Safiand Yu 2017

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 43

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/316077327

Online product review as an indicator of users’ degree of innovativeness and


product adoption time: a longitudinal analysis of text reviews

Article  in  European Journal of Information Systems · April 2017


DOI: 10.1057/s41303-017-0045-2

CITATIONS READS

4 144

2 authors, including:

Roozmehr Safi
University of Missouri - Kansas City
6 PUBLICATIONS   16 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Web Analytics View project

Framing View project

All content following this page was uploaded by Roozmehr Safi on 29 January 2019.

The user has requested enhancement of the downloaded file.


Online Product Review as an Indicator of Users’ Degree of Innovativeness and Product Adoption

Time: A Longitudinal Analysis of Text Reviews

Abstract

Online reviews have become extremely valuable sources of information about products and their

customers as electronic commerce continues to proliferate rapidly. Previous research has shown that

reviews of a product change and evolve over its life. Identifying and understanding patterns of change in

reviews and the forces that shape them is an underexplored topic with substantial potential for predicting

and improving the market performance of products. In this study, we analyze review text of nearly 50

products over the course of their lives. Our longitudinal analysis of reviews reveals changes in certain

personality-related characteristics of buyers in ways that are consistent with the predictions of product

adoption and diffusion theories. The main findings and conclusions still hold when we replicate the same

procedure on reviews of a different product category. Accordingly, based on online user-generated

content (UGC) in the form of online reviews, this research introduces a novel empirical method for

identifying the product adoption and diffusion stage. Implications of the study for theory, methodology,

and practice are discussed.

Keywords: online reviews, user generated content (UGC), product adoption, product life cycle

(PLC), content analysis

1
Online Product Review as an Indicator of Users' Degree of Innovativeness and Product Adoption

Time: A Longitudinal Analysis of Text Reviews

Introduction

Today, more and more buyers base their purchasing decisions on online reviews contributed by

other customers. Apart from the perceived unbiasedness and truthfulness of consumer reviewers (Bickart

& Schindler, 2001), these reviews are unique as they evaluate products in the context of everyday

purchasing and usage scenarios (Yubo & Jinhong, 2008). From this perspective, reviews reflect trends

and topics that are currently relevant to the general population of actual and potential buyers of a product

(Gamon et al., 2005). The user-centric and “live” nature of reviews provides many untapped opportunities

for vendors. Researchers analyze reviews to gain insight about buyers and their attitudes toward products

and to forecast product sales performance (Chevalier & Mayzlin, 2006; Dellarocas et al., 2007). Reviews

can be analyzed from different perspectives. One such perspective is the longitudinal study of reviews. A

typical product receives numerous reviews over time. It has been shown that these reviews are not just a

collection of thoughts arriving at random times; rather, their arrivals follow certain patterns and trends

(Godes & Silva, 2012). In this paper, we demonstrate that such trends can be the result of the changing

product customer base over time.

From a temporal perspective, buyers can be broadly grouped according to the time of their

purchase. Past research has shown that members of these different buyer groups share certain

characteristics (Moore, 1991). For example, while earlier groups of consumers (called innovators and

early majority) tend to be more risk seeking, optimistic, and affluent, later adopters (called the late

majority and laggards) tend to be more conservative, pessimistic, and concerned about price (Martinez et

al., 1998; Rogers, 1963; Wei, 2001). Demographic, psychographic, and economic characteristics such as

these collectively determine what researchers call a “buyer’s degree of innovativeness,” which is a key

determinant of timing of purchase (Rogers, 1976). Distinguishing between buyer groups from the

2
perspective of degree of innovativeness has significant theoretical and practical importance. In particular,

it allows for identifying the stage of products on product lifecycles (PLC) (Moore, 1991). A typical

product’s life can be divided into several stages, starting with the introduction stage and ending with the

decline stage. PLC stage is considered one of the most important contextual variables in product strategy

formulation and execution (Hoffer, 1975), and selecting the appropriate strategy in each product life stage

leads to a healthier and longer life for that product. However, determining the stage of a life cycle during

the life of a product has been a longstanding practical challenge (Day, 1981). In a recent critical review of

the literature of diffusion of innovation, scholars in this area have suggested that user-generated content

(UGC) from the Internet may have the potential to help identify the PLC stage (Peres et al., 2010). This

paper can be considered as one of the first to answer the call by those researchers.

In this article, we investigate if and how the major trait-like (innate) innovativeness-related

psychological characteristics of buyers, as reflected in their reviews, change over the life of a product. To

that end, we conduct a longitudinal text analysis of online reviews over the lifetime of nearly 50 different

products. From the methodological perspective, we apply text mining techniques to measure certain

innovativeness-related psychological characteristics of reviewers. Results show that when viewed

longitudinally, psychological characteristics of reviews exhibit trends that match the thought process of

different groups of buyers as described by the product adoption theory (Rogers, 2003). To test the

generalizability of our findings, we extended the proposed analysis to another group of buyers, who

purchased a different product category (90 different products), and observed similar patterns. Our work

can be regarded as a promising step in identifying the stage of product adoption, and subsequently,

product life stage using UGC. To the best of our knowledge, this is the first comprehensive study of its

type.

Adoption and diffusion theories are among the oldest, most intuitive, and most powerful tools for

researchers and practitioners. For the first time, this paper employs these theories to discover and explain

certain trends in user-generated content in the form of product reviews. Accordingly, this study provides a

novel way of observing and interpreting the evolution of reviews over time. The remainder of this article

3
is structured as follows. First, we review the literatures of online reviews, product diffusion, and product

life cycle (PLC). We then present a framework to demonstrate how reviews form and evolve over the life

of a typical product. Next we draw upon the product adoption theory to develop our hypotheses in the

context of online reviews. This is followed by the introduction of our dataset and a description of our

operationalization of variables. We then present our findings and discussion of the results. Finally, we lay

out the implications of our key findings for theory, methodology, and practice and discuss the limitations

of our study as well as opportunities for future research.

Literature Review and Theoretical Foundation

Online Reviews

Over the years, along with the increasing importance of online product reviews, the number and

diversity of studies in this area have risen sharply. Research in this domain can be classified under two

broad categories: (1) the study of impact of online reviews on sales and (2) the study of antecedents of

online reviews (Godes & Silva, 2012). While a large body of studies (e.g., Chevalier & Mayzlin, 2006;

Chintagunta et al., 2010; Dellarocas et al., 2007; Duan et al., 2008; Çelen & Kariv, 2004, 2005) has been

devoted to the study of the former category, according to Godes and Silva (2012), fewer studies have

investigated the latter, that is, the process behind the formation of reviews themselves. A recent survey of

the representative literature in the domain of online reviews and electronic word of mouth (Gao et al.,

2015), for example, has identified five times as many papers in the former category than in the latter.

Even though there have been many impactful publications in this domain, (Dellarocas et al., 2010; Goes

et al., 2014), more research is needed in this area since the relative scarcity of research into the

antecedents of reviews inevitably precludes us from obtaining a deep, holistic insight about the possible

impacts that reviews can ultimately have on commercial performance of products.

In the domain of product adoption, the literature has identified a host of intra- and interpersonal

factors influencing purchase and its timing (Rogers, 2003; Xu et al., 2008). Interpersonal factors deal

4
primarily with the effects of the social system on forming buyers’ purchase decisions. Such interpersonal

factors have been studied at the level of dyadic interactions among individual buyers (a micro-level view),

for example, in the form of (electronic) word of mouth (Hennig‐Thurau et al., 2004), or at a level beyond

dyadic interactions to study the broader effects of the social system on buyers (a macro-level view) (e.g.,

Rogers & Kincaid, 1981) (for a comprehensive review of the recent literaure, please see Gao et al., 2015).

Intrapersonal factors affecting the time of purchase, on the other hand, deal with personal characteristics

of individual buyers. The present study is concerned primarily with intrapersonal factors involved in

forming purchasing decisions.

Past research has shown that reviews evolve, or change systematically, over time (Godes & Silva,

2012). This evolution has been viewed from two main perspectives: sequential and temporal. While

sequential effects account for the impact of the already existing reviews on incoming reviews, temporal

effects deal with the mere effects of the passing of time on incoming reviews, in other words, the effect of

the length of time that a product has been available in the market. Among the works done in the latter

category is a notable stream of research that deals with the possible effects of self-selection in timing of

purchase on review content (Li & Hitt, 2008; Moe & Schweidel, 2012; Moon et al., 2010; Nan et al.,

2009). This type of self-selection occurs when individuals who share certain characteristics independently

make their purchasing decisions at a certain time. In this study, we seek to learn if different degrees of

innate innovativeness on the part of individuals can cause such a self-selection.

