BBA 3602 Unit VIII Scholarly Activity

RESEARCH NOTE

SYSTEM DYNAMICS MODELING FOR INFORMATION SYSTEMS RESEARCH: THEORY DEVELOPMENT

AND PRACTICAL APPLICATION1

Yulin Fang Department of Information Systems, College of Business, City University of Hong Kong,

Kowloon Tong, HONG KONG {[email protected]}

Kai H. Lim Department of Information Systems, College of Business, City University of Hong Kong,

Kowloon Tong, HONG KONG {[email protected]}

Ying Qian Department of Information Systems, School of Management, Shanghai University,

Shanghai CHINA {[email protected]}

Bo Feng Department of Management, School of Business, Soochow University,

Soochow CHINA {[email protected]}

Most information systems (IS) research develops theory for explanation and prediction based on a variance logical structure that assumes one-way, time invariant causal relationships. This approach largely misses the opportunity to extend theory from alternative logical structures that build upon reciprocal and temporal causal mechanisms; for example, the system perspective. This paper introduces system dynamics (SD), a modeling tool capable of capturing the reciprocal and temporal causal mechanisms that underlie many complex and dynamic systems, and demonstrates its ability to extend existing variance theory from a system perspective. To do so, we first describe the basic tenets of SD and discuss the status quo of existing SD applications in the field. Then, we demonstrate how to model SD’s unique theoretical logic of reciprocal and temporal causal structure to extend existing variance theory. To demonstrate the use of SD in theory development, we develop and validate an SD model of the e-commerce resource endowment of a click-and-mortar firm and simulate dynamic causal relationships between the e-commerce resource endowment and firm performance over time, under various scenarios. This case demonstrates how we can extend an existing variance theory by reconciling the inconsistent findings of prior research from a system perspective using the SD approach. The paper concludes by discussing how SD can help IS researchers develop dynamic theories.

1

Keywords: System dynamics, simulation, theory development, electronic commerce, resource-based view, firm performance, the system perspective

1H. Rao was the accepting senior editor for this paper. Samir Chatterjee served as the associate editor. Ying Qian and Bo Feng served as corresponding authors for this paper.

The appendices for this paper are located in the “Online Supplements” section of the MIS Quarterly’s website (http://www.misq.org).

DOI: 10.25300/MISQ/2018/12749 MIS Quarterly Vol. 42 No. 4, pp. 1303-1329/December 2018 1303

Fang et al./System Dynamics Modeling for IS Research

Introduction

Theory development has always been an important goal of information systems (IS) research. Past IS research has generally been successful in advancing theories in different subject domains (Grover et al. 2008; Straub 2012; Weber 2003), mostly for the purpose of explanation and prediction (i.e., Type IV theory, hereafter theory) (Gregor 2006; Gregor and Klein 2014; Grover and Lyytinen 2015). As the field has matured, theory extension (i.e., extending an existing theory) has become a common practice for further developing theory (Grover and Lyytinen 2015). Theory extension focuses on enriching explanation and enhancing prediction through theory-based works that “challenge the assumptions under- lying existing theories in some significant way” (Alvesson and Sandberg 2011, p. 247).

To date, most IS researchers have extended theory by focusing on substance of theory, such as introducing new constructs, developing nomological networks, or examining boundary conditions (Grover and Lyytinen 2015). However, the logical structure of theory—that is, the nature of causality underlying a theory—has been a far less common avenue for theory development (Burton-Jones et al. 2015; Markus and Robey 1988). Indeed, it has been found that IS studies devel- oping theory mostly use the variance logical structure (Parc et al. 2008), which assumes one-way, time invariant causal relationships (Markus and Robey 1988).

The variance logical structure is powerful; however, relying on it too heavily could lead researchers to miss the oppor- tunity to further develop IS theory through alternative logical structures (Burton-Jones et al. 2015). That is, researchers could identify theoretical tensions in an existing variance theory by examining the assumptions about the logical struc- ture that it incorporates, and then extending the theory through an alternative logical structure. For instance, the pro- cess perspective, which explains how outcomes unfold over time as discrete events in a sequential manner (Markus and Robey 1988), has been used as an alternative logical structure (Parc et al. 2008). Heated discussion has centered on how this logical structure can complement studies from the vari- ance perspective (Sabherwal and Robey 1995; Shaw and Jarvenpaa 1997), demonstrating the strong scholarly interest in developing theory from alternative logical structures. More recently, Burton-Jones et al. (2015) contend that scholars should treat logical structures (they call them theoretical perspectives) more flexibly by illustrating how different logi- cal structures can yield distinctive theoretical insights into the famous IS success model.

Our study ventures into another logical structure that has received much less attention in the field: the system perspec-

tive. In contrast to the variance perspective, the system per- spective assumes reciprocal, temporal causal relationships among parts of a holistic system (Von Bertalanffy 1972). As such, it looks at social systems holistically, stressing the dynamic interrelationships among concepts that represent the components of the system (Von Bertalanffy 1972). In this study, we suggest that the system perspective has the potential to extend studies using the dominant logical structure in the field (Burton-Jones et al. 2015; Klir 1991). To our knowl- edge, there is very little extant methodological guidance on how to extend an existing variance theory from the system perspective. As such, few researchers attempt to extend theory from the system perspective, and those who do may do so inappropriately (Burton-Jones et al. 2015). Thus, it is pertinent to develop a methodology that embraces the system logical structure and demonstrates how variance theories can be extended from the system perspective.

