HOW COMPLEXITY OF PRODUCTS AFFECTS
PREFERENCES OF INNOVATIVE CONSUMERS
IN A MIXED-BUNDLING SCENARIO
IMPROVING DIFFUSION OF NEW PRODUCTS THROUGH
PRODUCT BUNDLE OPTIMIZATION: A MENU-BASED CONJOINT
STUDY
BY
2
HOW COMPLEXITY OF PRODUCTS AFFECTS
PREFERENCES OF INNOVATIVE CONSUMERS
IN A MIXED-BUNDLING SCENARIO
IMPROVING DIFFUSION OF NEW PRODUCTS THROUGH
PRODUCT BUNDLE OPTIMIZATION: A MENU-BASED CONJOINT
STUDY
BY GERBEN TEMPELMAN Student number 2721775 Master thesis January 14, 2017 University of Groningen Faculty of Business and EconomicsDepartment of Marketing MSc Marketing Intelligence
Author: 1st supervisor 2nd supervisor
Gerben Tempelman Dr. F. Eggers Dr. K. Dehmamy g.tempelman@student.rug.nl f.eggers@rug.nl k.dehmamy@rug.nl Aquamarijnstraat 515 Nettelbosje 2 Nettelbosje 2
Duisenberg Building Duisenberg Building
DUI321 DUI
3
PREFACE
This master thesis was written to complete the Marketing Intelligence master program (Master of Science) of the University of Groningen. I have many debts of gratitude. To Sawtooth Software for granting me to use their Menu-Based Conjoint (MBC) software to analyze the data, which accelerated what would otherwise have been an analytical process of glacial pace; to the Sawtooth Software team that has provided me with excellent technical support while designing the study and analyzing the data; to my friends and family for all their encouragement. Above all, I would like to take the opportunity to express my profound gratitude to my supervisor Dr. Felix Eggers for all his guidance, support, and useful feedback.
Gerben Tempelman
ABSTRACT
This study examined if manipulating product complexity in a mixed-bundling scenario influences preferences of innovative consumers (and less innovative
consumers) in order to stimulate the diffusion of an innovation. More specifically, the objective of this study was to determine if consumers that are less innovative could be persuaded to adopt an innovation as part of a product bundle by manipulating the complexity of the products in those bundles. This is relevant since it provides
guidance for firms and practitioners in designing preference-based product bundles as part of a mixed-bundling strategy, to target innovative consumers and accelerate the diffusion of new products. A Menu-Based Conjoint (MBC) analysis was performed to estimate preferences of respondents varying in innovativeness. The results of the choice models show that none of the hypothesized effects were fully supported. Some significant differences in preference were found amongst respondents varying in innovativeness. However, these results cannot be generalized to the domain of
4
TABLE OF CONTENTS
PREFACE ... 3 ABSTRACT ... 3 1. INTRODUCTION ... 5 2. THEORETICAL FRAMEWORK ... 82.1 Product Bundling Strategies ... 8
2.2 Innovation Diffusion ... 11
2.3 Consumer Innovativeness ... 13
2.4 Synthesis, Hypotheses, and Conceptual Model ... 15
3. METHODOLOGY ... 18
3.1 Research Method: MBC ... 18
3.2 Data Collection ... 19
3.2.1 Selection of Stimuli ... 19
3.2.2 Measurement Scales ... 21
3.2.3 Sample and Survey Distribution ... 21
3.3 Study Design ... 22
3.4 Data Analysis and Hypothesis Testing ... 25
3.4.1 Plan of Analysis ... 25
3.4.2 Hypothesis Testing ... 26
4. RESULTS ... 28
4.1 Sample Results ... 28
4.2 Factor Analysis Results ... 29
4.3 Aggregate Logit and HB Results ... 30
4.4 Validation Results ... 31 4.5 Hypotheses Testing ... 32 5. DISCUSSION ... 34 5.1 Discussion ... 34 5.2 Summary ... 36 6. REFERENCES ... 38
APPENDIX A: PRE-TEST FOR STIMULI SELECTION ... 44
APPENDIX B: MEASUREMENT SCALES ... 45
APPENDIX C: STUDY DESIGN ... 47
APPENDIX D: R-SCRIPTS ... 51
APPENDIX E: FACTOR ANALYSIS RESULTS ... 64
APPENDIX F: ESTIMATION RESULTS AGGREGATE LOGIT ... 66
APPENDIX G: ESTIMATION RESULTS HB ... 70
APPENDIX H: ESTIMATION RESULTS REGRESSION ... 74
APPENDIX I: PRESENTATION DEFENSE ... 76
5
1. INTRODUCTION
A common marketing strategy employed by firms is to bundle two or more products/services in a single package for a single price (Guiltinan, 1987). A product is bundled with either related/complementary products, or unrelated/cross-categorical products, and offered as a pre-specified product bundle at a price discount
(Stremersch & Tellis, 2002). This benefits the consumer by saving money on products that the consumer might otherwise purchase separately (à la carte). When bundling products, a firm can adopt a pure-bundling strategy (only offer the products within a product bundle), a mixed-bundling strategy (offer the products separately as well as within a product bundle), or to adopt a non-bundling strategy (only offer the products separately; Guiltinan, 1987; Stremersch & Tellis, 2002). Although the optimality of a specific bundling strategy depends on many factors, mixed-bundling is considered a robust choice in many situations, and is often employed in practice (Guiltinan, 1987; Schmalensee, 1984; Stremersch & Tellis, 2002; Venkatesh & Kamakura, 2003).
6 Research on how preferences for bundled products relative to separate
products can be influenced is scarce (Harris & Blair, 2006a, 2006b; Stremersch & Tellis, 2002). Furthermore, how innovative consumers evaluate product bundles remains to be studied. If innovative consumers have distinct preferences, product bundles can be optimized so that new products can be diffused more effectively. The challenging aspect is therefore to determine if innovative consumers prefer certain types of products more than less innovative consumers, and if these types of products are still preferred when bundled. Therefore, the main research question of this thesis is defined as: “How does bundling of complementary products that vary in complexity affect preferences of innovative consumers for an innovative product?”.
A Menu-Based Conjoint (MBC) analysis has been conducted in order to examine how respondents varying in consumer innovativeness adopt an innovative high-tech product in a mixed-bundling scenario. From a methodological perspective, the use of MBC for bundling is new and interesting. MBC differs from the industry-standard Choice-Based Conjoint (CBC). CBC only observes choice for different combinations of attribute levels (i.e. choices among pre-configured products). MBC can also estimate preferences for individual attributes of which this product (or in this study: a bundle) is comprised, since MBC observes choice for each attribute
individually rather than for the composition of attributes as a whole. This is
interesting since it allows to estimate preferences for attributes (in bundling context: products) that vary in price among choice tasks, to examine both linear and non-linear price effects for each attribute, and estimate substitution and complementary effects amongst attributes. Studying menu-tasks using MBC enables practitioners and firms to optimize a customizable product or service (or a bundle of products). MBC was designed specifically to study preferences in multi-check menus, like in a mixed-bundling scenario. MBC is therefore a perfect solution to study mixed-mixed-bundling, as respondents can purchase pre-configured bundles or items à la carte.