Buyer Adoption and Product Life Cycle (PLC)

According to the literature of product adoption and diffusion, different classes of buyers adopt a

product or an idea at different times after its release (Rogers, 2003). Moore (1991) classifies buyers from

this perspective into five groups: innovators, early adopters (also called visionaries), early majority, late

majority, and laggards. Rogers attributes the observed differences in the timing of adoption to differences

in the degree of innovativeness on the part of buyers. He defines innovativeness as “the degree to which

an individual or other unit of adoption is relatively earlier in adopting new ideas than other members of a

social system” (2003, p. 267). He describes certain dimensions of innovativeness as personality traits,

5
while also acknowledging the role of contextual factors in adoption decisions. Others have made a clearer

distinction between different “levels” of innovativeness. In their conceptual work, Midgley and Dowling

(1978) explored the nature of innovativeness and its relationship to product adoption. In that work, they

propose conceptualizing innovativeness at different levels of abstraction. From their perspective, Rogers’s

definition, provided earlier in this paragraph, resides at the lowest level of abstraction. They refer to this

as “actualized innovativeness” (aka observed innovativeness). However, when treating innovativeness

strictly as a trait, or a general, global predisposition, Midgley and Dowling refer to it as “innate

innovativeness.” Innate innovativeness is innovativeness at its highest level of abstraction. Midgley

(1977, p.49) defines innovativeness at this abstract level as “the degree to which an individual makes

innovation decisions independently of the communicated experience of others,” or an adoption behavior

that one would see in the complete absence of contextual or social effects. According to this view, every

member of the social system possesses a higher or lower degree of innate innovativeness, and depending

on the amount of innate innovativeness that one possesses, it may take lesser or greater amounts of

contextual factors or interpersonal communication for that person to make the adoption/rejection decision.

Collectively, individual buyers’ adoption decisions influenced by their degree of innovativeness

form a product’s demand. Demand for a product can be predicted using the product life cycle framework,

which divides the life of a product into a number of conceptual stages. These are introduction, growth,

maturity, and finally the decline stage (Doyle, 1976), each with its special characteristics. The pattern of

product life cycles can take different forms for different types of products (Day, 1981). Some factors that

can affect the shape are unexpected takeoff or slowdown of adoption due to customer herding behavior

and informational cascading (Duan et al., 2009; Golder & Tellis, 2004), the introduction of new products

from competitors, the introduction of disruptive technologies (Bower & Christensen, 1995), or an overall

change in the target consumer’s disposable income (Golder & Tellis, 2004). However, it is generally

accepted that the form of the adoption curve is bell-shaped (Golder & Tellis, 2004). In fact, previous

research has shown that products that enjoy a higher growth rate at the introduction stage suffer a sharper

drop at the decline stage, which accounts for the symmetric shape of the curve (Golder & Tellis, 2004).

6
Product adoption and PLC are closely related concepts with significantly overlapping theoretical

foundations. In his study of the relationship between PLC and strategy, Fox (1977) identified the main

groups of adopters as the main groups of buyers during each stage of PLC (e.g., innovators are the main

buyer group in a product’s introduction stage, early adopters are the main group of buyers in the early

growth stage, etc.). In fact, he considers a near one-to-one mapping between different adopting groups

and PLC stages. Figure 1, adopted from Moore (1991), illustrates the point.

Early Early Late

Innovators Adopters Majority Majority Laggards

Introduction Early Growth Late Growth Maturity Decline

Figure 1. The correspondence between adoption stage and PLC curve.

Even though over the decades researchers have prescribed a host of tactics for properly managing

every stage of PLC--e.g., engaging in informative advertising during the introduction stage, engaging in

persuasive advertising during the growth stage, etc. (for a comprehensive discussion see Achrol & Kotler,

1999)-- to date there is no agreed-upon method for determining the stage of product life cycle in the first

place. Typically, sales value or volume has been used to assess the position of a product on the life cycle

(Achrol & Kotler, 1999). However, a rise or a decline in short- or even middle-term sales volume, per se,

does not necessarily reveal the underlying dynamics of product diffusion (Golder & Tellis, 2004). This

7
necessitates the development of more viable methods for assessing product adoption and consequently

PLC stages. In the following section, we present our new approach, which is based on exploiting the

information embedded in user reviews.

Rationale of the Study and Hypotheses

At any given time, online reviews reflect trends and topics that are currently relevant to the

general population of actual and potential buyers of that product (Gamon et al., 2005). This implies that

temporally adjacent reviews are similar in certain ways. We argue that among these similarities should be

those characteristics that collectively form buyers’ degrees of innate innovativeness. Rogers (1983) has

developed profiles for “ideal types” of adopter groups. His typography provides a detailed description of

personality (e.g., attitude toward change or uncertainly), demographic (e.g., age), social (e.g., education

level, social status), and economic (e.g., income) characteristics of each group of adopters. Researchers in

this domain have relied mainly on surveys or direct observation of subjects to obtain data about these

characteristics. In the online environment too, some researchers have attempted to identify certain

innovativeness-related characteristics of buyers such as their social characteristics (e.g., their level of

education) from the information that buyers explicitly disclose in their reviews (Pan & Zhang, 2011).

Nevertheless, buyers rarely express such characteristics in their writings. Therefore, the effectiveness of

this method is very limited.

To overcome these limitations, we focus on personality-related characteristics for determining

buyers’ degree of innate innovativeness. Innate innovativeness is conceptualized as a global personality

trait that influences all adoption behaviors regardless of product type or class. This approach is especially

valuable since compared to other buyer characteristics (e.g., economic, demographic), personality

variables have historically been examined the least in adoption studies (Rogers, 2003). In this study, we

draw inferences about buyer personality characteristics from their writings. Analyzing written text from

the online environment has been shown to be a fast, inexpensive, and unobtrusive technique for obtaining

information about user characteristics (Kozinets, 2002). Figure 2 summarizes the discussions up to this

8
point and depicts the logical chain of events involved in the process. It must be emphasized that similar to

other models, this figure presents a simplified picture of reality and does not include certain (potentially

influential) components. Innovativeness has been conceptualized on a spectrum of abstractness. On one

end of the spectrum, innovativeness is viewed as a latent personality trait, while on the other end, it is

studied as actualized, observable behaviors (that collectively form the lifecycle of a product) ( Midgley,

1976; Rogers, 2003). The translation of innate characteristics to action does not happen in a vacuum.

Rather, it is influenced by many explainable or random mediating factors. For example, it can be argued

that beside innate buyer personal characteristics, other environmental or product-specific factors (e.g.,

economic conditions, loyalty to a specific brand) influence the time to purchase. Likewise, it can be

argued that factors other than timing of purchase (e.g., the way the review website organizes and presents

previous reviews) can influence incoming review contents.

Innate Buyer Time to Review Stage of


(affects) (Reveals)
Innovatviness Purchase Content Adoption

Figure 2. The logical chain of events involved in inferring adoption stage from online reviews.

In this study, we aim to investigate if the patterns of change in buyer groups predicted by the

adoption theory can in fact be observed when their reviews are viewed longitudinally. To this end, we

extend the tenets pioneered by Rogers’s seminal works, as well as the findings from other researchers in

this area cited earlier in the article to the context of online reviews. Our arguments and predictions follow.

Hypotheses

By nature, earlier adopters of a technology or product are more open to risk and uncertainty than

late adopters. According to Rogers (2003), innovators “prefer venturesomeness to the respect of their

peers, who call them “strayed,” “experimenters,” or people with their “heads in the clouds.” He describes

innovators as preferring the “hazardous” and the “risky.” When thinking about adopting a product,

laggards are preoccupied with thoughts such as “before I start, I need to know the likely outcome.” This

propensity to certainty (certainty bias) is people’s bias, or tendency, towards taking too little risk. The

9
literature has identified the level of perceived risk and uncertainty associated with purchasing a product as

a key determinant of the timing of adoption (Hoeffler, 2003; Ostlund, 1974). This uncertainty takes

several forms. For example, buyers can be uncertain about how useful a product proves to be

(performance uncertainty), how others will view the purchase decision (symbolic uncertainty), how costly

it will be to switch to the new product (switching-cost uncertainty), and how high the emotional cost of

abandoning the current product or situation is (affective uncertainty) (Castaño et al., 2008). The

uncertainty associated with new products may discourage the individuals toward the end of the

innovativeness spectrum from making purchase decisions. In the context of online reviews, we posit the

following:

- Hypothesis 1 (H1): Compared to early buyers, late buyers show higher levels of certainty in

their reviews.

Products are typically unrefined and more difficult to use at the earlier stages of their

introduction. For example, they may have unidentified malfunctioning. As time goes by, companies

provide users with fixes and the needed technical information to get around problems. During earlier

stages of product introduction and in the absence of such technical information, early buyers must have

enough technical information to solve problems by themselves. However, as a product’s user base

expands over time, later buyers benefit from network externality (e.g., an operation guide provided by

third parties) and do not need to deal with technological complexities by themselves. Accordingly, earlier

buyers must “have the ability to understand and use complex technical information” (Rogers, 1963, p.

253). Apart from the complexities involved in using the product itself, the purchase situation and the

buying task itself can also be, or be perceived as being more complex during the early stages of a

product’s life. In general, high complexity implies higher learning and mental costs (Wood & Moreau,

2006). Prior research in the domain of adoption has demonstrated the effects of product and buying task

complexity on the timing of adoption (Tornatzky & Klein, 1982). Innovators may possess special mental

capacities that help them better cope with uncertainty and abstraction. An innovator has to conceptualize

10
abstract information and apply it to his or her special case while later adopters can simply observe the

outcome of earlier adopters’ decisions (Rogers, 2003). As such, we hypothesize the following:

- Hypothesis 2 (H2): Compared to early buyers, later buyers exhibit a less complex thought

process in their reviews.

