Online shopping vs in store shopping an analysis of choice behavior

We examine factors related to the time and monetary costs for shopping that affect the choice between online and in-store channel using a random utility framework.. A random utility mode

UNIVERSITY OF ECONOMICS ERASMUS UNVERSITY ROTTERDAM

HO CHI MINH CITY INSTITUTE OF SOCIAL STUDIES

VIETNAM – THE NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

ONLINE SHOPPING VS IN-STORE SHOPPING:

AN ANALYSIS OF CHOICE BEHAVIOR

BY

PHAM NHU MAN

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, DECEMBER 2017

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

ONLINE SHOPPING VS IN-STORE SHOPPING:

AN ANALYIS OF CHOICE BEHAVIOR

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By

PHAM NHU MAN

Academic Supervisor:

Dr TRUONG DANG THUY

HO CHI MINH CITY, DECEMBER 2017

ACKNOWLEDGEMENT

I would like to express my deep gratitude to my supervisor Dr Truong Dang Thuy, for his guidance, support and encouragement, whereby I could complete the program

I would like to thank Dr Pham Khanh Nam for his valuable advices for my thesis

I would like to thank Mr Do Huu Luat for his support and guidance during the process of data collection

I also would to express my thanks to VNP officers who support me during my thesis process

I would like to thank my family for their encouragement and support during my master program

ABSTRACT

This study examines the consumer’s choice between online and in-store shopping for

a specific product of books For this purpose, a survey of 352 book buyers was implemented

to collect revealed and stated preference data Conditional logit and mixed logit models are used to analyze the choice We found that 40 minute travelling to store and come back is worth VND 22,303 to VND 33,849, and that willingness to pay for one day earlier delivery is from VND 15,617 to VND 19,028 We also found that shoppers are more sensitive to online price than to price in store Those who are office workers and those who spend more time to access the internet are more likely to shop online Moreover, in-store shoppers with higher income are less sensitive to price than those who have lower income Especially, shoppers with high income are more concerned about travel time than those with low income The female shoppers are more patient waiting for the delivery In addition, respondents evaluated that the convenience of payment method, the quality of books, ability to read books before purchasing and large variety offered products, are important and effect partially on their choice

TABLE OF CONTENTS

LIST OF TABLES v

LIST OF FIGURES vi

CHAPTER 1: INTRODUCTION 1

1.1 Overview of Vietnam’s retail sector 1

1.2 Trend in shopping behavior in Viet Nam 2

1.3 Research questions and objectives 4

1.2 The scope of study 5

1.3 Organization of the thesis 6

CHAPTER 2: LITERATURE REVIEW 7

2.1 Binary Choice models 7

2.2 Frequency of online shopping 11

2.3 Random Utility Models (RUM) 13

2.4 Structural Equation Models (SEM) 15

CHAPTER 3: RESEARCH METHODOLOGY 17

3.1 The choice of online shopping versus traditional store shopping 17

3.2 Revealed preference data: methods of collection 26

3.3 The stated preference data: methods of collection including choice experimental design 29 3.3.1 The stated preference method 29

3.3.2 Choice of experimental design 31

3.4 Sample size and sampling 35

3.5 Survey 36

3.6 The questionnaire 37

3.7 Estimation methods 38

CHAPTER 4: RESEARCH RESULTS 41

4.1 Summary statistics 41

4.2 Estimation results 47

4.2.1 The Basic model 47

4.2.2 The Full model 52

4.2.3 Valuation indicators of time and cost attributes 56

4.2.4 Store attributes and the choice probability 61

CHAPTER 5: CONCLUSIONS AND POLICY IMPLICATIONS 69

REFERENCE 72 APPENDIX 76 THE QUESTIONNAIRE 76

LIST OF TABLES

Table 3 1: The store choice attributes description 22

Table 3 2: Individual characteristics variable description 24

Table 3.3: Attributes and levels in stated choice experimental design 30

Table 3.4: The attribute’s levels of online and physical store 32

Table 3.5: Designed choice sets 34

Table 3.6: The presentation of a sample choice set 1 in block 1 35

Table 4.1: Survey location and number of respondents 42

Table 4.2: Descriptive statistics of the sample 43

Table 4.3: Descriptive statistics analysis of consumer attitudes 45

Table 4.4: Revealed data between online shoppers and in-store shoppers 46

Table 4.5: Summary statistics of alternative-specific attributes 48

Table 4.6: Estimation results of the basic model 50

Table 4.7: Estimation results of the full model 53

Table 4.8: Willingness to pay of time and cost attributes 57

LIST OF FIGURES

Figure 1.1: Vietnam’s retail sales and online retail sales from 2013 to 2015 2

Figure 1.2: Percentage online shoppers of the online accessing in Vietnam 3

Figure 3.1: Choice model between online and in-store shopping 26

Figure 4.1: Frequency of Internet access and Internet access devices 44

Figure 4 2: The percentage change in online shopping probability given the change in purchase price, shipping fee and delivery time 62

Figure 4 3: The percentage change in online shopping probability given the change in purchase price, shipping fee and delivery time 64

Figure 4 4: The percentage change in online shopping probability given the change in purchase price, shipping fee and delivery time 67

CHAPTER 1:

INTRODUCTION

1.1 Overview of Vietnam’s retail sector

Vietnam has recently emerged as an attractive retail market thanks to the young population, with 70 percent in the age between 15 and 64 It is foreseen that the market size in Vietnam increases in terms of population as well as purchasing power The urban population, which is the most potential for the retail market, is expected to increase most rapidly at 2.6 percent annually from 2015 to 2020, the highest in the region (Worldbank, 2015 and United Nation, 2014) The urban population is further increased by the process of urbanization and rural-urban migration In addition, Vietnam appears to be the most energetic emerging economy with a rising living standards and intensifying disposable incomes The Boston Consulting Group (2013) has remarked that Vietnam has the highest growth speed of the wealthy and middle classes in the region, which is predicted to increase by 21 million compared to 12 million from 2012 to 2020 These consumers, whose monthly income is higher than VND 15 million (USD 714), will be a key group of potential customers for retailers

In addition, the improving infrastructure is also a factor that makes Vietnam an alluring market for retailers Many international retail groups as Metro Cash & Carry, BigC, Emart, Aeon, Takashimaya, AuchanSuper and many others have entered the market These international retailers, together with domestic players such as Co-op Mart, Vincom Mall have created intense competition in the retail sector of Vietnam CBRE (2014) remarked that among the cities in the Asian for retail market enlargement, Hanoi and Ho Chi Minh City were rated in the top 10 municipalities Hanoi was ranked third after Shanghai and Beijing as the metropolis with the vivid retail market in the area In recent years, Vietnam’s retail industry has manifested effective growing rates, with retail revenues rising by 60 percent in the period of 2009 – 2013, and is estimated to approach US$109 billion in 2017 Therefore, the retail sector has an enormous potential for new investors

