You are on page 1of 34

Transportmetrica A: Transport Science

ISSN: 2324-9935 (Print) 2324-9943 (Online) Journal homepage: https://www.tandfonline.com/loi/ttra21

Perceived Level of Service at Signalized


Intersections under Heterogeneous Traffic
Conditions

Darshana Othayoth, K.V. Krishna Rao & B.K. Bhavathrathan

To cite this article: Darshana Othayoth, K.V. Krishna Rao & B.K. Bhavathrathan (2020):
Perceived Level of Service at Signalized Intersections under Heterogeneous Traffic Conditions,
Transportmetrica A: Transport Science, DOI: 10.1080/23249935.2020.1737270

To link to this article: https://doi.org/10.1080/23249935.2020.1737270

Accepted author version posted online: 28


Feb 2020.

Submit your article to this journal

Article views: 7

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=ttra21
Publisher: Taylor & Francis & Hong Kong Society for Transportation Studies Limited
Journal: Transportmetrica A: Transport Science
DOI: 10.1080/23249935.2020.1737270

Perceived Level of Service at Signalized Intersections under

Heterogeneous Traffic Conditions

Darshana Othayoth1, K.V. Krishna Rao2 and B.K.Bhavathrathan3


1
Post Doctoral Fellow, Indian Institute of Technology Palakkad, Kerala 678623, India.

E-mail: odarshana@iitpkd.ac.in
2
Professor, Department of Civil Engineering, Indian Institute of Technology Bombay, Mumbai

400076, India.

E-mail: kvkrao@iitb.ac.in
3
Assistant Professor, Civil Engineering, Indian Institute of Technology Palakkad, Kerala 678623,

India.

Email: bhavathrathan@iitpkd.ac.in
Perceived Level of Service at Signalized Intersections under

Heterogeneous Traffic Conditions

Abstract

Level of Service (LOS) indicates the operational efficiency of traffic facilities. The method

recommended by Highway Capacity Manual (HCM) for signalized intersections uses control

delay as the service measure and does not incorporate users’ perception. As LOS is a measure

of customer satisfaction, it is important to incorporate users’ perceptions into it. Users of a

signalized intersection in vehicles constitute a heterogeneous mix, as traffic – in many parts

of the world – is a mixture of substantial proportions of different vehicle types. This paper

attempts to evaluate the influence of user’s vehicle class on perceived LOS at signalized

intersections. The study does not consider the perceptions of pedestrians/non-motorized

users of the signalized intersections. An extensive user perception survey involving

approximately 8500 users in motorized two-wheelers, three-wheelers, cars, and buses at

fifteen signalized intersections across India populates the database. Employing this database,

aggregate and disaggregate ordered probit models for users’ perceived LOS are developed.

The aggregate level model is an overall user-perceived LOS model, that doesn’t distinguish

between users’ vehicle class. From the model results it has been concluded that along with

perceived waiting time, other perception related factors such as road surface quality,

visibility of traffic signal from the queue, presence of signs and road marking, presence of

pedestrians, presence of heavy vehicles, obstructions, and aesthetics have significant effect
on the users’ perceived LOS. The findings highlight the necessity of incorporating qualitative

parameters in LOS analysis in addition to the conventional service measure (i.e. delay).

Keywords: level of service, user perception, signalized intersection, ordered probit

model, heterogeneous traffic

Introduction

A traffic facility is a ‘service’ supplied to make travel safe and comfortable. Among various such

facilities, ‘signalized intersections’ are notorious bottlenecks in road transportation networks.

Methodic planning is vital in maintaining their performance efficiency, and therefore it becomes

important to measure serviceability at signalized intersection. Highway Capacity Manual 2010

uses control delay as the service measure, which is based on the experts perception (Pecheux,

Pietrucha, and Jovanis 2000) but does not adequately reflect users’ perception of the operational

conditions being provided. Researchers have suggested that delay alone will not suffice as the sole

service measure as it ignores other quantitative and qualitative aspects that influence LOS

(Chakroborty and Kikuchi 1990; Pecheux, Pietrucha, and Jovanis 2000; Zhang 2004). Other than

delay, researchers have developed a wide variety of LOS rating systems considering safety,

combined safety and delay, etc. as the service measures. For example, Zhang and Prevedouros

(2002) used combined delay and safety index as the LOS indicator. By considering vehicle

conflicts, Pan et al. (2007) came up with safety LOS. In another study, Pan et al. (2008) developed

a comprehensive LOS performance measure by combining control delay and risk index.

Researchers have also expanded the existing six LOS classes into nine, or more (Baumgaertner

1996; Brilon and Estel 2010; Clark and MIHT 2008).

Despite LOS being associated with users’ perception of operational conditions, studies that

associate them are sparse (Sutaria and Haynes 1977; Pecheux, Pietrucha, and Jovanis 2000). A
pioneering study on perceived LOS at signalized intersection is apparently the one by Sutaria and

Haynes (1977), where they requested to rate factors before and after viewing the video films. While

they found delay to be the most influencing factor, other factors changed ranking before and after

viewing the videos. Pecheux et al. (2000) came up with 15 potential factors that may influence the

users’ LOS perception. Controlled laboratory studies were carried out for data collection at the

Pennsylvania Transportation Institute (PTI). Video data from 24 approaches and responses from

100 subjects were used. The factors that influence the perceived service quality at signalized

intersection were categorized into operational, geometrical and others by Flannery et al. (2006).

