This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Exploring managerial factors affecting ERPimplementation: an investigation of theKlein-Sorra model using regression splinesKweku-Muata Osei-Bryson,* Linying Dong,† Ojelanki Ngwenyama‡
*Department of Information Systems and the Information Systems Research Institute,Virginia Commonwealth University, Richmond, VA 23284, USA, email:[email protected], †Ted Rogers School of Information Technology Management,Ryerson University, Toronto ON M5B 2K3, Canada, email: [email protected], and‡Institute for Innovation and Technology Management, Ted Rogers School ofManagement, Ryerson University, Toronto ON M5B 2K3, Canada, email:[email protected]
Abstract. Predicting successful implementation of enterprise resource planning(ERP) systems is still an elusive problem. The cost of ERP implementation failuresis exceedingly high in terms of quantifiable financial resources and organizationaldisruption. The lack of good explanatory and predictive models makes it difficult formanagers to develop and plan ERP implementation projects with any assuranceof success. In this paper we investigate the Klein & Sorra theoretical model ofimplementation effectiveness. To test this model we develop and validate a datacollection instrument to capture the appropriate data, and then use multivariateadaptive regression splines to examine the assertions of the model and suggestadditional significant relationships among the factors of their model. Our researchoffers new dimensions for studying managerial interventions in IT implementationand insights into factors that can be managed to improve the effectiveness of ERPimplementation projects.
Keywords: ERP implementation, IT implementation management, informationsystems success, information systems management, multivariate adaptiveregression splines
1. INTRODUCTION
Enterprise resource planning (ERP) systems have the potential for operational, managerialand strategic benefits to the organization when successfully implemented (Murphy & Simon,2002; Shang & Seddon, 2002). However, recent studies (e.g. Robbins-Gioia, 2002) have foundthat more than 70% of ERP implementations have not met the expectations of managers.Various prescriptions have been proposed for improving ERP implementation success, includ-
ing: (a) standardize business practices to fit with the ERP software (Kremers & Dissel, 2000;Markus et al., 2000a; Sumner, 2000); (b) avoid customizing the software (Mabert et al., 2001;Murray & Coffin, 2001; Parr & Shanks, 2000); (c) provide appropriate training in the use of thesystem (Bingi et al., 1999; Sumner, 2000; Al-Mudimigh et al., 2001); and (d) provide projectteam education and proper project management (Bingi et al., 1999; Holland & Light, 1999;Kumar & Crook, 1999; Willcocks & Sykes, 2000). Some researchers have also attempted todevelop frameworks of essential characteristics of good ERP implementation management,generally referred to as critical success factors (CSFs; Bingi et al., 1999; Holland & Light, 1999;Nah et al., 2001), and other studies have adopted or extended various predictive models oftechnology acceptance to the problem of ERP systems (Kositanurit et al., 2006). However,studies examining managerial factors that influence ERP implementation success arescarce.
According to Markus et al. (2000a) ERP systems are ‘commercial software packages thatenable the integration of transactions-oriented data and business processes throughout anorganization’. They are a class of information systems (IS) designed to integrate the enter-prise’s core processes, information processing and information management (Holland &Light, 1999). Some common examples of these are SAP, BAAN and PeopleSoft. While ERPsystems have been around for more than two decades, research on their implementationstarted in the 1990s. Many of these ERP implementation studies have built upon the priorresearch on implementing IS in organizations (Markus et al., 2000a; Kositanurit et al., 2006).However, much of the IS implementation research has focused on the individual user’sperspective on the adoption and use (Venkatesh et al., 2003). A significant limitation of thisIS implementation research is the lack of attention to the managerial dimension; this is alsomanifest in the ERP implementation research which builds upon it (Parr & Shanks, 2000;Hong & Kim, 2002; Kumar, Maheshwari & Kumar et al., 2003). Recently, some researchershave pointed to the need for research on IS implementation that moves beyond the existingIS implementation models (Chin & Marcolin, 2001b). Others have pointed to the lack oftheories or models that investigate managerial perspective factors that affect successful ISimplementation (Holahan et al., 2004). The theoretical model of innovation implementationproposed by Klein & Sorra (1996) focuses on managerial factors affecting successful imple-mentation of new technologies. This model (see Section 3) is used as a basis for ourexploration of managerial factors that lead to successful ERP implementation. The primaryresearch question is: What are the managerial factors that affect successful ERP implemen-tation? Our exploration included collecting survey data from ERP environments, factor analy-sis using partial least squares (PLS), and hypothesis testing using a relatively new statisticalanalysis technique, multivariate adaptive regression splines (MARS). We developed a ques-tionnaire, conducted a survey in six companies and analyzed the data collected from 239respondents. The findings of our study suggest new and fertile areas for researching themanagerial perspective of IS implementation. The rest of the paper is organized as follows:in Section 2 we discuss prior research and still open questions about IS Implementation. InSection 3 we outline the key details of the Klein and Sorra model. In Section 4 we describethe research methodology and the steps we followed in conducting the study. In Section 5 we
present the MARS analysis and findings. Finally, in Section 6 we conclude with a discussionon the contributions of this research.
