Why PLS-SEM Is Suitable for Complex Modelling? PDF

Production Planning & Control The Management of Operations ISSN: 0953-7287 (Print) 1366-5871 (Online) Journal homepage: www.tandfonline.com/journals/tppc20 Why PLS-SEM is suitable for complex modelling? An empirical illustration in big data analytics quality Shahriar Akter, Samuel Fos...

Production Planning & Control The Management of Operations ISSN: 0953-7287 (Print) 1366-5871 (Online) Journal homepage: www.tandfonline.com/journals/tppc20 Why PLS-SEM is suitable for complex modelling? An empirical illustration in big data analytics quality Shahriar Akter, Samuel Fosso Wamba & Saifullah Dewan To cite this article: Shahriar Akter, Samuel Fosso Wamba & Saifullah Dewan (2017) Why PLS- SEM is suitable for complex modelling? An empirical illustration in big data analytics quality, Production Planning & Control, 28:11-12, 1011-1021, DOI: 10.1080/09537287.2016.1267411 To link to this article: https://doi.org/10.1080/09537287.2016.1267411 Published online: 11 Jul 2017. Submit your article to this journal Article views: 2593 View related articles View Crossmark data Citing articles: 179 View citing articles Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=tppc20 Production Planning & Control, 2017 VOL. 28, NOS. 11–12, 1011–1021 https://doi.org/10.1080/09537287.2016.1267411 Why PLS-SEM is suitable for complex modelling? An empirical illustration in big data analytics quality Shahriar Aktera, Samuel Fosso Wambab and Saifullah Dewanc a Sydney Business School, University of Wollongong, Australia; bToulouse Business School, Toulouse University, Toulouse, France; cFaculty of Business, Government and Law, School of Information Systems & Accounting, University of Canberra, Canberra, Australia ABSTRACT ARTICLE HISTORY The emergence of multivariate analysis techniques transforms empirical validation of theoretical concepts Received 3 June 2016 in social science and business research. In this context, structural equation modelling (SEM) has emerged Accepted 28 November 2016 as a powerful tool to estimate conceptual models linking two or more latent constructs. This paper shows KEYWORDS the suitability of the partial least squares (PLS) approach to SEM (PLS-SEM) in estimating a complex model PLS-SEM; big data; big data drawing on the philosophy of verisimilitude and the methodology of soft modelling assumptions. The analytics quality; business results confirm the utility of PLS-SEM as a promising tool to estimate a complex, hierarchical model in the value; satisfaction domain of big data analytics quality. 1. Introduction estimating such a complex-hierarchical model by eliminating the ambiguity of incorrect solutions (Wetzels, Odekerken- Foresight of phenomenon and power over them depend on knowl- Schroder, and Van Oppen 2009; Becker, Klein, and Wetzels 2012). edge of their sequences, and not upon any notion we may have formed respecting their origin or inmost nature. Mill (1865, p. 266) However, some researchers have questioned PLS-SEM’s rigor (e.g. Guide and Ketokivi 2015; Rönkkö et al. 2016) despite its Quantitative research has made an enormous impact on social well-established roots in management information systems science and business research through its positivist epistemo- (Chin and Gopal 1995; Chin 1998b; Chin, Marcolin, and Newsted logical belief since John Stuart Mill and the nineteenth-century 2003; Marcoulides, Chin, and Saunders 2009; Ringle, Sarstedt, experimental researchers. This impact has gained a momentum and Straub 2012a), strategic management (Bentler and Huang by satisfying the need for causal modelling and empirical vali- 2014; Sarstedt, Ringle, and Hair 2014) and marketing (Fornell dation of theories to explain complex concepts (Blalock 1964; and Bookstein 1982; Chin, Peterson, and Brown 2008; Hair Bagozzi 1980; Huber and McCann 1982; Sawyer and Peter 1983; et al. 2012). Thus, in an effort to illuminate the rigor of PLS- Iacobucci and Hopkins 1992; Stafford 2011; Hair et al. 2012). In SEM in estimating a complex model, we validate a hierarchical, this line of development, structural equation modelling (SEM) reflective–formative model in the context of big data analytics has emerged as a powerful multivariate analysis technique quality (BDAQ). over the last four decades combining the features of the prin- Big data have emerged as the new oil, new soil, the next man- cipal components and the regression analysis (Hair, Ringle, agement revolution (McAfee and Brynjolfsson 2012b) and the and Sarstedt 2012). Between the two approaches in SEM, the ultimate force behind ‘transforming management theory and covariance-based approach (CBSEM) is useful to confirm the- practice (George, Haas, and Pentland 2014). The extant literature oretically established relationships; however, this technique shows that more than 91% of Fortune 1000 companies have has distributional constraints (multivariate normality of the embraced big data analytics (BDA; Kiron, Prentice, and Ferguson observed indicators) in estimating a large model (Fornell and 2014) and achieved 5–6% higher growth in firm performance than Bookstein 1982; Chin 1998b; Hair, Ringle, and Sarstedt 2011). competitors (Akter and Fosso Wamba 2016). Despite its skyrock- As a result, the focus of CBSEM mostly is on small conceptual eting optimism, many firms still struggle to capitalise on BDA models, which results into hindering the development and to derive quality insights. Competitive advantage from BDA is validation of large, complex models (Chin, Peterson, and waning as managers grapple to understand the complexity of Brown 2008). analytics quality in the emerging data economy (Ransbotham, This study refers a complex model to a large-hierarchical Kiron, and Prentice 2016). Thus, for empirical illustration, the study model that consists of 10 or more constructs and 50 or first presents an overview of PLS-SEM applications to estimate a more items (Chin 2010). PLS-SEM (Wold 1975; Lohmoller complex model in the context of BDA and prediction-oriented 1989; Tenenhaus and Tenenhaus 2011) gains prominence in CONTACT Samuel Fosso Wamba [email protected] © 2017 Informa UK Limited, trading as Taylor & Francis Group 1012  S. AKTER ET AL. Table 1. Model parameters using PLS-SEM in PPC (2010–2015). Studies Sample size Constructs Items Context Clegg, Gholami and Omurgonulsen (2013)) 183 5 30 Quality management and performance Ren et al. (2015) 110 6 22 Inter-organisational value cocreation in supply chain Lee et al. (2015) 119 3 16 Greening the supplier, environmental performance and competitive advantage. Kotzab et al. (2015) 274 4 38 Supply chain management Ahmad and Mehmood (2016) 288 9 62 Enterprise systems and performance of future city logistics Average 194.8 5.4 33.6 analysis (Gefen, Straub, and Rigdon 2011; Rigdon 2014). We Sarstedt 2015), overall model fit using bootstrapping (Dijkstra define BDA as an integrated approach to collect and process and Henseler 2015a), interaction effects (Chin, Marcolin, and big data in order to provide actionable insights for managerial Newsted 2003; Fassott, Henseler, and Coelho 2016) and model decision-making. Second, the study applies PLS-SEM to develop specification (Sarstedt et al. 2016) providing a proof that it is able and validate a complex BDAQ model and its impact on business to define latent variables scores with well-defined statistical rela- value (BVAL) and BDA satisfaction (BDAS) in a nomological net- tions among them. work. We define BDAQ as user perceived analytics quality model One of the main characteristics of the PLS-SEM is that it is that measures overall excellence or superiority of the BDA plat- able to estimate a model with a large number of latent varia- form. Conceptually, this study extends quality modelling in big ble and indicators even with a small sample size (Chin, Peterson, data using the resource-based theory (RBT) and methodolog- and Brown 2008). For complex models, PLS-SEM ensures factor ically, it presents the rigor of PLS-SEM as the ultimate tool for determinacy by directly estimating latent variable scores, factor complex modelling. The