To address this need, this paper introduces system dynamics (SD) as a modeling methodology for extending existing vari- ance theory in the IS field from a system perspective. SD is a modeling tool that is able to model system structures that assume the logical structure of circular and delayed causality between system components (Sterman 2000). These two causality characteristics are at the core of the system perspec- tive. As such, we contend that SD is powerful for modeling how system logical structure generates system behaviors, particularly for complex systems with feedback mechanisms observed in many IS phenomena. Our introduction of SD here represents two important departures from the extant liter- ature. First, SD was originally used to model macro and relatively stable social systems with a long-term policy orien- tation (Barlas 1996; Meadows 1980). It was applied in business disciplines much later, to long-term to mid-term organizational changes, such as restructure and reorientation (Sastry 1997), quality improvement (Sterman et al. 1997), and innovation diffusion (Repenning 2002). Given that IS phe- nomena are more dynamic and short-lived, our introduction of SD could meaningfully extend the applicability of SD to this rather different discipline. Second, while there were limited SD applications in the field in the past (Abdel-Hamid 1988; Abdel-Hamid and Madnick 1989; Dutta and Roy 2005; Gallivan et al. 2003; Qian et al. 2012), these studies tended to develop SD models through empirical observations, or take SD as a simulation method to overcome empirical challenges, such as data unavailability and research context intractability (Prietula et al. 1998). No research has deliberately leveraged SD for theory development, particularly by realizing its poten- tial to extend the dominant variance theories from the system perspective. Our study focuses on theory development through SD, positing that SD can challenge the assumption underlying the variance logical structure by building on the system logical structure, thereby offering distinct theoretical insights.

1304 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

To achieve our objective, we demonstrate how SD can extend an existing variance theory in IS research by accounting for reciprocal (i.e., circular) and temporal (i.e., time variant) causality from the system perspective. Specifically, we use SD to extend variance theory by reconciling inconsistent findings observed in the existing literature (Gregor 2006). Resolving inconsistent findings is a usual point of departure for theory development, as conflicting findings often imply that there is tension in the existing theory; that is, theoretical explanation is insufficient or prediction imprecise (Smith and Hitt 2005). We show that SD can offer a plausibly more encompassing theoretical account by explaining how recipro- cal and temporal causal interactions could lead to contrasting, unexpected outcomes (Meadows 1980),2 thereby extending the original theory.

To provide a systematic roadmap for theory development using SD, we ground our demonstration in the e-commerce literature, a maturing area where the variance perspective dominates and where there are inconsistent results about the performance effects of e-commerce resource endowment. Building on the resource-based view, this area of research generally predicts that e-commerce resource endowment improves performance. Some empirical e-commerce research has established an overall positive relationship between e- commerce investment and performance (Hulland et al. 2007; Zhu 2004; Zhu and Kraemer 2002). However, there are inconsistent findings that question the explanatory power of the theory. For instance, other studies have found that firms deploying e-commerce initiatives derive no benefit (Martin- sons and Martinsons 2002), and in some cases, even put themselves at risk of total failure (Coughlan et al. 2001; Han and Noh 1999). We show how SD can reconcile such mixed findings by explaining why and how the reciprocal and tem- poral interactions between e-commerce resource endowment and firm performance (an indicator of the firm’s future resource endowment) could lead to different performance outcomes, hence advancing the existing variance theory.

Our study intends to contribute to research and practice in several ways. First, we demonstrate SD as a theory develop- ment tool in that SD can extend variance theories from the system perspective by modeling circular and delayed causal

relationships. Second, in demonstrating SD for theory devel- opment, we reconcile inconsistent findings observed in the e- commerce literature by developing a dynamic theory of e- commerce resource endowment, thus contributing to the e- commerce literature. Third, as will be pointed out later, there is a lack of rigorous SD modeling in the IS literature. We contribute by offering a systematic procedure for building, validating, and simulating an SD model, thus improving the methodological rigor of SD modeling for IS research. Finally, this study demonstrates the efficacy of the SD approach by extending it to the IS discipline.

The remainder of the paper is organized as follows. The next section introduces the system perspective, and reviews the SD method and its prior application in IS research. We then demonstrate how SD could be used to extend variance theory using the example of a traditional retailer initiating e- commerce capability. Next, we present simulation analyses to examine how a growing e-commerce capability interacts with existing marketing capability over time to influence firm performance outcomes. Finally, we discuss our major findings, the theoretical and practical implications of the study, and opportunities for future research.

System Logical Structure and System Dynamics Modeling

Logical structure is concerned with the causal formulation of theoretical reasoning, according to Markus and Robey (1988). Variance logical structure is concerned with predicting levels of outcome from levels of antecedents based on variation among the values of variable properties. In research models based on the variance logical structure, antecedents are posited to be necessary and sufficient to explain a specific outcome through a one-way, time invariant causal relation- ship. Unlike the variance perspective, the system perspective looks at social systems holistically, stressing the dynamic interrelationships among concepts that represent the compo- nents of the system (Von Bertalanffy 1972). Many IS phenomena can be viewed as open systems, characterized by a continuous exchange with their environment and inter- actions with other systems outside of themselves. As open systems, the phenomena are dynamically complex due to the mutually influencing and time-delayed effects between con- cepts. For instance, the relationship between e-commerce resource endowment and firm performance is not unidirec- tional, but mutually influencing. From the system perspec- tive, these open systems could be modeled in terms of concepts such as input, output, feedback, and environment (Von Bertalanffy 1973).

2Although time-lagged econometric models can also be used to assess delayed effects, econometric analysis is a static approach that assumes the past is merely and mostly linearly extrapolated into the future. In contrast, the system dynamics approach, which is process-oriented and causal- descriptive, embodies multiple circular and delayed effects in which key factors mutually influence and adapt over time, creating nonlinear and com- plex behaviors difficult to capture using econometric models (Forrester 1971).

MIS Quarterly Vol. 42 No. 4/December 2018 1305

Fang et al./System Dynamics Modeling for IS Research

Most IS research is conducted from the variance perspective;

however, it is noteworthy that many of the forerunners of our

field were systems theorists (Churchman 1979; Forrester

1968; Von Bertalanffy 1968), according to Burton-Jones et al.

(2015). It was just that the impact of the system perspective

somehow dissipated during the late 1980s as researchers were

drawn to the reasonable and parsimonious explanatory power

of the variance perspective (Burton-Jones et al. 2015). While

many research models were developed based on the variance

logical structure, their underlying theoretical foundations are

dynamic and feature feedback loops. This implies that it is

appropriate to develop theory using the system perspective

(Gregor 2006). For instance, Von Bertalanffy’s famous

general system theory (1973) and Shannon’s information

theory (1948) are “grand” Type IV theories of system nature

(Ashby 1956; Von Bertalanffy 1973), but they have been

adapted over time to yield many variance theories (Gregor

2006). In particular, general system theory suggests that the

structure of social systems is generally characterized by feed-

back loops, accumulation processes, and delays between

cause and effect (Lane 1999; Von Bertalanffy 1968), giving

important theoretical substance to the system logical structure.