2. PRIOR RESEARCH AND OPEN QUESTIONS
It has long been accepted that users and managers are two key players in IS implementation(Kwon & Zmud, 1987; Leonard-Barton & Deschamps, 1988). The role of managers is topromote system innovations and influence users to adopt them; individual users, on the otherhand, evaluate the system and social factors before making adoption decisions (Shang &Seddon, 2002; Chang, 2006). While research into IS implementation reflects these twoperspectives (individual user and managerial), more attention has been given to the individualuser perspective. Research on the individual perspective has a relatively long history and alarge and growing body of work, while the managerial perspective has been overlooked andonly few studies have focused on it. Due to space limitations we provide a brief overview ofthese two perspectives below, for more detailed discussions the reader should refer to thespecific references.
2.1 The individual user perspective
The major contributions to IS implementation research from the individual user perspective fallinto categories (a) user satisfaction and performance; and (b) task-technology fit (see Table 1).The user satisfaction and performance studies, focus on identifying the conditions under whichusers are satisfied with the IS they use. Doll & Torkzadeh (1988) define user satisfaction as theuser’s opinion of a IS application that she or he uses. The fundamental argument of thisapproach is that high levels of user satisfaction lead to high levels of user performance. Muchof the early work focused on developing instruments for measuring user satisfaction. Forexample, Bailey & Pearson (1983) conducted a literature review to identify influencing factorsand developed a questionnaire for investigating user satisfaction. Ives et al. (1983) replicatedand extended Bailey & Pearson’s (1983) study to provide evidence of the validity of theinstrument. Doll & Torkzadeh (1988) also developed an instrument that would measureend-user computing satisfaction. These studies paved the way for later research that linked theconstructs of user satisfaction, system characteristics and user performance. Some of thisresearch has also focused specifically on clarifying and confirming the relationship betweenuser satisfaction and end-user performance (DeLone & McLean, 1992). This argument wasalso the central point of the nomological net model of Igbaria & Tan (1997). In anotherimportant study, DeLone & McLean (1992) validate the constructs, system quality, informationquality, use, user satisfaction, and individual and organizational performance. Later, Etezadi-Amoli & Farhoomand (1996) developed a questionnaire instrument and tested the relationshipbetween end-user satisfaction and user performance.
The task-technology-fit (TTF) approach postulates that when the user’s task and the tech-nology are congruent, user performance will be high (Goodhue & Thompson, 1995; Goodhue,
1995; Dishaw & Strong, 1998; Mathieson & Keil, 1998). Consequently, studies falling underthis approach try to define task and technology characteristics and what is ‘goodness of fit’between specific technologies and end-user tasks (Dishaw & Strong, 1998; Mathieson & Keil,1998; Goodhue et al., 2000). Several researchers have used the TTF approach to explain theimpact of IS and task characteristics on individual performance (Goodhue & Thompson, 1995;Dishaw & Strong, 1998; Ferratt & Vlashos, 1998). This approach supports the argument thatwhen there is fit between user task characteristics and characteristics of the IS, utilization ofthe system will be high and user performance will also be high. Goodhue & Thompson (1995)found support for the relationships TTF and performance, and utilization and performance.
Recent research by Venkatesh et al. (2003) has attempted to integrate the technologyacceptance model and user acceptance into a unified model called, unified theory of accep-tance and use of technology. Further, Kositanurit et al. (2006) developed and tested anintegrated task-technology fit user performance model for the ERP environment. But neitherof these has addressed the managerial dimension of the ERP implementation problem.The managerial perspective remains under investigated, even though managers play a salientrole in diffusing an information technology into a user community (Kwon & Zmud, 1987,pp. 124–125).
2.2 The managerial perspective
IS implementation research that focused on managerial issues can be categorized into twogroups: (a) case studies that have identified critical success factors for IS implementation; and(b) survey studies focused on validating a narrow set of constructs. Management interventionshave been examined from different perspectives (see Table 2). For example, Thompson et al.(1991) developed the construct facilitating conditions that suggests factors to which managersmust attend in order to cultivate a supportive environment for individual technology adoption.Davis (1989) and Venkatesh et al. (2003) investigated how managers might affect useradoption behaviors by setting norms. Compeau & Higgins (1995a) and Igbaria et al. (1997)have investigated the importance of management support by providing resources and trainingfor users of new technologies. More recently however, managerial research on ERP imple-mentation has been dominated by case studies, which explore the complex and dramaticorganizational changes (Besson & Rowe, 2001; Lee & Myers, 2004) during and after imple-mentation of these systems (see Table 2). Some researchers have focused on explicatingvarious managerial problems from strategic alignment (Amrani et al., 2006; Clemons & Simon,2001; Grant, 2003; Soh & Sia, 2004) to change management (Besson & Rowe, 2001; Lee &Myers, 2004; Chae & Poole, 2005; Pozzebon & Pinsonneault, 2005) and knowledge transfer(Volkoff, 1999; Lee & Lee, 2000; Robey et al., 2002; Ko et al., 2005). Much of the managerialresearch on ERP implementation has been prescriptive or normative focusing on what wouldbe considered ‘best practice’ for successful implementation of ERP systems (Davenport, 1998;Besson & Rowe, 2001). Some researchers have attempted to develop frameworks to assistmanagers in defining and analysing critical success factors (Akkermans et al., 1999; Bingiet al., 1999; Holland & Light, 1999; Nah et al., 2001), risk factors (Sumner, 2000), to ensure
effective planning and decision making about ERP implementation projects, and for buildingintra-organizational coalitions for success (Pozzebon & Pinsonneault, 2005; Ward et al.,2005). Nah et al. (2001) have suggested what is good management practice and Willcocks &Sykes (2000) offer advice on the roles of Chief Information Officer and IT functions during ERPimplementations.
