Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
43
Zagreb International Review of Economics & Business, Vol. 27, No. 2, pp. 43-74, 2024
© 2024 Author(s). This is an open access article licensed under the Creative Commons
Attribution-NonCommercial-NoDerivs License (https://creativecommons.org/licenses/by-nc-nd/4.0/).
Faculty of Economics and Business, University of Zagreb and Sciendo. Printed in Croatia.
ISSN 1331-5609; UDC: 33+65
DOI: 10.2478/zireb-2024-0017
* Faculty of Transport and Trafc Engineering, University of Belgrade, Belgrade, Serbia.
+ Corresponding Author E-Mail: m.blagojevic@sf.bg.ac.rs
** Faculty of Transport and Trafc Engineering, University of Belgrade, Belgrade, Serbia.
*** Faculty of Transport and Trafc Engineering, University of Belgrade, Belgrade, Serbia.
Assessment of Customer Satisfaction with Postal Services
– a Statistical Approach
Mladenka Blagojević * +
Nikola Knežević **
Dejan Marković ***
Abstract: In this paper the objective is to seek and measure the level of customer satisfaction and
services rendered in the postal industry through the chosen methodology. The question of
assessment of customer satisfaction with postal services is treated using discriminant anal-
ysis as a statistical approach. It has been proven that by applying discriminant analysis
it is possible to separate users and not users, as well as to determine which parameters
are crucial for discrimination groups. The paper proves that is possible to separate, using
Fisher’s linear discriminant analysis, respondents who are classied as loyal users and
respondents who are classied as occasional users, as well as respondents who are clas-
sied as potential users and respondents who belong to the group of those who never use
postal services. Also, it has been proven that is possible, using predictive Fisher’s linear
discriminant analysis, to classify new respondents into one of the mentioned groups.
Keywords: Postal services; Management; Customers; Satisfaction; Discriminant analysis
JEL Classication: C44, L87, R49
44 Mladenka Blagojević, Nikola Knežević, Dejan Marković
Introduction
Services have become an integral part of human existence and they are one of the
important components of living standards. The service is produced and provided to
meet the needs of non-producers. The provision of the service cannot usually be sep-
arated in space or time, services are usually consumed at the time and place where
they are produced which results from their intangibility. When providing services,
live work is needed above all, and the results of this work are consumed in the pro-
vision.
The growth of services over the past decade has been remarkable. Services are
increasingly attracting attention of many authors. Nowadays in a competitive service
environment people who have the role of marketer in businesses are seeking custom-
er satisfaction to create and improve relationships between businesses and new or
existing customers (Webster and Sundaram, 2009). Customers are seen as the basis
of a company’s protability (Pishgar et al., 2013). Service quality has become a key
marketing tool for achieving competitive differentiation and fostering customer sat-
isfaction and loyalty.
Kotler and Keller (2016), as well as numerous marketing experts, state that suc-
cessful, modern companies in the era of globalization and consumerism should build
their strategies based on marketing and business orientation. That is, in such a way
that a quality offer implies matching the company’s offer and capabilities with the
user’s requirements, and not the other way around. They suggest ways that marketers
can implement to help their companies increase of business volumes and revenues
and state that in the age of consumerism, it is necessary to conduct your marketing
activities proactively. This type of marketing requires constant examination and anal-
ysis of the market to identify new customer needs and satisfy those needs.
There are several theoretical approaches to measuring customer satisfaction, in-
cluding, for example, the differential model of customer satisfaction, the model of
possible reactions, the Kano model, model GAP 5, and others. New and new recom-
mendations and procedures are constantly being created around the world and go
into absolute detail on the issue of achieving customer satisfaction. A small number
of scholarly researches to date have been carried out to classify quality elements
and complete features of services and their associations with customer satisfaction
(Zeithaml et. Al., 2002; Yang and Fang 2004). One of the more generally used tools
for measuring customer satisfaction is SERVQUAL extended by Parasuraman et al.
(1988). Researchers have concentrated more on the close relationship between service
quality and customer satisfaction (Bitner et al., 1990; Parasuraman et al., 1985; Para-
suraman et al., 1988).
Quality is a concept that probably has the most different denitions that can be
accepted as correct. The meaning and scope of this term have changed and expanded
over the years. Stuart (1969) says that there is no single denition of quality, i.e. that
45
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
quality represents the feeling that something is better than something else. It changes
during human life and depends on many aspects of human nature. Crosby (1989) de-
nes it as compliance with requirements. Feigenbaum (1999) says that he represents
the desire of customers. Juran (1979) says that quality is suitability for purpose and
use. According to the ISO standard, quality is dened as the sum of characteristics
of an entity that makes it possible to satisfy stated or unspecied needs. Regardless
of the adopted denition, it can be concluded that the quality of a product or service
represents a set of characteristics aimed at satisfying the expectations, needs, and
demands of clients.
The perception of services is strongly inuenced by their quality. What is the
quality of service? The most popular characteristics generally indicate that it can be
dened as a set of service features that meet customer needs (Urban, 2013). Concern-
ing the characteristics of services (diversity, immateriality, service enterprise-cus-
tomer interaction), the needs and expectations of customers are key aspects of service
quality (Bielawa, 2011). Given the information provided, improving the quality of
services that will allow meeting the changing needs of customers is a prerequisite
and basis for enabling service enterprises to operate in a competitive market (Dziad-
kowie c, 2007).
The literature distinguishes several different classications of service quali-
ty criteria. Kowalik (2020) states that worth mentioning is the classication of the
SERVQUAL (research instrument for measuring service quality) which groups qual-
ity factors into ve sets: features of tangibles, reliability, assurance, responsiveness,
and empathy. Also states that many modern classications derive from the denition
of Gronroos (1988), according to which the quality of services can have a technical
dimension (tech-quality) - which concerns the effect of the service process, and func-
tional (touch-quality) - concerning the course of the service provision process.
