Capturing heterogeneous decision making processes: the case with the E-book reader market

Abstract

Consumer decision-making within a specific product category can take many forms. This study investigates the individual-specific transferal processes in new product adoption through cognitive, affective, and conative stages. We propose a general and flexible Bayesian multivariate regression model and fit the model to survey data on dedicated E-book reader adoption. The results show that—among six possible transition paths—all paths are feasible for the decision-making process, except for the conative → cognitive → affective path. In terms of market share, those who follow the hierarchy of effect process describing the cognitive → affective → conative process has the biggest market share and those following affective → conative → cognitive path has the smallest share. The contribution of this research is twofold. From a theoretical perspective, this study developed an estimable model for capturing heterogeneity in consumers’ decision-making process. Practically, the study empirically shows that a variety of decision-making paths exist, using survey data on the Korean E-book reader market. In the substantive domain, capturing the heterogeneity of consumer decision-making could provide marketers with insights to help profile consumers and with a basis for customer segmentation.

Keywords

customer segmentation decision-making process diffusion of innovation Gaussian mixture hierarchy of effect model

Introduction

Understanding of the impact of communication messages on consumers’ purchase decision-making has long been an important research topic in the marketing field. Since the pioneering study that introduced the hierarchy of effect model by Lavidge and Steiner (1961), the hierarchy framework has been widely cited and applied in consumer behavior studies. The core of the traditional hierarchy framework is the claim that consumers respond to marketing communication or advertisement in a very ordered way that includes three stages: cognitively (“thinking”) first, affectively (“feeling”) second, and conatively (“doing”) third (Barry & Howard, 1990; Vakratsas & Ambler, 1999). While there is little disagreement on the three components of the hierarchy of effects model, there has been quite some critique on the order of the three stages in the hierarchy and there is no precise evidence about the sequence. This critique is directly linked to the following question: Does awareness of a product always lead to preference for the product and then to purchasing behavior (Barry, 1987)? From an academic perspective, this has been the area of the most intense criticism and debate among researchers concerning the hierarchy of effect.

The hierarchy of effect model used in consumer behavior is also referred to as the “decision-making perspective” in market research. From this decision-making perspective, prior works claim that there may exist different sequences of the three decision-making stages that differ from the traditional hierarchy of effects model (Barry & Howard, 1990; Vakratsas & Ambler, 1999). For example, Krugman (1965) argued that, in low involvement situations, massive repetition of advertisements eventually leads to a modified cognitive structure in consumers, leading some consumers to purchase a product based on cognition alone and only deciding afterward whether they prefer it or not. Krugman’s findings appear to suggest a cognition, conation, and affect sequence. Furthermore, Ray et al. (1973) suggested that any stage can be the initial processing stage in the consumer decision-making process. They argued that all three hierarchy stages are feasible and can be correct because consumer responses are likely to be individual and situation-specific. They also found that three separate hierarchy models (three orders model) were valid. Despite evidence that consumers may follow a different order of the three stages, little empirical research has been performed that addresses individual heterogeneity concerning the decision-making process.

This study therefore investigates a transferal process for new product adoption that allows for consumer heterogeneity in the decision-making process. This can be expressed by the following question:

Why is it important to incorporate heterogeneity in the decision-making process while studying the adoption behavior of new products?

This can be answered from two perspectives. First, while the traditional hierarchy of effect model has been heavily cited to describe consumers’ new product adoption, there have been many critiques of its sequence. We therefore allow a flexible pattern for the decision-making process in this research. This perspective includes explaining which paths are reliable and which are not. Second and more importantly, exploring and understanding different decision-making paths provides information on what communication strategies should be implemented according to each path, when marketers plan and implement communication strategies. In other words, once marketers identify individual-specific decision-making paths, they are able to learn different essential key drivers and bottleneck in the adoption of the new products and establish different communication strategies by targeting those who follow the same decision-making path. Consequently, capturing heterogeneous decision-making paths can help marketers when planning communication messages and advertising strategies to stimulate consumers to reach adoption behavior as fast as possible. Based on these two perspectives, this study aims to propose a statistical model for capturing all possible different orders of the three stages among individuals in the decision-making process regarding new products. We then apply the suggested model to data from the Korean E-book reader market. We expect that the application of the proposed model will provide a basis for customer segmentation and will identify key drivers or bottlenecks for stimulating adoption in marketers’ viewpoint.

Theoretical background

It is well known that in the “real world,” consumers do not immediately decide whether to purchase new products or not when they receive information on these products. They rather experience a series of hierarchical and sequential stages before reaching an actual adoption decision. This is called the hierarchy of effects model, a model that was theoretically developed to explain the sequential stages. This theoretical model was proposed by Lavidge and Steiner (1961), who described the accumulated effect of advertising on consumers’ purchase decisions. Their theory states that there are different sequential stages involved from when the product first enters target consumers’ awareness to the consumer developing purchase intention. The term “hierarchy” precisely represents the sequence of multiple steps a consumer passes through from the initial exposure to a product or an advertisement to their purchase decision. Researchers found that consumers went through a series of stages as they moved from awareness–knowledge to decision to adopt (Beal & Rogers, 1960). Adoption behavior is therefore a process that comprises three stages—cognitive, affective, and conative—that occur over time.