Accordingly, the system logical structure emphasizes that one

must incorporate feedback mechanisms and time delays in

order to understand causality mechanisms in social systems.

Systems dynamics is one such tool that addresses the system logical structure. System dynamics inherits the structural

principles of general system theory, making statements about

the principal mutual dependencies of elements in social sys-

tems. System dynamics postulates that dynamic processes in

social systems function in feedback loops and that the history

of systems accumulates in state variables. The accumulated

history influences the future development of a system—a

process that is often affected by time delays.

SD uses two fundamental structures, feedback loops and stocks and flows, to capture the aforementioned circular and delayed causality that exists among the elements of a system

(Randers 1980; Sterman 2000). Feedback loops capture

circular causality within a system, and stocks and flow struc-

tures identify potential delays in the system.

Feedback loops are formed by connecting a set of chained

causal links. A link is a directed curve denoting the causal relationship between two variables. The polarity of a link shows how the effect would change in response to a change

in that cause alone. A positive polarity means that the change

in the effect is in the same direction as the change in the

cause. A negative polarity means that the change in the effect

is in the opposite direction to the change in the cause. When

a chain of causal links closes up to constitute a loop, a circular

or endogenous effect is represented. A reinforcing loop con-

tains an even number of negative links (including zero) and a

balancing loop contains an odd number of negative links. Reinforcing loops amplify deviations and reinforce change,

producing path-dependent behavior, such as Sterman’s (2000)

notion of exponential growth. Balancing loops seek balance,

equilibrium, and stasis, and act to bring a system into line

with a goal or a desired state, such as goal-seeking behavior

(Sterman 2000). Most dynamic and nonlinear system

behaviors arise from the interaction of these two types of

feedback loops (Forrester 1971), illustrated in Figure 1.

Stocks and flows, defined as the accumulation and dispersal

of resources, are instrumental in modeling the time delay

inherent in organizational dynamics (Sterman 2001). The

fundamental distinctions between stocks and flows may be

illustrated by the “bathtub metaphor”: at any given moment,

the stock of water is shown by the level of water in the tub, which is the cumulative result of the flow of water into the tub (through the tap) and out of it (through a leak). This notion of

stocks and flows is a useful way to distinguish between stock

variables and flow variables. To illustrate, consider two

major concepts that we will later use in our simulation study:

capability and resource. Capabilities are complex bundles of accumulated skills and knowledge that enable firms to

coordinate activities (Day 1994; Kogut and Zander 1996);

resources are the physical, human, organizational, and mone- tary assets devoted to building capabilities (Wernerfelt 1984).

In other words, capabilities refer to the stocks accumulated by

choosing the appropriate time paths of resource flows over a

period of time (Dierickx and Cool 1989; Warren 2002). Al-

though resource flows can be adjusted instantaneously,

capabilities can only be changed through the time paths of

resource flows, delaying the effect of capability building

(Dierickx and Cool 1989). The SD simulation method typi-

cally uses a series of simple relationships with circular

causality (i.e., feedback loops), sometimes with time delay

through stocks and flow structures, to model a system. These

causal loops often share common constructs to intersect.

Differences between SD and other simulation methods used

in the IS and management fields are summarized in Appendix

A.

Given these distinctive modeling features, SD focuses on

understanding how causal relationships among constructs of

a research model representing a system can affect the

temporal behavior of the system (Forrester 1961; Sterman

2001). The SD method is useful for understanding how the

behavior of a system with complex causality changes over

time. Research questions are often framed as: How does an initial condition of a system affect the stability of the system over time? In particular, SD helps researchers find the initial conditions that could lead to abrupt, nonlinear changes, such

as tipping points, catastrophes, and the emergence of virtuous

1306 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Figure 1. Illustration of Causal Loops of SD Model

or vicious cycles that are often counterintuitive. In this regard, we suggest that SD is particularly useful when prior research on a particular subject exhibits theoretical tension (Smith and Hitt 2005); that is, when there is inconsistent evi- dence about the relationships between concepts and when existing theories cannot adequately account for this incon- sistency. This subject domain begs for tension to be resolved by improving the explanatory and predictive power of theory; SD could be a tool to meet this need by elaborating on the relationships concerned from the system perspective.

Despite its potential for theory development, SD has rarely been used in the IS field (Table 1 lists IS articles using SD), unlike the sibling fields of operations management (Größler et al. 2008) and general management (Repenning 2002; Sastry 1997). As noted earlier, this may be because SD was origin- ally used to model macro and stable social systems with long- term policy orientation. The fast-changing characteristics of many IS phenomena question the applicability of SD on the one hand, but also justify the need to extend SD to developing IS theory. Specifically, the SD approach to theory develop- ment that we advocate here differs from the way SD has been used in the IS field in two important aspects. First, prior studies develop SD models based on case studies or logical speculation (Abdel-Hamid 1988; Amitava 2001; Black et al. 2004; Dutta and Roy 2005; Qian et al. 2012; Quaddus and Intrapairot 2001), rather than by building on existing variance theories (see Table 2). Such an approach may be suitable for emerging areas of research with limited literature bases; how- ever, we argue that many research areas in our field have matured sufficiently to require theory extension by using a theory-based SD modeling approach. Second, in order to extend variance theory from the system perspective by resolving theoretical tensions, we need to observe behavioral outcome patterns through model simulation. However, we find that few past IS studies develop a formal SD simulation model, making behavioral pattern observation and discussion, which is an important stage of theory development through

SD, nearly impossible. A cascading issue is that most prior SD applications in our field are less rigorous in model vali- dation, a necessary exercise for establishing model validity before simulation (see Table 2). Thus, it is pertinent to draw on the original SD literature to enhance the techniques used for model validation and demonstrate systematic model simulation. Below, we demonstrate a practical application of SD modeling to address these inadequacies in the literature.

Modeling Circular and Delayed Causality: The E-Commerce Case

We choose the e-commerce area as a case to demonstrate use of SD for theory development because this area has been extensively studied in the last two decades from the variance perspective. E-commerce is defined as using the Internet to conduct business activities (Zhu 2004). In the marketing con- text, e-commerce can provide customers with multiple points of contact, personalized information, and convenient transac- tions (Zhu 2004). Thus, e-commerce has emerged as an important strategic resource for firms facing fierce competi- tion in traditional markets (Porter 2001). Traditional brick- and-mortar firms now face important strategic decisions about allocating resources to expanding their online presence. This issue has received much scholarly attention over the last decade.