Similarly, in this paper, through conducted survey the respondents evaluated
their satisfaction with postal services. The evaluation of the service was performed
through different types of attributes, and the research included users as well as not
users. The main hypothesis of this paper is determining the presence of signicant
differences in the respondents’ answers and is it possible to make a separation, using
Fisher’s linear discriminant analysis, between respondents who are users of postal
services and those who are not, as well as to determine which parameters (variables)
are crucial for discrimination and customer satisfaction. The respondents were indi-
viduals from the territory of the Republic of Serbia, with a sample of 800 respon-
dents. The contribution of the research presented in this paper is the combination
of respondents’ responses and discriminant analysis, to obtain a group of users of
postal services and a group that does not use postal services. Through discriminant
analysis, it would be possible to more precisely distinguish the service attributes that
create the greatest separation between groups and to clearly classication of users
into groups. This leads to the number of users and those who are not, with the cor-
46 Mladenka Blagojević, Nikola Knežević, Dejan Marković
responding subgroups. In the Republic of Serbia every two years the answers of the
respondents could be used for analysis by regulatory authority and for analysis by
discriminant analysis. The result would be monitoring changes in user satisfaction
from two independent sources and monitoring changes in the number of users by
dened groups. This information is important to the postal operator to undertake
marketing and business ventures for better positioning in the market and gaining
more users. Discriminant analysis served as a tool that a proactive marketer can use
to examine and analyze the postal services market, from the customer’s point of view.
In this research, application of discriminant analysis separated loyal (permanent) us-
ers, occasional users, respondents who used postal services but did not do so in the
last year (potential users), and respondents who do not use postal services (i.e. did
not use them in the last year). After discriminating respondents into the groups, how
potential users would be allocated was shown.
The paper is presented in seven sections. In the rst section, the basic principles
of the paper were presented. The second section of the paper presents the nature and
characteristics of postal services as subject of conducted research. The third section
provides the basic characteristics of discriminant analysis and a literature review of
the application of discriminant analysis in trafc and transport. The fourth section of
the paper contains the results of a survey that included 800 respondents. As part of
the third section, the quality attributes evaluated by the respondents are shown. These
attributes were selected based on the systematization of literature by experts in this
eld, the eld of postal trafc. This way of selecting attributes ensures the validity of
the questionnaire. The reliability of the research is ensured by Cronbach’s coefcient
α. In the fth section, the problem is formulated and presented in several stages.
Models based on discriminant analysis were developed to solve the mentioned prob-
lem, the discrimination of respondents into groups was carried out and the procedure
of allocation of new respondents was presented. In the sixth section, the research
discussion is presented, with a special reference to the previously developed alter-
native approach to the discrimination of respondents. In the seventh section general
conclusions are formulated.
The Key Parameters of Quality of Postal Services
The quality of postal services is an issue worthy of interest. The postal service mar-
ket has been dynamically changing. National trends coincide with the trends of glob-
al postal markets. In recent years, the number of traditional letters has been regularly
decreasing with the simultaneous increase in the number of parcels and the increase
in the popularity of courier parcels (Kowalik, 2019). Also, high requirements in
terms of exibility and adaptability to user requirements increase the cost of postal
infrastructure. Postal trafc is a business area in which a complete ‘’door-to-door’’
47
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
and ‘’just-in-time’’ service is provided, which is a rather demanding and expensive
process. It is necessary to adjust the business strategy of postal operators in a way to
maximize the efciency and productivity of work in the current conditions. Also, the
offer should be marketed in a way that recognizes and meets the needs of the envi-
ronment in which the operator operates.
The most important requirement of postal customers for the services provided is
their quality. Quality is an important aspect that has come to the attention of postal
service providers a long time ago. The quality of postal services serves as a basic
tool for the postal company to maintain and increase competitiveness in the market
(Rostasova et al. 2020). Quality is expressed as the conformity of correctly dened
requirements that satisfy customer needs. The denition of the quality of postal ser-
vices emphasizes the goal of quality of service, in which the needs and expectations
of customers are met through a price that represents the value of the service. As with
other service providers, the reasons for interest in quality issues of the postal items
are dened according to several aspects.
Generally, postal services are related to the delivery of parcels, letters, and doc-
uments. Also, the postal transportation process consists following activities: collec-
tion, input sorting, transportation, output sorting, and distribution (Noordin et al.,
2012) and they should be rated as high quality or not high quality.
The requirements of postal service users are constantly increasing, and if a postal
operator wants to expand its services and survive, it must constantly monitor the
demands of the users following their needs. To that end, it is necessary to conduct
continuous research and implement the obtained information into business decisions
(Pavlović et al., 2021).
The demands of users today are very high. They know exactly what they want and
how much they are willing to spend on it. If the operator provides them with a service
that will satisfy their needs, quality service has been achieved. From the aspect of
postal services quality can be dened as the ability to recognize the demands and
needs of users and to perform services within the legal framework. Postal operators
are obliged to transfer and deliver the postal items in the condition in which it was
collected, and to perform postal services under the conditions, in the manner, and
within the deadlines.
Quality control of postal services is one of the crucial elements in the postal mar-
ket. It can be done internally using company resources for that purpose and external-
ly if it is a service that has a monopoly status on the market, and it is of state interest.
To achieve this undertaking, it is necessary to determine adequate factors which
contribute to the overall quality of service for users (considered from the user’s point
of view), as well as the performance of the network, i.e. its ability to ensure the re-
alization of the service on the predetermined way. From the user’s point of view, the
quality of postal services is dened by more parameters such as speed of transport,
the convenience of providing the service, supplementary service capabilities, perfor-
48 Mladenka Blagojević, Nikola Knežević, Dejan Marković
mance capabilities (service availability, service preservation), etc. In the paper Lai et
al. (2022) the authors investigates the antecedents of customer satisfaction with par-
cel locker services in last-mile logistics based on the service quality (SERVQUAL)
model and logistics service quality (LSQ) model.