remainder of the paper is organised as identification by introducing flexible residual covariance structure follows: Section 2 focuses on literature review and the conceptual and above all, robust prediction in the context of small sample size, model; Section 3 discusses the research methodology; Section 4 asymmetric distribution and interdependent observations (Chin highlights the findings based on empirical illustration; Section 1998a, 1998b; Wetzels, Odekerken-Schroder, and Van Oppen 5 discusses the techniques for assessing a complex model and 2009). These distinctive methodological features make PLS-SEM finally, Section 6 concludes the paper with limitations and future a possible alternative to the more popular CBSEM approaches for research directions. complex modelling (Henseler, Ringle, and Sinkovics 2009; Hair et al. 2012). As such, PLS-SEM is more suitable in a complex setting to 2. Literature review validate large-hierarchical models by providing robust solutions (Chin, Peterson, and Brown 2008). This is echoed by Chin (2010, 2.1. PLS-SEM p. 661), ‘It is under this backdrop of high complexity that PLS, It struck me that it might be possible to estimate models with the regardless of whether applied under a strong substantive and same arrow scheme by an appropriate generalisation of my LS algo- theoretical context or limited/exploratory conditions, comes to rithms for principal components and canonical correlations. (Wold the fore relative to CBSEM’. 1982, p. 200) There is no doubt that PLS-SEM has become very popular for 2.2. Complex models and PLS-SEM survey research in recent years since its introduction in 1966 by Herman Wold. The development of PLS-SEM is largely driven by its Truth, Existence, Knowledge, Causality, Identity, Goodness: these advantages in distributional assumptions, the absence of factor are the principal notions which philosophers examine. Intelligent indeterminacy and models with more parameters than observa- persons normally have thoughtful and useful lives without paus- ing to look into these notions and into the connections between tions (Dijkstra and Henseler 2015a). The PLS-SEM is regarded as them. Once one starts to look into them, it is difficult to stop. Stuart a variance-based approach to SEM (Chin, Marcolin, and Newsted Hampshire (Pyke 2011) 2003; Tenenhaus 2008) becomes appreciated for its ability to esti- The extant literature in social science and relevant research mate both composites and factors (Henseler, Hubona, and Ray philosophies have contributed to the field of complex modelling 2016). As an alternative to CBSEM, PLS-SEM was developed to using the philosophy of verisimilitude (i.e. trust likeness or near- estimate complex relationships and emphasise prediction while ness to the truth). For example, Meehl (1990) states that most simultaneously relaxing the demands on data and specification models struggle to capture reality and suffer imperfection due to of relationships (Chin, Peterson, and Brown 2008; Dijkstra 2010). the imbalance between incompleteness and falseness. Whereas Differently from CBSEM, PLS-SEM aims at estimating latent varia- falseness represents the contradictions between the research ble proxies (also called latent variable scores) according to a pos- model and the real world, incompleteness focuses on the ability tulated model via an iterative sequence of ordinary least-squares to capture complex reality. Although these two philosophies play regressions (Wold 1975, 1985). Recent publications on PLS-SEM an important role in estimating reality, using hierarchical modelling (Wetzels, Odekerken-Schroder, and Most SEM studies seem to focus on the falsity of a model as opposed Van Oppen 2009; Akter, D’Ambra, and Ray 2010, 2011; Becker, to its completeness. In part because of algorithmic constraints, few Klein, and Wetzels 2012), consistent PLS (PLSc) for factor models SEM models are very complex (i.e. have a large number of latent (Dijkstra and Henseler 2015b), heterotrait-monotrait ratio of cor- variables). Emphasis on model fit tends to restrict researchers to relations (HTMT) for discriminant validity (Henseler, Ringle, and testing relatively elementary models representing either a simplistic  PRODUCTION PLANNING & CONTROL  1013 theory or a narrow slice of a more complex theoretical domain. (Chin, PPC used an average number of 194.8 samples. Although small Peterson, and Brown 2008, p. 294). sample size is the most frequently cited reason for using PLS-SEM, The philosophy of verisimilitude urges to recognise that ‘sci- the review indicates that operations researchers used relatively entific theories are never impeccably veridical in all aspects’ large sample size which is clearly immune to threats from data (Rozeboom 2005, p. 1335) and thus, practical theory adjudica- inadequacies (Ringle, Sarstedt, and Straub 2012a). Overall, this tion should concentrate more on how a research model is true review reflects the flexibility of PLS-SEM in handling large models and to what extent it is true rather than whether a research with fewer restrictions. model is true or not. In exploring CBSEM, Shah and Goldstein (2006) identified an 2.3. BDA quality average of 4.4 latent variables and a mean of 14 indicators per model in a review of 93 articles. By comparison, in exploring PLS- Big data refer to the massive amount of structured and unstruc- based models, Ringle, Sarstedt, and Straub (2012a) identified an tured data, which has four characteristics, that is, volume (i.e. average of 8.12 latent variables and 27.42 indicators per model in huge quantity), variety (i.e. number, text, image, voice and a review of 65 studies published in MIS Quarterly. Hair et al. (2012) video), velocity (i.e. speed) and veracity (i.e. reliability of data) identified an average of 29.55 indicators per PLS path model in (Samuel et al. 2015; Akter et al. 2016). According to Sanders 204 studies of top 30 marketing journals. These results highlight (2016, p. 28), ‘Big data without analytics is just a massive amount the suitability of PLS-SEM as a tool to estimate a large, complex of data. Analytics without big data are simply mathematical and model (Chin, Peterson, and Brown 2008). In a similar spirit, Ringle, statistical tools and applications’. We define BDA as an integrated Sarstedt, and Straub (2012b) comment data collection and analysis process to provide solid insights for … prior studies appearing in scholarly journals (e.g., Reinartz, managerial decision-making (Akter and Fosso Wamba 2016). Haenlein, and Henseler 2009) – including those more critical of the Although there is high adoption of BDA in recent years to obtain PLS-SEM method (e.g., Lu et al. 2011) – indicate that PLS-SEM over- competitive advantage, many companies face enormous chal- comes problematic model identification issues and that it is a power- lenges to derive quality insights from data (Ransbotham, Kiron, ful method to analyze complex models using smaller samples. and Prentice 2016). We define BDAQ as the overall excellence or Dijkstra and Henseler (2015a, p. 10) support that PLS-SEM has superiority of BDA platform perceived by its users. BDAQ also ‘the possibility of estimating models having more variables or refers to the distinctive attribute of the overall analytics platform parameters than Observations’. Although few studies in CBSEM to produce valuable insights for business (Ji-fan Ren et al. 2016). focused on developing a large model using small sample size; The extant literature shows that the quality of technology and these models are restricted by three items per LV to achieve information determine the extent of business value in big data goodness of fit (e.g. Marsh, Hau, and Wen 2004; Barendse, Oort, environment. In this regard, Barton and Court (2012) argue that and Garst 2010). This constraint is criticised by MacCallum and both technology and information quality work as an ecosystem Austin (2000) as it obstructs capturing the complexity of an in producing solid insights for managers. Technology quality empirical phenomenon. In this context, Blalock (1979, p. 881) refers to the quality of