There has been significant research attention paid to under- standing how e-commerce resource endowment affects overall firm performance, mainly by drawing on the resource-based view (RBV) (Barney 1991) and based on the variance logical structure. However, the results are not always consistent. Some empirical studies have found a positive relationship between e-commerce endowment and performance (Hulland et al. 2007; Zhu 2004; Zhu and Kraemer 2002), while others have found that no benefit accrues to firms that deploy e-

MIS Quarterly Vol. 42 No. 4/December 2018 1307

Fang et al./System Dynamics Modeling for IS Research

Table 1. Summary of SD Research in Major IS Journals Core Area Sub Area Research Question

IS development Software development (Abdel-Hamid 1988)

What were the impacts of QA on the project’s cost and schedule?

Software development (Rodrigues and Williams 1997)

How could SD model be used to better manage software development?

IS development (Luna-Reyes et al. 2005)

The causes of success and failure of ISD from social and organizational dynamics.

IS organization and strategy

Data warehouse (Quaddus and Intrapairot 2001)

What are the impacts of management policies on the diffusion of DW in the bank?

ERP system (Akkermans and van Helden 2002)

How do the critical success factors interrelate to explain ERP implementation success and failure?

Knowledge management system (Michael et al. 2003)

The reason knowledge management system initiatives fail to achieve their goals.

New technology--CT scanner (Black et al. 2004)

Why implementing new technologies often disrupts occupational roles in ways that delay expected benefits?

Knowledge system (Garud and Kumaraswamy 2005)

The challenges that organizations face in harnessing knowledge.

Economics of IS and IT

Network service provision (Amitava 2001)

A better understanding of the business process underlying network service provision.

IT offshoring (Dutta and Roy 2005) What are the mechanics to produce the observed growth in IT offshoring?

Table 2. Summary of SD Simulation Research in Major IS Journals Research Question Model Building Model Validation Theory Building

What were the impacts of QA on the project's cost and schedule? (Abdel-Hamid 1988)

Interviews and logical speculation

• Face validity • Replication of

reference modes • Extreme tests • Case study

Examines a previously unexplored relationship or process: the optimal QA expenditure level and its distribu- tion throughout the project's lifecycle

What are the impacts of management policies on the diffusion of DW in the bank? (Quaddus and Intrapairot 2001)

Case study Not mentioned Examines a previously unexplored relationship or process: increase level of training and decrease training delay that will significantly accelerate the diffusion

An integrated view to better understand the business process underlying network service provision. (Amitava 2001)

Logical speculation

• Face validity • Replication of

reference behavior

Introduces a new construct: the customer’s threshold of tolerance for performance degradation is important in balancing market share with profitability

Why implementing new technologies often disrupts occupational roles in ways that delay expected benefits? (Black et al. 2004)

Longitudinal case-study

Not mentioned Introduces a new construct: relative knowledge is a key factor benefit the adoption of new technology

What are the mechanics by which these factors interact to produce the observed growth in IT offshoring? (Dutta and Roy 2005)

Logical speculation

• Case study Developing and testing a computa- tional representation of the mechanics by which related factors interact to produce offshoring growth behavior

1308 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

commerce initiatives (Martinsons and Martinsons 2002). Indeed, some firms have been found to come to the brink of collapse (Coughlan et al. 2001; Han and Noh 1999).

These inconclusive results imply that the relationship between e-commerce and firm performance is too dynamic and com- plex to be accounted for by the variance perspective. One underlying reason for this could be related to the temporal nature of capability building that underlies RBV in general, and of e-commerce capability building in particular. That is, superior financial return on resource endowment may emerge slowly over time as capability develops (Barney 1991; Die- rickx and Cool 1989; Fang et al. 2007). For instance, an initial investment may trigger a self-reinforcing effect— success breeds success—creating a larger financial impact in the long-term (e.g., Dierickx and Cool 1989). However, the IS literature mostly appropriates RBV to develop variance models that are static and ex post, making it difficult to account for situations in which a firm’s decisions about resource endowment change its performance over time (i.e., how different resource endowment strategies evolve the firm toward failure or success) (Lawrence 1997; Miller and Shamsie 1996; Priem and Butler 2001). As such, it is useful to further develop theory on the potential circular and time- variant causal relationships between resource endowment and firm performance from the system perspective. Thus, we aim to use SD to address the question: How and why does the initial e-commerce resource endowment of a traditional firm improve or deteriorate the firm’s performance over time?

We develop a SD model to address this research question by taking the following four steps. First, we develop a concep- tual model with feedback loops (Randers 1980) by creating a base model and extending it with additional causal loops representing e-commerce resource endowment. In so doing, we also demonstrate to the reader how to model circular and delayed causality using SD. Second, we formalize the model by defining the mathematical relationship of variables. Third, we validate the model by conducting a series of tests. In so doing, we add to the existing literature on SD applications by demonstrating a comprehensive model-testing exercise. Finally, we simulate the model for analysis (Luna-Reyes and Andersen 2003; Richardson and Pugh 1981).

Base Model: Model Conceptualization

As a starting point, we model the marketing resource endow- ment process of a traditional retailer based on existing variance-based research. In so doing, we illustrate how to portray reinforcing and balancing loops and their interactions by building on the existing literature. For simplicity, we focus on the causal structure related to the firm’s marketing

capability and its effect on performance (see loop R1 in Figure 2). In the IS literature, marketing capability is defined as the ability to analyze markets, build and maintain brands, formulate plans to sell products, and achieve sales force effectiveness (Capron and Hulland 1999). As the customer base[1] grows, the firm’s revenue[2] increases, expanding the resources available for marketing and sales activities (i.e., marketing resources endowment[3]). Marketing resources endowment[3] is used to improve the firm’s marketing capa- bility[4] (Telser 1961), which in turn leads to a net increase in customers by converting potential customers away from com- petitors and retaining existing customers (i.e., customer net increase[5]) (Capron and Hulland 1999), which builds the customer base[1] (Day 1994; Rust et al. 2002). Following the polarities of the links around loop R1 reveals that R1 contains an even number of negative causal links (i.e., zero); hence, it is a reinforcing feedback loop. This loop (R1) illustrates the path-dependent nature of resource endowment behavior (Peteraf 1993).

