A Path To The Autonomous Driving, Begins with Chinese Consumers.

A Path To The Autonomous Driving, Begins

with Chinese Consumers.

Combining conjoint analysis and bass model to

investigate the future of autonomous vehicles in

China.

by Tiantian Wang

University of Groningen

Faculty of Economics and Business MSc Marketing Intelligence PO Box 800, 9700 AV Groningen (NL)

First Surpervisor: Keyvan Dehmamy Second Supervisor: Felix Eggers

17th June 2019

Management Summary

In China, with the quick financial development, the automobile industry is going through a blasting growth period as well as self-driving technology. China is expected to overtake the U.S. to turn into the world's biggest autonomous vehicle market by the time of 2020. China's "Twelfth Five-Year Plan" outline clearly proposes to focus on the development of new energy vehicles, including developing autonomous driving cars. Compared with traditional vehicles, autonomous vehicles have the advantages of better controllability, less potential human accidents and higher efficiency in solving traffic congestion. In addition, the large-scale development of autonomous vehicles has a positive effect on balancing the peak-to-valley load of the power grid and increasing the employment rate. Therefore, accurately predicting the quantity of autonomous vehicles has important practical significance for realizing the orderly development of autonomous vehicles and guiding the rational layout of upstream and downstream for firms to enter Chinese market.

This paper aims to build and develop suitable marketing strategy for firms who wants launch autonomous vehicles in China, by reviewing Chinese consumer behaviour and forecasting potential sales. By detecting consumer’s trade-off thinking process, this research is able to find out that Chinese consumers are indeed less responsive to price promotions, aligned with literature. On the contrary, Chinese consumers value the brand image over price which provides marketers opportunities to invest on building brand languages. The coefficient comes about of Bass model implies Chinese consumers are more imitative than innovative. Thus playing with word-of-mouth or relationship referrals are more effective during products being diffused.

Furthermore, conventional promotion is as well useful when it is applied in the areas with less population density where consumers are more price responsive. Meanwhile, firms need to consider demographics into their marketing strategy given China is the place varies in both social aspect and geographical aspect. Using the segments wisely will largely benefit firms to stand solid on Chinese market, especially for foreign car brands which already possesses in-advance advantages.

Key words: Conjoint Analysis, Innovation Diffusion, Bass Model, Autonomous Vehicles, Chinese Market, Marketing Strategy

Acknowledgment

I would like to take this opportunity to express my appreciation here officially. Not only for the people who have helped me during my master thesis but also my school, university of Groningen, who grants me valuable knowledge and skills. More specifically, I would like to thank Dr. Keyvan Dehmamy who supervises me as my first mentor. His on-time professional input has encouraged me to progress further on writing my thesis. More importantly, I consider himself not only a good teacher but also is easy going friend which lessen the burden of me, which makes my thesis process a happy journey. Moreever, I would like to thank my second supervisor Dr. Felix Eggers. I have been following his courses since my pre-master phase and his detailed, vivid description of conjoint analysis impressed me back then. He is also the first person who seeded conjoint analysis as one of the most important analysis I want to perform for my thesis. Lucky as I am, to be chosen to investigate Chinese consumers via conjoint analysis, I feel very motivated and appreciated for such an opportunity. Lastly, I would give my appreciation for my family, who supports me unconditionally no matter where I am. All in all, I am glad this journey has come to an end. This journey fosters myself to become a stronger and more professional person. Thank you.

Sincerely, Tiantian Wang

Contents

A.Innovative Diffusion Model 9

1) Innovativeness 9 2) Imitativeness 11 B. Preferences Measurements 12 3) Conjoint Analysis 12 4) Market Size 13 4. Methodology 14

A.Research Method Design 14

B. Model Structure 14 1) Measurement of coefficient 𝑝 16 2) Measurement of coefficient 𝑞 17 C. Model Parametrization 18 5. Results 20 Sample Overview 20 Customer preferences 20

Market Share Estimation 25

Bass Model Estimation 27

Hypothesis Testing 29

6. Discussion and Conclusion 31

1. Introduction

Given the economic boom in the 21st century, consumers are no longer solely content with the products offered by organizations. Up to date, consumers are more concerned with the products or services are delivered to them by their preferences. As both technology and society prospect continuously, innovation is of a great concern, to convey new products and services (Hauser, Tellis & Griffin, 2006).

In order to enhance sustainable competitive advantages for firms, looking into how the trend of innovation has developed is necessary. Over the last few years, the trend of looking into the definitions and motivations to innovate differs among various scholars and industrial practitioners. Subjects areas are as diverse as sociology and economics. As pioneers who researched innovation, Zaltman, Duncan & Holbek (1973) referred innovation as a process of generating fresh cognitive perspectives or with novel behavioral executions. Nevertheless, according to Goldsman and Foxall (2003), innovation was widely defined and differently in high-technology business, agriculture, medical business or educational research. Within boundary of this research paper, the research topic will be in the marketing context, as investigating to drive profitability by innovative products meanwhile satisfying consumer’s needs (Hauser et al., 2006). Car industry, as a technical innovation, has experienced a fluctuating trend up and down after the Industrial Revolution for nearly a century (Chan, 2007). Recently, traffic congestion and development in artificial intelligence, driven by urbanization, are speaking for a demand for a transportation solution that allows travellers to arrive at their destinations with efficiency. Autonomous car rises in response to the proper time and conditions. According to Viereckl et al. (2015), there will be a surprising growth in the automobile industry in China and the U.S.

innovation. The behaviours can be grouped as either innovator or imitator behaviours of adoption. The theory was applied to generic classes of products, removing the brand or new models. The main factors behind selecting the Bass model for the project are: the simplicity of the model, generalisation to a number of products, flexibility in estimating parameters and the extensions that capture more complex modelling scenarios.