the analytics platform that is reflected states, ‘reality is sufficiently complex that we will need theories in system reliability, system adaptability, system integration that contain upward of 50 variables if we wish to disentangle the and system privacy (Nelson, Todd, and Wixom 2005; Davenport, effects of numerous exogenous and endogenous variables on Barth, and Bean 2012). On the other hand, information quality the diversity of dependent variables that interest us’. He further represents the quality of data driven insights in terms of cur- adds that there is a natural imbalance between generalisability rency, format, accuracy and completeness (Nelson, Todd, and and parsimony in developing models, so ‘parsimony’ could be Wixom 2005). Wixom, Yen, and Relich (2013) show that the qual- sacrificed in building complex models to describe more diverse ity of technology and information in big data environment influ- settings and populations. In this case, PLS-SEM enjoys certain ence business value, which refers to the strategic benefits for advantages in estimating complex models because of its flex- firms. In addition to business value, scholars (Davenport 2006; ible iterative algorithm and the soft modelling assumptions. McAfee and Brynjolfsson 2012a) also identify that the quality Lohmoller (1989, p. 64) comments, ‘It is not the concepts nor the parameters influence user satisfaction, which determines sus- models more the estimation techniques which are “soft”, only tainability of the analytics platform. the distributional assumptions’. Because of its flexibility in mod- elling both composites and factors, McDonald (1996) identifies PLS as a sophisticated multivariate analysis platform, whereas 3. Research model for empirical illustration Hair, Ringle, and Sarstedt (2011) label it as a silver bullet. As such, Drawing the RBT, the study views analytics quality as an analyt- scholars across disciplines (e.g. Fornell and Bookstein 1982; ics resource only if it is rare and costly to imitate (Ray, Muhanna, Hulland 1999; Chin, Peterson, and Brown 2008; Chin 2010; Hair and Barney 2005). The RBT of BDA argues that BVAL and BDAS et al. 2012; Sarstedt et al. 2016; Henseler, Hubona, and Ray 2016) depend on the quality of resources that are valuable, rare, inim- put forward PLS-SEM as tool of trade for survey research to cap- itable and properly organised. The RBT also focuses on complex ture complexity in models. connections among the heterogeneous resources, such as sys- Our review examines all empirical studies using PLS-SEM tem and information quality, to examine BVAL and BDAS. Thus, published in the Production Planning & Control (PPC) journal from using RBT as a theoretical foundation, the study examined com- 2010 to 2015 indicates an average number of 5.4 constructs and monly found dimensions of BDA that influence quality percep- 33.6 indicators per PLS-SEM model to embrace the complexity tion. The review identified two primary dimensions of BDAQ, that in capturing reality. Table 1 also shows that PLS-SEM studies in is, technology quality and information quality (Ji-fan Ren et al. 1014  S. AKTER ET AL. System Reliability System Adaptability Technology BDA Quality Satisfaction System (BDAS) Integration H3 (+) System Privacy H2 (+) Big Data Analytics Quality (BDAQ) Completeness H1 (+) Business Value (BVAL) Accuracy Information Quality H4 (+) Format Currency Figure 1. Big data analytics quality model. 2016). BDAQ also emerged as a hierarchical construct through- The extant literature identifies that the excellence of BDAQ has out our review and theoretical exploration, which consists of a significant positive impact on business value, which ultimately two primary dimensions and eight subdimensions as shown drives satisfaction of BDA users (Ji-fan Ren et al. 2016). This study in Figure 1 (Davenport and Harris 2007; Davenport, Barth, and argues that the assessment of BDAQ results in an affective or emo- Bean 2012; McAfee and Brynjolfsson 2012b; Samuel et al. 2015). tional response, such as BVAL and BDAS. In this regard, Golder, Therefore, based on the systematic literature review, this study Mitra, and Moorman (2012) state that ‘[p]­ositive quality discon- identifies BDAQ as a complex construct model because of its firmation increases satisfaction; negative quality disconfirmation large number of dimensions and subdimensions under multiple decreases satisfaction’. Thus, this study explores the link between hierarchies (see Figure 1). quality-value-satisfaction and posits that: We specify the proposed BDAQ model as a higher-order, H2: Business value (BVAL) has a significant positive impact on user reflective–formative model as the first-order dimensions are satisfaction (BDAS). reflective (Mode A) and the higher-order dimensions are forma- H3: BDAQ has a significant positive impact on satisfaction. tive (Mode B) (Chin 2010; Ringle, Sarstedt, and Straub 2012b). We define the proposed quality model as a complex model because it The study identifies BVAL at the heart big data research involves large number of constructs and indicators under multiple because it will be directly influenced by BDAQ (Wixom, Yen, levels and dimensions (Law and Wong 1999; Edwards 2001; Jarvis, and Relich 2013). BVAL is identified as a mediator in the study MacKenzie, and Podsakoff 2003; Netemeyer, Bearden, and Sharma because, first, BDAQ (predictor) influences BVAL (mediator); sec- 2003; MacKenzie, Podsakoff, and Jarvis 2005). As part of embed- ond, BVAL influences BDAS and, finally, BDAQ influences BDAS (i.e. ding the higher-order quality model in a causal network, the the dependent variable) without any influence of the mediator study models it with criterion variables, such as, business value (Baron and Kenny 1986). Thus, the mediating role of BVAL in BDA and satisfaction. We define ‘satisfaction’ as the overall attitudinal research is important to explore: response by the big data analysts toward BDA and ‘business value’ H4: Business value, as a mediator, influences the relationship between BDAQ and satisfaction. as the degree of perceived benefits to the organisation at a stra- tegic level, e.g., competitive advantage (Wixom, Yen, and Relich 2013). The impact of BDAQ on BVAL is a dominant concern in big 4. Methodology data environment (Wixom, Yen, and Relich 2013). The significance of the association between BDAQ and BVAL was highlighted by 4.1. Data collection and sampling the extant literature (Lavalle et al. 2011; Wixom, Yen, and Relich Table 2 presents the operational definitions of all the dimensions 2013; Ji-fan Ren et al. 2016). Thus, we postulate that: and subdimensions of BDAQ. All the scales to measure BDAQ H1: BDAQ has a significant positive impact on business value (BVAL). were drawn from prior literature and adapted to suit the context.  PRODUCTION PLANNING & CONTROL  1015 Table 2. Construct and definitions. Construct and definitions Sources BDA technology quality is defined as systems reliability, system adaptability, system integration and system Nelson, Todd and Wixom (2005) and Parasuraman, privacy. System reliability refers to the degree to which the BDA is reliable over time; System adaptability refers Zeithaml and Berry (2005) to degree to which the BDA can adapt to a variety of user needs and changing conditions; system integration refers to the ability to integrate various sources of data to produce meaningful insights; and finally, system privacy refers to the degree to which the BDA system is safe and protects user information. BDA Information quality is defined as the completeness, accuracy, format and currency of information produced Wixom and Todd (2005) by BDA. Completeness indicates the extent to which the user perceives that BDA provide all the necessary information; accuracy focuses on the perceived correctness of information; format refers to the perception of how well the information is presented; and, finally, currency refers to the user’s perception of the extent to which the information is up to date. BDA Business value is defined as the strategic value refers to the degree of perceived benefits to the