According to Rogers (2003), to qualify as a member of the innovators group, one has to

frequently adopt innovative products. Therefore, such buyers must be able to absorb the cost associated

with occasional unsuccessful adoptions or at least perceive to have enough financial resources to deal

with such occasional unfortunate events. Consistent with what consumer theory and managerial practice

suggest, empirical research shows that early buyers exhibit lower levels of sensitivity to cost (i.e., lower

price elasticity) (Goldsmith, 1996). These observations lead us to hypothesize that those who adopt new

products earlier are less sensitive to monetary costs than later buyers. Accordingly, we expect to see fewer

price-related discussions in early buyers’ reviews. We hypothesize the following:

- Hypothesis 3 (H3): Compared to early buyers, late buyers reflect more sensitivity towards

cost in their reviews.

According to Rogers (2003), since innovators are the first group to buy a new product, at the time

of their purchases, there are not many people with whom they can consult or exchange ideas. As a result,

these buyers should secure their ideas and information needs from impersonal sources, such as technical

manuals, etc. Researchers have considered the degree to which individuals make their purchasing

decision independently of opinions of others and social norms as the primary factor that discriminates

innovators from others (Midgley & Dowling, 1993). Previous research has also identified a negative

relationship between collectivism and the degree of innovativeness (Steenkamp et al., 1999). Collectivism

implies commonality and a tendency to depend on others (Breckler et al., 2005, p. 316). Research also

shows that early adopters of products tend to be more inner-directed, while late adopters are typically

more other-directed (Donnelly Jr & Ivancevich, 1974).

Rogers (2003) proposes that innovators’ tendencies to venturesomeness “brings them out of their

local circles of peers,” whereas buyers on the other end of the adoption spectrums are essentially

11
“localistic” and “embedded” in their own networks. While early buyers stand out in a crowd with their

radical and revolutionary behaviors serving as opinion leaders, laggards tend to seek security and rapport

by melting into a crowd. Also, while innovators go beyond their early circles and their norms, laggards

tend to stay with those peers who share similar traditional values. As such, compared to early buyers, later

buyers are expected to possess more shared attributes and manifest a stronger sense of commonality with

other members of the social system. Accordingly, we hypothesize the following:

- Hypothesis 4 (H4): Compared to early buyers, later buyers exhibit higher degrees of

commonality in their reviews.

Rogers (2003) considers compatibility a mandate for adoption. According to him, compatibility is

the degree to which individuals perceive a new idea as being compatible with existing values and prior

experiences. Also according to him (Rogers, 2003), the point of reference for the people near the end of

the adoption spectrum is their past. In Rogers’s (2003) words, “while innovators look at the road of

change ahead, the laggards gaze at the rear-view mirror.” As such, people close to the end of the adoption

continuum can be described as traditional people who tend in their evaluations to compare novel products

or technologies to things from the past. Previous research has documented that late adopters of products

tend to be prone to nostalgia (Holbrook, 1993), which is manifested in the form of longing for, and

preferring products and other objects from the past (Steenkamp et al., 1999). As such, the literature has

proposed that buyers with higher levels of past orientation have lower levels of innovativeness while

buyers with higher levels of present orientation tend to exhibit more innovative behaviors (Karande,

Merchant, & Sivakumar, 2011). Accordingly, we hypothesize the following:

- Hypothesis 5 (H5): Compared to early buyers, late buyers show higher levels of past-

orientation in their reviews.

According to Rogers (2003), buyers towards the end of the adoption spectrum are suspicious of

innovations, innovators, and change agents. They see any new change as a potential threat to their

beloved status quo. In contrast, early adopters tend to be more positive and open to embrace change and

novelty. The Internet now facilitates everyone’s access to information about new products and

12
technologies. Given the same amount of information, those who process and weigh the available

information more positively tend to adopt the product or technology earlier. In the context of product

adoption, optimism is the belief and expectation that using the new product results in increased efficiency,

flexibility, and control (Parasuraman & Colby, 2007). Accordingly, we hypothesize the following:

- Hypothesis 6 (H6): Compared to early buyers, late buyers demonstrate lower levels of

optimism in their reviews.

The innovation-decision period is the time between obtaining information about a product or a

technology and making the adoption decision. Innovators are not the first group of people to adopt a new

product only because they become aware of its introduction somewhat earlier than others, but also

because it takes them a shorter time to move from the “knowledge state” to the “decision stage” (Rogers,

2003). Innovators are not happy with the status quo (Rogers, 2003). They are always actively in search of

new solutions and ideas. These buyers tend to be more actively involved at both the product purchase and

product usage stages (Ram & Jung, 1994). From the perspective of information search, innovativeness is

viewed as the “energy behind the changing, dynamic marketplace” (Vogt & Fesenmaier, 1998, p. 559)

that keeps buyers from buying from a limited set of established products. Late adopters, on the other

hand, tend to be passive. In fact, while innovators actively search for new ideas and adopt them, laggards

often adopt products by receiving them as gifts from friends or family. As such, we hypothesize the

following:

- Hypothesis 7 (H7): Compared to early buyers, late buyers exhibit a lower sense of dynamism

and activeness in their reviews.

Data

There are numerous online product review websites. Among the more renowned are websites

such as Amazon.com, epinions.com, cnet.com, consummerreview.com, and Ciao.com. Ciao.com is a

popular European web-based forum and e-commerce portal where individuals can share their purchase

experiences and provide their personal opinions about a wide range of products to help other buyers make

better purchase decisions. Membership to this website is open to the public and free. The website was

13
owned by Microsoft from 2008 to 2012 as Microsoft was trying to strengthen its e-commerce presence in

Europe. In January 2013 Ciao announced that it had achieved 28.4 million unique visitors per month in

Europe. Ciao has a multi-site strategy for serving different countries within Europe. We used user-

generated reviews from Ciao.co.uk, which is Ciao’s review website for the UK. Existing published

research in the domain of online reviews has used randomly-selected members of Ciao in a survey. The

survey indicates that the user base of Ciao is comprised of individuals with diverse demographic and

social backgrounds (Stöckl et al., 2006), making Ciao a representative community of online review

readers and writers for the purpose of this study. Previous research on online reviews has widely used

data from this website (Hennig‐Thurau et al., 2004; Khammash & Griffiths, 2011).

We used customer reviews posted from 2000 to 2011 for cell phones. These phones are various

models produced by established manufacturers (e.g., Nokia, Motorola, and Samsung). Over the past two

decades, cell phones have been continually evolving in terms of functionality (e.g., support for text

messaging, web browsing), form and shape (e.g., bar, clamshell, sliding), and features (e.g., camera,

touch-screen interface) (for details about the cell phone market around the time period of our

investigation, please see (Koski & Kretschmer, 2007)). Cell phones were chosen also because of their

diverse customer base as well as reasonable value and lifespan. Consistent with previous research (Wu &

Huberman., 2008), we excluded those products that had no more than 20 reviews. Also, for almost all of

the chosen products, at least six months had passed since the dates of their last reviews. Considering the

lifespan of a cell phone, we assume that if there have not been any new reviews received within a six-

month window, then the product life cycle has reached its end. This provision was to ensure that only

products with full or near-full life are included in the analysis. This resulted in the creation of a dataset

containing 2,065 complete reviews for 48 different products. It is noteworthy that the existence of

multiple reviews written about a product by the same reviewer at different points in time can potentially

bias the sample and complicate the analysis. Further investigation of the reviews in our dataset revealed

that almost all reviews for any given product were from a different user. The percentage of repeat reviews

was less than 1%, and the median length of the time between those repeat reviews was six days, which is

14
considerably shorter than the life span of products. Table 1 provides descriptive statistics about the data

(to eliminate the possible effects of extreme outliers, the last 5% of reviews were eliminated before

analyzing the temporal span of reviews).

Mean Std. Min Max

Number of reviews per product 43.02 36.40 20 240

Temporal span of reviews (months) 29.49 10.79 12.50 69.30

Review length (words) 492.14 453.89 90 4,193

Table 1. Descriptive statistics of cell phone reviews

Definition and Operationalization of Constructs

We employed different content analysis and text classification techniques to analyze reviews. To

measure certain psychological and personality-related characteristics of buyers (i.e., propensity to

certainty, degree of commonality, optimism, and activeness), we chose to use the previously validated

Diction software package. Diction is an automated, theory-based content and sentiment analysis program

for measuring and quantifying a variety of features in verbal or textual messages. The package has been

used extensively to analyze texts in different contexts, ranging from political speeches to user-generated

content on weblogs (Bordbar, 2010; Davis & Gardner, 2012). Recent research in the domain of “big data”

from the Web has recommended the use of this program to analyze large quantities of textual material

(Holbert, 2014). The program analyzes writing to develop indexes for 33 basic text sub-features (e.g.,

“Praise,” “Blame”). These sub-features are in turn used to create some general text features (e.g., degree

of optimism). As an example, sub-features “Praise,” “Satisfaction,” and “Inspiration” positively

contribute to the index of “Optimism,” whereas sub-features “Blame,” “Hardship,” and “Denials” affect

that index negatively.