With respect to transport, a number of studies have examined the purchase of new alternative fuel vehicles using the Bass diffusion model (Massiani and Gohs, 2015). The wide application of the model indicates it will be able to capture the diffusion of future technologies in the transport system. Existing research mostly concentrates on the influence on autonomous car market diffusion by external factors such as price and infrastructure (Park, Kim & Lee, 2011), rarely cooperating consumer preferences into the modeling process. Given by such a background, as including Chinese consumer preferences along with measuring the innovativeness and imitativeness, will shed some new light in this field.

2. Background

The Chinese car market provides an interesting context for this research, as it is an example of a typical market in an emerging economy. The Chinese car market has experienced exponential growth in the last decade compared to other more mature markets. In 2010, China overtook the United States to become the largest car market in the world when comparing annual car sales （Qian & Soopramanien, 2014). Annual sales of new cars in China, the United States and Japan are illustrated in Fig. 1 as a non-linear growth pattern.

Fig. 1 Sales data of new cars in USA, Japan and China from 1993-2009. Resource: Adopted from Qian & Soopramanien, 2014.

The Chinese car market is far from reaching market saturation. The average car ownership has increased from 100 cars per 1000 inhabitants in the early 1990s to 120 cars per 1000 inhabitants in 2005 (Qian & Soopramanien, 2014. ). Even though China has a much lower car ownership level: every 1000 inhabitants only owned 15 cars in 2005. This data has increased by more than 10 times since 1990.

benefits consumers are: 1)shorten travel time; 2)reduce accident risk; 3)reduce the cost per kilometer compared to renting or buying a car; 4) solve traffic congestion (Mckinsey China, 2018; Viereckl et al., 2015). In 2017, China Automotive Engineering Society released the “Technology Roadmap for Energy Saving and New Energy Vehicles”, which mentioned that by 2020, the scale of the automobile industry will reach 30 million, and the market share of driver assistance/partially autonomous vehicles will reach to 50% (Zhihu.com, 2018). As indicated by conjecture information, the worldwide offers of driverless vehicles will achieve 21 million units by 2035. McKinsey predicts that China is highly likely to become the world's biggest autopilot advertise later on. By 2030, autopilot-related new vehicle industry and travel administrations business will create more a value more than $500 billion (McKinsey China, 2018).

3. Literature Review

In the present dynamic and always showing signs of change business condition, the existence cycle of new innovation is abbreviated, and rivalry between players in the market is heightening. To design the feasible improvement, it is necessary to set up a model which allocates dynamic business environment for mid-term to long-term estimation. In order to compete strategically and grow sustainable advantages, employing innovation diffusion models to aid firms to reach such purposes is wise (Wang and Wang, 2016). Straightforward as it is, Bass model is founded as a diffusion process which characterizes the development dissemination as the procedure by which an advancement is imparted through specific channels after some time, among the individuals within a certain social framework (Roger, 2010). For the adoption and diffusion of innovative products and technologies, the Bass Diffusion Model was proposed by American management psychologist Frank M. Bass and its extension theory are often used as market analysis tools. New product and new technology needs can be forecasted. As one of the many market tools, the main function of the Bass model is to describe and predict the market purchases of newly developed consumer durables. Despite the fact that the Bass model is largely used, it also often gets criticized for its simplicity and robustness (Hauser et al., 2006).

Major parts of the conceptual model consist of combining the measurements of innovativeness (𝑝: innovator), imitativeness (𝑞: imitator) of Chinese consumers and market size (𝑚), which are formulated in the Bass Model. Each definition and measurement will be illustrated in the following sections.

1) Innovativeness

Goldsmith, 1993b). According to Hynes and Lo (2006), for companies who are ambitious to be more identifiable by consumers, developing innovativeness for new products is imperative.

● Definitions

Being a constant quality of personality which individual carries on for lifetime, innovativeness mostly is considered as a built-in personal traits. In other words, each individual at different extents of being innovative. This is often inferred as “innate innovativeness”. Specifically speaking, innate innovativeness is the extent for an individual of how he/she thinks of any new ideas and will he/she act on certain new ideas, regardless the experience generated whiling communicating with other individuals (Midgley and Dowling, 1978; Hirschman, 1980a). However, Roger and Shoemaker (1971) argued that there are many variables which can have impact on innovativeness. Such as previous educational level, working experience or cultural background etc. This argumentation brought innovativeness out of “born with” senario, instead, it adds some social atmosphere into it. Which later on this perspective was supported by Hirschman (1980a) in his works. In addition, not only can innovativeness be affected by external influences, a person may just own a high degree of innovativeness in one particular product category (for example, computer products) but he or she may show very little interest in other product categories (for example, clothing). This aspect brings difficulty to detect general innovators since only specific product category innovative spirits cannot speak for overall innovativeness (Gatignon and Robertson, 1985). Therefore, scholars advise specifying under which product domain innovativeness is measured can be seen as an improvement. (Foxall and Goldsmith, 1988; Goldsmith and Flynn, 1992).

● Measurement

Many existing literatures tend to distinguish innovators and imitators by the time of adoption happened. Often it is assumed that earlier individual who adopts an innovation, more innovative individual indicates (Goldsman & Foxall, 2003). However, such comprehension (time-of-approach) is open to debate. For an instance, Midgley & Dowling (1978) suggested an updated measurement (cross-sectional approach) which provides a list of current innovative products to respondents. Then calculating how many items out of all, were purchased beforehand by respondents. Personality trait is also involved with innovativeness.

plays an important part in measuring innovativeness. Therefore, indicating product-category knowledge of a consumer as a part of measuring innovativeness is as well of great interest. For instance, by asking consumer’s familiarity or certain knowledge, specifically in the field of autonomous vehicles, will sufficiently measure part of innovativeness.

