《电子商务 E-business》阅读文献：What do we know about mobile Internet adopters A cluster analysis

Availableonlineatwww.sciencedirect.col SCIENCE DIRECT. INFORMATION MANAGEMENT ELSEVIER Information Management 43(2006)127-141 www.elsevier.com/locate/dsw what do we know about mobile Internet adopters? a cluster analysis Shintaro okazaki Department of Finance and Marketing Research, College of Economics and Business Administration, Received 3 September 2004; received in revised form 8 March 2005; accepted 1 May 2005 Available online 16 june 200 Abstract consumers'attitude and their demographic characteristics have been only cursorily examined. The objective of our study was to fill this gap, by applying a two-step cluster analysis in profiling mobile Internet adopters in Japan. The findings suggest that four clusters exist: they exhibit distinct profile patterns. Paradoxical results were found within one, affluent single youth, which was further divided into two clusters: freelance, highly educated professionals had the most negative perception of mobile Internet adoption, whereas clerical office workers had the most positive perception. Married housewives and company executives also exhibited a positive attitude toward mobile Internet usage. C 2005 Elsevier B.V. All rights reserved Keywords: Mobile; Diffusion; Innovation; Internet: i-Mode; Japan; Uses and gratifications 1. Introduction number of Internet-connected PCs[44. Such dramatic convergence of the Internet and mobile telephony may In world markets, the rapid adoption of Web-enabled be attributed, in particular, to activity in Asian and mobile handsets has become increasingly important to Scandinavian countries, where penetration growth has Is professionals. A recent survey in 13 countries been meteoric. A recent survey indicated that roughly revealed an increase in usage of 145%0, reaching 79 70 million people in Japan(55% of the population)have million users in 2003, while the number of global signed up for Internet access from their cellular phone mobile Internet adopters has been predicted to reach compared with 12% of the population in the USA nearly 600 million by 2008 [19,35]. A pessimistic [13, 15]. In fact, the Japanese see cell phones or Keitai as forecast estimated that, by the year 2005, the number of devices for surfing the Internet while Americans use Internet-connected mobile phones would exceed the their laptops luch of this success can be traced to February 1999, when NTT DoCoMo, Japans leading mobile E-mailaddress:obarquitec(@coac.net. operator, launched the i-mode service. This isa mobile 0378-7206/S- see front matter 2005 Elsevier B V. All rights reserved doi:10.1016jm2005.05.001

What do we know about mobile Internet adopters? A cluster analysis Shintaro Okazaki * Department of Finance and Marketing Research, College of Economics and Business Administration, Autonomous University of Madrid, Cantoblanco, 28049 Madrid, Spain Received 3 September 2004; received in revised form 8 March 2005; accepted 1 May 2005 Available online 16 June 2005 Abstract Despite the increasing importance of wireless Internet use via Web-enabled mobile telephony, the relationship between consumers’ attitude and their demographic characteristics have been only cursorily examined. The objective of our study was to fill this gap, by applying a two-step cluster analysis in profiling mobile Internet adopters in Japan. The findings suggest that four clusters exist; they exhibit distinct profile patterns. Paradoxical results were found within one, affluent single youth, which was further divided into two clusters: freelance, highly educated professionals had the most negative perception of mobile Internet adoption, whereas clerical office workers had the most positive perception. Married housewives and company executives also exhibited a positive attitude toward mobile Internet usage. # 2005 Elsevier B.V. All rights reserved. Keywords: Mobile; Diffusion; Innovation; Internet; i-Mode; Japan; Uses and gratifications 1. Introduction In world markets, the rapid adoption of Web-enabled mobile handsets has become increasingly important to IS professionals. A recent survey in 13 countries revealed an increase in usage of 145%, reaching 79 million users in 2003, while the number of global mobile Internet adopters has been predicted to reach nearly 600 million by 2008 [19,35]. A pessimistic forecast estimated that, by the year 2005, the number of Internet-connected mobile phones would exceed the number of Internet-connected PCs [44]. Such dramatic convergence of the Internet and mobile telephony may be attributed, in particular, to activity in Asian and Scandinavian countries, where penetration growth has been meteoric. A recent survey indicated that roughly 70 million people in Japan (55% of the population) have signed up for Internet access from their cellular phones, compared with 12% of the population in the USA [13,15]. In fact, the Japanese see cell phones orKeitai as devices for surfing the Internet while Americans use their laptops. Much of this success can be traced to February 1999, when NTT DoCoMo, Japan’s leading mobile operator, launched the i-mode service. This is ‘a mobile www.elsevier.com/locate/dsw Information & Management 43 (2006) 127–141 * Tel.: +34 91 497 3552; fax: +34 91 802 0974. E-mail address: obarquitec@coac.net. 0378-7206/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.im.2005.05.001

