link length 0.04 0.02 0.05 0.06 0.02 0.02 0.06 0.02 0.02 0.05 0.02 0.02
Upstream V/C 0.63 0.17 <0.01 0.83 0.17 0.01 0.81 0.17 <0.01 0.69 0.17 <0.01
Downstream link length 0.06 0.03 <0.01 0.09 0.03 0.01 0.09 0.03 0.01 0.07 0.03 0.04
Downstream V/C 0.78 0.19 <0.01 0.93 0.19 0.01 0.93 0.19 <0.01 0.90 0.19 <0.01
Downstream no. of lanes 0.09 0.04 <0.01 0.08 0.04 0.18 0.08 0.04 0.17 0.10 0.04 0.06
Downstream speed limit 0.01 0.00 <0.01 0.01 0.00 0.01 0.01 0.00 <0.01 0.01 0.00 <0.01
QIC 33.45 33.45 33.45 42.48 42.48 42.48 29.68 29.68 29.68 1349.28 1349.28 1349.28
QICC 33.92 33.92 33.92 45.46 45.46 45.46 44.08 44.08 44.08 1282.77 1282.77 1282.77

Note: For negative-binomial and Poisson log-link, the average travel time was converted into seconds. In the case of linear and gamma log-link
distributions, the average travel time is in minutes.

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx 11

Table 9 e Comparison of model parameters for connecting arterial street links during the construction project period
related to average travel time before the construction project period.
Parameter Model
Linear model Gamma log-link Negative binomial Poisson log-link Linear model (no
log-link intercept)
C0 SE P C0 SE P C0 SE P C0 SE P C0 SE P
Intercept 0.05 0.05 0.30 0.98 0.10 <0.01 3.27 0.06 <0.01 3.12 0.10 <0.01 e e e
Average travel time before 1.01 0.03 <0.01 0.93 0.07 <0.01 0.01 <0.01 <0.01 0.01 0.01 <0.01 1.05 <0.01 <0.01
QIC 11.37 11.37 11.37 17.91 17.91 17.91 404.23 404.23 404.23 8.20 8.20 8.20 8.50 8.50 8.50
QICC 9.77 9.77 9.77 12.12 12.12 12.12 397.65 397.65 397.65 11.86 11.86 11.86 7.81 7.81 7.81

Note: For negative-binomial and Poisson log-link, the average travel time was converted into seconds. In the case of linear and gamma log-link
distributions, the average travel time is in minutes.

Table 8 summarizes the coefficients (C0 ), standard errors
(SE), significance values (P), QIC, and QICC for the freeway
links models during the construction project period. The
linear model has lower QIC and QICC when compared with
other distributions. They are also reasonably close to each
other for the linear model. In addition, most of the predictor
variables are significant at a 95% confidence level for the
linear model when compared with other distributions.
The developed linear model during the construction proj-
ect period on freeway links showed that the upstream link
length has a higher effect on the average travel time. However,
the upstream V/C has a smaller effect on the average travel
time at a 95% confidence level. In addition, downstream
characteristics such as the link length and the speed limit
have a smaller effect on the average travel time during the
construction project period. However, the downstream V/C
and the downstream number lanes have a higher effect on the
average travel time at a 95% confidence level.
An increase in the upstream link length increases the
average travel time. When the upstream V/C and the down-
stream link length decrease, the average travel time would
increase significantly. Also, an increase in the downstream V/
C has a significant effect on the average travel time. Further,
an increase in traffic volume on the downstream links would
increase the average travel time. Similarly, the average travel
time would also increase as the downstream number of lanes
increases. An increase in lane capacity would attract more
traffic on the downstream links. Further, the average travel
time is expected to increase if the speed limit is reduced on the
downstream links. Overall, from before to during the con-
struction project period on the freeway links, upstream and
downstream link lengths have a significant effect on the
average travel time. The upstream number of lanes and the
upstream speed limit do not have a significant effect during
the construction project period when compared with before
the construction project period on the freeway links.
The developed model was validated using data for the 71
randomly selected freeway samples. The RMSE and MAPE for
this model are 0.15 and 8.67%, respectively. The model was
observed to be accurately estimating the average travel time
during the construction project period.
4.4. Average travel time during the construction project
period for connecting arterial street links
A linear model, a linear model with no intercept, gamma log-
link, negative-binomial log-link, and Poisson log-link distri-
bution models for connecting arterial street links were
developed with the average travel time during the road con-
struction project period as the dependent variable and the
average travel time before the road construction project
period as the predictor variable. Data for 221 samples were
used for developing the model and 59 samples were used for
validating the model. A linear model with no intercept was
selected since the QIC and QICC are lower and close to each
other. The computed RMSE and MAPE for this model are 0.34
and 18.04%. The results indicate that the average travel time
during the construction project period is higher than the
average travel time before the construction project period on a
connecting arterial street link. Table 9 summarizes the
coefficients (C0 ), standard errors (SE), significance values (P),
QIC, and QICC for the various connecting arterial street link
models for the during the construction project period.
The Pearson correlation coefficients were computed, and
the average travel time model was developed for connecting
arterial street links during the construction project period.
The V/C during the construction on connecting arterial street
links is not known before the construction is commenced.