organisa- Gregor et al. (2006) tion at a strategic level, e.g. competitive advantage. BDA satisfaction refers to the users’ affect with (or, feelings about) BDAQ. Spreng, MacKenzie, and Olshavsky (1996) The study measured all the first-order constructs using 7-point Table 3 presents measurement properties of the first-order Likert scale (i.e. strongly disagree – strongly agree) except sat- model in order to examine reliability, convergent validity and isfaction, which was measured using a 7-point semantic dif- discriminant validity. The key psychometric properties including ferential scale (i.e. very dissatisfied–very satisfied). Data were loadings of manifest variables, Cronbach’s alphas, composite reli- collected from the 302 big data analysts in the United States abilities (CRs) and average variance extracted (AVEs) have con- and France using a leading market research firm. Specifically, the firmed scale reliability (Chin 2010) by successfully meeting the sample includes 150 valid responses from the France and 152 threshold of 0.7, 0.7, 0.8 and 0.5, respectively. The convergent from the United States. validity was ensured as all the items load much higher on their corresponding constructs than on other constructs. The study also calculated the square root of the AVE in the Table 4 to ensure 4.2. Data analysis discriminant validity (Fornell and Larcker 1981). As such, the find- The study applied PLS-SEM to estimate a hierarchical, reflective– ings of the measurement model provided adequate evidence of formative BDAQ construct in order to avoid the limitations of reliability, convergent validity and discriminant validity. These CBSEM regarding improper solutions or empirical under iden- findings provide the confidence to confirm all the hypothesised tification (Wetzels 2009; Chin 2010). Due to the soft modelling relationships of the structural model. assumptions, application of PLS-SEM helps in avoiding positively In Table 5, this study shows the findings of the complex-high- biased model fit indices for our large-complex model (Chin and er-order BDAQ model. The study estimated the third-order BDAQ Newsted 1999; Hair, Ringle, and Sarstedt 2011; Hair et al. 2012), construct, which consists of 2 second-order formative constructs which represents 13 latent constructs (i.e. 8 first-order + 2 sec- (technology quality and information quality) representing 24 ond-order + 1 third-order + 2 outcome constructs) and 82 items (3 + 3 + 3 + 3 + 3 + 3 + 3 + 3) valid items. Since both the second- (24 + 24 + 24 + 6 + 4). Indeed, this is a challenging context for and third-order constructs are formative; thus, we estimated the CBSEM-based studies ‘due to the algorithmic nature requiring weights of items of higher-order constructs that are significant at inverting of matrices’ (Chin 2010, p. 661). Therefore, the study p < 0.05. The findings show minimum evidence of collinearity as favoured PLS-SEM to remove the uncertainty of inadmissible the variance inflation factor of all items was less than 5. solutions for a large, complex model both in exploratory and Table 5 shows that the degree of explained variance of the in confirmatory settings (Hulland, Ryan, and Rayner 2010; Hair, third-order BDAQ construct is explained by second-order technol- Ringle, and Sarstedt 2011). In this context, Chin (2010, p. 660) ogy quality (58%) and information quality (49%). Accordingly, sec- states that ‘[i]­t should not be construed that PLS is not appro- ond-order constructs are explained by its first-order dimensions, priate in a confirmatory sense nor in well researched domains’. such as information quality is explained by completeness (27%), The study uses the approach of repeated indicators suggested currency (37%), format (26%) and accuracy (24%). The findings by Wold (cf. Lohmoller 1989, 130–133), Akter, D’Ambra, and Ray ensure that all the paths are significant at p < 0.001 both at the (2011) and Becker, Klein, and Wetzels (2012) in estimating the first-order and higher-order level. The study analyses the impli- hierarchical BDAQ model. cations of these results in the discussion section. 5. Findings 5.2. The structural model 5.1. The measurement model This study confirms the nomological validity of BDAQ model by examining its relationship with BVAL and BDAS. In order to The study used SmartPLS 3.0 (Ringle, Wende, and Becker 2015) assess the nomological validity, the study uses BVAL and BDAS to estimate the measurement properties of the complex, hierar- with the hierarchical BDAQ construct. In the main effects model chical BDAQ model. Specifically, the study applied nonparamet- (Figure 2), the findings provide a standardised beta of 0.797, ric bootstrapping (Efron and Tibshirani 1993; Chin 2010) with 0.175 and 0.681, respectively, from BDAQ-BVAL, BVAL-BDAS and 5000 replications to obtain the standard errors of the estimates BDAQ-BDAS. Path coefficients are significant at p < 0.05, thus (Hair et al. 2013) and a path weighting scheme for the inside support H1, H2 and H3 (see Table 6). These results also confirm approximation. 1016  S. AKTER ET AL. Table 3. Psychometric properties for first-order constructs. Dimensions Subdimensions Items Loadings Alpha CR AVE Technology Quality System reliability The system operates reliably for the analytics 0.928 0.952 0. 952 0.868 The system performs reliably for the analytics 0.933 The operation of the system is dependable for the analytics 0.935 System adaptability The system can be adapted to meet a variety of analytics needs 0.907 0.933 0.933.823 The system can flexibly adjust to new demands or conditions during 0.918 analytics The system is flexible in addressing needs as they arise during the 0.897 analytics System integration The system effectively integrates data from different areas of the 0.923 0.945 0.945 0.852 company The system pulls together data that used to come from different 0.908 places in the company The system effectively combines different types of data from all areas 0.938 of the company System privacy The system protects information about personal issues 0.912 0.948 0. 984 0.859 This system protects information about personal identity 0.942 The system offers a meaningful guarantee that it will not share 0.926 private information Information quality Completeness The business analytics used: 0.885 0.903 0.904 0.759 ____ provides a complete set of information 0.895 ____ produces comprehensive information 0.832 ____ provides all the information needed Currency ____ provides the most recent information 0.919 0.932 0.932 0.821 ____ produces the most current information 0.776 ____ always provides up-to-date information 0.883 Format The information provided by the analytics is ____ well formatted 0.936 0.952 0.952 0.869 The information provided by the analytics is ____ well laid out 0.933 The information provided by the analytics is ____ clearly presented 0.928 on the screen Accuracy The business analytics used: 0.913 0.894 0.896 0.742 ____ produces correct information 0.886 ____ provides few errors in the information 0.919 ____ provides accurate information BDA satisfaction (BDAS) I am satisfied with my use of BDA service 0.896 0.929 0.929 0.766 I am contented with my use of BDA service 0.879 I am pleased with my use of BDA service 0.835 I am delighted with my use of BDA service 0.890 Business value (BVAL) The BDA used by the firm: 0.847 0.937 0.937 0.712 Creates competitive advantage 0.859 Aligns analytics with business strategy 0.813 Establishes useful links with other organisations 0.909 Enables quicker response to change 0.834 Improves customer relations 0.795 Provides better products or services to customers Table 4. Mean, standard deviation (SD) and correlations of the latent variables for the first order constructs.a Constructs Mean SD SYRE SYAD SYIN SYPR COMP CURR FORM ACCU BVAL BDAS System reliability (SYRE) 4.894 1.008 0.932a System adaptability (SYAD) 4.858 1.145 0.441 0.907a System integration (SYIN) 5.045 1.139 0.343 0.451 0.923a System privacy (SYPR) 5.138 1.167 0.503 0.563 0.426 0.927a Completeness (COMP) 4.772 1.117 0.477 0.559 0.553 0.565 0.871a Currency (CURR) 5.084 