Availableonlineatwww.sciencedirect.co SCIENCE DIRECT INFORMATION MANAGEMENT ELSEVIER Information Management 43(2006)204-221 ww.elsevier. com/locate/dsw Migrating to internet-based e-commerce: Factors affecting e-commerce adoption and migration at the firm level Weiyin Hong a*. Kevin znu 6 nt information Systems, College of Business, University of Nevada, 4505 Maryland Parkway, P.O. Box 456034. Las Vegas, NV 89154, USA The Paul Merage School of Business, University of California, Irvine, CA, USA Received in revised form 18 March 2005; accepted 25 June 2005 Available online 26 August 2005 Abstract Web technology has enabled e-commerce. However, in our review of the literature, we found little research on how firms can better position themselves when adopting e-commerce for revenue generation. Drawing upon technology diffusion theory, we developed a conceptual model for assessing e-commerce adoption and migration, incorporating six factors unique to e-commerce. A series of propositions were then developed Survey data of 1036 firms in a broad range of industries were collected and used to test our model. Our analysis based on multi-nominal logistic regression demonstrated that technology integration, web functionalities, web spending, and partner usage were significant adoption predictors. The model showed that these variables could successfully differentiate non-adopters from adopters. Further, the migration model demonstrated that web functionalities, web spending, and integration of externally riented inter-organizational systems tend to be the most influential drivers in firms' migration toward e-commerce, while firm size, partner usage, electronic data interchange (EDI usage, and perceived obstacles were found to negatively affect commerce migration. This suggests that large firms, as well as those that have been relying on outsourcing or EDl, tended to be slow to migrate to the internet platform. C 2005 Elsevier B V. All rights reserved Keywords: Innovation adoption: Migration; Technology diffusion; E-commerce: Intemet technology; EDI; Outsourcing 1. Introduction field of information systems (IS). The potential of the In recent years, electronic-commerce (EC) has firms, such as Dell, Cisco, Wal-Mart, and Charles emerged as one of the most active research areas in the Schwab, have achieved tangible improvements in operational efficiency and revenue generation by integrating e-commerce into their value chain activ fax:+17028950802 ities [7]. not all firms have been uniformly successful 14, 11, 42]. Indeed, firms face a series of obstacles in 378-7206/$- see front matter c 2005 Elsevier B v. All rights reserved doi:l0.1016im.2005.06003
Migrating to internet-based e-commerce: Factors affecting e-commerce adoption and migration at the firm level Weiyin Hong a, *, Kevin Zhu b a Department of Management Information Systems, College of Business, University of Nevada, 4505 Maryland Parkway, P.O. Box 456034, Las Vegas, NV 89154, USA b The Paul Merage School of Business, University of California, Irvine, CA, USA Received in revised form 18 March 2005; accepted 25 June 2005 Available online 26 August 2005 Abstract Web technology has enabled e-commerce. However, in our review of the literature, we found little research on how firms can better position themselves when adopting e-commerce for revenue generation. Drawing upon technology diffusion theory, we developed a conceptual model for assessing e-commerce adoption and migration, incorporating six factors unique to e-commerce. A series of propositions were then developed. Survey data of 1036 firms in a broad range of industries were collected and used to test our model. Our analysis based on multi-nominal logistic regression demonstrated that technology integration, web functionalities, web spending, and partner usage were significant adoption predictors. The model showed that these variables could successfully differentiate non-adopters from adopters. Further, the migration model demonstrated that web functionalities, web spending, and integration of externally oriented inter-organizational systems tend to be the most influential drivers in firms’ migration toward e-commerce, while firm size, partner usage, electronic data interchange (EDI) usage, and perceived obstacles were found to negatively affect ecommerce migration. This suggests that large firms, as well as those that have been relying on outsourcing or EDI, tended to be slow to migrate to the internet platform. # 2005 Elsevier B.V. All rights reserved. Keywords: Innovation adoption; Migration; Technology diffusion; E-commerce; Internet technology; EDI; Outsourcing 1. Introduction In recent years, electronic-commerce (EC) has emerged as one of the most active research areas in the field of information systems (IS). The potential of the internet is now widely acknowledged. While some firms, such as Dell, Cisco, Wal-Mart, and Charles Schwab, have achieved tangible improvements in operational efficiency and revenue generation by integrating e-commerce into their value chain activities [7], not all firms have been uniformly successful [4,11,42]. Indeed, firms face a series of obstacles in www.elsevier.com/locate/dsw Information & Management 43 (2006) 204–221 * Corresponding author. Tel.: +1 702 895 2778; fax: +1 702 895 0802. E-mail address: whong@unlv.nevada.edu (W. Hong). 0378-7206/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.im.2005.06.003
w.Hong, K. Zhu/Information Management 43(2006)204-221 adopting e-commerce[23]. Researchers and managers 2. Theoretical development are struggling to determine the right conditions for adopting e-commerce, and what factors facilitate or 2.1. E-commerce adoption literature nhibit them in migrating to the internet from traditional physical channels[40] Table I provides a summary of ou While research has been advancing, literature on material on e-commerce adoption. The studies vary e-commerce adoption typically focuses on adoption in terms of the nature of the technology, research of either a specific technology, such as email and methodology, and measures of e-commerce adoption. web presence [5, 28, 39], or a specific e-commerce First, the e-commerce technologies investigate application [13]. In this study, we adopted an relatively simple, such as establishing corporate alternative view, and defined e-commerce as any websites or email systems. As many companies with application of web technologies that enable revenue- web presence do not actually conduct transactions generating business activities over the internet. This online, the adoption of these simple technologies differentiates our study in several ways: (1)we made are less likely to bring fundamental change to the a distinction between setting up a website and organization. Furthermore, the adoption of simple conducting e-commerce; (2) by focusing on revenue- internet technologies is relatively inexpensive and generating activities, we were able to observe the easy, which makes the adoption decision less xtent to which firms migrated from traditional controversial; for advanced e-commerce technologies, channels to the internet. We were interested in especially those involving online transactions and looking at the extent to which a firm migrated from integrated with internal business processes, the the traditional channel to the internet platform, adoption process is complicated and costly indicated by the percentage of revenue generated Second, there is little empirical data to characterize from the internet over the total revenue. A significant e-commerce or gauge the scale of its impact on firm portion of revenue generated through the internet is performance. This is especially true in"brick-and an indication of the organizations ability to leverage mortar" companies, because of the difficulty of the internet [48 developing measures and collecting data [47]. For We studied two related ways in which firms studies that collected empirical data from multiple move to the internet: e-commerce adoption and companies, the sample size was relatively small(from e-commerce migration. The former refers to whether 62 to 286) with a focus on a narrow industry sector or a the firm has started to use the internet for revenue- specific e-commerce application. These studies moved generating activities. The latter involves the extent the frontiers of knowledge forward, but the findings of revenues generated from the internet versus have less generalizability those from the traditional channel. These two Third, studies have used a variety of dependent measures complement each other: while adoption variables. One is an adoption measure, which typically provides a qualitative description of an organiza- uses the physical acquisition or purchase of the tions'behavior, migration captures the quantitative innovation. As noted by Fichman [17]. this is a property relatively"thin"measure as there can be significant We were particularly interested in examining the delay between the purchase and full implementation importance of technology-related factors and obst Richer information can be found in the assimilation cles associated with applying the technology. These literature, which deals with the extent to which the use factors are important in understanding e-commerce of a technology diffuses across organizational work adoption and migration at the firm level but have processes and becomes routinized in the activities seldom been studied [49]. By building upon earlier associated with those processes [18]. The literature theoretical models of technology adoption [34, 38] normally considers the adoption as a longitudinal especially the technology-organization-environment process that can be divided into a number of stages, (TOE) framework [43], we developed and tested a from awareness of the innovation to its full deployment model with an emphasis on firm level factors unique throughout the organization [14, 29]. Alternatively, to the internet organizational assimilation of e-commerce has been