According to the hierarchy of effect model, at the cognitive stage, consumers form beliefs about a product by accumulating knowledge regarding the relevant attributes. Next, at the affective stage, consumers evaluate these beliefs and develop a feeling about the new product. Over time, when the conative stage is reached, consumers engage in a relevant action or behavior, such as adoption or rejection of the new product, based on their evaluation. There is little doubt of the existence of the cognitive, affective, and conative reactions to advertisement. While all three components of an attitude are important, their relative impact may vary, depending upon a consumer’s level of motivation about the object. The question, however, is whether cognitive reactions must precede affective reactions which, in turn, must precede conative reactions. Existing literature relating to both psychology and marketing communication indicate that a cognitive response is often not a measurable precedent to either affective or conative responses (Bauer et al., 2006). Similarly, the affective–conative link has been questioned by numerous studies that addressed the attitude–behavior relationship. If we knew nothing about hierarchy, it would be reasonable to assume that the three stages could be ordered in all six possible permutations. In fact, some researchers have pointed out that consumers’ response to marketing communication messages can be represented by various styles, including the standard hierarchy of effect model (Jain & Sharma, 2013; Sprotles & Kendall, 1986). There is therefore a possibility that an individual can have a heterogeneous decision-making process. Considering the above, we briefly collate the alternative decision-making paths below.

Cognitive → affective → conative: traditional hierarchy of effect model

This model describes a consumer’s approach to product decisions as a problem-solving process (Lavidge & Steiner, 1961). First, they form a belief about a product by accumulating knowledge (beliefs) regarding its relevant attributes. Next, the consumer evaluates these beliefs and develops a feeling about the product. Over time they integrate information about alternative products. Finally, based on the evaluation, they engage in relevant behavior. This hierarchy assumes that a consumer is highly involved in making a purchase decision. The person is motivated to seek out a lot of information, carefully weigh alternatives, and come to a thoughtful decision. This process is likely to occur if the decision is important to the consumer.

Cognitive → conative → affective

This decision-making path has frequently been labeled as the “low involvement” hierarchy (Krugman, 1965, 1966). In this scenario, the consumer does not have a strong preference for a product but acts based on limited knowledge and then decide whether they like it or not. Following the findings of Krugman (1965), in the wake of the overwhelming repetition of television advertising, consumers may be better able to recall the concept of the product. Then, when the consumer is in a purchasing situation, that product comes to mind and they buy it after all. Their preference is then subsequently changed as a result of their experience with the product.

Affective → cognitive → conative

In this sequence, the affective stage usually, if not always, precedes the cognitive stage (Vaughn, 1980, 1986). This sequence is thought to typify the response of “feeling” consumers, who respond more to emotion than information when making purchase decisions resulting from advertisements. Vaughn (1986) posited that this hierarchy successfully explains the purchases of high involvement products and the model is the priority for buying emotional products such as cosmetics, jewelry, and fashion clothing.

Affective → conative → cognitive

Zajonc (1980) stresses the significance of the affective stage as the central aspect of purchasing behavior and found that a consumer’s affective response does not always require prior cognitions, but instead are primarily affectively based (Zajonc, 1980, 1984; Zajonc & Markus, 1982). The fact that preference can be decided by affective basis alone presents the potential for an affective and conative path. If an individual later saw the need to justify a preferred product, the affective, conative, and cognitive path could arise. This hierarchy can explain purchases such as that of a track listed in the billboard charts that may possess the same attributes as many other songs (e.g., dominant bass guitar, raspy vocal); however, beliefs about these attributes cannot explain why one song became classic while another disappeared in the chart leaving many people unaware of their presence.

Conative → cognitive → affective

Kiesler (1971) stated that choice behavior often wields the power of commitment, which results in the reorganization of cognitions to be consistent with that commitment (Kiesler, 1971). Affective response then follows commitment with both conative and cognitive responses. In the marketing context, purchasing a product may cause one to think about it in a manner that supports the choice, after which feelings are developed, consistent with that choice behavior and thoughts.

Conative → affective → cognitive

This decision-making path has been labeled “Dissonance-Attribution hierarchy” (Bem, 1972; Kelley, 1973). It is the exact opposite of the standard hierarchy of effect and typically occurs when the audience has been involved but the alternatives have been almost indistinguishable. In this situation, a consumer first purchases the product and preference is formed afterward to bolster the choice; next, selective learning follows to further support the behavior. Ray et al. (1973) explained that when a consumer is forced to make a choice on the basis of some non-media or non-marketing communication source, the conative response—a choice among undifferentiated alternatives—is made, after which attitude and cognitive responses are supposed to follow.

In short, existing research suggests many decision-making processes beyond the standard hierarchy of effect model. In this context, Ratchford (1987) proposed the Foote, Cone, and Belding (FCB) grid model which describes a planning model for advertising in which the decision-making process can be classified. Figure 1 describes the essence of the FCB grid model, which classifies consumers’ decision-making process based on the level of product involvement and characteristics of a product (thinking product/feeling product). The main claim of the FCB grid model is that a two-by-two combination matrix determines the decision-making process.

Figure 1.

Heterogeneous decision-making process based on the FCB grid model.