1.081 0.684 0.563 0.659 0.523 0.601 0.906a Format (FORM) 5.073 1.127 0.540 0.516 0.429 0.578 0.532 0.658 0.932a Accuracy (ACCU) 4.997 1.047 0.523 0.465 0.585 0.487 0.513 0.664 0.534 0.861a Business value (BVAL) 5.035 1.018 0.446 0.441 0.559 0.518 0.538 0.621 0.542 0.510 0.844a BDA satisfaction (BDAS) 4.897 1.022 0.558 0.533 0.381 0.591 0.590 0.601 0.537 0.529 0.414 0.875a a Square root of AVE on the diagonal. the significance of BVAL as a partial mediator between BDAQ PLS-SEM successfully validated the research model, this study and BDAS, which explains about 17% (0.797 × 0.175/0.797 × 0. investigated the significance of model fit, predictive relevance 175 + 0.681) of the total effect of BDAQ on BDAS. and unobserved heterogeneity to establish further rigor. Model fit is essential to establish conjectures (Tenenhaus et al. 2005; Henseler, Hubona, and Ray 2016), predictive relevance (Q2) is 5.3. An assessment of the PLS-SEM-based complex model critical to check the extent of reproduction of observed values This study applied PLS-SEM in estimating the complex, hier- and finally, unobserved heterogeneity is important for identi- archical research model with mediating effects. Although fying significant heterogeneity in data which can lead to bias  PRODUCTION PLANNING & CONTROL  1017 Table 5. Assessment of the higher-order, reflective-formative model. 1988, 1989; Barclay, Higgins, and Thompson 1995; Chin 1998a, Third-order formative Relationships with sec- 2010; Hulland 1999; Dijkstra 2010). construct ond-order dimensions β t-stat This study explained in detail the methodological gestalt of BDAQ Technology quality 0.583 8.278 complex modelling using PLS-SEM in order to demonstrate why Information quality 0.487 7.122 this study is a leap forward. Since the soft modelling assumptions Second-order formative Relationships with first-or- β t-stat of PLS-SEM facilitate developing complex models both in theo- constructs der dimensions retical and in applied research contexts, it has immense potential Technology quality System reliability 0.394 4.909 to capture the complexity of causal modelling. Indeed, PLS-SEM System adaptability 0.274 3.289 is best suited for complex models especially when the primary System integration 0.307 3.827 System privacy 0.158 2.493 objective is prediction, the focus is on explaining variance of Information quality Completeness 0.268 4.086 large number of variables and the sample size is small (Hulland, Currency 0.374 4.388 Ryan, and Rayner 2010). For example, the complex model in our Format 0.259 3.078 Accuracy 0.235 2.978 study has robustly explained variances of 13 latent variables, 82 (24 + 24 + 24 + 6 + 4) indicators with 302 samples. The applica- tion of PLS-SEM makes it possible to extend the theoretical and managerial contributions of the study. Theoretically, the study BDA contributes in several ways. First, the study offers a conceptual Satisfaction (R2 = 0.685) framework integrating RBT and BDAQ in order to provide a the- H3 (0.681) oretical synergy to work with big data against the backdrop of analytics studies that show mixed results in business value cre- ation. Second, extending the RBT, the study proposes a quality H2 (0.175) Big Data dominant logic in BDA research with two dimensions (i.e. tech- Analytics Quality nology quality and information quality) and eight subdimensions (BDAQ) (i.e. system reliability, system adaptability, system integration and system privacy as subdimensions of technology quality, and H1 (0.797) completeness, accuracy, format and currency as subdimensions Business of information quality). Third, the study has identified a full, yet Value (R2 = 0.635) tightly entangled, set of dimensions that help predict the qual- ity of BDA and their effects on business value and satisfaction. Finally, the research presents rigour by conceptualising all the H4 (+) dimensions, developing their scales and estimating the mediating Figure 2. The structural model. effects of BVAL on BDAQ-BDAS link. Highlighting the importance of mediating effects, Iacobucci (2009, p. 673) states ‘If mediation clarifies the conceptual picture somewhat, with the insertion of Table 6. Results of the structural model. just one new construct – the mediator – imagine how much richer Paths Path coefficients Standard error t statistic the theorising might be if researchers tried to formulate and test BDAQ → BVAL 0.797 0.028 28.504 even more complex nomological networks’. Practically, the pro- BVAL → BDAS 0.175 0.073 2.403 BDAQ → BDAS 0.681 0.065 10.454 posed BDAQ model presents practitioners with an instrument for investigating a holistic quality analysis and design of analytics. The results highlight that only having a sound technology plat- parameter estimates and invalid statistical conclusions (Esposito form is not adequate to ensure solid insight from analytics plat- Vinzi et al. 2008). The findings show that the study achieved an form. Although firms invest lot of resources to improve analytics adequate GoF value (>0.36), standardised root mean square platform, sophisticated insights explaining how BDA platform can residual (0.50). improve BVAL and BDAS deserve equal attention. The findings clearly show how to tap into BDAQ to influence business out- comes. Practitioners now can have a coordinated focus to ensure 6. Discussion the simultaneous quality of technology and information. Overall, The study answered the key question posed by the research these findings provide the blueprint to identify and improve a whether PLS-SEM can estimate a complex model. The findings specific quality dimension of BDA at different levels. illustrate that PLS-SEM entail the flexibility of soft modelling assumptions in validating a reflective–formative, hierarchical 7. Limitations and future research directions quality model in BDA research. According to Jacoby (1978, p. 91), ‘we live in a complex, multivariate world [and that] studying the Although PLS-SEM is a preferred technique for complex model- impact of one or two variables in isolation, would seem …… rel- ling in social science and business research, there are few chal- atively artificial and inconsequential’. Thus, there is huge possi- lenges that need to be addressed in order to establish it as an bility of PLS-SEM-based research in complex, predictive settings, esoteric method. For example, first, PLS-SEM should have the such as the big data environment (Wold 1980, 1985; Lohmoller flexibility of imposing constraints on model coefficients (weights, 1018  S. AKTER ET AL. loadings, path coefficients) in order to specify any information Notes on contributors or conjectures available a priori in estimating model parameters Shahriar Akter is a Senior Lecturer in the Sydney Business (Esposito Vinzi et al. 2010). Second, this approach should allow School, University of Wollongong, Australia. Shahriar was specific treatment of categorical variables, outliers, non-line- awarded his PhD from the Australian School of Business, arity and mutual causality both in measurement and in struc- University of New South Wales. As part of his doctoral tural models, which can lead toward estimation of interaction program, Shahriar received his methodological training from the Oxford Internet Institute, University of Oxford. and quadratic effects. Third, application of PLS methods should He has published in the leading international journals clarify both observed heterogeneity (Sarstedt, Henseler, and including Information & Management, International Ringle 2011) and unobserved heterogeneity departing from the Journal of Production Economics, Journal of Business assumption that all individuals act in a similar fashion (Becker Research, Electronic Markets, International Journal of Operations and et al. 2013). Fourth, PLS studies should pay serious attention to Production Management, Journal of the American Society for Information Science & Technology, International Journal of Production Research, Behavior & an adequate statistical power and representativeness of data IT, Production Planning & Control, Journal of Selected Areas in Communication, in the context of inferential