The two main retail channels are traditional retail and electronic tailing (e-tailing) Traditional retail in the form that exchanges of goods or services directly to customers in the general stores, convenient stores, supermarkets, hypermarkets (a superstore incorporate of

a department store into a supermarket) or shopping malls Thanks to the advancement of information technology and internet network facilities, e-tailing has been established E-tailing requires the customer’s access the Internet and visiting the stores’ websites in order

to purchase goods or services In online shopping, the purchasers can see the images of the products, explore information about the product specifications, the manufacturer, ingredients, texture, features, prices, and finally can choose the suitable payment methods All of these are completed by those mouse clicks By that way, the shoppers save their time, efforts and vehicles to transport products Figure 1.1 shows the online retail sales in comparison with total retail sales from 2013 to 2015 in Vietnam The sales of e-tailing, although being low, is increasing rapidly by more than 35 percent annually

Figure 1.1: Vietnam’s retail sales and online retail sales from 2013 to 2015

Source: General Statistics Office of Vietnam (2013, 2014, 2015)

1.2 Trend in shopping behavior in Viet Nam

The rapid growth of Information and Communication Technologies (ICT) resulted in the

remarkable change in consumer’s behavior and their preferences Consumers tend to prefer

products or services that best serve the convenience of their busy lives Online shopping meets some requirements of convenience Nielsen (2014) stated that Vietnam is ranked number three in ASEAN region after Singapore and Philippines regarding the uses of mobiles for online shopping, accounted for 58 percentage of people using mobile phones to access

the internet Furthermore, social networks such as Facebook, Twitter, Zalo, Youtube, and online shopping applications like Lazada, Adayroi, Shoppee, Tiki and many others are creating great community connections in the present society A huge percentage of Vietnam online shopping is conducted informally via these social networks and applications

The increasing busy life, the convenience of online shopping, and the development of ICT has made online shopping growing rapidly VECITA (2014) revealed that Vietnam online shopping sales revenue was valued US$2.97 billion with online purchase per capita was estimated at USD145 per year, accounting for 2,12% of total retail sales The most purchased products were fashion items and cosmetics (28% of total online sales), furniture and electronics (25%), computer and phones (16%), food and beverage (16%) and industry and construction (15%) VECITA (2015) indicates that the average per capita online purchase spending was estimated at 160 USD annually, revenue reached 4.07 billion USD, up 37% comparing to 2014 and accounting for around 2.8% of total retail sales The best purchases were clothing, footwear, cosmetic (32% of total online sales), followed by technology and electronic equipments; household appliances; books, stationery, flower, and gifts Figure 1.2 presents the percentage of online shoppers account for total number of Internet access in Vietnam

Figure 1.2: Percentage online shoppers of the online accessing in Vietnam

Source: VECITA (2013, 2014, 2015)

The online retail sales in Vietnam, ranging from 2 to 5% of total sales, is comparable to other countries in the region However that ratio is still lower than country such as China (13.5%), and Korea (11.2%) (iResearchina, 2015 and eMarketer, 2015) This implies there

may be great potential for the e-tailing sector in Vietnam And studying the behavior of shoppers to learn what drives them to online shopping may help the sector to grow

1.3 Research questions and objectives

The main goal of this study is to analyze the consumer choice between online shopping and in-store shopping Particularly, the study aims at identyfying attributes that affect the choice

of consumers In other words, what can push people are more likely to shop in-store? What can drive consumers towards online shopping?

We examine factors related to the time and monetary costs for shopping that affect the choice between online and in-store channel using a random utility framework Factors such as price, ordering time, travel time and costs, delivery time and other attributes of shopping are considered These factors determine utility and under the random utility theory, we assume that consumers would choose online or physical store in order to maximize utility A random utility model with time and money attributes is employed to examine the store choice for shopping books

Other factors related to the consumer such as socio-demographic and attitudes are also considered as factors affecting the choice These issues directly influence the choice of choosing either online shopping or in-store shopping. Hence, the aim of this study is to address the four main objectives:

(1) Evaluating the impacts of shopping channel’s attributes on the choice between online shopping versus in-store shopping

(2) Determining the effects of individual characteristics on the consumer’s choice

(3) Examining the trade-off between the alternative-specific attributes in the competition between online shopping and in-store shopping, to predict the change in the choice probability

(4) Measuring the willingless to pay between time and cost attributes of alternatives Therefore, this study raises the following questions:

(1) Which attributes affect on individual’s choice between online shopping and in-store shopping?

(2) How do the social-demographic affect the shopping mode choice?

(3) How does the probability of choosing shopping mode changes when changing the alternative-specific attributes?

(4) How much the consumers are willing to tradeoff between the time and cost attributes?

Applying the random utility models, this thesis analyzes observe how the consumer makes the decision for shopping between online channel and traditional store channel In so doing, we understand the driving forces that impact the consumer’s choice The results will support for the sellers in both shopping channels The sellers can conceive the key determinants impact on the choice behavior of consumers, and it is crucial not only build on the effective retailing strategies but also predict and investigate the valuation indicators This is to say that thesis could help the online sellers realize the benefits and obstacles, such as the issues in terms of increasing or decreasing of price of product, or the values about the travel time, travel cost or delivery time and cost Since, the online sellers can develop their sales and marketing strategy, as well as, indentify and remove the main obstacles to attract more shoppers Besides, the store sellers can appreciate which factors are significant competitive advantages for their channel that were drawn from the consumer’s choice

1.2 The scope of study

This thesis is a practical study about the consumer choice between online shopping and store shopping, through the application of random utility model This study conducted the survey under the paper-and-pencil form and face to face interview in August, 2017 The data

in-is collected by the combination of two methods of Revealed Preference data and Stated Preference approach The location of survey is Ho Chi Minh City, detail as Book Street Nguyen Van Binh, Walking Street Nguyen Hue, University of Economics Ho Chi Minh City, Ho Chi Minh City University of Pedagogy, University of Social Sciences and Humanities, Sai Gon University, Book Café, Coffee Store and some parks around Ho Chi Minh City Due to the limitations of finance and time, this study conducted a random survey of 352 respondents who are the decision maker of buying book in the past 12 months

The reason why this research attempts to focus on a single class of product, namely books, because of the three primary reasons:

(1) The target is to diminish the effects of product characteristics like sensory or sensory goods; or experience goods (clothing), so that the impacts of attributes of shopping channel can appear Besides that, if the consumer purchases any specific book

non-at a bookstore seems to be no difference from purchasing non-at an online store (Hsiao, 2009) Therefore, the effects of product characteristics can be eliminated from the model

(2) Vietnam E-Commerce Report (2016) showed that books are a kind of product that is usually purchased through online channel, ranked number fifth, after fashion, IT/mobile, kitchen/home app and food/drinks This statistical data indicated that most of the respondents who take part in the experiment will not be too strange with the online bookstores

(3) In addition, this study also conducted an experiment in terms of the portfolio of seller products that the consumers often purchase over the internet The result has shown that book is the third position among the products that consumers often buy via the online channel

best-1.3 Organization of the thesis

This thesis is structured into five chapters Chaper 1 provides an overview of Vietnam’s retail sector, the trend in shopping behavior in Vietnam, the research questions and objectives, the scope of study Chapter 2 discusses the existing literatures which are classified into four analysis approaches, including binary choice model, frequency of online shopping, random utility model and structural equation model Chapter 3 presents the research methodology which contains the choice model, the method of data collection by the revealed preference and the stated preference including choice experimental design, sample size and sampling and estimation method Chapter 4 presents the statistical description and the empirical results Chapter 5 gives the conclusion and policy implications as well as the limitations and further research direction