The FCB grid model has been used successfully in the real world and has provided useful implications for marketers. However, it has a critical limitation in that the model assumes that, within the same product category, one follows the same decision-making process. In other words, the product category determines the type of decision-making process in the FCB grid model (Park & Gretzel, 2008). It is natural, however, to allow heterogeneity in the decision-making process within the same product category. For example, one may follow the standard hierarchy of effect model to purchase an E-book reader, whereas another may follow the affective, cognitive, and conative stages by stimulating feeling rather than thinking. This study aims to overcome this limitation and propose a general and flexible model for capturing heterogeneity in the decision-making process in new product adoption.

Research design

As described above and based on our review, prior works have consistently argued that a variety of decision-making paths are conceivable in the real world. Against this background, this study investigates all possible transferal processes in decision-making through cognitive, affective, and conative stages. Figure 2 presents our research framework.

Figure 2.

The proposed theoretical research framework.

The research framework described in Figure 2 represents the potential transferal processes of a consumer’s decision-making path. For instance, if one follows the traditional hierarchy of effect decision-making path (Path 1) through the cognitive, affective, and conative stages, the regression model is as presented in equation (1)

\begin{array}{l} y_{c o g, i} = α_{c o g} + δ^{'} x_{i} + ε_{c o g} \\ y_{a f f, i} = α_{a f f} + γ_{c a} y_{c o g, i} + ε_{a f f} \\ y_{c o n, i} = α_{b e} + γ_{a c} y_{a f f, i} + ε_{b e} \end{array}

(1)

where $y_{c o g, i}, y_{a f f, i}, y_{c o n, i}$ are variables for cognitive, affective, and conative responses from each respondent $i (i = 1, \dots, I)$ ; $x_{i}$ is the variable for perceived innovation characteristics from each respondent; $α_{c o g}, α_{a f f}, α_{a f f}$ are the constants of each linear regression; and $δ_{0}, γ_{c a}, γ_{a c}$ represent the regression coefficients of the equation. Specifically, $δ$ indicates the regression coefficients of perceived innovation characteristics and $γ_{c a}, γ_{a c}$ indicate the transition weight of the prior stage. Similarly, we can express all six possible decision-making paths mathematically in the form of regression.

The decision-making path accounts for individual respondents. Equation (1) can be expressed again as equation (2) in a vector form

f o r τ_{i} = j Y_{i} = X_{i}^{(j)} β_{j} + ε_{i}

(2)

where $τ_{i}$ is the discrete latent variable indicating individual $i$ ‘s decision-making path $j (j = 1 \dots J)$ ; $X_{i}^{j}$ are the explanatory variables that correspond to the linear regression model of equation (1); and $ε_{i}$ is the error distribution that follows normal distribution with mean zero and variance-covariance matrix $Σ_{j}$ . We assume that the variance-covariance matrix has only diagonal elements, leaving off-diagonal elements zeros for identification; it has the form of $d i a g (σ_{c o g}^{2}, σ_{a f f}^{2}, σ_{c o n}^{2})$ .

The proposed model is an extended version of the traditional parametric Gaussian mixture model, which has been frequently applied in the marketing field (Allenby & Rossi, 1998; Rossi, 2014). Note that this equation is used only for individual transferal paths. When we aggregate data over the same path type group, the model identifies path-specific parameters of regression coefficients and error disturbance. We allow the model to tell us what portion $(π = (π_{1}, π_{2}, \dots, π_{J}))$ of each decision-making path can be obtained throughout the whole population by introducing hierarchical specification with respect to $τ_{i}$ .

We can therefore express the full posterior as a result of the product of the likelihood function and priors of unknown parameters, which are presented below as equation (3)

\begin{matrix} P (β, Σ, τ, π | y, x) \propto {Π_{i = 1}^{I} P (y_{i} | β, Σ, τ_{i})} \times {Π_{j = 1}^{J} P (β_{j})} \times {Π_{j = 1}^{J} P (Σ_{j})} \times {Π_{i = 1}^{I} P (τ_{i} | π)} \times P (π) \\ \propto [Π_{i = 1}^{I} Π_{j = 1}^{J} {| Σ |}^{- n_{j} / 2} \exp {- \frac{1}{2} \sum_{i \in Q_{j}} {(Y_{i} - {X_{i}}^{(j)} β)}^{'} Σ^{- 1} (Y_{i} - {X_{i}}^{(j)} β)}] \\ \times [Π_{j = 1}^{J} {| C |}^{- \frac{1}{2}} \exp {\sum_{i \in Q_{j}} - \frac{1}{2} {(β_{j} - B)}^{'} C^{- 1} (β_{j} - B)}] \\ \times [Π_{j = 1}^{J} {| Σ |}^{- (s + m + 1) / 2} \exp {- \frac{1}{2} t r (V {Σ_{j}}^{- 1})}] \times [Π_{i = 1}^{I} Π_{j = 1}^{J} {π_{j}}^{I_{τ_{i} = j}}] \times [π_{1}^{a_{1} - 1} π_{2}^{a_{2} - 1} \dots {π_{J}}^{a_{J} - 1}] \end{matrix}

(3)

where $P (y_{i} | β, Σ, τ_{i})$ is the individual likelihood function which has multivariate linear regression, $n_{j}$ is the number of observations fitted to path $j$ , $Q_{j}$ is the homogeneous path type group defined as $Q_{j} = {i | τ_{i} = j, i = 1, \dots, I}$ , and $π = {π_{1}, \dots, π_{j}}$ is the hyper-parameter of $τ_{i}$ , indicating a proportion of each path type over the total individual. The full posterior is not analytically tractable, and therefore, we used one of the MCMC methods. Further details regarding prior distribution, likelihood function, and estimation procedure are provided in the Appendix.