statistics using small sample size Business Process Management Journal etc. Shahriar’s research interests (Marcoulides, Chin, and Saunders 2009). Fifth, PLS models can include service systems evaluation, e-books, business/marketing analytics take into account feedback loops and model fit indices in order and complex modelling using PLS. to leverage its application as an SEM tool (Henseler, Hubona, and Samuel Fosso Wamba is Full Professor at Toulouse Ray 2016). Finally, future research can explore non-linear effects, Business School, France. Prior, he was Associate Professor parameter bias under Mode A and B, population type and data at Rouen Business School, Rouen, France and Senior lec- turer at the School of Information Systems & Technology conditions in the domain of complex modelling (Sarstedt et al. (SISAT), University of Wollongong, Australia. He earned 2016). The limitations mentioned in the study represent excit- an MSc in mathematics, from the University of Sherbrooke ing avenues for PLS-SEM researchers to establish it as a power- in Canada, an MSc in e-commerce from HEC Montreal, ful platform for complex modelling. Drawing on the arguments Canada, and a Ph.D. in industrial engineering, from the of Hair et al. (2011, 2012) and Reinartz, Haenlein, and Henseler Polytechnic School of Montreal, Canada. His current research focuses on business value of IT, inter-organisational system (e.g. (2009), Ringle et al. (2012b, vii) state that ‘… PLS-SEM can indeed RFID technology) adoption and use, e-government, supply chain manage- be a ‘silver bullet’ in certain research situations (e.g. when models ment, electronic commerce, mobile commerce, big data, business analytics are relatively complex and representative sets of data are rather and Internet of Things. He has published papers in a number of top interna- small’. In addition to the analysis tool, future research can evalu- tional conferences and journals. He has been (or will) served as mini-track ate the stability of the research model by using objective meas- organiser or co-organiser and chair for the Americas Conference on Information Systems and Hawaii International Conference on System ures, collecting longitudinal data and recruiting large samples Sciences. He is CompTIA RFID + Certified Professional, Academic co-Founder across various industries. Theoretically, the dimensions of the of RFID Academia and Founder and CEO of e-m-RFID.biz. BDAQ model could be extended by adding talent quality due to Saifullah Dewan has been an assistant professor at the ability of data scientists in generating meaningful insights University of Canberra. He has also undertaken impor- and gaining competitive advantages. Since BDA is transforming tant teaching and research roles at other universities operations and enhancing firm performance, future research including London School of Economics, City University, can also investigate the impact of analytics culture in achieving University of New South Wales, and Australian National business outcomes. University. His current research interests include big data, computer human interaction, digital inclusion, cyber security, electronic commerce, technology and work 8. Conclusion nexus, technology adoption and IT education. Saif has published widely in Information Systems outlets and received his PhD in Overall, our study of complex modelling answers a salient ques- Information Systems from Australian National University. Before joining aca- tion raised by Chin (2010, 645), ‘… whether the goal is to explain demia, Saif worked for several years in the software industry in Australia and Asia and consulted for governments, businesses and international organisa- the covariances of a relatively small set of measured items based tions. He also serves as a director on the boards for private schools and on a few underlying latent constructs or to focus on the com- NGOs. plex interrelationships among a large set of factors that more closely mirrors the study context’. Drawing on the philosophy of verisimilitude (Meehl 1990; Rozeboom 2005), we propose that References PLS-SEM may prove highly useful in developing and validating Ahmad, Naim, and Rashid Mehmood. 2016. “Enterprise Systems and complex models especially when the focus is on embracing Performance of Future City Logistics.” Production Planning & Control 27 (6): completeness (Meehl 1990), capturing reality (Cudeck and Henly 500–513. Akter, Shahriar, John D’Ambra, and Pradeep Ray. 2010. “Service Quality of 2003) or, reflecting the true parameters in the study. Therefore, MHealth Platforms: Development and Validation of a Hierarchical Model the study concludes that PLS-SEM is a modest and realistic tech- Using PLS.” Electronic Markets 20 (3–4): 209–227. nique to establish rigor in complex modelling, which reflects Akter, Shahriar, John D’Ambra, and Pradeep Ray. 2011. “Trustworthiness in Wold’s (1982) viewpoint: ‘There is nothing vague or fuzzy about mHealth Information Services: An Assessment of a Hierarchical Model soft modelling; the technical argument is entirely rigorous’. with Mediating and Moderating Effects Using Partial Least Squares (PLS).” Journal of the American Society for Information Science and Technology 62 (1): 100–116. Disclosure statement Akter, Shahriar, and Samuel Fosso Wamba. 2016. “Big Data Analytics in E-commerce: A Systematic Review and Agenda for Future Research.” No potential conflict of interest was reported by the authors. Electronic Markets 26(2): 173. doi:10.1007/s12525-016-0219-0.  PRODUCTION PLANNING & CONTROL  1019 Akter, Shahriar, Samuel Fosso Wamba, Angappa Gunasekaran, Rameshwar Dijkstra, Theo K. 2010. “Latent Variables and Indices: Herman Wold’s Basic Dubey, and Stephen J. Childe. 2016. “How to Improve Firm Performance Design and Partial Least Squares.” In Handbook of Partial Least Squares, Using Big Data Analytics Capability and Business Strategy Alignment?” edited by Vincenzo Esposito Vinzi, Wynne W. Chin, Jörg Henseler, and International Journal of Production Economics 182: 113–131. doi:10.1016/j. Huiwen Wang, 23–46. Heidelberg: Springer. ijpe.2016.08.018. Dijkstra, Theo K., and Jörg Henseler. 2015a. “Consistent and Asymptotically Bagozzi, R. P. 1980. Causal Models in Marketing. New York: Wiley. Normal PLS Estimators for Linear Structural Equations.” Computational Barclay, Donald, Christopher Higgins, and Ronald Thompson. 1995. “The Statistics & Data Analysis 81: 10–23. Partial Least Squares (PLS) Approach to Causal Modeling: Personal Dijkstra, Theo K., and Jörg Henseler. 2015b. “Consistent Partial Least Squares Computer Adoption and Use as an Illustration.” Technology Studies 2 (2): Path Modeling.” MIS Quarterly= Management Information Systems 285–309. Quarterly 39 (2):297-316. Barendse, M. T., F. J. Oort, and G. J. A. Garst. 2010. “Using Restricted Factor Edwards, Jeffrey R. 2001. “Multidimensional Constructs in Organizational Analysis with Latent Moderated Structures to Detect Uniform and Behavior Research: An Integrative Analytical Framework.” Organizational Nonuniform Measurement Bias; A Simulation Study.” AStA Advances in Research Methods 4 (2): 144–192. Statistical Analysis 94 (2): 117–127. Efron, Bradley, and Robert Tibshirani. 1993. An Introduction to the Bootstrap. Baron, R. M., and D. A. Kenny. 1986. “The Moderator–Mediator Variable Vol. 57. New York, NY: Chapman & Hall/CRC. Distinction in Social Psychological Research: Conceptual, Strategic and Esposito Vinzi, Vincenzo, Wynne W. Chin, Jörg Henseler, and Huiwen Statistical Considerations.” Journal of Personality and Social Psychology 51 Wang. 2010. Handbook of Partial Least Squares: Concepts, Methods and (6): 1173–1182. Applications. Berlin: Springer. Barton, D., and D. Court. 2012. “Making Advanced Analytics Work for You.” Esposito Vinzi, Vincenzo, Laura Trinchera, Silvia Squillacciotti, and Michel Harvard Business Review 90 (10): 78. Tenenhaus. 2008. “REBUS-PLS: A Response-Based Procedure for Detecting Becker, Jan-Michael, Kristina Klein, and Martin Wetzels. 