Firm growth, reinforced by marketing capability, cannot be sustained due to the finite size of the market; that is, the limit to growth (Meadows et al. 1972; Penrose 1959). Loop B1 in Figure 2 illustrates the causal structure of the firm’s limit to growth. As market size is not infinite, an enlarged customer base[1] of the focal firm implies slower or negative growth in the competitor’s customer base[6], which results in a smaller customer net increase[5] because the base for customer conver- sion is reduced (Sterman 2001). By following the polarities of the links, we can see that loop B1 contains an odd number of negative causal links (i.e., one negative pointing from customer base[1] to competitor’s customer base[6]); hence, it is a balancing feedback loop. This loop (B1) interacts with the reinforcing loop R1 by counteracting the customer base growth that is reinforced by building marketing capability (R1). That is, even though the path-dependent effect of loop R1 increases the customer base[1], loop B1 balances this growth so that the increase in customer base[1] will gradually reduce the size of the competitor’s customer base[6], resulting in a smaller pool of potential customers to be converted (Penrose 1959).

Figure 2 portrays a highly simplified conceptual SD model of the firm’s marketing resources endowment processes by referring to past variance-based research; in so doing, it illus- trates the feedback structure of a system. This manner of model representation is called causal loop diagram (CLD) (Sterman 2000). CLD is commonly used to understand and conceptualize endogenous forces and their interconnections. This simple causal structure can produce multiple, dramati- cally different dynamic behaviors with different performance implications, depending on which loop dominates at various

MIS Quarterly Vol. 42 No. 4/December 2018 1309

Fang et al./System Dynamics Modeling for IS Research

Figure 2. Reinforcing Loop “Marketing Resource Endowment” (R1) and Balancing Loop “Limit to Growth” (B1)

time points (a loop is called a dominating loop when its effect overrides others). Firm performance may exhibit exponential growth if the reinforcing loop R1 positively dominates. Con- versely, if loop R1 negatively dominates (i.e., a vicious cycle), the firm may experience exponential decay. If loops R1 and B1 are in balance, stable equilibrium behavior may result. If the reinforcing loop R1 dominates first and then the balancing loop B1 takes over, an S-shape growth pattern may result (i.e., exponential growth followed by goal-seeking behavior to achieve an equilibrium state) (Sterman 2000). Figure 3 presents the different behavior patterns that arise from the interaction of two loops.

Although one may draw plausible reference behaviors by tracing feedback loops via CLD, this approach is of limited use for addressing temporal effects. SD models account for the temporal effect by augmenting CLD with a stock and flow structure (Sterman 2000).

Base Model: Formal Model with Stocks and Flows

SD takes account of time delay via a stocks and flow struc- ture. The new symbols illustrated in Figure 4―boxes and pipes―representing stocks and flows respectively, capture the delayed effect of resource endowment on capability building. Theoretically, capabilities result from resource inflows over a period of time (Dierickx and Cool 1989; Warren 2002). Thus, we augment the causal loop diagram in Figure 3 with stocks and flow symbols that represent the delayed effect of marketing resources endowment[3] on marketing capability[4] (shown in Figure 4). The box of marketing capability[4] refers to capability’s accumulation status, while the pipe going to the box refers to marketing capability change[17]; that is, the rate of capability increase or decrease. The rate of capability

change is a function of marketing resources endowment[3] (Sterman 2000). The temporal accumulation of marketing capability, MC(t), can be represented in a differential equation form (Equation 1), where MC(0) is the initial value of marketing capability.

0

( ) ( ) (0) i

MC t MarketingCapabilityChange t dt MC= +

(Eq. 1)

Similarly, customer base[1] is also a stocks variable for which change is determined by the rate of customer net increase[5]. The temporal accumulation of customer base[1] can be repre- sented in a differential equation form (Equation 2).

0

( ) ( ) (0) t

C t CustomerNetIncrease t dt C= +

(Eq. 2)

where customer net increase(t) represents the net number of customers that flow into the customer base[1] at time t, and C(0) is the initial value of the customer base. Customer net increase[5] at time t is the difference between the number of competitor customers converted and the number of the focal firm customers lost to the competitor at time t.

The other variables in Figure 4 are either (1) auxiliaries repre- senting formulas providing information relevant to and mean- ingful within the system (these auxiliaries and formulas compute a derived value from other variables linking to them) or (2) constant variables representing numeric values exter- nally assigned. For instance, revenue[2] is computed as the product of customer base[1] and revenue per customer[13]. Marketing resources endowment[3] equals a portion of revenue[2] depending on the percentage for marketing[14] and

1310 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Figure 3. Selected Possible Firm Behavior Patterns Resulting from Interactions between a Reinforcing Loop and a Balancing Loop

Figure 4. Stocks and Flow Model of Marketing Resources Endowment Loop

is an input to marketing capability change[17] rate at time t. Table 3 summarizes the computational representations and their corresponding theoretical justifications. The causal loop diagram, augmented with stocks and flows (Figure 4), thus conceptualizes the circular and temporal causalities between firm resource endowment and performance.

The model shown in Figure 4 is based on the classic resource- based view (RBV). It is distinguished from the existing variance models in that it uses SD and is drawn from the system perspective. It provides a foundation for under- standing how e-commerce resource endowment in a tradi- tional firm would result in drastically different performance outcomes over time. Next, we seek to add the e-commerce resource endowment process to the base model for simulation analysis.

E-Commerce Model: Model Conceptualization

First, we define the key theoretical constructs relating to e- commerce resource endowment, using the roadmap of simulation methods for theory development in general (Davis et al. 2007) and the aforementioned SD modeling practices in particular (Sterman 2000). Then, we deduce the causal links among the constructs, drawing on RBV and the e-commerce literature, and add them to the base CLD form (shown in Figure 5). We conclude the model representation by listing computational representations for all the variables.

