Predicting the Bass diffusion model for drone delivery services: using measures of consumer innovativeness and consumer imitativeness in a choice-based conjoint analysis.

(1)

Predicting the Bass diffusion model for drone delivery services: using measures of

consumer innovativeness and consumer imitativeness in a choice-based conjoint

analysis.

by

(2)

Predicting the Bass diffusion model for drone delivery services: using measures of

consumer innovativeness and consumer imitativeness in a choice-based conjoint

analysis.

by

Annemijn Maria Sophia Geertje Steinmeijer

June 16, 2019

Master Thesis

MSc Marketing Intelligence & MSc Marketing Management University of Groningen

Faculty of Economics and Business Department of Marketing PO Box 800, 9700 AV Groningen

Supervisors:

First Supervisor: dr. K. (Keyvan) Dehmamy Second Supervisor: dr. F. (Felix) Eggers

Author: Annemijn Steinmeijer Student number: S2764490 Address: Jozef Israëlsstraat 41A

(3)

PREFACE

This master thesis was written to complete my MSc Marketing Intelligence and MSc Marketing Management programs at the University of Groningen, which I started in 2017. After starting the master in September 2017, I realized that both tracks fascinated me and I really enjoyed taking courses from both the Intelligence and Management track. Whereas I especially enjoyed learning about the branding and consumer psychology mechanisms and theories from the MSc Marketing Management courses, I experienced great pleasure from the MSc Marketing Intelligence courses by learning about how to deal with marketing data. Writing my MSc thesis was challenging, and that is exactly what I wanted it to be. I really enjoyed further developing and deepening my understanding of marketing analytics, especially being able to analyze data in R-studio. I would like to thank my supervisor dr. Keyvan Dehmamy for his help, feedback and ideas on how to improve my thesis. Due to his great guidance, I was able to take my thesis to the next level. In the end, I feel really proud of what I have accomplished and I can’t wait for my next adventure.

(4)

ABSTRACT

The Bass diffusion model, developed by Frank Bass (1969), has some estimation limitations when it comes to predicting the innovation diffusion of a fundamentally new product or service. This study uses a choice-based conjoint analysis and incorporates measures of consumer innovativeness and imitativeness, in order to estimate the diffusion of drone delivery services. This study gives insights into how a choice-based conjoint can be used to estimate the Bass diffusion model. Furthermore, this study unveils consumer preferences for drone delivery services, at an aggregated and individual level. Additionally, this study made consumer segments based upon differences in preferences for drone delivery services. This study shows that a choice-based conjoint analysis with measures of

innovativeness and imitativeness enables researchers to estimate the Bass diffusion model in situations where it might be hard to estimate the market share parameter (m), innovativeness parameter (p) and imitativeness parameter (q). Besides, this study provides managers with insights concerning consumer preferences for innovative autonomous drone delivery services.

(5)

(Regev, 2018). This might feel like something you would see in a science-fiction movie. However, with big companies like Amazon, UPS, 7-Eleven and even Domino’s investing in drone delivery services, it will probably become reality in the near future (Wingfield & Scott, 2016; Desjardins, 2018). The reasons behind these companies investing in drone delivery are based upon several trends occurring. First, consumers have ever-evolving delivery service demands to accommodate in their busy lifestyles, they expect delivery services to deliver their packages when they want it, where they want it and they demand that it will happen at the lowest possible cost (Schaupp & Bélanger, 2005). Furthermore, autonomous driving vehicles can be seen as the future, since they are able to offer a service at a lower cost and a lower environmental footprint (Manjoo, 2016; Joerss, Neuhaus & Schröder, 2016; Regev, 2018). Lastly, the current main form of delivery service is dependent on airports and roads, which are clogging up (Manjoo, 2016). The solution was found in drone delivery service, and with over half of the consumers being open to drones delivering their packages, it is an innovation bound to disrupt the whole delivery service market in the near future (Joerss, Schröder, Neuhaus, Klink & Mann, 2016; Hoogendoorn, 2017; Desjardins, 2018).

Organizations in the delivery service market are facing a highly competitive future, with drone delivery service disrupting the market, making it inevitable for organizations to invest in product innovations (Brown & Eisenhardt, 1995; Ernst, 2002). Innovations are equal to new product development, which is a process of offering a new product to the marketplace (Craig & Hart, 1992; Brown & Eisenhardt, 1995; Balbontin, Yazdani, Cooper & Souder, 1999; Ernst, 2000; Rubera, Chandrasekaran & Ordanini, 2015). Understanding customer needs and being able to predict the market potential for an innovation is found to be a critical factor determining the success of a new product (Cooper & Kleinschmidt, 1987; Mishra, Kim & Lee, 1996; Balbontin et al, 1999; Ernst, 2000). The question that arises is how this drone delivery service will spread among customers and under what conditions. In other terms, what attributes and characteristics are found important for customers when drone delivery service will become a reality and how will this innovation diffuse among customers.

(7)

behaviors, namely innovativeness and imitativeness (Bass, 1969; Massiani, 2015). Furthermore, to predict the innovation diffusion model of Bass (1969), the coefficient determining the market potential must be estimated.

Even though the Bass model is being applied extensively, it has raised some critique, especially for predicting the diffusion of fundamentally new products and services. It was found that the predictive validity of the Bass model is lacking in the early years of a product’s life circle, since the available sales data is either limited or not available (Heeler & Hustad, 1980). Since drone delivery service is a fundamentally new service, no prior sales data is available to estimate the diffusion. However, two solutions were found to this estimating issue. Estimating the parameters for

innovativeness (p), imitativeness (q) and market potential (m) without prior sales data of the actual innovation can be done in two ways. First, estimates can be obtained by accessing large databases containing historical sales data on comparable products or services, providing possible estimates of the innovativeness, imitativeness and market share parameters (Mahajan, Muller, Bass, 1990; Bass, 2004). This provides the approach of ‘’guessing by analogy’’, whereby the parameters of a new product will be determined by a similar product’s diffusion pattern available in the database (Bass, 2004).

Secondly, management can use their judgment to estimate innovation diffusion (Lawrence and Lawton, 1981; Mahajan, Sharma, 1986). This provides the approach of ‘’guessing without data’’, which provides intuitive interpretation (Bass, 2004).

However, there are some drawbacks to the two above mentioned solutions. First, the Bass model relies on an S-shaped curve for the product lifecycle, indicating that there will be both a takeoff and a declining phase (Massiani, 2012). However, since drone delivery service is expected to disrupt the whole delivery service market, it is expected to span a long lifetime (Massiani, 2012).

Additionally, all of the foregoing available data on product diffusion curves will be based on this S-shaped curve, which might not be the case with drone delivery service. Secondly, Bass himself mentions that both of the ‘’guessing’’ approaches are found to be imperfect, since they are in the end still guesses (Bass, 2004). Besides, it is extremely hard for managers to intuitively guess for a fundamentally new product, since they have nothing to base their guess on. Moreover, it was found that the available estimates for the behavior parameters have huge discrepancy’s, without consensus in literature about a selection criterion among all of these estimates (Massiani, 2015). More importantly, for a fundamentally new product, it seems impossible to use existing data from a similar product, since there are seemingly no similar products available.

