The Diffusion of Household Autonomous Robotic Vacuum Cleaners in China

The Diffusion of Household Autonomous Robotic Vacuum Cleaners in China

by

FANQIAO XU

University of Groningen

Faculty of Economics and Business MSc Marketing Intelligence

Schuitemakersstraat 2-56, 9711HW Groningen (06)43220916

f.xu.7@student.rug.nl

Student number S3311708

ABSTRACT

Household autonomous robotic vacuum cleaners (HARVCs) is a new phenomenon for the modern consumer, with its fast development, it is necessary for researchers and marketers to forecast the diffusion of this type of products. One of approaches to predict the product’s diffusion is by using Bass diffusion model, however, the approach to measuring the Bass diffusion model is controversial. This research aims at presenting measurements for the coefficients of the Bass diffusion model and predicts the performance of HARVCs with an automatic dirt disposal feature in the Chinese market using this model. Moreover, a conjoint choice-based (CBC) analysis is also conducted to examine consumers’ preferences with regard to HARVCs and to estimate the potential market share, which is one of the parameters in the Bass diffusion model. This study finds that the innovator coefficient is 0.02208037, the imitator coefficient is 0.1343183, and the predicted market share for one of the chosen products, Xiaomi New, is 211 million, and the predicted market share for another product, iRobot i7 Plus, is 55 million. Different with other distribution curves made by previous literatures, which have a slow increasing in sales at the beginning, the distribution curves for the chosen products show rapid increases in sales at the beginning and follow with a long tail until they reach saturation points. It is also found that the importance of the attributes considered in the CBC analysis were not influenced by different demographic segments. The results of this research provide novel insights into the adoption and diffusion of new product adoption and diffusion.

Keywords:

Bass diffusion model; choice-based conjoint; sales forecasting; household autonomous robotic

vacuum cleaners.

TABLE OF CONTENT

1. INTRODUCTION 4

2. THEORETICAL BACKGROUND 7

The Theory of the Bass Diffusion Model 7

Consumer Innovativeness 8

Consumer Imitativeness 10

The Formulation of Measures 12

Potential Market Share 14

The Conceptual Model 15

3. METHODOLOGY 18

Sample and Data 18

Measures 19

Data analytic strategy 20

The formulation of the Bass diffusion model 20

Calculation of parameters p and q 21

The formulation of the parameter m 23

4. RESULTS 25

Sample Characteristics 25

The measures of innovation p and imitation q 26

The estimation of market share m 30

Diffusion pattern of HARVCs 34

5. DISCUSSION 37

Implications 37

Limitations 39

Future Directions 41

6. CONCLUSION 42

7. ACKNOWLEDGEMENTS 42

8. REFERENCES 43

9. APPENDIXES 50

Appendix A - Questionnaire 50

Appendix B - Explanation of Items 56

Appendix C - Demographics results 57

Appendix D - The importance of attribute 58

Appendix E - Purchasing probability estimation 59

Appendix F - R-code 60

1. INTRODUCTION

Households now have a greater demand for automated products, as such products offer convenience and reduce the amount of time spent on household chores (Afonja, Alade, Asafa &

Olaniyan, 2018). Household autonomous robotic vacuum cleaners (HARVCs) are one of the most popular automated products. The demand for HARVCs has increased dramatically in recent years.

A report on the global demand for these products indicates that 54 million HARVCs, amounting to a value of approximately 2.1 billion dollar, were sold in 2015, and the sales volume is expected to increase to 429 million HARVCs, which are worth approximately 18.8 billion dollar between 2018 and 2020 (Guan, Huang, Xiao & Zhang, 2018). Compared to other types of autonomous vacuum cleaners, such as professional cleaners and special cleaners for companies, HARVCs occupied 70% of the total market share in 2015, which is expected to increase to 75% in 2019 (Chinese industry research institute, 2018). Given the increasing number of HARVCs among the general public, it seems that the market for these products has significant potential for development in the coming years (Franzmann, Mittendorf & Ostermann, 2018). It is thus necessary for business sectors and economists to develop a method for forecasting the development or diffusion of this type of product.

Various models for forecasting the diffusion of a product on a market, such as the logistic model,

the Gompertz model, and the Bass diffusion model is one of them which has been previously

investigated by scholars and applied in empirical studies (Gohs & Massiani, 2015; Elliot & Naseri,

2013, Kim, Lee, Lee & Park, 2013). However, there is no agreement on how the parameters in the

Bass diffusion model investigated should be measured. For example, the coefficient innovator, one

of the parameters in the Bass diffusion model, has been widely discussed before (Dowling &

Midgley, 1978; Eastman, Flynn & Goldsmith, 1996; Goldsmith & Hofacker, 1991; Hirschman, 1980), but there are different approaches to measuring this coefficient. Becker and Siduh (2009) conducted a simulation study in which they obtained values under three different scenarios;

however, Gohs and Massiani (2015) criticized Becker and Siduh’s study on the grounds that its justification was weak. Moreover, due to the difficulty of isolating the parameters of innovator and imitator, very little has been written about measures of the coefficient of imitator. Since approaches to measuring the Bass diffusion model are controversial, researchers have investigated these parameters and examined how they are measured.

These problems lead to the objective of this study: based on the Bass diffusion model, this research aims to identify appropriate measurements for the coefficients of the Bass diffusion model and attempts to forecast the acceptance in terms of the demand for HARVCs with an automatic dirt disposal feature (one of the newest features that have been added to HARVCs). The results of this study could help the business sector, policymakers, and organizations to develop better strategies for forecasting the development of innovative products. Hence, the approach of this research is to first obtain a clear overview of the influence of different factors on three important parameters - the innovator coefficient, the imitator coefficient, and the coefficient of market share - and to then measure these parameters through factor statements and conjoint analysis.

Addition to the application of the Bass diffusion model, a choice-based conjoint (CBC) analysis

test is conducted in this research. As a type of stated preference survey, it measures consumers’

preferences with regard to hypothetical combinations of attributes (Massiani, 2013). It has the advantage of allowing a researcher to measure the probability of a respondent choosing to adopt a new product, as well as the size of its potential market. Furthermore, this survey inquires as to respondents' preferences with regard to automatic dirt disposal, a feature that will soon be offered by HARVCs on the Chinese market. This feature has recently been added to HARVCs, and devices with this feature have only recently become available; HARVCs with this feature are only available in countries such as the Netherlands and the USA. After HARVCs with this feature was launched, it became a major success, and products with this feature have received many positive reviews (McCabe, 2019; McDonough, 2019; Brains, 2019; Seifert, 2018). In the near future, HARVCs with this feature will be launched in the Chinese market; given the differences in terms of culture and demand that exist between different continents and countries, the application of the conjoint analysis test would be intriguing. As this feature has not yet been made available in products on the Chinese market, it is unknown whether this feature will be considered essential by Chinese buyers and thus whether it can succeed.