The performance implications of e-commerce resource endowment can be understood from the RBV perspective, just as in the offline situation (Hulland et al. 2007; Zhu 2004). A

MIS Quarterly Vol. 42 No. 4/December 2018 1311

Fang et al./System Dynamics Modeling for IS Research

Table 3. Summary of SD Model Variables Variable Description Dynamic Relationship Potential Measure

Customer Base The group of customers who purchase goods or services from the focal company.

Customer base increases when more potential customers are converted to current customers and decreases when there is loss of current customers. Marketing capability is a basic force that builds customer base (Day 1994; Rust et al. 2002).

Number of customers of the focal company.

Competitor's Customer Base

The group of customers who purchase goods or services from the com- petitor (in an aggregate manner).

Competitor's customer base increases when there is net loss of the focal firm’s current cus- tomers. Marketing capability is a basic force that builds customer base (Day 1994; Rust et al. 2002).

Number of customers of the competitor.

Marketing Capability

The ability to attract new and retain existing customers.

Marketing resources endowment can improve the firm’s marketing capability (Telser 1961). High marketing capability leads to a net increase in customers by gaining new cus- tomers from competitors and retaining existing customers (Capron and Hulland 1999).

A scale variable initially set at 1. When ability increases, it exceeds 1. When ability decreases, it falls below 1.

Front-end Capability

The ability to deploy and leverage firm resources to support Internet- based interactions with customers .

Building this capability requires resources from both marketing and information technology functions (Hulland et al. 2007). Front-end capa- bility helps firms bypass fierce competition in traditional channels (Porter 2001; Zhu 2004).

A scale variable assessed as how much more efficient it could help marketing capability.

Figure 5. Full Causal Loop Diagram: E-Commerce Resource Endowment

1312 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

firm’s e-commerce initiative is a process of building e- commerce capability, defined as a firm’s ability to deploy and leverage e-commerce resources to support the customer order life cycle (Hulland et al. 2007). We focus on front-end e- commerce by capturing the notion of front-end e-commerce capability, which is defined as the ability to deploy and leverage firm resources to support Internet-based interactions with customers (hereafter, front-end capability). These inter- actions include information presentation (e.g., presenting useful information on the company website), transactions (e.g., facilitating online transactions such as taking orders, security identification, online payments), and customization (e.g., product configuration, content personalization, real-time support) (Zhu 2004). In accordance with the definition of resource (Wernerfelt 1984), front-end e-commerce resource endowment is defined as the physical, human, organizational, and monetary resources used to build front-end e-commerce capabilities, and it is measured in monetary terms.

With the e-commerce-related constructs defined, formal causal links are deducted and connected with the other constructs in the model. The causal loop diagram in Figure 5 extends Figure 2 by incorporating the two e-commerce-related constructs; namely front-end e-commerce resources endow- ment and capability. Reinforcing loop R2 represents the circular effect of front-end e-commerce resources endowment. Higher revenue[2] makes more financial resources available for front-end e-commerce resources endowment[7] (Chandy et al. 2003; Kraemer and Dedrick 2002), which can then be used to build front-end capability[8] (Zhu and Kraemer 2002). Front-end capability[8] tightly connects marketing resources and becomes a key component of the firm’s marketing pro- cess by providing timely and accurate information for market analysis, brand-building, and targeted marketing and sales activities, thus leading to enhanced marketing capability[4] (Hoffman and Novak 2000; Saini and Johnson 2005). As noted earlier, marketing capability[4] leads to net customer increases[5] and, subsequently, to a larger customer base[1] and higher revenue[2] (Day 1994; Rust et al. 2002). Thus, rein- forcing loop R2 again reproduces a path-dependent behavior (Peteraf 1993).

Up to now, reinforcing loops R2 (front-end capability building) and R1 (marketing capability building) both repli- cate the general principles of RBV. However, it is important to note that there is a potential negative association between front-end e-commerce resource endowment [7] and marketing resources endowment[3]. Front-end capability[8] represents the extent to which a firm uses Internet-based technology to enhance the customer-oriented downstream of the value chain. Building this capability draws on resources from both the marketing and information technology functions (Hulland et al. 2007) and this potentially compromises the amount of

resources that would otherwise be committed to traditional marketing activities (Chandy et al. 2003; Schmidt et al. 2000). Indeed, Hulland et al. (2007) suggest that a negative corre- lation exists between resource commitment to established offline channels and emerging online channels in click-and- mortar retailers. This effect is represented by adding a nega- tive causal link from front-end e-commerce resource endowment[7] to marketing resources endowment[3] in Figure 5. Moreover, there is another balancing loop associated with how competitors respond[9] to the firm which increases its marketing capability, B3. Competitors will execute a certain marketing strategy to win customers from the focal firm with a high marketing capability. The key concepts and how they are linked, based on the existing literature, are summarized in Table 3.

E-Commerce Model: Model Formulation

After specifying the full causal loop diagram, we augment the model with a stocks and flow structure (Figure 6 shows the stocks and flow model presenting key constructs). As noted earlier, capabilities develop through resource inflows over time and deplete as they become obsolete over time (Dierickx and Cool 1989; Warren 2002); that is, capabilities take time to change. Therefore, front-end capability[8] is represented as the accumulation of difference between the resource inflow rate and the resource depreciation rate. Table 4 lists the full mathematical equations for the stocks and flow variables, with justifications.

Model Validation and Simulation

To run the simulations, we built the full stocks and flow model (Figure 6) using Vensim (Ventana Simulation Environ- ment),3 a popular SD simulation software (Eberlein and Peterson 2000; Sterman 2000). Vensim is an interactive soft- ware environment for developing and simulating SD models. It features a graphical user interface that enables researchers to quickly sketch causal loop diagrams and capture the feed- back loops, stocks, and flows they identify (Sterman 2001). Vensim has been effectively used in many organizational studies (e.g., Garcia et al. 2003; Georgantzas 2003; Repenning 2002).

3Other SD-based simulation tools are Powersim (Sterman 2000), Stella (Rich- mond et al. 2006), and iThink (Dutta and Roy 2005).

MIS Quarterly Vol. 42 No. 4/December 2018 1313

Fang et al./System Dynamics Modeling for IS Research

Table 4. Model Variables, Equation (Description), Dimensions, and Explanation Category # Variable Equation (Description) Dimension Explanation on Validity

Stocks (Endogenous)

[1] Customer Base

= INTEG (Customer Net Increase, initial customer base)

Customer Valid by algebraic definition.