There are some examples of quality attributes of postal services based on the level
of achieved quality is determined (Table 1).
Table 1: Quality attributes of postal services
Paper Quality attributes of postal services
Kowalik (2020)
Price (price is an important quality feature of both products and services);
Service delivery time;
Scope of service digitization: traditional postal services, hybrid, completely digital;
Multichannel;
Ease of use;
The location/availability of posting/pickup points;
Additional services (parcel machines, home posting, applications, and many more);
Contact with the postal operator;
Postal operator’s image;
Experience with operator
Post of Serbia
(2021)
The availability of the postal service;
Speed and reliability of items transport;
Items security;
The effectiveness of resolving complaints;
User satisfaction with information;
Level of standardization and typication;
Organizational climate and job satisfaction
Heco (2015)
Security - risk reduction, physical and nancial security, guarantee;
Reliability in the provision of services - fulllment of promises, consistency in the provision
of services;
Affordability of prices to users of postal services;
Professionalism and responsibility - willingness and availability of employees to provide a
certain service;
Competence - knowledge and skills, the expertise of staff who communicate with service users;
Accessibility - service availability (suitable working hours, location, waiting time for service);
Friendliness - kindness, respect, understanding, and cordiality;
Communication with users - understandably informing users and respecting opinions;
Credibility - respect;
Professionalism, reputation, and trust enjoyed by the company.
Sengazani
Murugesan et al.
(2020)
Diversity and range of service;
Intensity and depth of service;
Digital and physical security;
Service availability;
Convenient operating time;
Effectiveness of employee skill;
Prompt service to customers;
Employees’ proper behavior;
Consistently pleasing and courteous;
Simplied delivery process;
Structured delivery process;
49
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
Paper Quality attributes of postal services
Sengazani
Murugesan et al.
(2020)
Fool-proof procedure;
Adequate facilities provision;
Adequate personnel provision;
Comfortable HVAC provision;
Equipment/physical layout;
Housekeeping;
The appearance of visual sign boards;
Neat and professional appearance;
Equal treatment;
Service transcendence;
Availability of service in all places;
Sense of public responsibility;
Zhang (2019)
Service convenience;
Service responsiveness;
Service care;
Service tangibility;
Service economy;
Kowalik (2019)
The attractiveness of the institution;
The modernity of equipment;
Staff’s appearance;
Availability of materials;
Punctuality of service delivery;
Faultlessness of service delivery;
Staff’s help in problem-solving;
Compliance with the offer;
Staff’s competence;
Staff’s politeness;
Staff’s trust inspiration;
Ensuring security;
The efciency of service delivery;
Transmission of all the information;
Immediate response to requests;
Individualized treatment;
Willingness to help;
Paying attention to customers;
Customers need understanding
Rostasova et al.
(2020)
Adequacy of postal fees;
Opening hours during the afternoon;
Opening hours during weekends;
Waiting time at the compartment;
Availability of post ofces and mailboxes;
Post ofce interior;
Parking spots nearby;
Willingness and helpfulness of employees;
Identication of employees by logo;
Informativeness of providing services;
Security of shipment delivery;
Handling of complaints and grievances;
Electronic services offer
50 Mladenka Blagojević, Nikola Knežević, Dejan Marković
The evaluation of customer satisfaction of postal service providers consists of the
application of various methods of measuring quality. Customers expect the postal pro-
vider willing to help them and to provide prompt service. Thus, postal providers are
expected to be very responsive toward their customers and to be prompt in addressing
their requests, queries, and complaints (Roopchund and Boojhawon, 2014). In the paper
Khairunnisa et al. (2018) customer satisfaction and loyalty on customer delivered value
of postal and shipping service were analyzed through causal research by Structural
Equation Modeling. Each method of measuring the quality of services has its specif-
ic features, which evaluate the specic features of the provided postal service. These
aspects concern: time availability, availability of contact and access points, security
during the relocation process, staff expertise, affordability of postal services, waiting
time at post ofces, handling of complaints, and information on postal services (Dia-
belkova, 2013). The objective of this paper is to seek and measure the level of customer
satisfaction and services rendered in the postal sector in the Republic of Serbia through
the chosen methodology that the researchers consider to be effective and suitable for its
application. The chosen methodology is discriminant analysis. Research has been car-
ried out on customer satisfaction with postal services in Serbia up to now. Only a few
researchers, in not very recent literature, have paid attention to the service quality of
postal services in Serbia. Past studies from Serbia relate the quality of postal services,
as perceived by customers, with their satisfaction with specic features of these services
(Pavlović et al., 2021; Šarac et al., 2017; Lečić-Cvetković et al. (2012), Stojanović-Višić
et al. (2012); Marković et al. (2011); Ratković and Pavlović (2017)). According to the
author’s knowledge, by analyzing the Clarivate Analytics Web of Science database,
there is no evidence in the literature about the application of discriminant analysis in
the process of assessing customer satisfaction with postal services.
Research on the quality of postal services in Serbia is conducted every two years
by the Regulatory Authority for Electronic Communications and Postal Services.
Through that survey some aspects of the universal postal service provided by the
public postal operator were analyzed. The area of express services provided by other
operators was also considered. A special group of questions relates to complaints
regarding universal and express/courier services, as well as the impact and conse-
quences of COVID-19 on individuals in terms of the use of postal services, e-com-
merce, and complaints to the regulatory authority.
Discriminant Analysis
Discriminant analysis is an important statistical instrument whose application is
wide and consistent. The theoretical denition of discriminant analysis dates back to
the thirties of the 20th century. It was mentioned for the rst time in the paper of the
Indian scientist Mahalanobis (1936) and the British scientist Fisher (1936).