For assessing user perception of a transportation facility, a generalized fuzzy approach was

put forward by Lee (2006). Traffic signal waiting time, the length of gaps in traffic on the cross-

street, traffic signal operations, traffic signal visibility, information guidance systems, physical

features of the intersection etc. were the factors considered in their study. The application of fuzzy

neural networks–based approach to estimate the user perceived LOS of signalized intersection was

initiated by Chen et al. (2009). A methodology based on visualization was applied to collect user

perceptions of LOS. VISSIM was used to simulate the subject intersection and the animation

videos were recorded. For evaluating the user perceived turning movement LOS they considered

factors like waiting time, the volume of pedestrians, the volume of bicycles, etc. Zhang and

Prevedouros (2011) came up with the concept of composite LOS measure by incorporating the

most influencing factors that affect the users’ perception. A study on drivers’ perception of LOS

at signalized intersections was carried out by Jou et al. (2013) especially for motorcycle riders and

car drivers in Taiwan. Pre-recorded videos of different LOSs were presented to the respondents in

the survey. In their study, the factors influencing user perceived LOS were divided into seven
categories namely-traffic flow, pavement condition, geometric highway design, traffic control,

distribution of vehicle types, hardware facility, and weather.

There are very few studies that considered the effect of vehicle type on user perceived LOS

at signalized intersections. In India, people are from different economic backgrounds and have

different living standards. Moreover, the presence of wide variety of vehicle classes and the

absence of any lane discipline all together make the scenario complex. Delay may not be the only

factor that influences their perception. A study by Sekhar et al. (2016) stated that the increase in

traffic volume and the underdeveloped road network along with the poor maintenance system in

India resulted in deteriorations to the existing pavement leading to poor pavement surface quality.

The presence of potholes, cracks, and rutting not only causes congestion and accidents but also

leads to the reduction in the speed of the vehicles and causes severe discomfort to the users. In

such cases, the users may give higher priority to road surface quality than delay. Similar is the case

with poor pedestrian facilities. Poorly maintained foot-over bridges and the longer cycle time at

the signalized intersections lead to the pedestrian noncompliance with traffic signal. This would

significantly cause discomfort to the motorized vehicle users. Therefore, the use of control delay

as the sole service measure cannot be justified for Indian traffic scenario. Hence, there is a need to

identify the various factors influencing the users’ perception of LOS at signalized intersection

under heterogeneous traffic condition. Further, the identified factors need to be incorporated to

develop a user perceived LOS model at signalized intersections. Hence, the objectives of the

present work are to develop user perceived level of service model in order to quantify the impact

of the various influencing factors and to evaluate the influence of vehicle type on the users’

perceived level of service of signalized intersection.


Ordered Probit for Perceived LOS

When a user identifies a LOS category, the user is in fact making a discrete choice, made after

accounting for various factors. So in order to evaluate the user perceived LOS, a statistical

approach that accounts ordered and discrete data is needed. The traditional multinomial logit and

probit model doesn’t consider the ordered nature of the data and can result in the reduction of

estimation efficiency (Amemiya 1985). Ordered probit model is a better tool for modeling ordered

and discrete data (Greene and Hensher 2010; Washington, Karlaftis, and Mannering 2010). An

ordered probit model is a type of ordinal regression, where the dependent variable is discrete and

ordered in nature and the error term follows standard normal distribution. Hence, it has been

adopted here to model users’ perceived LOS at signalized intersection. The general form of the

model is given by Equation 1.

𝑦𝑖∗ = 𝛽𝑋𝑖 + 𝜀𝑖 (1)

where, 𝑦𝑖∗ is an unobserved latent continuous dependent variable, 𝑋𝑖 is a vector of variables that

explains 𝑦𝑖∗ and are assumed to be independent of 𝜀𝑖 , 𝛽 is a vector of estimable parameters and 𝜀𝑖

is an error term or the disturbance, assumed to follow normal distribution N [0,1].

The perceived LOS, which takes rating from excellent to very poor are converted to ordinal

data and that can be obtained from the model as follows:

0, 𝑖𝑓 𝑦𝑖∗ ≤ 0 (𝐸𝑥𝑐𝑒𝑙𝑙𝑒𝑛𝑡)
1, 𝑖𝑓 0 ≤ 𝑦𝑖∗ ≤ µ1 (𝑉𝑒𝑟𝑦 𝑔𝑜𝑜𝑑)
2, 𝑖𝑓 µ1 ≤ 𝑦𝑖∗ ≤ µ2 (𝐺𝑜𝑜𝑑)
𝑦𝑖 = (2)
3, 𝑖𝑓 µ2 ≤ 𝑦𝑖∗ ≤ µ3 (𝐹𝑎𝑖𝑟)
4, 𝑖𝑓 µ3 ≤ 𝑦𝑖∗ ≤ µ4 (𝑃𝑜𝑜𝑟)
{ 5, 𝑖𝑓 𝑦𝑖∗ ≥ µ4 (𝑉𝑒𝑟𝑦 𝑝𝑜𝑜𝑟)
where µj represent the threshold values that are estimated along with the vector β, to categorize the

observed variable y. The probability of users’ perception towards different LOS can be expressed

as

𝑃(𝑦𝑖 = 0) =  (0 − 𝛽𝑋𝑖 )

𝑃(𝑦𝑖 = 1) =  (µ1 − 𝛽𝑋𝑖 ) −  (0 − 𝛽𝑋𝑖 )

𝑃(𝑦𝑖 = 2) =  (µ2 − 𝛽𝑋𝑖 ) −  (µ1 − 𝛽𝑋𝑖 )

𝑃(𝑦𝑖 = 3) =  (µ3 − 𝛽𝑋𝑖 ) −  (µ2 − 𝛽𝑋𝑖 )

𝑃(𝑦𝑖 = 4) =  (µ4 − 𝛽𝑋𝑖 ) −  (µ3 − 𝛽𝑋𝑖 )

𝑃(𝑦𝑖 = 5) = 1 −  (µ4 − 𝛽𝑋𝑖 ) (3)

where  is the cumulative distribution function of the standard normal distribution for 𝜀𝑖 .