adopting e-commerce [23]. Researchers and managers are struggling to determine the right conditions for adopting e-commerce, and what factors facilitate or inhibit them in migrating to the internet from traditional physical channels [40]. While research has been advancing, literature on e-commerce adoption typically focuses on adoption of either a specific technology, such as email and web presence [5,28,39], or a specific e-commerce application [13]. In this study, we adopted an alternative view, and defined e-commerce as any application of web technologies that enable revenuegenerating business activities over the internet. This differentiates our study in several ways: (1) we made a distinction between setting up a website and conducting e-commerce; (2) by focusing on revenuegenerating activities, we were able to observe the extent to which firms migrated from traditional channels to the internet. We were interested in looking at the extent to which a firm migrated from the traditional channel to the internet platform, indicated by the percentage of revenue generated from the internet over the total revenue. A significant portion of revenue generated through the internet is an indication of the organization’s ability to leverage the internet [48]. We studied two related ways in which firms move to the internet: e-commerce adoption and e-commerce migration. The former refers to whether the firm has started to use the internet for revenuegenerating activities. The latter involves the extent of revenues generated from the internet versus those from the traditional channel. These two measures complement each other: while adoption provides a qualitative description of an organizations’ behavior, migration captures the quantitative property. We were particularly interested in examining the importance of technology-related factors and obstacles associated with applying the technology. These factors are important in understanding e-commerce adoption and migration at the firm level but have seldom been studied [49]. By building upon earlier theoretical models of technology adoption [34,38], especially the technology–organization–environment (TOE) framework [43], we developed and tested a model with an emphasis on firm level factors unique to the internet. 2. Theoretical development 2.1. E-commerce adoption literature Table 1 provides a summary of our review of material on e-commerce adoption. The studies vary in terms of the nature of the technology, research methodology, and measures of e-commerce adoption. First, the e-commerce technologies investigated are relatively simple, such as establishing corporate websites or email systems. As many companies with web presence do not actually conduct transactions online, the adoption of these simple technologies are less likely to bring fundamental change to the organization. Furthermore, the adoption of simple internet technologies is relatively inexpensive and easy, which makes the adoption decision less controversial; for advanced e-commerce technologies, especially those involving online transactions and integrated with internal business processes, the adoption process is complicated and costly. Second, there is little empirical data to characterize e-commerce or gauge the scale of its impact on firm performance. This is especially true in ‘‘brick-andmortar’’ companies, because of the difficulty of developing measures and collecting data [47]. For studies that collected empirical data from multiple companies, the sample size was relatively small (from 62 to 286) with a focus on a narrow industry sector or a specific e-commerce application. These studies moved the frontiers of knowledge forward, but the findings have less generalizability. Third, studies have used a variety of dependent variables. One is an adoption measure, which typically uses the physical acquisition or purchase of the innovation. As noted by Fichman [17], this is a relatively ‘‘thin’’ measure as there can be significant delay between the purchase and full implementation. Richer information can be found in the assimilation literature, which deals with the extent to which the use of a technology diffuses across organizational work processes and becomes routinized in the activities associated with those processes [18]. The literature normally considers the adoption as a longitudinal process that can be divided into a number of stages, from awareness of the innovation to its full deployment throughout the organization [14,29]. Alternatively, organizational assimilation of e-commerce has been W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 205
206 w. Hong, K. Zhu/ Information Management 43(2006) 204-22 Literature review on e-commerce adoption Study Methodology Factors/major findings Beatty et al.(2001)(&M) Innovation diffusion Survey (N=286) DV: Entry timing(pioneer, early adopter. early majority, late majority, laggard) IT adoption Various industries Medium-to-large technical compatibility, organization ompatibility, top management support Chatterjee et al. [11](MIsQ) tional theory Survey(N= 62) DV: EC activities and strategies Manufacturing IV: Championship, strategic investment service firms rationale. extent of coordination Chiru and Technology Case study: Online Framework: Value flows- potential Kauffman [13(MIS diffusion theory vel reservation value→ realized value Limits.to-value model systems Valuation barriers: Industry barriers knowledge barriers, usage barriers Zhu et al. [49](EIS) TOE framework Survey (N= 3100) DV: Intent to adopt e-business Technology competence, firm scope, siz onsumer readiness, partner readiness, and competitive pressure Kowtha and Resource- based view Survey(N= 135) DV: Website development-four generation hoon [24 (information and catalog, database. transaction, integrated site) IT adoption Travel. financial IV: Prior competencies, firm size, firm age, d It sectors competitive intensity, strategic commitment to e-commerce Mehrtens et al. [28(&M) Innovation literature Case studies DV: Decision to adopt-dichotomy (Y/N) SMEs IV: Perceived benefits, organizationa Teo et al. 39(EC Co Survey (N= 188) theory, TOE (adopters with website, adopters Various industries IV: Technological factors, organizational Small and large firms factors, environmental factors TOE framework Case study Innovation-specific characteristics Ramamurthy [44(EC) (the social and technological context) Zhu and Kraemer [47](ISR) It business valu DV: Firm performance measures Resource-based view IV: EC capability(infor transaction, interaction, supplier integration): IT infrastructure ote: DV, dependent variables; IV, independent variables: EDI, electronic data interchange. Studies focusing on consumer acceptance of internet ommerce, website design, and price comparison are not included. measured by the degree of different activities being impact of the innovation In our study, we chose both an mplemented on the corporate website. While these adoption measure(adopter, potential-adopter, and non- measures depict the extent to which the innovation has adopter) and a direct impact measure(the portion of been used, they do not provide information on the revenue generated by e-commerce). The two measures