2) Imitativeness

Research has shown that spreading word of mouth and generate customer referrals in various situations, is particularly crucial during new product diffusion (Mahajan, Muller, and Bass, 1995). When consumers make general purchase decisions, they often seek others’ opinions or social approval (Feick and Price, 1987).

● Network Externalities

Starting from literatures 20 years ago, adoption of innovation has been described as a substantial process where with sequential adopters are the vital players (Burt and Friedkin, 1991). Sequential adopters are individuals who will adopt innovations based on opinions from previous adopters. This effect of network externalities is not only found in economics but also in technology development. For example, by taking social judgement into product development may as well result in a bandwagon effect in innovating products (Strang and Macy, 2001; Abrahamson and Rosenkopf, 1993). Within the scope of this research paper, network externalities is referred as a direct network externality. Specifically with an increase of the number of same product will create a network effect with additional value. (Hauser et al., 2006). Real life application of this network effect can be found in the fax machine industry. In terms of measurement of this effect, previous researches can be acknowledged. Researched from Tucker (2007) confirmed that adoption within a certain network is more effective in a form of direct contacts instead of indirect contacts. Therefore, in this paper ‘direct contact’ can be defined as the number of each respondent’s contacts that they have from a certain social platform.

● NPS

Reichheld (2003), NPS is the most popular indicator of customer loyalty analysis, focusing on how the word-of-of consumers mouth affects business growth. Naturally, for firms who keeps track of net recommendations, companies can make themselves more successful. The nature of NPS as simplicity and usability have made NPS one of the most practical metrics in various industries, for firms to promote and forecast future business (De Haan, Verhoef & Wiesel, 2014; Morgan and Rego, 2006). In this paper, NPS is used as a mirror tool. For instance, by provided detailed information of NPS of different companies, reflecting how other consumers evaluate these brands, later on ask respondents whether will they consider purchasing solely relied on NPS scores. By doing so, it captures each individual’s imitative ability, the ease that each respondent rely on others opinions on brands.

B. Preferences Measurements

As some researches indicated before, most of them strived to one cultural factor, such as ‘mianzi’ (similar to vanity), will affect the consumer behavior during the process of purchase goods (Hu, 2006). Unfortunately, those researches only included individual factor separately, so that their results cannot prove the real purchase where features and characteristics are involved. Therefore, it is essential to arrange product characteristics in a way to detect consumer’s trade-off thinking process, which characteristic or feature each consumer will choose in the end. Another reason why measuring consumer preference in China matters, is the potential market size. China’s car market is enlarging nowadays as a new boom market so that understanding Chinese consumer preferences will advance the efficiency of strategy.

3) Conjoint Analysis

As the few scholars who initiated conjoint analysis, Luce and Tukey (1964) firstly applied conjoint measurement to the area of mathematical psychology. Green and Rao (1971) developed further on Luce and Tukey's model, then introduced it into marketing research. In the 1980s, conjoint analysis and computer programming technologies such as adaptive conjoint analysis developed rapidly, resulting in the development of commercial conjoint analysis programs (Orme, 2006).

procedure provide a relatively more explicit information of what attribute is mostly preferred by consumers. Nevertheless, such procedures can be hardly linked to the real life situation since consumers often face actual choices that are full of alternatives, not directly to a single attribute. Moreover, likert scale such as “1-7” of evaluation might be less meaningful in terms of measuring consumer’s purchase intention. In real life situations, a “7” does not necessarily guarantee an absolute purchase willingness. As for choice-based analysis (CBC), it compensates the disconnection that rank-based or rating-based procedure have with real-life situations, respondents will face with various alternatives. It also solves the problem that when using rating-based or rank-based procedure, different cultures might process numeric numbers differently, such as that some cultures would avoid certain numbers or have round-off system (Eggers and Sattler, 2011).

4) Market Size

4. Methodology

The general methodology employed in this research paper is summarized in Fig. 2 and will be explained in details afterwards.

Two methods are applied in this research to analyze the research questions formulation: the demand forecast of the AV market in China, and the consumer preferences of Chinese consumers in the AV market. Bass Model is employeed to forecast the future of AV market and its diffusion process in China. Conjoint analysis is made in understanding the Chinese consumers preferences of several ‘choices’ of combination of attributes and also possibly to identify plausible marketing strategy in the discussion stage. By implementing the result from Conjoint Analysis into the Bass Model, Conjoint Analysis serves as an embedded aspect in this research to complete the estimation of the diffusion process, e.g market share.

Two main research questions are drawn with sub-questions.

1. What is the potential market size for autonomous car in China, when taken Chinese consumers preferences into consideration?

1.1 Will consumer preferences be moderated by demomgrahpic charistics, e.g. gender? 2. How successful can AV products can be diffused in China?

2.1 What is the innovativeness coefficience in this research ? 2.2 What is the imitativeness coefficience in this research ? B. Model Structure

imply changes in consumers’ behavior, as the autonomous cars, does restrict the measurement of 𝑝 and 𝑞 in this case (Santa Eulalia, Neumann & Klasen, 2011). In order to solve such a dilemma, new way of measuring 𝑝 and 𝑞 is established, by asking respondents questions through survey. Variables that are chosen for measurement are illustrated in Fig. 3.