[4] Marketing Capability

= INTEG (Marketing Capability Change, 1) Dimensionless Valid by algebraic definition. Level of 1 assumed as the initial equilibrium.

[6] Competitor’s Customer Base

= INTEG (Front-End Capability Building – Front- End Capability Depreciation, 0)

Customer Valid by algebraic definition.

[8] Front-End Capability

= INTEG (– Customer Net Increase, initial competitor customer base)

Dimensionless Valid by algebraic definition. Level of 0 as no front-end capability exists at the beginning.

Flows (Endogenous)

[5] Customer Net Increase

= INTEG (Front-End Capability Building – Front- End Capability Depreciation, 0)

Customer/ Month

Customer net increase is the difference between new customers attracted and lost. The equation adapted from the market diffusion model (Sterman 2000, 2001); marketing capability attracts new and retains existing customers (Day 1994).

[17] Marketing Capability Change

= (Marketing Resource Endowment / Required Marketing Resource) × (1 + Front-End Capability) – Marketing Capability) / Time to Adjust Marketing Capability

Dimensionless/ Month

Capability is accumulated by com- mitting resources and depreciated due to obsolesence (Dierickx and Cool 1989); Front-End Capability adds to matching capability (Hoffman and Novak 2000; Saini and Johnson 2005).

[23] Front-End Capability Building

= Delay fixed (“Front-End E-Commerce Resources Endowment” / “Total Desired E- Commerce Resources” / “Time to Develop Front-End”, 3, 0)

Dimensionless/ Month

Capability is accumulated over time by committing resources (Dierickx and Cool 1989), delayed fixed is a Vensim function that returns the value of the input delayed by a fixed delay time, 3 months reasonably assumed in our case.

[26] Front-End Capability Depreciation

= Front-End Capability / Front-End Depreciation Time

Dimensionless/ Month

Capability is depreciated over time due to obsolesence (Telser 1961)

Auxiliaries (Endogenous)

[2] Revenue = Revenue per Customer × Customer Base DOLLAR Valid by algebraic definition. [3] Marketing

Resource Endowment

= Revenue × Percentage for Marketing – Front- End E-Commerce Resources Endowment × Percentage of E-Commerce Resources from Marketing

DOLLAR Traditional marketing resources may be compromised in favor of e-commerce initiatives (Chandy et al. 2003; Schmidt et al. 2000).

[7] Front-End E-Commerce Resource Endowment

= Revenue × Percentage of Revenue to E-Commerce

DOLLAR Valid by algebraic definition.

[9] Competitor’s Response

= -0.25 × Marketing Capability + 1.25 Dimensionless The linear function models the com- petitor’s reaction to the focal company by increasing or decreasing its capa- bility. The line goes through (1, 1) and (2, 0.75) representing two situations: (1) if the focal company does not change its capability, the competitor remains unchanged too, and (2) if the focal company increases its capability by 100%, the competitor reacts by increasing its capability by 33% (1/0.75). This simple linear function was used based on the panel input.

[10] Conversion Rate

= Norman Conversion Rate × Competitor’s Response

Dimensionless Valid by algebraic definition.

[13] Revenue per Customer

= Norman Revenue per Customer × Revenue Change Over Time

DOLLAR Valid by algebraic definition.

1314 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Table 4. Model Variables, Equation (Description), Dimensions, and Explanation (Continued) Category # Variable Equation (Description) Dimension Explanation on Validity

[15] Required

Marketing

Resource

= Revenue per Customer × Percentage for

Marketing × Customer Base

DOLLAR/

Customer

Valid by algebraic definition.

[18] Market Share = Customer Base / (Competitor’s Customer

Base + Customer Bas)

Dimensionless Valid by algebraic definition.

[19] Profit = Revenue × (1 – Other Cost) – Front-End

E-Commerce Resources Endowment –

Marketing Resource Endowment

DOLLAR Valid by algebraic definition.

[22] Total Desired

E-Commerce

Resources

= Revenue per Customer × Initial Customer

Base × 0.2

DOLLAR Total desired e-commerce resource is

about 20% of total revenue, advised by

the marketing panel.

[30] Initial

Customer

Base

= Initial Market Share × Total Market Size Customer Valid by algebraic definition.

[33] Initial Competi-

tor Customer

Base

= Total Market Size × (1 – Initial market share) Customer Valid by algebraic definition.

[34] Revenue

Change Over

Time

= 1 + 0.42% × month Dimensionless Revenue per customer increases 5%

every year, reasonably assumed based

on the panel input.

Constants

(Exogenous)

[11] Loss Rate Natural customer loss rate (= 10%) /Month Consulted with the marketing panel.

[12] Contact Rate Number of contacts made between an existing

customer and a potential customer (= 1/3)

Contact/Month/

Customer

Consulted with the marketing panel.

[14] Percentage for

Marketing

Percentage of revenue allocated to marketing (=

20%)

Dimensionless Consulted with the marketing panel.

[16] Time to Adjust

Marketing

Capability

Months required to fully transform new

marketing resources to capability (= 3)

Month Consulted with the marketing panel.

[20] Other Cost Total other cost occurred (= 60%) Dimensionless Consulted with the marketing panel.

[24] Time to

Develop Front-

End

Months required to fully implement e-commerce

resources( = 12)

Month Consulted with the marketing panel.

[25] Front-End

Depreciation

Time

Months needed for full depreciation (= 12) Month Consulted with the marketing panel.

[27] Normal

Revenue per

Customer

Revenue gained per customer at the beginning

(= 500)

DOLLAR/

Customer

Scenario setting.

[29] Normal

Conversion

Rate

Normal possibility of converting a potential

customer through word of mouth (= 40%)

/Contact Consulted with the marketing panel.

[31] Total Market

Size

The whole market size (= 800) Customer Scenario setting for simplicity. The

behavioral patterns of interest do not

change qualitatively even when this

assumption is relaxed.

[32] Initial Market

Share

Market share of the local company at the

beginning (25%)

Dimensionless Scenario setting.

Constants [21] Percentage of

Revenue to

E-Commerce

Percentage of revenue allocated to e-commerce

(= 5%)

Dimensionless Policy variable for scenario generation.