measured by the degree of different activities being implemented on the corporate website. While these measures depict the extent to which the innovation has been used, they do not provide information on the impact of the innovation. In our study, we chose both an adoption measure (adopter, potential-adopter, and nonadopter) and a direct impact measure (the portion of revenue generated by e-commerce). The two measures 206 W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 Table 1 Literature review on e-commerce adoption Study Theory Methodology Factors/major findings Beatty et al. (2001) (I&M) Innovation diffusion Survey (N = 286) DV: Entry timing (pioneer, early adopter, early majority, late majority, laggard) IT adoption Various industries IV: Perceived benefits, complexity, technical compatibility, organizational compatibility, top management support Medium-to-large U.S. firms Chatterjee et al. [11] (MISQ) Institutional theory Survey (N = 62) DV: EC activities and strategies Structuration theory Manufacturing & service firms IV: Championship, strategic investment rationale, extent of coordination Chircu and Kauffman [13] (JMIS) Technology diffusion theory Case study: Online travel reservation systems Framework: Value flows ! potential value ! realized value Limits-to-value model Valuation barriers: Industry barriers, organizational barriers Conversion barriers: Resource barriers, knowledge barriers, usage barriers Zhu et al. [49] (EJIS) TOE framework Survey (N = 3100) DV: Intent to adopt e-business IV: Technology competence, firm scope, size, consumer readiness, partner readiness, and competitive pressure Kowtha and Choon [24] (I&M) Resource-based view Survey (N = 135) DV: Website development—four generation (information and catalog, database, transaction, integrated site) IT adoption Travel, financial, and IT sectors IV: Prior competencies, firm size, firm age, competitive intensity, strategic commitment to e-commerce Mehrtens et al. [28] (I&M) Innovation literature Case studies DV: Decision to adopt—dichotomy (Y/N) SMEs IV: Perceived benefits, organizational readiness, external pressure Teo et al. [39] (IJEC) Contingency theory, TOE Survey (N = 188) DV: Decision to adopt—trichotomy (adopters with website, adopters without website, non-adopters) Various industries IV: Technological factors, organizational Small and large firms factors, environmental factors Vadapalli and Ramamurthy [44] (IJEC) TOE framework Case study Innovation-specific characteristics (the social and technological context) Organization-specific characteristics (organization boundaries, transaction cost economics, and organizational cognition) Zhu and Kraemer [47] (ISR) IT business value Survey (N = 260) DV: Firm performance measures Resource-based view Manufacturing firms IV: EC capability (information, transaction, interaction, supplier Dynamic capability integration); IT infrastructure measures Note: DV, dependent variables; IV, independent variables; EDI, electronic data interchange. Studies focusing on consumer acceptance of internet commerce, website design, and price comparison are not included
w.Hong, K. Zhu/Information Management 43(2006)204-221 helped us gain a more balanced understanding of potentially affected and the innovation may have e-commerce adoption and migration strategic relevance to the firm 2.2. The technology diffusion framework We consider e-commerce to be a Type In innovation because it is often embedded in the firms Tornatzky and Fleischer [43] proposed the core business processes or is extending basic business technology-organization-environment framework to products and services, and integrating suppliers and doption of technological innovations; it ustomers in the value chain identified three aspects of a firms contexts that influenced adoption and implementation. (1) Techno- logical context-the existing and emerging technol 3. Conceptual model and theoretical ogies relevant to the firm; (2)organizational context- propositions in terms of several descriptive measures: firm size and scope, managerial structure, and internal resources; 3. 1. An integrated model of e-commerce adoption ( )environmental context-the macro arena in which and migration a firm conducts its business: industry, competitors, and dealings with government. We developed an integrated model to address the The TOE framework has been utilized for studying adoption and migration of e-commerce. As shown in different types of innovations [12, 22, 26, 41]. Accord- Fig. 1, this posited six predictors for e-commerce ing to the typology of Swanson, innovations can be adoption( technology integration, web spending, web classified into three categorie functionalities, electronic data interchange(EDi)use, outsourcing partner usage, and perceived obstacles) Type 1: technical innovations restricted to the Is while controlling for firm size and industry types functional tasks. These factors were chosen because they were believed Type Il: applying IS products and services toto be important in understanding and explaining the support administrative tasks of the business phenomenon of interest. Type I: integrating Is products and services with ide range of factors was found in the literature the core business where the whole business is Instead of repeating them, we chose to focus on a few Pla Technology Integration Web Spending Web Functionalities ed EC Migration EDI Use Partner Usage Perceived Obstacles Fig. 1. Conceptual framework for e-commerce(EC)adoption and migration
helped us gain a more balanced understanding of e-commerce adoption and migration. 2.2. The technology diffusion framework Tornatzky and Fleischer [43] proposed the technology–organization–environment framework to study the adoption of technological innovations; it identified three aspects of a firm’s contexts that influenced adoption and implementation. (1) Technological context—the existing and emerging technologies relevant to the firm; (2) organizational context— in terms of several descriptive measures: firm size and scope, managerial structure, and internal resources; (3) environmental context—the macro arena in which a firm conducts its business: industry, competitors, and dealings with government. The TOE framework has been utilized for studying different types of innovations [12,22,26,41]. According to the typology of Swanson, innovations can be classified into three categories: Type I: technical innovations restricted to the IS functional tasks; Type II: applying IS products and services to support administrative tasks of the business; Type III: integrating IS products and services with the core business where the whole business is potentially affected and the innovation may have strategic relevance to the firm. We consider e-commerce to be a Type III innovation, because it is often embedded in the firm’s core business processes or is extending basic business products and services, and integrating suppliers and customers in the value chain. 3. Conceptual model and theoretical propositions 3.1. An integrated model of e-commerce adoption and migration We developed an integrated model to address the adoption and migration of e-commerce. As shown in Fig. 1, this posited six predictors for e-commerce adoption (technology integration, web spending, web functionalities, electronic data interchange (EDI) use, outsourcing partner usage, and perceived obstacles), while controlling for firm size and industry types. These factors were chosen because they were believed to be important in understanding and explaining the phenomenon of interest. A wide range of factors was found in the literature. Instead of repeating them, we chose to focus on a few W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 207 Fig. 1. Conceptual framework for e-commerce (EC) adoption and migration
w Hong, K. Zhu/ Information Management 43(2006) 204-22 factors that are part adoption and are represented on the web platform. This extends the migration. While other factors could have been technology notion from a purely technical measure to selected, the factors in our model emphasized the fit (1)the relationship between the new technology and the between the new capability and legacy systems, the existing base and (2) the use of the technology by the web-related technological capability that firms pos- organization. As a type Ill technology, e-commerce sess, the is budget devoted to web-related spending, requires close coordination of various components the firms prior technology base, the strategy that firms along the value chain. Similarly, the more integrated take to develop their IT applications, and the perceived these existing applications are with the internet obstacles associated with applying the technology. platform, the more capacity the organization has to These factors are more related to the technological conduct its business over the internet. As a result those than the organizational and environmental contexts. firms would enjoy greater e-commerce migration. This First, organizational factors have already been studied leads to the following propositions: 32, 33] and environmental factors have been exam ed [25]. Essential organizational characteristics, Pla. Technology integration is positively associated such as firm size and industry type, were included in with e-commerce adoption the model as control variables Second, the study of Is adoption requir PIb. Technology integration is positively associated consideration of the specific technology and its use with e-commerce migration. [6], partly because innovation diffusion theories did not provide specific innovation attributes for Is 3. 