51
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
The discriminant analysis deals with the problem of separating groups and allo-
cating observations into previously dened groups. The application enables the iden-
tication of the variable that most contributed to the separation of the groups, as well
as the prediction of the probability that the object will belong to one of the groups,
based on the value of a set of independent variables. It is an adequate technique in
cases where the dependent variable ( is categorical (nominal, descriptive), and the
independent variables, , , etc., are numerical. In most cases, the dependent variable
consists of two groups or categories (in this paper, the group of users and the group
of not users). In a rarer number of cases, it consists of several groups (in this paper,
loyal users, occasional users, potential users, and respondents who never use postal
services and their attraction requires radical changes and large costs). Discriminant
analysis is the classication of individuals into groups according to certain criteria.
The most common case of application of this analysis involves the selection of
several variables (usually two), which contribute in the best way to the separation
between predened groups. After that, Fisher’s linear discrimination function is
formed, the discrimination score is determined for each respondent, and the means
of the discrimination scores for each group. Then the mean of the class means, i.e.
pooled mean that contributes the most to the separation is determined. Based on the
discrimination score and the pooled mean (cutting score) the respondents are dis-
criminated into pre-dened groups (Lovrić et al., 2009). In the end, the results are
presented graphically, on a two-dimensional graphic for the case with a smaller num-
ber of respondents or on a one-dimensional graphic for the case of a large number of
respondents for better visibility of the results.
The application of discriminant analysis can have a dual purpose, so a distinction
can be made between descriptive (canonical) discriminant analysis and predictive (lin-
ear) discriminant analysis. Descriptive (canonical) discriminant analysis is used to de-
termine whether there is a difference between two or more groups concerning a set of
quantitative characteristics. Therefore, in descriptive analysis, a mathematical function
is dened, which, under the conditions of certain assumptions and limitations, makes
the greatest possible difference between two or more populations or groups (Tenjović,
2021). This analysis is very similar to the multivariate analysis of variances (MANO-
VA), but in addition to answering the question of whether there are differences between
groups and how big they are, it provides answers to some other questions that will be
presented in the further part of the paper. Predictive or linear discrimination analysis
serves to classify individuals into one of two or more well-dened groups, using math-
ematical rules. Classication discriminant analysis procedures minimize classication
errors and they are based on maximum posterior probabilities (Tenjović, 2021). In this
paper, Fisher Linear Discriminant Analysis (also called Linear Discriminant Analysis)
was used for the case of two or more groups.
There are many diverse areas in which discriminant analysis was applied, and
some of the papers related to the application of discriminant analysis in trafc and
52 Mladenka Blagojević, Nikola Knežević, Dejan Marković
transport are mentioned below. Aksoy et al. (2003) conducted a study aimed at deter-
mining whether there is a signicant difference between passengers using domestic
and international airline companies at Alanya Airport. Lee et al. (2005) analyze the
demand for Thai railways using discriminant analysis. Saffan and Rizki (2018) use
discriminant analysis to nd out why railway passengers use OJEK for the realization
of the rst and last mile of travel. Li et al. (2020) analyzed the attitude of rail service
users toward rail trafc safety. The goal of the paper by Kuljanin et al. (2015) was to
determine the characteristics and behavior of passengers of traditional and low-cost
airlines on competitive lines, i.e. lines on which both types of carriers provide their
services.
In marketing, this analysis is used to determine factors that distinguish types of
customers based on data collected in surveys. Its application is generally carried out
through several sequential phases that include: formulation of the problem, identi-
cation of key attributes (variables), collection of responses from respondents, problem
solving, and interpretation of results (Huberty, 1994).
Discriminant analysis is described in more detail in the papers by Johnson and
Wichern (2007), Manly and Navarro Alberto (2016), Timm (2002), and Walde (2014).
When the observed area is slightly narrowed, i.e. when only the postal sector is ob-
served, there is no paper on this topic. Therefore, the further part of the paper and the
application of discriminant analysis can serve as a pioneer in research of this kind.
Determining the Key Parameters and Survey Research
Information on the satisfaction of users of postal services was collected through a
survey. Respondents rated the various attributes. The questions in the survey were of
a closed type, and the answers are rated numerically on a Likert (Likert, 1932) scale,
where the lowest number represents the worst rating for the observed attribute. This
scale is used to quantify the qualitative characteristics of services and compare them.
For this paper, this scale has been modied so that the answers overlap as little as
possible, so instead of the usual 5 divisions, it now includes 10.
The survey also contained questions that requested the following information from
users: gender, age, business status, education, and region. For this reason, the survey
had a longitudinal design (that is, the survey was conducted at different geographical
areas and during a certain period) and was conducted through different survey chan-
nels (besides tête-à-tête the survey was also conducted in electronic form by lling
out questionnaires). The research aimed to determine how respondents perceive the
quality of postal services (group of users) and what expectations (or opinions) they
have (group of not users), to conclude which attributes are key when choosing the
type of service and how the postal operator could direct its resources and marketing
activities to retain existing and attract new users.
53
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
The research was conducted in the period from September 2022 until October
2022 throughout the Republic of Serbia. 800 respondents participated in the research.
Reliability and Validity of the Questionnaire
Bolarinwa (2015) denes reliability as the degree of repetition of data obtained by a
measuring instrument, while he denes validity as the degree to which a measuring
instrument measures what it is intended to measure. Emphasizes the difculty in
quantifying abstract and intangible concepts (such as postal services, as opposed to a
product). To observe and measure such concepts, it is necessary to choose an adequate
‘’measuring instrument’’. Accordingly, the most common problem when formulating
such questionnaires is determining a reliable and valid measuring instrument.
Reliability is a necessary condition for validity, but it is not sufcient. A ques-
tionnaire can be reliable without being valid and vice versa. In any case, the more
reliable the questionnaire, the higher the chances that it is valid. The validity of the
questionnaire can be ensured in two ways, theoretically and empirically. For this pa-
per, the rst method was used, the sublimation of papers compiled by experts in the
eld of research (so-called face validity), as well as using a Likert scale that contains
questions in the form of statements and is easy to interpret.