The remainder of this study first presents a theoretical background on the Bass diffusion model and its three important parameters, which is followed by the development of a conceptual model.

Second, the methodology and the results are critically discussed. Finally, the implications and

limitations of this work and directions for future research are presented at the end of this study.

2. THEORETICAL BACKGROUND The Theory of the Bass Diffusion Model

Compared to traditional cleansing tools, HARVCs are considered to be a new propulsion technology; advanced models are required to forecast the sales of such new technologies. The diffusion of this new type of product is assumed to follow an S-shaped curve (Rogers, 1983), and the populations in a social system are considered to be heterogeneous in terms of their propensity to adopt (Rogers, 1983). Within this setting, the groups of adopters are divided into innovators, early adopters, early majority, late majority, and laggards (Rogers, 1983).

Recently, various models that have attempted to explain the actual shape of this diffusion curve

have been widely discussed (Young, 1993; Ord & Young, 1989; Chu & Wu, 2010; Islam & Meade,

1995). Conclusions with regard to the performance of various models have been inconclusive; for

example, Ord and Young (1989) tested both the logistic and Gompertz models and found that

neither demonstrated consistent performance, while, in Elliott and Naseri’s (2013) study, which

analysed time series data concerning online shopping in Australia, found that the Bass diffusion

model outperforms these two models. The present study adopts the view of Elliot and Naseri (2013)

and uses the Bass diffusion model to forecast the development of HARVCs. This model assumes

an S-curve-shaped of distribution, when there are more adopters, the probability of purchasing by

imitators increases, and, after the market becomes saturated, the number of potential adopters

decrease, as does the probability of purchasing (Kim et al., 2013). The Bass diffusion model

calculates the timing of initial purchases of new products based on the number of previous buyers

(Bass, 2004). It is assumed that potential buyers could be classified as falling into one of two

groups, namely innovators and imitators (Bass, 2004). Beyond these two elements, the diffusion pattern is also highly dependent on the value of the market potential (Gohs & Massiani, 2015).

The subsequent sections of this thesis present more details concerning the parameters of innovator

Consumer Innovativeness

The innovator parameter is one of the essential elements in the Bass model. Innovators are defined as “Individuals [who] decide to adopt an innovation independently of the decisions of other

unlike imitators, innovators only represent 1.5 to 2% of adopters in diffusion phases. This group of consumers is essential in understanding the diffusion of innovative products because they have a predisposition to adopt new products earlier than other consumers (Truong, 2013). Thus, consumer innovativeness is first evaluated in order to obtain a more accurate value for the parameter of innovators.

The extant literature defines innovativeness as the degree to which an individual is likely to make purchasing decisions or to adopt new products earlier than other members of a social system (Roehrich, 2004; Rogers and Shoemaker, 1971). This definition focuses on the level of consumer innovativeness that is observable in early adopters’ behaviour, while consumers who are high in innovativeness do not always adopt new items or are not the earliest adopters (Geuens &

Vandecasteele, 2010). This study focuses on HARVCs with features that are not yet available on

the market; thus, analysing innovativeness according to actual or observable behaviour is not realistic. Therefore, this study defines consumer innovativeness as the tendency to adopt innovative products and brands rather than to continue exhibiting previous consumption patterns and choices (Hofstede, Steenkamp & Wedel, 1999).

Studies measuring consumer innovativeness have been widely analysed over the last few decades (Baumgartner & Steenkamp, 1996; Kirton, 1976; Raju, 1980; Roehrich, 2004). Recent studies on consumer innovativeness (Bearden, Hunter & Tian, 2001; Grohmann, Spangenberg & Voss, 2003) have suggested that multidimensional motivations are appropriate scales for measuring consumer innovativeness, as motivations encourage goal-oriented purchasing behaviours, and they are activated by the goals that individuals pursue (Geuens & Vandecasteele, 2010). The functional motivational dimension is one type of multidimensional motivations; it indicates that individuals adopt innovation to improve their work performance or to increase their productivity in a qualitative way (Geuens & Vandecasteele, 2010). In addition, Geuens and Vandecasteele (2010) also examined the hedonic dimension and found that it motivates consumers to make innovative purchasing decisions based on feelings of excitement, joy, and satisfaction.

Furthermore, the social motivational dimension assumes that the self-assertive social need for

differentiation motivates consumer innovativeness (Geuens & Vandecasteele, 2010). Specifically,

purchasing innovation helps consumers to fulfil their social needs, such as those for individuality,

self-determination, superiority, and resource acquisition, which allows them to enhance their social

status, increase their likelihood of success, or to obtain approval from other members of a social

system (Geuens & Vandecasteele, 2010). Finally, the cognitive motivational dimension indicates that consumers are motivated by the need for mental stimulation; they tend to expand their cognitive limits, such as those concerning exploration, understanding, and intellectual creativity, through pursuing knowledge and engaging in thought (Geuens & Vandecasteele, 2010). The innovativeness of individuals can be measured with these four motivational dimensions.

Consumer Imitativeness

Another important parameter in the Bass diffusion model is the imitator parameter. Since the Bass diffusion model describes not only the time of adoption but also how consumers are influenced by other members of a social system (Bass, 2004), this study not only measures the value of the innovator parameter but also that of imitator. This segment of buyers plays a major role in diffusion pattern because it is a relatively larger segment than that of innovators (Jain & Takada, 1991).

Rogers (1983) divided adopters into five classes, Bass (2004), except the first-class innovator,

aggregated four classes of adopters, from early adopters through laggards, as imitators. Most of

the extant literature concerned with the adoption of innovation has focused on identifying and

understanding consumers who exhibit innovative behaviour or the early adoption behaviour of

new products (Adner and Kapoor, 2010; Laciana, Rovere, & Podesta, 2013; Semadeni and

Anderson, 2010; Mahajan, Muller, & Peres, 2010), while other categories of adopters, mainly

imitators, have been rarely discussed (Singh, 2006). Singh (2006) explains that this oversight is

due to the difficulty of identifying the different adopter categories. Bass (2004) defines imitators

as adopters whose decisions concerning purchasing new products are influenced by other members

of their social systems. Unlike innovators, imitators are influenced by or “learn” from previous adopters (Bass, 2004). Singh (2006) notes that imitators are influenced by pressures in a social system, which come from external sources. When individuals seek information on new products, they will exhibit different levels of reliance on communication channels such as mass media (Singh, 2006).