2.2. Web spending ption in firms. Fichman argued that classical Financial resources are an important factor for diffusion variables are unlikely to be strong predictors technology adoption. We defined web spending as the of adoption and diffusion for complex Type portion of financial resources devoted to web-based Initiatives innovations, suggesting that factors more specific to ding hardware, software, IT services the technology should be added. consulting, and employee training. Firms with Third, the literature on e-commerce adoption higher web budget are better positioned to adopt e-commerce. This may also be an indication of suggested that a large variety of variables have been the importance that top management places on studied. However, there is little consistency in terms of the inclusion of variables in these studies because (1) e-commerce. Hence, firms with greater web spending there was a large span of variables that could be are more likely to adopt e-commerce as well as migrate offine transactions to the online platform. studied;(2)the nature of the technologies differed;(3) This leads to the following propositions the organizational contexts varied. Therefore, we did not attempt to conduct a comprehensive study but P2a. Web spending is positively associated with e- focused on technology-related variables that apply to commerce adoptio e-commerce technologies P2b. Web spending is positively associated with 3.2. Theoretical propositions commerce migration. 3.2.1. Technology integration 3. 2.3. Web functionalities Prior to the internet, firms had been usin Web technologies offer a variety of functionalities technologies to support business activities along their ranging from static presentation of content to dynamic value chain, but many were"islands of automation capture of transactions with provisions for security and they lacked integration across applications[45]. The personalization [10]. Firms must make use of these characteristics of the intemet may help remove the technologies and decide how to draw upon their incompatibilities and rigidities of legacy IS and achieve capabilities for e-commerce. Web functionalities help technology integration among various applications and firms provide real-time information to customers, databases. We define technology integration as the update product offerings and make price change, extent to which various technologies and applications facilitate self-service via online account management
factors that are particularly relevant to adoption and migration. While other factors could have been selected, the factors in our model emphasized the fit between the new capability and legacy systems, the web-related technological capability that firms possess, the IS budget devoted to web-related spending, the firm’s prior technology base, the strategy that firms take to develop their IT applications, and the perceived obstacles associated with applying the technology. These factors are more related to the technological than the organizational and environmental contexts. First, organizational factors have already been studied [32,33] and environmental factors have been examined [25]. Essential organizational characteristics, such as firm size and industry type, were included in the model as control variables. Second, the study of IS adoption requires consideration of the specific technology and its use [6], partly because innovation diffusion theories did not provide specific innovation attributes for IS adoption in firms. Fichman argued that classical diffusion variables are unlikely to be strong predictors of adoption and diffusion for complex Type III innovations, suggesting that factors more specific to the technology should be added. Third, the literature on e-commerce adoption suggested that a large variety of variables have been studied. However, there is little consistency in terms of the inclusion of variables in these studies because (1) there was a large span of variables that could be studied; (2) the nature of the technologies differed; (3) the organizational contexts varied. Therefore, we did not attempt to conduct a comprehensive study but focused on technology-related variables that apply to e-commerce technologies. 3.2. Theoretical propositions 3.2.1. Technology integration Prior to the internet, firms had been using technologies to support business activities along their value chain, but many were ‘‘islands of automation’’— they lacked integration across applications [45]. The characteristics of the internet may help remove the incompatibilities and rigidities of legacy IS and achieve technology integration among various applications and databases. We define technology integration as the extent to which various technologies and applications are represented on the web platform. This extends the technology notion from a purely technical measure to (1) the relationship between the new technology and the existing base and (2) the use of the technology by the organization. As a type III technology, e-commerce requires close coordination of various components along the value chain. Similarly, the more integrated these existing applications are with the internet platform, the more capacity the organization has to conduct its business over the internet. As a result, those firms would enjoy greater e-commerce migration. This leads to the following propositions: P1a. Technology integration is positively associated with e-commerce adoption. P1b. Technology integration is positively associated with e-commerce migration. 3.2.2. Web spending Financial resources are an important factor for technology adoption. We defined web spending as the portion of financial resources devoted to web-based initiatives, including hardware, software, IT services, consulting, and employee training. Firms with higher web budget are better positioned to adopt e-commerce. This may also be an indication of the importance that top management places on e-commerce. Hence, firms with greater web spending are more likely to adopt e-commerce as well as migrate offline transactions to the online platform. This leads to the following propositions: P2a. Web spending is positively associated with ecommerce adoption. P2b. Web spending is positively associated with ecommerce migration. 3.2.3. Web functionalities Web technologies offer a variety of functionalities ranging from static presentation of content to dynamic capture of transactions with provisions for security and personalization [10]. Firms must make use of these technologies and decide how to draw upon their capabilities for e-commerce. Web functionalities help firms provide real-time information to customers, update product offerings and make price change, facilitate self-service via online account management 208 W. Hong, K. Zhu / Information & Management 43 (2006) 204–221
w.Hong, K. Zhu/Information Management 43(2006)204-221 and research tools, and conduct online transactions with P4a. The use of EDI is negatively associated with e- suppliers [46]. Therefore, firms that are capable of commerce adoption providing more web functionalities are better posi and capable web functio P4b. The use of EDI is negatively associated with e- alities will make customers as well as trading partners commerce migration more willing to conduct transactions online with greater migration to e-commerce. This suggests the proposi 3.2.5. Partner usage Many firms have relied on partners or contractors P3a. Web functionalities are positively associated for their is design and implementation tasks. An with e-commerce adoption outsourcing approach has been popular in driving the growth of applications service providers. Relying on P3b. Web functionalities are positively associated partners for e-commerce lementation may speed with e-commerce migration up the initial adoption of e-commerce, bypassing the potentially slow process associated with in-house development [3, 27]. However, this may slow down an 3.24. EDI use organization's subsequent migration to e-commerce. Electronic data interchange was an antecedent of e- Outsourcing may seem to be a"shortcut"for e- commerce. As an interorganizational information commerce adoption, but business processes may not system, it had some features in common with be fully aligned with the internet; employees may not internet-based e-commerce [30), but it also exhibited get the exposure of e-commerce and thus lack of"buy significant differences, as EDI was typically a in", and organizational culture may remain separated proprietary technology over a private network con- fre trolled by one large manufacturer or supplier [50] here are conflicting views about the effect of pr P5a. Greater partner usage is positively associated technology, such as EDI, on the adoption of internet- with e-commerce adoption based e-commerce. On the one hand, the experience with EDI made organizations more familiar with P5b. Greater partner usage is negatively associated electronic media. Prior It infrastructure such as edi with e-commerce migration was necessary to leverage and integrate the new technologies [16]. In addition, the organizations 3.2.6. Perceived obstacle culture and operational processes may have already The adoption of e-commerce requires a substantial been adapted to fit to an electronic platform, thus degree of technical and organizational competence for reducing the adjustment cost and shortening the smooth transition. This assumes a