S Okazaki/Information Management 43(2006)127-141 phone service offering continuous, always-on Internet Second, despite obvious cultural differences,an access based on packet-switching technology'[6]. empirical investigation of cons can serve as a Through a handset, users can access a micro-browser useful case study for other markets. DoCoMo's i-mode that offers services such as e-mail, data search, instant for example, has expanded to European countries; the messaging,Internet, and i-Imenu E-mail is considered adopters include E-Plus(Germany), KPN Mobile the most popular killer app, and 71% of i-mode users (Netherlands ), BASE (Belgium), Bouygues Telecon receive an e-mail newsletter [29, 38]. The i-menu acts( France), Telefonica Moviles( Spain), wind (Italy), and as a mobile portal resulting in approximately 4100 COSMOTE( Greece), and total subscribers reached 1.5 fficial and 50,000 unofficial sites offering diverse million by the end of 2003, from 270,000 at the end of additional functions 321 2002 [1]. The software platform and its content have One of the unusual features of i-mode is the way it been converted into added value resulting from the develops i-mode content. Instead of purchasing it, solution and partner network, possibly providing wider DoCoMo allows'designated third parties to provide implications [28]. Therefore, it is importan fee-based content and services with collection through a theoretical basis of the attitudes and demographics of the monthly phone bill. By October 2003, more than mobile Internet adopters 0 million subscribers to 2G and 3G i-mode Internet services existed. The key to understanding this growth lies in the profiles of mobile consumer segments but 3. Mobile content creation in the m-commerce little effort has addressed the fundamental question: value chain what are the attitudinal and demographic characte istics of mobile Internet adopters a chain of value-adding activities in The purpose of our study was to fill this gap, by commerce involves two global perspectives: conducting a two-step cluster analysis to identify and infrastructure-and-services [5]. The curren m ts of mobile Internet adopters in by global mobile players, however, pla ace more Our method overcomes the limitations of traditional strategic emphasis on content, the value chain of cluster analysis by: (1) considering both continuous content aggregation, its management, and access [25] and categorical variables and (2)automatically DoCoMo's i-mode, for example, is a ' semi-walled determining the number of clusters or segments on garden'controlled by its packet network and server the basis of objective statistical criteria. system; in this many different contents may be structured into official(approved) and unofficial(non- approved) providers(Fig. 1). However, only official 1gnilcance marge for content through DoCoMo's subscription billing system, which offers multiple .i. This study contributes to electronic commerce incentives for active content creation. literature in two ways. First, research on mobile The mobile portals play a key role in adding value to Internet adopters has primarily focused on the area of mobile market-making [5]. For instance, when users direct marketing: user profiles have rarely been select i-mode, they are presented with i-menu with links considered. The majority of the studies result from to personal information management applications, sporadic industry reports that leave an important offering users a one-stop shop solution. The increasing question unanswered: in developing effective busi- sophistication of mobile handsets has accommodated ness-to-consumer m-commerce strategies, how do diverse killer apps, such as built-in GPS, music information managers identify mobile Internet adop- downloads, videos, e-coupon for discounts. bill ters? And: what kind of demographic and psycho- payment, and even karaoke machines. Strategically, graphic segments do they have? An effort to classify this portal imposes no additional infrastructure costs, may therefore help in driving the development and number of third-party content providers and aggra e mobile Internet adopters based on specific attributes because the content creation can be arranged with execution of customer strategy and targeting of tors. As a result, in 2004 DoCoMo's group net profits more than tripled to 650 billion yen [2] and its

phone service offering continuous, always-on Internet access based on packet-switching technology’ [6]. Through a handset, users can access a micro-browser that offers services such as e-mail, data search, instant messaging, Internet, and i-menu. E-mail is considered the most popular killer app, and 71% of i-mode users receive an e-mail newsletter [29,38]. The i-menu acts as a mobile portal resulting in approximately 4100 official and 50,000 unofficial sites offering diverse additional functions [32]. One of the unusual features of i-mode is the way it develops i-mode content. Instead of purchasing it, DoCoMo allows ‘designated third parties to provide fee-based content and services with collection through the monthly phone bill’. By October 2003, more than 40 million subscribers to 2G and 3G i-mode Internet services existed. The key to understanding this growth lies in the profiles of mobile consumer segments but little effort has addressed the fundamental question: what are the attitudinal and demographic character￾istics of mobile Internet adopters? The purpose of our study was to fill this gap, by conducting a two-step cluster analysis to identify specific segments of mobile Internet adopters in Japan. Our method overcomes the limitations of traditional cluster analysis by: (1) considering both continuous and categorical variables and (2) automatically determining the number of clusters or segments on the basis of objective statistical criteria. 2. Significance of the study This study contributes to electronic commerce literature in two ways. First, research on mobile Internet adopters has primarily focused on the area of direct marketing; user profiles have rarely been considered. The majority of the studies result from sporadic industry reports that leave an important question unanswered: in developing effective busi￾ness-to-consumer m-commerce strategies, how do information managers identify mobile Internet adop￾ters? And: what kind of demographic and psycho￾graphic segments do they have? An effort to classify mobile Internet adopters based on specific attributes may therefore help in driving the development and execution of customer strategy and targeting of customers. Second, despite obvious cultural differences, an empirical investigation of consumers can serve as a useful case study for other markets. DoCoMo’s i-mode, for example, has expanded to European countries; the adopters include E-Plus (Germany), KPN Mobile (Netherlands), BASE (Belgium), Bouygues Telecom (France), Telefonica Moviles (Spain), Wind (Italy), and COSMOTE (Greece), and total subscribers reached 1.5 million by the end of 2003, from 270,000 at the end of 2002 [1]. The software platform and its content have been converted into added value resulting from the solution and partner network, possibly providing wider implications[28]. Therefore, it is important to establish a theoretical basis of the attitudes and demographics of mobile Internet adopters. 3. Mobile content creation in the m-commerce value chain A chain of value-adding activities in mobile￾commerce involves two global perspectives: content and infrastructure-and-services [5]. The current moves by global mobile players, however, place more strategic emphasis on content, the value chain of content aggregation, its management, and access [25]. DoCoMo’s i-mode, for example, is a ‘semi-walled garden’ controlled by its packet network and server system; in this many different contents may be structured into official (approved) and unofficial (non￾approved) providers (Fig. 1). However, only official providers can charge for content through DoCoMo’s subscription billing system, which offers multiple incentives for active content creation. The mobile portals play a key role in adding value to mobile market-making [5]. For instance, when users select i-mode, they are presented with i-menu with links to personal information management applications, offering users a one-stop shop solution. The increasing sophistication of mobile handsets has accommodated diverse killer apps, such as built-in GPS, music downloads, videos, e-coupons for discounts, bill payment, and even karaoke machines. Strategically, this portal imposes no additional infrastructure costs, because the content creation can be arranged with a number of third-party content providers and aggrega￾tors. As a result, in 2004 DoCoMo’s group net profits more than tripled to 650 billion yen [2] and its 128 S. Okazaki / Information & Management 43 (2006) 127–141