The program uses a 10,000-word corpus to processes English texts. The word lists used to

measure the 33 sub-features have no overlap, and these mutually-exclusive wordlists are open for

15
inspection by researchers. The program thus achieves variable reliability (Holsti, 1969; Krippendorff,

2008). Nevertheless, around 10% of the words in the word lists are homographs: words with different

meanings that are spelled the same (for instance, the word “second” can refer to a unit of time or to

ordinal numbering). This can potentially cause complications in text analysis. However, not all

homographs cause problems. Benign homographs are homographs that are essentially different words but

still belong to the same semantic family (for example, the word judge as a noun or as a verb). The

program ignores these types of homographs. As for confounding homographs, it applies statistical

weighting based on frequency of use (Easton, 1940) to determine the degree to which a word is likely to

belong to a class. This method is often referred to as “soft categorization” (Sebastiani, 2002). This

provision improves content validity (Krippendorff, 2008). Diction provides the statistical results

alongside the original text being analyzed; thus, users can readily cross-check text with its quantified

version, which provides further validity and reliability of the analysis. Studies performed and published

by both the creators of the program and independent authors suggest that the program is both valid and

reliable in analyzing English text (Ober et al., 1999).

To obtain an index for buyers’ sensitivity toward cost, automated text classification was used.

This essentially involves assigning text into a number of pre-determined classes (in our case, text that is

price-related and text that is not). A common method of representing text in classification tasks that is

both simple and effective is the use of bag of words. We performed review text classification based on the

term list (unigrams) that we developed for the purpose of this study. We started off building the term list

by considering a small number of price-related “seed words” (e.g., “price” and “money”) and then used a

thesaurus and WordNet (Miller, 1995) to “grow” the seeds into a larger set containing price-related

words. The absence or presence of words from this price-related word set in a review determines the

index of sensitivity to cost for that review. Even though more complicated methods can be used to

measure reviewers’ attitudes toward cost (e.g., sentiment analysis), the proposed approach is adequate for

the purpose of the analysis since our purpose is limited to identifying text that discusses price. Table 2

16
provides definitions of constructs used in the hypothesis as well as a summary of the method used to

operationalize and measure the respective variables.

Variable Operationalization

The Oxford dictionary defines certainty as “a firm conviction that something is the

case.” We used Diction’s general text feature “Certainty” to assess this feature.

Certainty is defined by Diction as “Language indicating resoluteness, inflexibility,


Propensity to
and completeness and a tendency to speak ex cathedra.” Diction uses the following
certainty
formula to create an index for certainty:

Certainty = [Tenacity + Leveling + Collectives + Insistence] - [Numerical Terms +

Ambivalence + Self Reference + Variety].

The idea used in deriving this index is borrowed from Flesch (1951). The idea is that

convoluted paraphrasing is positively related to higher levels of complexity and

abstraction. Flesch’s method does not actually analyze the meaning of the words used
Complexity
in corpus. Rather, it takes into account word and sentence length. Text featuring

longer words/sentences tends to be more abstract and complex. Diction uses the same

logic in creating its computed variable “Complexity.”

Automated text classification was used to identify reviews that discussed price. The

index of sensitivity to price is created for each review using the frequency of
Sensitivity
appearances of cost-related terms in reviews. A cost-related term list was developed
toward cost
by growing an initial set of seed words (e.g., “expensive,” “price”) using a thesaurus

and WordNet. We then fed the list to Diction which created the related index.

17
This is based on Diction’s text feature “Commonality” defined as “Language

highlighting the agreed-upon values of a group and rejecting idiosyncratic modes of


Degree of
engagement.”
commonality
The program uses the following formula to calculate the index: Commonality =

[Centrality + Cooperation + Rapport] - [Diversity + Exclusion + Liberation].

This index is calculated by extrapolating the present and past tense of the verbs in the

Present list of general public terms developed by the psycholinguist Ogden (1968). The list

concern to includes verbs referring to general physical activity (e.g., “take,” “touch,” “see”),

past concern social activities (e.g., “meet,” “discuss”) and task-related activities (e.g., “make,”

ratio “do”). Diction uses this method to calculate and report variables “present concern”

and “past concern.”

Optimism conveys hopefulness or positive attitude about the outcome or the future of

something. For the purpose of our analysis, we adopt the following definition for

optimism from Diction: “Language endorsing some person, group, concept or event
Optimism
or highlighting their positive entailments.”

Diction calculates the index based on the following formula:

Optimism = [Praise + Satisfaction + Inspiration] - [Blame + Hardship + Denial].

Activeness is based on the Diction variable “Activity.” Diction defines activeness as

“Language featuring movement, change, the implementation of ideas and the

avoidance of inertia.”
Activeness
The following formula is used to calculate the index:

Activity = [Aggression + Accomplishment + Communication + Motion] - [Cognitive

Terms + Passivity + Embellishment].

Table2. Definition and operationalization of constructs.

Statistical Analysis and Results

18
Dependent variables in our analysis are measures of review characteristics such as optimism and

sensitivity to price. For our independent variable, we labeled reviews in the first and the last quartiles

(first and last 25%) as “Early” or “Late,” respectively. In addition to this categorical independent variable,

we constructed a continuous time-related variable, taking into account that the launch date, lifespan, and

number of reviews vary from product to product. For every product, we first calculated the time span of

reviews by counting the number of days between the first and the last review. Then, for every individual

review, we calculated the number of days between the arrival day of that review and the date of the first

review for that particular product. Visual investigation of the frequency plot of this continuous variable

for some of the products revealed that review arrival times are not evenly distributed across the life of

products; rather, review arrival points are typically concentrated towards the beginning of a product’s life

(positively skewed). To address this, we log-transformed the variable. Next, to obtain an overall

understanding of data, we plotted review characteristics (e.g., propensity to certainty, optimism) against

the log-transformed time-related variable. Figure 3 shows the observed pattern of change in two of the

dependent variables for Motorola V3 as a representative product. Plots feature a LOESS (locally weighted

scatter plot) smoothing curve. LOESS curves are fixable tools for exploration and detection of trends in

data (Cleveland & Devlin, 1988). On every plot, a linear regression line has also been superimposed.

19
Figure 3. Measures of psychological aspects of reviews for Motorola V3.

Visual investigation of the data reveals that LOESS does not show signs of pronounced deviation

from the linear trend line. A similar investigation on some other products in data set reveals the same,

which provides support for using linear models. It is worth mentioning that some researchers (Moore,

1991) have argued that in some contexts innovativeness in different adopter groups (i.e., innovators and

early adopters) can take a discontinued form. However, in line with what the above graphs show, prior

empirical research does not show strong evidence for there being discontinuations, or “chasms,” between

adjacent adopter groups in terms of their degree of innovativeness (Rogers, 2003). We now turn to our

empirical analysis.

Empirical Model

To test the hypotheses, we employed a mixed effects model (a.k.a. multi-level model, or

hierarchical liner model) (Bates, 2010). As the name implies, this model is useful for analyzing statistical

parameters that vary at multiple levels (here, at adopter group level and at product level). Specifically,

membership in adopter groups has two fixed levels (early or late) and in each case, is hypothesized to

have a systematic and predictable influence on the dependent variables of interest (e.g., degree of

optimism, certainty). We, therefore, model membership in adopter groups as a fixed effect (Bates, 2010).

On the other hand, the idiosyncrasies in various products present in our dataset (i.e., different phone

models) are expected to have unpredictable and nonsystematic effects. After all, the models in our dataset

represent a random sample from the population of all possible phones, and unlike levels of a fixed effect

variable, are by no means close to “exhausting the population of interest” (Winter, 2013). In particular,

our purpose in this analysis is to generalize the (systematic and fixed) effect of belonging to early or late

buyers groups over the idiosyncrasies of various products (that can have different manufacturers, price

ranges, etc.). Accordingly, we modeled product as a random effect in our mixed model (for a concise

overview, please see Winter, 2013).

20
It is noteworthy that product can be introduced to the model as a fixed effect too. While either of

these methods have their own strengths and weaknesses (Greene, 2000), modeling variables such as this

as a random effect is generally more efficient and generalizable. Nonetheless, to investigate whether

modeling product as a fixed effect can actually outperform modeling it as a random effect in our specific

case, we conducted Hausman’s specification test (Hausman, 1978). Hausman test’s null hypothesis is that

the preferred model is the random effects model, and the alternative is that the fixed effects model is more

appropriate. Results from Hausman’s test did not reject the null hypothesis, indicating that modeling

product as a random effect is preferred.

To summarize, we used a mixed effects model to explore the relationship between memberships

in different buyer groups (early or late) and the innovativeness-related characteristics of the buyers

outlined in the hypothesis section. To our model, we entered group membership as the fixed effect and

had intercept for products entered as a random effect. This procedure was performed separately for all

response variables mentioned in the hypothesis section.

To fit the model, we used the linear mixed effect model (lmer) function from the lme4 package

(Bates et al., 2013) in R environment for statistical computing (version 3.2.3; R Development Core

Team, 2015). Table 3 summarizes the result of the statistical analysis.

Dependent Statistics associated with the fixed effect of group

Variable (Early or Late)

Parameter Standard t Value P value

Estimate Error

Certainty 0.440 0.176 2.50 .012

Complexity -0.005 0.017 -0.33 .741

Price concern 0.313 0.143 2.18 .029

21
Commonality 0.328 0.165 2.00 .046

Present to past -0.398 0.018 -2.20 0.028

concern ratio

Optimism -0.479 0.167 -2.90 .004

Activity -1.798 0.978 -1.84 .066

Table 3. Parameter estimates associated with the fixed effect of membership to early or late buyer
groups of cell phones.

Results from the statistical analysis show that compared to reviews by early buyers, reviews by

late buyers show higher levels of propensity to certainty by 0.440 units ± 0.17 (standard errors).