Table 10 e Model during the construction project period for estimating V/C on connecting arterial street links.
Parameter Model
Linear model Linear model (no intercept) Gamma log-link
0 0 0
C SE P C SE P C SE P
Intercept 0.16 0.04 <0.01 e e e 0.15 0.08 0.03
V/C before 2.06 0.18 <0.01 1.59 0.12 <0.01 0.87 0.10 0.21
QIC 20.93 20.93 20.93 21.15 21.15 21.15 42.48 42.48 42.48
QICC 9.12 9.12 9.12 8.41 8.41 8.41 45.46 45.46 45.46

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
12 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx

Table 11 e Comparison of model parameters for connecting arterial street links during the construction project period.
Parameter Model
Linear model Gamma log-link Negative binomial Poisson log-link
log-link
C0 SE P C0 SE P C0 SE P C0 SE P
Intercept 2.99 0.23 <0.01 2.10 0.23 <0.01 6.19 0.23 <0.01 5.89 0.23 <0.01
Divided/undivided 0.29 0.04 <0.01 0.22 0.04 <0.01 0.21 0.04 <0.01 0.20 0.04 <0.01
V/C during 0.33 0.05 <0.01 0.29 0.05 <0.01 0.28 0.05 <0.01 0.23 0.05 <0.01
Speed limit (mph) 0.04 0.00 <0.01 0.03 0.00 <0.01 0.03 0.00 <0.01 0.03 0.00 <0.01
Upstream link length 0.11 0.02 <0.01 0.06 0.02 0.05 0.06 0.02 0.05 0.07 0.02 <0.01
Upstream V/C 0.25 0.06 <0.01 0.35 0.06 0.01 0.34 0.06 0.02 0.20 0.06 0.19
Upstream no. of lanes 0.04 0.04 <0.01 0.08 0.04 0.23 0.07 0.04 0.23 0.04 0.04 0.42
Downstream no. of lanes 0.07 0.05 <0.01 0.11 0.05 0.22 0.11 0.05 0.22 0.06 0.05 0.31
D (mi) 0.03 0.02 <0.01 0.01 0.02 0.93 0.01 0.02 0.95 0.01 0.02 0.65
QIC 32.88 32.88 32.88 106.6 106.6 106.6 19.57 19.57 19.57 1186.67 1186.67 1186.67
QICC 32.21 32.21 32.21 32.51 32.51 32.51 31.86 31.86 31.86 875.88 875.88 875.88

Note: For negative-binomial and Poisson log-link, the average travel time was converted into seconds. In the case of linear and gamma log-link
distributions, the average travel time is in minutes.