7 consumers. The results of this study have relevant implications regarding the general problem of how to target innovative consumers via product bundling strategies to improve diffusion of innovations. This study provides a unique contribution by focusing on the intersection of innovation diffusion, consumer innovativeness and product bundle preferences by demonstrating a relatively new conjoint method. This is relevant since it provides guidance for firms and practitioners in designing
preference-based bundles as part of a mixed-bundling strategy, to target innovative consumers and accelerate the diffusion of new products. If innovative consumers have distinct preferences regarding product complexity, firms are able to manipulate
complexity in a mixed-bundling scenario, so that the offering appeals to innovative consumers in the introduction phase to improve innovation diffusion.
This thesis is structured in the following manner: chapter 2 reviews the literature regarding product bundling, innovation diffusion, and consumer
8
2. THEORETICAL FRAMEWORK
As highlighted in the introduction, this study focuses on product bundling, innovation diffusion, and consumer innovativeness. In this chapter, a theoretical overview is presented of these subjects. To finalize this chapter, a synthesis of the subjects is provided along with a conceptual framework that visualizes the
hypothesized effects.
2.1 Product Bundling Strategies
The definition of product bundling by Guiltinan (1987: 74) is often used in the literature on bundling: “the practice of marketing two or more products and/or
services in a single "package" for a special price”. In accordance with the definition by Stremersch and Tellis (2002), this thesis assumes that separate markets exist for all products that are bundled. Designing bundles that appeal to consumers depends on understanding how consumer preferences are structured (Janiszewski & Cunha, 2004). A common assumption in bundling is that of “strict additivity” (Guiltinan 1987). It assumes that a consumer’s reservation price for the product bundle equals the sum of the reservation prices of the individual products (Schmalensee, 1984; Venkatesh & Kamakura, 2003). Since bundling involves offering products at a “special price”, bundling strategies are regularly considered pricing strategies and are employed as a device for price discrimination (Stremersch & Tellis, 2002; Venkatesh & Kamakura, 2003). This special price is often the product of a discount that is given over the sum of the prices of the individual products in the bundle.
9 research suggests that a mixed-bundling strategy is in general slightly superior to a pure-bundling strategy or a non-bundling strategy (Stremersch & Tellis, 2002), this study will focus on bundle evaluations in a mixed-bundling context.
Product bundling has been studied from numerous perspectives and in many fields, such as marketing, economics, and behavioral psychology. For instance, based on economic theory, research has focused on optimality of product bundles and optimal pricing (Hansen & Martin, 1987; Venkatesh & Mahajan, 1993). However, studies that focus on consumer preferences and how consumers evaluate product bundle attractiveness have taken a more behavioral perspective. These studies suggest that the consumer’s evaluation of a product bundle is sensitive to the way information on price and the price discount is framed (Janiszewski & Cunha, 2004; Johnson, Herrmann, & Bauer, 1999; Soman & Gourville, 2001; Yadav, 1994, 1995; Yadav & Monroe, 1993). Numerous studies have endeavored to examine how the attractiveness of a bundle can be increased by manipulating information on the price of the bundle and on the bundle discount (Janiszewski & Cunha, 2004; Johnson et al., 1999; Khan & Dhar, 2010; Mazumdar & Jun, 1993; Stremersch & Tellis, 2002; Yadav & Monroe, 1993). In bundling, a price can be presented for each individual product in the bundle, or for the bundle as a whole. Also, a price discount can be distributed among each individual product in the bundle, framed on one specific item, or presented as a saving on the bundle as a whole (Janiszweski & Cunha, 2004; Khan & Dhar, 2010). More specifically, researchers examined how the segregation (versus integration) of both prices and discounts affect the attractiveness of bundles. The foundation of most of these studies is provided by principles from the field of behavioral research, namely mental accounting principles (Thaler, 1985) and the prospect theory value function (Kahneman & Tversky, 1979).
10 al., 1999, Morwitz, Greenleaf, & Johnson, 1998). These inconsistencies indicate that consumers are sensitive to the way in which information on price and discount is presented in bundles, and have a large influence on the extent to which consumers use mental accounting principles to determine product bundle attractiveness (Chakravarti et al., 2002; Heath, Chatterjee, & France, 1997; Johnson et al., 1999). For example, Heath et al. (1997) found that percentage-based discounts cause consumers to rely less on mental accounting principles. Also, framing the discount on one specific item (rather than equally distributing the discount over all items in the bundle) can increase the attractiveness of the entire product bundle, if framed on the item that is perceived as most important by the consumer (Khan & Dhar, 2010, Yadav 1994, 1995). This prediction is conceptualized in the Weighted Additive Model (Yadav 1994, 1995). This model assumes that consumers weight the evaluations of each product in the product bundle differently (Janiszewski & Cunha, 2004; Mazumdar & Jun 1993; Yadav, 1995). Furthermore, research provides evidence that consumers often edit the product bundles that are presented to them in a flexible manner (Chakravarti et al., 2002; Thaler & Johnson et al., 1999). As Chakravarti et al. (2002) mentioned: “even if prices are partitioned by component, consumers can easily add them to determine the total price of the bundle and then evaluate the associated loss. Such editing leaves the mental account on the price (loss) side identical, regardless of whether the
presentation is partitioned or consolidated” (Chakravarti et al., 2002: 217).
11 2003). Not only knowledge about the product category can decrease uncertainty, but also the characteristics of the innovation itself (Rogers, 2003). The subsequent section elaborates on this from an innovation diffusion perspective.
2.2 Innovation Diffusion
Research on the adoption of new innovations is rich and has been studied for over 40 years. One of the most important and widely used frameworks on innovation adoption is that of Rogers, called the “diffusion of innovations theory” (Rogers, 2003). Rogers’ definition of innovation diffusion is: “the process in which an
innovation is communicated through certain channels over time among the members of a social system” (Rogers, 2003: 5). Rogers defines an innovation as “an idea, practice, or project that is perceived as new by an individual or other unit of adoption” (Rogers, 2003: 12). Stremersch and Tellis (2002) mention that bundling an innovation is useful to enhance market penetration and diffusion, but research on this is scarce.
The innovation-diffusion process is described as “an uncertainty reduction process” (Rogers, 2003: 232). Rogers proposes five characteristics of an innovation that can reduce uncertainty: relative advantage, compatibility, complexity, trialability, and observability (Gatignon & Robertson, 1985; Rogers & Shoemaker, 1971; Rogers, 1983, 2003). Perceived risk is also negatively related to the speed of diffusion
(Gatignon & Robertson, 1985; Ostlund, 1974; Venkatraman, 1991). Rogers (2003) showed that the rate at which an innovation is adopted is primarily explained by how these characteristics of the innovation are perceived. Due to the scope of this study, this section focuses on the complexity characteristic and that of perceived risk.