Shopping mode choice between online shopping and store shopping has been analyzed by some researchers Most of the existing literatures in analyzing shopping mode choice apply one of the four main approaches The first approach is the binary choice model, which is usually applied to analyze the choice between online versus in-store shopping The second method is count data models, which are used to analyze the frequency of online and in-store shopping This describes a number of shopping trips or frequencies of the interesting events in the certain period of the past The third analytical approach that has been applied that is the Random Utility Models (RUM) which assumes that decision makers look at the attributes of the shopping modes and choose the one that maximize utility The last one is frequently applied in recent literatures, namely Structural Equation Model (SEM), for identifying factors affecting the choice of online shopping The following sub-sections present a review the four approaches

2.1 Binary Choice models

Binary choice models are used to analyze the choice between two alternatives (Train, 2002)

In the context of analyzing online shopping decision, the two alternatives are usually online and in-store shopping The outcome variable of interest, or the dependent variable 𝑦, is binary The dependent variable 𝑦 gets the value equal to “one” when the decision makers have taken the action or the event of interest has transpired, otherwise, 𝑦 obtains the value equal to “zero” for non taking an action or event of interest has failed In these models, the

explanatory variables are attributes of the decision makers, hence no attributes of the alternatives is included

For the case in this research, 𝑦 was described for two alternatives of shopping mode between online shopping and in-store shopping Typically, for the choice of online shopping alternative will be gotten the value equal to one (𝑦 = 1), because the decision makers have taken the choice of interest being online shopping mode has happened Otherwise, 𝑦 obtains the value equal to zero (𝑦 =0) due to the person does not taking the choice of online shopping, meaning the choice of the rest alternative has in-store shopping The model, or particularly the probability of purchasing an item online, is then specified as

Pr(𝑦 = 1) = 𝑓(𝛽|𝑋) where 𝛽 is the parameter to be estimated, and 𝑋 is the vector of explanatory variables

Choice probabilities in the binary choice situation, in the fact that the choice probability of the decision maker or individual consumer between online and store shopping, which can be expressed the succinct generally form as (Ben-Akiva and Lerman, 1985)

So far, a few studies utilizing binary choice models for analyzing the choice of shopping channels Degeratu et al (2000) explored the effects of brand name, price and information availability on the consumer choice behavior in online stores versus supermarkets Chocarro et al (2013) determined the impacts of key purchase situation variables on the choice between traditional outlet versus online store, through the survey in Pamplona, Spain Meanwhile, Crocco et al (2013) implemented an online survey on the citizens of Cosenza and Rende, located in the South of Italy to examine the factors that influences the inclination of people to use online shopping versus in-store shopping Suel et

al (2015) investigated the disaggregate relationships in online and store shopping, focusing

on grocery shopping activities of residents in London Using 952 Internet users for an online survey in two cities in Northern California in 2006, Zhai et al (2016) explored the interactions between e-shopping and store shopping in the four stages of shopping process (consciousness of the product, searching for information, product testing, and transaction) for two categories of products: search goods (books) and experience goods (clothing) Additionally, Arce-Urriza et al (2017) also examined the distinct impacts of price promotions

on orange juice brand choice between online and offline channels

In general, most of these studies conducted survey to collect data Degeratu et al (2000) conducted a longitudinal experiment from a separate sample of online shoppers and store shoppers While, Croco et al (2013) asked each respondent to depict their last purchase by revealing the adopted shopping channel (online or in-store), together with the category of purchased products They in turns gathered information about socio-demographic variables (i.e gender, age, education, income and other things), the product’s information and information of consumer attitudes towards shopping, which was assessed

by five-point scale Similarly, Zhai et al (2016) concentrated on four variables that encompass 42 attitudinal statements pertained to five-point scale In addition, Arce-Urriza et

al (2017) used data in the past purchases over six months in 2007 of a supermarket chain in Spain

Logit or Logistic regression was frequently applied as the econometrics model Logistic regression allows to examining the multivariate effects of independent variables on the agent’s choice of two alternatives, and to predict the probability to shop online or in-store Explanatory variables usually include:

 The perception and attitude towards online shopping, usually including the risk received

by consumers concerning online shopping (for example risk relating credit card), interesting advantages of online channel (having online assistant in real time, taking the discounts, large variety of providing products and procuring more information), online shopping experience, service quality elements and trust of online transactions (Chang et al., 2005; Chocarro et al., 2013 and Crocco et al., 2013)

 The characteristics of consumers, normally containing socio-demographic variables, Internet or computer usage knowledge and psychological variables (Chang et al., 2005; Chocarro et al., 2013; Crocco et al., 2013 and Zhai et al., 2016)

 The product or website characteristics, regularly comprising the cost, trial and tangibility

of the products, and website features as design to compare the products to each other (Chang et al., 2005; Chocarro et al., 2013; Crocco et al., 2013 and Zhai et al., 2016) Degeratu et al (2000) found an interesting result that online shoppers have less time

to go to store but more money, hence they are less price-sensitive, and their opportunity cost tend to be higher to search the vouchers Arce-Urriza et al (2017) also indicated that offline channel is more elastic to promotions than online, meaning the same consumers are more likely to respond to promotions in offline than online Online shoppers favor to online shopping since the value of time is evaluated highly (Degeratu et al., 2000) The same results are also found in Chocarro et al (2013), where the probability of choosing online shopping was affected highly by distance to store and time pressure For search goods, shoppers are more likely to shop online due to distance to store, meanwhile, social-demographic only impacts on high-involvement goods Zhai et al (2016) also stated that search good as book tend to purchase via online store than experience good as clothing Because the experience goods (clothing), one-third of the Internet buyers tend to travel to stores than online shopping due to the demand of information searching and product trial

Additionally, according to Croco et al (2013) shown the interacting with shop assistants and risk concerning credit card have a negative impact on online shopping This implies that consumers who consider the interaction with shop assistants and risk concerning credit card are important are less likely to shop online Meanwhile, negotiating has a positive effect on the choice of online shopping, which implies that those who think the negotiation is important are more likely to shop online The opportunity of immediately having products

has a negative impact on online shopping, meaning that if the consumers concern about the immediately having products attribute is important, they are less likely to shop online

2.2 Frequency of online shopping

The identification feature of this type of analysis is the dependent variable, which usually embraces two manners The first form of 𝑦 receives the values in a manner of ordinal scale, for instance, the frequencies of online searching will be assigned ordered values like less than one month or never; infrequent (at least one a month) and frequent (at least one a week) (Farag et al., 2005); or online buying [never, infrequent (1-5 times in the past year) and frequent (at least 6 times in the past year)] or frequencies of in-store shopping The second kind of dependent variable 𝑦 labels frequency of shopping trips or (average) numbers of trips in a last week or last month Hence, 𝑦 recognizes the discrete variable value, and if the mean values of 𝑦 are small, then the appropriate model is Poisson, the data therefore follows Poisson distribution