Fig. 3 Conceputal Model Hypothesis:

1. Car attributes wil have different utility on preferences, moderated by demongraphics. 1a. Brands will have a positive utility on preferences

1b. Price will have a negative utility on preferences 1c. Car size will have a positive utility on preferences

2. Estimates of innovativeness and imitativeness will be differenyly across gender.

Where:

ℎt is the hazard rate, the probability a consumer will adopt the technology at time 𝑡 given they have not previously adopted the technology;

𝑓t is the probability density function of adoption at time 𝑡; 𝐹t is the cumulative distribution function;

𝑌𝑡 is the number of adopters at time 𝑡, such that 𝑌t = 𝑚𝐹t ; 𝑝 is the parameter referred to as the coefficient of innovation; 𝑞 is the parameter referred to as the coefficient of imitation;

𝑚 is the ultimate market potential, ie the estimate number of adopters following complete market saturation.

1) Measurement of coefficient 𝑝

Followed from literature review, it is clear that ‘innovativeness’ or ‘innovation’ is a broad concept depends on ways of interpretation. For the new durable goods, such as transportation, consumers' purchase decisions are not directly affected by other consumers, thus they do not pay much attention to the network external effects of infrastructure (Becker, 2009). Such consumers can quickly adapt to the new mode of transportation, e.g. autonomous vehicles. The innovation coefficient p can reflect the number of consumers who quickly adopt this innovative product without social influence. p is between 0.00 and 1.00 (Bass, 1969). The closer the value is to 1, the faster the innovators adopt. For products with sufficient sales history data, the estimated values of p can be obtained directly by the least squares method, the maximum likelihood method, etc. According to KPMG (2018), the Chinese government allowed first approved AV tests in 2018, but there is not enough historical data. Therefore, a method of analogy is adopted. In short, the p is measured as a mean with the sum of the values (where measure innovativeness) divided by the number of respondents, It is given by:

where:

a1 = the re-scale weight coefficient defined uniquely for the score obtained from βi.

a2 = the re-scale weight coefficient defined uniquely for the scores obtained from 𝛾i. a3 = the re-scale weight coefficient defined uniquely for the scale transfored from 𝛌i

βi = the score given by each individual to one question which represents variable ‘innovative

purchase recency’;

𝛾 i = the accumulated scores given by each individual to three questions which are relevant to their openness (part of innovativeness) extent;

𝛌i= to what extent that each individual have sufficient interest or knowledge of AV market;

It is noteworthy that a1, a2 and a3 are re-scaled differently but are within the same range, based on previous coefficients of 𝑝. Thus the V is divided by 3 to eliminate repeat rescaling. Literature shows that using the Bass model to predict the sales of various durable goods, the empirical value of p is concentrated in 0.01~0.03 (Sun, 2009). Furthermore, Zheng et al., (2013) concluded their prediction for Chinese car market, this p coefficient will be maintained in a range between 0.01-0.0258. Thus, for this paper, integrated with literature, a1, a2 and a3 are re-set in range between 0.01-0.025.

2) Measurement of coefficient 𝑞

Different from innovators, imitators’ purchase decision does not only depends on the effect of network externalities (how many people are already on one platform), but also likely to be influenced what other people think of products. The imitation coefficient q is also between 0.00 and 1.00, the closer the value is to 1, the faster the autonomous car spreads in the potential user group. Similar to the coefficient p, analogy method is applied as well: the q is measured as a mean with the sum of the values (where measure imitativeness) divided by the number of respondents, It is given by: Where x̅ is calculated by individual weight as an average imitative ability measured by each individual’s score on NPS and how many direct contacts he/she owns on one social media platform. The logic behind is even two persons who give same score of the likability of recommondating AV products, whoever has more direct contacts will receive heavier weights in terms of imitation of social influence. Specifically,

Xi = the score given by each individual of the likibility of recommondating AV product; Yi = the number of direct contacts/friend each individual owns on Wechat;

Yj = the total number of contacts which all respondents sum up;

b1 = the re-scale weight coefficient defined uniquely for the score obtained from Xi ;

According to the prediction of sales volume by using the Bass model for various durable goods in China previous years, the empirical value of q is concentrated at 0.3~0.7, and q is rarely greater than 0.5 (Sun, 2009). Referred from Zhang et al., (2013), here the range for re-scaling b1 is between 0.3-0.5.

C. Model Parametrization

A conjoint analysis was used to determine the attribute that has the most influence on consumers interested in buying an autonomous car in the Chinese market. However, respondents might consider attributes all equally important or sometimes they will experience difficulty to choose the best option according to the relative importance (Moore & L, 2004). Thus, this model is established on the basis of works from Eggers and Eggers (2011), where utilized a choice-based joint adoption model to predict potential market size without the help of actual historical data.

Not full factorial design is used because it would indicate that respondents need to choose between all available choice sets. For example, between 3 brands names, 3 price levels and 3 car sizes, which would indicate 3*3*3 = 27 choice sets. In order to keep survey interesting for respondents and cost less effort, orthogonal design is applied in this study (fractional factorial design) to reduce the number of profiles. In choice-based conjoint studies, utilities (U) can be estimated by summing up several product attributes with their preferences estimates (𝞑) as a parameter vector. A multinomial logit model (MNL) is used to estimate each preference. Preferences are estimated using the multinomial logit (MNL) model (Louviere, Hensher & Joffre, 2000), which models preferences in terms of choice probabilities P. For example, observing that an alternative i has been chosen from a set of Z alternatives would be integrated into the likelihood function as given by:

Uij = Overall utility of alternative j for individual i; aij = the utility on the jth attribute for the ith individual; xij = The level on attribute j describing the ith alternative;

As stated in literatures, the market share for i can be presented as the purchase choice probability (P) among all available choice sets, multiple with forecasted data of car demand in China.