S. Okazaki/Information Management 43(2006)127-141 DoCoMo Network Provider Packet Radio tower i-mode server Unofficial Fig. 1. Functions of i-mode Internet service. Source: Reprinted by permission of Ref [28]- Table 1 Mobile Internet services in Japan Vodafone NTT DoCoMo Vodafone Au(KDDI Market share 60.5% 20.1% Number of subscribers 14.36 Markup language HDML 3G/2.5G network W-cdma Cdma2000 IX mum capacity(packet) 384 kbps er of official sites 50 >2000 1000 yen monthly charge 300 yen monthly charge 300 yen monthly charge +0.1 yen per packet +0.3 yen per packet +0.2 yen per packet In million [43]. Ref. [41 Refs.[17,32] d refs.[3,31,46 competitors, Vodafone and au(KDDI), implemented 4.1. Demographic profiling similar mobile Internet services. albeit using differe underlying technologies [17](Table 1) Demographic profiling is the process of the market by considering personal similarit differences, such as gender, age, marital 4. Profiling mobile Internet adopters occupation, monthly allowance, and household struc- ture. Such descriptive attributes have been used In pursuing this study, it was necessary to establish most industry surveys a conceptual framework for assessing the structure of Earlier industry reports have indicated that mobile the mobile Internet market. In general, markets consist Internet penetration was highest among young affluent of a number of segments, each of which is made up of males [42]. Most of them(83%)were found to use the natural groupings of customers[14]. Consumers can mobile Internet for personal purposes, but a sub- be split into different segments or clusters, within stantial portion (49%) also used it for work. WAP which customers have similar characteristics and adopters in Taiwan were predominantly young single needs [27]. The combined benefits were sought by males(21-40 years old) with middle income [18]. A adopting a two-step cluster analysis recent US industry report also indicated that a typical

competitors, Vodafone and au (KDDI), implemented similar mobile Internet services, albeit using different underlying technologies [17] (Table 1). 4. Profiling mobile Internet adopters In pursuing this study, it was necessary to establish a conceptual framework for assessing the structure of the mobile Internet market. In general, markets consist of a number of segments, each of which is made up of natural groupings of customers [14]. Consumers can be split into different segments or clusters, within which customers have similar characteristics and needs [27]. The combined benefits were sought by adopting a two-step cluster analysis. 4.1. Demographic profiling Demographic profiling is the process of splitting the market by considering personal similarities and differences, such as gender, age, marital status, occupation, monthly allowance, and household struc￾ture. Such descriptive attributes have been used in most industry surveys. Earlier industry reports have indicated that mobile Internet penetration was highest among young affluent males [42]. Most of them (83%) were found to use the mobile Internet for personal purposes, but a sub￾stantial portion (49%) also used it for work. WAP adopters in Taiwan were predominantly young single males (21–40 years old) with middle income [18]. A recent US industry report also indicated that a typical S. Okazaki / Information & Management 43 (2006) 127–141 129 Fig. 1. Functions of i-mode Internet service. Source: Reprinted by permission of Ref. [28]. Table 1 Mobile Internet services in Japan i-Mode Vodafone live EZweb Operator group NTT DoCoMo Vodafone Au (KDDI) Market share 60.5% 19.5% 20.1% Number of subscribersa 41.32 12.95 14.36 Markup language cHTML MML HDML 3G/2.5G networkb W-cdma W-cdma Cdma2000 1X Maximum capacity (packet)b 384 kbps 384 kbps 144 kbps Number of official sitesc 4100 650 >2000 Charged 1000 yen monthly charge + 0.1 yen per packet 300 yen monthly charge + 0.3 yen per packet 300 yen monthly charge + 0.2 yen per packet a In million [43]. b Ref. [41]. c Refs. [17,32]. d Refs. [3,31,46]

S Okazaki/Information Management 43(2006)127-141 user was male, between 18 and 34 years old, with a advantage over the idea it supersedes; (2)compatibility household income of uss 60,000 or more. Such with existing technology; (3)perceived complexity of findings suggest that the likelihood of adopting mobile its understanding and use;(4)trialability; (5)observa IT innovations is dependent on age and income, while bility to others. These variables collectively result in the effect of gender on mobile Internet service user attitude toward the mobile Internet, which in turn adoption remains uncertain. For example, a survey affects consumers' behavioral intent to use it of mobile text messaging by 500 British young adults Furthermore, prior research on Japanese i-mode found hardly any differences due to age or gender [7]. adoption suggested that the credibility of a new In addition, a survey of mobile banking adoption in communication channel was key to the choice of South Africa found that the majority(67%)of the access. This was important, because more than 20% of respondents were young, educated groups, either Japanese mobile Internet adopters accessed news and employed or studying or both, with the gender city guides, which often acted as a trusted information distribution approximately equal [9] source in their daily lives. Also, with logo branding sponsorship campaigns increasingly popul 4.2. Attitudinal profiling mobile sites, users may also have wanted to be sure of the trustworthiness of such paid-publicity. By choo While many industry surveys have focused on ing the i-menu, users could access a content-based descriptive characteristics, little diagnostic informa- platform, Tokusuru Menu(menu to your advantage, in tion has been provided on mobile Internet adopters, Japanese), which featured text banner ads from where similar demographic data may be differentiated sponsoring companies. By clicking, consumers could by the adopters'psychological motives. We therefore browse further detailed information pages that offered attempted to uncover profiles on the basis of: (1)the discounts, coupons, free-samples, sweepstakes, and uses and gratifications of adopters with the new media ring-tone downloads, etc. Therefore, in forming basic and (2)the diffusion of the mobile Internet as an attitudes toward and intention to access the mobile innovation Internet, individuals may have been reminded of the The uses and gratifications theory is axiomatic; it underlying information credibility and trustworthi argues that psychological needs shape an audiences ness adoption of the media [23]. This theory is primarily grounded on three basic tenets: media adopters are goal-directed, active media-users, and aware of their 5. Methodology needs. Because the mobile Internet service has been characterized as being highly personal, interactive, 5. 1. The questionnaire and immediate [10], important attributes can be found by profiling individuals according to the degree to Our study was part of an omnibus research project which they spontaneously perceive the medium to be conducted by an advertising research foundation in irritating, informative, or entertaining. These have Tokyo. The survey instrument included face, common been identified as principal motivations in wired and specific questions. The face questions covered Internet service adoption [11, 22, 24], while prior general demographic information, such as gender, research on the mobile Internet has made similar occupation, marital status, monthly allowance, and suggestions [4, 21) hours spent outside home. The common questions Rogers [36] defines the diffusion as 'the process by were related to general perceptions of media selection, which an innovation is communicated through certain leisure activity, consumption attitude, etc. Finally, the channels over time among the members of a social specific questions addressed attitudinal dimensions system. Therefore, it is a communication of new ideas, with respect to i-mode platforms, including content nich'participants create and share information with and source credibility, informativeness, entertainment, one another in order to reach a mutual understanding,. irritation, general liking, and willingness to access Thus, the mobile Internet seems to satisfy the five emed re asonable to assume that usage of e-mail principal characteristics of innovation:(1)relative messaging and access to mobile portals were two key