More specifically, the general function type for these dependent variables is usually expressed in two forms The first one is ordinal probit model, for example:

Let 𝑦 = 1, 2, 3, 4 be the frequencies of online searching or online buying

Let 𝑦∗ a continuous latent variable of online searching or online buying frequency 𝑦∗ can not observe, but we can observe the categories of frequency instead

 Pr(𝑦 = 1) = 1 − ∅(𝛽𝑥 − 𝑢1)

 Pr(𝑦 = 2) = ∅(𝛽𝑥 − 𝑢1) − ∅(𝛽𝑥 − 𝑢2)

 Pr(𝑦 = 3) = ∅(𝛽𝑥 − 𝑢2) − ∅(𝛽𝑥 − 𝑢3)

 Pr(𝑦 = 4) = ∅(𝛽𝑥 − 𝑢3)

The second model is Poisson distribution with a random variable 𝑦 is said to have a Poisson distribution with parameter 𝜆 > 0 if it takes integer values 𝑦 = 1, 2, 3, … , 𝑘 is the numbers

of online shopping time or frequency of online searching in the past 12 months intervalwith probability

A significant number of studies have investigated the frequency of online shopping or shopping trips up to now In the series studies by Farag and his colleges, Farag et al (2005) has conducted a research about the relationship between online and in-store shopping by analyzing the correlation among online searching frequency, online purchasing, and nondaily shopping trips, controlling for socio-demographic, land use, behavioral, characteristics and attitudinal Analyzing the basic factors of online buying and their relationship with in-store shopping by using empirical data obtained from Minneapolis, USA, and Utrecht, the Netherlands (Farag et al., 2006) Farag et al (2007) examined the relationship between frequencies of online buying, online searching and non daily shopping trips, at the same time considering the effects of attitudes, behavior and land use features on them Furthermore, Dijst at el (2008) analyzed the factors related to demographic, land uses, ownership of technology, internet skills and personality traits on the frequency of online buying Farag et

al (2005) requested 826 respondents participate in the shopping survey in one municipality (Utrecht) and three surrounding suburban in the Netherlands A shopping survey and a 2-day travel diary were designed Farag et al (2006) conducted the experiment in two ways First, they defined two categories of individuals: those who have and have not bought online

Second, for multivariate analysis, the frequency of online buying for those who ever bought online was also analyzed In-store shopping is identified as the average number of trips (per week for daily shopping; per month for non-daily shopping) and the average shopping duration in minutes per trip

The econometrics models for this approach include Poisson model and the like (i.e negative binomial), Ordered Logit/Probit, and interval regression

The common findings indicated that online searching has a positive impact on both store shopping and online buying Also, Internet usage has a positive influence on the frequency of shopping trips The implication is that the relationship between online and in-store shopping are complementary, not substitute [Farag et al., (2005); Farag et al (2006); Farag et al (2007)] Moreover, in the key result of Zhou and Wang (2014) stated that online shopping promotes shopping trips, in contrary to shopping trips tend to prevent the online shopping propensity

in-2.3 Random Utility Models (RUM)

In this method, the decision makers choose one alternative which provides the highest level

of utility from a set of more than two alternatives (Marschak, 1960 and McFadden, 1974) Utility level of each alternative is determined by attributes of the alternative This part of utility is called the systematic component The other is the random component, which cannot be observed by the researcher (Train, 2002) As a result, utility from alternative 𝑖 includes the systematic component 𝑉𝑖 and the random component 𝜀𝑖

et al., 2016)

Under the assumption that the error component follows Gumbel distribution, the choice probability that the person chooses either store shopping or e-shopping is defined as

𝑃𝑖 = 𝑒

𝑉 𝑖

∑ 𝑒𝑉𝑗𝑗

Very few studies apply RUM to analyze the choice of shopping mode because of difficulties in collecting data Nevertheless in a detached segment of the literature, Lee and Tan (2003) have used RUM to analyze consumers’ choice between online and in-store shopping Retail’s reputation and product’s risk were the two factors that impact on consumer’s choice Hsiao (2009) explored the consumer’s choice between physical store and e-shopping, constrained by the time attributes as value of travel time and value of product delivery time A recent study of Schmid et al (2016) investigated the choice of 339 participants in Zurich, Switzerland through trade-off different attributes between online versus in-store shopping for experience and search goods

Most of the studies using RUM in analysis conducted experimental design to survey the stated preference data based on revealed preference data

Using conjoint analysis as estimation method, Lee and Tan (2003) found that consumers with shopping experience are more likely to purchase the goods of low purchase risks on the Internet On the other hands, people are more likely to purchase the products of famous brands online store than less well-known ones Using the conjoint analysis, Hsiao (2009) found that value of travel time is $5.29/hour, while value of delivery time is

$0.53/day This implies an online bookstore will have to lower a value of book’s price by

$0.53 to capture a shopper in physical bookstore when one day is delayed for delivery time This value of one waiting day, however, is much lower than that of one hour travelling to store Meanwhile, Schmid et al (2016) found that in Switzerland, the value of travel time saving (40 CHF/hour) is also high comparing to the value of delivery time saving (16 CHF/hour), showing a comparative advantage for the online channel

Moreover, studies found that shoppers with high income have more positive attitudes toward online shopping (Schmid et al., 2016) Hsiao (2009) found that individuals with e-shopping experience are more likely to purchase online to save travel time and cost However, people who do not have e-shopping experience also consider buying more via

online when travel cost increases and delivery time decreases The monetary cost of a way trip to a bookstore amounted to USD 5.58 and average 5.48 days waiting for a purchased book that is equivalent a monetary value of USD 2.9 In other words, if a consumer purchases a book over the internet instead of in-store, they will save USD5.58 of a trip to bookstore, however, they have to wait around 5.48 days that equal to USD2.9 of monetary cost

two-2.4 Structural Equation Models (SEM)

SEM modeling is most commonly applied in exploring consumer shopping channel choices because of the convenience of collecting data of the dependent variables In this method, 𝑦

is defined by a Likert scale For example, 𝑦 takes the values of a rating scale by placing the code as “1” for never, “2” for less frequent, “3” for frequent and “4” for very frequent, which

to answer for the question of “how frequent do you shop online?” The general equation of SEM as follows

𝑌 = 𝛽𝑌 + 𝑋𝜏 + 𝜀 Where:

 𝑌 contains endogenous variables such as online shopping frequency

 𝑋 includes exogenous variables, which normally include demographics, Internet experience and shopping attitudes

 𝛽 coefficients representing direct effects of endogenous variables on other endogenous variables

 𝜏 coefficients representing direct effects of exogenous variables on endogenous variables

 𝜀 the errors

Many existing literatures have investigated the frequent of online buying which is estimated by SEM method Cao et al (2012) explored the relationship between online buying, in-store shopping and the searching of product information via the Internet by using

539 adults Internet users in the Minneapolis-St Paul metropolitan area In addition, Cao et al (2013) used the data from the shopping survey in the Twin Cities of 585 adults Internet users

to investigate the association between spatial attributes and online shopping frequency