Bass, Mahajan, and Muller (1990) reviewed the Bass diffusion model and mentioned that only

potential imitators are influenced by word-of-mouth, as favourable user experiences of innovative

products can motivate the spreading of positive information, thereby stimulating adoption among

later adopters (Bass et al., 1990; Gatignon & Robertson, 1986; Rogers, 1983). Kawakami, Kishiya,

and Parry (2003) distinguished between personal word-of-mouth (pWOM), which refers to the

sharing of information between individuals who know each other, and virtual word-of-mouth

(vWOM), which refers to communicating with individuals one has never met in real life

(Kawakami et al., 2003). The authors indicate that pWOM helps to stimulate product adoption and

product use, as innovation users share experiences with their families, friends, and acquaintances

(Kawakami et al., 2003). Virtual WOM helps to improve consumers’ perceptions of product value,

as consumers tend to reply on online reviews, online information, and other vWOM sources when

evaluating innovative products. (Kawakami et al., 2003). Therefore, this study adopts these views

and use two variables to measure consumer imitativeness.

The Formulation of Measures

Given that the objective of this study is to forecast the demand for HARVCs with a new feature using the Bass diffusion model, the following two research questions are formulated:

What is the coefficient of innovator?

What is the coefficient of imitator?

What is the number of market share?

First, the factors that measure innovativeness and imitativeness have been discussed in previous sections: motivational dimensions are used to measure innovativeness. People with higher levels of agreement on different motivational dimensions are more likely to be innovative. This study measures four motivational dimensions that are indirectly explained by different statements; those statements make it possible to identify the innovativeness of an individual. Similarly, the WOM effect has an influence on imitativeness; thus, consumers who are affected by social influences tend to be imitative. This study measures two types of WOM; it allowed to measure the coefficient of imitation through various statements which indirectly explain consumer imitativeness.

Furthermore, it is known that the values of p and q in the Bass model should be between 0.00 and

1.00 (Griffin, Hauser & Tellis, 2006). Table B1 (Appendix B) summarizes the original items

(statements) and their sources. More detailed information about the measurement method can be

found in Chapter 3.

Since this study uses several statements to measure the coefficients of p and q, it is necessary to review and compare the values for these parameters that have been found or tested in previous studies. First, the bass diffusion model has been widely used to forecast the sales of electric vehicles. Kim et al. (2013) applied a forecasting model to hydrogen fuel cell vehicles in Korea;

they measured the estimated factors by incorporating results from a comparison of the values of these parameters for other Pacific Rim countries. They found that the purchase probability of innovators was 0.0037, while the purchase probability of imitators was 0.3454 (Kim et al., 2013).

Cherchi, Jensen, Mabit, and Ortuzar (2017) estimated parameters and implemented the model for

Danish electric cars by using the number of newly registered electric cars in Norway. They found

the value of innovators to be 0.002 and that of imitators to be 0.23. However, there is a critical

issue in that the difference in the diffusion of a product in one country's market can lead to a major

difference in supply and demand in that country when compared to others. This can result in the

transfer of parameters being invalid (Gohs and Massiani, 2015). Finally, Lamberson (2008)

conducted research on sales data in the USA concerning hybrid electric vehicles; he used these

sales data to calibrate the parameters of innovators, which he found to be 0.000618, and the

parameter of imitators, with was found to be 0.8736 for the same market. In addition to electric

vehicle products, estimates of parameters for products related to the household industry can also

be compared. Trkman and Turk (2012) applied the Bass model to broadband products across

European OECD member countries; data were collected from households in Slovenia and the

OECD database. The authors’ estimate for the coefficient of innovation p was 0.056, while that of

imitation q was 0.5666 on average. Islam (2014) measured the adoption probabilities of renewable

solar photovoltaic panels in Canada based on discrete choice experiments and Bass model

diffusion. She found that diffusion is predominantly driven by the coefficient of innovation, which was 0.0914, while the coefficient of imitation was only 0.1036. Table 1 presents a summary of the main results concerning the Bass coefficients that were found in the literature.

Table 1: Estimates from the literature for the parameters of p and q

References Method Innovation

coefficient p

Imitation coefficient q Kim et al. (2013) Hydrogen fuel cell vehicles sales data

for the Korean market

0.0037 0.3454 Cherchi et al. (2017) Registered electric car data from

Norway the Danish market 0.002 0.23

Lamberson (2008) Hybrid electric vehicles sales data in the USA

0.000618 0.8736 Trkman, & Turk (2012) Broadband products, based on a

survey in Slovenia and the OECD database

0.056 0.5666

Islam (2014) A discrete choice experiment concerning solar photovoltaic panels

in Canada

0.0914 0.1036

Potential Market Share

The potential market is defined as the cumulative market potential of a product’s life cycle (Bass, 2004). In forecasting the adoption of HARVCs, it is important to estimate the relative values of innovator and imitator, as well as the size of the market (Bass, 2004). The Bass diffusion model encounters an issue when considering the total cumulative sales for innovative products, as it is difficult to determine for how many years sales should be computed. It is risky to apply the Bass diffusion model to products with a long lifetime horizon, as it assumes a declining phase in sales.

To solve this problem, Massiani (2013) developed the stated preference method, which, in this

study, is referred to as choice-based conjoint analysis (CBC analysis). He explains that CBC

analysis is a survey method where preferences are analysed based on hypothetical attribute combinations; in other words, consumers are asked to evaluate certain attributes, such as price and brand, of various alternatives (Massiani, 2013). By investigating consumers’ preferences, researchers can gather information about their preferences with regard to price and other attributes;

this makes it possible to compute customers' probability of purchasing a product, their willingness to pay for it and the market share for a particular product (Massiani, 2013). Moreover, CBC analysis has an advantage in that it measures the actions that consumers have actually undertaken based on their preferences; it also offers a number of advantages in terms of forecasting product diffusion. First, it provides information about non-monetary attributes (Massiani, 2013), which is helpful in understanding how those attributes influence consumers’ choices regarding different products (Massiani, 2013). Second, the results of CBC analysis can assist in understanding the effects of certain attributes (Massiani, 2013); in this study, it is particularly important to analyse the influence of the feature that has been added to HARVC products. Third, CBC analysis considers both alternative attributes of a product and competitors (Massiani, 2013). Based on those advantages, this study measures the market potential of HARVCs in China using a CBC analysis.