higher dimension of learning curve On the other hand, the implementation organizational knowledge and the role of human factors of EDI often involved relationship-specific investment in facilitating the adoption process. The notion of between the firm and its trading partners. This could learning by doing argues that it takes time and expertise translate into switching costs in migrating to the to incorporate complex technologies in an organization internet. Also, the incremental benefits of investing in [1]. Adoption of a complex technology can also be e-commerce may be viewed as small described as a process of knowledge accumulation, Considering the conflicting effects of EDI use on e- especially when the innovations( 1)have an abstract and commerce adoption and migration, we expected the demanding scientific base, (2)are fragile because they negative effects to be stronger than the positive ones. do not always function as expected, (3)are difficult to Adopting internet-based e-commerce would induce try before the end system is implemented, and (4) ignificant switching costs on both end users and cannot be treated as a black box but it must be business processes [36]. Even after adoption of e- incorporated into the business processes to become commerce, some portion of the transactions may still be effective [2]. The adoption of e-commerce seems to conducted over EDI. This leads to the following possess these properties. When managers perceive such propositions obstacles, they become reluctant to adopt e-commerce
and research tools, and conduct online transactions with suppliers [46]. Therefore, firms that are capable of providing more web functionalities are better positioned to adopt e-commerce and capable web functionalities will make customers as well as trading partners more willing to conduct transactions online with greater migration to e-commerce. This suggests the propositions: P3a. Web functionalities are positively associated with e-commerce adoption. P3b. Web functionalities are positively associated with e-commerce migration. 3.2.4. EDI use Electronic data interchange was an antecedent of ecommerce. As an interorganizational information system, it had some features in common with internet-based e-commerce [30], but it also exhibited significant differences, as EDI was typically a proprietary technology over a private network controlled by one large manufacturer or supplier [50]. There are conflicting views about the effect of prior technology, such as EDI, on the adoption of internetbased e-commerce. On the one hand, the experience with EDI made organizations more familiar with electronic media. Prior IT infrastructure such as EDI was necessary to leverage and integrate the new technologies [16]. In addition, the organization’s culture and operational processes may have already been adapted to fit to an electronic platform, thus reducing the adjustment cost and shortening the learning curve. On the other hand, the implementation of EDI often involved relationship-specific investment between the firm and its trading partners. This could translate into switching costs in migrating to the internet. Also, the incremental benefits of investing in e-commerce may be viewed as small. Considering the conflicting effects of EDI use on ecommerce adoption and migration, we expected the negative effects to be stronger than the positive ones. Adopting internet-based e-commerce would induce significant switching costs on both end users and business processes [36]. Even after adoption of ecommerce, some portion of the transactions may still be conducted over EDI. This leads to the following propositions. P4a. The use of EDI is negatively associated with ecommerce adoption. P4b. The use of EDI is negatively associated with ecommerce migration. 3.2.5. Partner usage Many firms have relied on partners or contractors for their IS design and implementation tasks. An outsourcing approach has been popular in driving the growth of applications service providers. Relying on partners for e-commerce implementation may speed up the initial adoption of e-commerce, bypassing the potentially slow process associated with in-house development [3,27]. However, this may slow down an organization’s subsequent migration to e-commerce. Outsourcing may seem to be a ‘‘shortcut’’ for ecommerce adoption, but business processes may not be fully aligned with the internet; employees may not get the exposure of e-commerce and thus lack of ‘‘buy in’’; and organizational culture may remain separated from e-commerce. This leads to two propositions: P5a. Greater partner usage is positively associated with e-commerce adoption. P5b. Greater partner usage is negatively associated with e-commerce migration. 3.2.6. Perceived obstacles The adoption of e-commerce requires a substantial degree of technical and organizational competence for smooth transition. This assumes a higher dimension of organizational knowledge and the role of human factors in facilitating the adoption process. The notion of learning by doing argues that it takes time and expertise to incorporate complex technologies in an organization [1]. Adoption of a complex technology can also be described as a process of knowledge accumulation, especially when the innovations (1) have an abstract and demanding scientific base, (2) are fragile because they do not always function as expected, (3) are difficult to try before the end system is implemented, and (4) cannot be treated as a black box but it must be incorporated into the business processes to become effective [2]. The adoption of e-commerce seems to possess these properties. When managers perceive such obstacles, they become reluctant to adopt e-commerce. W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 209
w. Hong, K. Zhu/ Information Management 43(2006) 204-22 Such obstacles could also make it more difficult for Table 2 business to migrate to the internet. These lead to a pair Sample description(sample size= 1036) of P6a. Perceived obstacles are negatively associated United States 838(80.9% with e-commerce adoption 98(19.1%) Title of the respondent P6b. Perceived obstacles are negatively associated President, Owner, or Managing Director 157(152%) with e-commerce mig gration Chief Information Officer(CIOMChief 92(8.9%) Technology Officer/VP of 4. Research methodology IS Manager. Director, Planner 281(27.1%) Other manager in Is department 99(96%) 4.1. Data and sample usiness Operations Manager 52(50%) Administration/Finance Manager 76(7.3%) We adopted a field survey methodology. Telephone Firm size interviews were conducted. which allowed trained Fewer than 50 271(26.2%) interviewers to clarify questions to ensure accurate 74(7.1%) )-199 144(13.9%) and meaningful responses. A computer-aided tele 200.499 171(16.5%) phone interviewing system(CATD) was used. The 85(82%) system provided various automatic data checks while 00-4999 137(13.2%) the respondent remained on the line. The survey was 5000-9999 44(4.2%) 10000 nd April 2001 9609.3%) professional firm specialized in IT-related survey research. Eligible respondents were executives or /wholesale senior managers best qualified to speak about the firms overall e-commerce activities. At the start of the interview, a screening question asked the respondents whether they would felt qualified to answer questions (3)organizations belonging to the public sector, such (see Appendix). Only those who answered positively as government and education, because their main ontinued with the interview purpose was not in generating revenues. The sample frame was obtained from a Dun Bradstreet database that contained representative firms This left us with 1036 observations in the usable in the entire local market, regardless of computerization sample replacement was used, with a predetermined number of Table 2 presents the descriptive statistics of the or web access. A stratified sampling method without al sample. Titles of the respondents reflect large firms selected randomly from each firm size and variations in the nature of the It management dustry category to ensure an unbiased representation responsibilities in these organizations. There was also of the sample distribution. It was considered important a wide range of firm size(from fewer than 50 to more to include both adopters and non-adopters in the sample than 10,000 employees). The firms represent three so that we could examine how the proposed factors industries: manufacturing, retail/wholesale, and ser- infuenced both adopters and non-adopters. More than vice. Overall, the sample represented a wide range of 2000 interviews were conducted in the United States firms, increasing the generalizability of the results and Canada. Samples that had any of the following characteristics were excluded 4.2. Variables and constructs (1) interviews that had missing values on the 4.2.1. Dependent variable dependent variable The dependent variable in the model was a cate- (2)firms without a public website: gorica variable with three groups, i.e., non-adopters