Likewise, Zhou and Wang (2014) examined the relationship between online versus shopping trips, and consider whether online shopping reduces the demand of in-store shopping due to door-to-door delivery services Data is derived from the National Household Travel Survey (NHTS) in 2009 database

The estimation methods employed is maximum likelihood with Poisson distribution Cao et al (2012) applied the SEM with the main variables such as shopping behavior comprises frequency of online purchases, frequency of in-store trips and frequency of product information search via the internet, as well as 15 statements questions of shopping attitudinal variables which are judged by five-point Likert scale from strongly disagree to strongly agree Internet experiences were also considered in the model Still apply SEM in analysis, however, Cao et al (2013) implemented to test two hypotheses The first hypothesis includes the endogenous variables that are urban area indicator and e-shopping frequency (categorical variables), while the exogenous variables are social demographics, Internet experience and shopping attitudes Implementing the weighted least squares with robust standard errors and mean adjusted chi-square to estimate the model The second hypothesis evaluated the effects of shopping accessibility on the online shopping frequency

is separated into three equations of urban, suburban and exurban neighborhoods location Cao et al (2012) found that online shopping frequency has positive impacts on in-store shopping and online searching frequency has also positive effects on online shopping and in-store shopping frequencies These interactions shown that online shopping has a complementary to in-store shopping, besides that it seems to be a challenge to travel reduction Online buying encourages in-store shopping trips also pointed out by Zhou and Wang (2014) Furthermore, Zhou and Wang (2014) claimed that shopping trips tend to restrain the proclivity of online shopping Besides, the factors of demographics, regional and household attributes impacted on both online and in-store shopping frequency The main findings of Cao et al (2013) shown that the geographical areas have considerably affected

on the e-shopping because of the shopping accessibility tend to be various and accidental positions in metropolitan locations Additionally, urban region has the high Internet accessibility, which promotes more online shopping than the counterparts with low shopping accessibility Whereas, the low Internet accessibility in exurban areas have not stimulated to e-shopping

CHAPTER 3:

RESEARCH METHODOLOGY

As discussed, binary logit model limits to one specific choice and does not account for stores’ attributes The SEM which usually use of Likert-scale shopping frequency does not really explain the behavior of online shopping and its drivers Meanwhile, the frequency of online shopping cannot be reflected exactly when collecting data because of the difficulty in revealed preference data All these three approaches cannot analyze the effect of store attributes on consumer’s choice Therefore, this paper does not apply the binary choice model, SEM as well as the frequency of online shopping to examine the impacts of those main variables on shopping mode choice behavior Instead we apply choice experiment in order to deeper investigate the impacts of store’s attribute such as purchase price, delivery time and shipping fee, as well as travel time and travel cost on the choice of the two shopping channels

3.1 The choice of online shopping versus traditional store shopping

Random utility theory stem from the Law of Comparative Judgment proposed by Thurstone

(1927) in the seminal on psychophysical discrimination with explaining dominance judgments among pairs of alternatives (Adamowicz et al., 1998) This law postulated that consumers when facing choice among mutually exclusive alternatives would choose the alternative 𝑖 that has the best stimulus level 𝑈𝑖 The stimulus comprises of two components, the systematic and random components, or 𝑈𝑖 = 𝑉𝑖+ 𝜀𝑖 When the perceived stimulus is interpreted as utility, the theory became an economic theory (McFadden, 2001) Marschak (1960) has developed a theoretical framework for the choice probabilities of utility

maximization that included random components, which is called Random Utility

Maximization (RUM) (McFadden, 2001)

According to McFadden (2001), utility function is specified as 𝑈(𝑥), where 𝑥 is the vector of consumed quantities of various goods If the consumer prefers a bundle of goods

𝑋1 = {𝑥11, … , 𝑥𝑛1} to 𝑋2 = {𝑥12, … , 𝑥𝑛2}, then 𝑈(𝑋1) > 𝑈(𝑋2) The utility maximization subject to a budget constraint 𝑝𝑥 ≤ 𝐼, where 𝐼 is consumer’s income and 𝑝 is the vector of price with demand function 𝑥 = 𝑑(𝐼, 𝑝) The general problem was established as following

𝑀𝑎𝑥{𝑥1,…,𝑥𝑛} 𝑈(𝑥1, … , 𝑥𝑛)

S.t ∑𝑛𝑖=1𝑝𝑖𝑥𝑖 ≤ 𝐼

If we know the utility function and income of every consumer, the problem will be resolved for each consumer From that, we will find out the total market demand for many consumers and 𝑋 = {𝑥1, … , 𝑥𝑛} is the consumer demand from good 1 to good n

In discrete choice framework, 𝑈𝑛𝑖 is the utility that agent of choice (decision maker, person, firm) 𝑛 gets from a set of bundle options (alternatives) The utility-maximizing is that

a consumer 𝑛 has to face with a choice set 𝐶 which composes 1, … , 𝑗𝑛 numbers of alternatives Each consumer must choose only one alternative from a choice set The choice

of the consumer is indicated by a binary variable, 𝑌𝑛𝑖, for each alternative:

𝑌𝑛𝑖 = {1 𝑖𝑓 𝑈𝑛𝑖 > 𝑈𝑛𝑗 ∀𝑗 ≠ 𝑖

0 𝑖𝑓 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒The discrete model can be operationalized by specifying the utility function, usually

in a linear form where the utility of consumer 𝑛 from alternative 𝑖

𝑈𝑛𝑖 = 𝛽𝑥𝑛𝑖+ 𝜀𝑛𝑖

Ben-Akiva and Lerman (1985) supposed that attributes determine the utility and

these attributes are characteristics of each alternative In consumer choice behavior of choosing shopping mode, there are many attributes that determine the utility of shopping channel choice Nonetheless, the most prominent attributes are purchase price, time and cost (Bateman et al., 2002)

In studies of store choice, purchase price probably plays an important role in making decision Purchase price in online stores are typically lower than traditional stores This may

be because those selling online can save the overheads in terms of reducing the cost of hiring sales staff and renting space However, the online sellers have to charge the shipping cost, in addition to the selling price of commodities The final price is probably lower than the price of the same item outside the store (Hsiao, 2009) Furthermore, a major consumers subsist on urban areas tend to search information related to commodities online before they travel to store (Farag et al., 2007) The aim of this activity is to compare the sell price of the same good between online and in-store channel, and gather other necessary information As Grewal et al (2003) revealed that the asymmetric information of selling price in traditional

stores and the possibilities to make up the price difference of marketers can be significantly eliminated, which thanks to the transparency of price information on the Internet Besides, Koyuncu and Bhattacharya (2004) concluded that since the online shopping channel offers better prices, the consumers therefore prefer to buy more via the Internet

Travel is another attribute that creates the differences between online and traditional in-store shopping (Hsiao, 2009) The consumers must spend travel time and travel cost in order to conduct a shopping trip to store In the meantime, if the consumers choose

to shop online, they need to spend time for ordering, time waiting for the delivery of products and may be pay for shipping fee Truong and Hensher (1985) pointed out that disutility of individuals was caused by travel time In other words, Hsiao (2009) stated that if the consumers waste their travel time and travel cost, these things will have direct effects on their values, and therefore leading to the diminishing utility of consumers In this case, shopping online appears to be a better choice for consumers