While much of the managerial research has been oriented to developing concepts fromempirical field observations, some studies have attempted to use organizational influence theory(Avital & Vandenbosch, 2000; Chae & Poole, 2005) to understand organizational power shiftsand institutional theory (Markus et al., 2000b; Soh & Sia, 2004; Boersma & Kingma, 2005), tounderstand mutual adaptation between ERP technologies and the organization. There is also asmall body of work that investigates problems of knowledge transfer between various actors inthe ERP implementation project and their implications for success (Lee & Lee, 2000; Robeyet al., 2002; Volkoff et al., 2004). In summary, ERP studies on management issues, by applyingcase study method, offer an in-depth understanding of challenges faced by management in ERPimplementations. However, there is a lack of empirical studies focused on developing predictivemodels to assist managerial decision-making and intervention to improve ERP implementationoutcomes. Presently, few studies have focused on developing predictive models for ERPimplementation outcomes (Amoako-Gyampah & Salam, 2004; Kositanurit et al., 2006).Amoako-Gyampah & Salam (2004) extended the technology acceptance model for ERPimplementation environments and Kositanurit et al. (2006) proposed and tested a user perfor-mance and task technology model for the ERP environment. Both of these studies took theindividual user perspective but did not focus attention on the managerial perspective.
3. THE KLEIN AND SORRA MODEL
The Klein and Sorra model (see Figure 1), is one of few proposed predictive models thataddressed the managerial perspective of IS implementation. Klein and Sorra argue that keydeterminants of implementation effectiveness are: (a) climate for implementation, whichdescribes employees’ ‘shared perceptions of the events, practices, and behaviors that arerewarded, supported, and expected in a setting’ and (b) innovation-values fit, which depicts theextent to which the innovation is consistent with shared employee values. A strong organiza-tion implementation climate, is one that includes organizational influence and compliancemechanisms, establishes necessary user skills and provides appropriate incentives for use(and disincentives for non-use), and removes obstacles to use (Klein & Sorra, 1996). Thestronger the organizational climate for implementation is, the more likely it is that targetedusers will apply the system frequently, consistently and enthusiastically (more on this later).
Two studies have tested parts of the Klein & Sorra (1996) model. Klein et al. (2001)confirmed the relationship between implementation climate and implementation effectivenessacross 39 factory environments. And Holahan et al. (2004) confirmed the influence of imple-mentation climate on implementation effectiveness but failed to confirm innovation-values fit asa key determinant of implementation effectiveness. Neither of these tested the complete set of
factors that Klein & Sorra (1996) postulated as impacting implementation effectiveness. Con-sequentially, the Klein and Sorra model still remains a set of conjectures based on theoreticalspeculation about the phenomena. Our interest is in empirically examining their conjectures inorder to determine how useful this model is in explaining or accounting for successful ISimplementation. The basic objective of this investigation is on testing the conjectures of themodel, proposing alternative conjectures and finding the ones that can offer the potential for abetter explanation of the empirical evidence within the context of ERP systems. This approachto scientific investigation of models and conjectures with the intention of identifying the betterexplanatory model or theory is well discussed in philosophy of science (cf. Day & Kincaid,1994). The basic principle of this approach suggests that a model M1 is preferred when itsconjectures or hypotheses offer a better explanation of the evidence than the conjectures orhypotheses of another model M0 with regard to the same empirical evidence.
The Klein and Sorra model of implementation is based on Sussmann & Vecchio (1982), workon organizational influence, which in turn is built upon the social influence theory of Kelman
(1961). Social influence theory suggests three processes through which organizations influ-ence employees’ behavior: (a) compliance, in which an employee complies with an organiza-tional influence to gain specific rewards and to avoid punishments; (b) identification, in whichan employee accepts the organizational influence to engage in a satisfying role-relationshipwith another person or a group; and (c) internalization, in which an employee internalizes theorganizational influence because it is congruent with his value systems or because it isintrinsically rewarding. Klein and Sorra argue that internalization and compliance are twoimportant processes for effective implementation of new technologies. In their model, imple-mentation effectiveness is defined as ‘quality and consistency of use of an adopted innovation’(Klein & Sorra, 1996, p. 1056). Two determinants of implementation effectiveness are: (a)climate for implementation, which describes employees’ ‘shared perceptions of the events,practices, and behaviors that are rewarded, supported, and expected in a setting’ (Schneider,1990, p. 384) and (b) innovation-values fit, which depicts the extent to which the innovation isconsistent with the values of the organization members (Klein & Sorra, 1996, p. 1063).
4. METHODOLOGICAL DETAILS OF OUR INVESTIGATION
Our methodology for investigating the Klein-Sorra model can be described as a four-phasedprocess: (a) development and validation of an instrument for collecting data about theconstructs defined in the model; (b) survey data collection that includes 209 participants insix organizations which implemented enterprise systems from different venders (e.g. SAP,PeopleSoft, Oracle, Bann, JDE); (c) the use of PLS analysis to validate the relevant constructsdefined in Klein and Sorra’s model; and (d) the use of MARS analysis to evaluate and describecausal links existing between factors in Klein and Sorra’s model. We discuss this final phasein Section 4 of the paper.