Data Collection

Council of Scientific and Industrial Research-Central Road Research Institute (CSIR-CRRI) in

coordination with prominent academic institutes in the country has taken up a multi-institutional

project named “Development of Indian Highway Capacity Manual (Indo-HCM)”. The data

collected for that project has been used for the present study. Fifteen signalized intersections from

cities of New Delhi, Kolkata, Mumbai, Surat, Baroda, and Ahmedabad from India are selected for

the data collection. The selected intersections are either three or four-legged. All the signals are

pre-timed and are not manually controlled. Some intersections included in the study has very good

flow characteristics with wide approaches, flared geometry at the stop-line, and exclusive left-turn

lanes, whereas the traffic flow at some of the intersections is influenced by the pavement

conditions, roadside activities and parking. Hence, this data represents the wide variation of traffic
and geometric characteristics at the intersections prevailing in the Indian context. Table 1 gives

the location details of the study intersections.

An extensive user perception questionnaire survey was carried out at the selected

signalized intersections. Trained personnel who are native speakers and good at English conducted

the survey. Survey work was carried out from 22nd September 2015 to 14th March 2016. The user

perception survey was designed to understand the users’ overall perceived LOS at the signalized

intersection, the factors influencing their perceived LOS and the users’ satisfaction level with these

factors. The exact survey locations include commercial buildings, workplaces, fuel stations and

bus depots, around the intersection where the users can respond comfortably. The socio-

demographic (gender, age), travel-related (frequency and purpose of trip) and perception-related

information was collected from the user perception questionnaire survey. Only the perceptions of

adult users of motorized two-wheelers, motorized three-wheelers, cars, and buses are included in

this study. A total sample of 8458 responses was obtained from the user perception questionnaire

survey.

(Table 1 here)

From previous studies (Othayoth and Krishna Rao 2017; Othayoth and Rao 2017), factors

influencing the perceived LOS were identified and incorporated in the questionnaire. The factors

considered include waiting time, road surface quality, clear/legible signs and road marking,

visibility of the signal from the queue, presence of pedestrians, presence of heavy vehicle,

obstructions due to bus stops and parking, and aesthetics. Instead of using technical terms, more

general terms were used in the questionnaire. For example ‘waiting time’ was used instead of

‘delay’, ‘road surface quality’ was used instead of ‘pavement surface quality’. Qualitative (poor,
moderate or good) data were collected from the respondents regarding the level of satisfaction with

each factor for the subject intersection. A retrospective approach (the users’ time estimation being

from memory) with category rating technique is used to obtain the perceived waiting time from

the users. In category rating, the surveyor presents the temporal interval and the respondents locate

the perceived waiting time in one of the ‘m’ ordered categories. Finally, the users were requested

to give the overall rating of the intersection in A-F scale, where ‘A’ indicates ‘very good’ and ‘F’

indicates ‘very poor’ operating condition. Before interviewing, it was ensured that the respondent

has traveled through the concerned intersection approach (leg) the very same day or the day before

the interview. While interviewing the bus users, it was made clear that the factor ‘presence of

heavy vehicle’ means the vehicle type other than bus. For other vehicle type users this was not

explained. Similarly, some of the factors considered in the questionnaire, e.g. ‘visibility of traffic

signal from the queue’ may not be important for bus users compared to other vehicle type users.

However, all the considered factors are made common for all the users irrespective of their vehicle

type. Table 2 gives the list of variables along with their descriptions collected through the

questionnaire survey.

(Table 2 here)

Data Analysis

Figure 1 shows the classification of users based on their vehicle type. Of the total survey

respondents, just over half are motorized two-wheeler users. The results of socio-demographic

analysis is shown in Figure 2. In the study sample, 85% respondents were male. Age was

categorized into three groups -18 to 35, 35 to 55 and over 55 years. Of all the respondents, 49% of

the users were from the age group of 35-55 years. Trip purposes considered in the study are work
trips, education trips and discretionary trips (social, shopping, recreation etc.). Figure 3 shows the

purpose-wise percent trips made by the users. Figure 4 shows the trip frequency in percentage and

only 9% of the respondents’ trips are discretionary. This shows that majority of the respondents

are commuters indicating they are very familiar with the subject intersection. The distribution of

user perceived LOS of the whole study sample is shown in Figure 5. From the survey; it was

observed that a large majority of the users perceived that the intersections are operating at LOS C

followed by LOS D and LOS B. Among the 8458 respondents, 47.3% (4003) users perceived LOS

C category, whereas the percent number perceived for LOS A and LOS F is 1% and 2.5%

respectively. Figure 6 shows the various LOS categories perceived by different vehicle type users.

All the vehicle type users perceive all the six LOS categories.

(Figure 1 here)

(Figure 2 here)

(Figure 3 here)

(Figure 4 here)

(Figure 5 here)

(Figure 6 here)
For model development, it is important to select those independent variables that are

significantly contributing to the dependent variable. Including more number of variables in the

model may give a better fit but results in model with high variance. Analysis was carried out to

understand the strength and direction of association between socio-demographic, travel-related

and perception related factors with the perceived LOS. Spearman’s correlation analysis has been

carried out for the same and from the results it was found that gender (p-value = 0.790), age (p-

value = 0.431), and frequency of trip (p-value = 0.80) are not significantly influencing the users’

perceived LOS at signalized intersection. Whereas, along with purpose (p-value = 0.056), all the

perception related factors considered in the study are having a significant correlation with the

users’ perceived LOS at 5% significance level. The sign of the Spearman’s correlation coefficient

indicates the direction of association between the independent and dependent variable. The

correlation coefficient between waiting time and LOS is 0.6 indicating that there is a strong

positive correlation between them. It means that there is a shift towards LOS F category when the

perceived waiting time increases. Also, it was found that none of the independent factors

considered are significantly correlated with each other. For modeling the user perceived LOS, only

those factors that are found to have significant correlation with it have been further considered.