Cronbach’s Coefcient Α (Reliability of the Questionnaire)
Cronbach’s alpha or coefcient α, developed by (Cronbach, 1951), measures reliabil-
ity or internal consistency. Cronbach’s tests are used to determine whether surveys
with multiple-question Likert scales are reliable.
This coefcient α is calculated by the following formula:
where:
N is the number of questions,
c is average covariance among groups (questions),
is average variance within the group.
Cronbach’s coefcient α gives the lower limit of the reliability of the question-
naire. Also, its potentially low value may suggest that the questionnaire has a small
number of questions (evaluation criteria). In the case of large values of this coef-
cient, there is a possibility of a high correlation between the questions, that is, it sug-
gests that some questions may be redundant. The following Table 2 shows the values
of this coefcient and how they can be interpreted.
narrowed, i.e. when only the postal sector is observed, there is no paper on this topic. Therefore, the
further part of the paper and the application of discriminant analysis can serve as a pioneer in research of
this kind.
Determining the Key Parameters and Survey Research
Information on the satisfaction of users of postal services was collected through a survey. Respondents
rated the various attributes. The questions in the survey were of a closed type, and the answers are rated
numerically on a Likert (Likert, 1932) scale, where the lowest number represents the worst rating for the
observed attribute. This scale is used to quantify the qualitative characteristics of services and compare
them. For this paper, this scale has been modified so that the answers overlap as little as possible, so
instead of the usual 5 divisions, it now includes 10.
The survey also contained questions that requested the following information from users: gender,
age, business status, education, and region. For this reason, the survey had a longitudinal design (that is,
the survey was conducted at different geographical areas and during a certain period) and was conducted
through different survey channels (besides tête-à-tête the survey was also conducted in electronic form by
filling out questionnaires). The research aimed to determine how respondents perceive the quality of
postal services (group of users) and what expectations (or opinions) they have (group of not users), to
conclude which attributes are key when choosing the type of service and how the postal operator could
direct its resources and marketing activities to retain existing and attract new users.
The research was conducted in the period from September 2022 until October 2022 throughout
the Republic of Serbia. 800 respondents participated in the research.
Reliability and Validity of the Questionnaire
Bolarinwa (2015) defines reliability as the degree of repetition of data obtained by a measuring
instrument, while he defines validity as the degree to which a measuring instrument measures what it is
intended to measure. Emphasizes the difficulty in quantifying abstract and intangible concepts (such as
postal services, as opposed to a product). To observe and measure such concepts, it is necessary to choose
an adequate ‘’measuring instrument’’. Accordingly, the most common problem when formulating such
questionnaires is determining a reliable and valid measuring instrument.
Reliability is a necessary condition for validity, but it is not sufficient. A questionnaire can be
reliable without being valid and vice versa. In any case, the more reliable the questionnaire, the higher the
chances that it is valid. The validity of the questionnaire can be ensured in two ways, theoretically and
empirically. For this paper, the first method was used, the sublimation of papers compiled by experts in
the field of research (so-called face validity), as well as using a Likert scale that contains questions in the
form of statements and is easy to interpret.
Cronbach’s Coefficient Α (Reliability of the Questionnaire)
Cronbach’s alpha or coefficient α, developed by (Cronbach, 1951), measures reliability or internal
consistency. Cronbach’s tests are used to determine whether surveys with multiple-question Likert scales
are reliable.
This coefficient α is calculated by the following formula:
=
∗
+−1∗
where:
N is the number of questions,
is average covariance among groups (questions),
is average variance within the group.
54 Mladenka Blagojević, Nikola Knežević, Dejan Marković
Table 2: Values of Cronbach’s coefcient α
Cronbach’s α Rating
α ≥ 0,9 Perfect
0,9 > α ≥ 0,8 Good
0,8 > α ≥ 0,7 Acceptable
0,7 > α ≥ 0,6 Questionable
0,6 > α ≥ 0,5 Bad
0,5 > α Unacceptable
The Application of Discriminant Analysis
Based on the received answers, a discriminant analysis was carried out for discrim-
ination of users (separation between groups) and classication (or allocation) of po-
tential users.
First, it is necessary to check the validity and reliability of the questionnaire be-
cause if one of these two conditions is not met, the results and conclusions obtained
from the analysis could be rejected a priori.
Validity is provided theoretically, through sublimation of the attributes that ex-
perts in the eld of postal trafc cite as key in assessing quality. By systematizing the
mentioned attributes in Table 1, the key criteria were selected:
• The effective procedure of collection and delivery
• Security of transport
• Number of counters
• An acceptable price
• Speed of service provision
• Attitude towards users
• The possibility of items tracking (Track&Trace)
• Waiting time for service
• Availability and accessibility.
The reliability of the questionnaire is ensured by the method of checking the inter-
nal consistency by calculating the Cronbach coefcient α, as described in subsection
Cronbach’s Coefcient α (Reliability of the Questionnaire). The calculated value is
0.75 and according to Table 1 this result as acceptable. After ensuring the reliability
and validity of the questionnaire, the formulation of the problem is approached.
The problem is presented in several stages (Figure 1). The rst represents the clas-
sication of respondents into a group of users and respondents who do not use postal
services (group of not users). To solve this model, Fisher’s linear discriminant analysis
was used to separate the respondents into those two groups. Respondents who stated
that they used postal services at least once in the last few months were classied as
postal service users. All remaining respondents were classied into the second group.
55
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
The second stage includes the classication of respondents into 4 groups. That is,
the category of users is divided into the subcategory of loyal users (the frequency of
using postal services is at least once a week) and the subcategory of occasional users.
The group of respondents who do not use postal services is further divided into the
subcategories of respondents who used postal services at least once in the last year
(potential users) and the group of respondents who have not used them in the last year
(or never).