To apply the proposed theoretical and statistical model, we collected survey data related to cognitive, affective, and conative responses regarding consumers’ attitude toward a dedicated E-book reader device. In addition, we measured five perceived attributes of innovation toward a dedicated E-book reader to identify significant key innovation attributes affecting E-book reader adoption behavior according to the specific transferal process. Our reason for choosing a dedicated E-book reader for empirical application of the proposed model is twofold. First, our literature review for the theoretical background showed that previous studies suggest the possibility of different decision-making processes depending on product involvement. This means that if we select products with exceedingly low product involvement called habitual products, such as cigarettes, the product can be strongly linked to a conative first pattern (Nayeem & Casidy, 2015). Conversely, if we choose fashion products, for example, that are too expensive or luxurious, it is very likely to be strongly linked to the affective first pattern. Therefore, to avoid such strong linkages, we chose a product category which can sit across the four FCB quadrants. This implies that the purchase of a dedicated E-book reader does not depend on a specific decision-making process and can appear in various forms, depending on the consumer. Second and more importantly, a few studies attempted to identify factors affecting awareness, interest, and intention to use within the Korean E-book reader market (Jung et al., 2012; Lee, 2013; Shim et al., 2016). In those studies, authors assumed the standard hierarchy of effect model in their research designs, neglecting the possibility of any other decision-making process. Such a strong assumption may lead to biased results by aggregating different decision-making paths into the hierarchy of effect model. This study will correct this limitation and attempt to empirically show the possibility of various decision-making processes in the E-book reader market in Korea.

We conducted a survey online and 536 respondents completed the questionnaire. Respondents were aged from 20 to 59 years and living in the Seoul area. The age range was set at 20 to 59 because consumers older than 60 years have relatively low rates of use for smart devices and consumers younger than 19 years have relatively low purchasing power (Shin et al., 2016). Equal numbers were assigned to males and females and 268 of each gender were surveyed. The sample statistics of this study are shown in Table 1.

Table 1.

Characteristics of the survey sample.

Attributes	Respondents	Percentage
Gender
Male	268	50.0
Female	268	50.0
Age
20–29	179	33.4
30–39	180	33.6
40–49	107	20.0
50–59	70	13.0
Education level
High school	68	12.7
College	103	19.2
University	313	58.4
Above graduate school	52	9.7
Monthly household income level (10,000 KRW)
Below 100	72	13.4
100–199	95	17.7
200–299	128	23.9
300–399	140	26.1
Above 400	101	18.9
Total	536	100%

Data were collected using a structured survey questionnaire with questions answered on a Likert-type scale to measure the research variables. This survey allows us to gauge the three components of the decision-making process and perceived innovation attributes that could otherwise not be directly quantified. Specifically, a Likert-type scale of seven points was used to measure the construct, with 1 indicating “completely disagree” and 7 indicating “strongly agree.” The survey was conducted by a specialized survey company that has more than 200,000 panelists in Korea. Respondents were offered a small amount of electronic cash as payment for participating in the survey (Table 2).

Table 2.

Variable measurement items.

Variable	Operational definition	Mean (variance)
Component of attitude
Cognitive response	I know about E-BR device	4.2994 (2.1254)
Affective response	I have a favorite with E-BR device	3.9683 (1.8255)
Conative response	I have intention to use E-BR device	4.0933 (2.0211)
Perceived innovation attributes
Relative advantage	Using E-BR makes it easier to read text than paper-based book	4.1940 (1.925)
Compatibility	Using E-BR is compatible with my lifestyle	3.7761 (2.0282)
Complexity	E-BR is not easy to use	3.6063 (1.1753)
Observability	I have seen E-BR demonstrations at stores	3.3731 (1.9541)
Trialability	I would like to have an opportunity to try E-BR	3.4235 (1.9046)

To estimate the suggested Bayesian model in an empirical analysis, Gibbs sampling was applied. Gibbs sampling is one of the popular MCMC methods typically used when direct sampling of the joint posterior distribution is intractable, but sampling from the full conditional distribution of each parameter is reasonably straightforward. By iteratively sampling from each conditional distribution in turn, samples of the joint posterior distribution are indirectly obtained. In this way, parameter samples are repeatedly drawn until it is decided that a reasonable representation of the joint posterior distribution has been obtained. To do this, we ran 80,000 draws from the conditional posterior distributions. Every 10th value was chosen from the draws for parameter inference after discarding the initial 40,000 draws as the burn-in period. See Appendices 1 and 2 for more details about model specification and estimation procedure using Gibbs sampling.