Compared to reviews by early buyers, reviews by late buyers reflect more sensitivity toward cost by 0.313

units ± 0.143 (standard errors). Also, compared to early buyers, late buyers exhibit higher degrees of

commonality in their reviews as reflected by 0.328 units ± 0.165 (standard errors) units increase in the

respective measure. The composite measure of present to past concern was also lower by 0.398 units ±

0.018 (standard errors) in late buyers. For each review, this measure was created by dividing a review’s

present concern to the sum of its present concern and past concern indices, resulting in a ratio ranging

between zero and one. The closer to zero the ratio, the more concerned the author about the past, and

vice-versa. Since this outcome measure is a proportion, we transformed it to normality by performing

arcsine (angular) transformation (Kabacoff, 2015) before conducting our analysis. It is noteworthy that

when analyzed individually, the index of present concern shows a decrease by 0.481 units, and the index

of past concern shows an increase by 0.317 units over time. These individual changes too are in the

predicted direction and therefore provide further evidence supporting our hypothesis, even though the

trends in these individual indexes are not statistically significant. Finally, compared to early buyers, the

measure of optimism was lower in late buyers by 0.479 units ± 0.167 (standard errors). The measure of

activity also showed a decline of 1.798 ± 0.978 from early to late groups even though this decline is only

significant at α of 0.1. Even though the change in the measure of complexity is in the hypothesized

22
direction, the difference is not statistically significant. A plausible explanation for this observation is that

the purchasing behavior of many early adopters is derived by an urge for novelty and a desire for new

experiences (Hirschman, 1980) and is not necessarily the result of a complex decision process.

Results presented above show the average change associated with the psychological trends across

all products in our dataset. Nevertheless, it could be that some individual products show stronger or

weaker trends than the average. In the next step of our analysis, in addition to accounting for differences

in baselines (random intercepts), we account for possible differences in slopes (random slopes) across

different products. Figure 4 shows the frequency histograms of slopes for the 48 different products

present in our dataset.

23
Figure 4. Frequency histogram of the slopes of psychological trends belonging to individual cell phones

(Slopes less than zero indicate a downward trend and slopes more than zero indicate the opposite.)

Visual investigation of the results reveals that other than the average slopes across all products,

individual slopes for single products also are mostly in the hypothesized direction. For example, it can be

seen in the figure above that reviews of all but three of the 48 different models of cell phones in our

dataset (that is, 93% of the products in our sample) show a decline in the measure of optimism over time

(observable in the form of negative slopes of different degrees).

Discussion

Results from hypothesis testing are summarized in table 4.

Supported /Not
Hypothesis Comments
supported

H1:
As hypothesized, the degree of propensity to certainty is
Propensity to Supported
higher in reviews written by late buyers.
certainty

Our sample data did not support the hypothesis that the
H2:
Not supported degree of complexity of reviews written by early buyers
Complexity
is higher than those written by late buyers.

24
As hypothesized, cost-related concerns of late reviewers,
H3:
Supported as reflected in their product reviews, are higher than that
Price concern
of their earlier counterparts.

As hypothesized, the sense of commonality and


H4:
Supported relatedness among late buyers tends to be higher than
Commonality
that of early buyers.

H5: As hypothesized, early reviews reflect a higher

Temporal Supported fascination with the present while late reviews reflect a

concern higher fascination with the past.

H6: As hypothesized, compared to late reviews, early


Supported
Optimism reviews reflect higher levels of optimism.

H7: As hypothesized, reviews written by earlier reviewers


Partially supported
Activity reflect higher degrees of activity.

Table 4. Results of hypothesis testing for cell phones.

Taken together, these findings are largely consistent with the general notion that reviews, when

viewed chronologically, evolve in a way that reflects the thought process of different groups of buyers in

terms of their degrees of innovativeness. The observed trend is also consistent with the general tenet of

buyer adoption theory and the personality and psychological characteristics that one would expect to see

(Everett M Rogers, 2003) from different groups of buyers along the product life cycle.

Gauging the Generalizability of Findings

To gauge the generalizability of these findings, we conducted a similar analysis on a different

product belonging to a different category. This chosen product is console & computer games (e.g., Xbox

games or PC games, hereafter referred to as “video games”). On one hand, similar to cell phones, video

games have reasonable prices and lifespans and are therefore appropriate for the purpose of our study. On

25
the other hand, the extended category has its own traits compared to cell phones, providing an opportunity

to compare and contrast trends across different categories.

Models based on economics of information search categorize goods to search and experiential,

depending on the cost of information search (Nelson, 1970, 1974). According to Nelson (1970, p. 783),

“goods can be classified by whether the quality variation was ascertained predominantly by search or by

experience.” Interpreted more liberally, search goods are those whose quality can be reasonably assessed

before consumption, and experiential goods are the goods whose quality is difficult to assess unless

consumed. Other researchers have used the degree of tangibility as the discriminating factor between

search and experience goods, with search goods being more tangible, and experience good being less so

(Levitt, 1981). However, similar to many other dichotomies, categorizing a product to search or

experiential may not be straightforward (Rao & Ruekert, 1994) since a product can have both search and

experiential features. The more recent literature, therefore, recognizes that products have a mix of search

and experiential “attributes”. Products such as cameras and music are among the classic examples that fall

near the two ends of the search-experiential continuum, respectively (Nelson, 1970). In the case of cell

phones, for example, camera resolution, display size, and battery life would be search attributes, while the

degree of “friendliness” of the interface would be an experiential feature. Since search attributes are

dominant in cell phones, they “belong more” to the search category. On the other hand, games belong

more to the experiential category, since they are dominated by experiential attributes.

In the domain of online reviews, the dichotomy of search and experiential has been studied from

different perspectives (see for example Mudambi & Schuff, 2010). A key reason this distinction is

relevant is the fundamental differences in the way the attributes of search and experiential goods are

evaluated, and subsequently expressed. Specifically, while the key characteristics of search goods are

objectively measurable and expressible (e.g., hours of talk time, or camera resolution for a cell phone),

the characteristics of experiential goods tend to be fairly subjective (e.g., the degree of customizability,

26
randomness, or engagement of a video game) (see Elias et al., 2012 for more details about key

characteristics of digital games).

Our dataset contained reviews from 90 different products of game. Table 5 provides descriptive

statistics about these games.

Mean Std. Min Max

Number of reviews per product 38.9 22.75 20 154

Temporal span of reviews (months) 40.12 23.67 7 105

Review length (words) 571.65 596.30 12 6,218

Table 5. Descriptive statistics of game reviews

We conducted an analysis of game data similar to that which we conducted on cell phone data.

Results of the statistical analysis are presented in Table 6.

Dependent Statistics associated with the fixed effect of group

Variable (Early or Late)

Parameter Standard t Value P value

Estimate Error

Certainty 1.382 0.162 8.56 <.001

Complexity 0.016 0.012 1.3 0.194

Price concern 0.461 0.079 5.818 <.001

Commonality -0.182 0.115 -1.60 0.110

Present to past -0.016 0.008 -2.03 0.042

concern ratio

Optimism -0.320 0.135 -2.4 0.016

Activity 0.108 0.217 0.5 0.617

27
Table 6. Parameter estimates associated with the fixed effect of membership to early or late buyer
groups of video games.

As Table 6 demonstrates, there are remarkable similarities between the patterns observed in cell

phone reviews and in game reviews. Similar to what was observed with cell phones, earlier reviews of

games reflect lower levels of propensity to certainty on the part of their writers. Likewise, early reviews

of games tend to show less sensitivity toward price. Earlier buyers of games also tend to be more

concerned about the present than the past, and are also more optimistic in their reviews. Similar to the

case of cell phones, we failed to show any significant difference between the degrees of complexity

reflected in early or late reviews of game buyers.

Nonetheless, the increase in the degree of commonality in late buyers of cell phones was not

observed with late buyers of games. Even though we cannot ascertain with confidence why these

differences exist between reviews for cell phones and games, we can conjecture some possible reasons.

As discussed earlier, a key discriminating factor between search and experience goods is their degree of

tangibility (Levitt, 1981). Cell phones are tangible and highly visible, while games are intangible, and

their use, for the most part, is a private experience. For a highly visible product such as a cell phone, late

buyers can readily see the adoption behavior of earlier buyers, and they may tend to conform and manifest

more common attributes with early groups. In fact, the tendency to attend to social comparison

information has been shown to have a negative relationship with innate innovativeness (Clark &

Goldsmith, 2006). Also, even though the decrease in activity of late buyers compared to early buyers was

partially supported for cell phone data, a similar pattern is not observable for games. We postulate that the

observed difference may be rooted in some of the unique features of video games as a product category.

Games, by their very nature, are about motion and continuous change. Therefore, our measure of activity

may not have been able to pick up the differences between levels of activity in early and late reviews of

games whose main subject is already primarily about motion and activeness (e.g., building, running,

28
attacking). Table 7 provides a summary, along with a comparison between the results from the two

products.

Supported /Not

Hypothesis supported (and a Comments

comparison)

H1: Supported Similar to results from cell phones, the degree of

Propensity to (similar to results propensity to certainty is higher in reviews written by

certainty from cell phones) late buyers of video games.

Similar to the case of cell phones, our sample data did


Not supported
H2: not support the hypothesis that the degree of complexity
(similar to results
Complexity of reviews written by early buyers is higher than those
from cell phones)
written by late buyers of video games.