Such obstacles could also make it more difficult for business to migrate to the internet. These lead to a pair of propositions: P6a. Perceived obstacles are negatively associated with e-commerce adoption. P6b. Perceived obstacles are negatively associated with e-commerce migration. 4. Research methodology 4.1. Data and sample We adopted a field survey methodology. Telephone interviews were conducted, which allowed trained interviewers to clarify questions to ensure accurate and meaningful responses. A computer-aided telephone interviewing system (CATI) was used. The system provided various automatic data checks while the respondent remained on the line. The survey was conducted between February and April 2001 by a professional firm specialized in IT-related survey research. Eligible respondents were executives or senior managers best qualified to speak about the firm’s overall e-commerce activities. At the start of the interview, a screening question asked the respondents whether they would felt qualified to answer questions (see Appendix). Only those who answered positively continued with the interview. The sample frame was obtained from a Dun & Bradstreet database that contained representative firms in the entire local market, regardless of computerization or web access. A stratified sampling method without replacement was used, with a predetermined number of firms selected randomly from each firm size and industry category to ensure an unbiased representation of the sample distribution. It was considered important to include both adopters and non-adopters in the sample so that we could examine how the proposed factors influenced both adopters and non-adopters. More than 2000 interviews were conducted in the United States and Canada. Samples that had any of the following characteristics were excluded: (1) interviews that had missing values on the dependent variable; (2) firms without a public website; (3) organizations belonging to the public sector, such as government and education, because their main purpose was not in generating revenues. This left us with 1036 observations in the usable sample. Table 2 presents the descriptive statistics of the final sample. Titles of the respondents reflect large variations in the nature of the IT management responsibilities in these organizations. There was also a wide range of firm size (from fewer than 50 to more than 10,000 employees). The firms represent three industries: manufacturing, retail/wholesale, and service. Overall, the sample represented a wide range of firms, increasing the generalizability of the results. 4.2. Variables and constructs 4.2.1. Dependent variable The dependent variable in the model was a categorical variable with three groups, i.e., non-adopters, 210 W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 Table 2 Sample description (sample size = 1036) Variables Frequency (%) Country United States 838 (80.9%) Canada 198 (19.1%) Title of the respondent President, Owner, or Managing Director 157 (15.2%) Chief Information Officer (CIO)/Chief Technology Officer/VP of information systems 92 (8.9%) IS Manager, Director, Planner 281 (27.1%) Other manager in IS department 99 (9.6%) Business Operations Manager 52 (5.0%) Administration/Finance Manager 76 (7.3%) Firm size Fewer than 50 271 (26.2%) 50–99 74 (7.1%) 100–199 144 (13.9%) 200–499 171 (16.5%) 500–999 85 (8.2%) 1000–4999 137 (13.2%) 5000–9999 44 (4.2%) 10000+ 96 (9.3%) Industry Manufacturing 590 (56.9%) Retail/wholesale 70 (6.8%) Services 376 (36.3%)
w.Hong, K. Zhu/Information Management 43(2006)204-221 potential adopters, and adopters Depending whether a 5. Data analysis and results company has generated revenues from internet sales in the prior 12 months and whether it expected to generate 5.1. Measurement model revenues from internet sales in the next 12 months. we defined non-adopters as firms having O% revenue from Three independent variables, technology integra- internet sales in both years, potential adopters as firms tion, perceived obstacles, and partner usage, were having 0% online revenue in the prior year but measured by multiple indictors. Therefore, we needed expecting to have positive online revenues the current to assess their psychometric properties. Before the year,and adopters as firms having greater than 0% assessment, missing data were removed by listwise online revenues in both years. The dependent variable deletion, so that only complete observations would be in the migration model was operationalized as the used [20). This resulted in an N=806 sample for revenue from the internet divided by the total revenue testing the measurement model (see Appendix A combination of exploratory and confirmatory factor analysis methods was used in validating the 4.2.2. Independent variables measurement of the multi-item constructs. The sample chnology integration was measured by asking with complete data on the multi-item constructs was respondents the level of integration of various It randomly split into two sub-samples to enable cross- systems with the internet. The systems covered validation. The first sub-sample(N=406) was used major IT functional areas, including ERP, CRM, for exploratory factor analysis(EFA)while the second SCM, materials management, inventory control, (N=400) was used for confirmatory factor analysis counting, and financial management. Together (CFA). The reason for conducting both EFA and CFA they reflect how well the IT systems are connected on is that the Efa procedure allowed us to drop some a common platform Web spending was measured by invalid items from the scale while CFA then refined the percentage of the firms Is operating budget the scale to ensure the validity of the instrument using devoted to web hardware, software, IT services, a different sample To ensure that the two sub-samples consulting, and employee training. Web functional- were comparable and unbiased, we tested the equity of ities was measured as the number of web-services the means of the demographic data in the two sub- provided by the website based on Zhu and Kraemer samples. The results of t-tests showed that there was definition and listed in Appendix. EDI use was a no significant difference between the two sub- binary variable: companies that were currently using samples EDI were coded as one. otherwise zero. partner usage was measured as the degree of using partners 5.1.1. Exploratory factor analysis or contractors in IT-related areas(website design and a principal component analysis with oblimin operation, implementation, and application devel- rotation was used to examine the factor structure of opment). Perceived obstacle was measured by asking the measures in the first sub-sample. Three factors respondents to describe their perceived level of emerged with eigenvalue above 1.0, explaining a total ficulty in managing their websites along a 5-point of 64% of the variance in the data. Items with low Likert-scale loadings on the intended factor or high cross-loadings on other factors were removed. The resulting instruments(as presented in Appendix) were eval- Some of the cross-sectional variations can uated for reliability, convergent validity, and dis- explained only if controls are appropriately applied. criminant validity To control for firm-and industry-specific effects, we The reliability or internal consistency was assessed employed two control variables: firm size [15,351, by computing Cronbachs alpha and composite otal number of employees, and industry type reliability. They were all above 0.8, which is higher coded into three dummy variables, representing than the 0.7 threshold normally considered as retail/wholesale, service, and manufacturing indus- minimum 31]. Convergent and discriminant validities were examined by both factor loadings and a