Empirical analysis

Using the proposed statistical model, we were able to successfully identify reliable decision-making paths using the survey data set on a dedicated E-book reader. In the proposed model, a decision-making path is feasible if the transferal process between each stage and the next is statistically significant. For example, if the decision-making path from the cognitive to the affective stage and from the affective to the conative stage has a significance influence, such a decision-making path—the hierarchy of effect model—is a feasible decision-making path. This means that some consumers follow such a decision-making path in the E-book reader market. Similarly, for decision-making Path 5 (Conative → Cognitive → Affective) to be a reliable path, both the transition from conative stage to cognitive stage and from cognitive stage to affective stage must be significant. This means that the “transition weight” in Tables 3 to 5 are the variables that determine whether a transferal process from one stage to the next stage is feasible for a hypothetical decision-making path. Accordingly, if both transition weight variables are statistically significant, it can be interpreted that the decision-making path is feasible in the E-book reader market. Conversely, if the transition weight variable is not statistically significant, the decision-making path is not feasible. The estimation results for six different decision-making paths are provided in Tables 3 to 5 and are described in diagram form in Figures 3 to 5.

Table 3.

Estimation results of decision-making Path 1 and decision-making Path 2.

	Path 1 (cognitive → affective → conative)			Path 2 (cognitive → conative → affective)
	Cognitive	Affective	Conative	Cognitive	Affective	Conative
Constant	2.6565 * (0.4166) 2[1.971, 3.341]			1.8251 * (0.6118) [0.817, 2.838]
Constant		2.2707 * (0.2716) [1.824, 2.717]			−1.0002* (0.0435)[−1.071, −0.928]
Constant			0(0.0144)[−0.024, 0.024]			3.597 * (0.3212) [3.073, 4.127]
Relative Advantage	0.0179(0.0811)[−0.115, 0.152]			0.1828 * (0.1009) [0.018, 0.350]
Compatibility	0.1253(0.0796)[−0.006, 0.257]			0.0017(0.096)[−0.157, 0.160]
Complexity	−0.1245(0.0828)[−0.261, 0.012]			−0.0039(0.1084)[−0.181, 0.176]
Observability	0.3076 * (0.0818) [0.172, 0.443]			0.2812 * (0.1007) [0.115, 0.446]
Trialability	0.1585 * (0.084) [0.020, 0.296]			0.1739(0.1061)[−0.002, 0.347]
Transition weight		0.4343 * (0.0603) [0.335, 0.5331]				0.289 * (0.0729) [0.169, 0.408]
Transition weight			1 * (0.0033) [0.995, 1.005]		1 * (0.0089) [0.985, 1.015]
Sigma	0.5334 * (0.7597)	0.4704 * (0.6715)	0.0015 * (0.0021)	0.4639 * (0.6664)	0.003 * (0.0043)	0.2932 * (0.4214)
Phi (Market share)	0.35 (0.0173) *		0.2 (0.0148) *

Significant at 90% credible interval; number parentheses is standard error; square bracket is 90% credible interval.

Bold values are Significant at 90% credible interval.

Table 4.

Estimation results of decision-making Path 3 and decision-making Path 4.

	Path 3 (affective → cognitive → conative)			Path 4 (affective → conative → cognitive)
	Cognitive	Affective	Conative	Cognitive	Affective	Conative
Constant	1.326 * (0.561) [0.444, 2.272]			2.662 * (0.980) [0.993, 4.203]
Constant		0.008(0.894)[−1.419, 1.502]			3.105 * (1.074) [1.376, 4.873]
Constant			2.176 * (0.5766) [1.428, 3.2373]			−1.786* (0.418)[−2.449, −1.085]
Relative Advantage		0.116(0.237)[−0.245, 0.529]			0.217(0.208)[−0.130, 0.552]
Compatibility		−0.079(0.286)[−0.578, 0.358]			0.012(0.246)[−0.376, 0.433]
Complexity		0.174(0.225)[−0.203, 0.538]			−0.089(0.210)[−0.428, 0.256]
Observability		0.087(0.203)[−0.241, 0.423]			0.481 * (0.198) [0.152, 0.802]
Trialability		0.490 * (0.186) [0.207, 0.805]			−0.101(0.228)[−0.484, 0.267]
Transition weight	0.922 * (0.207) [0.564, 1.251]					0.921 * (0.085)
Transition weight			0.6293 * (0.1461) [0.3539, 0.81]	0.5082 * (0.3074) [0.010, 1.021]
Sigma	0.3072 * (0.5007)	0.3273 * (0.5205)	0.1579 * (0.3167)	0.807 * (1.2291)	0.3204 * (0.5084)	0.0845 * (0.1316)
Phi (Market share)	0.0913 (0.021) *			0.0958 (0.0117) *

Significant at 90% credible interval; number parentheses is standard error; square bracket is 90% credible interval.

Bold values are Significant at 90% credible interval.

Table 5.

Estimation results of decision-making Path 5 and decision-making Path 6.