4.1 Phase 1: scale development
Three of the seven constructs (i.e. innovation-values fit, implementation effectiveness andcommitment) in Klein and Sorra’s model have related measures developed by previous studies(for details see Appendix A). In particular, we adapted measures of TTF (i.e. quality describingthe extent to which an IS provides current, updated, and useful information and locatibilitydescribing how easily it is to find data in the system) by Goodhue (1995; 1998) and measuresof work processes evaluating degree of repetitiveness by Valle et al. (2000) to evaluate asystem’s fit with users’ task-related values. Implementation effectiveness was evaluated bythe measures developed by Klein et al. (2001). Other constructs including implementationclimate, skills, absence of obstacles, and commitment became the focal point of the scaledevelopment.
To develop measurement items for these constructs, we first created a pool of raw mea-surement items based on the review of prior studies. In particular, the measurements forimplementation Climate were created based on four core dimensions of implementation
Climate – task support (i.e. the extent to which employees perceive that they are beingsupplied with the material, equipment, services and resources necessary to perform their jobs),reward emphasis (i.e. the extent to which employees perceive that various organizationalrewards are to be allocated on the basis of their job performances), mean emphasis (i.e. theextent to which managers make known the methods and procedures that employees areexpected to use in performing their jobs) and goal emphasis (i.e. the extent to which managersmake known the types of outcomes and standards that employees are expected to accom-plish; Kopelman et al., 1990). Measures for users’ Skills were developed to evaluate whethera user has comprehended basic concepts, applications, and products of a system, how wellthe user inputs and interprets information in the system, and whether the user can do somesimple troubleshooting (Simon et al., 1996). For incentives and absence of obstacles, wefocused on their definitions and created measures to examine whether users are givenincentives to use an IS (incentives) and the extent to which individual use of an IS is deterredby the organization (absence of obstacles). Measures for commitment were developed basedon seven-item measures for affective commitment of Meyer & Allen (1991) to assess users’emotional attached to a newly.
The created measurement items were then reviewed by four IS researchers who wereknowledgeable about the IS innovation and implementation and three people who participatedin IS implementations, and then were entered into the card sorting after rewording unclearitems, deleting the items that did not make sense, and adding the items suggested by thereviewers. The remaining candidate items were further explored using the popular four-roundcard-sorting method of Moore & Benbasat (1996). In comparing the card-sorting results of thefirst round with those of the fourth round, we notice not only that the number of itemsdramatically reduced to 35 from 50, but also the Cohen’s Kappa index dramatically increased,suggesting that the coherence among other items under their assigned constructs wasenhanced (see Table 3). As a result, the card sorting improved the validity and reliability of theconstruct measurements. With the refined items, we went ahead with the full survey.
4.2 Phase 2: data collection
We randomly selected and contacted 800 midsized or large Canadian companies from twosources: ‘Canada Top 1000’ online database (from The Globe and Mail 2002), and SCOTT, a
Table 3. Results of the four rounds of card sorting
Results of card sorting Judges A–D First round Second round Third round Fourth round
comprehensive database of Canadian companies. Fifteen organizations qualified havingrecently implemented ERP systems, and consented to participate in our research. After a fewrounds of communications clarifying the responsibilities of the companies, only six of themwere committed to the research project. We asked the six companies to provide us with thename of ERP system recently implemented and the contact information of the targeted users.All together 422 users were identified and we sent out questionnaires to each of them. Wereceived 239 responses (a 56.6% response rate). Of the 239 responses only 209 were usable.We then assessed non-response bias by comparing the responses from the first batch(received within the first 2 weeks after the survey questionnaire was delivered) with those fromthe last batch (received 2 months after) (Armstrong & Overton, 1977), but did not findsignificant differences in age, work experience, or job tenure (Hotelling’s Trace = 1.034,p = 0.235), nor did we find any significant discrepancies in position (Chi-square = 1.29,p = 0.256), gender (Chi-square = 1.718, p = 0.190), or education (Chi-square = 2.992, p =0.224). The sample size for each organization varied from 30 to 42. In total, the averagerespondent age was 40 years, the average time spent in service of the organization wasapproximately eight and one half years, and the average time holding the current position wasa little more than 5 years. Of all the participants, 81 (38.8%) had a completed university orgraduate degree, and 54 (25.8%) had completed high school or some college or universitystudy. Approximately 56% of respondents were male.
4.3 Phase 3: validation of the model constructs
We applied PLS, a second-generation statistical method, to obtain factor scores for eachconstruct in the model. PLS was adopted in this paper due to its rigor, ease to use, and itsability to analyze both formative and reflective constructs (Chin, 1998a; 1998b; Gefen et al.,2000). The sample exceeded the ‘10 times’ PLS sample size heuristic described by Chin et al.(2003). In our analysis implementation climate is the construct with the greatest number offormative indicators, it contained four items. Implementation effectiveness is the construct withthe greatest number of structural paths (i.e. two) leading into it. The model required a minimumof 40 cases to be estimated. To obtain the factor scores for our analysis we used the PLS 3.0bootstrap method with 200 re-sampling. And two criteria were adopted to evaluate constructconvergent validity: (a) each measurement item of every construct should have loadings above0.70 (Chin, 1998b) and (b) average variance extracted score for each construct should beabove 0.50 (Fornell & Larcker, 1981). However, we relaxed the 0.70 threshold level to 0.60 dueto the exploratory nature of the study (Wixom & Watson, 2001). Internal consistency reliabilitywas evaluated by composite reliability (Werts et al., 1974), a better estimate than Cronbach’salpha for composite reliability uses actual item loadings to calculate the reliability (Chin &Gopal, 1995). The common cutoff point for composite reliability is 0.70.