Therefore, the V/C was estimated using the V/C before the
construction project period on the connecting arterial street
links as the predictor variable. Data for 221 samples were used
for developing the model, while data for 59 samples were used
for validating the developed model. Linear model, a linear
model with no intercept and Gamma log-link distribution
models were developed. The developed V/C model was vali-
dated with the V/C from during the construction project
period data for connecting arterial street links. The co-
efficients (C0 ), standard errors (SE), significance values (P), QIC,
and QICC are shown in Table 10.
The QIC and QICC are lower for the linear model with no
intercept. In addition, to avoid the negative intercept, a linear
model with no intercept was selected for estimating the V/C
during the construction project period on a connecting arterial
street link.
For the developed V/C model, the QIC and QICC values are
lower and close to each other. The V/C before the construction
project period influences the V/C during the construction
project period on the connecting arterial street links. The V/C
model was validated by comparing with the V/C during the
construction project period on the connecting arterial street
links. The RMSE and MAPE for this model are 0.08 and 3.06%,
respectively. The developed V/C model during the construc-
tion project period was then used as one of the predictor
variables when developing the Pearson correlation co-
efficients and the average travel time model for the connect-
ing arterial street links during the construction project period.
A Pearson correlation coefficient analysis was conducted
for the connecting arterial street links similar to the freeway
links. Data for 280 samples were used to compute the Pearson
correlation coefficients. The Pearson correlations indicate that
the majority of the variables are correlated to the average
travel time on the connecting arterial street links during the
construction project period.
The V/C, the number of lanes, the upstream link length, the
upstream number of lanes, the downstream link length, the
downstream V/C, the downstream number of lanes, and
weekday evening peak are positively correlated with the
average travel time during the construction project period. As
the V/C increases, the average travel time would increase
since the traffic volume would increase while the capacity is
less. As the number of lanes increases, arterial streets would
attract more traffic volume, which in turn increases the
average travel time. Similarly, the average travel time would
increase if the upstream V/C and the number of lanes
increase.
On the other hand, divided/undivided, the speed limit, the
upstream V/C, the downstream speed limit, the distance from
the study corridor, weekend day, and evening off-peak hours
are negatively correlated with the average travel time during
the construction project period. The average travel time would
increase as the speed limit reduces. If the link is not a divided
section, the average travel time could increase due to the
reduced comfort level from close oncoming traffic. Likewise,
the upstream V/C and the downstream speed limit could have
a similar effect. In addition, the average travel time would
reduce as the vehicles move away from the study corridor.
The predictor variables such as, whether the link is divided
or undivided, the V/C, the speed limit, the upstream link
length, the upstream V/C, the upstream number of lanes, the
downstream number of lanes, and the distance of a link from
the construction project were considered for model develop-
ment. All the predictor variables considered are correlated to
other predictor variables and are not correlated to each other.
The logic applied for previous models while selecting the
predictor variables was applied for selecting the predictor
variables in developing the models for during the construction
project period.
Data for 221 samples were used for developing the model,
while data for 59 samples were used for validating the
developed model. The maximum significance level consid-
ered was 0.05 (at a 95% confidence level). The predictor vari-
ables with a significance value greater than 0.05 were
eliminated, except V/C, one after another when developing
the models. The elimination process was repeated until all
the predictor variables in the models have a significance
value less than or equal to 0.05.