	Path 5 (conative → cognitive → affective)			Path 6 (cognitive → affective → cognitive)
	Cognitive	Affective	Conative	Cognitive	Affective	Conative
Constant	7.186 * (0.453) [6.447, 7.900]			1.7311 * (0.5265) [0.865, 2.6032]
Constant		3.459 * (0.521) [2.600, 4.310]			1.0001 * (0.0366) [0.9402, 1.0604]
Constant			1.852(1.325)[−0.306, 4.050]			0.543(0.5085)[−0.2907, 1.38]
Relative Advantage			0.576 * (0.298) [0.095, 1.071]			0.2773 * (0.0989) [0.114, 0.438]
Compatibility			0.134(0.283)[−0.324, 0.598]			0.183 * (0.0969) [0.0228, 0.3404]
Complexity			−0.023(0.292)[−0.488, 0.471]			−0.2084* (0.1197)[−0.402, −0.0086]
Observability			0.073(0.262)[−0.352, 0.502]			0.3599 * (0.0976) [0.1979, 0.5203]
Trialability			−0.297(0.279)[−0.759, 0.148]			0.136(0.1057)[−0.0375, 0.3105]
Transition weight	−0.797* (0.096)[−0.949, −0.641]				1 * (0.0103) [0.9831, 1.017]
Transition weight		−0.188(0.128)[−0.394, 0.022]		0.6393 * (0.1156) [0.4471, 0.8292]
Sigma	0.3338 * (0.5062)	0.4942 * (0.7866)	1.0898 * (1.6553)	0.495 * (0.7152)	0.0043 * (0.0062)	0.2606 * (0.3771)
Phi (Market share)	0.1058 (0.0134) *		0.1558 (0.0132) *

Significant at 90% credible interval; number parentheses is standard error; square bracket is 90% credible interval.

Bold values are Significant at 90% credible interval.

Figure 3.

Diagram description of transferal decision-making path: Path 1 and Path 2.

Figure 4.

Diagram description of transferal decision-making path: Path 3 and Path 4.

Figure 5.

Diagram description of transferal decision-making path: Path 5 and Path 6.

Next, we determine how many people will follow a decision-making path. In other words, when the market is segmented based on consumers’ decision-making path, the size of market—or market share—is one of the important factors in the implementation of market segmentation and targeting strategy. Because the proposed statistical model is an extended application of the standard Gaussian mixture model, the market share can be easily evaluated statistically (Choi et al., 2013; Fader, 1993). We can perform statistical inference on market share using the conjugate prior relationship between Dirichlet—a multinomial distribution (see fourth step, Appendix 2). Consequently, the variable “Phi” in Tables 3 to 5 indicates market share when we evaluate the segmented market in terms of heterogeneous decision-making paths.

Remarkably, all decision-making paths were feasible, except Path 5: conative → cognitive → affective. Moreover, the traditional hierarchy response model had the largest market share with 35% while the affective → cognitive → conative path ranked at the bottom, with 9.13%. In addition, based on the decision-making paths, we determined key innovation attributes which must be emphasized by marketers. For example, enhancing observability and trialability is an effective strategy for those who follow the cognitive → affective → conative path, to stimulate consumers to formulate a positive attitude toward the new product.

Tables 3 to 5 show the estimation results in detail. Based on the estimation results, Figures 3 to 5 depict diagrams of the decision-making paths.

Path 1 depicts the standard hierarchy of effect model. Those who follow this path may consider a dedicated E-book reader as a high involvement product, where thinking and economic considerations prevail (Vaughn, 1986). Our estimation showed that observability and trialability had a positive and significant impact on the cognitive stage in this path; this implies that emphasizing observability and trialability in advertisements stimulates purchase intention in consumers who follow Path 1. Moreover, the fact that observability and trialability are key factors to be emphasized in marketing communication means that the E-book reader is still an unfamiliar product in the market and that consumers are not aware of the usage and features of this product. The potential market share of Path 1 was estimated to be 35%, the biggest share among the six different paths. Accordingly, potential buyers are most likely to follow the standard hierarchy of effect path in their E-book reader adoption.

Path 2 depicts the purchase decision-making path in the order of cognitive, conative, and affective stages; this means that once consumers understand the usage of E-book readers, they experience intention to use and then rationalize this intention by building preference for it. In our empirical analysis, both the transferal process between cognitive and conative stages and the transferal process between conative and affective stages were statistically significant. This confirms that Path 2 is a reliable decision-making path for a certain market share in the E-book reader market. Relative advantage and observability were statistically significant as innovation characteristics affecting the cognitive stage in Path 2. In terms of market share, the market size for Path 2 was estimated to be 20%. This places Path 2 second among the decision-making paths.

Path 3 depicts the decision-making path of affective, cognitive, and conative. This path was found to be statistically significant, as the transferal process between affective and cognitive and between cognitive and conative were all statistically significant. Of the five innovation characteristics, trialability is the only significant factor influencing the affective stage in Path 3. One possible explanation of this result is that allowing consumers to try out E-book reader devices induces a positive attitude toward an E-book reader and stimulates the adoption of devices for those who follow decision-making Path 3. The market share of Path 3 was 9.13%. Although this figure is the smallest, it is not negligible.