Similar to results from cell phones, cost-related concerns


Supported
H3: of late reviewers of video games, as reflected in their
(similar to results
Price concern product reviews, are higher than that of their earlier
from cell phones)
counterparts.

Not supported
Unlike the observed pattern in cell phones, the sense of
H4: (different from
commonality and relatedness among late buyers was not
Commonality results from cell
higher than that of early buyers of video games.
phones)

Similar to results from cell phones, early reviews of


H5: Supported
video games reflect a higher fascination with the present
Temporal (similar to results
while late reviews reflect a higher fascination with the
concern from cell phones)
past.

29
Supported, Similar to results from cell phones, compared to late
H6:
(similar to results reviews, early reviews of video games reflect higher
Optimism
from cell phones) levels of optimism.

Not supported,
Unlike results from cell phones, reviews written by
H7: (different from
earlier reviewers of video games did not reflect higher
Activity results from cell
degrees of activity.
phones)

Table 7. Results of hypothesis testing for video games and a comparison with cell phones.

In sum, results from the analysis of review data from cell phones and games are for the most part

consistent. More specifically, four out of the five hypotheses supported for cell phones hold for video

games too. This is particularly remarkable when considering the fact that the two products belong to two

distinct product categories.

Observations About Star Ratings

These findings can have implications for interpreting trends in review stars, among other things.

Every text review in our dataset is accompanied by a star review (e.g., a numerical rating of the product).

We conducted further analysis on trends in these stars. In several previous studies, it has been observed

that review stars tend to decline gradually over both time and order (Chevalier & Mayzlin, 2006; Li &

Hitt, 2008; Wu & Huberman., 2008). Consistent with the results of these studies, our correlation analysis

on (pooled) star ratings and review arrival orders and on star ratings and review arrival times for cell

phones show statistically significant negative correlations (r = -0.180, p<0.001 and r = -0.136, p<0.001,

respectively).

The decline in review stars has been explained in different ways. Li & Hitt (2008), for example,

attribute it to different degrees of product liking by early and late buyers. This means that, on average,

those buyers who have a higher valuation of a product make their purchases and subsequently post their

30
reviews earlier than other buyers, and this eventually results in a decline in average user review stars over

time. Even though this is a plausible reason for the observed decline, this explanation is based on two

implicit assumptions: (1) buyers who have above-average expected utility also end up gaining above-

average experienced utility, and (2) post-purchase star rating accurately reflects the actual experienced

utility (Godes & Silva, 2012). There are reasons to question the validity of these assumptions, though. For

example, buyers are delighted only when their experienced utility exceeds what they initially anticipated

(Achrol & Kotler, 1999). Therefore, it may be more difficult to please the early buyers who presumably

have high expectations of the products.

Clearly there is a need for a comprehensive and controlled study to better explain the observed

decline in star ratings over time and over arrival order. Nonetheless, in light of the findings in this study,

it can be argued that the observed trend in star ratings can also be the result of differences in innate

personality characteristics of buyers who self-select to buy the product at different times after its release.

As discussed earlier, in the context of product adoption, optimism is the belief and expectation (not

necessarily an actualized experience) that using the new product results in increased efficiency,

flexibility, and control (Parasuraman & Colby, 2007). Given the fact that innovators tend to be more

optimistic, it can be argued that one reason for higher star ratings during the earlier stages of a product

release is that adopters in that stage are generally more optimistic, hopeful, and positive. In fact, we

observed a highly significant positive correlation between our quantified measure of review text optimism

and star review rating (r=0.190, p<0.001). While this topic merits a more thorough investigation, the

above discussion shows how viewing trends in reviews from the perspective of buyer adoption behaviors

can provide us with novel insights.

General Discussion and Conclusion

The Web now allows buyers to share their personal experiences and attitudes towards products

with virtually everyone else, including vendors. This has provided new opportunities for marketers to

31
form near real-time insights into various aspects of consumer behavior. In this study, we took a novel

perspective by analyzing review content from the standpoint of user’s predispositioned, or innate

innovativeness. Our work has significant implications for theory, methodology, and practice.

From the theoretical perspective, we showed that certain characteristics of text reviews change

over time in ways that reflect the changes in psychological characteristics of different groups of buyers as

predicted by the adoption theory. The literature has conceptualized and studied innovativeness at different

levels of abstraction (Midgley & Dowling, 1993). At its least abstract level, innovation is referred to as

“actualized innovativeness” and is viewed as the actual, observable act of adopting a new idea or product

(e.g., purchasing a device). Using a behavioral approach, innovativeness at this level is measured in terms

of the relative time of acquiring a new product with respect to other members of the social system

(Rogers, 1976; 2003). Studies that have investigated innovativeness at this level have used measures such

as ownership of specific goods (or the intention thereof), the total number of products owned by a person,

or other similar methods. Yet, at its most abstract level, innovativeness is viewed as an innate,

unobservable global personality trait influencing adoption decisions across all product classes (Goldsmith

& Hofacker, 1991; Midgley & Dowling, 1993). By being global, this level of innovativeness is distinct

from domain-specific innovativeness that is conceptualized to influence the adoption of products

belonging to certain product types or classes (Goldsmith et al., 1995). Innovativeness-related research

aimed at segmenting buyers has treated buyers with different levels of innate innovativeness as possessing

distinct cognitive styles and as adopting different approaches for information processing. The relation

between latent innovativeness-related personal characteristics (innate innovativeness) and actual

observable adoption behavior (actualized innovativeness) is complex as it is mediated by many

explainable and random variables, including a host of demographic, economic, and other contextual

factors such as exposure to ads or to social media. Although these factors can complicate the relationship

between trait and behavior, using a novel methodology, we were able to empirically show that several key

innate innovativeness-related characteristics of buyers, inferable from review text, correspond directly to

actualized innovativeness.

32
In their critical review of innovation diffusion and new product growth models, Peres et al.

(2010) describe data from online communities and other forms of user-generated information from the

Web as a source that provides “numerous new possibilities for exploring individual adoption decisions

and linking them to overall diffusion patterns” (p. 103). From the methodological perspective, the current

work can be considered as the first answer to that call. In exploring diffusion patterns, we considered

personality-related and psychological characteristics of reviewers as reflected in their reviews. Due to

practical issues, these types of characteristics are under-researched in diffusion studies (Rogers, 2003).

Nevertheless, our unique methodological approach (i.e., review text mining) allowed us to unobtrusively

measure such characteristics.

Our work also has important implications for practice. The product adoption and PLC

frameworks are among the most intuitive and appealing methods for explaining and predicting market

dynamics and have a wide range of applications for strategy formulation and execution (Hoffer, 1975).

However, unambiguously determining the current stage of PLC remains one of the main issues in using

PLC (Day, 1981). Without reliable metrics for identifying the stage of the product on its life cycle, or at

the least the “critical turning points” on the life cycle, the model has little predictive validity or empirical

value (Golder & Tellis, 2004). Our research provides preliminary, yet promising empirical evidence to

address the long-standing challenge involved in identifying the stage of product adoption and life cycle.

Among its many applications, being able to reasonably assess the stage of adoption and PLC can

help in designing effective marketing messages. For instance, as discussed earlier, the literature has

prescribed using informative and persuasive advertising messages during early and later stages of life

cycle, respectively. In addition to that, firms can tailor their description of products to be more appealing

to the target group of prospective buyers during each stage of life cycle. For example, they can describe

the product as having novel features during early stages and as contributing to ease-of-use, reliability, or

cost saving during late stages. Another benefit of having a reasonably accurate assessment of the stage of

products in their lifecycles for firms with multiple offerings is that it helps them avoid accidental product

cannibalization, whereby a newly-released product “eats” the market shares of the firm’s own incumbent

33
product due to poor product release timing. Having a reasonably accurate assessment of product life stage

can also help firms engage in intentional cannibalization, whereby the strategically-timed introduction of

a new product results in a larger overall market share, in increased customer loyalty, or in both.

Our work also has important implications for online retailers and review websites. Websites such

as amazon.com and ebay.com attract large volumes of viewer traffic for the product reviews they post on

their websites. In 2013, 67% of consumers read fewer than six reviews when making a purchase

(Anderson, 2013). In fact, the number of reviews consumed by buyers has been constantly dropping over

recent years. As a remedy, matching the information displayed with buyers’ personality characteristics

(Jahng et al., 2002) can encourage customers to browse for more information. In order to achieve this

goal, an important practical issue is which reviews to show and in what order, as this can significantly

influence customers’ perception of two of the review websites’ key attributes: its usefulness and its ease-

of-use (Van der Heijden et al., 2003). Findings from this study stress the importance of displaying the

more recent reviews first, as authors of those reviews tend to be more similar to prospective buyers in

terms of attitude toward products and, therefore, tend to discuss topics that are perceived as being more

relevant to the new prospective buyer.

Life cycle analysis can furnish different results and insights when applied to different “levels” in

the product hierarchy (e.g., to entire industries, to product classes, to product forms, or to variations of a

brand and single models). The current study applied the analysis to the lower levels of the hierarchy,

providing insights that are useful for product-level management. But applying the concept of lifecycle to

lower levels of the product hierarchy inevitably means dealing with shorter time horizons. In extreme

cases, the timing of product update and release may be pre-determined at model level. So the question is

whether or not identifying the stage of adoption or life cycle is useful in such cases. We are persuaded

that having a reasonable assessment of the stage of adoption is useful even if release dates are set in

advance, since the inclusion of many product features is usually decided late in the development or

production process with respect to market situations and trends. Having a reasonable assessment of

34
adoption and life cycle stages, therefore, will allow managing or revitalizing product life cycle by

releasing products with the right set of features.