If the consumer’s purchase online, after finalizing the payment transaction for the product’s value, they have to wait for product delivery Only when the products are music (mp3 file), software or certain kinds of services like Internet banking and online consultation, then no delivery time is needed However, the satisfaction of products will be decreased if the products are delayed in terms of delivery In addition, an uncertainty about the products will be created, because it will be associated with product’s quality which the consumers often predict (Liu and Wei, 2003) Moreover, Koyuncu and Bhattacharya (2004) found that delivery time is was one of the reasons that has direct effect on the choice of consumers, if longer delivery time would cause less purchase to buy over the Internet from consumers

Besides the main effects of these attributes, the individual characteristics and attitudinal forward shopping channels play an important role According to Adamowicz etal (1998), the choices can vary systematically from individual to individual, and to compute for this differences as much as possible, the set of explanatory variables can be extended to include individual differences such as demographic and psychological factors, 𝑧, with respect

to the vector of coefficients 𝛾.These individual difference measures may be hypothesized to influence utility levels via intercept and/or slope coefficients in the 𝛽 vector

Based on RUM, the customers make the decision to choose either online shopping or in-store shopping whichever generates the higher level of utility The two utility functions for in-store and online shopping are functions of attributes:

 𝐴𝑆𝐶𝑜𝑙 is the alternative specific constant of online shopping

 𝑝𝑟𝑖𝑐𝑒𝑜𝑙: The effect of purchase price of online shopping on online shopping utility

 𝑜𝑟𝑑𝑒𝑟𝑡𝑖𝑚𝑒𝑜𝑙: The effect of ordering time on online shopping utility

 𝑑𝑒𝑙𝑖𝑣𝑒𝑟𝑦𝑡𝑖𝑚𝑒𝑜𝑙: The effect of delivery time on online shopping utility

 𝑠ℎ𝑖𝑝𝑝𝑖𝑛𝑔𝑓𝑒𝑒𝑜𝑙: The effect of shipping fee (or delivery cost) on online shopping utility

In the equation (2), 𝑉𝑖𝑠 is the indirect utility function of in-store shopping channel, which contains the key variables as

 𝑝𝑟𝑖𝑐𝑒𝑖𝑠: The effect of in-store purchase price on utility of in-store shopping

 𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔𝑡𝑖𝑚𝑒𝑖𝑠: The effect of shopping time on in-store shopping utility

 𝑡𝑟𝑎𝑣𝑒𝑙𝑡𝑖𝑚𝑒𝑖𝑠: The effect of travel time (to store) on in-store shopping utility

 𝑡𝑟𝑎𝑣𝑒𝑙𝑐𝑜𝑠𝑡𝑖𝑠: The effect of travel cost on in-store shopping utility

The Full model in this study is estimated with attributes interacted with individual characteristics:

 Online ASC (𝐴𝑆𝐶𝑜𝑙) is interacted with dummy variable for office worker, and internet access frequency (hours/day) This is to allow for difference in online shopping among office workers and non-office workers, and among those with different internet access frequency

 Prices of both alternatives (online and in-store) are interated with income This is to allow for different price sensitivity among shoppers with different income

 Delivery time is interacted with gender to allow for female shoppers may be more patient in waiting for the delivery

 Travel time (to store) is interacted with income, as shoppers with higher income may have higher opportunity cost of time

The utility function of the Full model is then:

 𝑑𝑒𝑙𝑖𝑣𝑒𝑟𝑦𝑡𝑖𝑚𝑒𝑜𝑙× 𝑔𝑒𝑛𝑑𝑒𝑟: People who are female or male with respect to the sensitivity of delivery time, which have the additional effect on utility of online shopping

 𝑝𝑟𝑖𝑐𝑒𝑖𝑠× 𝑖𝑛𝑐𝑜𝑚𝑒: People who have the different levels of income with respect to the sensitivity of in-store purchase price, which have the additional effect on utility of online shopping

 𝑡𝑟𝑎𝑣𝑒𝑙𝑡𝑖𝑚𝑒𝑖𝑠× 𝑖𝑛𝑐𝑜𝑚𝑒: People who have the different levels of income with respect to the sensitivity of travel time, which have the additional effect on utility of online shopping

From the estimated coefficients, we can calculate the willingness to pay (WTP) of time and cost attributes WTPs can be calculated from the Basic model as follows

𝑊𝑇𝑃𝑜𝑟𝑑𝑒𝑟𝑡𝑖𝑚𝑒 = − 𝛼2

𝛽1 (5) 𝑊𝑇𝑃𝑑𝑒𝑙𝑖𝑣𝑒𝑟𝑦𝑡𝑖𝑚𝑒 = −𝛼3

𝛽1 (7) 𝑊𝑇𝑃𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔𝑡𝑖𝑚𝑒 = −𝛽2

𝛽 1 (6) 𝑊𝑇𝑃𝑡𝑟𝑎𝑣𝑒𝑙𝑡𝑖𝑚𝑒 = − 𝛽3

𝛽 1 (8)

We use the 𝛽1 (coefficient of price in the in-store utility function) as the marginal utility of money instead of 𝛼1 (the online shopping utility function), as people have to pay immediately when shopping in-store while they may pay later when shopping online and as

a result, 𝛼1 may not really reflect the marginal utility of instant money Equation (5) shows the WTP of one minute spend for searching and placing the order over the internet (VND/min) Equation (6) exhibits the WTP of one minute spend for finding and buying the items of product at the store (VND/min) Equation (7) shows the WTP of one day waiting for the delivery of purchased products (VND/day) Equation (8) exhibits the WTP of one minute spend for travelling to a store (VND/min)

Table 3 1: The store choice attributes description

𝑝𝑟𝑖𝑐𝑒𝑜𝑙 VND 1000 Purchase price of online shopping:

the price when buying online

Negative (-)

𝑜𝑟𝑑𝑒𝑟𝑡𝑖𝑚𝑒𝑜𝑙 Minute Ordering time: the time spends for

searching and placing the order of books over the internet

Negative (-)

𝑑𝑒𝑙𝑖𝑣𝑒𝑟𝑦𝑡𝑖𝑚𝑒𝑜𝑙 Days Delivery time: the day of waiting for

the delivery of purchased books

Negative (-)

(within 1 day, 2-3 days and >= 3 days)

𝑠ℎ𝑖𝑝𝑝𝑖𝑛𝑔𝑓𝑒𝑒𝑜𝑙 VND 1000 Shipping fee (or delivery cost): the

monetary value which pays for the delivery of purchased books (VND 0, VND 15 and VND 22)

Negative (-)

𝑝𝑟𝑖𝑐𝑒𝑖𝑠 VND 1000 Purchase price of in-store shopping:

The purchase price at the bookstore

Negative (-)

𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔𝑡𝑖𝑚𝑒𝑖𝑠 Minute Shopping time: the time spent for

finding and buying books at the bookstore

Negative (-)