Figure 1: Research approach
With the model dened in this way, for the case with four groups, the results in-
dicate that the success rate in three of the four groups is higher compared to the case
when the classication was done randomly. Also, the overall success rate of correct
discrimination is higher than the random classication rate in the case where the a
priori classication probabilities are equal (0.25 for the case with four groups), but
such results are characterized as unsatisfactory according to Kovačić (1994), i.e. they
provide only slight improvements. For this reason, that model calculation is not pre-
sented in this paper.
After discriminating the respondents into groups, how potential users would be
allocated is presented.
Each of the mentioned stages also contains several steps, as part of which the
following is calculated:
• Which variable (attribute) has the greatest importance when allocating respon-
dents to one of the groups;
• Percentage of error when allocating respondents into groups;
• Degree of the signicance of each of the discrimination functions;
• To which of the above groups do the new 100 respondents belong?
Solving the Problem Using Discriminant Analysis for the Case with Two Groups
Due to the robustness of the data containing the responses of all respondents, the av-
erage scores for each attribute will be presented here. The analysis was conducted us-
ing the discriminant procedure in Statistical Package for the Social Sciences (SPSS).
Table 1 this result as acceptable. After ensuring the reliability and validity of the questionnaire,
the formulation of the problem is approached.
The problem is presented in several stages (Figure 1). The first represents the
classification of respondents into a group of users and respondents who do not use postal
services (group of not users). To solve this model, Fisher's linear discriminant analysis was used
to separate the respondents into those two groups. Respondents who stated that they used postal
services at least once in the last few months were classified as postal service users. All remaining
respondents were classified into the second group.
The second stage includes the classification of respondents into 4 groups. That is, the
category of users is divided into the subcategory of loyal users (the frequency of using postal
services is at least once a week) and the subcategory of occasional users. The group of
respondents who do not use postal services is further divided into the subcategories of
respondents who used postal services at least once in the last year (potential users) and the group
of respondents who have not used them in the last year (or never).
With the model defined in this way, for the case with four groups, the results indicate that the
success rate in three of the four groups is higher compared to the case when the classification
was done randomly. Also, the overall success rate of correct discrimination is higher than the
random classification rate in the case where the a priori classification probabilities are equal
(0.25 for the case with four groups), but such results are characterized as unsatisfactory
according to Kovačić (1994), i.e. they provide only slight improvements. For this reason, that
model calculation is not presented in this paper.
After discriminating the respondents into groups, how potential users would be allocated
is presented.
Each of the mentioned stages also contains several steps, as part of which the following is
calculated:
• Which variable (attribute) has the greatest importance when allocating respondents to one of
the groups;
• Percentage of error when allocating respondents into groups;
• Degree of the significance of each of the discrimination functions;
• To which of the above groups do the new 100 respondents belong?
Solving the Problem Using Discriminant Analysis for the Case with Two Groups
Due to the robustness of the data containing the responses of all respondents, the average scores
for each attribute will be presented here. The analysis was conducted using the discriminant
procedure in Statistical Package for the Social Sciences (SPSS).
56 Mladenka Blagojević, Nikola Knežević, Dejan Marković
It can be seen from the respondents’ answers that the ‘’waiting time for service’’ is
the worst-rated attribute among users of postal services (Table 3). Among respondents
who are not users of postal services, the worst rated attribute is ‘’attitude towards us-
ers’’. ‘’Availability and accessibility’’ and ‘’acceptable price’’ are the best-evaluated
attribute (criteria) by both groups of respondents.
As discriminating criteria (attributes which create the biggest separation) in this
case gure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
Table 3: Mean values of the attributes by groups (case with two groups: users and not
us ers)
Group
The
effective
procedure
of
collection
and
delivery
Security of
transport
Number of
counters
An
acceptable
price
Speed of
service
provision
Attitude
towards
users
The
possibility
of items
tracking
(Track&
Trace)
Waiting
time for
service
Availability
and
accessibility
Users 5.84 5.94 6.90 7.88 6.72 6.26 7.38 5.62 8.16
Not users 5.22 5.26 5.84 7.2 5.04 4.40 6.34 4.94 7.84
Difference 0.62 0.68 1.06 0.68 1.68 1.86 1.04 0.68 0.32
After the attributes that have the greatest importance in the discrimination of re-
spondents into groups have been determined, the next step is to calculate the sample
indicators. The most important indicators will be presented here.