Complexity is defined as “the degree to which an innovation is perceived as relatively difficult to understand and use” (Rogers, 2003: 15). This element is
12 The literature on product diffusion suggests that over time, a new product gets adopted first by a small portion of the market before the majority of the market adopts it. Adopters of an innovation are often considered to be member of a specific adopter category (Bass, 1969; Rogers, 2003). Adopters are categorized based on their
innovativeness, resulting in five distinct adopter categories: innovators, early adopters, early majority, late majority, and laggards (Rogers, 2003). Within each adopter category, consumers are assumed to be homogeneous in terms of
innovativeness. The adopter category that adopts the product before all other categories is referred to as the “innovators”. Innovators exercise social influence (through word-of-mouth and observational learning) on the potential market, which is captured by the imitation parameter in the classic diffusion model developed by Bass (Bass, 1969; van den Bulte & Stremersch, 2004). By adopting the product first, innovators are the ones that first determine the number of adoptions, based on which new adopters will consider “imitating” the innovators. Less innovative consumers (imitators) often use innovators as sources of information when it comes to new products (Bass, 1969). If the innovators do not adopt the new product, there is an increasing likelihood that other consumers will not adopt either.
Innovators make up roughly 2,5% of the total market, and can be
characterized as being highly innovative compared to the other segments that might adopt the product later (Bass, 1969; Rogers, 2003). Innovators are considered to be venturesome, which requires them to developed expertise and gain complex technical knowledge (Rogers, 2003). The diffusion literature suggests that technological and learning aspects are distinct elements that set innovators apart (Midgley & Dowling, 1993; Bass, 1969; Moore, 1991). Also, their venturesomeness causes them to be risk-seeking, as they cannot rely on information from other people who adopted previously (Rogers, 2003: 248). Innovators are able to cope better with higher degrees of risk and uncertainty about an innovation when compared to the other adopter categories
(Rogers, 2003). Since innovation complexity is negatively related to the rate of adoption and innovators are the first category to adopt an innovation, one can
13
2.3 Consumer Innovativeness
Research shows that consumer innovativeness is one of the most important determinants of the adoption of new products (Gatignon & Robertson, 1985;
Hirschman, 1980; Im, Bayus, & Mason, 2003; Im, Mason, & Houston, 2007). In the literature, a distinction is being made between actualized innovativeness and
dispositional innovativeness. Steenkamp and Gielens (2003) found dispositional innovativeness to be a significant predictor of actual innovative behaviour.Actualized innovativeness can be defined as “the degree to which an individual is relatively earlier in adopting an innovation than other members of his system” (Rogers & Shoemaker, 1971: 27). Logically, actualized innovativeness thus implies the actual adoption of the new product. Dispositional innovativeness is often alternatively phrased as innate innovativeness (Hirschman, 1980; Im et al., 2007; Midgley & Dowling, 1993; Steenkamp & Gielens, 2003). Dispositional innovativeness can be defined as “a generalized unobservable predisposition toward innovations applicable across product classes” (Im et al., 2003: 62). Predisposition implies that this concept considers the innovativeness of an individual as a personal characteristic that is more or less innate (Hirschman, 1980, Midgley & Dowling, 1978). An alternative definition of dispositional innovativeness comes from Midgley (1997: 49): “the degree to which an individual makes innovation decisions independently of the communicated
experience of others”. This definition also more or less assumes that all consumers possess innovativeness as a personal trait to some extent, as being a product of the individual’s genes (Hirschman & Stern, 2001; Midgley, 1978; Midgley, 1977). This trait inhibits the need for new experiences (Hirschman, 1980). New products can help the individual attain these new experiences by stimulating the consumer cognitively or sensually (Hirschman, 1984; Venkatraman & Price, 1990; Venkatraman, 1991). Previous research shows that both actualized innovativeness (Hirschman, 1980) and dispositional innovativeness (Im et al., 2007) are positively related to the acquisition of information, gaining knowledge, and novelty seeking. These elements are
considered to be part of the process of new product adoption, as mentioned in the previous section (Im et al., 2007). However, consumers that are considered to be highly innovative in a specific product category are not necessarily highly innovative in other product categories. This is often referred to as domain-specific
14 only if the consumer is interested in the product category. If consumers are interested in the product category, they are generally more informed about new products and therefore have to exert less effort to evaluate the product’s attractiveness and value (Gatignon & Robertson, 1985; Midgley & Dowling, 1987).
As mentioned in the previous section, some innovative consumers
(categorized as innovators in diffusion theory) gather information and knowledge on new products and spread this information via word-of-mouth to the market. In the literature on consumer innovativeness, these consumers are often referred to “market mavens” (Alba & Hutchinson, 1987; Feick & Price, 1987). Market mavens are those consumers that are used as sources for information since they obtain knowledge through first-hand experience by adopting the product, and actively sharing this knowledge with others. Market mavenism helps the new product to become more familiar in the market and increases product diffusion (Alba & Hutchinson, 1987; Feick & Price, 1987). Whereas dispositional innovativeness is related to the degree in which consumers prefer innovative products rather than non-innovative alternatives, market mavenism is related to the degree in which innovative consumers participate in word-of-mouth to spread information about the innovative product to others. This study focuses on the constructs of dispositional innovativeness, market mavenism, and interest in product category, in order to determine if consumers are actually innovative and their behavior resembles that of the innovator adoption category. The methodology section elaborates on how this study measures consumer innovativeness.
Since consumers with high dispositional innovativeness are generally more knowledgeable (if interested in the product category), they might have a better understanding on how the complementary products that are bundled with an innovative product provide value. Research indeed suggests that consumers with significant experience in the product category have better-developed schemas and knowledge structures, allowing them to predict possible outcomes more accurately (Gatignon & Robertson, 1985; Hirschman, 1980). Gatignon and Robertson (1985) propose that “the completeness and complexity of these information structures or schema determine the ability to mentally represent a problem, to isolate solution criteria, to identify decision strategies, and to evoke the products with the relevant attributes” (Gatignon & Robertson, 1985: 861; Hirschman, 1980). This would reduce uncertainty about products offered in mixed-bundling and cause innovative
15 (Johnson et al., 1999). This might cause preferences between innovative consumers and less innovative consumers to be significantly different since they rely on different concepts in their decision-making.
2.4 Synthesis, Hypotheses, and Conceptual Model
The main research question of this thesis is defined as: “How does bundling of complementary products that vary in complexity affect preferences of innovative consumers for an innovative product?”.
This study proposes that the knowledge that innovative consumers acquire decreases their uncertainty about the products’ attributes in the choice situation. Because these consumers are more familiar with the product category, they have to exert less effort to evaluate the product’s attractiveness and value (Gatignon & Robertson, 1985; Midgley & Dowling, 1987). In general, uncertainty about product attributes might arise especially for complex products. Consumers are presumed to have the same knowledge level regarding simple products regardless of their innovativeness, as these products don’t require much knowledge to evaluate. However, this might not be true for complex products since they require more (technical) knowledge to fully
comprehend. The uncertainty that less innovative consumers experience regarding complex products might evoke risk aversion, causing simple products to be more preferred. However, research has shown that consumers with high dispositional innovativeness (and specifically those that are categorized as innovators) are
generally less risk averse and even risk-seeking, which might be an indication of the proposed relationship between knowledge and uncertainty (Bartels & Reinders, 2011; Robertson & Myers, 1969; Rogers, 2003). Given that innovative consumers are more knowledgeable, less risk averse, and novelty seekers, they are expected to prefer complex items when presented in the à la carte situation more than less innovative consumers. Therefore, we derive the following hypothesis.