This leads to problems in estimating the parameters innovativeness (p), imitativeness (q) and market share potential (m) without prior sales data available for a fundamentally new and disruptive innovation. This study will integrate measures of consumer innovativeness and consumer

imitativeness into a choice-based conjoint analysis, to improve the estimation of the Bass diffusion model, using drone delivery service as an innovation.

(8)

problem of unreliable estimates for the parameters innovativeness (p) and imitativeness (q). Consumer innovativeness is found to be a valid predictor for adoption of innovations (Hauser, Tellis and Griffin, 2006). These consumers will decide to adopt a product independently of their social system. The degree of consumer innovativeness will be the parameter estimate for innovativeness (p) in the Bass model. The innovativeness parameter will measure innovativeness in terms of dispositional

innovativeness (Steenkamp & Gielens, 2003). On the other hand, consumer imitativeness is the degree to which individuals get influenced by others in their choice to adopt a new product or not (Bass, 1969). Since no validated scales are available for consumer imitativeness, the closely related concept of consumer susceptibility to interpersonal influence will be used, which entails both normative social influence and informational social influence. The degree of consumer susceptibility to interpersonal influence will be the parameter estimate for imitativeness (q) in the Bass model (Bearden, Netemeyer & Teel, 1989).

To solve the issue of ‘guessing’ the market potential incorrect, a choice-based conjoint

analysis will be conducted concerning drone delivery service. This choice-based conjoint analysis will show the preferences of consumers in drone delivery service services, on the bases of the attributes delivery time, delivery price and delivery brand. The utility function, made by including the aforementioned attributes, will be able to predict purchase decisions (Eggers, Sattler, Teichert & Völckner, 2018). The outcomes of a choice-based conjoint analysis will be able to reflect actual outcomes, which can be translated to estimating market shares (Wiley & Bushnell, 1979; Desarbo, Ramaswamy & Cohen, 1995). This means that the choice-based conjoint analysis will be able to uncover consumer preferences for drone delivery service and estimate market shares using a conjoint market simulation.

This study is structured in the following manner; chapter 2 will review the literature regarding drone delivery service, new product development, consumer innovativeness, consumer imitativeness and conjoint analysis, resulting in hypothesis and a conceptual model. Chapter 3 contains the

(9)

2. LITERATURE REVIEW 2.1 Drone Delivery

With the introduction of e-commerce, which enabled retailers to sell their products through the use of internet, delivery services became more important. The global costs of package delivery

nowadays is being estimated at 70 billion euro’s, and has grown tremendously by the introduction of e-commerce (Joerss, Neuhaus & Schröder, 2016). Consumers have an ever increasing demand for increased convenience and lower costs when it comes to delivery services, and they keep expecting more (Pooler, 2016; Hoogendoorn, 2017). Delivery service operators are trying to accommodate to these demands, resulting in a large amount of options concerning package delivery services, including free delivery, pick-up-points, delivery at work, weekend and evening delivery, same-day delivery, tracking services and Apps (PostNL, Amazon, DHL). However, even with all of these available add-ons, the price of a delivery service was still found to be the most important for customers (Joerss et al, 2016). However, the importance of price might be shifting, since 25% of consumers are willing to pay a price premium for same-day or instant delivery services, with the percentage being even higher for younger consumers, over 30% (Joerss, Neuhaus & Schröder, 2016).

Last-mile delivery, the movement of goods from a transportation hub to the consumer’s preferred delivery destination, is currently dependent on a vehicle operating in a delivery area, packed with as many packages as possible (Hoogendoorn, 2017). Traditional delivery service firms like UPS and FedEx are receiving heightened competition from technology firms like Amazon, Google and Uber by being able to disrupt the delivery service market with innovations like autonomous driving vehicles (Popper, 2016; Soper, 2016; Aquino, 2016). Autonomous driving vehicles, like drones and autonomous driving cars, can be seen as the future of delivery services (Joerss, Neuhaus & Schröder, 2016). Drones can deliver packages faster at lower costs with less energy needed, which is a win-win situation for both consumers and retailers (Desjardins, 2018; Shimer, 2018). It is for instance predicted by the Deutsche Bank that Amazon’s delivery costs can be cut in half with the introduction of drones, making the landscape extremely competitive (Manjoo, 2016). In terms of faster delivery times, Jeff Bezos, CEO of Amazon, first announced his Amazon Prime Air delivery service in 2016, where drones will be able to deliver packages within 30 minutes (Amazon.com/Amazon-Prime-Air). It is expected that in the future, current forms of delivery service will take longer to deliverer packages that nowadays, since all traditional forms of transportation, like for instance roads and airports, are

expected to be jammer by the year 2045 (Manjoo, 2016).

Looking back at what consumers find important, which was mostly a low price and a newly raised shift towards very fast delivery, drone delivery services might be able to fulfill both of these customer needs. Besides, it was found that over half of the consumers are open to the innovation of drones delivering packages (Joerss et al, 2016; Desjardins, 2018).

(10)

partly or fully replace traditional delivery services (Wingfield & Scott, 2016).

2.2 New product development

Organizations are facing a competitive landscape nowadays, with a continuous development and introduction of new products and services (Ernst, 2002). New product development is the process of offering a new product or service to the marketplace, also often revered to as product innovations (Craig & Hart, 1992; Brown & Eisenhardt, 1995; Balbontin et al, 1999; Ernst, 2000; Rubera,

Chandrasekaran & Ordanini, 2015). Product innovations seem inevitable for organizations, especially in fast-paced and competitive markets (Brown & Eisenhardt, 1995). Nonetheless, new product development is a risky process, considering that new products and services can either produce beneficial outcomes for an organization in terms of increased profits, or cause financial losses when the new product fails (Sivadas & Dwyer, 2000). A critical factor determining the success of new product development is found to be customer and market orientation, which concerns understanding customer needs and accurately predicting the market potential (Cooper & Kleinschmidt, 1987; Mishra, Kim & Lee, 1996; Balbontin et al, 1999; Ernst, 2000). Since drone delivery services are a new product development, it is interesting to predict the market potential and how this innovation will diffuse in the market under the current customer needs and preferences for drone delivery services.

2.3 Innovation diffusion

(11)

1990). In other terms, diffusion models focus on two behavior mechanisms that influence adoption, namely word-of-mouth communication, which is communication through interpersonal channels, and mass-media communication, which is communication through mass-media channels (Mahajan & Muller, 1979). Mass-media communication primarily creates knowledge about innovations, while word-of-mouth influences the formation and altering of attitudes towards the innovation (Rogers, 1962). It is assumed that mass-media will generate awareness about the innovation, while word-of-mouth within a social system will influence other individuals to adopt the innovation (Midgley & Dowling, 1978). The time concerns the timeline of adoption, starting at the initial gathering of knowledge towards either rejection or adoption of the innovation (Rogers, 1962). Furthermore, diffusion models focus on modeling the flow from the untapped market to the potential market to ultimately get customers to the current market (Mahajan & Muller, 1979). This flow of innovation diffusion is often captured in a curve. The innovation diffusion models are mostly first-purchase diffusion models, which entails that there are no repeat buyers and that each buyer will have a purchase volume of only one unit (Mahajan & Muller, 1979).