Perceived LOS Model

Ordered probit models were developed using NLOGIT 4.0 software package. The data

collected from twelve signalized intersections (from New Delhi, Mumbai, and Kolkata) which

comprises 85% of the whole data has been used for calibrating the model. The remaining data has

been used for model validation. A conversion procedure was first performed to convert the user

perceived LOS rating to ordinal data. Ordinal data corresponding to LOS A is zero, and that for

LOS F is five. The result of the ordered probit model estimation is presented in Table 3. The model
is tested at 5% significance level. A positive sign of the coefficient indicates that an increase in

that variable increases the probability of perceiving LOS F, and a negative sign indicates the

increase in that variable increases the probability of perceiving LOS A. From the results of the

ordered probit modeling, it can be observed by comparing the estimated coefficients with their

standard errors that all the explanatory variables considered for modeling are statistically

significant. The higher coefficient for variables like perceived waiting time indicates that increase

in the perceived waiting time increases the probability of perceiving worst level of service like

LOS E and F. Purpose of trip is having a negative coefficient, which indicate that the users

performing discretionary trips perceives better LOS than users who perform work trips. Frequency

of discretionary trip is less compared to work trips. Hence, users performing work trips are more

familiar with the subject intersection and the combined effect of all the factors considered may

have made them to rate LOS as poor. The negative sign of the coefficient for the road surface

quality shows that better road surface quality leads the users to perceive better LOS grades (LOS

A, B etc.). Similar is the case with visibility of traffic signals, presence of clear/legible road signs

and markings, and aesthetics. The presence of appropriate signboards and road markings reduces

the dilemma of the users. Hence, if proper signs and road markings are present users tend to

perceive better LOS grades. Visibility of traffic signals from the queue is another factor influencing

users’ LOS perception. Good visibility of the signals from the queue leads the users to perceive

better LOS. Scenery/aesthetics has the least significance in influencing the users’ LOS perception

at signalized intersections, but its level of significance is nevertheless high (t-statistic = -6.67) in

influencing the users’ LOS perception at signalized intersection. This may be because there are

other factors like waiting time and road surface quality, which are more important to the users than

aesthetics.
(Table 3 here)

Presence of heavy vehicles, presence of pedestrians and obstructions have negative

coefficients. This indicates that increase in these variables increases the probability of perceiving

worst LOS. The presence of heavy vehicles has a higher impact on the maneuverability of vehicles

like motorized two-wheelers and motorized three-wheelers. Hence, the users of these vehicle types

may rate the overall quality of service of the signalized intersection as poor once the presence of

heavy vehicles increases. Presence of pedestrians actually means those pedestrians who are

crossing the road. Even if foot-over bridges are available it is common that most of the pedestrians

cross when the signal is green for the vehicles. Longer cycle time may be one of the reasons for

pedestrian noncompliance with traffic signal. This will lead to an increase in the number of

accidents with associated increase in waiting time, reduction in the speed of the vehicles etc.

Obstructions considered in the study includes the obstruction to through traffic by side frictions

like parking, bus stops etc. The number of unauthorized on-street parking utilizes the width of the

road and the effective width available to the motorized vehicle users decreases. Bus stops in the

near vicinity of the intersection cause blockage to the through movement of other vehicles. Hence,

there is a chance of users rating LOS at poor grades.

Pseudo 𝑅 2 of the model is 0.415, which shows that the model accounts for a good deal of

the variation in the observed data. From the model estimation result, it can be concluded that other

than delay there are significant factors influencing the users’ rating of LOS. The estimates of the

ordered probit model imply,


𝑦 ∗ = 0.587 − 0.026 𝑃𝑢𝑟𝑝𝑜𝑠𝑒 + 1.142𝑃𝑊𝑇 − 0.145 𝑅𝑆𝑄 − 0.278𝑉𝑇𝑆
(4)
− 0.178 𝑆𝑅𝑀 + 0.194 𝐻𝑉 + 0.198𝑃𝐸𝐷 + 0.224 𝑂𝐵𝑆

− 0.200 𝐴𝐸𝑆

where, 𝑦 ∗ = unobserved latent continuous dependent variable, PWT = perceived waiting time, RSQ

= road surface quality, VTS = visibility of traffic signals, SRM-presence of signs and road

markings, HV = presence of heavy vehicles, PED = presence of pedestrians, OBS = presence of

obstructions, AES = aesthetics.

The user perceived LOS expressed in ordinal scale ( 𝑦𝑖 ) can be obtained from the above model by

using the estimated threshold values in Table 3 as follows:

0, 𝑖𝑓 𝑦𝑖∗ ≤ 0
1, 𝑖𝑓 0 ≤ 𝑦𝑖∗ ≤ 2.494 (5)
2, 𝑖𝑓 2.494 ≤ 𝑦𝑖∗ ≤ 5.076
𝑦𝑖 =
3, 𝑖𝑓 5.076 ≤ 𝑦𝑖∗ ≤ 7.211
4, 𝑖𝑓 7.211 ≤ 𝑦𝑖∗ ≤ 8.071
{ 5, 𝑖𝑓 𝑦𝑖∗ ≥ 8.071

Marginal Effects

Marginal effects are computed to assess the impact of each independent variable (Xi) on the user

perceived LOS. The marginal effects indicate the relative importance of each variable in

determining the likelihood to select a particular LOS category. These are computed in order to

evaluate the effect of a unit change in the independent variable on each of the perceived LOS. It

gives the direction and magnitude of the effect of change in the independent variable. The marginal

effects are computed using Equation 6.


∂P(yi = j)
= (𝑖 (µ𝑗−1 − 𝛽𝑋𝑖 ) − 𝑖 (µ𝑗 − 𝛽𝑋𝑖 )) β, j = 0,1,2.. (6)
∂x

All the variables are previously defined. Table 4 gives the marginal effect of the various

independent factors considered in the ordered probit model on the perceived LOS. It can be seen

that marginal effects are almost zero for LOS A and LOS F. This may be because there were very

small number of observations for LOS A and LOS F in the study sample compared to other LOS

categories. The positive sign shows that keeping all the other variables constant a unit increase in

the variable increases the probability of perceiving the corresponding LOS and vice-versa.