potential adopters, and adopters. Depending whether a company has generated revenues from internet sales in the prior 12 months and whether it expected to generate revenues from internet sales in the next 12 months, we defined non-adopters as firms having 0% revenue from internet sales in both years, potential adopters as firms having 0% online revenue in the prior year but expecting to have positive online revenues the current year, and adopters as firms having greater than 0% online revenues in both years. The dependent variable in the migration model was operationalized as the revenue from the internet divided by the total revenue (see Appendix). 4.2.2. Independent variables Technology integration was measured by asking respondents the level of integration of various IT systems with the internet. The systems covered major IT functional areas, including ERP, CRM, SCM, materials management, inventory control, accounting, and financial management. Together they reflect how well the IT systems are connected on a common platform. Web spending was measured by the percentage of the firm’s IS operating budget devoted to web hardware, software, IT services, consulting, and employee training. Web functionalities was measured as the number of web-services provided by the website based on Zhu and Kraemer’ definition and listed in Appendix. EDI use was a binary variable: companies that were currently using EDI were coded as one, otherwise zero. Partner usage was measured as the degree of using partners or contractors in IT-related areas (website design and operation, implementation, and application development). Perceived obstacle was measured by asking respondents to describe their perceived level of difficulty in managing their websites along a 5-point Likert-scale. 4.2.3. Control variables Some of the cross-sectional variations can be explained only if controls are appropriately applied. To control for firm- and industry-specific effects, we employed two control variables: firm size [15,35], total number of employees, and industry type, coded into three dummy variables, representing retail/wholesale, service, and manufacturing industries. 5. Data analysis and results 5.1. Measurement model Three independent variables, technology integration, perceived obstacles, and partner usage, were measured by multiple indictors. Therefore, we needed to assess their psychometric properties. Before the assessment, missing data were removed by listwise deletion, so that only complete observations would be used [20]. This resulted in an N = 806 sample for testing the measurement model. A combination of exploratory and confirmatory factor analysis methods was used in validating the measurement of the multi-item constructs. The sample with complete data on the multi-item constructs was randomly split into two sub-samples to enable crossvalidation. The first sub-sample (N = 406) was used for exploratory factor analysis (EFA) while the second (N = 400) was used for confirmatory factor analysis (CFA). The reason for conducting both EFA and CFA is that the EFA procedure allowed us to drop some invalid items from the scale while CFA then refined the scale to ensure the validity of the instrument using a different sample. To ensure that the two sub-samples were comparable and unbiased, we tested the equity of the means of the demographic data in the two subsamples. The results of t-tests showed that there was no significant difference between the two subsamples. 5.1.1. Exploratory factor analysis A principal component analysis with oblimin rotation was used to examine the factor structure of the measures in the first sub-sample. Three factors emerged with eigenvalue above 1.0, explaining a total of 64% of the variance in the data. Items with low loadings on the intended factor or high cross-loadings on other factors were removed. The resulting instruments (as presented in Appendix) were evaluated for reliability, convergent validity, and discriminant validity. The reliability or internal consistency was assessed by computing Cronbach’s alpha and composite reliability. They were all above 0.8, which is higher than the 0.7 threshold normally considered as minimum [31]. Convergent and discriminant validities were examined by both factor loadings and a W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 211
w. Hong, K. Zhu/ Information Management 43(2006) 204-22 Fit indices for the measurement model Fit indices Recommended value Measurement model Goodness-of-fit(GFD) Adjusted goodness-of-fit(AGFI) normalized fit index(NFD Non-normalized fit index (NNFD) >0.90 Comparative fit index(CF Root mean square residual(RMSR) 0.60) on or 0.70 following a stricter criterion [19]. All the factor their associated factors, showing convergent validity. loadings were greater than 0.50 with a majority of Furthermore, all items loaded much higher on the them above 0. 70. Also, all items loaded significantly associated factors than on any other(with no cross-(p<0.01)on their underlying construct, indicating loading greater than 0. 25), demonstrating discriminant high convergent validity. Discriminant validity was tested by examining whether the shared variance 5.1.2. Confirmatory factor analysis We used lisreL8.30 to perform the confirmatory Table 4 factor analysis on the second sub-sample. The fit of the Results of the confirmatory factor analysis overall measurement model was estimated by various Factor loadings SE I- Value indices(see Table 3). The x--statistics was not used (standardized) because of its sensitivity to large sample size. Instead, Technology integration(Cronbach' s alpha =0.85; composite the ratio of x to degrees-of-freedom(d f )was used. A value of 2.33 was obtained. which was within the recommended value of 3 9). Also all fit indices were 0.64 above 0.90, indicating good model fit. Finally, the root INTA 0.53 mean square residual(RMSR), which indicates the INTS proportion of the variance not explained by the model was 0.06, implying a good fit between the observed 0.71 data and the proposed model [8]. RMs error of approximation(RMSEA), which describes the dis- rusage( Cronbachs alpha=0.91; composite reliability =0.91) crepancy between the proposed model and the population covariance matrix, was 0.06, which was also lower than the recommended cut-off value of 0.08 0.83 Next, we proceed to examine the instruments Perceived obstacle (Cronbach's alpha =0.89: composite reliability, convergent validity and discriminant eliability =0.89) validity [37]. Similar to the results of the EFA, Cronbachs alpha (composite reliability) was 0.85 (0. 81)for technology integration, 0.91(0.91) fc 0.06 partner usage, and 0. 89(0.89)for perceived obstacles, indicating high reliability of the measures. Convergent Note N/A, the first item of each construct is specified as a fixed validity was tested by factor loadings(see Table 4), parameter having a value of 1.0. Therefore, the standard errors(SE) which are considered as significant if greater than 0.50 and the t-values cannot be estimated
correlation matrix. All items load highly (>0.60) on their associated factors, showing convergent validity. Furthermore, all items loaded much higher on the associated factors than on any other (with no crossloading greater than 0.25), demonstrating discriminant validity. 5.1.2. Confirmatory factor analysis We used LISREL8.30 to perform the confirmatory factor analysis on the second sub-sample. The fit of the overall measurement model was estimated by various indices (see Table 3). The x2 -statistics was not used because of its sensitivity to large sample size. Instead, the ratio of x2 to degrees-of-freedom (d.f.) was used. A value of 2.33 was obtained, which was within the recommended value of 3 [9]. Also all fit indices were above 0.90, indicating good model fit. Finally, the root mean square residual (RMSR), which indicates the proportion of the variance not explained by the model, was 0.06, implying a good fit between the observed data and the proposed model [8]. RMS error of approximation (RMSEA), which describes the discrepancy between the proposed model and the population covariance matrix, was 0.06, which was also lower than the recommended cut-off value of 0.08. Next, we proceed to examine the instruments’ reliability, convergent validity and discriminant validity [37]. Similar to the results of the EFA, Cronbach’s alpha (composite reliability) was 0.85 (0.81) for technology integration, 0.91 (0.91) for partner usage, and 0.89 (0.89) for perceived obstacles, indicating high reliability of the measures. Convergent validity was tested by factor loadings (see Table 4), which are considered as significant if greater than 0.50 or 0.70 following a stricter criterion [19]. All the factor loadings were greater than 0.50 with a majority of them above 0.70. Also, all items loaded significantly ( p < 0.01) on their underlying construct, indicating high convergent validity. Discriminant validity was tested by examining whether the shared variance between constructs was lower than the average 212 W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 Table 4 Results of the confirmatory factor analysis Factor loadings (standardized) S.E. t-Value Technology integration (Cronbach’s alpha =0.85; composite reliability = 0.81) INT1 0.58 N/A N/A INT2 0.71 0.11 10.53 INT3 0.64 0.10 9.82 INT4 0.53 0.11 8.65 INT5 0.62 0.10 9.64 INT6 0.76 0.11 10.97 INT7 0.71 0.12 10.54 INT8 0.60 0.10 9.41 Partner usage (Cronbach’s alpha = 0.91; composite reliability = 0.91) PAR1 0.81 N/A N/A PAR2 0.73 0.06 16.07 PAR3 0.89 0.05 20.80 PAR4 0.84 0.06 19.26 PAR5 0.83 0.06 19.08 Perceived obstacle (Cronbach’s alpha = 0.89; composite reliability = 0.89) OBS1 0.76 N/A N/A OBS2 0.83 0.06 17.14 OBS3 0.83 0.05 17.17 OBS4 0.79 0.06 16.38 OBS5 0.74 0.06 15.13 Note: N/A, the first item of each construct is specified as a fixed parameter having a value of 1.0. Therefore, the standard errors (S.E.) and the t-values cannot be estimated. Table 3 Fit indices for the measurement model Fit indices Recommended value Measurement model x2 /d.f. 3 2.33 Goodness-of-fit (GFI) 0.90 0.92 Adjusted goodness-of-fit (AGFI) 0.80 0.90 Normalized fit index (NFI) 0.90 0.92 Non-normalized fit index (NNFI) 0.90 0.95 Comparative fit index (CFI) 0.90 0.95 Root mean square residual (RMSR) 0.10 0.06 Root mean square error of approximation (RMSEA) 0.08 0.06 Note: N/A means not applicable.