One of the first articles proposing a model for innovation diffusion was in the field of

sociology, which proposed that innovation diffusion was a normally distributed curve (Rogers, 1962). This curve divided consumers into different categories, ranging from innovators, early adopters, early majority, late majority to laggards. However, the problem with this model is that it assumes 100% adoption, which seems unlikely for marketing innovations, since the consumer lacks objective measuring criteria to assess whether the innovation is superior (Robertson, 1967). The most acknowledged first-purchase diffusion models in marketing are from Fourt and Woodlock (1960), Mansfield (1961) and Bass (1969). The first innovation diffusion model by Fourt and Woodlock (1960) proposed that innovation diffusion is an exponential curve based on the number of customers who have bought at time t. Furthermore, the Fourt and Woodlock model (1960) found the behavior mechanism of using mass-media communication to seek information about the innovation the driver of innovation diffusion (Mahajan, Muller & Bass, 1990). Contrary, Mansfield (1961) proposed a logistic curve for innovation diffusion and assumed that the behavior mechanism of using word-of-mouth communication to seek information about the innovation most important in innovation diffusion (Mahajan, Muller & Bass, 1990).

The most cited and well-known diffusion theory is by Frank M. Bass (1969), who extended the basic diffusion model with the two-step flow of communication, where mass-media influences individuals who are highly influential who in turn influence individuals less influential, who were influenced by word-of-mouth (Katz, 1957; Robertson 1967). In this study, the Bass model for

(12)

2.4 Bass model

As mentioned before, the Bass model of innovation diffusion is one of the most prominent innovation diffusion models, popular due to its simplicity (Bas, 1969; Bass, 2004). Bass innovation diffusion model (1969) combined the foregoing innovation diffusion models by Fourt and Woodlock (1960) and Mansfield (1961). Taking a step back, Fourt and Woodlock (1960) proposed that mass-media was the main driver of innovation diffusion, while Mansfield (1961) on the other hand proposed that word-of-mouth was the main driver of innovation diffusion. Bass (1969) combined the two, including both mass-media communication and word-of-mouth. Furthermore, Bass (1969) introduced an s-shaped curve of innovation diffusion, which is in-line with the Mansfield (1961) innovation diffusion model, but different from the Fourt and Woodlock model (1960), who proposed an exponential curve.

Bass (1969) proposed that the group of potential adopters could be segmented into two groups, who are either being influenced by mass-media communication or word-of-mouth. The potential customers who were being influenced by mass-media communication were called innovators, while the potential customers relying on word-of-mouth communication were imitators (Bass, 1969). Bass elaborated on these two types of behavior, mentioning that the core difference between the two relies on the buying influence. He said that imitators are influenced in their timing of adoption by the decisions of other members within their social system, while innovators are not. The adoption process of imitators can be seen as learning behavior, whereby they learn from others who have already adopted (Bass, 1969). The innovation parameter is called (p), and the imitation parameter is called (q). Besides, another parameter required for forecasting the diffusion of an innovation is the market potential, which is called (m) (Mahajan, Muller & Bass, 1990). Summarizing, the Bass model of innovation diffusion states that the diffusion of an innovation is highly dependent on the

innovativeness (p), imitativeness (q) and the potential market share (m) (Bass, 1969).

(13)

managements’ judgement, whereby they forecast the diffusion pattern based on intuitive interpretation (Bass, 2004). However, both of these ‘’guessing’’ manners for parameter estimation have found to be lacking, whereby the estimates often have huge discrepancy’s with ad hoc outcomes (Bass, 2004). Besides, ‘’guessing’’ parameter estimates for a fundamentally new product is even harder, since there is literally no historical data or past experience to base the estimates on.

Some studies have tried to overcome this estimation issue. Several studies have used external sources, like secondary data, management judgements and market surveys to estimate the parameter potential market share (m) (Heeler & Hustad, 1980; Lawrence & Lawton, 1981; Mahajan & Sharma, 1986; Teotia & Raju, 1986; Mesak & Mikhail, 1988). Massiani (2015) used a discrete choice model to predict the potential market share (m) within the Bass model. Besides, an article published by

Massiani and Gohs (2015) tried to find a solution on how to adequate choose estimates for the parameters used in the Bass model for forecasting purpose. Unfortunately, they found large

discrepancies between their reported estimates and ad hoc estimates, implying that there is no ultimate selection criterion for Bass’s parameters p, q and m (Massiani & Gohs, 2015). Anand, Bansal & Aggrawal (2018) used a technique similar to this study, namely using a conjoint analysis to measure the choice probability of a consumer purchasing a product, to forecast the potential market share (m). Furthermore, the Bass model was criticized for not having a behavioral description of the parameters innovators and imitators, giving limited attention to the mechanisms of the choice processes involved (Massiani & Gohs, 2015). Moreover, it remains unclear whether innovativeness and imitativeness are characteristics of an individual known up front, or if individuals are clustered afterwards (Massiani & Gohs, 2015).

This study will use a choice-based conjoint analysis on drone delivery to predict the potential market share (m) for drone delivery. The choice-based conjoint analysis will include measures of consumer innovativeness and consumer imitativeness, to estimate the innovativeness (p) and imitativeness (q) parameters. The next subchapters will elaborate on the concepts of consumer innovativeness, consumer imitativeness and the conjoint analysis as solutions for the wrong estimate parameters in the Bass model (1969).

2.5 Consumer innovativeness

The main goal of innovations is that the innovation will get adopted by customers, preferably as many as possible, so the innovation can diffuse through the whole market (Hauser, Tellis & Griffin, 2006). This study will use consumer innovativeness as input for the innovativeness (p) parameter in the Bass model (1969), because it has been found that a valid predictor for innovation adoption is the concept of consumer innovativeness (Gatignon & Robertson, 1985; Im, Bauys & Mason, 2003; Hauser, Tellis and Griffin, 2006; Im, Mason & Houston, 2007).

(14)

innovativeness as the degree to which an individual adopted an innovation earlier in time than other members of his social system (Rogers & Shoemaker, 1971). Important in the foregoing definition is the mention that adopting earlier in time is meant in terms of an actual timeline, rather than individuals perceiving that they adopted an innovation earlier in time (Rogers & Shoemaker, 1971). Crucial within the foregoing definition of consumer innovativeness is the relative time of adoption of the innovation. The basis for this focus on the relative time of adoption to explain consumer innovativeness are the adopter categories by Robertson (1967), which range from innovators to laggards based on their relative time of adoption of an innovation. Besides the basic definition before, consumer

innovativeness as a concept has been defined in terms of a personality trait and as actual behavior. First, several authors see consumer innovativeness as a trait possessed to some degree by every individual, which can be placed on a continuum (Midgley, 1977; Midgley & Dowling, 1978;

Goldsmith, 1995; Steenkamp, ter Hofstede, & Wedel, 1999; Steenkamp & Gielens, 2003). Consumer innovativeness in terms of a personality trait is called innate innovativeness or dispositional

innovativeness, and is defined as the predisposition of individuals to buy new and different products and brands, instead of remaining with previous choices and patterns, independently of the

communication information of others (Midgley, 1977; Midgley & Dowling, 1978; Steenkamp, ter Hofstede & Wedel, 1999; Steenkamp & Gielens, 2003). However, some authors discuss consumer innovativeness as a general trait, allowing for differences in consumer innovativeness between categories. Research suggests that these differences occur due to differences in interest in a product category, with interested consumers gathering more information and thereby knowledge about a certain product group. These interested consumers will be able to evaluate the value and attractiveness of an innovation, therefore having to exert less effort to decide whether to adopt or not (Gatignon & Robertson, 1985; Midgley & Dowling, 1987). This domain-specific innovativeness is being defined as an individual’s predisposition to adopt a new product or service within a specific domain of interest (Goldsmith & Hofacker, 1991; Roehrich, 2004).