user was male, between 18 and 34 years old, with a household income of US$ 60,000 or more. Such findings suggest that the likelihood of adopting mobile IT innovations is dependent on age and income, while the effect of gender on mobile Internet service adoption remains uncertain. For example, a survey of mobile text messaging by 500 British young adults found hardly any differences due to age or gender [7]. In addition, a survey of mobile banking adoption in South Africa found that the majority (67%) of the respondents were ‘young, educated groups, either employed or studying or both’, with the gender distribution approximately equal [9]. 4.2. Attitudinal profiling While many industry surveys have focused on descriptive characteristics, little diagnostic informa￾tion has been provided on mobile Internet adopters, where similar demographic data may be differentiated by the adopters’ psychological motives. We therefore attempted to uncover profiles on the basis of: (1) the uses and gratifications of adopters with the new media and (2) the diffusion of the mobile Internet as an innovation. The uses and gratifications theory is axiomatic; it argues that psychological needs shape an audiences’ adoption of the media [23]. This theory is primarily grounded on three basic tenets: media adopters are goal-directed, active media-users, and aware of their needs. Because the mobile Internet service has been characterized as being highly personal, interactive, and immediate [10], important attributes can be found by profiling individuals according to the degree to which they spontaneously perceive the medium to be irritating, informative, or entertaining. These have been identified as principal motivations in wired Internet service adoption [11,22,24], while prior research on the mobile Internet has made similar suggestions [4,21]. Rogers [36] defines the diffusion as ‘the process by which an innovation is communicated through certain channels over time among the members of a social system’. Therefore, it is a communication of new ideas, in which ‘participants create and share information with one another in order to reach a mutual understanding’. Thus, the mobile Internet seems to satisfy the five principal characteristics of innovation: (1) relative advantage over the idea it supersedes; (2) compatibility with existing technology; (3) perceived complexity of its understanding and use; (4) trialability; (5) observa￾bility to others. These variables collectively result in user attitude toward the mobile Internet, which in turn affects consumers’ behavioral intent to use it. Furthermore, prior research on Japanese i-mode adoption suggested that the credibility of a new communication channel was key to the choice of access. This was important, because more than 20% of Japanese mobile Internet adopters accessed news and city guides, which often acted as a trusted information source in their daily lives. Also, with logo branding and sponsorship campaigns increasingly popular in mobile sites, users may also have wanted to be sure of the trustworthiness of such paid-publicity. By choos￾ing the i-menu, users could access a content-based platform, Tokusuru Menu (menu to your advantage, in Japanese), which featured text banner ads from sponsoring companies. By clicking, consumers could browse further detailed information pages that offered discounts, coupons, free-samples, sweepstakes, and ring-tone downloads, etc. Therefore, in forming basic attitudes toward and intention to access the mobile Internet, individuals may have been reminded of the underlying information credibility and trustworthi￾ness. 5. Methodology 5.1. The questionnaire Our study was part of an omnibus research project conducted by an advertising research foundation in Tokyo. The survey instrument included face, common, and specific questions. The face questions covered general demographic information, such as gender, occupation, marital status, monthly allowance, and hours spent outside home. The common questions were related to general perceptions of media selection, leisure activity, consumption attitude, etc. Finally, the specific questions addressed attitudinal dimensions with respect to i-mode platforms, including content and source credibility, informativeness, entertainment, irritation, general liking, and willingness to access. It seemed reasonable to assume that usage of e-mail messaging and access to mobile portals were two key 130 S. Okazaki / Information & Management 43 (2006) 127–141

S. Okazaki/Infonnation Management 43(2006)127-14 indicators of mobile Internet adopters. Therefore In the first step, original cases are grouped into photographic images of i-menu portal sites were preclusters that are then used in place of the raw data nserted in the questionnaire in asking respondents for in the hierarchical clustering Based upon its similarity their general opinions on the use of such services. to existing preclusters, each successive case is added With regard to e-mail usage, the questionnaire to form a new precluster, using a likelihood distance included a filtering question: did respondents use e- measure as the similarity criterion. Cases are assigned mail messaging via mobile telephony? to the precluster that maximizes a log-likelihood function. In the second step, the preclusters are 5.2. sample grouped using the standard agglomerative clustering algorithm, producing a range of solutions, which is The sample involved stratified random sampling then reduced to the best number of clusters on the basis according to age and gender distribution. The popula- of Schwarz's Bayesian inference criterion BIC) tion was based on the Citizens Registry Book of the which is known as one of the most useful and objective Tokyo Metropolitan District. Questionnaires were selection criteria, because it essentially avoids the distributed to 1623 residents in the greater Tokyo area. arbitrariness in traditional clustering techniques. In A professional marketing organization was employed addition, both background noise and outliers can be for this task, and researchers visited each respondent to identified and screened out. leave the questionnaire. A total of 786 responses were collected in the next month, giving an effective response 6.2. Categorical and continuous variables rate of 48.4%. However, only 612 responses were included in the data analysis: those who regularly used ne categorical and continuous variables are shown the e-mail message service via the mobile phone in Tables 2 and 3, respectively. The categorical vari- ables involve gender, age, marital status, occupation, monthly allowance, and household structure. Monthly 6. Statistical treatment allowance was chosen over monthly income, on the assumption that the level of mobile usage expenditure is 6.1. The two-step cluster analysis a function of disposable allowance rather than of total income. Each variable was assessed on a categorical Traditionally, cluster analysis has been used for scale with no multiple responses allowed. The conti empirical classification of objects [16]. It is an explo- nuous variables were associated with general percep- ratory technique that has been widely applied in diverse tions of the mobile platform: content credibility, source ciplines credibility, informativeness, entertainment, irritation, and Ghose[8]applied a latent class modeling approach general liking, and willingness to access Each measure to segment Web shoppers based on demographics and consisted of a multiple-item scale, as indicated benefit sought while Jih and Lee [20] attempted to Table 3 segment cellular phone users according to their retail To ensure the adequacy of the selected variables, two shopping motives. Thus, this technique was deemed preliminary analyses were conducted in an attempt to appropriate in forming groups according to the simi- identify significant differences between mobile e-mail larity of their demographic and attitudinal variables users and non-users. This was needed because if there Our study used a statistical program, TwoStep were no significant differences between the two groups, Cluster in SPSS 12.0; this had been suggested as then profiling made little sense. First, the Pearson chi appropriate in clustering large data sets with mixed se square test was performed for each of the the categoric attributes [30]. The method is based on a distance variables across the two groups. The expected values in measure that enables data with both continuous and each cell were greater than 1 and most cells had categorical attributes to be clustered. This is derived expected values greater than 5. Significant differences from a probabilistic model in which the distance were detected at P<0.00l for all variables between e between two clusters is equivalent to the decrease in mail users and non-users. Second, a MANOVA was log-likelihood function as a result of merging [12] conducted with type of use or non-use as independent