(Table 4 here)

The results of marginal effects show that the perceived waiting time at the signalized

intersection is quite influential on user perceived LOS. Keeping all the other factors constant, unit

increase in perceived waiting time inflates the probability of perceiving LOS D by 0.3618 units

and declines perceiving LOS B by 0.0691units. With the improvement in the road surface quality,

the probability of perceiving LOS B increases by 0.0088 units, LOS C by 0.0384 units, whereas

the probability of perceiving LOS D reduces by 0.046 units. As the presence of pedestrians, heavy

vehicles and obstructions increases, users tend to perceive worst grades of LOS. The marginal

effect shows that a unit increase in all the above factors leads to a decrease in the probability of

perceiving LOS B and LOS C whereas elevates the probability of perceiving LOS D and LOS E.

On the other hand, a unit increase in the visibility of traffic signal from the queue, presence of

signs and road markings and aesthetics, inflates the probability of perceiving LOS B and LOS C

and reduces the probability of perceiving LOS D and LOS E.


Model Validation

The data collected from three intersections in Gujarat state, India has been used for validating the

developed ordered probit model. A total of 1268 responses (15% of the collected data) were

available for model validation. LOS distribution of the validation data set reveals that the

perception towards LOS A and LOS F is negligible. A large majority of the users perceived that

the intersections are operating at LOS C. The model validation results are represented in the form

of a prediction-success table. It is a cross-classification between the observed LOS and the

predicted LOS. Table 5 gives the results of the model validation. The unobserved variable 𝑦𝑖∗ is

obtained by substituting the values of the independent variables collected from the three

intersections in the calibrated model. The observed variable yi can be obtained from the thresholds

for each unobserved variable 𝑦𝑖∗ . The observed share for LOS A is 2% whereas the predicted share

by the model is 1%. This is because; the calibration data set had very few LOS A observations.

LOS C has the highest percentage of correct predictions at 91%. The percentage correctly predicted

for LOS B is 68% whereas that of LOS D is 74%. The sum of the diagonal elements of the

prediction-success table gives the overall prediction success. For the developed model the overall

prediction success rate is 81%. Hence, it can be concluded that the developed model accounts for

the users’ perception of level of service quite well in the heterogeneous traffic condition.

(Table 5 here)

Perceived LOS Model Disaggregated Among Users of Different Vehicle Class

In heterogeneous traffic, it is logical to assume that the perception of motorized two-wheeler users

toward LOS differs from that of car users. The factors that influence the perceived LOS of one
vehicle type user may be different from that of other vehicle type user. Hence, an attempt is made

to evaluate the influence of users’ vehicle type on their perceived LOS of signalized intersection

at a disaggregate level. Separate models were developed for LOS perceptions of users of each class

of vehicles. The objective is to check whether the assumption that users of different vehicle type

perceives differently is true or not. Table 6 gives the ordered probit model estimation result for

each class of vehicle.

Purpose of trip has a positive coefficient for motorized two-wheeler (coefficient = 0.0247)

and motorized three-wheeler (coefficient = 0.023) users whereas it is negative for car (coefficient

= -0.057) and bus (coefficient = -0.090) users. Positive sign of purpose of trip by motorized two-

wheeler and three-wheeler users indicates that users making discretionary trips by two-wheeler

and three-wheeler perceives worse LOS like LOS D, E and F. Whereas users who make work trips

by the above modes perceive better LOS. In case of car and bus users, as the discretionary trips

made increases users perceive better LOS grades and as the work trips made increase users

perceive worst LOS grades. All the factors considered in the study are significantly influencing

the motorized two-wheeler users perceived LOS. Positive coefficient of perceived waiting time

and presence of heavy vehicles shows that an increase in these variables increases the probability

of perceiving LOS F. Negative coefficient for presence of signs and road marking shows that

presence of clear/legible signs increases the probability of perceiving LOS A. The pseudo 𝑅 2 of

the model is 0.429, which shows that the model provides a good fit of the predictions to the

observed data.

Similar to motorized two-wheeler users perceived LOS, all the perception-related factors

are significantly influencing the motorized three-wheeler users perceived LOS. As expected, it

was found that perceived waiting time is the most significantly influencing factor. Positive
coefficient of perceived waiting time shows that an increase in the magnitude of that variable

increases the probability of perceiving LOS F. Negative coefficient for road surface quality shows

that good pavement condition increases the probability of perceiving LOS A. The pseudo 𝑅 2 of

the model is 0.389. All the considered factors are significantly influencing the car users perceived

LOS. Perceived waiting time is having most impact on the users perceived LOS. The presence of

heavy vehicles and the presence of pedestrians and presence of obstructions are having positive

coefficient indicating an increase in the magnitude of these variables increases the probability of

perceiving LOS F. Road surface quality, the presence of signs and marking, visibility of traffic

signals from the queue and aesthetics has negative coefficients. The pseudo 𝑅 2 of the model is

0.419. Except for visibility of traffic signals from the queue and presence of heavy vehicles, all

the perception-related factors are significantly contributing to the bus users perceived LOS. This

may be because, for bus users, the visibility of traffic signal may not be an influencing factor.

Perceived waiting time is the most significantly influencing factor, followed by road surface

quality and presence of pedestrians. The presence of obstruction is the least significant factor. The

pseudo 𝑅 2 of the model is 0.448.

(Table 6 here)

Comparing the results of all the four vehicle classes, it was found that, except bus users,

for all the other vehicle classes considered, all the factors considered in the study are significant.

For bus users, the visibility of traffic signals from the queue and presence of heavy vehicles are

not significant. The bus being a heavy vehicle, the presence of other heavy vehicles may not be

influencing the perception of bus users. Among the perception related factors, while the perceived
waiting time is the most significant factor for all the vehicle type users, the other factors also rank

somewhat similarly for all the vehicle types. From the results, it can be concluded that there is no

need for separately assessing the user perceived LOS at signalized intersection based on vehicle

types.