The behavioral view of consumer innovativeness as actual innovative behavior expressed by an individual is called actualized innovativeness (Midley, 1977; Midgley & Dowling, 1978).

Actualized innovativeness is defined as the degree of interest in new ideas and making decisions concerning innovation adoption independently of the communication information of others (Midgley, 1977; Midgley & Dowling, 1978). Both of the foregoing definitions focus on the communication process, by which innovators adopt an innovation independent of their communication system. This is in line with the definition of innovators by Bass (1969) in terms of communication, who defines innovators as individuals who decide to adopt an innovation independently of other individuals in their social system. Bass (1969) adds that these innovative individuals can only be influenced by mass-media, which can be seen as external information.

(15)

can be done in three ways; by the time-of-adoption measure, cross-sectional method and self-reported measurement scales. None of these measuring strategies are found to be outperforming the others, considering they all have both strengths and weaknesses (Goldsmith & Hofacker, 1991). The time-of-adoption measure is based on the definition by Rogers and Shoemakers (1971) mentioned earlier, whereby they define consumer innovativeness as the degree to which an individual adopted an innovation earlier in time than other members of his social system. Since this definition is already in terms of measuring consumer innovativeness, it can be seen as an operational definition (Midgley & Dowling, 1978). Even though the time-of-adoption measure is fairly easy to measure, it has been criticized since time-of-adoption is a temporal concept that is being associated to the construct of innovativeness, but there is no actual one-to-one relationship between the two concepts (Midgley & Dowling, 1978). This unfolds in the time-of-adoption measure not being able to predict actual future behavior (Midgley & Dowling, 1978; Goldsmith & Hofacker, 1991). Another way consumer innovativeness can be measured is using the cross-sectional method, or in other terms looking at the ownership of new products of individuals or using new services (Robertson and Myers, 1969;

Summers, 1971; Darden & Reynolds, 1974; Baumgarten, 1975). The cross-sectional method measures innate innovativeness, which is consumer innovativeness in terms of a personality trait, which can be observed through the interaction with other personality traits, characteristics of the innovations and situational factors (Midgley & Dowling, 1978). However, this method does not improve the predictive validity and in addition would be challenging to both control as well as develop (Goldsmith &

Hofacker, 1991). Lastly, the self-reported measurement scales of consumer innovativeness, which were designed since they would be simple and easy to administer, the scales could be easily adapted towards several domains, and they can be used in a survey. A disadvantage is that self-reporting scales are found to be less suitable for unknown products or service domains, since consumers have trouble reporting on anticipated behavior which they never engaged in before (Goldsmith & Hofacker, 1991). The self-reporting scales can be divided into two groups, namely the life innovativeness scales and the adoptive innovativeness scales (Roehrich, 2004). Life innovativeness scales measure innovativeness at a general behavioral level, which is described as feeling an attraction to any kind of newness, not to the adoption of a specific product or service (Kirton, 1976; Roehrich, 2004). Adoptive innovativeness scales on the other hand focus on the tendency to adopt new products or services, which is appropriate for this study. Numerous scales have been developed to measure adoptive innovativeness (Teotia & Raju, 1986; Goldsmith and Hofacker, 1991; Baumgartner & Steenkamp, 1996; Steenkamp & Gielens, 2003), yet researchers have not agreed about a measurement scale. The adoptive innovativeness scales use different dimensions to measure innovativeness, with attraction to newness and speed of adoption as the two most popular dimensions used.

(16)

research has already tried to overcome the estimation issue for the innovativeness parameter (p) by including measures of innovativeness. Wong, Yap, Turner & Rexha (2011) combined the Bass model (1969) with domain-specific innovativeness (Gatignon & Robertson, 1985; Goldsmith & Hofacker, 1991) and the adoption categories (Rogers, 1962). Massiani and Gohs (2015) have tried to estimate the innovativeness parameter (p) by relating it to the potential market share (m) parameter. However, no article has found a way to estimate the innovativeness parameter (p) for a future product, that has not yet been launched on the marketplace, with no data on market size and peak times yet.

For this study, the self-reporting dispositional innovativeness scale by Steenkamp and Gielens (2003) will be used, which defines consumer innovativeness as a general consumer trait which has been found to have a positive effect on actual trial probability of innovations (Steenkamp, ter Hofstede & Wedel, 1999; Steenkamp & Gielens, 2003). The decision to use the dispositional innovativeness scale (Steenkamp & Gielens, 2003) is based on several reasons. First, dispositional innovativeness has been mostly found to have a significant positive effect on new product adoption and innovative behavior in several studies (Baumgartner & Steenkamp, 1996; Steenkamp & Burgess, 2002; Im, Bayus & Mason, 2003; Steenkamp & Gielens, 2003; Shannon & Mandhachitara, 2008), or a partly positive relationship (Venkatraman, 1991; Hirunyawipada & Pawan, 2006; Im, Mason & Houston, 2007). Even though the domain-specific scale shows a less ambiguous positive relationship in different domains with new product adoption and innovative than the dispositional innovativeness scale (Varma Citrin, Sprott, Silverman & Stem Jr, 2000; Hirunyawipada & Pawan, 2006), it would be difficult to construct a domain-specific scale for drone delivery service, since it is a novel and yet unknown domain for customers. Besides, it has been found that dispositional innovativeness correlates with domain-specific innovativeness (Goldsmith, Freiden & Eastman, 1995; Hirunyawipada & Pawan, 2006, Roehrich, 2004). Furthermore, dispositional innovativeness has been found to have a positive relationship with purchase intention (Jin & Gu Suh, 2005; Pallister, Wang & Focall, 2007).

Furthermore, dispositional innovativeness has been found to relate to both products and services, across a widespread range of categories (Im, Bayus & Mason, 2003; Steenkamp & Gielens, 2003; Im, Mason & Houston, 2007; Jin & Gu Suh, 2005; Pallister, Wang & Focall, 2007).