The measurements for the five first-order constructs (skills, commitment) in the model weredirectly entered into PLS. Since PLS could not handle second-order constructs directly, weassessed the two second-order constructs (i.e. implementation climate, innovation-values fit)by deploying a two-step procedure proposed by Chin & Gopal (1995). PLS was first run
between first-order factors and the adjacent construct(s) to obtain factor scores for thesefirst-order factors. The generated factor scores are deemed to ‘more accurately [reflect] theunderlying constructs than any of the individual items by accounting for the unique factors anderror measurements that may also affect each item’ (Chin & Gopal, 1995, p. 50). These factorscores are then treated as indicators of second-order constructs and entered into PLS.
4.3.1 Results from construct validation
The first run of factor analysis indicated that a couple of items had weak loadings (e.g. one ofmeasurement items of the construct incentive ‘I am given incentives to use the system’ had afactor loading of 0.2646), and thus these items were discarded. For the second round of PLSanalysis all constructs were shown to possess satisfactory internal consistency reliability andconvergent validity. The lowest composite reliability was 0.78 and the lowest factor loading was0.65, which surpasses the threshold level 0.60 for exploratory studies (Wixom & Watson,2001). Further, the discriminant validity of the constructs was validated since factor loading foreach measurement item is higher than the item’s correlations with other constructs. With thevalid and reliable measurements, we used PLS to generate factor scores, which were thenentered into MARS for additional analysis.
5. MARS ANALYSIS OF THE DATA
5.1 Overview on MARS
MARS is a technique for discovering, evaluating and describing the causal links betweenfactors in any theoretical model. Ordinary regression equations model the relationship betweenoutcome and predictor variables using a single function (e.g. linear and log linear) of thepredictor variables, describing the contribution of each predictor (independent) variable with asingle coefficient. As Hastie & Tibshirani (1990) point out non-linearity in ordinary regressionis captured in higher order terms (x2, x3, etc.) but the coefficients of these terms are estimatedusing the data globally and thus, local features of the true function might not be captured.However, the regression splines (RS) approach models the relationship between outcome andpredictor variables as a piecewise polynomial function f(x) which can be obtained by dividingthe range of each predictor variable into one or more intervals and representing f by a separatepolynomial in each interval (Hastie & Tibshirani, 1990). A regression spline function can beexpressed as a linear combination of piecewise polynomial basis functions (BF) that are joinedtogether smoothly at the knots, where a knot specifies the end of one region of data and thebeginning of another (Steinberg & Martin, 1999). The coefficient of each BF is estimated byminimizing the sum of square errors, which is similar to the estimation process of linearregression, but involving local data for the given region.
Although regression splines have only been used recently in IT studies, RS based analysishas been successfully applied in various fields including software engineering (Briand et al.,
2000a; 2000b), electrochemistry (e.g., Carey & Yee, 1992), geography (Abraham & Steinberg,2001), communication (Ekman & Kubin, 1999), chemical studies (De Veaux et al., 1993;Nguyen-Cong & Rode, 1996), cancer research (Mallick et al., 1997), genetics (York & Eaves,2001), engineering (Jin et al., 2000), geochemistry (Griffin et al., 1997), epidemiology (Kuhnertet al., 2000), finance (Abraham, 2002) and biological sciences (Prasad & Iverson, 2000). Itshould be noted that both regression and regression splines could identify the order ofimportance of the independent variables in a predictive model, and estimate the value of thecoefficient for each independent variable. However, if the impact of an independent variable onthe dependent variable is conditional, then regression splines can identify such conditionswhile regression cannot. Thus, some questions cannot be answered using regression since itdoes not provide means for exploring those questions. On the other hand, regression splinescan provide the means for exploring our research questions in greater depth than would havebeen possible using regression.
5.2 MARS analysis of the Klein-Sorra model
In the following subsections we discuss the MARS analysis. In Subsection 5.2.1 we discuss thecausal links between the independent factors (climate and fit) and the mediators (skill, incen-tives, absence of obstacles and user commitment) of the Klein-Sorra model. In Subsection5.2.2 we discuss the causal links between the mediator factors (skill, incentives, absence ofobstacles and user commitment) and implementation effectiveness. For these analyses weuse the Salford System’s MARS software. For the forward phase modeling we set theparameters as follows:
• Maximum number of BFs: 180
• Minimum number of observations between knots: 1
• Method for determining degrees-of-freedom charge per knot: 10-fold cross validation
A high value for the maximum number of BFs parameter setting reduces the chance thatrelevant BFs might not be identified. While our desire was to generate as many relevant BFsin the forward phase as was permitted by computer memory constraints, we did not want tohave too much redundancy; consequently we set the minimum number of observationsbetween knots to one. If we had more observations in the dataset we might have increased thevalue of this parameter.