Conclusions

Users are the integral component of any transportation system. Incorporating user perceptions into

LOS would provide a more practically useful and theoretically sounder basis for design and

operation of signalized intersections. Also, an understanding of various factors and their effect on

user perception would increase the likelihood of successful projects by better informing investment

decisions. This study evaluates user perception of LOS at signalized intersection under

heterogeneous traffic conditions. The research contribution in this study is built on a massive user

perception survey data set collected from a real-world setting.

An ordered probit model was employed to elucidate the factors that influence the LOS

perception of different vehicle type users. Ordered probit modeling of users’ perceived LOS has

been carried out at aggregate level as well as disaggregate level. The aggregate level model is an

overall user-perceived LOS model, that doesn’t distinguish between users by vehicle class.

Waiting time is the most significant factor contributing to the users’ perception of service quality.

Presence of heavy vehicles, pedestrians etc., have a significant negative influence on perceiving

better LOS grades. In other words, the model results suggest that the perceived LOS deteriorates

quickly with an increase in the perceived waiting time, number of heavy vehicles, number of

pedestrians and obstructions. Also, road surface quality, presence of clear/legible signs and road

markings, visibility of traffic signals from queue and aesthetics have a positive influence on

perceived LOS. From the analysis, it was found that socio-economic variables (gender, age) and
trip frequency are not significantly influencing the users perceived LOS at signalized intersection.

All the perception-related factors and purpose of trip considered in the study are significantly

influencing the user rating of LOS. A planner may thus use the marginal effects to make informed

decisions on apportioning of budget to maximize user satisfcation.

At a disaggregate level, the model has been developed for LOS as perceived by users of

different vehicle classes separately. The result not only includes identifying the most influencing

factors but also quantifies the effect of these variables on perceived LOS by different vehicle

classes. Motorized two-wheelers, motorized three-wheelers, cars, and buses are the vehicle types

considered for the present study. Among the perception related factors, the perceived waiting time

is the most significant factor for all the vehicle type users. All the other factors also rank somewhat

similarly for all the vehicle types. From the results, it can be concluded that there is no need for

separately assessing perceived LOS at signalized intersection based on users’ vehicle types.

Given the number of factors involved in model development, it may be difficult to assess

user perceived LOS in general. However, the developed model considers the universe of factors

that influences the users’ perceived LOS at signalized intersection under heterogeneous traffic

condition. Any improvements made in these identified factors can have a significant influence on

the users’ perceived LOS. The application of the study methodology ranges from setting priority

for the improvement or modification of the various factors like road surface quality, pedestrian

facility, road marking etc. The present work has not studied the influence of motorized two-

wheelers on the perception of users of other vehicle types. Study of the same may leads to value

addition, as the presence of motorized two-wheelers causes annoyance to users of other vehicle

types. Pedestrians and non-motorized users are a significant set of road users in India whose

perceptions are not incorporated in this study. Further studies can be carried out by incorporating
the perceptions of pedestrians and non-motorized users to obtain a unified LOS measure at

signalized intersection.

Acknowledgments

This work is a part of the sponsored research project “Development of Indian Highway Capacity

Manual (Indo-HCM)” coordinated by CSIR-CRRI, New Delhi, India The authors gratefully

acknowledge the support provided by the members of this sponsored research project. The authors

would also like to express their sincere thanks to respondents of the questionnaire survey.

References

Amemiya, T. 1985. Advanced Econometrics. Harvard University Press.

Baumgaertner, W.E. 1996. “Levels of Service--Getting Ready for the 21st Century.” ITE Journal

66 (1).

Brilon, Werner, and Anja Estel. 2010. “Differentiated Analysis of Level of Service F within the

German Highway Capacity Manual.” Transportation Research Record: Journal of the

Transportation Research Board 2173: 36–44. doi:10.3141/2173-05.

Chakroborty, P., and S. Kikuchi. 1990. “Application of Fuzzy Set Theory to the Analysis of

Capacity and Level of Service of Highways.” In In First International Symposium on

Uncertainty Modeling and Analysis, 146–50. doi:10.1109/ISUMA.1990.151343.

Chen, X, D Li, N Ma, and C Shao. 2009. “Prediction of User Perceptions of Signalized Intersection

Level of Service Based on Fuzzy Neural Networks.” Transportation Research Record:

Journal of the Transportation Research Board 2130: 7–15. doi:10.3141/2130-02.

Clark, Ian, and M MIHT. 2008. “Level of Service F: Is It Really Bad as It Gets?” In IPENZ

Transportation Group Conference.


Flannery, A., D. McLeod, and N. J. Pedersen. 2006. “Customer-Based Measures of Level of

Service.” Institute of Transportation Engineers. ITE Journal 76 (5): 17.

Greene, W. H., and D. A Hensher. 2010. “Modeling Ordered Choices: A Primer.” In . Cambridge

University Press.

“Highway Capacity Manual.” 2010. Transportation Research Board, National Research Council,

Washington, DC.

Jou, Rong Chang, Cheng Chen Kou, and Yi Wen Chen. 2013. “Drivers’ Perception of LOSs at

Signalised Intersections.” Transportation Research Part A: Policy and Practice 54 (1): 141–

54. doi:10.1016/j.tra.2013.07.013.

Lee, Dongmin. 2006. “A Generalized Approach for Analyzing Transportation User Perception

Using Fuzzy Sets.” (Doctoral Thesis, Pennsylvania State University).

Othayoth, Darshana, and K. V. Krishna Rao. 2017. “Factors Influencing Level of Service for

Motorized Vehicles at Signalized Intersection under Mixed Traffic Condition.”

Transportation in Developing Economies 3 (2). Springer International Publishing: 16.

doi:10.1007/s40890-017-0045-x.