H1: Innovative consumers have a significantly higher preference for complex products in the à la carte situation than less innovative consumers.
16 contain complex products. Harris and Blair (2006a; 2006b) show that the relative preference for a product bundle when compared to separate products (à la carte) is positively influenced when consumers experience functional compatibility risk, uncertainty about product knowledge, and have a low motivation for searching and processing information. Functional compatibility risk is defined as “risk associated with the joint functionality or performance of bundle elements” (Harris & Blair, 2006a: 19). When the products in the choice situation are complementary, one could question how the products complement each other and what combination of products provides the best solution. Since less innovative consumers are not as involved in the product category, finding an answer to this question might require expertise or
extensive cognitive effort. Less innovative consumers might lack the knowledge as to how complex items are compatible or complement each other, since they might only have sufficient knowledge regarding simple products. Purchasing a product bundle could prevent extensive cognitive efforts and risk based on the assumption that
product bundles often contain products that are highly compatible with each other and hence provide superior value (Guiltinan, 1987; Wilson, Weiss & John, 1990).
Therefore, product bundles containing complex products may simplify decision-making by providing a “shortcut” that reduces uncertainty and risk (Harris & Blair, 2006a; Urbany, Dickson & Wilkie, 1989). Moreover, preference for product bundles increases as the bundle’s ability to reduce search costs increases (Guiltinan, 1987; Harris & Blair, 2006b). Since simple products require less information search than complex products, this might imply that complex bundles are more preferred by less innovative consumers than innovative consumers. Innovative consumers might not need to search for information about complex products as extensively, since their participation in novelty seeking motivates them to actively search for information (Hirschman, 1980; Roehrich, 2004). Furthermore, research suggests that the extent to which bundles are able to reduce costs for searching for information is most likely moderated by the complexity of the products and the consumer’s knowledge (Brucks 1985; Harris & Blair, 2006b). Therefore, the following hypotheses are formulated in order to answer the research question.
17 H3: Innovative consumers do not have a significantly different preference for a
discount bundle that contains simple products than less innovative consumers. H4: Innovative consumers have a significantly lower preference for the no-choice option than less innovative consumers.
It is important to note that the “Innovative consumers” that are mentioned in the hypotheses are consumers that score relatively high on all three innovativeness constructs (dispositional innovativeness, market mavenism, interest in product category) when compared to other consumers. These three constructs were chosen specifically to find innovative consumers that resemble innovators, so that the results can be generalized to the field of innovation diffusion.
Figure 1 contains the conceptual model including the hypothesized effects. One can notice that the hypotheses only concern moderating effects. The reason for this is that based on the literature, one does not expect the main effects to be either significantly positive or negative a priori, without controlling for individual
differences.
18
3. METHODOLOGY
This chapter elaborates on the methods that were used to collect data, design the study, and test the hypotheses. First, this chapter elaborates on why the Menu-Based Conjoint (MBC) method is chosen for this study, and how it is different from the industry-standard Choice-Based Conjoint (CBC) method. The subsequent section explains what measures are used to collect data. Thirdly, the study design is
explained. The final section explains how the data were analyzed and how the hypotheses were tested.
3.1 Research Method: MBC
Menu-Based Conjoint (MBC) analysis is an advanced discrete choice modeling method that is designed to study preferences in multi-check menus. Sawtooth Software released their first version of the MBC software in 2012, which made analyzing menu-based choice tasks a lot less time consuming (Orme, 2013). The development of MBC software was driven by the need to study menu-based tasks due to mass customization. Consumers are confronted increasingly more often with choice tasks that involve making menu-like choices. Examples include fast-food menus, telecom services, and insurances, where consumers can configure their product/service by selecting each component (attribute) themselves (Cohen & Liechty, 2007; Liechty, Ramaswamy & Cohen, 2001). This empowers consumers to design and customize their personal product solution, rather than choosing amongst several pre-configured product solutions. This requires a consumer to think in terms of “and” rather than “or”.
There are some important differences between MBC and other discrete choice methods such as CBC or MaxDiff. In CBC, product alternatives are composed of several attributes, each attribute having several levels (i.e. prices). CBC observes choice for different combinations of attribute levels (i.e. choices among pre-configured products). Therefore, discrete choice modeling methods such as CBC focus on the “or” of the choice. This allows researchers to estimate preferences for products that are composed of specific levels, but not specifically for individual attribute levels, as choice is observed on product-level and not on attribute-level.
19 individual attributes influence what respondents pick. Choice is observed for each attribute individually, rather than for the composition of attributes as a whole. In MBC, one is able to estimate complementary and substitution effects amongst attributes, which cannot be done using CBC or MaxDiff (Orme, 2013). Estimating substitution and complementary effects is useful for revenue optimization and more specific marketing actions like substituting products when they are out of stock or focusing on cross-selling using complementary products. Studying menu-tasks using MBC enables practitioners and firms to optimize a customizable product or service (or a bundle of products). MBC is therefore a perfect solution to study mixed-bundling, as respondents can choose pre-configured bundles or items à la carte.
MBC uses many of the same statistical procedures as other discrete choice methods to build models. These statistical procedures include Multinomial Logit models and Binary Logit models. However, MBC allows specifying more complex models and combinations of models than CBC. To estimate utilities, the user of the MBC software has several options (Orme, 2013). One can use counting analysis to count choice frequencies and count the choice combinations of items that most often occur. Also, Aggregate Logit can be employed to obtain aggregated utilities for the whole sample. To obtain individual-level estimates, Hierarchical-Bayes (HB) estimation can be used. The software also includes a market simulator that can be used to simulate choice probabilities based on specific attribute levels.
3.2 Data Collection
3.2.1 Selection of Stimuli. To study adoption behavior, it is of great importance that
an innovation is selected that is relatively new and has not been adopted by the majority of the market. For this study, the innovation of interest is a 360-degree camera. This is a spherical camera that allows the consumer to film in 360 degrees to capture all surroundings. The produced content can be viewed via a smartphone or even on a Virtual-Reality headset to get a total immersive experience. Many established brands such as LG, Samsung, and Nikon released their first 360-degree cameras in early-mid 2016.
20 asked to indicate the extent to which they perceived certain products to be complex (using a five-point likert scale). An overview of the pre-test is included in Appendix A. The pre-test was distributed using an anonymous link, which was available to fellow students at the University of Groningen, and for friends and family members of the author. The survey was distributed both in an English version and in a Dutch version. In addition to evaluating each product, the age of each respondent was recorded. The age distribution shows a mean age of 29 years, with a standard deviation of 11 years, with the minimum age being 20 years, and the maximum age being 57 years. Table 1 shows a summary of the pre-test results. Table 1 is derived from the more detailed table in Appendix A.