indicators of mobile Internet adopters. Therefore, photographic images of i-menu portal sites were inserted in the questionnaire in asking respondents for their general opinions on the use of such services. With regard to e-mail usage, the questionnaire included a filtering question: did respondents use e￾mail messaging via mobile telephony? 5.2. Sample The sample involved stratified random sampling according to age and gender distribution. The popula￾tion was based on the Citizens Registry Book of the Tokyo Metropolitan District. Questionnaires were distributed to 1623 residents in the greater Tokyo area. A professional marketing organization was employed for this task, and researchers visited each respondent to leave the questionnaire. A total of 786 responses were collected in the next month, giving an effective response rate of 48.4%. However, only 612 responses were included in the data analysis: those who regularly used the e-mail message service via the mobile phone. 6. Statistical treatment 6.1. The two-step cluster analysis Traditionally, cluster analysis has been used for empirical classification of objects [16]. It is an explo￾ratory technique that has been widely applied in diverse disciplines for its partitioning ability; e.g., Bhatnager and Ghose [8] applied a latent class modeling approach to segment Web shoppers based on demographics and benefit sought while Jih and Lee [20] attempted to segment cellular phone users according to their retail shopping motives. Thus, this technique was deemed appropriate in forming groups according to the simi￾larity of their demographic and attitudinal variables. Our study used a statistical program, TwoStep Cluster in SPSS 12.0; this had been suggested as appropriate in clustering large data sets with mixed attributes [30]. The method is based on a distance measure that enables data with both continuous and categorical attributes to be clustered. This is derived from a probabilistic model in which the distance between two clusters is equivalent to the decrease in log-likelihood function as a result of merging [12]. In the first step, original cases are grouped into preclusters that are then used in place of the raw data in the hierarchical clustering. Based upon its similarity to existing preclusters, each successive case is added to form a new precluster, using a likelihood distance measure as the similarity criterion. Cases are assigned to the precluster that maximizes a log-likelihood function. In the second step, the preclusters are grouped using the standard agglomerative clustering algorithm, producing a range of solutions, which is then reduced to the best number of clusters on the basis of Schwarz’s Bayesian inference criterion (BIC), which is known as one of the most useful and objective selection criteria, because it essentially avoids the arbitrariness in traditional clustering techniques. In addition, both background noise and outliers can be identified and screened out. 6.2. Categorical and continuous variables The categorical and continuous variables are shown in Tables 2 and 3, respectively. The categorical vari￾ables involve gender, age, marital status, occupation, monthly allowance, and household structure. Monthly allowance was chosen over monthly income, on the assumption that the level of mobile usage expenditure is a function of disposable allowance rather than of total income. Each variable was assessed on a categorical scale with no multiple responses allowed. The conti￾nuous variables were associated with general percep￾tions of the mobile platform: content credibility, source credibility, informativeness, entertainment, irritation, general liking, and willingness to access. Each measure consisted of a multiple-item scale, as indicated in Table 3. To ensure the adequacy of the selected variables, two preliminary analyses were conducted in an attempt to identify significant differences between mobile e-mail users and non-users. This was needed because, if there were no significant differences between the two groups, then profiling made little sense. First, the Pearson chi￾square test was performed for each of the categorical variables across the two groups. The expected values in each cell were greater than 1 and most cells had expected values greater than 5. Significant differences were detected at P < 0.001 for all variables between e￾mail users and non-users. Second, a MANOVA was conducted with type of use or non-use as independent S. Okazaki / Information & Management 43 (2006) 127–141 131