𝑡𝑟𝑎𝑣𝑒𝑙𝑡𝑖𝑚𝑒𝑖𝑠 Minute Travel time: the time spends to go to

the bookstore

Negative (-)

𝑡𝑟𝑎𝑣𝑒𝑙𝑐𝑜𝑠𝑡𝑖𝑠 VND 1000 Travel cost: the monetary value

which depends on the reported transportation and the distance to the store for the last purchase in the questionnaire

Negative (-)

Calculating the WTPs using the Full model is more complicated In the Full model, the marginal utility of money is 𝛽1+ 𝜑1𝑖𝑛𝑐𝑜𝑚𝑒̅̅̅̅̅̅̅̅̅̅, where 𝑖𝑛𝑐𝑜𝑚𝑒̅̅̅̅̅̅̅̅̅̅ is an average income The WTPs for these attributes is now

𝑊𝑇𝑃𝑜𝑟𝑑𝑒𝑟𝑡𝑖𝑚𝑒 = − 𝛼2

𝛽1+𝜑1𝑖𝑛𝑐𝑜𝑚𝑒 ̅̅̅̅̅̅̅̅̅̅̅ (9) 𝑊𝑇𝑃𝑑𝑒𝑙𝑖𝑣𝑒𝑟𝑦𝑡𝑖𝑚𝑒 = −𝛼3 +𝛼 4 ×𝑔𝑒𝑛𝑑𝑒𝑟̅̅̅̅̅̅̅̅̅̅̅

𝛽1+𝜑1𝑖𝑛𝑐𝑜𝑚𝑒 (10) 𝑊𝑇𝑃𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔𝑡𝑖𝑚𝑒 = − 𝛽2

𝐼𝑓 𝑈𝑂𝐿,𝑛 > 𝑈𝐼𝑆,𝑛: 𝑐ℎ𝑜𝑖𝑐𝑒𝑛 = { 𝑂𝑛𝑙𝑖𝑛𝑒 𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔

𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 𝐼𝑛 − 𝑠𝑡𝑜𝑟𝑒 𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔Therefore, the probability of choosing shopping channel is calculated as

This study use both Revealed Preference (RP) data and Stated Preference (SP) The data collection used for this paper is organized in two stages Stage I contains the questionnaires based on RP method to obtain the revealed data of the respondents Stage II comprises the chocie experiment, also known as SP method which was developed from stage I By doing so, we can get the stated choice from the respondents According to Louviere et al (2000), revealed preference method collect the shopping mode data in “the world as it is”, meanwhile stated preference method identified “the world as it could be” to infer the shopping channel choice behavior In other words, RP survey points out that what the respondents actually did, while, SP survey expresses what the respondents would do in the hypothetical scenarios

Table 3 2: Individual characteristics variable description

Occupation 1 = students; 2 = housewife; 3 = government

officers; 4 = freelancers; 5 = company employees;

6 = researchers; 7 = unemployment; 8 = retired and 9 = others

Monthly income In million

VND

1 = Under 5; 2 = 5 – 10; 3 = 11 – 15; 4 = 16 – 20; 5

= 21 – 25; 6 = 26 – 30; 7 = 31 – 35; 8 = 36 – 40; 9 =

41 – 45; 10 = 46 – 50 and 11 = Over 50

Internet access

frequency

In hours 1 = 2 – under 4; 2 = 4 – under 6; 3 = 6 – under 8; 4

= 8 – under 10 and 5 = Over 10

Figure 3.1 describes the conceptual framework of the choice between online shopping versus in-store shopping

Each method of Revealed and Stated Preference has their own advantages and disadvantages, so, in order to understand why the researches often incorporate both approaches, we first examine shortly about the two methods in the way how to collect research data

Figure 3.1: Choice model between online and in-store shopping

3.2 Revealed preference data: methods of collection

Revealed Preference Theory is developed by Paul Samuelson (1938), a Nobel Prize winner, with the aim to identify the preferences of an individual (utility function) by observing actual choice behavior The greatest advantage of RP data is that the respondents revealed the actual or observed choices made in authentic existence situations (Morikawa, 1989)

Utility: Online vs

in-store shopping

Choice of shopping mode

Online shopping attributes

Online ASC* Internet access frequency

Therefore, revealed preference technique can avoid potential issues linked with hypothetical responses (McFaddden, 1974) which usually do not consider accurately about the choice behavior and strategic responses In a decision making processing for utility maximizing, the agents often compare the alternatives among the others by estimating the benefits from observed actual risks (McFadden, 1974a) Additionally, Louviere et al (2000) designated that revealed preference is an adequate method for small number of observations without any assumptions of functional forms

Nevertheless, the application of revealed preference means that data merely observes what the choice was and the implication is that we cannot know surely what the choice set was Consequently, Ben-Akiva and Lerman (1985) indicated that revealed preference survey collects less observation because of high cost Moreover, the strong correlation between the attributes concerning the value of travel time, value of delivery time and travel cost, hence the effect’s estimation of independent variables on the utility maybe bias The range of attributes is limited due to the dependence in the range of the actual decision made by particular consumption of existing products situations (Adamowicz et al., 1997), which may result in multicollinearity, estimation efficiency and endogeneity Moreover, Adamowicz et al (1994) claimed that revealed preference can better estimate the use value and market value, while RP data with intangible attributes associated to reliability, convenience and comfort are difficult to measure by the researcher Revealed preference survey cannot be applied to estimate the demand for new modes, since the information for new modes do not available or non-existing alternatives, instead stated preference survey would does (Morikawa, 1989, Ben-Akiva et al., 1994 and Louviere et al., 2000) Hence, the preference indicator is usually as choice (Morikawa, 1989)

Nevertheless, the application of revealed preference means that data merely observes what the choice was and the implication is that we cannot know surely what the choice set was Consequently, Ben-Akiva and Lerman (1985) indicated that revealed preference survey collects less the observations because of high cost Moreover, the strong correlation between the attributes concerning the value of travel time, value of delivery time and travel cost, hence the effect’s estimation of independent variables on the utility maybe bias The range of attributes is limited due to the dependence in the range of the actual decision made by particular consumption of existing products situations (Adamowicz et al.,

1997), which may result in multicollinearity, estimation efficiency and endogeneity Moreover, Adamowicz et al (1994) claimed that revealed preference can better estimate the use value and market value, while RP data with intangible attributes associated to reliability, convenience and comfort are difficult to measure by the researcher Revealed preference survey cannot be applied to estimate the demand for new attributes, since the information for new attributes are not available or non does not exist, instead stated preference survey would do (Morikawa, 1989, Ben-Akiva et al., 1994 and Louviere et al., 2000)

The respondent’s shopping mode choices were conducted by a personal interview with the questionnaire to record the revealed responses in real-life Following previous studies, we assume that different shopping modes give shoppers different utility levels, depending on their intrinsic attributes This study elicits the utility function of shopping channel, online and in-store, from the choice decisions The questionnaire is designed to collect respondent’s choices on choosing chopping modes from their book purchases in the last 12 months