Product Attributes/levels ● Brand Nationality

Brand nationality means the origin country of each brand, which is one of main attributes. Data (Autohome, 2013) shows that brands with top 3 sales are all German brands, also indicates that Chinese consumers may choose AV products with a preference on brand nationality. Thus, in this research, three brands are taken as different levels, which are German brand: BMW, American brand: Tesla and lastly Chinese brand : Baidu.

● Car Type

Car type means the type of cars sold in China. It determines the size and exterior of the car, since Chinese consumers consider exterior as one of the most important factors (Sha, 2013). In this research, car types is defined into three levels regarding with the existed car type currently, which are SUV (Sports Utility Vehicle), Mini and Hatchback.

● Price

Price is constantly a key factor which impact purchase decisions. Study from Erickson and Johansson (1985) showed that price and quality have a mutual impact. It found that premium-priced cars sometimes are perceived with additional quality. Furthermore, cars with better quality naturally are more pricey than they actually worth. Thus, in this research, the price is divided into three levels as 250,000 CNY1_{, 450,000 CNY and 1,000,000 CNY.}

Screen shots of survey (English & Chinese Versions) are attached in Appendix I.

5. Results

Data were collected from April 20th until May 15th 2019 via online surveys, in Chinese. There were 415 surveys distributed in total; there were 408 participants (98.3%) and provided valid responses; the remaining 7 responses (2.7%) were discarded due to repetitive survey testing. Among 408 responses, 110 completed responses were documented, indicating respondents answered all questionnaires without skip or exit survey. Whereas the incomplete surveys were surprisingly high, with 298 respondents. Specifically, most respondents who did not finish survey stopped at beginning questions, such as asking demographics. Other incomplete surveys started at where conjoint experiments began. Of all 408 respondents, there were 279 respondents did not answer CBC questions. Thus they were excluded for the part of measuring customer preferences (n = 129). One plausible reason could be going through multiple choice scenarios exhaust respondent’s attention. The descriptive statistics of the respondents are presented in Table 1. A total of 46.4% of the respondents were female and 53.6% were male; most of respondents (52.7%) were between 46-65 years old, second majority (38.2%) were between 18-30 years old; over half of respondents (53.6%) hold a bachelor degree; and largest portion of respondents (42.7%) live in four-tier cities2 _{or below; lastly, regarding income level,} 36.4% of respondents earn over 8000 CNY ( 1045.75 euros) per month.

Customer preferences

Conjoint analysis is used to analyze consumer preferences. Table 2 shows attributes (and their values for conjoint analysis) on an average level noted as important considerations for consumers purchasing an AV: prices, car sizes and car brands. These attributes were identified from background section. The attributes can be combined in a full factorial design, as a set of 3 brands names, 3 price levels and 3 car sizes. This yielded 27 alternatives (3×3×3=27). An orthogonal design to reduce the number of alternatives to 9 by using the conjoint analysis module of SawTooth Software, to retain respondents’ interest. Based on 129 respondents who answered conjoint analysis section, the result reveals that Chinese consumer prefer car brands first (degree of importance = 36.92%, p < 0,05) when purchasing a new AV. Within each three

2 _{Four-tier cities (Four-tier City) are mostly medium-sized cities with urban scale, economic and social}

attributes, different levels are calculated as zero-centered. For example, in terms of specific brands, Chinese consumers would prefer BMW (B = 34.22, p <0,05) , Tesla (B = 6.28, p >0.05) and Baidu Apolo (B = -40.50, p <0,05), in such an order. Then the second important attribute for Chinese consumers is price (degree of importance = 35.97%, p < 0,05), which is very close to the car brands. Furthermore, although price attribute is calculated as a part-worth utility, with the increase in price, the preference estimates does drastically drop. For instance, a car with 250,000 CNY is found to be the most preferred (B = 32.84, p <0,05) whereas with 1,000,000 CNY is the least preferred (B = -36.86, p <0,05). The least important attribute Chinese consumer found is car size (degree of importance = 27.11%, p < 0,05). And SUV is to be the most preferred car size for Chinese consumers (B = 17.65, p<0.05). Nevertheless, among all the options, not choosing an autonomous car is also widely preferred, despite it appears not to be significant by average (B = 13.43, p >0.05). Looking together, what attracts them as a combination would be a BMW car with price of 250,000 CNY with a size of SUV.

However, there are also extreme cases. For example, although Baidu Apollo is least preferred by average, still respondent 114 granted the highest utility of this brand whereas he/she least preferred brand Tesla. Same situation is detected in car type. Respondent 65 appears to be fond of mini size of car the most where he/she gives hatchback and SUV both negative evaluation. For further methodology operation, please refer to the help file of SawTooth.

Table 1. Descriptive characteristics of respondents (N =110)

Item Frequency Size

Age Below 18 years old 2 1.81%

Between 18-30 42 38.2% Between 31-45 8 7.3% Between 46-65 58 52.7% Above 65 0 0% Gender Male 59 53.6% Female 51 46.4%

Education High school or less 11 10%

College 16 14.5%

University Bachelor Degree 59 53.6%

University Master Degree or above 21 19.1%

Others 3 2.7%

Income Below 2000 CNY 7 6.4%

2001-4000 CNY 11 10%

4001-6000 CNY 22 20%

6001-8000 CNY 17 15.5%

Over 8001 CNY 40 36.4%

Prefer not to say 13 11.8%

Current

Third-Tier City 7 6.4%

Four-Tier City or Below 47 42.7%

Prefer not to say 6 5.5%

Market Share Estimation

Figure 5. Products Simulation and Potential Market Share.

Figure 6. Potential Market Share by Genders.