S Okazaki/Information Management 43(2006)127-141 Categorical variables used for the cluster analysis 1)Male:(2)female (1)15-19:(2)20-29:(3)30-39;(4)4049:(5)50-59;(6)6065 (1) Executive; (2)managerial; (3) clerical; (4)administrative staff )self-employed;(6) freelance professional; (7)part-time worker; )housewife:(9) student; (10)unemployed; (11) others Monthly allowance (yen) 1)150.000;(9) unknown Household structure (1) Single:(2)married couple: (3)married couple and children; (4)extended family: (5)others 00JPY≈USs0942≈0.744EUR. variable and all continuous variables as dependent dence of the respondents was ensured by the random variables. Using Wilks'criterion, the continuous sampling plan. Second, to assess the normality of each variables were significantly affected by e-mail use or continuous variable, both skewness and kurtosis tests non-use,F(6,721)=2.45,P<0.05 were determined: in neither test did any of the calculated z-values exceed a critical value +1.96 6.3. Test of assumptions, reliability indicating the normality of the distribution at P<0.05 Finally, because our data were not sequential in nature, Before starting the cluster analysis, the missing the multinomial distribution was assumed for each values were replaced with their means by using SPSs categorical variable. The literature suggests, however, 12.0; this was because the missing values seriously that the two-step clustering algorithm behaves robustly distorted the multivariate analysis results. According even if this assumption was not met. to the recommendations of hair et al., the followin The overall construct validity and reliability of all were examined: (1)the representativeness of the the multiple-item measures (i.e, content credibility, sample and (2)the absence of multicollinearity. First, source credibility, informativeness, entertainment, the sample was considered to be representative, given irritation, general liking, and willingness to access) that our study employed a sufficiently large random were assessed using maximum-likelihood confirmatory sampling procedure, which meant that the results were factor analysis using the covariance-based program, generalizable to the population of interest. Second, the AMOS 5.0. First, convergent validity was supported by level of multicollinearity was exa amines hrough the all the items loadings on respective constructs being tolerance value. The tolerance was found to be within statistically significant. Second, all composite relia- an acceptable range, with all scores between 0.70 and bility estimates exceeded 0.97, while the amount of 0.96, indicating low collinearity among variables. variance extracted by each construct ranged between Next, the specific assumptions of the two-step 0.64 and 0.90. Third, despite the significance of the chi clustering algorithm were assessed. First, the indepen- square value(X=442. 4, P<0.001 ), more pragmatic nuous variables used for the cluster analysis Variance extracted Source credibility Entertainment general liking 3233342 78 0.70 0.99 0.88 Note: All measures were assessed on a 7-point semantic differential scale from 1(strongly disagree) to 7(strongly agree)

variable and all continuous variables as dependent variables. Using Wilks’ criterion, the continuous variables were significantly affected by e-mail use or non-use, F(6, 721) = 2.45, P 150,000; (9) unknown Household structure (1) Single; (2) married couple; (3) married couple and children; (4) extended family; (5) others 100 JPY US$ 0.942 0.744 EUR.

S. Okazaki/Information Management 43(2006)127-141 indices indicate a good fit for the model: goodness-of-fit index(GFD)=0.93, comparative fit index (CFD=0.98 and a root mean square error of approximation (RMSEA)=0.058. Collectively, these tests indicated that the measures of continuous variables were reliable 燃)( and valid reflectors of intended constructs 7. Results The auto-clustering algorithm indicated that a four cluster solution was the best model. because it minimized the bic value and the change in them Cluster between adjacent numbers of clusters(Table 4). In addition, the multidimensional map, as shown in Fig. 2, indicated the clear separation of the four clusters Fig. 2. Multidimensional map of the four-cluster solution. The resulting clusters 1, 2, 3, and 4 contained 147, 135, 152, and 178 cases, which corresponded to 24.0, monthly allowance was relatively high and more than 22.1,24.8,and29.1% half of those who could afford expenses of more than 100,000 yen belonged to this cluster. The proportion of g male respondents was substantially greater than that of female respondents. In terms of household structure, Tables 5 and 6 summarize the frequency distribu- they were primarily unmarried in single households tions for the categorical variables within and across however, a relatively large proportion of others, which clusters, respectively. Cluster 1 consisted mainly of may have included respondents who still live with their younger respondents who were in their 20s or 30s, and parents, constituted this cluster. Similarly, cluster 2 w in three groups of occupations: (1)clerical and characterized by a majority of teenage students living administrative,(2) freelance professionals, and (3) alone or with parents. Also, clerical, administrative, or other or unemployed. In the first two groups, the part-time job segments represented mportan proportion of this cluster. The majority were singles in their 20s, whose monthly allowance was relatively esults of auto-clustering high. In fact, this cluster occupied the second largest Number of BIC proportion of those with a monthly allowance more than change 100.000 All married women were part of cluster 3 and their monthly allowance was relatively lower than that of other groups. Also, more than half of the part-time workers were in this cluster. They were primarily in their 30-50s. In contrast cluster 4 consisted of almost all married men who were corporate executives, managerial, or self-employed. This cluster contained 1. most of those whose monthly allowance was between 1408-0.43 102 10 3980.1 30.000and99,999yen 142.1 -0.43 The changes are from the previous number of clusters in the table The ratios of changes are relative to the change for the two- 7. 2. Attitudinal profiling cluster solution The ratios of distance measures are based on the current number Fig. 3 shows the mean values of seven continuous of clusters against the previous number of clusters. variables for each cluster. Clearly, cluster 2 showed the