The Conceptual Model

Figure 1 presents the structure of the Bass diffusion model for HARVCs and the three parameters

of innovator, imitator, and potential market. It also depicts the factors that are used to identify the

values of these three factors. Given that the addition of demographic variables allows for

differences in terms of gender, age, level of education, and level of income to be compared, these

variables are included.

Figure 1: The conceptual model

In addition, when conducting a CBC analysis to determine the scope of the potential market,

decisions concerning which HARVC attributes are considered should be also discussed. Brand and

price are two basic but very important features when making a purchasing decision concerning a

smart device (Djatmiko, & Pradana, 2016). Djatmiko and Pradana (2016) found that with an

increase in the brand image or price of Samsung smart phones, the purchase decision value also

increases. Chang and Wildt (1994) found that perceived price has a significant but negative impact

on purchase intention. Therefore, this research will include brand and price as two attributes to

determine whether they are important utilities for HARVCs. In addition, mobile WIFI monitor is

a basic feature of a robotic vacuum cleaner (Nadie, 2018); it allows one to remotely control such

a machine using an app. With this function, consumers can open, close, switch the cleaning mode

of, recharge, and monitor an HARVC remotely. Furthermore, as mentioned previously, automatic

dirt disposal is the newest feature on HARVCs that have become available on the market; a robot

vacuum cleaner with this feature can automatically empty its bin into an enclosed disposable bag,

which makes cleaning more convenient. iRobot is the first brand to offer an HARVC with this

feature on the marketplace (irobot.com); however, this product is not currently available on the

Chinese market. With ongoing technological development, it is possible that people may come to

perceive this feature as valuable, leading to other HARVC brands adopting it. More detailed

information about attributes and levels can be found in the Chapter 3, and Table B2 in Appendix

B explains all attributes and levels.

3. METHODOLOGY Sample and Data

To measure the values of innovator, imitator, and the potential market, a structured questionnaire, which can be found in Appendix A, was designed. This questionnaire used the non-probability sampling method, as it is difficult to gather a large random sample within a limited period of time.

Moreover, as this study is not limited to certain types of participant, snowball sampling was adopted. The survey was mainly distributed online by sending it to a small group of participants on Chinese social network platforms such as WeChat and Weibo. Thereafter, the selected participants were requested to forward the online survey to others to generate a referral effect.

The procedure of the questionnaire was to first explain the purpose of the study to the participants;

thus, they were well-informed at the beginning of the survey. Thereafter, participants were asked to provide basic demographic information in the first part of survey, including gender, age, level of education, and income. The participants subsequently moved to the second section. It consisted of several multiple-item questions which indirectly measured innovativeness and imitativeness.

The participants were required to share their opinions on the given statements based on their

behaviours. The last section of the survey focused on a CBC analysis. The participants needed to

choose their preferred products from several options. To further elaborate on the manner in which

the survey was completed, all the questions were set as forced response, which means that the

participants had to completely answer all of them. The following subchapter presents the specific

measures for these three sections.

Measures

In the first part of the survey, questions concerning age, income, and education level were designed as single choice. Therefore, every option for each variable had a selection range; this range was determined with reference to Tan, Zeng, and Zhu (2017) and Chinapower.csis.org. All variables were coded using the dummy coding technique. For instance, for the question concerning gender, 1 was coded as male, and 2 was coded as female.

In its second section, this survey featured a seven-point Likert-type scale, ranging from 1 (strongly disagree) to 7 (strongly agree). All factors, which included four motivational dimensions and two types of WOM, were indirectly assessed through a set of items (statements) which have been tested.

Specifically, the items for measuring innovativeness were adopted from Geuens and Vandecasteele (2010), and those for measuring imitativeness were adopted from Kawakami et al.

(2013). They measured pWOM and vWOM for “portable game players”; given the context, this study changed this to “HARVCs” and “an innovative product”. Moreover, the items were translated from English to Chinese; to ensure that the meaning of the text was consistent, the survey was pre-tested by 10 people who have knowledge of both English and Chinese.

The third section focused on the CBC analysis. First, this study chose brand, price, mobile WIFI

monitor, and automatic dirt disposal as attributes, the combination of those attributes could

represent HARVCs. Moreover, each attribute had two to three levels; brand had three levels,

namely iRobot, Ecovacs, and Xiaomi, as these are the three best-selling brands in the Chinese

market (Chinese industry research institute, 2018). Prices also had three levels, 1499, 3499, and

6499; those prices are in Chinese yuan. The selection of prices was based on the average prices of Ecovacs and Xiaomi products and the selling price of the iRobot with the automatic dirt disposal function. In addition, both mobile WIFI monitor and automatic dirt disposal had two levels, which were with and without each feature. Furthermore, there was a no-choice option in the survey;

participants could choose this option if they did not prefer any of the given options.

The experiment followed a fractional factorial design, which means that it featured a subset of the all possible attribute-level combinations and used cue cards and images as forms of presentation to motivate consumers to complete all of the choice sets. Since this survey included four attributes, the complexity of the options could cause a fatigue effect; as such, the number of alternatives per choice set was limited to three, including a no-choice option, and only nine choice sets, which were presented in a balanced and orthogonal manner, were available.

Data analytic strategy

This section introduces the formula for analysing the Bass diffusion model and the methods used to calculate the parameters of innovator, imitator, and potential market. To gather these data, this study used the R programming language to collect the coefficients of innovation and imitation; in addition, the survey website was used to measure market share.

The formulation of the Bass diffusion model

The Bass diffusion model is used to forecast the market potential for innovative propulsion

products, and it is often applied in situations in which diffusion plays an essential role (Gohs &

Massiani, 2015). This research deals with products that will be soon launched in the Chinese market; therefore, the sales of new products coincide with the number of initial purchases during the beginning of period T, which also means that there are no sales have been made before (Bass, 2004). Following this context, the likelihood of product purchases at time T can be written as (Gohs

& Massiani, 2015) follows:

Moreover, the number of sales S(T) at time T can be expressed as follow:

Where, 𝑃(𝑇) indicates the probability of an initial adoption at time T, assuming that the new product has not been adopted previously; 𝑓(𝑇) indicates the likelihood of purchase at time T; 𝐹(𝑇) indicates the cumulative distribution function of adoption at time T; p indicates the coefficient of innovation; p indicates the coefficient of innovation; q indicates the coefficient of imitation; m indicates the total number of purchases over the period interest; and 𝛾(𝑇) indicates the total number of purchases in the (0, 𝑇) interval, 𝛾(𝑇)= 𝑚𝐹(𝑇).