Summarizing, consumer innovativeness is a general consumer trait, possessed by every individual to some degree. Consumers with a high degree of innovativeness will adopt an innovation early in time, because they hold the predisposition to explore new and different products and brands, instead of remaining with previous choices. Furthermore, these high innovativeness consumers will make adoption decisions independent of their social system. This study will measure consumer innovativeness, which will uncover the innovativeness parameter (p) that will be used in the Bass model (1969), needed to predict the innovation diffusion of drone delivery service. The research questions that arises is: How to measure consumer innovativeness to accurately predict the diffusion of

(17)

2.6 Consumer imitativeness

Consumer imitativeness is another important parameter used to estimate innovation diffusion in the Bass Model (1969). Imitators are individuals being influenced in their timing of adoption by decisions of other members in their social system (Bass, 1969). Consumer imitativeness concerns the degree that individuals are being influenced by word-of-mouth, or in other terms internal

communication. An imitator will decide to buy a product or service if other people have bought that product or service already (Mahajan, Muller & Bass, 1990). This can be compared to a learning process, where imitators learn from the adoptive behavior of others (Bass, 1969).

A concept very closely related to consumer imitativeness is consumer susceptibility to

interpersonal influence, which arises from Deutsch and Gerard’s (1955) distinction between two types of interpersonal influence: normative and informational, which are both forms of social influence. Informational influence is based upon individuals accepting information from others, whereas normative influence concerns individuals conforming to the expectations of others (Deutsch & Gerards, 1955; Burnkrant & Cousineau, 1975). Informational social influence is being defined as ‘’an influence to accept information obtained from another as evidence about reality’’ (Deutsch & Gerards, 1955, 629). Normative social influence is defined as ‘’an influence to comply with the positive

expectations of another’’ (Deutsch & Gerards, 1955, 629). Research on social influence has further deconstructed normative influence into utilitarian and value-expressive influence (Burnkrant & Cousineau, 1975; Bearden & Etzel, 1982; Bearden, Netemeyer & Teel, 1989, 1990). Utilitarian social influence is a compliance process, which happens when individuals conform to the expectations of other when they want to avoid a punishment or receive a reward (Burnkrant & Cousineau, 1975; Bearden & Etzel, 1982; Bearden, Netemeyer & Teel, 1989, 1990). Value-expressive influence occur since individuals want to enhance their self-concept, by identifying themselves with positive

evaluative groups, and distancing themselves from negative evaluative groups through actions (Burnkrant & Cousineau, 1975; Park & Lessig, 1977; Bearden & Etzel, 1982).

(18)

are susceptible to interpersonal influence will perceive buying new products and services as a risk, which makes them less likely to try out a new product (Bearden & Etzel, 1982; Steenkamp & Gielens, 2003). This risk is perceived to be higher for more novel products and services, since the norms surrounding the new product or service have not yet been formed (Steenkamp, ter Hofstede & Wedel, 1999). Consumers susceptible to interpersonal influence rely on their social system and whether they accept or reject a new product. The process of innovators adopting a new product or not is crucial for consumers susceptible to interpersonal influence, since the word-of-mouth that spreads from the innovators to the consumers susceptible to interpersonal influence is what influences them in also adopting or not.

There are several scales measuring consumer susceptibility to interpersonal influence (Cacioppo & Petty, 1982; Swap & Rubin, 1983; Bearden, Netemeyer & Teel, 1989). The most used scale for measuring consumer susceptibility to interpersonal influence is the scale by Bearden, Netemeyer and Teel (1989). This scale is based upon the research by Deutsch and Gerard (1955) on normative and informational influence and the research by McGuire (1968)on influence. The susceptibility to interpersonal influence has been split into two parts, with 8 items measuring normative interpersonal influence, abbreviated as SNI, and 4 items measuring the informational interpersonal influence. This two-dimensional measure was found to be reliable and valid across several studies, both for the normative influence (Bearden, Netemeyer and Teel, 1989, 1990; Sen, Gürhan-Canli & Morwitz, 2001; Steenkamp & Gielens, 2003; Wooten & Reed II, 2004; Mangleburg, Doney & Bristol, 2004; de Jong, Steenkamp & Fox, 2007), as well as the informational influence (Mangleburg, Doney & Bristol, 2004; Chen, Teng, Yu & Yu, 2014; Bearden, Netemeyer and Teel, 1989, 1990).

Summarizing, this study tries to overcome the issue of having to guess estimates for the imitativeness parameter (q) in the Bass model (1969) to predict the innovation diffusion of a fundamentally new service, namely drone delivery service, by including measures for the consumer susceptibility to interpersonal influence. Since consumer imitativeness as defined by Bass (1969) lies very close to the consumer susceptibility to interpersonal influence scale by Bearden, Netemeyer & Teel (1989), it is assumed that the measurement scale will uncover the imitativeness parameter (q) needed to predict the innovation diffusion of drone delivery service. The research questions that arises is: How to measure consumer imitativeness to accurately predict the diffusion of drone delivery

service?

2.7 Conjoint analysis

(19)

products or services (Eggers et al, 2018). This preference measurement is done with the assumption that products and services are bundles of attributes, with these attributes having different levels (Orme, 2002; Eggers et al, 2018). An attribute is a characteristic of a product, and the levels are various degrees of that characteristic. These attributes will discover utilities that each attribute will add to the overall preference utility, by varying in the levels of the attributes (Orme, 2002). Conjoint analysis can be seen as a decompositional method, as the overall utilities for the product or service gets

decomposed into utilities for each attribute, also called part worth utilities (Green & Rao, 1971; Eggers et al, 2018). Conjoint analysis is also defined as a statistical technique where respondents express their preferences for combinations of attributes, which results in respondents’ utilities of the attributes (American Marketing Association, 2015). Conjoint analysis will uncover a preference utility function, formed out of the attribute bundle preferences of respondents (Eggers et al, 2018). The preference utility is the function of all of the part-worth utilities (Wiley & Bushnell, 1979). Consumers will choose the offering that will maximize their utility (Jun & Park, 1999; Eggers et al, 2018). In other terms, the utility function of drone delivery service will show the specific characteristics customers prefer for drone delivery services (Eggers & Sattler, 2011).

Conducting a conjoint analysis can be done by either ranking, rating or choice-based methods, which differ in forming the overall utility function (Eggers et al, 2018). The first two, ranking and rating conjoint analysis, are the traditional approaches first described in literature which use experimentally controlled combinations of attribute levels (Green & Rao, 1971; Chrzan & Orme, 2000). Choice-based conjoint analysis uses sets of attributes and choices (Louviere and Woodworth, 1983; Louviere, 1988; Chrzan & Orme, 2000; Eggers et al, 2018). In this study, the choice-based conjoint analysis will be used, since it is the most popular one due to its advantages over ranking and rating methods (Louviere and Woodworth, 1983; Louviere, 1988; Orme, 2016). The first advantage of choice-based conjoint analysis is that it simulates actual purchase decisions, especially when a no-choice option is included, simulating realistic no-choice behavior and predicting behavior in the marketplace (Elrod, Louviere & Davey, 1992; Desarbo, Ramaswamy & Cohen, 1995; Eggers et al, 2018). This higher level of realistic marketplace behavior equals a higher validity (Eggers et al, 2018). The no-choice option will make sure that choosing an option will not be forced upon the respondent, which will reduce respondents opting out of making a decision (Eggers et al, 2018). Another

advantage is that product attributes and levels can easily be accommodated within the choice-based conjoint analysis (Desarbo, Ramaswamy & Cohen, 1995). Besides, choice-based conjoint analysis requires a higher complexity of deducting the experiment, making it possible to estimate more than only main effects (Chrzan & Orme, 2000). Furthermore, and most important for this study, the

(20)

will spread under different levels of attributes.