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx
Table 11 summarizes the coefficients (C0 ), standard errors
(SE), significance values (P), QIC, and QICC for the various
connecting arterial street links models for during the
construction project period. The linear model was selected
as the best-fit model for estimating the average travel time.
The QIC and QICC are lower and close to each other.
Furthermore, most of the predictor variables are significant
at a 95% confidence level for the linear model when
compared with other model distributions.
The developed linear model during the construction proj-
ect period on the connecting arterial street links shows that if
an undivided link and V/C have a higher effect on the average
travel time at a 95% confidence level. However, the speed limit
has a smaller effect on the average travel time. The upstream
link length and the upstream number of lanes have a higher
effect on the average travel time at a 95% confidence level. The
upstream V/C has a smaller effect on the average travel time
during the construction project period. In addition, the
downstream number of lanes and the distance of a link from
the study corridor has a smaller effect on the average travel
time at a 95% confidence level.
Most of the predictor variables that are observed to be
significant before the construction project period is observed
to be significant during the construction project period on
connecting arterial street links. The average travel time would
increase on the connecting arterial street links as the V/C in-
creases. During the construction project period, the average
travel time would increase if the speed limit is reduced. In
addition, the upstream link length and the upstream number
of lanes has a significant effect on the average travel time. The
average travel time would increase as the length of the link
increases. Similarly, upstream V/C and the downstream
number of lanes have a significant effect on the average travel
time. From before to during the construction project period,
the upstream link length and the downstream number of
lanes have a significant effect in increasing or decreasing the
average travel time. The effect of other predictor variables
remained the same from before to during the construction
project period on the connecting arterial street links.
The developed average travel time model was then vali-
dated using data for 59 randomly selected samples. The
computed RMSE and MAPE for this model are 0.44 and 11.42%,
respectively. The developed average travel time model is
accurately estimating travel times during the construction
project period on the connecting arterial street links.
Overall, the validation results from the freeway and con-
necting arterial street link models indicate that the developed
models accurately estimate link-level travel time when
compared with the conventional BPR equation.
5. Conclusions
A road construction project increases delay, congestion, and
lowers travel time on the freeway links and the connecting
arterial street links. By modeling the effect of a road con-
struction project period on the average travel time at link-
level, this research provides insights pertaining to factors that
influence the average travel time on the freeway and the
Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
13
connecting arterial street links, before and during the con-
struction of a resurfacing project period.
The average travel times are significantly different from
estimates obtained using the BPR travel time equation. It was
observed that the average travel time during the construction
project period decreased on the freeway links and increased
on the connecting arterial street links. This could be attributed
to the change in the posted speed limits during the study
period. Furthermore, to avoid delay during the construction
project period, drivers would have shifted from the freeway
links to the connecting arterial links. At the connecting arte-
rial links, the speed limit, capacity and the number of lanes
will be lower when compared with the freeway links. In
addition, the increase in traffic volumes would have affected
the average travel times on the connecting arterial links due to
a construction project on freeways.
The V/C varies with the time-of-the-day and the day-of-
the-week and was observed to have a higher correlation with
the time-of-the-day and the day-of-the-week. Therefore, V/C
was forced into the models while the time-of-the-day and the
day-of-the-week were not considered when developing the
models.
The predictor variables such as the V/C, the speed limit, the
upstream and downstream link characteristics have a signif-
icant effect on the average travel time on the freeway and the
connecting arterial street links. The predictor variables, re-
lationships, and the effect differ for before and during the
construction project periods. They also differ for freeway and
connecting arterial street links. Practitioners should consider
these factors into consideration, in addition to work zone
characteristics, when planning a resurfacing construction
project on the freeways.
More importantly, the road construction project in-
fluences the average travel time on connecting arterial street
links. The effect on the connecting arterial street links should
be considered along with the effect on the freeway links
when developing temporary traffic control actions and
detour plans.
The data for the entire road construction project period
was considered for this research. However, the data related to
construction activity or actual construction work times by
location and data were not available. Some aspects such as
the exact location of lane closures and the reduced speed limit
varied spatially and temporally during the road construction
project. Reasonable assumptions were made for this research
purposes. Collecting the actual start and end times of the
construction activity and considering these details along with
traffic control practices at micro-level (for example, reduced
speed limit, number of lanes open, etc.), by location, for
analysis and modeling would improve the accuracy of
estimates.
The type of road construction project and other ongoing
projects within its vicinity may influence the travel time on
the freeway and connecting arterial street links. The
geographic location and land use characteristics within the
vicinity of the road construction project may influence the
travel time on the freeway and connecting arterial street links.