Calculation of parameters p and q

This study reports the Bass parameter estimates based on the survey data. Based on Chapter 2, four

motivational dimensions were used to measure innovativeness, and two types of WOM were used

to measure imitativeness. Each variable was indirectly described by two statements; participants

chose an answer ranging from 1 (strongly disagree) to 7 (strongly agree) based on how accurately the statements described them. Those scales together can help to understand the extent of consumers’ innovativeness and imitativeness. Moreover, from previous literatures, the coefficient of innovation p was ranged from 0.000618 to 0.0914, and the range of the coefficient of imitation was between 0.1036 and 0.8736. These values provide guidance for this research, as the coefficients of innovation and imitation should be within these ranges. Therefore, there is a need to rescale the values of innovativeness and imitativeness obtained from the multi-item scale into an appropriate form. Then, by calculating the mean of those re-scaled scores, a number indicating each individual's innovativeness or imitativeness (the individual coefficient of p or q) could be obtained. For example, innovativeness was measured by four variables, and each variable included two statements, therefore, in total eight statements together can measure the innovator coefficient by calculating the mean of eight re-scaled scores. The average of the coefficients of p or q for all participants are calculated as the sum of the individual innovator coefficients or imitator coefficients divided by the number of respondents.

Below, the method of analogies is presented:

Where, i indicates the ith participant; the p indicates the parameter of innovation; a indicates the

rescaled weighted coefficient of individual i given by the functional motivational dimension; b

indicates the rescaled weighted coefficient of individual i given by the social motivational

dimension; c indicates the rescaled weighted coefficient of individual i given by the hedonic

motivational dimension; d indicates the rescaled weighted coefficient of individual i given by the

cognitive motivational dimension; N indicates the number of participants, which is 250 participants in this research; and 𝛴

indicates that the sum of the coefficient of innovation assigned to each participant; the results give a total innovation value for the 250 participants.

As with the calculation of the coefficient of p, the coefficient q is calculated as follows:

Where, i indicates the ith participant; q indicates the parameter of imitation; e indicates the rescaled weighted coefficient of individual i given by pWOM; f indicates the rescaled weighted coefficient of individual i given by vWOM; N indicates the number of participants, which is 250 participants in this research; and 𝛴

indicates that the sum of the coefficient of imitation assigned to each participant; the results give a total imitation value for the 250 participants.

The formulation of the parameter m

In order to forecast the development of HARVCs with the new feature of automatic dirt disposal, the potential market should be measured. Since this feature is still new to the market, there is no historical data concerning how many HARVCs with this feature have been sold in the current market. Thus, this study conducted a CBC analysis and its result applies the multinomial logit model (MNL model) to determine the value of m. The MNL model predicts the probability that an individual will choose the alternative i from a choice set J (Mcfadden, 1974).

This study used four options in choice set J, one of which was the combination of levels which

have the highest utilities for “brand”, “price” and “mobile WIFI monitor” attributes and with

automatic dirt disposal function. By calculating the utility and the probability of this most preferable product, the market share of this product can be predicted. The random utilities can be expressed as follows:

Uij = aj +

𝛽

Where, j indicates the jth choice; i indicates the ith participant; U indicates the utility of each level;

a indicates the constant estimation coefficient; x indicates the product attribute vector;

𝛽

the estimation coefficient vector; ɛ indicates the error term or the random part of utility, which is a probability variable; and utility function follows a S-curve. The MNL model assumes the independence assumption of the error term and it distributes with a cumulative distribution function (Bijmolt, Leeflang, Pauwels, Wieringa, 2015).

The probability of that an individual will choose the alternative i is written as follows:

Where, i indicates an alternative choice; J indicates the choice set, 𝐽 = {1, . . . , 𝑖, . . . , 𝑚}; U indicates

the systematic or rational utility component; and m indicates the number of options within a choice

set. The formula is explained as the probability of choosing option i from a set of J options equals

the systematical utility of option i divided by the sum of the systematical function of the options

that are within the choice set (Mcfadden, 1974). The probability of choosing option i is between 0

to 1.

4. RESULTS

The results of this study are analysed in the following four sections. The first section described the characteristics of samples, while the second and third section tested the coefficients of innovation

coefficient was first determined for validation, as the variables used to measure innovativeness and imitativeness are consist of multiple items. Therefore, pre-testing those items is necessary to determine whether they can explain one or more variables. The fourth section described the results of the Bass diffusion model, which is used to predict the diffusion of HARVCs. Furthermore, since the data also include irrelevant information, such as starting time, ending time, and IP addresses, it was decided to delete this information before performing further analysis.

Sample Characteristics

After the two-week data collection period, the total number of participants was 722; however, due to the complexity of the survey, 472 participants were dropped out in the middle of the questionnaire. Ultimately, only 250 reliable and valid questionnaires were obtained and used for data analysis. Table C1, which can be found in Appendix C, presents information on the demographic variables of the 250 participants. The results showed that 56.8% of the participants were female, while 43.2% were male. 5.2% of the participants were under 18 years old, 43.6%

were between 18 and 30 years old, 30.4% were between 31 to 45 years old, 20.4% were between

46 and 56 years old, and only 0.4% were older than 65 years of age. The majority of the participants

(around 55.2%) had attained a bachelor’s degree or a higher qualification. Furthermore, 54.4% of

the participants earned more than 4,001 Chinese yuan every month. It is interesting to note that the

income distributions differ between the male and female participants (Appendix C, Figure C1).

The income distribution for male tended to be a normal distribution, while the income distribution for female tended to be a right-skewed distribution. This indicates that the men were more concentrated in the middle-income group, while the women were more concentrated in the lower- to middle-income groups. This difference may influence the results of the conjoint analysis, because one of the attributes in conjoint analysis is price, men and women may show different perceptions with regard to pricing. In the following sections, estimations of the values of innovator coefficient, imitator coefficient, and the coefficient of market share are made, taking gender and other demographic variables into account.