w.Hong. K. Zhu/ Information Management 43(2006)204-221 213 Table 5 Discriminant validity Technology integration Partner usage Perceive obstacle echnology integration artner usage 0.68 erceived obstacle Note: Diagonals represent the average variance extracted. Other entries represent the shared variance. variance extracted of the individual constructs they 82()=In P(Y minant validity(see Table 5). Thus, the measurement - B20+ B2 Firmsize B22Indusl+ P23Indus2 model demonstrated adequate psychometric validity and internal consistency pa4TechInte B2s WebSpenc B2 WebFunc +P27EDI+ B28 PartUsg 5.2. Adoption model: multi-nominal logistic +pagPercobs=xB2 where Bis are the coefficients; the names of the Multi-nominal logistic regression was performed to independent variables are self-explanatory. Then, the test our propositions, with non-adopters(O), potential conditional probability P(Y=jix)forj=0, 1, 2, can be adopters(1), and adopters(2)as the three-category expressed as dependent variable. We chose this technique over multiple regressions because our dependent variable p(Y=/A, 1+ Ers+esa(r), forj=0,1,2 was categorical rather than continuous. Also, logistic regression requires fewer assumptions than discrimi (3) nant analysis and was, thus more robust in the face of data conditions that could negatively impact dis We classified the sample (N= 1036)into non- criminant analysis adopters, potential adopters, and adopters(see Table 6) We denote p covariates (including control vari Cases with missing data were removed, leaving us with ables) and a constant term by the vector, x, of length a total of 627 cases for regression analysis. The p+l, where Xo= 1. The two logit functions are randomness of the missing data across the six inde- defined as: pendent variables and two control variables were tested by the correlations among the dichotomous variables 8I(c)=In/ P(=Olr) generated from the original variables by replacing valid data with a value of one and missing data with a value of P(Y=2|x) zero. The majority of the correlations between any pair B1o+BuFirmsize+BuIndusl +Bu3Indus2 of dichotomous variables were below 0.1, and the pu4TechInte+ Pis WebSpend highest correlation was below 0.3. Therefore. we onsidered the deletions to be +B1 Web Func +Bu7EDI+ BusPartUsg To assess the overall fit of the model. we used the PigPercobs=xp, )) chi-square test and pseudo R[21]. First, we adopted Percent of revenue from internet sales =0%o in past year >O% in past year =0%in 595(non-adopter) 7(quitter) >O% in next year 106(potential adopter)
variance extracted of the individual constructs they were, and thus the instrument demonstrated discriminant validity (see Table 5). Thus, the measurement model demonstrated adequate psychometric validity and internal consistency. 5.2. Adoption model: multi-nominal logistic regression Multi-nominal logistic regression was performed to test our propositions, with non-adopters (0), potential adopters (1), and adopters (2) as the three-category dependent variable. We chose this technique over multiple regressions because our dependent variable was categorical rather than continuous. Also, logistic regression requires fewer assumptions than discriminant analysis and was, thus more robust in the face of data conditions that could negatively impact discriminant analysis. We denote p covariates (including control variables) and a constant term by the vector, x, of length p + 1, where x0 = 1. The two logit functions are defined as: where bi’s are the coefficients; the names of the independent variables are self-explanatory. Then, the conditional probability P(Y = jjx) for j = 0, 1, 2, can be expressed as: PðY ¼ jjxÞ ¼ 1 1 þ eg1ðxÞ þ eg2ðxÞ ; for j ¼ 0; 1; 2 (3) We classified the sample (N = 1036) into nonadopters, potential adopters, and adopters (see Table 6). Cases with missing data were removed, leaving us with a total of 627 cases for regression analysis. The randomness of the missing data across the six independent variables and two control variables were tested by the correlations among the dichotomous variables generated from the original variables by replacing valid data with a value of one and missing data with a value of zero. The majority of the correlations between any pair of dichotomous variables were below 0.1, and the highest correlation was below 0.3. Therefore, we considered the deletions to be appropriate. To assess the overall fit of the model, we used the chi-square test and pseudo R2 [21]. First, we adopted W. Hong, K. Zhu / Information & Management 43 (2006) 204–221 213 Table 5 Discriminant validity Technology integration Partner usage Perceive obstacle Technology integration 0.42 Partner usage 0% in past year Total =0% in next year 595 (non-adopter) 7 (quitter) 602 >0% in next year 106 (potential adopter) 328 (adopter) 434 Total 701 335 1036 g1ðxÞ ¼ ln PðY ¼ 0jxÞ PðY ¼ 2jxÞ ¼ b10 þ b11Firmsize þ b12Indus1 þ b13Indus2 þ b14TechInte þ b15WebSpend þ b16WebFunc þ b17EDI þ b18PartUsg þ b19PercObs ¼ x0 b1 (1) g2ðxÞ ¼ ln PðY ¼ 1jxÞ PðY ¼ 2jxÞ ¼ b20 þ b21Firmsize þ b22Indus1 þ b23Indus2 þ b24TechInte þ b25WebSpend þ b26WebFunc þ b27EDI þ b28PartUsg þ b29PercObs ¼ x0 b2 (2)