5.2.1 Relationships between independent factors and mediators
The Klein and Sorra model postulates that each mediator factor (skill, incentives, absence ofobstacles and user commitment) has a single independent factor (climate or fit) as its predictor.However, since Klein and Sorra did not subject their model to empirical validation, theseassertions about the causal links are nothing more than conjectures. Consequently, wecommence our analysis by examining possible causal links for each of the mediator factorsdescribed in the model. We explore the possible causal links from two perspectives: (a) we
investigate the relationships as they are described in the Klein and Sorra model; and (b) weinvestigate the possibility that either or both of the independent factors could be potentialpredictors of any of the mediator factors. Our analysis for the second case is well supported byMARS as it allows us to retrieve both a model with two predictors of SKILL, namely CLIM andFIT (cf. Table 4.1a), and a model with a single predictor of SKILL, namely FIT (cf. Table 4.1a).We then compare these models from the first perspective (i.e. only one independent factor isa predictor) to the models of the second perspective (i.e. both independent factors arepredictors) based on their R2 values. However, in order to allow for a fair comparison, we selectonly models that have the same number of BF. Our interest here is identifying the better model.The criteria for identifying the better model is as follows: If the adjusted R2 of the two-predictormodel is greater than the adjusted R2 of the single predictor model suggested by Klein andSorra, and if this difference is greater than 0.05 then we take the position that data suggeststhat the given two predictor model provides a better description of the causal relationship thanthe given single predictor model suggested by Klein and Sorra.
Given our interest in not overburdening the reader with all the details of our analysis, we willlimit ourselves to providing a full discussion of our analysis and results for one of theindependent factors (i.e. skill), and have a reduced, but nevertheless substantive discussionfor the other variables. Table 4.1a displays models for predicting the factor Skill. The readermay observe that the R2 (0.48) of the two predictor model (Input: CLIM, FIT; Output: CLIM, FIT)is greater than the R2 (0.25) of the single predictor model suggested by Klein and Sorra (Input:CLIM; Output: CLIM), and this difference is greater than 0.05 (i.e. 0.23) thus we conclude thatthe evidence suggests that the given two-predictor model provides a better description of thecausal relationship than the single predictor model suggested by Klein and Sorra. Interestinglythe R2 (0.42) of the other single predictor model (Input: CLIM, FIT; Output: FIT) is also greater
Table 4.1a. MARS models for predicting SKILL
Target Input Output R2 MARS model: BFs and equation
than the R2 (0.25) of the single predictor model suggested by Klein and Sorra (Input: CLIM;Output: CLIM) even though the former has fewer BF. Together these alternate models stronglysuggest that FIT should also be considered to be a predictor of SKILL. We therefore select thetwo-predictor model (Input: CLIM, FIT; Output: CLIM, FIT) as the ‘best’ of these models.
The MARS results displayed in Table 4.1b suggests that the rate of impact of climate on skillis not uniform, but rather depends on the value of the climate factor. When the climate CLIM)is below -2.420, climate has no impact on skill. Beyond this point, there is positive impact ofclimate on skill but the rate of impact decreases at the point where the factor score of the CLIMis above -2.176. Similarly the results for the impact of the fit on skill suggest that its rate ofimpact is not uniform, and could be positive or zero. The positive impact of fit on skill occurswhen the factor score of fit is above -1.220.
Table 4.2, illustrates our results for the impacts on the target variables incentives, absenceof obstacles and user commitment respectively. In the left column of the table the targetvariable is given followed by the predictor variable, then the region of the regression splinefollowed by the b coefficient and the direction of the impact. The last column of the tableprovides the data on the model fit. The reader will notice that the relationship betweenINCENTIVE and CLIM is complex; in two regions of the regression spline it is positive and theother it is negative. We will discuss this more later.
5.2.2 Relationships between mediators and implementation effectiveness
The Klein and Sorra model postulates that each mediator factor (skill, incentives, absence ofobstacles and user commitment) is a predictor of implementation effectiveness. But as statedearlier none of these conjectures were subjected to empirical validation, consequently, we willnow examine the possible causal links between each of the mediator variables and implemen-tation effectiveness as they are described in the model. We commence our exploration byexamining the impact of each mediator on implementation effectiveness then examine theircollective impact of implementation effectiveness. We then compare the models from the firstcase (i.e. a single mediator as a predictor of implementation effectiveness) to the model of the
second case (i.e. all mediators are predictors) based on their R2 values. The results of ourMARS analysis are four models for predicting implementation effectiveness (cf. Appendix B).An examination of these models points to the four–predictor model postulated by Klein andSorra as the most appropriate with an R2 of 0.46. The basic characteristics of this model aredisplayed in Table 4.3 below. The model illustrates that for each mediator, the correspondingrate of impact is complex and non-uniform, and depends on the value of the mediator.
For example, on different regions of the regression spline defining the relationship betweenskill and implementation effectiveness, the impact of skill on implementation effectiveness ispositive or negative, depending on the value of this factor. The results also suggest absenceof obstacles has no impact on implementation effectiveness when the value of this factor isbelow -1.898 but has a negative impact above this value. The results also show that usercommitment has no impact on implementation effectiveness when the value of this factor is inthe range (-�, -0.738), a positive impact when it is in the range (-0.738, -0.568), a negativeimpact when in the range (-0.568, 0.146), and a positive impact when it is greater than 0.146.