Path 4 depicts the decision-making sequence of affective, conative, and cognitive. The empirical analysis showed that Path 4 was also a feasible decision-making path by confirming that the transferal processes between affective and conative and between conative and cognitive were statistically significant. This path illustrates that an affective response does not always require prior cognition and that it is possible to adopt a product without an understanding of its usage and features. Given that in Path 4, preference is decided on an affective basis alone and it directly influences conative reaction without a cognitive stage, consumers following this decision-making path perceives the E-book reader as primarily expressive or for delivering sensory pleasure, rather than being utilitarian. Observability is the only significant innovation characteristic affecting the affective stage in Path 4. Simply put, this result implies that enhancing observability in offline stores and online shops can stimulate decision-making for those who follow Path 4, since this group would not want to understand specific utilitarian benefits related to feature and usage. The market share of Path 4 over population is 9.58%. This figure is small but cannot be ignored in market segmentation.

Path 5 represents the conative, cognitive, and affective sequence, whereas Path 6 represents the conative, affective, and cognitive sequence. The assumption that the conative factor first comes to consumers’ minds is the same in Path 5 and Path 6. The difference between two paths is that the next is cognitive in Path 5 and affective in Path 6. In the estimation results, the decision-making process of Path 5 was not supported, whereas Path 6 was supported. In Path 5, the transition between conative and cognitive is significantly negative and the transition between cognitive and affective is not statistically significant. Berger (1986) stated that Path 5 is more likely to apply to habitual products for which deeper learning is not necessary, such as household cleaners or gasoline, for example. In contrast, an E-book reader purchase is not related to routinized consumers and their purchase decision-making. Therefore, we can state that this decision-making path is not feasible in E-book reader adoption. It is notable that Path 6 is supported, even though both Path 5 and Path 6 start with the conative stage. The fact that the transferal process of Path 5 is insignificant can be interpreted as follows: if consumers intend to use E-book readers first, it is more likely that they will transfer to the affective stage, rather than cognitive stage, as consumers tend to form attitudes to bolster action. Moreover, selective learning follows to support that behavior (Table 6).

Table 6.

Summary of empirical analysis findings on the Korean E-book reader market.

Decision-making paths	Feasibility	Market share (%)	Meaningful attributes influencing decision-making
Path 1	Feasible	35	Observability, Trialability
Path 2	Feasible	20	Relative advantage
Path 3	Feasible	9.1	Trialability
Path 4	Feasible	9.6	Observability
Path 5	Unfeasible
Path 6	Feasible	15.6	Relative advantage, Compatibility, Complexity, Observability

Discussion and conclusion

The results of our study show that among the six possible transferal paths of the decision-making process, every path was statistically significant, except for Path 5: conative → cognitive → affective. Path 5 is an understandable and logical path for low involvement and thinking products with routinized consumer behavior that does not require deep learning about a product. Therefore, this decision-making process is not fit for E-book reader adoption behavior. In terms of market share, Path 1—cognitive → affective → conative—has the largest market share, followed by Path 2. According to our results, 55.02% of potential consumers form cognition about a product by accumulating knowledge regarding the relevant attributes of E-book readers. In other words, the cognitive stage is the first that comes to mind for more than half of the respondents.

An interesting point highlighted by the empirical results is that the conative → cognitive → affective path is not reliable; this path’s non-significance implies that consumers do not perceive E-book readers as a habitual product, such as food or household cleaning items. This decision-making path has often been referred to as “passive learning hierarchy” because it does not require cognitive effort and is often selected out of habit. Therefore, passive learning is not a factor when consumers purchase an E-book reader. Another possible interpretation of this result is that when consumers experience impulsive purchase intention regarding an E-book reader in which conative reaction comes first, they are more likely to follow it with an affective and a cognitive reaction, sequentially, rather than a cognitive and an affective reaction.

Our study also successfully identifies key innovation characteristics which marketers should emphasize, depending on the type of decision-making path. For instance, enhancing observability and trialability in the market is an effective communication strategy for those who follow the traditional hierarchy of effect model (Path 1).

This study contributes to marketing literature in both theoretical and practical dimensions. From a theoretical perspective, we developed an estimable model for capturing the possible transferal path in the decision-making process. Using survey data, we also empirically show that there exists heterogeneity in individual decision-making paths. From a practical perspective, capturing the heterogeneity of the decision-making process provides a good basis for customer segmentation. Furthermore, our model enabled us to obtain insights into the advertisements for new product adoption. In the context of innovation diffusion, the impact of perceived innovation attributes in new product adoption can vary according to a consumer’s decision-making path.

We must note, however, that our study has several limitations. First, the model only includes perceived innovation characteristics as factors that influence certain decision-making process. It is possible that not only innovation attributes but also source credibility, such as reliability and expertise, may influence decision-making process. Extensive data collection related to message characteristics will contribute to market research. Second, we did not model consumers’ motivation for selecting a certain path; what makes a consumer follow a certain decision-making process, rather than others. Incorporating underlying motivations for choosing different decision-making paths would be a good avenue for future research, as little is known about the reasons of individual heterogeneity in decision-making paths.

It should be noted that the dedicated E-book reader market was chosen to include as many as possible decision-making paths among the six possible transferal paths in this study. However, as the proposed statistical model is a very general and flexible model, it can be applied to any product purchase situation. In the case of purchasing an expensive electric vehicle, there is high possibility that the majority of potential consumers would follow the standard hierarchy of effect decision-making path. The model proposed in this study can be applied to this situation to test if that decision-making path is reliable or not. Furthermore, not only innovation attributes—such as observability and compatibility—but also promotion—such as advertising and PR—can be chosen as stimuli for consumer decision-making. For example, instead of choosing innovation characteristics, advertisement characteristics such as reliability and trustworthiness can be included in the model to determine what decision-making paths are affected by such message characteristics.