The measures of innovation p and imitation q

Since the variables were indirectly measured by different statements, the reliability test was carried

out first when evaluating the responses to the multiple-item questions. Internal consistency can be

tested by using Cronbach’s alpha. Table 2 indicates that the Cronbach’s alphas are acceptable for

the social motivational dimension (𝛼 = 0.79), the hedonic motivational dimension (𝛼 = 0.74), and

the cognitive motivational dimension (𝛼 = 0.73). However, the Cronbach’s alphas are poor for the

functional motivational dimension (𝛼 = 0.44), pWOM (𝛼 = 0.53) and vWOM (𝛼 = 0.53). Possible

reasons for low Cronbach’s alphas include a limited number of questions or poorly interrelated

items. The low Cronbach’s alpha values were also confirmed by the correlation coefficients. Since

each variable contains two items, the correlation coefficient can be used to test the relationship

between these two items. Table 2 and Figure 2 present the results of the correlation test for the

innovativeness and imitativeness items. The relationships between two items that measure one

variable are found to be significant (p < 0.05). In particular, the items used to measure the social

motivational dimension (r = 0.66), the hedonic motivation dimension (r = 0.59), and the cognitive

motivational dimension (r = 0.57) are strongly and positively related to each other; however, the items measuring the functional motivational dimension are weakly and positively associated with each other (r = 0.29). Furthermore, the items that measure pWOM (r = 0.49) and vWOM (r = 0.37) also exhibit weak positive relationships. Since the results show significant relationships among these items, it was decided to use all of them in the subsequent calculation of the coefficients of innovativeness and imitativeness.

Table 2: The correlation coefficients of the items investigated in this study Variables Items codes Cronbach’s

alpha

r Innovativeness

Social 1 and 2 0.79 0.66*

Functional 3 and 4 0.44 0.29*

Hedonic 5 and 6 0.74 0.59*

Cognitive 7 and 8 0.73 0.57*

Imitativeness

pWOM 1 and 2 0.53 0.49*

vWOM 3 and 4 0.53 0.37*

Note: * p< 0.05

Figure 2: The correlation coefficients and p-values obtained

A further test on the correlation coefficients was conducted to ensure that the variables are correlated with each other and that, together, they can be used to measure consumer innovativeness and imitativeness. Table 3 and Figure 3 present the results of the correlation test for the innovativeness and imitativeness variables. The results show that all variables are significantly related to each other (p < 0.05). Specifically, the correlations between the social and function variables (r = 0.51), the social and hedonic variables (r = 0.54), the social and cognitive variables (r = 0.54), the functional and hedonic variables (r = 0.62), the functional and cognitive variables (r = 0.57), and the hedonic and cognitive variables (r = 0.57) are strong and positive. Only the correlations between pWOM and vWOM are moderate and positive, with r = 0.37. However, since all of the relationships are significant and positive, together, these variables can therefore measure customer innovativeness and imitativeness.

Table 3: The correlation coefficients of variables investigated

Variables r

Innovativeness

Social and functional 0.51***

Social and hedonic 0.54***

Social and cognitive 0.54***

Functional and hedonic 0.62***

Functional and cognitive 0.57***

Hedonic and cognitive 0.57***

Imitativeness

pWOM and vWOM 0.37***

Note: *** p< 0.001

Figure 3: The results of the analysis of the correlation coefficients and p-values

Having examined the correlation coefficients among the items and variables, the coefficients of p

and q can be determined based on the formulas presented in Chapter 3. The item scales were first

rescaled using values for p and q obtained from previous literature (Table 4). Thereafter, the

average individual innovator coefficient or imitator coefficient was calculated using the means of

the total re-scaled scores given to the multiple-item questions by the participants. The average

imitator or innovator coefficient for the 250 participants was calculated by taking the sums of the

individual innovator coefficients (or imitator coefficients) and dividing them by the number of

participants, which is 250. These calculations yielded an average coefficient for innovator p of

0.02208037 and an average coefficient for imitator q of 0.1343183. A more detailed discussion of

the calculation procedure can be found in the R-code in Appendix F.

Table 4: The rescaled item scales

Variables Original item scales Re-scaled scores

Innovativeness 1 0.000618

2 0.01574833

3 0.03087866

4 0.04600899

5 0.06113932

6 0.07626965

7 0.0914

Imitativeness 1 0.1036

2 0.2948833

3 0.4861666

4 0.6774499

5 0.8687332

6 1.060016

7 0.8736

The estimation of market share m

Market share is estimated using the survey website tool developed by Sawtooth Software. First, the importance of every attribute was segmented by gender (Table 5, Figure 4). Overall brand and price were considered most important by both male and female participants. Men perceived brand as the most important attribute (Importance

= 33.44%, p < 0.05; Importance

= 31.74%, p

< 0.05), whereas women perceived price as most important (Importance

= 31.56%, p < 0.05;

Importance

= 33.49%, p < 0.05). In addition, the importance of a mobile WIFI monitor

(Importance

= 17.67%, p < 0.05; Importance

= 17.58%, p < 0.05) is similar to the

importance of the automatic dirt disposal feature for both men and women (Importance

=

17.15%, p < 0.05; Importance

= 17.38%, p < 0.05). Although men and women value these

attributes differently, these differences are relatively minor. Hence, this study does not segment

the product functions and purchasing probabilities by gender. More segments were created

according to other demographic variables (Appendix D); the results show that the importance of

certain attributes for different segment groups is similar. Therefore, this study does not segment the sample.

Table 5: The importance of attributes by gender

Attribute Importance SD Lower 95%

CI

Upper 95%

CI Male

Brand 33.44% 13.5 30.89 35.98

Price 31.74% 13.3 29.23 34.25

Mobile WIFI monitor 17.67% 10.65 15.66 19.68

Automatic dirt disposal 17.15% 10.87 15.10 19.20

Female

Brand 31.56% 14.83 29.12 34.00

Price 33.49% 13.85 31.21 35.76

Mobile WIFI monitor 17.58% 10.36 15.88 19.28

Automatic dirt disposal 17.38% 11.72 15.45 19.30

Figure 4: The importance of attributes based on gender

To calculate the probability of purchasing any set of choices, the utility of every level was first

calculated. Table 6 indicates that, compared to the no-option choice ( 𝛽 = -41.25, p < 0.05),

participants are more likely to choose one of the HARVC products. More specifically, the

participants preferred Xiaomi brand ( 𝛽 = 24.85, p < 0.05) to the brand of iRobot ( 𝛽 = -19.01, p <

0.05) and Ecovacs ( 𝛽 = -5.85, p > 0.05). iRobot was the least preferred brand, as the estimation of its utility is significant and negative. Unsurprisingly, the participants preferred an HARVC with a price of 1,499 Chinese yuan ( 𝛽 = 33.55, p < 0.05) to one with a price of 3,499 ( 𝛽 = -5.77, p > 0.05) or 6,499 Chinese yuan ( 𝛽 = -27.78, p < 0.05), while the HARVC with a price of 6,499 Chinese yuan was least preferred. Furthermore, compared to an HARVC without a mobile WIFI monitor ( 𝛽 = -14.83, p < 0.05), participants preferred the product with this feature ( 𝛽 = 14.83, p < 0.05).