Summarizing, a choice-based conjoint analysis can be used to uncover the preferences of individuals towards tangible and intangible products and services. This can be done by decomposing a service into attributes and levels. This study will use a choice-based conjoint analysis to uncover the preferences of individuals on drone delivery service and overcome the issue of having to guess estimates for the market share parameter (m) in the Bass model (1969). Furthermore, the choice-based conjoint analysis enables analyzing preferences of individuals on drone delivery service, for instance willingness-to-pay, attribute importance and latent-class analysis.

However, to determine what will have an influence on the preferences of individuals for drone delivery service, attributes need to be determined. The utility function for drone delivery service will consist of three different attributes, namely delivery brand, delivery price and delivery time. The first attribute is delivery brand, since several brands like DHL, Amazon and Google have already

announced their future interest in drone delivery service services (Rose, 2013; Bryan, 2014; Madrigal, 2014). The model used for the delivery brand will be a part worth model, since each level of the attribute delivery brand will need to be estimated separately, since it cannot be assumed that there is a functional relationship between brands (Eggers et al, 2018). Secondly, it was found that even though consumers are willing to pay significant price premiums for same-day delivery services, many are reluctant to pay even higher costs for even quicker delivery (Joerss et al, 2016). This shows that consumers find delivery time fundamental in choosing their delivery service, while still staying highly price sensitive (Joerss et al, 2016). Accordingly, the second and third attribute for drone delivery service are delivery time and delivery price. Since delivery time and delivery price are assumed to have a proportional positive (negative) effect on the utility function when the attribute level will be increased (decreased), a vector model is appropriate (Eggers et al, 2018). Lastly, next to the three attributes brand, price and time, a no-choice option will be added to the choice-based conjoint to heighten the level of realism for the respondents (Wlömert & Eggers, 2014; Eggers & Sattler, 2011). After respondents have determined their preferred alternative, they will have the option to either adopt the alternative that they choose, or not adopt. This study will include the no-choice option by making respondents choose between adopting the chosen drone delivery service alternative, or not adopting and preferring to persist with traditional delivery services. This study assumes that the majority of consumers are willing to adopt drone delivery services when drone delivery services offers the right features and has benefits over traditional drone delivery services. Besides, it was found that out of all future autonomous driving delivery vehicles, already 35 percent of customers prefer drones over for instance autonomous guided vehicles, which are portable robots. Furthermore, it was found that 60% stands neutral towards drone delivery services (Joerss et al, 2016).

(21)

H1: The attribute delivery time will negatively affect the preference utility of drone delivery service. H2: The attribute delivery price will negatively affect the preference utility of drone delivery service. H3: The attribute delivery brand will affect the preference utility of drone delivery service.

H4: The no-choice option will have a positive effect on the preference utility of drone delivery service.

Besides determining the preference utility function, as mentioned before, the choice-based conjoint analysis will also include measures of consumer innovativeness and consumer imitativeness. Since consumer innovativeness and consumer imitativeness are both consumer traits, it is assumed that they will have an influence on how the attributes affect the preference utility for drone delivery

service. For instance, research has shown that consumers high on innovativeness are less risk-averse, flexibility, independent, impulsive and novelty-seekers, it is expected that these customers feel a stronger urge for experiencing new products and services (Venkatraman & Price, 1990; Steenkamp & Baumgartner, 1992). Consumer susceptibility to interpersonal influence is found to be related to a lower self-esteem and lower confidence, and higher self-monitoring (Bearden, Netemeyer & Teel, 1990). This suggests that these two traits, innovativeness and imitativeness, can be seen as opposites, with the way they influence the preference utility being opposing as well.

Consumers high on innovation are feeling a stronger urge to engage with new products and services. Furthermore, they are higher on impulse buying behavior. Besides, a higher price might indicate a novel product. On the other hand, consumers high on imitativeness express more self-monitoring, which shows lower levels of impulse buying due to controlling behavior. This could indicate that innovative consumers will pay higher price for innovations, due to the strong urge they feel combined with impulsive buying behaviors, while imitators are assumed to monitor their behavior more, perceive a higher risk and be less confident in making the right decision, preferring a lower price to adopt a new service. This leads to the following hypotheses:

H5a: Innovativeness will have a positive moderating effect on delivery price, where innovative consumers prefer to pay a higher price compared to imitative consumers

H6a: Innovativeness will have a negative moderating effect on delivery price, where innovative consumers prefer to pay a lower price compared to innovative consumers

Furthermore, it is assumed that innovative consumers prefer a more innovative brand, since it is in line with their self-concept. On the other hand, imitative consumers are assumed to prefer less innovative brands for the same reason as stated before, since it will be in line with their self-concept. This leads to the following hypotheses:

H5b: Innovativeness will have a moderating effect on brand, with innovative consumers preferring an innovative brand like Google and Amazon.

(22)

Additionally, the last attribute is delivery time. This study will not make any assumption on the differences between innovators and imitators on their preference for delivery time, since it cannot be related to past research. However, it is still assumed that these traits will lead to differences in preferences, since they are opposing traits. This leads to the following hypotheses:

H5c: Innovativeness will moderate the effect of the attribute delivery time on the preference utility of drone delivery service

H6c: Imitativeness will moderate the effect of the attribute delivery time on the preference utility of drone delivery service

Lastly, the no choice option will present respondents with the option to not adopt their

preferred alternative, enabling them to stay with traditional delivery services rather than drone delivery services. The hypothesis concerning the no choice option will be based upon the adoption categories of Rogers (1962), where innovators are the first group adopting a new technology or innovation. Moreover, consumer innovativeness can be defined as the degree to which the consumer has adopted the innovation earlier in time than others (Rogers and Shoemakers, 1971). Therefore, it is expected that innovativeness will have a positive moderating effect on the no-choice option, with innovative consumers significantly more deciding to adopt drone delivery services compared to imitative consumers. On the other hand, it is expected that imitativeness will have a negative moderating effect on the no-choice option, with imitative consumers significantly more deciding to opt for traditional delivery services. This leads to the following hypothesis:

H5d: Innovativeness will have a positive moderating effect on the no choice option, where innovativeness will increase the preference for adopting drone delivery services

H6d: Imitativeness will have a negative moderating effect on the no choice option, where imitativeness will decrease the preference for adopting drone delivery services

2.8 Synthesis and Conceptual Model

The main research question that this study will tackle is: How can the diffusion of an

innovative future service be predicted using the Bass diffusion model?

This study will try to answer this question based on using a fundamentally new and disrupting service, namely drone delivery service. A choice-based conjoint analysis will unveil the preference utility for drone delivery service, which can be used to estimate the potential market share (m) parameter in the Bass model (1969) for drone delivery service. Measures of consumer disruptive innovativeness will be able to estimate the consumer innovativeness parameter (q) in the Bass model (1969) for drone

(23)

service. Furthermore, this study assumes that the consumer traits, consumer innovativeness and consumer imitativeness, have an influence on the preferences for drone delivery service. These two traits are assumed to moderate the effect of the attributes on the preference utility. The conceptual model is presented below in Figure 1.