Limitations and Directions for Future Research

Similar to any other micro-level empirical study, this study has limitations that need to be

considered when interpreting the results. In our analysis we used data collected from a single review

website. It would be useful to perform similar analysis of reviews posted on other websites to see how

different platform features may magnify or dampen the effects observed in the current study. In addition,

our approach is mainly based on the effect of time on review evolution and does not directly account for

sequential dyadic effects of reviews. Sequential effect is the effect that the already existing reviews can

have on an incoming review. Nevertheless, while some researchers have stressed the importance of

considering both temporal and sequential effects on the evolution of reviews (Godes & Silva, 2012),

others have questioned the primacy of sequential effect on product purchasing and adoption (Van den

Bulte & Lilien, 2001).

Another prevalent potential issue in all adoption studies is their inability to efficiently deal with

products that fail (Rogers, 2003). To counter this potential problem, following the existing literature of

online reviews (Wu & Huberman., 2008), in our sample, we included products that had at least 20

reviews, with the aim of including only those products that have been received reasonably well by buyers;

however, it would be insightful to investigate and compare products that receive considerably fewer, or

considerably more reviews. The other challenge in studying the dynamic and evolution of online reviews

is posed by products that are said to have “international life cycles.” These products have several

concurrent life cycles going on in different markets. For example, a product may already be in its maturity

phase in one market when it is introduced to another market. In such situations, it is easy to have mixed

reviews from different buyer groups. This, however, is not a threat to our analysis because of the limited

geographical area from which reviewers come (the UK).

35
According to Hofer (1975), “The most fundamental variable in determining an appropriate

business strategy is the stage of the product life cycle” (p. 798). Previous research (Anderson &

Zeithaml, 1984) has tested propositions about the contingency of the choice of optimum strategy to stage

of the product on its life cycle and found support for those propositions. However, the idea of PLC is of

little empirical use unless one can track the stage of a product on its life cycle. Future research can use the

trend in review text observed in the present study to determine the critical turning points (inflection

points) of product life cycle. These turning points are (1) the point of product “takeoff” (the end of

introduction stage and the beginning of growth) when pessimism abounds and managers are under

pressure to “pull the plug” on the new, presumably failing, product; and (2) the point of “shutdown,” the

peak in product life cycle, after which the product decline starts but managers are typically overoptimistic

about the product’s future (Golder & Tellis, 2004). While in our empirical model we operationally

defined “early” and “late” as the only two levels for buyer groups, by operationalizing group membership

at a more fine-grained level (e.g., early adopters, early majority, late majority, laggards), future research

can be directed at assessing the location of the inflection point of the adoption curve based on trends in

review characteristics of all buyer groups. Another useful topic for future research is the effects of prior

sales performance on the trends identified by this study. Finally, the reported results reflect changes in

reviews of “average products.” Compared to an average product, different products can follow the

identified trends to different degrees. Identifying features that differentiate products from this perspective

would be very insightful for both research and practice.

Our work is one of the first attempts to capture the potential of text reviews in the study of

product adoption and life cycle. Adoption and PLC are highly important areas of business research and

practice with numerous direct empirical applications. The above-mentioned topics, along with many

others, remain promising areas for future research.

36
References

ACHROL RS and KOTLER P (1999) Marketing in the network economy. The Journal of Marketing,
146-163.

ANDERSON C and ZEITHAML C (1984) Stage of the Product Life Cycle, Business Strategy, and
Business Performance. Academy of Management Journal 27(1), 5-24.

ANDERSON M (2013) Local consumer review survey 2013. BrightLocal Blog. Retrieved from
BrightLocal blog http://searchengineland.com/2013-study-79-of-consumers-trust-online-reviews-as-
much-as-personal-recommendations-164565.

BATES D, MAECHLER M, BOLKER B, WALKER S and OTHERS (2013) lme4: Linear mixed-effects
models using Eigen and S4. R package version 1(4). Available at: http://keziamanlove.com/wp-
content/uploads/2015/04/StatsInRTutorial.pdf

BATES DM (2010) lme4: Mixed-effects modeling with R. URL http://lme4. r-forge. r-project. org/book.

BICKART B and SCHINDLER RM (2001) INTERNET FORUMS AS INFLUENTIAL SOURCES OF


CONSUMER INFORMATION. Journal of Interactive Marketing (John Wiley & Sons) 15(3), 31-40.

BORDBAR F (2010) Gender, identity, and language Use: A case of Tehrani teenage bloggers’ virtual
speech community. International Journal of Language Studies 4(2), 1-62.

BOWER JL and CHRISTENSEN CM (1995) Disruptive technologies: catching the wave. Harvard
Business Review Video Available at:
http://immagic.com/eLibrary/ARCHIVES/GENERAL/JOURNALS/H950130C.pdf.
BRECKLER S, OLSON J and WIGGINS E (2005) Social psychology alive. Cengage Learning

CASTAÑO R, SUJAN M, KACKER M and SUJAN H (2008) Managing consumer uncertainty in the
adoption of new products: temporal distance and mental simulation. Journal of Marketing Research
45(3), 320-336.

ÇELEN B and KARIV S (2004) Distinguishing Informational Cascades from Herd Behavior in the
Laboratory. American Economic Review 94(3), 484-498.

ÇELEN B and KARIV S (2005) An experimental test of observational learning under imperfect
information. Economic Theory 26(3), 677-699.

CHEVALIER JA and MAYZLIN D (2006) The Effect of Word of Mouth on Sales: Online Book
Reviews. Journal of Marketing Research (JMR) 43(3), 345-354.

CHINTAGUNTA PK, GOPINATH S and VENKATARAMAN S (2010) The Effects of Online User
Reviews on Movie Box Office Performance: Accounting for Sequential. Marketing Science 29(5),
944-957.

CLARK RA and GOLDSMITH RE (2006) Interpersonal influence and consumer innovativeness.


International Journal of Consumer Studies 30(1), 34-43.

CLEVELAND WS and DEVLIN SJ (1988) Locally weighted regression: an approach to regression


analysis by local fitting. Journal of the American Statistical Association 83(403), 596-610.

DAVIS KM and GARDNER WL (2012) Charisma under crisis revisited: Presidential leadership,
perceived leader effectiveness, and contextual influences. The Leadership Quarterly 23(5), 918-933.

37
DAY GS (1981) THE PRODUCT LIFE CYCLE: ANALYSIS AND APPLICATIONS ISSUES. Journal
of Marketing 45(4), 60-67.

DELLAROCAS C, GAO G and NARAYAN R (2010) Are consumers more likely to contribute online
reviews for hit or niche products? Journal of Management Information Systems 27(2), 127-158.

DELLAROCAS C, XIAOQUAN Z and AWAD NF (2007) Exploring the value of online product reviews
in forecasting sales: The case of motion pictures. Journal of Interactive Marketing (John Wiley &
Sons) 21(4), 23-45.

DONNELLY JR JH and IVANCEVICH JM (1974) A Methodology for Identifying Innovator


Characteristics of New Brand Purchasers. Journal of Marketing Research (JMR) 11(3), 331-334.

DOYLE P (1976) The realities of the product life cycle. Quarterly Review of Marketing 1(4), 1-6.

DUAN W, GU B and WHINSTON A (2009) Informational cascades and software adoption on the
internet: an empirical investigation. MIS quarterly 33(1), 23-48.

DUAN W, GU B and WHINSTON AB (2008) Do online reviews matter?-An empirical investigation of


panel data. Decision Support Systems 45(4), 1007-1016.

EASTON H (1940) Word frequency dictionary. New York: Dover

ELIAS GS, GARFIELD R, GUTSCHERA KR and WHITLEY P (2012) Characteristics of games. MIT
Press

FLESCH RF (1951) How to test readability. New York: Harper

FOWLER GA and AVILA JD (2009) On the Internet, everyone’s a critic but they’re not very critical.
Wall Street Journal Available at: http://www.wsj.com/articles/SB125470172872063071.

FOX HW and RINK DR (1977) Coordination of purchasing with sales trends. Journal of Purchasing and
Materials Management 13(4), 10-18.

GAMON M, AUE A, CORSTON-OLIVER S and RINGGER E (2005) Pulse: Mining customer opinions
from free text. In: Advances in Intelligent Data Analysis VI. pp. 121-132, Springer.

GAO GG, GREENWOOD BN, AGARWAL R and JEFFREY S (2015) Vocal Minority and Silent
Majority: How Do Online Ratings Reflect Population Perceptions of Quality? MIS Quarterly 39(3),
565–589.

GARVIN DA (1988) Managing quality: The strategic and competitive edge. Free Pr.

GODES D and SILVA JC (2012) Sequential and Temporal Dynamics of Online Opinion. Marketing
Science 31(3), 448-473.

GOES PB, LIN M and AU YEUNG C (2014) ‘Popularity Effect’ in User-Generated Content: Evidence
from Online Product Reviews. Information Systems Research 25(2), 222-238.

GOLDER PN and TELLIS GJ (2004) Growing, Growing, Gone: Cascades, Diffusion, and Turning Points
in the Product Life Cycle. Marketing Science 23(2), 207-218.