Simple
Neither simple,
nor complex Complex
Count % Count % Count %
A protective hardcase (bag) 33 97,06% 1 2,94% 0 0,00% A rechargeable battery 32 91,43% 2 5,71% 1 2,86% A 128GB memorycard 28 80,00% 5 14,29% 2 5,71% A selfie-stick/tripod (2-in-1) 28 80,00% 3 8,57% 4 11,43% An external microphone 20 57,14% 9 25,71% 6 17,14% A 360 degree camera 19 54,29% 9 25,71% 7 20,00% A set of adhesive mounts 18 51,43% 10 28,57% 7 20,00% A Virtual-Reality headset 12 34,29% 11 31,43% 12 34,29% A drone 7 20,00% 7 20,00% 21 60,00% Professional editing software 3 8,57% 4 11,43% 28 80,00%
Table 1: summary of pre-test results
Comparing the more detailed table in Appendix A with Table 1, one can notice that the response categories “somewhat simple” and “simple” were aggregated to a
category called “simple”, as well as “somewhat complex” and “complex” which were aggregated to a category called “complex”. Table 1 is an ordered table that shows the products that are perceived as most simple in the top rows and the products that are perceived as most complex in the bottom rows. This study aimed to use a cut-off point of 80% to classify a product as either simple or complex. Products that were classified by 80% of the sample or more as either simple or complex are marked in bold in Table 1. From Table 1 we conclude that four products are considered to be simple, and one product to be complex.
21 note that there is some lack of validity in the classification of this specific product, since it did not meet the pre-determined cut-off point. This notion was used as an important note when testing hypotheses for this product. Also, in retrospect the author decided to not to include “A rechargeable battery” in the study design, as it is
generally supplied with the 360-degree camera in order for it to function properly. Therefore, among the remaining products, the two products that were perceived as most simplistic and the two perceived as most complex were selected to be included in the product bundles. These products were: a protective hardcase (bag), a 128GB memorycard, a drone, and professional editing software.
3.2.2 Measurement Scales. This study used three scales to measure consumer
innovativeness, namely that of dispositional innovativeness (Steenkamp & Gielens, 2003), market mavenism (Feick & Price, 1987), and interest in the product category (Laurent & Kapferer, 1985). The aim of this study was to examine if these three scales load onto the same dimension using Factor analysis, and to create one
consumer innovativeness measure that is internally consistent. Principal Component Analysis (PCA) was used as the extraction method and Varimax was used as the rotation method. The Cronbach’s Alpha statistic was studied to determine the internal consistency of scales. Appendix B contains the scale items that were used for this study. Item 3.4 functioned as an attention check to increase construct validity and detect respondent fatigue. Respondents who did not respond to the attention check correctly were not included for the analysis. To measure dispositional innovativeness, this study used the scale that was used by Steenkamp and Gielens (2003). To measure market mavenism, this research used a scale from the paper by Feick and Price
(1987). To measure the consumer’s interest in the product category, the last three items of the consumer involvement scale by Laurent and Kapferer (1985) were used. All of these scales are measured using a seven-point likert scale. In addition to obtaining measurements to measure consumer innovativeness, several demographic measures were included in the survey. These measures include age, gender, income, and education (Appendix B).
3.2.3 Sample and Survey Distribution. The survey was translated into Dutch since
Virtual-22 Reality glasses were randomly given away among respondents who completely and correctly filled in the survey. This give-away was used as an incentive for respondents to participate in the study and to motivate respondents to answer the questions
accurately. Within the survey respondents were carefully instructed regarding the tasks that ought to be performed. The data were collected during a three-week period in November/December 2016. Sawtooth Software’s Lighthouse Studio software was used to design the survey and collect the data. The survey was hosted using a
Sawtooth Web Server. Thereafter, Sawtooth Software’s MBC software was used to analyze the data, create choice models, and obtain individual level utilities.
3.3 Study Design
This MBC study was designed so that it resembled a mixed-bundling choice situation. This implies that consumers were able to adopt the innovation as part of a pre-configured product bundle, or to adopt it by choosing the products separately through a “mix and match” approach in the à la carte section. Also, a no-choice option could have been chosen, which enabled the consumer to not adopt the
innovation at all. Figure 2 displays a screenshot of the study design. A more detailed (English) version is included in Appendix C. Figure 2 includes the eighth choice task of the study, which is a fixed hold-out set.
23 The study design contains four (discount) product bundles at the top position of the choice task. The two bundles on the left side contain complex products, and the two bundles on the right side contain simple products. In each bundle, the innovation (360-degree camera) is bundled with only one product. The bundle size is determined based on an online study examining product bundle offerings for 360-degree cameras by major online retailers in The Netherlands. From this study, one can conclude that bundling the 360-degree camera with one additional product is the most common bundling approach by many retailers (Figure 3). Many online retailers do not offer bundle sizes of three products or more. The innovation is therefore bundled with only one complementary product from the à la carte section.
Figure 3: Examples of bundling size Dutch retailers (Bol.com and Coolblue).
This study used a percentage-based segregated bundle discount rather than an
absolute discount in euros, since many online retailers use this approach. Respondents receive a 5% discount on the innovation and a 5% discount on the additional item when purchasing a bundle (in accordance with the mental accounting principles described in Chapter 2.1). The percentage-based discount is remained constant for all product bundles through all choice tasks. The total price of each product bundle in each choice task is calculated as follows: (price of innovation * 0,95)+ (price of complementary product * 0,95)).
24 or no-choice) were coded as categories of one variable. The second-stage of the
decision process concerns the products in the à la carte section. These four complementary products were coded as single binary variables (they are either selected or not). To be allowed to select any of these products, the respondent should have selected the 360-degree camera in the à la carte section. In other words: the choice for any of the à la carte products is conditional on the choice to adopt the innovation out-of-bundle (marked in green in Figure 4). Table C1 in Appendix C contains a useful and comprehensive overview of all variables in the study and how they relate to each other.
Figure 4: Two-staged choice process coding
In practice, consumers are able to purchase the innovation as part of a bundle, and still select other products à la carte. However, these prohibitions were coded specifically to examine how different product bundles are able to persuade respondents to adopt the innovation. If one removes these prohibitions, it is not possible to determine what (type of) products might have caused the respondent to adopt the innovation as part of a product bundle.
This study design holds (5 + (4*(44))) = 1029 possible combinatorial
25 (approximately the expected number of respondents). The study design was tested using the Lighthouse Studio software. The design efficiency test is included in
Appendix C. It indicates that the design was near-orthogonal and level-balanced. Each respondent was presented with 12 choice tasks. Of these choice tasks, two were fixed choice tasks, and the remaining (random) choice tasks were used to estimate utilities. Choice task 8 (Figure 2) and 11 were fixed and function as hold-out choice tasks for validation purposes to increase predictive validity of the study.
3.4 Data Analysis and Hypothesis Testing
3.4.1 Plan of Analysis. An R-script was written to transform the exported survey data
from the Sawtooth Web Server into an appropriate format that is compatible with the MBC software. R was preferred over other methods such as Microsoft Excel because it enhances reproducibility. Another R-script was used to test the hypotheses and validate the results using the hold-out choice tasks. Both R-scripts are included in Appendix D.