indices indicate a good fit for the model: goodness-of-fit index (GFI) = 0.93, comparative fit index (CFI) = 0.98, and a root mean square error of approximation (RMSEA) = 0.058. Collectively, these tests indicated that the measures of continuous variables were reliable and valid reflectors of intended constructs. 7. Results The auto-clustering algorithm indicated that a four￾cluster solution was the best model, because it minimized the BIC value and the change in them between adjacent numbers of clusters (Table 4). In addition, the multidimensional map, as shown in Fig. 2, indicated the clear separation of the four clusters. The resulting clusters 1, 2, 3, and 4 contained 147, 135, 152, and 178 cases, which corresponded to 24.0, 22.1, 24.8, and 29.1%, respectively. 7.1. Demographic profiling Tables 5 and 6 summarize the frequency distribu￾tions for the categorical variables within and across clusters, respectively. Cluster 1 consisted mainly of younger respondents who were in their 20s or 30s, and in three groups of occupations: (1) clerical and administrative, (2) freelance professionals, and (3) other or unemployed. In the first two groups, the monthly allowance was relatively high and more than half of those who could afford expenses of more than 100,000 yen belonged to this cluster. The proportion of male respondents was substantially greater than that of female respondents. In terms of household structure, they were primarily unmarried in single households; however, a relatively large proportion of others, which may have included respondents who still live with their parents, constituted this cluster. Similarly, cluster 2 was characterized by a majority of teenage students living alone or with parents. Also, clerical, administrative, or part-time job segments represented an important proportion of this cluster. The majority were singles in their 20s, whose monthly allowance was relatively high. In fact, this cluster occupied the second largest proportion of thosewith a monthly allowance more than 100,000 yen. All married women were part of cluster 3 and their monthly allowance was relatively lower than that of other groups. Also, more than half of the part-time workers were in this cluster. They were primarily in their 30–50s. In contrast, cluster 4 consisted of almost all married men who were corporate executives, managerial, or self-employed. This cluster contained most of those whose monthly allowance was between 30,000 and 99,999 yen. 7.2. Attitudinal profiling Fig. 3 shows the mean values of seven continuous variables for each cluster. Clearly, cluster 2 showed the S. Okazaki / Information & Management 43 (2006) 127–141 133 Table 4 Results of auto-clustering Number of clusters BIC BIC changea Ratio of BIC changesb Ratio of distance measuresc 1 3846.2 2 3517.9 328.3 1.00 1.49 3 3372.7 145.2 0.44 1.33 4 3319.1 53.6 0.16 1.83 5 3392.4 73.3 0.22 1.01 6 3466.6 74.1 0.23 1.23 7 3569.1 102.5 0.31 1.26 8 3697.1 128.1 0.39 1.15 9 3837.9 140.8 0.43 1.02 10 3980.1 142.1 0.43 1.29 a The changes are from the previous number of clusters in the table. b The ratios of changes are relative to the change for the two￾cluster solution. c The ratios of distance measures are based on the current number of clusters against the previous number of clusters. Fig. 2. Multidimensional map of the four-cluster solution.

S Okazaki/Information Management 43(2006)127-141 Table 5 Composition of demographic profiles within clusters (n=612) istics Cluster 1 Cluster 2 Cluster 3 (n=147) =135) (n=152) (n=178) 55.1 0 50.4 96.6 .9 5.4 178 0.6 14.1 40-49 178 60-65 3.4 0.0 2.7 15 95.5 92 67.3 98.5 Occupational category Executive 0.0 14.9 193 13.5 Administrative staff 14.2 19.0 .2 Part-time worker 11.6 Housewife Student 10.9 31.9 .6 5.6 Monthly allowance(yen) 9.6 28.3 3.9 0,000-19,999 123 20,000-29,999 167 21. 30,000-49999 27.1 13.2 33.7 50000-69,999 14.2 17.8 5.3 15.7 70.000-99,999 69 3.0 0.7 11.8 8.2 0.7 0.0 7.0 structure .9 12.2 13.3 Married couple 12.2 Married with children 54.8 71.7 Extended family 127 16 Others Note: The numbers indicate percentages that vertically sum to 100% 100JPY≈USs0942≈0.744EUR most positive attitudes toward credibility, informa- intentions. It should be noted that, while a two-step tiveness, and entertainment. In contrast, cluster 1 cluster analysis clearly separated four groups, the showed the most negative attitudes toward the mobile seven continuous variables seemed to be significantly platform. Thus, within the same youth groups, there correlated. Therefore, it seemed reasonable to infer was a clear separation in attitudinal and behavioral that the more positive the respondents perceptions of

most positive attitudes toward credibility, informa￾tiveness, and entertainment. In contrast, cluster 1 showed the most negative attitudes toward the mobile platform. Thus, within the same youth groups, there was a clear separation in attitudinal and behavioral intentions. It should be noted that, while a two-step cluster analysis clearly separated four groups, the seven continuous variables seemed to be significantly correlated. Therefore, it seemed reasonable to infer that the more positive the respondents’ perceptions of 134 S. Okazaki / Information & Management 43 (2006) 127–141 Table 5 Composition of demographic profiles within clusters (n = 612) Characteristics Total Cluster 1 (n = 147) Cluster 2 (n = 135) Cluster 3 (n = 152) Cluster 4 (n = 178) Gender Male 55.1 66.0 50.4 0.0 96.6 Female 44.9 34.0 49.6 100.0 3.4 Age (years) <20 5.4 5.4 17.8 0.0 0.6 20–29 27.9 49.0 61.5 9.2 1.1 30–39 27.0 29.3 14.1 35.5 27.5 40–49 17.8 9.5 2.2 28.3 27.5 50–59 16.8 3.4 4.4 21.1 33.7 60–65 5.1 3.4 0.0 5.9 9.6 Marital status Married 60.8 32.7 1.5 100.0 95.5 Single 39.2 67.3 98.5 0.0 4.5 Occupational category Executive 3.8 2.7 3.0 2.0 6.7 Managerial 7.5 3.4 0.0 0.0 23.0 Clerical 14.9 24.5 19.3 3.3 13.5 Administrative staff 14.2 19.0 15.6 0.7 20.8 Self-employed 8.2 6.1 1.5 2.0 20.2 Freelance professional 3.6 7.5 0.7 0.0 5.6 Part-time worker 15.8 11.6 19.3 34.2 1.1 Housewife 13.9 0.0 0.0 55.9 0.0 Student 9.8 10.9 31.9 0.0 0.6 Unemployed 2.1 2.7 3.0 0.0 2.8 Others 6.2 11.6 5.9 2.0 5.6 Monthly allowance (yen)a <10,000 10.8 2.0 9.6 28.3 3.9 10,000–19,999 12.3 6.1 12.6 23.0 7.9 20,000–29,999 16.7 12.9 15.6 21.1 16.9 30,000–49,999 27.1 30.6 30.4 13.2 33.7 50,000–69,999 14.2 18.4 17.8 5.3 15.7 70,000–99,999 6.9 10.9 3.0 0.7 11.8 100,000–149,999 3.4 8.2 5.2 0.0 1.1 <150,000 1.6 4.1 2.2 0.7 0.0 Unknown 7.0 6.8 3.7 7.9 9.0 Household structure Single 6.9 12.2 13.3 1.3 2.2 Married couple 10.3 12.2 1.5 13.2 12.9 Married with children 59.3 46.3 54.8 71.7 62.9 Extended family 12.7 15.6 8.1 9.9 16.3 Others 10.8 13.6 22.2 3.9 5.6 Note: The numbers indicate percentages that vertically sum to 100%. a 100 JPY US$ 0.942 0.744 EUR.