Similar results were found for the automatic dirt disposal feature: Participants tended to prefer a product with this feature ( 𝛽 = 17.14, p < 0.05) to one without ( 𝛽 = -17.14, p < 0.05).

Table 6: The utilities of the various levels

Levels Utilities SD Lower 95%

CI

Upper 95%

CI Brand

iRobot -19.01 58.58 -26.27 -11.75

Xiaomi 24.85 59.21 17.51 32.19

Ecovacs -5.85 55.06 -12.67 0.98

Price

1,499 33.55 56.76 26.52 40.59

3,499 -5.77 53.37 -12.39 0.84

6,499 -27.78 56.37 -34.77 -20.79

Mobile WIFI monitor

Yes 14.83 38.26 10.08 19.57

No -14.83 38.26 -19.57 -10.08

Automatic dirt disposal

Yes 17.14 37.66 12.47 21.81

No -17.14 37.66 -21.81 -12.47

No option -41.25 207.38 -66.96 -15.54

After determining the values of the utilities, several image sets of choices were made to determine the probability of purchasing an HARVC with or without the automatic dirt disposal feature. The alternatives were chosen based on the utilities. The first alternative includes the listed iRobot product, called the iRobot i7 Plus, with a price of 6,499 Chinese yuan, with a mobile WIFI monitor and the automatic dirt disposal feature. The second alternative includes all of the levels with the least preferable utilities. The third alternative only includes the no-option choice, while the last alternative contains the levels with the highest utilities. The more preferable product referred to as Xiaomi New includes the levels Xiaomi brand and has the mobile WIFI monitor and automatic dirt disposal features, with a price of 1,499 Chinese yuan. Figure 5 illustrates the results of the probability calculations for the four alternatives. The iRobot i7 Plus has a similar purchasing probability (prob = 14.7) to that of the HARVC product with the least preferable utilities (prob = 14.6), as well as that of the no-option choice (prob = 14.5). However, the HARVC product with the most preferable levels (Xiaomi New) exhibits the highest probability, which reaches 56.3%.

Furthermore, it is also interesting to compare the purchase probability for Xiaomi New with and without the automatic dirt disposal feature. To make a comparison, a new product, Xiaomi Old, is created and consists the levels that were remained the same for Xiaomi New, only the level of

“automatic dirt disposal feature” is set to “without”, the purchase possibility decreases

approximately 20% (Figure E1, Appendix E). This indicates the importance of having automatic

dirt disposal as a new feature, as the purchase probability will significantly increase should this

feature be included.

Figure 5: The purchasing probabilities for the four alternatives

Since the probabilities of purchasing the Xiaomi New (56.3%) and the iRobot i7 Plus (14.7%) are known, the total market capacity can be estimated. The potential customers of these HARVCs will be households in China, and the average number of households is approximately 374 million over the past five years (stats.gov.cn). Hence, the market capacity for Xiaomi New will be 211 million, while that of the iRobot i7 Plus will be 55 million in total. Table 7 presents the result of all coefficients in the Bass diffusion model.

Table 7: The coefficients in the Bass diffusion model

Parameters p q m

Coefficients 0.02208037 0.1343183

Xiaomi New 211,000,000*

iRobot i7 Plus 55,000,000*

Note: * p< 0.05

Diffusion pattern of HARVCs

Once the parameters are known, the predicted sales can be calculated. To estimate the diffusion

patterns for two products, Xiaomi New and the iRobot i7 Plus, it is necessary to calculate their

sales and cumulative sales. First, the expected sales and cumulative sales are calculated based on the parameters identified in previous sections and the formula of the Bass diffusion model, which was presented in Chapter 3. The initial market is set to zero, since the new product does not have any historical sales. Figure 6 presents the sales of Xiaomi New and iRobot i7 Plus until saturation point is reached. The predicted sales of Xiaomi New are significantly larger than that of the iRobot i7 Plus; in addition, the sales curve of Xiaomi New shows a steep slope, while that of the iRobot i7 Plus shows a stable slope. Moreover, both curves show right-skewed distributions; this indicates that the mean is to the right of the median, so for both brands, sales increase rapidly at the beginning and then become steady after the peak. Xiaomi New sales start at 4.66 million in 2019, reach a peak of 9.61 million in 2031, and finally shows a long tail in the positive direction until 2079, with sales of 0.06 million. Sales of iRobot i7 Plus start at 1.21 million and reach a peak of 2.5 million sales in 2031. The final sales made in 2072 reach 0.05 million.

Figure 6: The sales curves for the Xiaomi New and the iRobot i7 Plus

The cumulative sales for Xiaomi New and iRobot i7 Plus are compared in Figure 7. As with the results presented in Figure 6, the cumulative sales for these two products are also dramatically different, while the distributions of the curves are similar. The slopes of the two curves both start from zero and increase steeply until 2040, at which point the growth rate becomes slow and stable.

The cumulative sales curve of Xiaomi New reaches its peak at around 2040 with 170.54 million;

the cumulative sales curve of iRobot i7 Plus also reaches its peak at this point, but only with 44.45 million. The Xiaomi New will require 61 years to reach saturation point and complete its distribution pattern, while, for the iRobot i7 Plus, the diffusion is predicted to be complete within 54 years.

Figure 7: The cumulative sales curves for the Xiaomi New and the iRobot i7 Plus

5. DISCUSSION Implications

This research identified certain theoretical implications with regard to the study of the adoption and diffusion of new products. First, the diffusion patterns can vary for different products. Kim et al. (2013) predicted the diffusion pattern for an electric vehicle in the Korean market; the sales curve that they forecasted showed a very slow penetration at the beginning, and follow with a rapid growth in a later stage, however, the sales curves in this research rapidly increases at the beginning, and then slowly decreases after the peak. One of possible reasons for different sales curves might be electric vehicles are a “high learning product”, consumers need time to make individual behavioural adjustments. Once consumers realize that electric vehicles represent a useful alternative to the conventional vehicle, these products start obtaining a greater market share in later stages. However, the sales curves show in the present study shows a rapid market penetration at the beginning; the reason behind this may be that Chinese consumers are already familiar with HARVC products, they need less time to get used to this type of product. Should a new HARVC product with more useful options become available on the market, Chinese consumers would be more willing to consider it as a supplement to traditional HARVCs.