MBC provides several statistical methods to estimate preferences. First, counting analysis was used. Counting analysis simply compares how many time an item in the menu was chosen to how many times it was actually available to be
26 consumer preferences. Furthermore, the Aggregate Logit method assumes
Independence of Irrelevant Alternatives (IIA). Since the objective of this thesis is to study heterogeneous preferences based on individual consumer innovativeness, this requires a method that takes individual preferences into account. Therefore, a third method was used: Hierarchical-Bayes (HB). HB is an estimation method that is generally preferred over Aggregate Logit because it reduces the IIA, accounts for respondents with heterogeneous preferences, allows for a priori or post hoc segmentation of consumers, and provides slightly more accurate estimation results (Orme, 2000; Orme, 2013). Since the results of this study should allow practitioners to target specific consumers (innovators), HB was used to obtain individual level estimates and account for preference heterogeneity. An HB model is essentially a Logit-specified model that measures its fit in terms of Log-Likelihood (Orme, 2000). HB provides parameter estimates for each respondent by borrowing information from the total population of respondents. It borrows this information from a multivariate normal distribution, which is comprised of all respondent utilities (Orme, 2000). The extent to which HB borrows information from the population to populate individual parameter estimates is often referred to as “Bayesian Shrinkage”. If choices of
respondents are consistent, little information is borrowed from the population and the Bayesian Shrinkage is relatively low. If choices of respondents are inconsistent, the information for each respondent is low so more information is borrowed from the population to determine the individual parameter estimates. The latter case indicates a relatively high amount of Bayesian Shrinkage. This process of borrowing information is called “Gibbs Sampling”, which stabilizes the estimates for each individual over a series of iterations (Orme, 2000). For each hypothesis in this study, one HB model was estimated. Each HB model provided an ASC (Alternative-Specific Constant) when the point estimates were exported. The ASC captures the desirability of the hypothesized choice/products (Orme, 2013). A One Sample t-test was performed on these ASC variables of the HB models to make sure that the variables were
significantly different from zero. The ASC’s and the respondent’s consumer innovativeness scores were used to test the hypotheses.
3.4.2 Hypothesis Testing. Hypothesis 1 concerns the preference for complex products
27 drone, or a bundle containing the professional editing software. Hypothesis 3
28
4. RESULTS
This chapter reports the results of this study. First, this chapter provides information on the respondent sample. The subsequent sections report the results of the Factor analysis and the choice models. In section 4.4, the predictive validity of these choice models is examined. Finally, section 4.5 reports the results of the Regression models that were used to test the hypotheses.
4.1 Sample Results
The data for this study were collected between November 23rd 2016 and December 9th 2016. A total of 146 respondents took part in the survey. Of all responses, 104 were recorded as complete responses, whereas 42 responses were incomplete. Of all 104 complete responses, 3 respondents failed to fill in the attention check correctly (see item 3.4, Appendix B). The incomplete responses and the
29 Education LBO/MAVO/VMBO 7 6,9% HAVO or VWO 12 11,9% MBO 18 17,8% HBO 33 32,7% WO 31 30,7%
Table 2: Demographic statistics of respondent sample
4.2 Factor Analysis Results
Factor analysis and the Cronbach’s Alpha measure of internal consistency were used to determine the number of latent constructs underlying the total set of items that were used to measure consumer innovativeness. All items were included, except for item 3.4 which served as the attention check. Several component-solutions were studied to determine the optimal number of dimensions to measure consumer innovativeness. The results of the Factor analysis suggest that the items do not all load onto the same component. Furthermore, when studying the internal consistency of the scales using Cronbach’s Alpha, a three-component solution is considered to be the most optimal solution. This results in one component per consumer innovativeness construct (dispositional innovativeness, product category interest, market mavenism). The internal consistency of the dispositional innovativeness component could be improved from a Cronbach’s Alpha of 0,820 to a Cronbach’s Alpha of 0,851 by deleting item 1.3. Therefore, this item was excluded to improve internal consistency. Furthermore, the internal consistency of the interest in product category component could be improved from a Cronbach’s Alpha of 0,849 to a Cronbach’s Alpha of 0,876 by deleting item 2.1. However, since this component consists of only three items and the internal consistency is already very good, it was decided not to exclude this item in order to retain valuable information in this component. The Cronbach’s Alpha for the market mavenism component was 0,911, which is excellent.
30 three construct-specific components. Therefore, three new variables were created from this Factor analysis in order to test their effect on the hypothesized preferences.
4.3 Aggregate Logit and HB Results
Table F1 in Appendix F contains an overview of the variables that were included in the choice models. The base model for each hypothesis and the functional form of the variables in these models were chosen based on counting analysis and by studying the t-ratio’s. Subsequently, for each model several Likelihood-Ratio tests were performed to determine if the model could be significantly improved if another variable is added to the model. The most significant choice model for each hypothesis is used for estimation. Table 3 presents information on the goodness of fit of the Aggregate Logit models. The first model (H1) includes 52 respondents rather than the full sample of 101 respondents. This hypothesis models the choice of complex
products à la carte. As mentioned in the previous chapter, this choice is conditional upon the innovation being adopted in the à la carte section. This implies that only those respondents were included in the estimation process that met this condition. In other words: this model is a two-stage model that only includes those respondents who passed the first stage.
H1 H2 H3 H4
Log-Likelihood Null model -149,02664 -700,07865 -700,07865 -700,07865
Log-Likelihood -116,77430 -606,12564 -627,16346 -458,10220
Chi-Square 64,50469 187,90603 145,83039 483,95291
RLH 0,58092 0,54874 0,53743 0,63536
Percent Certainty 21,64200 13,42035 10,41529 34,56418
Consistent Akaike Info
Criterion (CAIC) 246,28988 1228,08668 1278,08003 932,03981
Respondents included 52 101 101 101
Table 3: Goodness of fit Aggregate Logit models
Model Effect Std Err t-Ratio Variable
H1 -1,16355 0,16122 -7,21725 ASC (H1 chosen) -0,60067 0,42400 -1,41666 IV3.Hardcase.Price [Linear] H2 -0,89908 0,06961 -12,91679 ASC (H2 chosen) -0,38467 0,18631 -2,06468 IV4.Drone.Price [Log-linear] H3 -0,75454 0,06817 -11,06874 ASC (H3 chosen) -0,62964 0,18320 -3,43687 IV1.Camera.Price [Linear] 0,36596 0,18279 2,00203 IV4.Drone.Price [Log-linear] H4 -1,59726 0,08600 -18,57370 ASC (H4 chosen) 0,96533 0,23012 4,19492 IV1.Camera.Price [Linear]
31 Table 4 provides the parameter estimates of the models. From this table one can
observe that the t-ratio of all variables is larger than |1,282|, indicating that the included variables are significant at the 80% confidence level. In addition, this table shows that many of the standard errors are larger than 0,10.
After the Aggregate Logit models were estimated, HB models were estimated to obtain individual level estimates. The general settings that were used along with the full estimation report of the HB analysis are included in Appendix G. Table 5 shows the results of the estimation procedure. An average RLH of 0,8348 and an average Percent Certainty of 73,58% show that these models are able to predict individual choices adequately. Since all hypotheses concern binary choices, it outperforms the random prediction chance of 50%. Table 5 also shows that HB estimation took less than 2 minutes for all models.