GOLDSMITH RE, FREIDEN JB and EASTMAN JK (1995) The generality/specificity issue in consumer
innovativeness research. Technovation 15(10), 601-612.

38
GOLDSMITH RE and HOFACKER CF (1991) Measuring consumer innovativeness. Journal of the
Academy of Marketing Science 19(3), 209-221.

HAUSMAN JA (1978) Specification tests in econometrics. Econometrica: Journal of the Econometric


Society, 1251-1271.

HENNIG‐THURAU T, GWINNER KP, WALSH G and GREMLER DD (2004) Electronic word‐of‐


mouth via consumer‐opinion platforms: What motivates consumers to articulate themselves on the
Internet? Journal of interactive marketing 18(1), 38-52.

HIRSCHMAN EC (1980) Innovativeness, novelty seeking, and consumer creativity. Journal of consumer
research, 283-295.

HOEFFLER S (2003) Measuring preferences for really new products. Journal of Marketing Research
40(4), 406-420.

HOFFER CW (1975) Toward a Contingency Theory of Business Strategy. Academy of Management


Journal 18(4), 784-810.

HOLBERT RL (2014) DICTION as a Tool for Studying the Mass Media. Communication and Language
Analysis in the Public Sphere, 462.

HOLBROOK MB (1993) Nostalgia and consumption preferences: Some emerging patterns of consumer
tastes. Journal of Consumer research, 245-256.

HOLSTI OR (1969) Content analysis for the social sciences and humanities. Addison-Wesle.

KARANDE K, MERCHANT A and SIVAKUMAR K (2011) Erratum to: Relationships among time
orientation, consumer innovativeness, and innovative behavior: the moderating role of product
characteristics. AMS review 1(2), 99-116.

KHAMMASH M and GRIFFITHS GH (2011) ‘Arrivederci CIAO.com, Buongiorno Bing.com’-


Electronic word-of-mouth (eWOM), antecedences and consequences. International Journal of
Information Management 31(1), 82-87.

KOSKI H and KRETSCHMER T (2007) Innovation and dominant design in mobile telephony. Industry
and Innovation 14(3), 305-324.

KOZINETS RV (2002) The field behind the screen: using netnography for marketing research in online
communities. Journal of marketing research, 61-72.

KRIPPENDORFF K (2008) Testing the reliability of content analysis data: What is involved and why.
The content analysis reader, 350-357.

KU Y-C, WEI C-P and HSIAO H-W (2012) To whom should I listen? Finding reputable reviewers in
opinion-sharing communities. Decision Support Systems 53(3), 534-542.

LARSEN K (2005) Generalized naive Bayes classifiers. ACM SIGKDD Explorations Newsletter 7(1), 76-
81.

LEVITT T (1981) Marketing intangible products and product intangibles. Cornell Hotel and Restaurant
Administration Quarterly 22(2), 37-44.

39
LI N and WU DD (2010) Using text mining and sentiment analysis for online forums hotspot detection
and forecast. Decision support systems 48(2), 354-368.

LI X and HITT LM (2008) Self-selection and information role of online product reviews. Information
Systems Research 19(4), 456-474.

LIU B (2010) Sentiment analysis and subjectivity. In: Handbook of Natural Language Processing. pp.
627-666.

MAHAJAN V, MULLER E and BASS FM (1990) New product diffusion models in marketing: A review
and directions for research. The Journal of Marketing, 1-26.

MARTINEZ E, POLO Y and CARLOS F (1998) The acceptance and diffusion of new consumer
durables: differences between first and last adopters. Journal of Consumer Marketing 15(4), 319-342.

MCGLOHON M, GLANCE NS and REITER Z Star Quality: Aggregating Reviews to Rank Products and
Merchants.

MIDGLEY DF (1976) A Simple Mathematical Theory of Innovative Behavior. Journal of Consumer


Research 3(1), 31-41.

MIDGLEY DF (1977) Innovation and new product marketing. Croom Helm.

MIDGLEY DF and DOWLING GR (1978) Innovativeness: The concept and its measurement. Journal of
consumer research, 229-242.

MIDGLEY DF and DOWLING GR (1993) A longitudinal study of product form innovation: The
interaction between predispositions and social messages. Journal of Consumer Research, 611-625.

MILLER GA (1995) Word Net: A Lexical Database for English. Communications of the ACM 38(11),
39-41.

MOE WW and SCHWEIDEL DA (2012) Online Product Opinions: Incidence, Evaluation, and
Evolution. Marketing Science 31(3), 372-386.

MOON S, BERGEY PK and IACOBUCCI D (2010) Dynamic Effects Among Movie Ratings, Movie
Revenues, and Viewer Satisfaction. Journal of Marketing 74(1), 108-121.

MOORE GA (1991) Crossing the chasm: Marketing and selling high-tech products to mainstream
consumers. New York: HarperBusiness.

MUDAMBI SM and SCHUFF D (2010) What makes a helpful review? A study of customer reviews on
Amazon.com. MIS quarterly 34(1), 185-200.

NAN HU, PAVLOU PA and JIE Z (2009) Overcoming the J-shaped Distribution of Product Reviews.
Communications of the ACM 52(10), 144-147.

NELSON P (1970) Information and consumer behavior. Journal of political economy 78(2), 311-329.

NELSON P (1974) Advertising as information. Journal of political economy 82(4), 729-754.

OBER S, ZHAO JJ, DAVIS R and ALEXANDER MW (1999) Telling It like It Is: The Use of Certainty
in Public Business Discourse. Journal of Business Communication 36(3), 280-300.

OGDEN CK (1968) Basic English: international second language. Harcourt, Brace & World.

40
OSTLUND LE (1974) Perceived innovation attributes as predictors of innovativeness. Journal of
consumer research, 23-29.

PAN Y and ZHANG JQ (2011) Born Unequal: A Study of the Helpfulness of User-Generated Product
Reviews. Journal of Retailing 87(4), 598-612.

PANG B and LEE L (2004) A sentimental education: Sentiment analysis using subjectivity
summarization based on minimum cuts. In: Proceedings of the 42nd annual meeting on Association
for Computational Linguistics. p. 271, Association for Computational Linguistics.

PANG B and LEE L (2008) Opinion mining and sentiment analysis. Foundations and trends in
information retrieval 2(1-2), 1-135.

PARASURAMAN A and COLBY CL (2007) Techno-ready marketing: How and why your customers
adopt technology. The Free Press.

PERES R, MULLER E and MAHAJAN V (2010) Innovation diffusion and new product growth models:
A critical review and research directions. International Journal of Research in Marketing 27(2), 91-
106.

RAM S and JUNG H (1994) Innovativeness in product usage: A comparison of early adopters and early
majority. Psychology & Marketing 11(1), 57-67.

RAO AR and RUEKERT RW (1994) Brand alliances as signals of product quality. Sloan management
review 36(1), 87.

ROGERS EM (1963) What are innovators like? Theory into Practice 2(5), 252-256.

ROGERS EM (1976) New Product Adoption and Diffusion. Journal of Consumer Research 2(4), 290-
301.

ROGERS EM and KINCAID DL (1981) Communication networks: toward a new paradigm for research.

ROGERS EM (2003) Diffusion of innovations. New York., NY, Free Press.

SEBASTIANI F (2002) Machine learning in automated text categorization. ACM computing surveys
(CSUR) 34(1), 1-47.

STEENKAMP J-BE, HOFSTEDE F ter and WEDEL M (1999) A cross-national investigation into the
individual and national cultural antecedents of consumer innovativeness. The Journal of Marketing,
55-69.

STÖCKL R, KOSYAK A, WALTER B and HESS T (2006) Success factors of communities for user
driven content: The case of Ciao.com. AMCIS 2006 Proceedings, 525.

TORNATZKY LG and KLEIN KJ (1982) Innovation characteristics and innovation adoption-


implementation: A meta-analysis of findings. Engineering Management, IEEE Transactions on (1),
28-45.

VAN DEN BULTE C and LILIEN GL (2001) Medical Innovation Revisited: Social Contagion versus
Marketing Effort. American Journal of Sociology 106(5), 1409-1435.

VOGT CA and FESENMAIER DR (1998) Expanding the functional information search model. Annals of
Tourism Research 25(3), 551-578.

41
WEI R (2001) FROM LUXURY TO UTILITY: A LONGITUDINAL ANALYSIS OF CELL PHONE
LAGGARDS. Journalism & Mass Communication Quarterly 78(4), 702-719.

WINTER B (2013) Linear models and linear mixed effects models in R with linguistic applications. arXiv
preprint arXiv:1308.5499.

WOOD SL and MOREAU CP (2006) From fear to loathing? How emotion influences the evaluation and
early use of innovations. Journal of Marketing 70(3), 44-57.

WU F and HUBERMAN BA (2008) How public opinion forms. In: Internet and Network Economics. pp.
334–341, Springer. Available at: http://link.springer.com/10.1007%2F978-3-540-92185-1_39.

XU YC, ZHANG C, XUE L and YEO LL (2008) Product adoption in online social network. ICIS 2008
Proceedings, 200.

YUBO C and JINHONG X (2008) Online Consumer Review: Word-of-Mouth as a New Element of
Marketing Communication Mix. Management Science 54(3), 477-491.

42

View publication stats

You might also like