Based on Figure 6, it is clear that female prefer products by large difference whereas male do (P= 49.9%). It can be seen that even with the same price, the influence of brand BMW is extra obvious for female consumers. For Baidu Apollo, product 1, the difference is not obvious. The potential market share of product 1 is least optismitic for male consumers. Another interesting observation is, there are 23.3% male consumers would still choose non-autonomous vehicles, nearly twice the number of females (P = 12.5%). This might indicate current models are not satisfying male consumers demands.

Table 5. Product and Market Simulation Summary

Products Market Size (unit :cars)

Product 1: Baidu Apollo with price of 250,000 CNY

in size of mini 10.3% * 30,000,000 = 3,090,000

Product 2: Tesla with price of 1,000,000 CNY in size

of hatchback 15.6% *30,000,000 = 4,680,000

Product 3: BMW with price of 250,000 CNY in size

of SUV 56.0% * 30,000,000 = 16,800,000

Product 4: Any NON-Autonomous product 18.1% * 30,000,000 = 5,430,000

Bass Model Estimation

Three important parameters should be determined in this model estimation: the ultimate market potential M, the coefficient of innovation p and the coefficient of imitation q. In order to answer all three parameters, respondents who completed the survey are acceptable, which are 110 responses.

1) The coefficient of innovation p is calculated by measuring the means of combination of three aspects within innovation scores individually: the recency of purchased innovative products, the initiative innovation knowledge towards cars and the innovative personality traits, while each aspect has been rescaled according to the literature beforehands. After calculation, innovativeness p for all 110 respondents is 0.01565341. Respectively, for males, where p is 0.01617232 and for females this value is slightly lower, 0.0150531. In order to investigate whether innovativeness be different across demographics, t-test is performed. Result shows that there is highly significant difference between males and females in terms of how innovative they are (Df = 1, p = 0,00869 < 0,05).

difference between males and females indicating former assumption (Df = 1, p = 0,838 > 0,05).

Table 6. Summarized innovativeness and imitativeness across genders.

Overall Male Female Gender

Difference

Innovativeness p 0.01565341 0.01617232 0.0150531 0.00869**

Imitativeness q 0.4040979 0.7545421 0.8700639 0.838

**indicates significance at 5% confidence interval.

3) In order to predict potential sales curve of AV product in Chinese market, the ultimate potential market size plays an essential role. Given by situations that Baidu Apollo will be officially available in China within a year and to simulate a “best choice” product according to Chinese consumers preferences, product 1 and 3 will be taken into considerations (marked as M1 and M3). From Bass (1969), if p and q for a product are known then forecasting its adoption over time will be possible. And thereby generate a time path of sales. For instance, given M1 is 3,090,000, p = 0.01565341 and q = 0.4040979, Then the adoption rate of product 1 its diffusion curve are shown in the Figure 7 below.

Figure 7. Baidu Apollo Diffusion Curve and Adoption Rate

Figure 8. Sale Forecast of Product 1 and Product 3. Hypothesis Testing

6. Discussion and Conclusion

The main purpose of this study was to determine the dominant factors in consumers’ preferences when choosing among autonomous cars, in China particularly. The research examined the product attributes of car brands, price and car types. Respondents were asked to choose their most preferred among two car models plus one non-option. Using a multinomial regression method via Sawtooth, the preferences from respondents are presented as the log-odds ratio, which later can be transformed into potential market share. Without taking any interaction effect, the multinomial regression analysis gives a clear picture that, indeed Chinese consumers are affected by different car attributes and car brand is playing the most important factor. However, when personal estimates were included in the analysis, consumer preferences were moderated by demographics. On the individual level, particular cases were as well spotted, such as, despite of majorities are more fond of foreign cars, there are still minor cases which shows extreme fondness/dislike towards domestic brand.

Price

Price can affect purchase intentions both negatively and positively, indicating from methodology. However, the results of this study does not support the idea that price will have a positive impact of consumer preferences. On the contrary, it proves the global assumption that consumers enthusiasm will decrease by increasing price. Despite of price being significant factor, its influence differs across gender and residency. For instance, females are more easily influenced by how much an AV costs, over males, showing a higher price sensitivity. Meanwhile, for consumers who live in the First-Tier cities, price becomes the least important factor to persuade purchase decisions. The limitation of this variable is discussed later in this thesis.

Car Brands

Based on the results from Conjoint Analysis, domestic and German brands are evaluated very differently in terms of preferences. As literature suggested, car brand is often linked to brand nationality. In this research, German brand BMW is the most popular, then American brand Tesla, lastly Chinese brand Baidu. One plausible explanation could be German brand is entitled with premium quality which is often found in other products. Furthermore, the fact that Chinese consumers are not confident with domestic brands might open another marketing strategy which will be discussed in later chapter.

Car type is the least important factor to influence Chinese consumers in terms of buying an autonomous vehicle. Furthermore, results reveal that SUV perceives the highest utility, suggesting it is the most popular car types more than other types. Nevertheless, car type hatchback ranks second with minor difference and the least preferred is mini type by large difference.

Chinese Market Simulation

A first-choice is considered as an efficient simulation model where hypothetically assume the most promising product set among all alternatives is generated from a combination of highest utilities. Given by the conjoint analysis, three hypothetical products, plus a non-option are generated based on the sum of corresponding attribute’s utilities. Clearly, a product which is designed with BMW brand, sold as 250,000 CNY and in a size of SUV, seems being able to capture 56% market share. This result is insightful because if certain market conditions are reached, this hypothetical choice probability can be really close to the ideal market share where firms actually are interested in. On the other hand, a product from Baidu Apollo with same price but in size of mini, scores poorly in this simulation, with only 10.3% market share potentially. It appears that although price remain the same, Chinese consumers will still consider a car from BMW. This result is also aligned with literatures earlier, that in China price is not the first dominant factor in terms of purchasing important products. This insight can be used later on for marketing startegy.