Figure 1: Conceptual Model

(24)

3. METHODOLOGY

This chapter will present the measures and methods used, the study design and how the data was collected. First, the measures of consumer innovativeness and consumer imitativeness are

presented, as well as the reasons why those measures are used. Secondly, the chapter will elaborate on the choice-based conjoint, the attributes and the levels. Thirdly, the choice design will be presented. Lastly, the estimation procedure will explain how the data will be analyzed.

3.1 Measures of innovativeness and imitativeness

(25)

3.2 Choice based conjoint analysis

Subsequent to the self-reporting measurement scales, respondents will be presented with an introduction text which introduces the topic of drone delivery service (Appendix A). This introduction will offer respondents basic information and knowledge concerning drone delivery service, enabling respondents to contemplate on the subject. Afterwards, respondents will be presented with the choice based conjoint analysis. The following two paragraphs will go into detail concerning the choice-based conjoint analysis.

3.2.1 Attributes & Levels.

When deciding on which attributes to include in the choice-based conjoint analysis, it is important to keep in mind that attributes should be relevant, discriminate, manageable and not interrelated (Orme, 2002; Eggers et al, 2018). Subsequently, when determining the attribute-levels, it is important to cover more levels than the full range of possibilities, have clear descriptions, be generally acceptable and be mutually exclusive. Attributes should have between three to four levels, whereby the linearity or non-linearity of the utility function should be kept in mind (Eggers et al, 2018).

The attributes for this study are delivery brand, delivery price and delivery time. The

corresponding levels were chosen based upon an extensive online study, and are presented in table 1.

Dispositional innovativeness (Steenkamp & Gielens, 2003)

1) When I see a new product on the shelf, I’m reluctant to give it a try. *

2) In general, I am among the first to buy new products when they appear on the market. 3) If I like a brand, I rarely switch from it just to try something new. *

4) I am very cautious in trying new and different products. * 5) I am usually among the first to try new brands.

6) I rarely buy brands about which I am uncertain how they will perform. * 7) I enjoy taking chances in buying new products.

8) I do not like to buy a new product before other people do. *

Consumer susceptibility to interpersonal influence (Bearden, Netemeyer & Teel, 1989)

1) If I want to be like someone, I often try to buy the same brands that they buy. 2) It is important that others like the products and brands I buy.

3) I rarely purchase the latest fashion styles until I am sure my friends approve of them. 4) I often identify with other people by purchasing the same products and brands they purchase. 5) When buying products, I generally purchase those brands that I think others will approve of. 6) I like to know what brands and products make good impressions on others.

7) If other people can see me using a product, I often purchase the brand they expect me to buy. 8) I achieve a sense of belonging by purchasing the same products and brands that others purchase. 9) To make sure I buy the right product or brand, I often observe what others are buying and using. 10) If I have little experience with a product, I often ask my friends about the product.

(26)

Table 1. Attributes and levels for drone delivery service

3.2.2 Choice Design.

This study will apply a randomized fractional factorial experimental design, whereby each respondent is being exposed to only a subset of stimuli. The full factorial design on the other hand will expose respondents to all sets of all potential stimuli, which is every combination of all attribute levels (Wlömert & Eggers, 2014; Eggers & Sattler, 2011). However, since exposing respondents to all combinations will be nearly impossible, this is not an effective design. Besides, the fractional factorial experimental design will still allow for main effect estimations, when the design is both balanced and orthogonal. A balanced design ensures that each level will be displayed an equal number of times, while an orthogonal design will ensure that each level combination will be displayed an equal number of times. Respondents were presented with 12 choice tasks, with 3 alternatives per choice set and a no-choice option. The no-no-choice option is included to increase the realism of the study (Wlömert & Eggers, 2014; Eggers & Sattler, 2011). Due to the no-choice option, the decision is not forced upon respondents. Besides, when adding a no-choice option to a study which has price as an attribute, the willingness-to-pay can be calculates (Eggers & Sattler, 2009). A dual-response no-choice option will be used instead of having the no-choice option as a separate alternative, since it will acquire more information (Appendix C) (Wlömert & Eggers, 2014). Mypreferencelab, a software tool created by Felix Eggers, will be used to create the choice-based conjoint analysis and the choice design. The software package makes sure that the design is balanced, orthogonal and that respondents will be randomly allocated to the choice sets.

3.3 Plan of Data Collection

In order to gather data on consumer preferences for drone delivery services, a survey has been created in Preference Lab. The survey starts out with an introductory text, to introduce the respondent to the subject of drone delivery services. Thereafter, the respondent is asked to fill in some

demographic questions, followed by two self-reported measuring 1-7 Likert scales, which will measure consumer innovativeness and consumer imitativeness. Lastly, the respondent will be

presented with the choice-based conjoint analysis, which consists of 12 choice sets. The survey set-up and questions can be found in the Appendix (A-C). The survey was distributed through Amazon

Attribute Description Level 1 Level 2 Level 3 Level 4 Brand The brand of the drone delivery service

service operator.

Amazon DHL Google UPS

Price per delivery

The price consumers will have to pay to use the drone delivery service per order

$5 $10 $15 $20

Time of delivery

The time (in minutes) consumers want their package to be delivered by a drone delivery service

Within 15 minutes Within 30 minutes Within 1 hour

(27)

Mechanical Turk, offering respondents $0.20 for a completely filled in survey. As a control check, respondents had to fill in their Amazon Worker ID at the end of the survey. This control check controlled for 1) respondents completely filling in the survey, 2) respondents reading through instructions carefully.

3.4 Plan of Analysis

The focus of this study is the Bass diffusion model (1969), where the number of adopters at time t can be measured by including estimates for innovativeness (p), imitativeness (q) and potential market size (m) in the basic Bass formula below.

! ! = ! + ! ∗ !

The first step that needs to be taken is rescaling the seven-point Likert-scale measurements of the consumer innovativeness (p) and consumer imitativeness (q), so they can be used as parameters inputs in the Bass diffusion model (1969). This will be done by rescaling the scale from a 1-7 Likert scale to a 0.004 – 0.039 scale for innovativeness, and a 0.25-1.76 scale for imitativeness (Sultan, Farley & Lehmann, 1990). However, for the estimation of the Bass diffusion model (1969), the potential market share (m) needs to be estimated. This can be done based upon a market scenario simulation, which can be estimated using the utilities for attribute levels. The market scenario will consist of 2 potential products and 1 outside option, which is the none option, where consumers prefer to stick with

traditional delivery services instead of opting for the drone delivery service. The market share (m) for option (i) out of (S) alternatives can be calculated using:

! ! = exp !! exp! !! !∈!

!

with:

p = probability of choosing service option i out of S alternatives

i = service choice

S = all alternatives within the choice set

j = all choices

v = combined utility function

This study will use two methods to estimate the preference utilities for drone delivery service, namely: an aggregate multinomial logit model (MNL) and the Hierarchical-Bayes (HB) estimation method.