Moreover, a product's diffusion pattern is significantly influenced by that product's utilities. This study developed two diffusion curves—one for the Xiaomi New and one for the iRobot i7 Plus.

While the distributions of these two curves are both right-skewed, the sales volumes of these two

products differ dramatically. The possible reason for this is that Xiaomi New includes those

utilities that participants rated most desirable, while the iRobot i7 Plus has the fewest of these

utilities, such as brand and price; the availability of these utilities can influence the possibility of

purchase as well as predicted market share, and these differences will finically influence the sales volumes.

In addition to theoretical implications, this study also identifies various practical implications for the marketing field. First, this research determined that the importance of attributes among Chinese participants did not differ significantly based on gender, age, level of education, and income level.

Participants from different gender and age groups, and of varying levels of education and income generally place more emphasis on brand and price, as the analysis indicated that the importance of these two attributes reached approximately 60% in total. Therefore, when considering purchasing an HARVC product, consumers generally tend to make a decision based on brand and price. These results may prove valuable to marketers when attempting to develop strategies intended to enhance brand awareness and penetration. Once a particular brand of HARVC is familiar to consumers, products from that brand are more likely to be chosen by consumers when compared to other products with similar features.

The high utility of price also indicates that consumers’ purchase behavior is influenced by the price of an HARVC. The results presented in Chapter 4 indicate that 6,499 Chinese yuan has the lowest utility and 1,499 Chinese yuan has the highest utility; thus, Chinese consumers are price sensitive.

An increase in price may significantly decrease the demand for an HARVC. It is essential for

marketers to make correct pricing decisions based on their target customers. Moreover, the

importance of pricing varies relative to other purchasing criteria; in this research, branding was

found to be relatively more important than pricing among male participants than female

participants. This indicates that branding is considered to be more important than pricing, which makes male consumers less sensitive to pricing when making purchasing decisions.

In addition, the utilities of the automatic dirt disposal and mobile WIFI monitor features were also found to be significant and positive. These findings could help marketers and product designers to better understand perspectives on HARVCs in the Chinese market. Although brand and price are the two most important attributes, when HARVCs offer more useful options, they are likely to become more appealing to consumers, and the purchase probability will also increase. Therefore, it is important that product designers oversee the development of the products and understand customer needs and that marketers focus on designing effective marketing strategies intended to target customers and stimulate their purchasing sentiment by promoting these features.

Limitations

This research failed to consider the validity of the coefficient ranges of p and q. In Chapter 2, the

rescaled p and q coefficients were measured using coefficients for electric vehicles drawn from

previous literature. Since this type of product distinct from household products, consumers may

behave differently on two types of products, the coefficients of innovation and imitation may

therefore differ substantially. Cherchi et al. (2017) also suggested that the distribution of car and

household products can differ; for example, the market share of microwave increased to 80% after

consumers discovered that microwave is a supplement to heat food. However, unlike microwaves,

it is difficult to consider an electric car as a supplement to a conventional car, as the market

penetration in Denmark is only 60%. Therefore, based on the authors' suggestion, different types

of products could have varying degrees of market penetration. If the coefficients of p and q for

household products--in the context of this study, HARVCs--are measured with reference to the coefficients of electric cars, any findings obtained could be invalid.

Another serious problem is that this research only used two items to measure variables that can measure the coefficients of p and q. A low number of items could lead to poor interrelatedness among items (Malhotra, 2009). As noted in Chapter 4, the internal consistencies of the functional motivational dimensions, pWOM and vWOM are considered to be poor. This problem has a direct impact on the results of the measurement of the p and q coefficients, as these coefficients were calculated with reference to the averages of the rescaled item scales. When the items used are not be trustworthy, the measures can be criticized, and the results may be unreliable.

Furthermore, the ultimate market share m, which affects the prediction of the diffusion pattern of

HARVCs, is highly influenced by the total ultimate future market, therefore, if the total ultimate

market is inaccurately predicted, the market share for a specific product will also be wrongly

forecasted, and the diffusion pattern of that product will be invalidated. For example, if the time

taken to reach saturation is longer or shorter than predicted, the forecasted diffusion pattern will

give the wrong information to marketers when they develop strategies in different stages of product

life cycle. For instance, the point at which market saturation for electric vehicles was estimated to

be reached in the United States was in 2007; however, it was found out that more than 2 million

of electric vehicles were sold between 2010 and 2015 (Lamberson, 2008, as cited in Cherchi et al.,

2017). Possible reasons for an under- or overestimation of market saturation may be improvements

to a product's characteristics or changes in economic or political incentives; these changes may

impact the demand for HARVCs and therefore extend or shorten the product life cycle (Cherchi et al., 2017).

Future Directions

This study only applied a simplified version of the Bass diffusion model to analyse HARVCs in

China; a more complex model that includes other parameters, such as price, competitors could be

used to obtain more accurate results. Moreover, this research focused on forecasting the diffusion

of HARVCs with an automatic dirt disposal feature; however, it would also be interesting to

analyse the effect(s) of adding other attributes on the penetration of HARVCs in general, as doing

so can cause a change in consumer preferences. Because Cherchi et al. in 2017 indicated that

changes in individual preferences can have an effect on the predicted market share, as well as the

diffusion pattern, of a product.

6. CONCLUSION

This research provides empirical evidences concerning the distribution pattern of HARVCs with an automatic dirt disposal feature in the Chinese market. These findings provide useful insights in terms of understanding the innovativeness and imitativeness of Chinese customers, as well as their preferences with regard to HARVCs and the likelihood of purchasing such products. It was suggested that future research could involve developing a more robust model for analysing the Chinese HARVC market or the development of HARVCs under various scenarios.

7. ACKNOWLEDGEMENTS

Gratefully thank Dr. K. (Keyvan) Dehmamy for his useful ideas and constructive comments

regarding computational assistance and an earlier version of the manuscript. And I would like to

express my appreciation to Dr. F. (Felix) Eggers for spending the time to evaluate my work and

act as the second supervisor.