However this market share prediction also differs in genders. Although Baidu Apollo is the least preferred option, among female respondents it is more popular compared to male respondent (11.4% vs 9.2%). As for product 3 from BMW, female shows propotional larger interest in preference over male (62.7% vs 49.9%). This observation as well matches earlier results, which is for females the first factor to consider is price then followed by brands. Whereas males first consider brands then prices.

How will AV products be diffused in China

means is smaller than imitation, across genders as well. This observation infers that currently in China for durable products, word-of-mouth or relationship referrals are more effective during products being diffused. Especially given by the large population China has, the power of imitation is undoubtedly critical if firms focus on customer relationship management. But does this result also imply firms should focus less on innovation? The answer is obviously no. From results, males are more innovative than females but also less price sensitive. And brands like BMW will experience a explosive increase in numbers at the beginning of product diffusion whereas domestic brands will also grow, but in steady pace. Therefore, having sufficient early adoptors or innovators lay the foundation of succeeding market.

7. Limitation

In order to diffuse autonomous vehicles successfully in China, this research applies conjoint analysis to grasp chinese consumer behaviours then using bass model to forecast potential sales of various AV products. Considering to both analysis, there are several limitations drawn.

Firstly, the design of this conjoint analysis may have generated influence on how each respondent evaluate every alternative among choice set. Since respondents are given by textual description of car models, which is less vivid and effortful to transform words into actual images. Another downside is, although the number of choice sets has been cut down to 9 (which is considered as optimal), respondents still show partial impatience filling out surveys or directly exiting survey which caused missing data. A future suggestion regarding this matter will be, present choice sets visually to present car models. This might also ease respondent’s patience to complete survey.

Secondly, the effect of price attribute is measured as part-worth method. Thus, it is unclear by how much increase/decrease in price will affect consumer’s willingness-to-pay in a linear regression model. Also it is ambiguous to judge the true impact of price attribute due to the fact that there is no “real” purchase. Whole simulation is based on hypothetical setting where exists hypothetical bias, which in real life decisions might be completely different. One solution to this matter could be, aligning incentives to respondent’s choices.

Thirdly, the estimates of imitativeness and innovativeness are based on analogies test where there is no historical sales data to validate. During analogies test, scope of each coefficient is pre-selected from previous literatures where those estimates were made across all industries. There could exist minor differences in terms of different industry has different innovativeness and imitativeness. One remedy for further research could be, to validate innovativeness and imitativeness coefficients by the help of historical data, if it is available.

8. Implications

Despite of the topic of this research is autonomous vehicle only, but by substituting various attributes and levels, this paper can also be used as analyzing other products or industries. For instances, house-cleaning robots, doctor robots or food-delivery packages. Theref, this research can also contribute to generating other insights from other countries, to capture local consumer behaviors.

Managerial Implications

Firms can utilize conjoint analysis in their benefits since it provides them a clear path of how should they design their products in order to maximize their interests, given a highly competitive environment. For firms whose goals are to reach the maximum market share, within conjoint simulation, firms can provide products which combine highest utilities (best features) by offering the lowest price. However, this often is seen as a bold and unrealistic solution due to the fact that firms need to take production costs into consideration, as the best quality often costs more. Then the second best solution could be simulate incremental benefits based on current product attributes, relative to the marginal cost offering to consumers.

References

Abrahamson, E., & Rosenkopf, L. (1993). Institutional and competitive bandwagons: Using mathematical modeling as a tool to explore innovation diffusion. Academy of management review, 18(3), 487-517.

Assets.kpmg. (2019). 2019 Autonomous Vehicles Readiness Index. [online] Available at:

https://assets.kpmg/content/dam/kpmg/xx/pdf/2019/02/2019-autonomous-vehicles-readiness-index.pdf [Accessed 15 Apr. 2019].

Bass, F. M. (1969). A new product growth for model consumer durables. Management science, 15(5), 215-227.

Burt, R. S. (1980). Innovation as a structural interest: Rethinking the impact of network position on innovation adoption. Social Networks, 2(4), 327-355.

Chan, C. C. (2007). The state of the art of electric, hybrid, and fuel cell vehicles. Proceedings

of the IEEE, 95(4), 704-718.

Cho, H. J., Jin, B., & Cho, H. (2010). An examination of regional differences in China by socio-cultural factors. International Journal of Market Research, 52(5), 613-633.

De Haan, E., Verhoef, P. C., & Wiesel, T. (2015). The predictive ability of different customer feedback metrics for retention. International Journal of Research in Marketing, 32(2), 195-206.

Deroıan, F. (2002). Formation of social networks and diffusion of innovations. Research policy,

31(5), 835-846.

Eggers, F., & Eggers, F. (2011). Where have all the flowers gone? Forecasting green trends in the automobile industry with a choice-based conjoint adoption model. Technological Forecasting and Social Change, 78(1), 51-62.

Eggers, F., & Sattler, H. (2011). Preference measurement with conjoint analysis. Overview of state-of-the-art approaches and recent developments. GfK Marketing Intelligence Review, 3(1), 36-47.

Eggers, F., Sattler, H., Teichert, T., & Völckner, F. (2018). Choice-Based Conjoint Analysis. In C. Homburg, M. Klarmann, & A. Vomberg (Eds.), Handbook of Market Research Springer.

Erickson, G. M., & Johansson, J. K. (1985). The Role of Price in Multi-Attribute Product

Evaluations . Journal of Consumer Research , 12 (2), 195- 199.

Feick, L. F., & Price, L. L. (1987). The market maven: A diffuser of marketplace information. Journal of marketing, 51(1), 83-97.