2012. “Hierarchical Unit Segments in PLS Path Modelling.” Applied Stochastic Models in Latent Variable Models in PLS-SEM: Guidelines for Using Reflective- Business and Industry 24 (5): 439–458. Formative Type Models.” Long Range Planning 45 (5): 359–394. Fassott, Georg, Jörg Henseler, and Pedro S. Coelho. 2016. “Testing Becker, Jan-Michael, Arun Rai, Christian M. Ringle, and Franziska Völckner. Moderating Effects in PLS Path Models with Composite Variables.” 2013. “Discovering Unobserved Heterogeneity in Structural Equation Industrial Management & Data Systems 116 (9):1887–1900. doi: 10.1108/ Models to Avert Validity Threats.” MIS Quarterly 37 (3): 665–694. imds-06-2016-0248. Bentler, Peter M., and Wenjing Huang. 2014. “On Components, Latent Fornell, Claes, and Fred L. Bookstein. 1982. “Two Structural Equation Models: Variables, PLS and Simple Methods: Reactions to Rigdon’s Rethinking of LISREL and PLS Applied to Consumer Exit-voice Theory.” Journal of PLS.” Long Range Planning 47 (3): 138–145. doi:10.1016/j.lrp.2014.02.005. Marketing Research 19: 440–452. Blalock, Hubert M. 1964. “Causal Inferences in Nonexperimental Research.” Fornell, Claes, and David F. Larcker. 1981. “Evaluating Structural Equation New York: Norton (1972, c1964) Models with Unobservable Variables and Measurement Error.” Journal of Blalock, Hubert M. 1979. “The Presidential Address: Measurement and Marketing Research 18: 39–50. Conceptualization Problems: The Major Obstacle to Integrating Theory Fosso Wamba Samuel, Shahriar, Akter, Andrew Edwards, Geoffrey Chopin, and Research.” American Sociological Review 44 (6): 881–894. and Denis Gnanzou. 2015. “How ‘Big Data’ Can Make Big Impact: Clegg, Ben, Roya Gholami, and Mine Omurgonulsen. 2013. “Quality Findings from a Systematic Review and a Longitudinal Case Study.” Management and Performance: A Comparison Between the UK and International Journal of Production Economics 165: 234–246. doi:10.1016/j. Turkey.” Production Planning & Control 24 (12): 1015–1031. doi:10.1080/0 ijpe.2014.12.031. 9537287.2011.642486. Gefen, David, Detmar W. Straub, and Edward E. Rigdon. 2011. “An Update Chin, Wynne W. 1998a. “Commentary: Issues and Opinion on Structural and Extension to SEM Guidelines for Admnistrative and Social Science Equation Modeling.” Management Information Systems Quarterly 22 (1): Research.” Management Information Systems Quarterly 35 (2): iii–xiv. vii–xvi. George, Gerard, Martine R. Haas, and Alex Pentland. 2014. “Big Data and Chin, Wynne W. 1998b. “The Partial Least Squares Approach for Structural Management.” Academy of Management Journal 57 (2): 321–326. Equation Modeling.” Modern Methods for Business Research 295 (2): 295– Gregor, Shirley, Michael Martin, Walter Fernandez, Steven Stern, and Michael 336. Vitale. 2006. “The Transformational Dimension in the Realization of Chin, Wynne W. 2010. “How to Write up and Report PLS Analyses.” In Business Value from Information Technology.” The Journal of Strategic Handbook of Partial Least Squares, edited by V. Esposito Vinzi, W. W. Chin, Information Systems 15 (3): 249–70. J. Henseler, & H. Wang, 655–690. Berlin: Springer. Golder, Peter N., Debanjan Mitra, and Christine Moorman. 2012. “What is Chin, Wynne W., and Abhijit Gopal. 1995. “Adoption Intention in GSS: Relative Quality? An Integrative Framework of Processes and States.” Journal of Importance of Beliefs.” ACM SIGMIS Database 26 (2–3): 42–64. Marketing 76 (4): 1–23. Chin, Wynne W., Barbara L. Marcolin, and Peter R. Newsted. 2003. “A Partial Guide Jr., V. Daniel R., and Mikko Ketokivi. 2015. “Notes from the Least Squares Latent Variable Modeling Approach for Measuring Editors: Redefining Some Methodological Criteria for the Journal.” Interaction Effects: Results from a Monte Carlo Simulation Study and an Journal of Operations Management 37: v–viii. doi:10.1016/S0272- Electronic-mail Emotion/Adoption Study.” Information Systems Research 6963(15)00056-X. 14 (2): 189–217. Hair, Joe F., G. Tomas M. Hult, Christian Ringle, and Marko Sarstedt. 2013. A Chin, Wynne W., and Peter R. Newsted. 1999. “Structural Equation Modeling Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). Analysis with Small Samples Using Partial Least Squares.” Statistical Thousand Oaks, CA: Sage. Strategies for Small Sample Research 2: 307–342. Hair, Joe F., Christian M. Ringle, and Marko Sarstedt. 2011. “PLS-SEM: Indeed Chin, Wynne W., Robert A. Peterson, and Steven P. Brown. 2008. “Structural a Silver Bullet.” The Journal of Marketing Theory and Practice 19 (2):139– Equation Modeling in Marketing: Some Practical Reminders.” The Journal 152. of Marketing Theory and Practice 16 (4): 287–298. Hair, Joseph F., Christian M. Ringle, and Marko Sarstedt. 2012. “Partial Least Cudeck, Robert, and Susan J. Henly. 2003. “A Realistic Perspective on Pattern Squares: The Better Approach to Structural Equation Modeling?” Long Representation in Growth Data: Comment on Bauer and Curran (2003).” Range Planning 45 (5–6): 312–319. Psychological Methods 8(3): 378–383. Hair, Joe F., Marko Sarstedt, Christian M. Ringle, and Jeannette A. Mena. 2012. Davenport, T. H. 2006. “Competing on Analytics.” Harvard Business Review 84 “An Assessment of the Use of Partial Least Squares Structural Equation (1): 98–107. Modeling in Marketing Research.” Journal of the Academy of Marketing Davenport, Thomas H., Paul Barth, and Randy Bean. 2012. “How ‘Big Data’is Science 40 (3): 414–433. Different.” MIT Sloan Management Review 54 (1): 43–46. Henseler, Jörg, Geoffrey Hubona, and Pauline Ash Ray. 2016. “Using PLS Path Davenport, Thomas H., and Jeanne G. Harris. 2007. Competing on Analytics: Modeling in New Technology Research: Updated Guidelines.” Industrial The New Science of Winning. Boston, MA: Harvard Business School Press. Management & Data Systems 116 (1): 2–20. 1020  S. AKTER ET AL. Henseler, Jörg, Christian M. Ringle, and Marko Sarstedt. 2015. “A New Criterion McAfee, A., and E. Brynjolfsson. 2012a. “Big Data: The Management for Assessing Discriminant Validity in Variance-based Structural Equation Revolution.” Harvard Business Review 90 (10): 60–68. Modeling.” Journal of the Academy of Marketing Science 43 (1): 115–35. McAfee, Andrew, and Erik Brynjolfsson. 2012b. “Big Data: The Management Henseler, Jörg, Christian Ringle, and Rudolf Sinkovics. 2009. “The Use of Revolution.” Harvard Business Review 90: 60–66, 8, 128. Partial Least Squares Path Modeling in International Marketing.” Advances McDonald, Roderick P. 1996. “Path Analysis with Composite Variables.” in International Marketing (AIM) 20: 277–320. Multivariate Behavioral Research 31 (2): 239–270. Huber, Joel, and John McCann. 1982. “The Impact of Inferential Beliefs on Meehl, Paul E. 1990. “Appraising and Amending Theories: The Strategy of Product Evaluations.” Journal of Marketing Research 19: 324–333. Lakatosian Defense and Two Principles That Warrant It.” Psychological Hulland, John. 1999. “Use of Partial Least Squares (PLS) in Strategic Inquiry 1 (2): 108–141. Management Research: A Review of Four Recent Studies.” Strategic Mill, John Stuart. 1865. Auguste Comte and Positivism. London: N Trubner & Management Journal 20 (2): 195–204. Co. Hulland, John, Michael J. Ryan, and Robert K. Rayner. 2010. “Modeling Nelson, R. Ryan, Peter A. Todd, and Barbara H. Wixom. 2005. “Antecedents Customer Satisfaction: A Comparative Performance Evaluation of of Information and System Quality: An Empirical Examination within Covariance Structure Analysis versus Partial Least Squares.” In Handbook the Context of Data Warehousing.” Journal of Management Information of Partial Least Squares, edited by V. E., Vinzi, W. W. Chin, J. Henseler and H. Systems 21 (4):199–235. Wang, 307–325. Berlin: Springer. Netemeyer, Richard G., William O. Bearden, and Subhash Sharma. 2003. Kotzab, Herbert, Christoph Teller, David B. Grant, and Anders Friis. 2015. “Supply Scaling Procedures: Issues and Applications. London: SAGE. Chain Management Resources, Capabilities and Execution.” Production Parasuraman, A., V. A., Zeithaml, and L. L., Berry. 2005. “E-S-QUAL a Multiple- Planning & Control 26 (7): 525–542. doi:10.1080/09537287.2014.927932. item Scale for Assessing Electronic Service Quality.” Journal of Service Iacobucci, Dawn. 2009. “Everything You Always Wanted to Know about Research 7 (3): 213–233. SEM (Structural Equations Modeling) but Were Afraid to Ask.” Journal of Pyke, Steve. 