6. CONCLUSION
In this research we set out to conduct a complete and systematic empirical evaluation of theinnovation implementation model postulated by Klein & Sorra (1996) which until now was onlypartially tested. Our reason for selecting this model is that it offered a conceptualization of themanagerial dimension of implementation that is missing from existing models. The results fromour research suggest that the causal relationships postulated by model all exist, but in somecases they are more complex (e.g. non-linear) than suggested by Klein & Sorra (1996).Further, our MARS analysis offers evidence for the existence of additional causal relationshipsnot postulated in the original model. For example, in the original model, implementation climateis believed to be a sole predictor of skills, incentives, and absence of obstacles; but ouranalysis indicates that innovation-values fit also affects these variables as well (see Figure 2).In particular, our findings indicate that both implementation climate and innovation-values fithave strong influence on implementation effectiveness. implementation climate affects imple-mentation effectiveness through enhancing user skills, reducing obstacles, and increasingincentives. Further, innovation-values fit also impacts on implementation effectiveness byincreasing the users’ commitment to the new ERP system. We have also found that innovation-values fit is the best predictor of implementation effectiveness and it also has direct impact onincentives, absence of obstacles and user commitment. Our findings offer three main contri-butions to the IS literature: (a) to ERP implementation theory; (b) to IS implementation; and (c)to ERP implement management practice; which we discuss below.
6.1 Contributions to ERP implementation research
Recent research on predictive models of ERP implementation build on research findings fromvoluntary system usage context, and do not consider the managerial dimension (Kositanurit
et al., 2006; Amoako-Gyampah & Salam, 2004) that is relevant to the context of mandatorysystem usage. However, given the mission-critical role of ERP systems in organizations,managers need to pay careful attention to cultivating and managing factors that could have apositive impact on the implementation. Our study if the Klein and Sorra model suggests two keyfactors, implementation climate and innovation-values fit, that managers can manipulate toachieve effective ERP implementation. An unexpected finding of this study is the influence ofinnovation-values fit on implementation effectiveness. Our MARS analysis suggest that highinnovation-values fit influences users to obtain better skills, perceive less obstacles, and feelmore motivated in using the ERP system. A possible explanation is that if users perceive that theERP will help them solve their work related problems; they internalize the benefits of the ERP.Consequently, they are more open to learning and mastering the system, thus becomingintrinsically motivated. Some studies have found that when system usage is not intrinsicallydriven, system suffers underutilization or users’ intentional sabotage (Markus & Keil, 1994;
Figure 2. Model suggested by empirical exploration.
Brown et al., 2002). As ‘higher level of intrinsic motivation typically leads to willingness to spendmore time on the task’ (Venkatesh, 2000, pp. 348–349), it is understandable that users whoperceive high innovation-values fit will be more skillful and highly motivated in using the ERP.This finding is particularly useful for the implementation where system usage is mandated byorganizations. Therefore, in mandatory system usage contexts, it is particularly important formanagers to help individual users understand how the ERP system fits with the organization’splans for innovating the business, for the expected benefits from the system to be realized.
6.2 Contributions to IS implementation research
Our empirical evaluation of the Klein and Sorra model in an ERP environment is a contributionto the general literature on IS implementation research for the following reasons: First, a keycontribution of the Klein and Sorra model is the formalization of the idea that it is committedadoption that enables organizations to obtain expected benefits when implementing newinformation technologies (Shrednick et al., 1992), and it is crucial to understand the antecedentsof committed adoptions. The Klein and Sorra model recognizes that individual users makecompromises, which could undermine their system usage, and consequently affect organiza-tional performance negatively. By highlighting key determinants of users’ commitment andquality system usage the model offers insights into effective management interventions in ISimplementations. Second, the model introduces unexplored constructs such as implementationclimate and innovation-values fit, which can be applied to various implementation contexts andhelp enrich our understanding of IS adoption and implementation (Agarwal, 2000; Chin &Marcolin, 2001a). Other researchers have also suggested that attitudinal elements (e.g.,enthusiasm) and quality elements (e.g., skillful and consistent use) are also implicated in ISimplementation outcomes (Bailey & Pearson, 1983; Gelderman, 1998; Kositanurit et al., 2006).Third the model offers new possibilities for extending research into the managerial factors thataffect ERP (and more generally IS) implementation success. As stated earlier much of the ISimplementation research has focused on the individual user perspective or the task technology-fit perspective (see Table 1), and there has been an attempt to unify these perspectives (e.g.Venkatesh et al., 2003), little attention has focused on expending beyond these two perspec-tives. As others have pointed out (Parr & Shanks, 2000; Hong & Kim, 2002; Kumar, Maheshwari& Kumar et al., 2003), this is a significant limitation of this IS implementation research. Ourresearch offers the potential of adding a managerial perspective to the existing implementationmodels. Thus making them more complete and better predictors of IS implementation success.Our complete evaluation of the Klein and Sorra model is a step in the development of a moregeneral model of IS implementation that would include managerial factors.