Fetscherin, M., & Toncar, M. (2010). The effects of the country of brand and the country of manufacturing of automobiles: An experimental study of consumers' brand personality perceptions. International Marketing Review, 27(2), 164-178

Flynn, L. R., & Goldsmith, R. E. (1993a). A validation of the Goldsmith and Hofacker innovativeness scale. Educational and Psychological Measurement, 53(4), 1105-1116.

Flynn, L. R., & Goldsmith, R. E. (1993b). Identifying innovators in consumer service markets.

Service Industries Journal, 13(3), 97-109.

Friedkin, N. E. (1991). Theoretical foundations for centrality measures. American journal of Sociology, 96(6), 1478-1504.

Goldsmith, R. E., & Flynn, L. R. (1995). The domain specific innovativeness scale: theoretical and practical dimensions. In Association for Marketing Theory and Practice Proceedings (Vol. 4, No. 1995, pp. 177-182).

Goldsmith, R. E., & Foxall, G. R. (2003). The measurement of innovativeness. The

international handbook on innovation, 321-330.

Goldsmith, R. E., & Goldsmith, E. B. (1996). An empirical study of overlap of innovativeness. Psychological Reports, 79(3_suppl), 1113-1114.

Green, P. E., & Rao, V. R. (1971). Conjoint Measurement-for Quantifying Judgmental Data.

Journal of Marketing research, 8(3), 355-363.

Horsky, D., & Simon, L. S. (1983). Advertising and the diffusion of new products. Marketing Science, 2(1), 1-17.

Hynes, N., & Lo, S. (2006). Innovativeness and consumer involvement in the Chinese market.

Singapore Management Review, 28(2), 31-46.

Jackson, D. N. (1976 ). Jackson Personality Inventory Manual. Port Huron, MI: Research Psychologists Press.

Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated choice methods: analysis and applications. Cambridge university press.

Luce, R. D., & Tukey, J. W. (1964). Simultaneous conjoint measurement: A new type of fundamental measurement. Journal of mathematical psychology, 1(1), 1-27.

Mahajan, V., Muller, E., & Bass, F. M. (1995). Diffusion of new products: Empirical generalizations and managerial uses. Marketing science, 14(3_supplement), G79-G88.

Massiani, J., & A Gohs. (2015). The choice of Bass model coefficients to forecast diffusion for innovative products: an empirical investigation for new automotive technologies. Research in Transportation Economics: 17–28.

McKinsey & Company. (2013). Upward mobility: The future of China's premium car market.

[online]Available

at:https://www.mckinsey.com/industries/automotive-and-assembly/our-insights/upward-mobility-the-future-of-chinas-premium-car-market [Accessed 17 Apr. 2019].

Midgley, D. F. & Dowling, G. R. (1978). Innovativeness: The concept and its measurement.

Journal of Consumer Research, 4 (March): 229–242.

Moore, & L., W. (2004). A cross-validity comparison of rating-based and choice-based conjoint analysis models. International Journal of Research in Marketing , 21 (3), 299–312.

Morgan, N. A., & Rego, L. L. (2006). The value of different customer satisfaction and loyalty metrics in predicting business performance. Marketing Science, 25(5), 426-439.

Orme, B. K. (2006). Getting started with conjoint analysis: strategies for product design and

pricing research. Research Publishers, LLC.

Park, S. Y., Kim, J. W., & Lee, D. H. (2011). Development of a market penetration forecasting model for Hydrogen Fuel Cell Vehicles considering infrastructure and cost reduction effects.

Energy Policy, 39(6), 3307-3315.

Popkins, N. C. (1998). The five-factor model: Emergence of a taxonomic model for

personality psychology. Available at: http:/ /www.personalityresearch.org.

Qian, L., & Soopramanien, D. (2014). Using diffusion models to forecast market size in emerging markets with applications to the Chinese car market. Journal of Business Research, 67(6), 1226-1232.

Reichheld, F. F. (2003). The one number you need to grow. Harvard business review, 81(12), 46-55.

Rogers, E. M. (2010). Diffusion of innovations. Simon and Schuster.

Sha, S., Huang, T., & Gabardi, E. (2013). Upward mobility: The future of China’s premium car market. McKinsey & Company: Automotive & Assembly Practice.

Tucker, C. (2007). Interactive and option-value network effects in technology adoption. Mimeo, MIT.

Viereckl, R., Ahlemann, D., Koster, A., Jursch, A. (2015). Connected Car Study 2015: Racing

ahead with autonomous cars and digital innovation. Available at:

http://www.strategyand.pwc.com/reports/connected-car-2015-study. Retrieved on 27th Feb, 2019.

Wang, H. T., & Wang, T. C. (2016). Application of the grey Lotka–Volterra model to forecast the diffusion and competition analysis of the TV and smartphone industries. Technological

Forecasting and Social Change, 106, 37-44.

Zaltman, G., Duncan, R., & Holbek, J. (1973). Innovations and organizations. John Wiley &

Sons.

Zhang, S. S., van Doorn, J., & Leeflang, P. S. (2014). Does the importance of value, brand and relationship equity for customer loyalty differ between Eastern and Western cultures?.

International business review, 23(1), 284-292.

Zhuanzhi.ai. (2018). 无 人 驾 驶 产 业 发 展 现 状 及 影 响 - 专 知 . [online] Available at:

http://www.zhuanzhi.ai/document/42626467e65c8a581468d7f2a53f37ca [Accessed 15 May.

APPENDIX II

Lower 95% CI Upper 95%CI

Car Brands 36.92%** _{0.1599 0.3416 0.3968}

Price (exc.Tax) 35.97%** _{0.1565 0.3327 0.3867}