H1 H2 H3 H4
RLH 0,910 0,787 0,762 0,880
Percent Certainty 0,864 0,654 0,609 0,816
Average variance 33,304 13,833 5,222 15,549
Parameter root mean
square (RMS) 6,790 4,274 2,838 6,365
Time elapsed 11 seconds 1 minute and 23 seconds 1 minute and 23 seconds 1 minute and 22 seconds
Respondents included 52 101 101 101
Table 5: HB Results
4.4 Validation Results
Two fixed hold-out choice tasks were used in this study for validation purposes. The Hit Rate for each hypothesis was calculated to provide an indication of the goodness of fit and predictive validity (Table 6). One might notice that the total number of observations for H1 (104 observations) is less than the other models (202
32 predictions is relatively high for the model concerning H1 when compared to the other models.
Table 6: Hit Rate fixed tasks
Table 7: Mean Absolute Error
4.5 Hypothesis Testing
To test the hypotheses, four Regression models were estimated. These Regression models examine the causal relationship between the consumer innovativeness constructs and preference for a specific choice. The three
innovativeness variables were included as independent variables, and the ASC of the corresponding HB model was included as the dependent variable. The Regression results are included in Appendix H. Table 8 presents an overview of the Regression coefficients and their significance. Hypothesis 1 expected a positive effect of consumer innovativeness on the preference for complex products in the à la carte situation. From Table 8 one can observe that both the overall model and all its coefficients are insignificant. Hypothesis 2 proposed a negative effect of consumer innovativeness on the preference for a discount bundle containing complex products. Table 8 shows that the overall model significant at the 90% confidence level. The coefficient of dispositional innovativeness is significant at the 95% confidence level. The effect of this coefficient is however positive rather than negative, indicating that
33 contradictory to the expectations, dispositional innovativeness has a positive effect on the preference for product bundles containing complex products.
Significance (p-value): *** < 0.001, ** < 0.01, * < 0.05, . < 0.1 Table 8: Hypothesis testing results
Hypothesis 2 proposed that consumer innovativeness has no significant effect on the preference for discount bundles containing simple products. Table 8 shows that both the overall model and the coefficients are not significant. Hypothesis 4 concerns the negative effect of consumer innovativeness on the preference for the no-choice option. The results show that the overall model for this hypothesis and the negative coefficient for the Interest variable are significant at the 95% confidence level. The R2 of this model is 10.93%.
H1 H2 H3 H4
Market Mavenism 0.2227 0.2949 -0.2040 -0.2692
Dispositional Innovativeness 0.7362 0.7844* 0.2143 0.6977
Interest in product category 0.9055 0.4231 0.3934 -1.3551**
Model sig. 0.6777 2.317 . 0.6875 3.97**
34
5. DISCUSSION
This chapter draws conclusions from the results, discusses the major limitations of this study, and elaborates on what issues might have caused the hypothesized effects to turn out insignificant. Furthermore, this chapter summarizes the implications of this study.
5.1 Discussion
A drawback of this study is that both counting analysis and the
Log-Likelihood Ratio tests of the Logit models suggest very few effects to be included in the models. Many of the main effects did not significantly contribute to the model, whereas some cross-effects did (Table 1F, Appendix F). As adding these
non-significant main effects to the model would weaken the quality and significance of the model(s), for this thesis only the most significant effects were included. However, from a managerial perspective one would expect that the price of a certain product would affect its choice. The author has decided to focus on the theoretical exhibition of the MBC method by proceeding with the most significant models (as this
theoretical exhibition was one of the objectives of this study), rather than to focus on the managerial implications in terms of optimization.
The overall goodness of fit of the Logit models and their explanatory power are quite poor. One can conclude this by interpreting the Root Likelihood (RLH) or the Percent Certainty (Orme, 2013). However, the Logit models were mainly used to formally test what variables to include in the model. When studying the results of the Logit models (Table 4), one can observe that most of the standard errors of the
35 minutes to finish (Orme, 2013). This too is an indication that the sample size of this study was too small. The HB model that concerns H1 even finished after 11 seconds (including only 52 respondents due to a conditional dependent variable).
When interpreting the Regression analysis results in Table 8, we cannot fully accept any of the hypotheses. The overall models for H1 and H3 are not significant. For H2, we observe that there is some significance in the quality of the overall model (p<0.1). For this hypothesis, we find an unexpected positive effect for dispositional innovativeness. This implies that consumers that have a higher dispositional
innovativeness have a higher preference for bundles that contain complex products. This is somewhat in line with the expectations stated under H1 since it is expected that innovative consumers prefer complex products, but not when these products are presented in product bundles (H2). This finding implies that consumers with high dispositional innovativeness actually prefer complex products when they are bundled. This positive effect found under H2 does not confirm the expected mechanism that less innovative consumers would have a higher preference for complex discount bundles due to their lack of knowledge (as proposed in the theory section). This study expected that manipulating complexity could persuade less innovative consumers to adopt the innovation through bundling, but this result indicates that the preference of less innovative consumers is not influenced by this bundling manipulation. Since the effect is positive rather than negative, one cannot accept H2. The author proposes three possible causes for why this hypothesis is not supported.
First, this study assumes to a large extent that in general, less innovative consumers do not prefer products they do not fully understand. However, this might not always be the case since some products are not primarily purchased based on knowledge (e.g. hedonic products; Hirschman, 1980). Certain innovations might be more hedonic, causing product knowledge to be less important in the purchase decision. Also, consumers might have been overconfident in assessing their product knowledge (Arts et al., 2011). This might have distorted the proposed effect between knowledge and preferences of items that vary in complexity.
36 tendency to prefer products that are complex when the main product is also
considered complex. This is based on the notion that in general consumers respond more positively to bundles with a high-perceived fit or relationship between items (Guiltinan, 1987). This form of congruency behavior within product bundling might be an interesting topic for future research.
Third, this study does not account for the perceived innovativeness of the complementary products. The preference for complex products might not have been due to its complex nature, but due to the fact that complex products might be
perceived as more innovative than simplistic products, and that they are therefore more preferred by innovative consumers. Research indeed suggests that complexity might signal higher quality, newness and advancement (Arts et al., 2011; Taylor & Todd, 1995). This might be a very plausible explanation for the positive relationship found under H2.
For H4, we find that the overall model is significant, and we find a somewhat strong negative effect for interest in the product category. This effect implies that consumers that are more interested in the product category have a lower preference for the no-choice option. This also implies - due to the way in which the study was coded - that consumers that are more interested in the product category have a higher chance of adopting the innovation than less interested consumers. Since none of the other two variables are significant, one can only partially accept this hypothesis. More specifically, we cannot generalize any significant effects to the domain of diffusion theory since this finding does not directly apply to innovators as an adopter category, which was the aim of this research. Moreover, the explanatory power of the model for H4 is very low: only 10,93% of the variance in the preference of the no-choice option is explained. From the results one can conclude that the explanatory power of all models is very poor (less than 10% for most models). This lack of explanatory power is a major limitation in terms of its implications and generalizability.