2011. Philosophers. Oxford: Oxford University Press. Consumer Psychology 19 (4): 673–680. Ransbotham, Sam, David Kiron, and Pamela Kirk Prentice. 2016. “Beyond the Iacobucci, Dawn, and Nigel Hopkins. 1992. “Modeling Dyadic Interactions Hype: The Hard Work behind Analytics Success.” MIT Sloan Management and Networks in Marketing.” Journal of Marketing Research 29: 5–17. Review 57 (3): 1–16 Jacoby, Jacob. 1978. “Consumer Research: A State of the Art Review.” Journal Ray, Gautam, Waleed A. Muhanna, and Jay B. Barney. 2005. “Information of Marketing 42: 87–96. Technology and the Performance of the Customer Service Process: A Jarvis, Cheryl Burke, Scott B. MacKenzie, and Philip M. Podsakoff. 2003. Resource-based Analysis.” MIS Quarterly 29 (4): 625–652. “A Critical Review of Construct Indicators and Measurement Model Reinartz, Werner, Michael Haenlein, and Jörg Henseler. 2009. “An Empirical Misspecification in Marketing and Consumer Research.” Journal of Comparison of the Efficacy of Covariance-Based and Variance-Based Consumer Research 30 (2): 199–218. SEM.” International Journal of Research in Marketing 26 (4):332–344. Ji-fan Ren, Steven, Samuel FossoWamba, Shahriar Akter, Rameshwar Ren, Steven Jifan, Hu Caihong, E. W. T. Ngai, and Mingjian Zhou. 2015. “An Dubey, and Stephen J. Childe. 2016. “Modelling Quality Dynamics, Empirical Analysis of Inter-organisational Value Co-creation in a Supply Business Value and Firm Performance in a Big Data Analytics Chain: A Process Perspective.” Production Planning & Control 26 (12): 969– Environment.” International Journal of Production Research 90: 1–16. doi: 980. doi:10.1080/09537287.2014.1002022. 10.1080/00207543.2016.1154209. Rigdon, Edward E. 2014. “Rethinking Partial Least Squares Path Modeling: Kiron, David, Pamela Kirk Prentice, and Renee Boucher Ferguson. 2014. “The Breaking Chains and Forging Ahead.” Long Range Planning 47 (3): 161– Analytics Mandate.” MIT Sloan Management Review 55 (4): 1–25. 167. Lavalle, S., E. Lesser, R. Shockley, M. S. Hopkins, and N. Kruschwitz. 2011. Ringle, Christian M., Marko Sarstedt, and Detmar W. Straub. 2012a. “Editor’s “Big Data, Analytics and the Path from Insights to Value.” MIT Sloan Comments: A Critical Look at the Use of PLS-SEM in MIS Quarterly.” MIS Management Review 52 (2): 21–32. Quarterly 36 (1): iii-xiv. Law, Kenneth S., and Chi-Sum Wong. 1999. “Multidimensional Constructs in Ringle, Christian, Marko Sarstedt, and Detmar Straub. 2012b. “A Critical Structural Equation Analysis: An Illustration Using the Job Perception and Look at the Use of PLS-SEM in MIS Quarterly.” MIS Quarterly (MISQ) 36 Job Satisfaction Constructs.” Journal of Management 25 (2): 143–160. (1): iii–xiv. Lee, Voon-Hsien, Keng-Boon Ooi, Alain Yee-Loong Chong, and Binshan Lin. Ringle, C. M., S. Wende, and J. M. Becker. 2015. “Smart PLS 3. Boenningstedt: 2015. “A Structural Analysis of Greening the Supplier, Environmental SmartPLS GmbH.” http://www. smartpls. com. Performance and Competitive Advantage.” Production Planning & Control Rönkkö, Mikko, Cameron N. McIntosh, John Antonakis, and Jeffrey R. 26 (2): 116–30. doi:10.1080/09537287.2013.859324. Edwards. 2016. “Partial Least Squares Path Modeling: Time for Some Lohmoller, Jan-Bernd. 1988. “The PLS Program System: Latent Variables Path Serious Second Thoughts.” Journal of Operations Management 47: Analysis with Partial Least Squares Estimation.” Multivariate Behavioral 9–27. Research 23 (1): 125–127. Rozeboom, Wm W. 2005. “Meehl on Metatheory.” Journal of Clinical Lohmoller, Jan-Bernd. 1989. Latent Variable Path Modeling with Partial Least Psychology 61 (10): 1317–1354. Squares. Heidelberg: Physica-Verlag. Sanders, Nada R. 2016. “How to Use Big Data to Drive Your Supply Chain.” Lu, Irene R. R., Ernest Kwan, D. Roland Thomas, and Marzena Cedzynski. California Management Review 58 (3): 26–48. 2011. “Two New Methods for Estimating Structural Equation Models: Sarstedt, Marko, Joseph F. Hair, Christian M. Ringle, Kai O. Thiele, and Siegfried An Illustration and a Comparison with Two Established Methods.” P. Gudergan. 2016. “Estimation Issues with PLS and CBSEM: Where the International Journal of Research in Marketing 28 (3): 258–268. Bias Lies!” Journal of Business Research 69 (10): 3998–4010. MacCallum, Robert C., and James T. Austin. 2000. “Applications of Structural Sarstedt, Marko, Jörg Henseler, and Christian M. Ringle. 2011. “Multigroup Equation Modeling in Psychological Research.” Annual Review of Analysis in Partial Least Squares (PLS) Path Modeling: Alternative Psychology 51 (1): 201–226. Methods and Empirical Results.” In Measurement and Research Methods in MacKenzie, Scott B., Philip M. Podsakoff, and Cheryl Burke Jarvis. 2005. “The International Marketing, edited by M. Sarstedt, M. Schwaiger, C. R. Taylor, Problem of Measurement Model Misspecification in Behavioral and 195–218. Bingley: Emerald Group Publishing Limited. Organizational Research and Some Recommended Solutions.” Journal of Sarstedt, Marko, Christian M. Ringle, and Joseph F. Hair. 2014. “PLS-SEM: Applied Psychology 90 (4): 710–730. Looking Back and Moving Forward.” Long Range Planning 47 (3): 132–137. Marcoulides, George A., Wynne W. Chin, and Carol Saunders. 2009. “A Critical doi:10.1016/j.lrp.2014.02.008. Look at Partial Least Squares Modeling.” MIS Quarterly 33 (1): 171–175. Sawyer, Alan G., and J. Paul Peter. 1983. “The Significance of Statistical Marsh, Herbert W., Kit-Tai Hau, and Zhonglin Wen. 2004. “In Search of Golden Significance Tests in Marketing Research.” Journal of Marketing Research Rules: Comment on Hypothesis-testing Approaches to Setting Cutoff 20: 122–133. Values for Fit Indexes and Dangers in Overgeneralizing Hu and Bentler’s Shah, Rachna, and Susan Meyer Goldstein. 2006. “Use of Structural Equation (1999) Findings.” Structural Equation Modeling: A Multidisciplinary Journal Modeling in Operations Management Research: Looking Back and 11 (3): 320–341. Forward.” Journal of Operations Management 24 (2): 148–169.  PRODUCTION PLANNING & CONTROL  1021 Spreng, Richard A., Scott B. MacKenzie, and Richard W. Olshavsky. 1996. “A Wixom, B. H., B. Yen, and M. Relich. 2013. “Maximizing Value from Business Reexamination of the Determinants of Consumer Satisfaction.” Journal of Analytics.” MIS Quarterly Executive 12: 111–123. Marketing 25 (2): 15–32. Wold, Herman. 1975. “Soft Modelling by Latent Variables: The Non-linear Stafford, T. F. 2011. “Special Research Commentary Series on Advanced Iterative Partial Least Squares (NIPALS) Approach.” In Perspectives in Methodological Thinking for Quantitative Research.” Management Probability and Statistics, in Honor of MS Bartlett, edited by M. S. Bartlett, Information Systems Quarterly 35 (2): xv–xvi. 117–144. London: Academic Press.. Tenenhaus, Michel. 2008. “Component-based Structural Equation Modelling.” Wold, Herman. 1980. “Model Construction and Evaluation When Theoretical Total Quality Management 19 (7–8): 871–886. Knowledge is Scarce.” In Evaluation of Econometric Models, edited by J. Tenenhaus, Michel, Vincenzo Esposito Vinzi, Yves-Marie Chatelin, and Carlo Kmenta and J. B. Ramsey, 47–74. London: Academic Press. Lauro. 2005. “PLS Path Modeling.” Computational Statistics & Data Analysis Wold, Herman. 1982. “Models for Knowledge.” In The Making of Statisticians, 48 (1): 159–205. 189–212. New York: Springer. Tenenhaus, Arthur, and Michel Tenenhaus. 2011. “Regularized Generalized Wold, Herman. 1985. “Partial Least Squares.” Encyclopedia of Statistical Canonical Correlation Analysis.” Psychometrika 76 (2): 257–284. Sciences, edited by S. Kotz and N. L. Johnson, 581–591. New York: Wiley. Wetzels, Martin, Gaby Odekerken-Schroder, and Claudia Van Oppen. 2009. Wixom, Barbara H., and Peter A. Todd. 2005. “A Theoretical Integration of User “Using PLS Path Modeling for Assessing Hierarchical Construct Models: Satisfaction and Technology Acceptance.” Information Systems Research Guidelines and Empirical Illustration.” MIS Quarterly 33 (1): 177. 16 (1): 85–102.