The mean values from the group samples are:
Sample covariance matrices by groups:
General covariance matrix:
Fisher’s linear discriminant function is as follows:
!=6.72
6.26 ;
!= 5.04 ;
!=4.33 1.21
1.21 4.18 ;
!=6.30 1.76
1.76 3.81
;
=5.42 50.97
50.97 4.08
Fisher’s linear discriminant function is as follows:
y=0.34!+0.29!;
Means of discrimination scores by groups are:
!=0.41;
!=0.30;
The mean of the class means (pooled mean) is:
!=0.35;
After the mentioned procedure, the discrimination of users according to groups is carried out
(Table 4). All attributes of the services figure in the calculation, and in this part of the analysis, only the
attributes that create the greatest separation are shown. As discriminating criteria (attributes which create
the biggest separation) in this case figure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
If the discrimination score is higher than the pooled mean, then observation takes ‘’good’’, otherwise
‘’bad’’. Due to the robustness of the data, all data and results cannot be presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Users
6
8
0.35
good
Not users
6
4
0.35
good
9
10
good
3
3
good
10
4
good
7
10
bad
7
8
good
7
3
good
6
3
bad
10
7
bad
4
3
bad
7
8
bad
4
5
bad
6
3
good
7
7
good
4
6
good
5
7
good
3
1
good
7
6
good
7
2
good
8
10
good
4
2
good
8
6
good
6
6
bad
!=6.72
6.26 ;
!= 5.04 ;
!=4.33 1.21
1.21 4.18 ;
!=6.30 1.76
1.76 3.81 ;
=5.42 50.97
50.97 4.08
Fisher’s linear discriminant function is as follows:
y=0.34!+0.29!;
Means of discrimination scores by groups are:
!=0.41;
!=0.30;
The mean of the class means (pooled mean) is:
!=0.35;
After the mentioned procedure, the discrimination of users according to groups is carried out
(Table 4). All attributes of the services figure in the calculation, and in this part of the analysis, only the
attributes that create the greatest separation are shown. As discriminating criteria (attributes which create
the biggest separation) in this case figure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
If the discrimination score is higher than the pooled mean, then observation takes ‘’good’’, otherwise
‘’bad’’. Due to the robustness of the data, all data and results cannot be presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Users
6
8
0.35
good
Not users
6
4
0.35
good
9
10
good
3
3
good
10
4
good
7
10
bad
7
8
good
7
3
good
6
3
bad
10
7
bad
4
3
bad
7
8
bad
4
5
bad
6
3
good
7
7
good
4
6
good
5
7
good
3
1
good
7
6
good
7
2
good
8
10
good
4
2
good
8
6
good
6
6
bad
!=6.72
6.26 ;
!= 5.04 ;
!=4.33 1.21
1.21 4.18 ;
!=6.30 1.76
1.76 3.81 ;
=5.42 50.97
50.97 4.08
Fisher’s linear discriminant function is as follows:
y=0.34!+0.29!;
Means of discrimination scores by groups are:
!=0.41;
!=0.30;
The mean of the class means (pooled mean) is:
!=0.35;
After the mentioned procedure, the discrimination of users according to groups is carried out
(Table 4). All attributes of the services figure in the calculation, and in this part of the analysis, only the
attributes that create the greatest separation are shown. As discriminating criteria (attributes which create
the biggest separation) in this case figure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
If the discrimination score is higher than the pooled mean, then observation takes ‘’good’’, otherwise
‘’bad’’. Due to the robustness of the data, all data and results cannot be presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Users
6
8
0.35
good
Not users
6
4
0.35
good
9
10
good
3
3
good
10
4
good
7
10
bad
7
8
good
7
3
good
6
3
bad
10
7
bad
4
3
bad
7
8
bad
4
5
bad
6
3
good
7
7
good
4
6
good
5
7
good
3
1
good
7
6
good
7
2
good
8
10
good
4
2
good
8
6
good
6
6
bad
!=6.72
6.26 ;
!=
5.04
4.40
;
!=4.33 1.21
1.21 4.18 ;
!=6.30 1.76
1.76 3.81 ;
=5.42 50.97
50.97 4.08
Fisher’s linear discriminant function is as follows:
y=0.34!+0.29!;
Means of discrimination scores by groups are:
!=0.41;
!=0.30;
The mean of the class means (pooled mean) is:
!=0.35;
After the mentioned procedure, the discrimination of users according to groups is carried out
(Table 4). All attributes of the services figure in the calculation, and in this part of the analysis, only the
attributes that create the greatest separation are shown. As discriminating criteria (attributes which create
the biggest separation) in this case figure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
If the discrimination score is higher than the pooled mean, then observation takes ‘’good’’, otherwise
‘’bad’’. Due to the robustness of the data, all data and results cannot be presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Users
6
8
0.35
good
Not users
6
4
0.35
good
9
10
good
3
3
good
10
4
good
7
10
bad
7
8
good
7
3
good
6
3
bad
10
7
bad
4
3
bad
7
8
bad
4
5
bad
6
3
good
7
7
good
4
6
good
5
7
good
3
1
good
7
6
good
7
2
good
8
10
good
4
2
good
8
6
good
6
6
bad
57
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
Means of discrimination scores by groups are:
The mean of the class means (pooled mean) is:
After the mentioned procedure, the discrimination of users according to groups
is carried out (Table 4). All attributes of the services gure in the calculation, and
in this part of the analysis, only the attributes that create the greatest separation are
shown. As discriminating criteria (attributes which create the biggest separation) in
this case gure the ‘’speed of service provision’’ and ‘’attitude towards users’’. If the
discrimination score is higher than the pooled mean, then observation takes ‘’good’’,
otherwise ‘’bad’’. Due to the robustness of the data, all data and results cannot be
presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Group
Speed of
service
provision
Attitude
towards
users
Pooled
mean
Classi-
cation Group
Speed of
service
provision
Attitude
towards
users
Pooled
mean
Classi-
cation
Users
6 8
0.35
good
Not
users
6 4
0.35
good
9 10 good 3 3 good
10 4 good 7 10 bad
7 8 good 7 3 good
6 3 bad 10 7 bad
4 3 bad 7 8 bad
4 5 bad 6 3 good
7 7 good 4 6 good
5 7 good 3 1 good
7 6 good 7 2 good
8 10 good 4 2 good
8 6 good 6 6 bad
2 4 bad 4 4 good
9 6 good 7 2 good
4 6 bad 5 6 good
8 2 bad 6 7 bad
10 10 good 4 5 good
9 7 good 7 3 good
5 6 bad 5 6 good
7 9 good 4 5 good
3 5 bad 2 3 good
10 8 good 6 4 good
!=6.72
6.26 ;
!= 5.04 ;
!=4.33 1.21
1.21 4.18 ;
!=6.30 1.76
1.76 3.81 ;
=5.42 50.97
50.97 4.08
Fisher’s linear discriminant function is as follows:
y=0.34!+0.29!;
Means of discrimination scores by groups are:
!=0.41;
!=0.30;
The mean of the class means (pooled mean) is:
!=0.35;
After the mentioned procedure, the discrimination of users according to groups is carried out
(Table 4). All attributes of the services figure in the calculation, and in this part of the analysis, only the
attributes that create the greatest separation are shown. As discriminating criteria (attributes which create
the biggest separation) in this case figure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
If the discrimination score is higher than the pooled mean, then observation takes ‘’good’’, otherwise
‘’bad’’. Due to the robustness of the data, all data and results cannot be presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Users
6
8
0.35
good
Not users
6
4
0.35
good
9
10
good
3
3
good
10
4
good
7
10
bad
7
8
good
7
3
good
6
3
bad
10
7
bad
4
3
bad
7
8
bad
4
5
bad
6
3
good
7
7
good
4
6
good
5
7
good
3
1
good
7
6
good
7
2
good
8
10
good
4
2
good
8
6
good
6
6
bad
!=6.72
6.26 ;
!= 5.04 ;
!=4.33 1.21
1.21 4.18 ;
!=6.30 1.76
1.76 3.81 ;
=5.42 50.97
50.97 4.08
Fisher’s linear discriminant function is as follows:
y=0.34!+0.29!;
Means of discrimination scores by groups are:
!=0.41;
!=0.30
;
The mean of the class means (pooled mean) is:
!=0.35;
After the mentioned procedure, the discrimination of users according to groups is carried out
(Table 4). All attributes of the services figure in the calculation, and in this part of the analysis, only the
attributes that create the greatest separation are shown. As discriminating criteria (attributes which create
the biggest separation) in this case figure the ‘’speed of service provision’’ and ‘’attitude towards users’’.