Othayoth, Darshana, and K.V.Krishna Rao. 2017. “Statistical Analysis of User Perceived Level of

Service at Signalized Intersection.” In 96th Annual Meeting of the Transportation Research

Board, Washington, D.C.

Pan, F., Lu, J., Xiang, Q., and McAvoy, D. S. 2008. “Safety Based Signalized Intersection Level

Service.” The First International Symposium on Transportation and Development–

Innovative Best Practices (TDIBP 2008). doi:10.1017/CBO9781107415324.004.

Pan, F., Lu, J., Xiang, Q., and Zhang, G. 2007. “Safety Level of Service at Highway Signalized

Intersections.” International Conference on Transportation Engineering, ASCE, 1499–1504.


Pecheux, K.K., M.T. Pietrucha, and P.P. Jovanis. 2000. “User Perception of Level of Service at

Signalized Intersections: Methodological Issues.” In Transportation Research Circular E-

C018: Proceedings of the Fourth International Symposium on Highway Capacity, 322–335.

http://onlinepubs.trb.org/onlinepubs/circulars/ec018/28_43.pdf.

Sekhar, Ch Ravi, J. Nataraju, S. Velmurugan, Pradeep Kumar, and K. Sitaramanjaneyulu. 2016.

“Free Flow Speed Analysis of Two Lane Inter Urban Highways.” Transportation Research

Procedia 17: 664–73.

Sutaria, T.C., and J.J. Haynes. 1977. “Level of Service at Signalized Intersections.” Transportation

Research Record 644: 107–13.

Washington, S.P., M.G. Karlaftis, and F.L. Mannering. 2010. “Statistical and Econometric

Methods for Transportation Data Analysis.” In . CRC Press.

Zhang, Lin. 2004. “Signalized Intersection Level-of-Service That Accounts for User Perceptions.”

(Doctoral dissertation, University of Hawai’i).

Zhang, Lin, and Panos D Prevedouros. 2011. “User Perceptions of Signalised Intersection Level

of Service Using Fuzzy Logic.” Transportmetrica 7 (4): 279–96.

doi:10.1080/18128601003667460.

Zhang, Lin, and Panos D. Prevedouros. 2002. “Signalized Intersection LOS That Accounts for

Safety Risk.” Transportation Research Record 3288.

Table 1 Location Details of the 15 Study Intersections

Intersection
Name of Intersection City Coordinate
Identity
A Aashirwad Chowk New Delhi 28° 35' 47.9'' N 77° 2' 59.9'' E
B Deepali Chowk New Delhi 28° 41' 53.1'' N 77° 7' 11.6'' E
C Depot Chowk New Delhi 28° 35' 41.3'' N 77° 4' 18.7'' E
Firozshah-KG Marg
D New Delhi 28° 37' 22.1'' N 77° 13' 31.5'' E
Junction
E NTPC Chowk New Delhi 28° 36' 3.6'' N 77° 22' 20.9'' E
F PTS Chowk New Delhi 28° 31' 58.1'' N 77° 11' 45.0'' E
G Stadium Chowk New Delhi 28° 35' 23.6'' N 77° 20' 9.5'' E
H Vardhaman Chowk New Delhi 28° 35' 29.2'' N 77° 3' 27.6'' E
I Vijay Char Rastha Ahmedabad 23°02'34.1"N 72°32'56.01"E
J GEV Circle Vadodara 22°18'37.7"N 73°9'54.3"E
K Rangila Park Intersection Surat 21°10'29.9"N 72°48'18.9"E
IIT Bombay Main Gate
L Mumbai 19° 7' 30.3'' N 72° 54' 59.7'' E
Intersection
M Shivaji Chowk Mumbai 19° 4' 28.3'' N 72° 59' 52.0'' E
N Kona Intersection Kolkata 22°34'31.6"N 88°18'06.5"E
O Rashbehari Intersection Kolkata 22°31'02.4"N 88°21'08.6"E

Table 2 Variable Descriptions

Variable Description
Gender 1 = ‘male’, 2 = ‘female’
1 = ‘18-35 years old’ , 2 = ‘35-55 years old’, 3 = ‘more
Age
than 55 years old’
Purpose of trip 1 = ‘work’, 2 = ‘education’, 3 = ‘discretionary trips’
Frequency of trip 1 = ‘daily’, 2 = ‘occasionally’
1 = ‘less than 30 seconds’, 2 = ‘30-60 seconds’, 3 =
‘60-120 seconds’, 4 = ‘120-180 seconds’, 5 = ‘180-240
Perceived waiting time
seconds’, 6 = ‘240-480 seconds’, 7 = ‘480-600
seconds’, 8 = ‘more than 600 seconds’
Road surface quality 0 = ‘poor’, 1 = ‘moderate’, 2 = ‘good’
Visibility of traffic signals from
0 = ‘poor’, 1 = ‘moderate’, 2 = ‘good’
queue
Signs and road marking 0 = ‘poor’, 1 = ‘moderate’, 2 = ‘good’
Presence of heavy vehicles 0 = ‘low’, 1 = ‘medium’, 2 = ‘high’
Presence of pedestrians 0 = ‘low’, 1 = ‘medium’, 2 = ‘high’
Obstructions due to parked vehicles,
0 = ‘low’, 1 = ‘medium’, 2 = ‘high’
bus stops
Scenery/aesthetics 0 = ‘poor’, 1 = ‘moderate’, 2 = ‘good’
Perceived LOS Based on users rating on A-F scale
Table 3 Ordered Probit Model Results