Conducting analysis and modeling using other types of road
construction projects and at various geographic locations also
merit investigation.
14 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx
Disclaimer
This paper is disseminated in the interest of information ex-
change. The views, opinions, findings, and conclusions re-
flected in this paper are the responsibility of the authors only
and do not represent the official policy or position of the
USDOT/OST-R, or any other state, or the University of North
Carolina at Charlotte or other entity. The authors are
responsible for the facts and the accuracy of the data pre-
sented herein. This paper does not constitute a standard,
specification, or regulation.
Conflict of interest
The authors do not have any conflict of interest with other
entities or researchers.
Acknowledgments
This paper is prepared based on information collected for a
research project funded by the United States Department of
Transportation - Office of the Assistant Secretary for Research
and Technology (USDOT/OST-R) University Transportation
Centers Program (Grant # 69A3551747127). The authors
sincerely thank the staff of the North Carolina Department of
Transportation (NCDOT), the Regional Integrated Trans-
portation Information System (RITIS), and the city of Charlotte
Department of Transportation (CDoT) for their help with data
required for the study.
references
Abdelmohsen, A.Z., El-Rayes, K., 2016. Optimal trade-offs
between construction cost and traffic delay for highway
work zones. Journal of Construction Engineering and
Management 142 (7), 05016004.
Al-Deek, H., Emam, E.B., 2006. New methodology for estimating
reliability in transportation networks with degraded link
capacities. Journal of Intelligent Transportation Systems 10
(3), 117e129.
Bogers, E.A.I., Viti, F., Hoogendoorn, S.P., et al., 2006. Valuation of
different types of travel time reliability in route choice: large
scale laboratory experiment. Transportation Research Record
1985, 162e170.
Chen, C., Skabardonis, A., Varaiya, P., 2003. Travel-time reliability
as a measure of service. Transportation Research Record 1885,
74e79.
Dowling, R.G., Parks, K.L., Nevers, B., et al., 2015. Incorporating
Travel-Time Reliability into the Congestion Management
Process: a Primer. FHWA-HOP-14-034. FHWA, Washington DC.
Du, B., Chien, S., Lee, J., et al., 2017. Predicting freeway work zone
delays and costs with a hybrid machine-learning model.
Journal of Advanced Transportation 2017, 6937385.
Federal Highway Administration (FHWA), 2005a. Traffic
Congestion and Reliability: Trends and Advanced Strategies
for Congestion Mitigation. FHWA, Washington DC.
Federal Highway Administration (FHWA), 2005b. Travel Time
Reliability: Making it There on Time, All the Time. Available
Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002
at: https://ops.fhwa.dot.gov/publications/tt_reliability/TTR_
Report.htm (Accessed 15 September 2018).
Fitzpatrick, K., Dixon, K., Avelar, R., 2016. Evaluating operational
implications of reduced lane and shoulder widths on freeways.
Journal of Transportation Engineering 142 (11), 04016052.
Fontaine, M.D., Chun, P., Cottrell Jr., B.H., 2014. Using private
sector travel time data for project-level work zone mobility
performance measurement. In: Transportation Research
Board 93rd Annual Meeting, Washington DC, 2014.
Ghosh-Dastidar, S., Adeli, H., 2006. Neural network-wavelet
microsimulation model for delay and queue length
estimation at freeway work zones. Journal of Transportation
Engineering 132 (4), 331e341.
Hohmann, S., Geistefeldt, J., 2015. Assessment of travel time
losses in freeway work zones for different geometric layouts.
In: Transportation Research Board 94th Annual Meeting,
Washington DC, 2015.
Imran, Md S., Pulugurtha, S.S., 2014. Assessment of the effect of
freeway posted speed limit reduction on vehicular delay at a
work zone. In: Transportation Research Board 93rd Annual
Meeting, Washington DC, 2014.
Jiang, Y., 2002. Dynamic prediction of traffic flow and congestion
at freeway construction zones. Journal of Construction
Engineering 7 (1), 45e57.
Koilada, K., Pulugurtha, S.S., Mane, A.S., 2018. Risk factors
affecting crash injury severity by work zone area. In:
Transportation Research Board 97th Annual Meeting,
Washington DC, 2018.
Kukkapalli, V.M., Pulugurtha, S.S., 2018. Effect of road
construction projects on travel time reliability. In:
International Conference on Transportation & Development
(ICTD), Pittsburgh, 2018.
Lomax, T., Turner, T., Margiotta, R., 2004. Monitoring Urban
Roadways in 2002: Using Archived Operations Data for
Reliability and Mobility Measurement. Texas Transportation
Institute, Texas A&M University, College Station.
Lyman, K., Bertini, R., 2008. Using travel time reliability measures
to improve regional transportation planning and operations.
Transportation Research Record 2046, 1e10.
Mane, A.S., Pulugurtha, S.S., 2018. Link-level travel time
prediction using artificial neural network models. In: 21st
IEEE International Conference on Intelligent Transportation
Systems, Maui, 2018.
Petersen, S., Minge, E., Iverson, C., 2014. Queue warning and
travel time estimation near a work zone. In: Transportation
Research Board 93rd Annual Meeting, Washington DC, 2014.
Pu, W., 2011. Analytic relationships between travel time reliability
measures. Transportation Research Record 2254, 122e130.
Pulugurtha, S.S., Duddu, V.R., 2014. Assessing transportation
system reliability. In: The 2nd T&DI Congress, Orlando, 2014.
Pulugurtha, S.S., Duddu, V.R., Thokala, V.R., 2016. Travel-
time-based performance measures: examining
interrelationships and recommendations for analysis. In:
Transportation Research Board 95th Annual Meeting,
Washington DC, 2016.
Pulugurtha, S.S., Penmetsa, P., Duddu, V.R., 2017. Monetizing
Reliability to Evaluate the Impact of Transportation
Alternatives. FHWA/NC/2015-07. North Carolina Department
of Transportation, Raleigh.
Qu, T., Jin, P.J., Yu, C., et al., 2015. Travel time reliability based
highway work zone scheduling. In: Transportation Research
Board 94th Annual Meeting, Washington DC, 2015.
Ramezani, H., Benekohal, R., 2011. Analysis of queue formation
and dissipation in work zones. Procedia-Social and
Behavioral Sciences 16, 450e459.
Sisiopiku, V.P., Islam, M.D.S., 2012. A freeway travel time
reliability study. International Journal of Engineering
Research and Development 3 (10), 83e101.
 J. Traffic Transp. Eng. (Engl. Ed.) xxxx; xxx (xxx): xxx 15