S. Okazaki/Information Management 43(2006)127-141 Table 6 Composition of demographic profiles across clusters(n=612) Characteristics Cluster I Cluster 2 Cluster 3 Cluster 4 (n=135) (n=152) (n=178) 22.1 x2=64,d.=3 P=0.09 Gender x2=318.8,df.=3 Male 28.8 1.0 24.2 72.7 0.0 3.0 8.2 115 40-419 39.4 6065 Marital status 436.1.df.=3 12 40.9 45.7 Occupational category x2=634.0.df.=30 Executive 174 Administrative staff elf-employed Freelance professional 45.5 Part-time worker Housewife 100.0 Student 38.5 Others x2=1660.df.=24 150,000 11.6 Household structure 83.3.df.=12 429 31.7 with childre 30.0 family Others Note: The numbers indicate percentages that horizontally sum to 100%0 a100JPY≈USs0.942≈0.744EUR Significant at P< 0.C0I level using Pearson chi-square. No cells have expected count less than 5, with an exception of"monthly allowance in which 5 cells(13.9%)have expected count less than 5

S. Okazaki / Information & Management 43 (2006) 127–141 135 Table 6 Composition of demographic profiles across clusters (n = 612) Characteristics Cluster 1 (n = 147) Cluster 2 (n = 135) Cluster 3 (n = 152) Cluster 4 (n = 178) x2 Total 24.0 22.1 24.8 29.1 x2 = 6.4, d.f. = 3 P = 0.09 Gender x2 = 318.8, d.f. = 3 *** Male 28.8 20.2 0.0 51.0 Female 18.2 24.4 55.3 2.2 Age (years) x2 = 327.2, d.f. = 15 *** 150,000 60.0 30.0 10.0 0.0 Unknown 23.3 11.6 27.9 37.2 Household structure x2 = 83.3, d.f. = 12 *** Single 42.9 42.9 4.8 9.5 Married couple 28.6 3.2 31.7 36.5 Married with children 18.7 20.4 30.0 30.9 Extended family 29.5 14.1 19.2 37.2 Others 30.3 45.5 9.1 15.2 Note: The numbers indicate percentages that horizontally sum to 100%. a 100 JPY US$ 0.942 0.744 EUR. *** Significant at P < 0.001 level using Pearson chi-square. No cells have expected count less than 5, with an exception of ‘‘monthly allowance’’, in which 5 cells (13.9%) have expected count less than 5.

S Okazaki/Information Management 43(2006)127-141 00 3.00 pility Informativeness Entertainment Irritation eneral liking Willingness to 母 Chuster2403 一 Cluster3 Fig 3. Mean values of continuous variables. Note: Means of each continuous variable are significantly different across four clusters at P< 0.001 level using univariate F-test. informativeness, entertainment, and credibility of 7.3. validation of the cluster solution mobile Internet service, the more likely they were to exhibit a positive attitude and intention to acce Although the number of clusters was objectively Undoubtedly, perceived irritation was inversely determined on the basis of the BIC, it was essential to correlated with both general liking and willingness assess the stability of the cluster solution and to to access determine whether the cluster members were indeed erestingly, clusters 3 and 4 showed a similar ten- homogeneous within clusters, while being hetero- dency in terms of attitudinal or perceptual dimensions geneous between them. To this end, two types of gesting that they had relatively positive opinion of dependent multivariate analyses were employed for the value of the mobile Internet service. Nevertheless the validation of the four-cluster solution married women (cluster 3) indicated more negative First, a multinomial logistic regression was per perceptions than married men, and this was consistent formed with eight categorical variables that were not with research on gender difference in wired Internet used in the cluster analysis: average hours spent out of office/home per day, average sleeping hours, time to Table Likelihood-ratio tests in multinomial logistic regression 2 log-likelihood Average hours spent outside Average sleeping hours Time to sleep 904.5 verage hours spent watching TV average hours spent listening to radio erage hours spent reading newspaper 911.1 21198 Monthly household expenses 919.0 Significant at P< 0.05 level. Significant at P< 0. 001 level

informativeness, entertainment, and credibility of mobile Internet service, the more likely they were to exhibit a positive attitude and intention to access. Undoubtedly, perceived irritation was inversely correlated with both general liking and willingness to access. Interestingly, clusters 3 and 4 showed a similar ten￾dency in terms of attitudinal or perceptual dimensions, suggesting that they had relatively positive opinion of the value of the mobile Internet service. Nevertheless, married women (cluster 3) indicated more negative perceptions than married men, and this was consistent with research on gender difference in wired Internet adoption. 7.3. Validation of the cluster solution Although the number of clusters was objectively determined on the basis of the BIC, it was essential to assess the stability of the cluster solution and to determine whether the cluster members were indeed homogeneous within clusters, while being hetero￾geneous between them. To this end, two types of dependent multivariate analyses were employed for the validation of the four-cluster solution. First, a multinomial logistic regression was per￾formed with eight categorical variables that were not used in the cluster analysis: average hours spent out of office/home per day, average sleeping hours, time to 136 S. Okazaki / Information & Management 43 (2006) 127–141 Fig. 3. Mean values of continuous variables. Note: Means of each continuous variable are significantly different across four clusters at P < 0.001 level using univariate F-test. Table 7 Likelihood-ratio tests in multinomial logistic regression Categorical independent variables 2 log-likelihood x2 d.f. P Average hours spent outside 1028.7 150.0 18 *** Average sleeping hours 898.0 19.3 15 0.20 Time to wake up 931.8 53.1 18 *** Time to sleep 904.5 25.8 12 ** Average hours spent watching TV 912.8 34.1 21 ** Average hours spent listening to radio 919.4 40.6 21 ** Average hours spent reading newspaper 911.1 32.4 9 *** Monthly household expenses 919.0 40.3 18 ** ** Significant at P < 0.05 level. *** Significant at P < 0.001 level.