Variable Coefficient Standard Error


Constant 0.587 0.101
Purpose of trip -0.026 0.009
Perceived waiting time 1.142 0.047
Road surface quality -0.145 0.009
Visibility of traffic signals from queue -0.278 0.026
Clear/legible signs and road marking -0.178 0.023
Presence of heavy vehicles 0.194 0.026
Presence of pedestrians 0.198 0.016
Obstructions due to parked vehicles, bus stops 0.224 0.029
Scenery/aesthetics -0.200 0.030
Other Estimates Value
Threshold 1 (µ1) 2.494 0.099
Threshold 2 (µ2) 5.076 0.086
Threshold 3 (µ3) 7.211 0.091
Threshold 4 (µ4) 8.071 0.125
Log-likelihood at convergence -5547.08
Log-likelihood at zero -9846.10
McFadden pseudo R-squared 0.415
Number of observations 7190
Table 4 Marginal Effects for the Users’ Perceived LOS

Marginal Effects
Variable
LOS A LOS B LOS C LOS D LOS E LOS F
Purpose of trip 0.0001 0.0016 0.0069 -0.0082 -0.0002 -0.0001
Perceived waiting time -0.0001 -0.0691 -0.3024 0.3618 0.0091 0.0006
Road surface quality 0.0001 0.0088 0.0384 -0.046 -0.0012 -0.0001
Visibility of traffic signals
0.0001 0.0168 0.0736 -0.088 -0.0022 -0.0001
from queue
Signs and road marking 0.0001 0.0107 0.0471 -0.0563 -0.0014 -0.0001
Presence of heavy vehicles -0.0001 -0.0117 -0.0514 0.0615 0.0015 0.0001
Presence of pedestrians -0.0001 -0.0121 -0.0525 0.0628 0.0016 0.0001
Obstructions due to parked
-0.0001 -0.0135 -0.0593 0.0709 0.0018 0.0001
vehicles, bus stops
Scenery/aesthetics 0.0001 0.0121 0.053 -0.0634 -0.0016 -0.0001
Table 5 Prediction-Success Table

Predicted LOS
Observed
Observed Row
Y= 0 Y=1 Y=2 Y=3 Y=4 Y=5 Share
LOS Sum
LOS A LOS B LOS C LOS D LOS E LOS F (%)

Y=0 (LOS A) 7 18 0 0 0 0 25 2
Y=1 (LOS B) 0 104 44 4 0 0 152 12
Y=2 (LOS C) 0 22 681 45 0 0 748 59
Y=3 (LOS D) 0 1 60 225 10 9 305 24
Y=4 (LOS E) 0 0 1 13 10 1 25 2
Y=5 (LOS F) 0 0 0 4 5 4 13 1
Column Sum 7 145 786 291 25 14 1268
Predicted
1 11 62 23 2 1
Share (%)
% Correctly
28 68 91 74 40 31
Predicted
Table 6 Model Results of Perceived LOS of Different Vehicle Type Users

Motorized two- Motorized three-


Car Bus
wheeler wheeler
Std. Std. Std. Std.
Variable Coefficient Coefficient Coefficient Coefficient
Error Error Error Error
Constant 0.669 0.150 1.347 0.225 0.776 0.163 0.446 0.278
Purpose of trip 0.0247 0.018 0.023 0.020 -0.057 0.013 -0.090 0.029
Perceived waiting time 1.117 0.045 0.941 0.062 1.091 0.060 1.207 0.080
Road surface quality -0.213 0.015 -0.123 0.012 -0.186 0.016 -0.154 0.017
Visibility of traffic signals from
-0.358 0.041 -0.431 0.063 -0.194 0.059 -0.282* 0.257
queue
Clear/legible signs and road marking -0.206 0.027 -0.251 0.068 -0.195 0.038 -0.076 0.018
Presence of heavy vehicles 0.16 0.023 0.101 0.028 0.3 0.074 0.174 *
0.130
Presence of pedestrians 0.198 0.018 0.24 0.033 0.253 0.043 0.368 0.039
Obstructions due to parked vehicles,
0.287 0.039 0.235 0.054 0.195 0.039 0.158 0.066
bus stops etc.
Scenery/aesthetics -0.187 0.036 -0.156 0.079 -0.223 0.061 -0.274 0.060
Other Estimates Value Value Value Value
Threshold 1 (µ1) 2.364 0.083 2.389 0.118 2.4512 0.106 2.532 0.120
Threshold 2 (µ2) 5.114 0.091 4.987 0.094 5.155 0.097 5.253 0.094
Threshold 3 (µ3) 7.374 0.094 7.051 0.097 7.23 0.095 7.589 0.104
Threshold 4 (µ4) 8.132 0.130 7.936 0.162 8.107 0.137 8.652 0.160
Log-likelihood at convergence -2306.47 -1112.34 -2015.10 -807.82
Log-likelihood at zero -4040.61 -1820.95 -3471.09 -1464.64
McFadden pseudo R-squared 0.429 0.389 0.419 0.448
Number of observations 3205 1424 2676 1153
*
not significant at 5% significance level
17%

38%
13%

32%
Motorized-three wheelers Bus Car Motorized-two wheelers

Figure 1 Distribution of users based on vehicle type

Gender Age

12%
15%

49%
39%
85%

Male Female 18-35 35-55 >55

Figure 2 Distribution of users based on gender and age


9%
15%

76%

Work Education Discretionary

Figure 3 Purpose wise percent trips made by the users

19%

81%

Daily Occasionally

Figure 4 Trip frequency


50 47.3
45
40
35
29.3

Percentage
30
25
20 15.3
15
10
4.6
5 1.0 2.5
0
LOS A LOS B LOS C LOS D LOS E LOS F
User Perceived LOS

Figure 5 Distribution of user perceived LOS for the whole sample

100
Percent Number Perceived

80

60

40

20

0
Motorized two- Motorized Car Bus
wheeler three-wheeler

LOS A LOS B LOS C LOS D LOS E LOS F

Figure 6 Perceived LOS by different vehicle type users


List of Figures

Figure 1 Distribution of users based on vehicle type

Figure 2 Distribution of users based on gender and age

Figure 3 Purpose wise percent trips made by the users

Figure 4 Trip frequency

Figure 5 Distribution of user perceived LOS for the whole sample

Figure 6 Perceived LOS by different vehicle type users

You might also like