If the discrimination score is higher than the pooled mean, then observation takes ‘’good’’, otherwise
‘’bad’’. Due to the robustness of the data, all data and results cannot be presented in the paper.
Table 4: Mean values of the attributes by groups (the part of results)
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Grou
p
Speed
of
service
provisio
n
Attitud
e
toward
s users
Poole
d
mean
Classificati
on
Users
6
8
0.35
good
Not users
6
4
0.35
good
9
10
good
3
3
good
10
4
good
7
10
bad
7
8
good
7
3
good
6
3
bad
10
7
bad
4
3
bad
7
8
bad
4
5
bad
6
3
good
7
7
good
4
6
good
5
7
good
3
1
good
7
6
good
7
2
good
8
10
good
4
2
good
8
6
good
6
6
bad
58 Mladenka Blagojević, Nikola Knežević, Dejan Marković
Group
Speed of
service
provision
Attitude
towards
users
Pooled
mean
Classi-
cation Group
Speed of
service
provision
Attitude
towards
users
Pooled
mean
Classi-
cation
Users
8 9
0.35
good
Not
users
8 4
0.35
bad
8 5 good 6 4 good
8 7 good 4 6 good
9 5 good 7 2 good
6 7 good 6 5 good
7 6 good 6 4 good
8 8 good 5 2 good
7 7 good 5 5 good
9 9 good 4 4 good
1 5 bad 5 6 good
9 8 good 8 3 bad
7 10 good 7 3 good
8 5 good 6 8 bad
7 5 good 5 6 good
5 2 bad 6 5 good
6 4 bad 2 6 good
9 5 good 5 3 good
3 7 bad 4 3 good
8 6 good 1 3 good
6 6 good 3 4 good
6 5 bad 2 4 good
4 8 good 4 4 good
6 7 good 1 2 good
6 8 good 5 4 good
7 6 good 5 7 bad
8 6 good 2 5 good
6 5 good 5 5 good
6 2 bad 5 3 good
... ... ... ... ... ...
Classication of Respondents into the Group of Users or into the Group of Respond-
ents Who Do Not Use Postal Services
To determine the classication success rate, a confusion matrix is dened that
shows the number of correctly and incorrectly classied observations by a group.
In this case, a confusion matrix is formed for the case with two groups (Table 5).
The elements on the main diagonal represent the number of observations that were
correctly allocated, while the off-diagonal elements represent the number of obser-
vations that were incorrectly allocated. In the literature, this term is also known as
the hit ratio. The hit ratio gives the correctly classied observations divided by the
number of observations.
59
Assessment of Customer Satisfaction with Posta l Services – a Statistical Approach
Table 5: Confusion matrix
True class membership
Users Not users
Predicted class
membership
Users 250 100
Not users 150 300
Sample size 400 400
Assessing group membership prediction accuracy – hit ratio (HR):
That is, the overall allocation success rate is:
As in the discrimination of respondents in two groups with equal observations,
the observations can be randomly allocated to one of the groups with equal probabili-
ty, this means that in the case of classication, randomly, the error rate would be 50%.
As the error rate of 69% (overall allocation success rate) was obtained using Fish-
er’s classication procedure, it can be concluded that the used observations classica-
tion procedure is a good tool for their reliable allocation by groups. That is, accord-
ing to the answers received in the survey, it is possible to make a clear distinction
between respondents who declared themselves as users of postal services and those
who were not.
The following graph shows the allocation between these two groups (Figure 2). To
make it easier to see the allocated respondents, their discrimination scores are shown
on a one-dimensional graphic. Users are represented in blue, respondents who are not
users are in red, while the green triangle represents the mean of the class means, i.e.
pooled mean μy. Well-allocated respondents in the group of users are all those whose
discrimination score is above the pooled mean, while in the group of not users, the
case is reversed (i.e., all those whose discrimination score is below the pooled mean
are well-allocated).
To determine the classification success rate, a confusion matrix is defined that shows the number of
correctly and incorrectly classified observations by a group. In this case, a confusion matrix is formed for
the case with two groups (Table 5). The elements on the main diagonal represent the number of
observations that were correctly allocated, while the off-diagonal elements represent the number of
observations that were incorrectly allocated. In the literature, this term is also known as the hit ratio. The
hit ratio gives the correctly classified observations divided by the number of observations.
Table 5: Confusion matrix
True class membership
Users
Not users
Predicted
class
membership
Users
250
100
Not
users
150
300
Sample size
400
400
Assessing group membership prediction accuracy –