In addition to a better understanding of two key determinants for implementation effective-ness under the mandatory IS usage context, our research contributes to the existing literatureby offering a set of validated constructs for future studies of IT implementation climate. Agarwal(2000) specifically mentioned the lack of studies on managerial interventions such as climate.The constructs developed here can be deployed in other studies on the effect of managementinterventions in various IS implementation contexts. Furthermore, we demonstrate MARS as
an effective tool to not only testing existing causal relationships conjectured in Klein and Sorramodel, but also exploring the existence of additional causal relationships. For any suchrelationship that is suggested by our statistical analysis, we also offered theoretical justificationfor its existence. The result is an extended model that offers a richer description and expla-nation of the impacts of managerial factors on implementation effectiveness that is supportedby both theory and data analysis. It appears to us that even beyond this research project, suchan approach to the validation and development of theory seems to be potentially valuable toresearchers given the fact that it is burdensome and almost impossible for the researcher tospecify and evaluate all relevant hypotheses. Thus, the researcher cannot discover additionalimportant relationships that may exist in the data if they were not explicitly included in the setof hypotheses.
6.3 Contributions to ERP implementation practice
As the first study exploring the two key determinants for implementation effectiveness, ourresearch offers several important implications to practitioners. First, our findings shed light oneffective strategies for management intervention during an ERP implementation. In particular,we indicate that managers need to focus on cultivating a strong climate for ERP implementa-tion and helping users identify the innovation-values fit. The four dimensions of implementationclimate (i.e. goal emphasis, mean emphasis, task support, and reward emphasis) highlightwhat managers can do to foster a strong climate. For example, managers need to emphasizevisions behind the system implementation, explain to users changes to their daily tasks, andoffer rewards and necessary assistance. The three dimensions of innovation-values fit (i.e.,quality, business processes and locatibility) suggest that managers need to stress the benefitsof a new ERP in these areas to help users recognize the value of the new system andinternalize its usage. An ERP implementation without a strong climate fails to equip users withnecessary skills and motivation, and an ERP implementation that is not perceived (innovation-values fit) by users as solving the organization’s problems negatively influences users resultingin users becoming disinterested in the system.
Second, our research findings inform managers and researchers that mandatory systemusage does not necessarily equate to users’ commitment and quality system usage. In fact, wehave found that users realizing the fit like to pay extra effort to master the system, and as aresult, tend to be more skilful, enthusiastic, and committed to the system than do users who failto see the fit. It is reasonable to argue that implementations with committed and skillful usershave a higher possibility to realize the benefits of the ERP than do implementations withdisinterested and unskillful users. Accordingly, in promoting a new ERP system, managersshould not neglect users’ internalization of an enterprise system just because its usage ismandatory. Rather, considering the importance of fit in implementation effectiveness, manag-ers need to direct their efforts at helping users internalize the benefits of a new system. Finally,it is well known that organizations have been experiencing high failure rates in implementingERP systems. Despite the wealth of findings gained through the existing IS implementationliterature, there has been lacking a strong theory conceptualizing about the managerial
dimension in studies of ERP (or IS) implementation effectiveness. Most of the studies of ERP(or IS) implementation have taken the individual user focus. Our research has contributed tothe advancement of understanding how management interventions can positively impact ERPimplementations. Our research offers insights immediately useful for managers and futureresearch on IT implementation, a primary topic in the field of IS.
REFERENCE
Abraham, A. (2002) Analysis of hybrid soft and hard
computing techniques for forex monitoring systems.
In: Proceedings of the 2002 IEEE International Con-
ference on Fuzzy System (IEEE FUZZ’02), 2002
IEEE World Congress on Computational Intelligence,
pp. 1616–1622. IEEE Press, Piscataway, NJ,
USA.
Abraham, A. & Steinberg, D. (2001) Is neural network a
reliable forecaster on Earth? A MARS query. In: Inter-
national Work Conference on Artificial and Natural
Neural Networks, Mira, J. & Prieto, A. (eds.), pp. 679–
686. Springer-Verlag, Berlin, Germany.
Agarwal, R. (2000) Individual acceptance of information
technologies. In: Framing the Domains of IT
Management: Projecting the Future Through the Past,
Zmud, R.W. & Price, M.F. (eds.), pp. 85–104. Pinnaflex
Educational Resources, Cincinnati, OH, USA.
Akkermans, H., Bogerd, P. & Vos, B. (1999) Virtuous and
vicious cycles on the road towards international supply
chain management. International Journal of Operations
and Production Management, 19, 565–581.
Al-Gahtani, S.S. & King, M. (1999) Attitude, satisfaction
and usage: factors contributing to each in the accep-
tance of information technology. Behavior and Informa-
tion Technology, 18, 277–297.
Al-Mudimigh, A., Zairi, M. & Al-Mashari, M. (2001) ERP
software implementation: an integrative framework.
European Journal of Information Systems, 10, 216–
226.
Amoako-Gyampah, K. & Salam, A.F. (2004) An extension
of the technology acceptance model in an ERP imple-
mentation environment. Information and Management,
41, 731–745.
Amrani, R.E., Rowe, F. & Geffroy-Maronnat, B. (2006) The
effects of enterprise resource planning implementation
strategy on cross-functionality. Information Systems
Journal, 16, 79–104.
Armstrong, J.S. & Overton, T.S. (1977) Estimating non-
response bias in mail surveys. Journal of Marketing
Research, 14, 396–402.
Avital, M. & Vandenbosch, M. (2000) SAP Implementation
at Metalica: an organizational drama in two acts. Journal
of Information Technology, 15, 183–194.
Bailey, J.E. & Pearson, S.W. (1983) Development of a tool
for measuring and analyzing computer user satisfaction.
Management Science, 29, 530–545.
Besson, P. & Rowe, F. (2001) ERP project dynamics and