3.4.1 Aggregated Multinomial Logit

(28)

importance, willingness-to-pay and lastly the market share by means of the market simulation scenario (Eggers, Sattler, Teichers & Völckner, 2018). The formula for the MNL model, including only main effects is:

! = !!+ !!+ !!+ !!

with:

U = preference utility of drone delivery service

!_! = part worth utility of delivery time

!! = part worth utility of delivery price

!_! = part worth utility of delivery brand

The MNL model above can estimate the aggregated preference utility for drone delivery service, assuming that all attributes have a part worth utility. Even though the literature review assumed that delivery time and delivery price will estimate better when having a vector specification, the model starts out with a part worth utility for every attribute. This will enable the calculation of the relative attribute importance. Thereafter, the functional form of the MNL will be explored.

Nevertheless, before being able to estimate the full utility function, the utility functions for each attribute need to be estimated. This can be done with the formula’s below:

!_!= !!_!"!!_!"

with:

!! = partworth utility for delivery time for drone delivery service

βtm = utility vector for delivery time for level m for drone delivery service Xtm = numeric value of level m of attribute delivery time in drone delivery

service

m = levels (1, 2, 3, 4) corresponding to 15, 30, 45 or 60 minutes

!! = !!"!!!"!

with:

!_! = partworth utility for delivery price for commercial drone delivery βpm = utility vector for delivery price for level m for drone delivery service Xpm = numeric value of level m for delivery price in drone delivery

m = levels (1, 2, 3, 4) corresponding to $5, $10, $15 or $20 price

!

!!!!= !!!!!!!"!!"!

with:

!! = partworth utility for delivery brand for commercial drone delivery

(29)

Xbm = effect coding variable with value level m for drone delivery service

of delivery brand, otherwise 0

m = levels (1, 2, 3) corresponding to UPS, FedEx, DHL & Google,

using effect-coding

After estimating the aggregated part-worth utilities, the attribute importance of each attribute will be calculated. This will reveal the attribute importance !! in a percentage, relative to other attributes. The attribute importance will be calculated using:

!_!= max !! − min !! max !_! − min !_! ! !!! !! with:

!! = attribute importance of attribute n

!! = parthworth utility for attribute n

!! = parthworth utility for all attributes

Since this study includes a price attribute, the willingness-to-pay can be calculated, which is able to analyze the difference in utility when the price changes, and relate that change to an attribute level. To estimate the willingness-to-pay, the attribute requires the use of a vector model, or in other terms, linearity. It will be tested whether price is linear, to subsequently express non-price attributes in monetary terms. Besides, the attribute time can also be linear, which enables calculating the

willingness-to-time, to express non-time attributes in durational terms. The willingness-to-pay and willingness-to time will be calculated using:

WTP!"= β_!"

β! !

with:

WTP/T!" = willingness-to-pay/time for level m of attribute n

β_!" = parthworth utility for level m of attribute n

β!/! = utility vector for the price/time attribute

(30)

3.4.2 Hierarchical Bayes

The hierarchical Bayes (HB) model will be able to predict preference utilities for each individual. By not aggregating the data, HB includes consumer heterogeneity into the model (Orme, 2000). This study will include a HB model, because it has been found that a conjoint analysis can profoundly improve from the use of HB estimation (Orme, 2000; Johnson, 2000). The HB model has two levels, hence it is called hierarchical, with the normal distribution of utility preferences across all customers as a high level and the probability of an individual to achieve a certain outcome based upon their individual utility preferences estimated by the MNL model at the lower level (Rossi & Allenby, 1993; Orme, 2000; Arora & Huber, 2001). The HB iteration process will produce a posterior

distribution of the individual level utility parameters, based upon the differences between the individual’s preferences and preferences from other individuals in the sample data (Orme, 2000; Howell, 2009). A fitted HB model will depend mostly on the choices of the individual, while a poorly fitted model will depend more on all respondents. Bayes’ rule is the probability estimation of HB, which is based upon obtaining posterior probabilities by using likelihoods from the dataset to update the prior probabilities (Johnson, 2000). In practical applications, the Bayes’ rule can produce posterior probabilities, which are proportional to the likelihoods times the priors, by updating the prior

probabilities with the likelihoods obtained from the dataset used. This can be calculated using:

!"#$ ! ! ∞!prob Y β ∗ !prob(β)

!! The formula above will be done for several thousand of iterations. Furthermore, for this study, the Monte Carlo Markov Chain (MCMC) will be used, which will result in a burn in and thousands of draws per respondents, which will be transformed into the point estimate, which is the average of each respondent’s draws (Orme, 2000; Johnson, 2000). Estimating individual utilities will not only improve the conjoint analysis, it will also allow for easy clustering and segmentation, based upon the

differences in preferences between individuals (Howell, 2009).

3.4.2.1 Clustering.

(31)

4. RESULTS

This chapter will report the results of this study in the following manner; First, the characteristics of the data will be analyzed. Subsequently, a factor analysis will report upon the innovativeness and imitativeness scale validity. Afterwards, the results of the aggregate logit model and hierarchical Bayes model will be reported on. Then, market shares will be predicted for both the aggregate logit and the hierarchical Bayes model. Lastly, the Bass diffusion model will be estimated based upon the innovativeness (p), imitativeness (q) and market share (m) parameters.

4.1 Sample characteristics

The data for this study were collected on the 13th of April 2019 by posting the survey on Amazon Mechanical Turk. A total of 306 respondents completely filled in the survey. Respondents had to fill in their Amazon Worker ID to verify their ‘’identity’’, which was used confidentially to check whether respondents might have filled in the survey more than once. Data exploration uncovered that 10 respondents filled in the survey twice, and 1 respondent even filled in the survey three times. Their second and third times were deleted from the dataset, leaving the dataset with a total of 294 completed cases, with 10584 filled in choice sets. The respondents age ranged from 18-74, with an average of 33.7 years and a standard deviation of 11.6 years. Respondents spend an average of 5 minutes and 15 seconds on completing the survey, with a standard deviation of 3 minutes and 48 seconds. A boxplot uncovered that none of the respondents fell without the interquartile range, which is why no respondents were deleted based upon their average time of filling in the survey.

Respondents nationality’s range from North-American (51.7%), Indian (34.7%) or European (8.8%), with a few other (4.8%). The education of the respondents is quite high, with 67.2% having at least a Bachelor’s degree or higher. Table 1 provides an overview of the demographics of the dataset.

Frequency Percentage Gender Male (0) 177 60% Female (1) 117 40% Nationality North-American 152 51.7% Indian 102 34.7% European 26 8.8% Other (Nigerian, Asian, Jamaican,

Brazilian, White)

14 4.8%

Education

Less than a high school degree (0) 3 1% High school degree (1) 67 22.8%

Predicting the Bass diffusion model for drone delivery services: using measures of consumer innovativeness and consumer imitativeness in a choice-based conjoint analysis.

Predicting the Bass diffusion model for drone delivery services: using measures of

consumer innovativeness and consumer imitativeness in a choice-based conjoint

analysis.

by

Predicting the Bass diffusion model for drone delivery services: using measures of

consumer innovativeness and consumer imitativeness in a choice-based conjoint

analysis.

by

Annemijn Maria Sophia Geertje Steinmeijer

June 16, 2019

Table of Contents