SPSS, 2008. SPSS 16.0 for Windows. SPSS Inc., Chicago. In: The 4th Conference of Transportation Research Group of
Srivastava, A., Chitturi, M.V., Ahn, S., et al., 2018. Analytical India (CTRG), Mumbai, 2017.
Methods for Work Zone Travel Time Reliability. InTrans
Project 06-277. Iowa Department of Transportation, Ames.
Tanvir, S., Ahmed, I., Karmakar, N., et al., 2017. Determination of
Dr. Venu Madhav Kukkapalli completed his
diversion sensitivity at urban freeway work zones using
PhD in infrastructure and environmental
Bluetooth devices. In: Transportation Research Board 96th
systems at the University of North Carolina
Annual Meeting, Washington DC, 2017.
at Charlotte in 2018, and is currently working
Tanvir, S., Karmakar, N., Rouphail, N.M., et al., 2016. Modeling
as transportation engineer at Transcend
freeway work zones with mesoscopic dynamic traffic
Engineers & Planners, LLC., Houston, TX. He
simulator: validation, gaps, and guidance. Transportation
earned his master's degree from the National
Research Record 2567, 122e130.
Institute of Technology, Trichy, India. His
van Lint, J.W.C., Tu, H., van Zuylen, H.J., 2004. Travel time
areas of interest are transportation planning
reliability on freeways. In: The 10th World Conference on
and traffic safety.
Transportation Research (WCTR), Istanbul, 2014.
van Lint, J.W.C., van Zuylen, H.J., 2005. Monitoring and predicting
freeway travel time reliability: using width and skew of day-
to-day travel time distribution. Transportation Research
Record 1917, 54e62.
Wakabayashi, H., Matsumoto, Y., 2012. Comparative study on
travel time reliability indexes for highway users and
operators. Journal of Advanced Transportation 46 (4), 318e339. Dr. Srinivas S. Pulugurtha is currently
Williams, B.M., Chase, R.T., Xu, Y., et al., 2013. Mobility and working as professor and research director
Reliability Performance Measurement. North Carolina of the Department of Civil and Environ-
Department of Transportation, Raleigh. mental Engineering at The University of
Yang, M., Liu, Y., You, Z., 2010. The reliability of travel time North Carolina at Charlotte (UNC Charlotte).
forecasting. IEEE Transactions on Intelligent Transportation He is also currently directing the Infra-
Systems 11 (1), 162e171. structure, Design, Environment, and Sus-
Yesantarao, V.R.S.S., Pulugurtha, S.S., 2017. Evaluating the tainability (IDEAS) Center on UNC Charlotte
influence of a freeway capacity improvement project on campus.
travel time based performance measures within its vicinity.

Please cite this article as: Kukkapalli, V.M., Pulugurtha, S.S., Modeling the effect of a freeway road construction project on link-
level travel times, Journal of Traffic and Transportation Engineering (English Edition), https://doi.org/10.1016/j.jtte.2019.11.002