Firstly, to identify whether will the respondents be the right individual for this interview Hence, respondents will be asked: “In the last 12 months, did you buy book in-store or online?” If yes, they will be continue with “for the last purchase, are you the decision maker to buy?” If they are the decision maker, they keep going on the interview

Next, the general information of respondents as frequency of buying books will be displayed by asking: “How many books buying times have you bought in the last 12 months?” and “How far months ago have you bought to the last purchase?” Then, the respondents were asked to reveal the past experience about book shopping and Internet use The Internet use experience related to some questions like: “what is the most common device have you often accessed the Internet?” (in respect of mobile phone/ Laptop, PC/ Tablet/ Others one), and “how many hours per day have you accessed the Internet?” To identify the categories of individuals, those who have ever bought books via online or not,

“Did you ever use the Internet for purchasing books”

After all, to perceive the purpose of buying book of respondents, we asked them “the reason why you decided to buy the book?” Finally, respondents were also required to answer the question about “Where did you buy the books in the last purchase?”, either

online channel or in-store channel From that, respondents will be moved to the questions in terms of the last purchase of in-store shopping or online shopping

If respondents made their shopping on in-store, they were also requested to respond

to the questions about online shopping with associated attributes according to their knowledge (Hsiao, 2009), and vice versa The attributes of revealed preference method were classified for the last purchase in both channels by these questions: “how much the total purchase price have you spent for the last book purchase?(VND)” Moreover, we also asked:

“if you have bought via online (or in-store) for the last purchase, the total purchase price will

be more expensive or cheaper? About how many percent? These questions are to identify the purchase price for the last time

The aim of collecting the information in value of time and cost that will be asked through these questions: “how many minutes have you spent for shopping time (ordering time)?” Besides, to visibly ascertain the shopping time or ordering time that respondents have spent, we have: “Did you only buy books in the last purchase?”; “How many minutes have you spent for travel time in one trip? Or “how many days for delivery time?” Besides that, the travel cost can be defined by “how much travel cost have you spent for the last purchase?” In the case that respondents cannot estimate the value of travel cost, we can suggest that “what kind of transport have you used for the last purchase?” and “how far for distance to bookstore?” Respectively, we have “how much delivery cost for the last purchase if you have bought via online?”

Moreover, each respondent was asked to provide the personal information such as gender, age, marital status, education, occupation and income level In addition, they were also asked to rate 18 attitudinal statements with five-point Likert scales From that, this study can collect the database which is a foundation for stated preference data

3.3 The stated preference data: methods of collection including choice

experimental design

3.3.1 The stated preference method

Stated preference method base on hypothetical scenarios to ask the respondents give their choice This against to what the respondents actually did in the real world Hence, a stated

preference survey has more observations for an experiment And the process to make respondent’s choice is controlled by the researcher

Morikawa (1989) summarized the advantage characteristics of stated preference Since the respondent’s choices in the survey were presented by the researcher, we, therefore, can define surely what the choice set was Not only that, the level of attributes can be extended because of independent of variables in real-world scenarios Additionally, the high collinearity between the variables as time and cost which can be avoided or minimized due to the researcher examined to combine the levels of attributes when design the experiment So, the measurement error of all variables can be evaded Besides, the intangible variables can be gathered thanks to the rating of respondents We can infer the preferences for new alternatives Consequently, the preferences indicators can be elicited reasonably as choice, rate and rank The implication is that the information is collected from stated preference survey which is larger than revealed preferences

However, the biggest weakness of stated preference data encounters lesser validity than revealed preferences (Morikawa, 1989) Because stated preferences collect the reported data through presented hypothetical scenarios to the respondents and asking for their choice (Morikawa, 1989 and Fifer et al., 2014), resulting in biases in estimation

Table 3.3: Attributes and levels in stated choice experimental design

Each variable (attribute) has their levels Using the information of attributes received from respondent’s answer in stage I revealed preferences, in which is to derive the levels of each attributes This is to enhance individual preference revelation (Rose et al., 2008 and Schmid et al., 2016) Table 3.3 shows the attributes and attribute’s levels were hypothesized

to impact on the choice between online versus in-store shopping

Based on Table 3.3, the pivot design produces the different individual choice situations The attribute’s levels were modified depending on the reference value what we collected from revealed preference data

3.3.2 Choice of experimental design

According to Louviere et al (2000) to build up the stated choice questionnaires, the alternative scenarios or descriptions must be constructed to be presented to respondents

(Bateman et al., 2002) Different combinations of attribute levels generate the profiles when

the researcher applies some forms of orthogonal design (Louviere, 1988 and Adamowicz et al., 1998)

We code the variables of in-store shopping as purchase price 𝑋1, shopping time 𝑋2, travel time 𝑋3 and travel cost 𝑋4 𝑋1 has four levels, while 𝑋2, 𝑋3 and 𝑋4 three levels All possible combinations can be generated from these attribute levels 4x33=108 alternatives (profiles) In the same manner for online shopping, we also have 108 possible profiles That

is so-called full factorial design Full factorial design has the property of orthogonality, in the

sense that the sample correlation between any two variables is zero Nonetheless, a large number of alternatives were produced and unfeasible designs for performing The reason is that we cannot ask the respondents to finish simultaneously 108 profiles in practical because

it is unrealistic

Therefore, methods of experimental design are developed to solve the problems of complete factorial design, which select a sample of profiles which have a specific set of statistical properties that can be used to estimate the utility specification (Adamowicz et al., 1998) The design selects a subset of full factorial design for using in a questionnaire in a statistically efficient manner (Bateman et al., 2002) Because a number of scenario

acombinations were reduced in a fractional factorial design, thus, in which the existing of all

or some interactions will not be detected However, fractional factorial design becomes effective in practical when presenting to respondents

In the experimental design, thus, we can estimate two types of effects of attributes

on the choices (Bateman et al., 2002), which contains main effects and interaction effects The main effects present the effects of each presented personal attributes On the other side, main effects are the orthogonal subset of the full factorial design, in which the researcher can estimate only the effect of independent variables on other variable’s values (Adamowicz et al., 1998) The interaction effects imply the connection level between behavior and variations in the combinations of the various attributes provided (Bateman et al., 2002) On the other hands, the interactions are the place that the effect of a variable is depicted as a function of other variable’s values

This study obtains the choice experimental design by using D-optimal designs,

including linear and quadratic function First, each attribute is treated by coding response to number from one to four in Table 3.4

The answer for the question of why we need to construct the choice sets is that because after experimental design has produced the profiles, then it will be grouped into the choice sets to be presented to respondents (Bateman et al., 2002) Choice sets which can be developed by combining a pair of alternatives or more alternatives together In this study, choice sets are the combination of two alternatives between online shopping and in-store shopping When the respondent makes their choice among two shopping channels, meaning that they are conducting the choice scenarios or the choice tasks

Table 3.4: The attribute’s levels of online and physical store

Level Purchase

Price

(OL)

Purchase Price (IS)

Ordering time (OL)

Shopping time (IS)

Shipping fee (OL)

Travel cost (IS)

Delivery time (OL)

Travel time (IS)