In conclusion, in this study, we have taken the initial steps to account for heterogeneity in consumers’ decision-making path when considering adoption behavior of a high-tech new product. We hope that this encourages future researchers to address some of the unresolved issues raised here.

Footnotes

Appendix 1

Appendix 2 Acknowledgements

This paper was supported by Konkuk University in 2020.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Dongnyok Shim

References

Allenby

G. M.

Rossi

P. E.

(1998). Marketing models of consumer heterogeneity. Journal of Econometrics, 89(1–2), 57–78.

Barry

T. E.

(1987). The development of the hierarchy of effects: An historical perspective. Current Issues and Research in Advertising, 10, 251–295.

Barry

T. E.

Howard

D. J.

(1990). A review and critique of the hierarchy of effects in advertising. International Journal of Advertising, 9, 121–135.

Bauer

H. H.

Sauer

N. E.

Becker

(2006). Investigating the relationship between product involvement and consumer decision-making styles. Journal of Consumer Behaviour, 5, 342–354.

Beal

G. M.

Rogers

E. M.

(1960). The adoption of two farm practices in a central Iowa community. https://core.ac.uk/download/pdf/83024227.pdf

Bem

D. J.

(1972). Self-perception theory. In Berkowitz

(Ed.), Advances in experimental social psychology (pp. 183–200). Academic Press.

Berger

(1986). Theory into practice: The FCB grid. European Research, 14, 35–46.

Choi

J. Y.

Shin

Lee

(2013). Strategic management of new products: Ex-ante simulation and market segmentation. International Journal of Market Research, 55, 289–314.

Fader

P. S.

(1993). Integrating the Dirichlet-multinomial and multinomial logit models of brand choice. Marketing Letters, 4, 99–112.

10.

Jain

Sharma

(2013). A review on Sproles & Kendall’s consumer style inventory (CSI) for analyzing decision making styles of consumers. Indian Journal of Marketing, 43, 40–47.

11.

Jung

Chan-Olmsted

Park

Kim

(2012). Factors affecting e-book reader awareness, interest, and intention to use. New Media & Society, 14, 204–224.

12.

Kelley

H. H.

(1973). The processes of causal attribution. American Psychologist, 28, 107–128.

13.

Kiesler

C. A.

(1971). The psychology of commitment: Experiments linking behavior to belief. Academic Press.

14.

Krugman

H. E.

(1965). The impact of television advertising: Learning without involvement. Public Opinion Quarterly, 29, 349–356.

15.

Krugman

H. E.

(1966). The measurement of advertising involvement. Public Opinion Quarterly, 30, 583–596.

16.

Lavidge

R. J.

Steiner

G. A.

(1961). A model for predictive measurements of advertising effectiveness. Journal of Marketing, 25, 59–62.

17.

Lee

(2013). An integrated adoption model for e-books in a mobile environment: Evidence from South Korea. Telematics and Informatics, 30, 165–176.

18.

Nayeem

Casidy

(2015). Australian consumers’ decision-making styles for everyday products. Australasian Marketing Journal, 23, 67–74.

19.

Park

Y. A.

Gretzel

(2008). Investigating the effects of product type on online decision-making styles. In O’Connor

Höpken

Gretzel

(Eds.), Information and communication technologies in tourism 2008 (pp. 509–520). Springer.

20.

Ratchford

B. T.

(1987). New insights about the FCB grid. Journal of Advertising Research, 27, 24–38.

21.

Ray

M. L.

Sawyer

A. G.

Rothschild

M. L.

Heeler

R. M.

Strong

E. C.

Reed

J. B.

(1973). Marketing communication and the hierarchy-of-effects. In Clarke

(Ed.), New models for mass communication research (pp. 147–176). SAGE.

22.

Rossi

(2014). Bayesian non-and semi-parametric methods and applications. Princeton University Press.

23.

Shim

Kim

J. G.

Altmann

(2016). Identifying key drivers and bottlenecks in the adoption of E-book readers in Korea. Telematics and Informatics, 33, 860–871.

24.

Shin

Park

Lee

(2016). Strategic management of over-the-top services: Focusing on Korean consumer adoption behavior. Technological Forecasting and Social Change, 112, 329–337.

25.

Sprotles

G. B.

Kendall

E. L.

(1986). A methodology for profiling consumers’ decision-making styles. Journal of Consumer Affairs, 20, 267–279.

26.

Vakratsas

Ambler

(1999). How advertising works: What do we really know? Journal of Marketing, 63, 26–43.

27.

Vaughn

(1980). How advertising works: A planning model. Journal of Advertising Research, 20, 27–33.

28.

Vaughn

(1986). How advertising works: A planning model revisited. Journal of Advertising Research, 26, 57–66.

29.

Zajonc

R. B.

(1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35, 151–175.

30.

Zajonc

R. B.

(1984). On the primacy of affect. American Psychologist, 39, 117–123.

31.

Zajonc

R. B.

Markus

(1982). Affective and cognitive factors in preferences. Journal of Consumer Research, 9, 123–131.