5.2 Summary
The research question of this paper was: “How does bundling of
37 for an innovation. The three innovativeness constructs used in this study were
supposed to capture the characteristics of the innovators adopter category, which is so often referred to in innovation diffusion theory. Since none of the hypotheses
provided significant effects for all innovativeness constructs, the results cannot be generalized to the field of diffusion theory. The positive finding under H2 implies that consumers with higher dispositional innovativeness have a higher chance of adopting the innovation when it is bundled with complex products than consumers with lower dispositional innovativeness. From a managerial perspective, practitioners would manipulate product complexity in the mixed-bundling scenario to optimize the offering for the innovators adopter category. However, since high dispositional innovativeness is just one of the characteristics of a consumer from the innovators adopter category, bundling the innovation with complex products (H2) will not necessarily accelerate diffusion of the innovation. Alternatively, if all three innovativeness constructs would have had a positive effect under H2, the results would have had important implications for the diffusion of the innovation. Namely, that the consumer group that is most likely to adopt the innovation first has a significantly higher preference for the innovation when it is bundled with complex products than compared to other adopter categories. This would have allowed the practitioner to target consumers in the innovator category by using product bundling as a strategy to persuade these consumers to adopt the innovation.
But since none of the hypotheses are fully supported, the results of this study imply that manipulating product complexity in a mixed-bundling scenario with the objective to diffuse an innovation is not an effective strategy. It would have been interesting to examine if more (or all) innovative constructs would have been significant if the sample size of this study was sufficiently large.
This study contributed to the field of choice modeling methods by providing a demonstration of the capabilities and applications of MBC. It showed how
preferences for specific products in a mixed-bundling scenario can be measured, and how they differ amongst respondents that have different characteristics. Furthermore, this MBC study showed how consumer preferences in a multi-select menu are
38
6. REFERENCES
Alba, J. W., & Hutchinson, J. W. 1987. Dimensions of Consumer Expertise. Journal
of Consumer Research, 13(March): 411-454.
Arts, J. W. C., Frambach, R. T., & Bijmolt, T. M. A. 2011. Generalizations on Consumer Innovation Adoption: A Meta-analysis on the Drivers of Intention and Behavior. International Journal of Research in Marketing, 28(2): 134-144.
Bartels, J., & Reinders, M. J. 2011. Consumer innovativeness and its correlates: a propositional inventory for future research. Journal of Business Research, 64: 601– 609.
Bass, F. 1969. A New Product Growth Model for Consumer Durables. Management
Science, 15(5): 215-227.
Brucks, M. 1985. The Effects of Product Class Knowledge on Information Search Behavior. Journal of Consumer Research, 12 (June): 1-15.
Chakravarti, D., Rajan, K., Paul, P., & Srivastava, J. 2002. Partitioned Presentation of Multi-component Bundle Prices: Evaluation, Choice and Underlying Processing Effects. Journal of Consumer Psychology, 12(3): 215-230.
Cohen, S. H., & Liechty, J. C. 2007. Have it Your Way: Menu-based conjoint analysis helps marketers understand mass customization. Marketing Research
Magazine, Fall 2007: 28-34.
Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. 1989. User acceptance computer technology: A comparison of two theoretical models. Management Science, 35(8): 982-1003.
Drumwright, M. E. 1992. A demonstration of anomalies in evaluations of bundling.
39 Feick, L. F., & Price, L. L. 1987. The Market Maven: A Diffuser of Marketplace Information. Journal of Marketing, 5(January): 83–97.
Gatignon, H., & Robertson, T. S. 1985. A Propositional Inventory Of New Diffusion Research. Journal of Consumer Research, 11(March): 849-867.
Guiltinan, J. P. 1987. The Price Bundling of Services: A Normative Framework.
Journal of Marketing, 51(April): 74–85.
Hansen, W., & Martin, R. K. 1987. Optimal Bundle Pricing. Management Science, 36(February): 155-174.
Harris, J., & Blair, E. A. 2006a. Functional Compatibility Risk and Consumer Preference for Product Bundles. Journal of the Academy of Marketing Science, 34(1): 19-26.
Harris, J., & Blair, E. A. 2006b. Consumer Preference for Product Bundles: The Role of Reduced Search Costs. Journal of the Academy of Marketing Science, 34(4): 506–513.
Heath, T. B., Chatterjee, S., & France, K. R. 1993. Mental accounting and changes in price: The frame dependence of reference dependence. Journal of Consumer
Research, 22: 90–97.
Hirschman, E.C. 1980. Innovativeness, Novelty Seeking, and Consumer Creativity.
Journal of Consumer Research, 7(December): 283-295.
Hirschman, E. C. 1984. Experience Seeking: A Subjectivistic Perception of Consumption. Journal of Business Research, 12: 115-136.
40 Johnson, M. D., Herrmann, A., & Bauer, H. H. 1999. The Effects of Price Bundling on Consumer Evaluations of Product Offerings. International Journal of Research
in Marketing, 16(2): 129-142.
Khan, U., & Dhar, R. 2010. Price-framing effects on the purchase of hedonic and utilitarian bundles. Journal of Marketing Research, 47(6): 1090–1099.
Kahneman, D., & Tversky, A. 1979. Prospect Theory: An Analysis of Decision Under Risk. Econometrica, 47(March), 263-291.
Laurent, G., & Kapferer, J. 1985. Measuring Consumer Involvement Profiles.
Journal of Marketing Research, 22(1): 41 -53.
Lehmann, D. R., Moore, W. L., & Elrod, T. 1982. The development of distinct choice process segments over time: a stochastic modeling approach. Journal of Marketing, 46(2), 48–59.
Liechty, J., Ramaswamy, V., & Cohen, S. H. 2001. Choice Menus for Mass
Customization: An Experimental Approach for Analyzing Customer Demand with an Application to a Web-Based Information Service. Journal of Marketing Research, May: 183-196.
Mazumdar, T., & Jun, S. Y. 1993. Consumer evaluations of multiple versus single price change. Journal of Consumer Research, 20(3): 441–450.
Midgley, D. F., & Dowling, G. R. 1978. Innovativeness: The Concept and its Measurement. Journal of Consumer Research, 4(2): 229-242.
Moore, W. L., & Lehmann, D. R. 1980. Individual differences in search behavior for a nondurable. Journal of Consumer Research, 7(3), 296–307.
41 Orme, B. 2000. Hierarchical Bayes: Why All the Attention? Retrieved from
www.sawtoothsoftware.com/download/techpap/hbwhy.pdf. First accessed on January 12th 2017.
Orme, B. 2013. Menu-Based Choice (MBC) for multi-check choice experiments. Retrieved from http://www.sawtoothsoftware.com/download/mbcbooklet.pdf. First accessed on September 8th 2016.
Ostlund, L. E. 1974. Perceived Innovation Attributes as Predictors of Innovativeness.
Journal of Consumer Research, 1(2): 23-29.
Puto, C. P. 1987. The framing of buying decisions. Journal of Consumer Research, 14(3), 301–315.
Robertson, T. S., & Myers, J. H. 1969. Personality correlates of opinion leadership and innovative buying behaviour. Journal of Marketing Research, 6(2): 164-168.
Roehrich, G. 2004. Consumer innovativeness: Concepts and measurements. Journal
of Business Research, 57(6): 671-677.
Rogers, E. M. 2003. Diffusion of innovations (5th ed.). New York: Free Press.
Soman, D., & Gourville, J. T. 2001. Transaction Decoupling: How Price Bundling Affects the Decision to Consume. Journal of Marketing Research, 38(February). 30-44.
Schmalensee, R. 1984. Gaussian Demand and Commodity Bundling. Journal of
Business, 57(1): S211-S230.