[28] Percentage of

E-Commerce

Resources

from Marketing

Percentage of e-commerce resources that are

drawn from the marketing function (= 0, 100%,

50%)

Dimensionless Policy variables from scenario

generation.

Note: *INTEG represents integration function.

MIS Quarterly Vol. 42 No. 4/December 2018 1315

Fang et al./System Dynamics Modeling for IS Research

Figure 6. Full Stocks and Flow Model: E-Commerce Resource Endowment

Key Assumptions

We assume that the firm described in the earlier section “System Logical Structure and System Dynamics Modeling” operates in a market of a fixed size for the sake of model sim- plicity.4 We assume that the firm holds 25% of market share, while the aggregated competitor holds 75%. The firm’s resource endowment budget for both traditional marketing and e-commerce is drawn from revenue only (i.e., other financing means are excluded).

Model Validation

Model validation is an important aspect of any model-based methodology (Davis et al. 2007) including SD (Barlas 1996; Richardson and Pugh 1981). Model validation takes place after the individual model has been formulated, and before any behavior analysis is conducted or policy recommenda- tions made. SD models are causal-descriptive, so they should not only reproduce or predict behavior patterns but also explain how the behavior patterns are generated; that is, the

right output behavior for the right reasons. SD deploys three test categories: direct structure tests, structure-oriented behavior tests, and behavior pattern tests (Barlas 1996). Our model passed all three. Table 5 lists the tests conducted, the procedures recommended by the literature, and how they were applied in the current study.

Before conducting these tests, it is worth assessing the overall validity of the model through a case study. The case in point is a real click-and-mortar firm, Best Buy (NYSE:BBY), a leading consumer electronics retailer. The U.S. consumer electronics industry has been saturated with fierce competition since the late 1980s. As an industry leader, Best Buy launched its e-commerce Web site in 1998 and e-commerce has been part of its core strategy ever since. Best Buy is typical of traditional companies moving into the e-commerce world. We retrieved historical news reports on the company’s e-commerce initiative and interviewed a former executive who was involved in the e-commerce project from the beginning. The interview revealed that Best Buy considered its Web site to be an essential marketing and sales channel that complemented its chain of about 1,300 physical stores (Radigan 2001) and that it had spent about 2 years developing the online channel. At the time, the e-commerce project was funded by an extra Board-approved budget allocation (i.e., the e-commerce budget was not redirected from existing mar- keting budgets). However, marketing and sales staff had to spread themselves across both channels, which is consistent

4We examined the assumption of fixed market size by conducting sensitivity analyses with varied market growth rates; the resulting behavioral patterns from model runs under different market growth rates remain qualitatively comparable, thus leading to the similar theoretical insights we yield in the paper.

1316 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Table 5. Model Validation

Category Test What to Test Common Tools and

Procedures Examples from This Study

Direct

Structural

Tests

(Forrester

and Senge

1980;

Richardson

and Pugh

1981)

Structure

Assessment

Model structure is

consistent with

relevant descriptive

knowledge of the

system.

Use causal diagrams, stock

and flow maps or direct

inspection of model equations.

Use interviews, workshops to

solicit expert opinion, archival

materials, direct inspection of

system processes.

Major elements are built upon

prior literature. A panel of

experts and professionals

review the causal diagram and

stock and flow model.

Parameter

Assessment

Parameter values

are consistent with

and reasonable to

descriptive and

numerical knowledge

of the system.

Use statistical methods to

estimate parameters.

Use judgmental methods

based on interviews, expert

opinion, focus groups, direct

experience.

Judgmental methods are

applied by interviewing with a

group of marketing and

professionals.

Boundary

Adequacy

The important con-

cepts for addressing

the problem are

endogenous to the

model.

Use model boundary charts,

variable list, and/or CLD to

explicitly present endogenous

and exogenous variables for

inspection.

Key concepts of resources and

performance are endogenous in

the model. Boundary variables

are explicitly listed

Dimension

Consistency

Each equation is

dimensionally consis-

tent without the use

of parameters having

no real-world

meaning.

Use dimensional analysis

software.

Inspect model equations for

suspect parameters.

Model passed dimension

consistency check utility in

Vensim DSS.

An expert inspected model

equations.

Structure-

Oriented

Behavior Test

(Barlas 1996;

Forrester and

Senge 1980;

Sterman

2000)

Extreme

Conditions

Model responds

plausibly when

extreme policies

apply.

Test response to extreme

values of each input, alone

and in combination.

Model exhibited anticipated

behaviors when extreme values

were assigned to the constants

in the model.

Sensitivity

Test

This is the extent to

which numerical

values and behaviors

change significantly.

Test response to a set of

varying numeric values and

see if model-generated

behavior is consistent with the

real system.

Tested the model with a set of

varying values on key resource

variables (e.g., marketing

expense) The model generated

anticipated behaviors.

Integration

Error

Results are not

sensitive to the

choice of time step or

numerical integrating

method.

Cut the time step in half and

test for changes in behavior;

use different integration

methods and test for changes

in behavior.

Cut the time step in half; model

was run using both Euler and

Runge-Kutta Methods (Sterman

2000).

Behavior Test

(Barlas 1996)

Behavior

Reproduction

Model reproduces

the behavior of

interest in the

system.

Compare model output and

data qualitatively; such as

modes of behavior, shape of

variables, relative amplitudes.

Compute statistical measures

of correspondence between

model and data.

Our model replicated Best Buy’s

behavior pattern over ten years

from the year of starting e-

commerce, in terms of front-end

e-commerce capability-building

and revenue, respectively.

MIS Quarterly Vol. 42 No. 4/December 2018 1317

Fang et al./System Dynamics Modeling for IS Research

with our model in which traditional resources subsidize the online channel. The executive we interviewed agreed that our model is generally consistent with his/her experience of the mechanisms of resource allocation between offline and online channels. With this result, we went on with the other model validation tests. Note we also used this case study for behavioral pattern tests, reported later in the paper.

Direct structure tests include assessments for structure, parameter, boundary, and dimension (Forrester and Senge 1980). Structural assessment is used to examine the validity of the model structure by theoretical and empirical com- parison with the real system structure. In the foregoing section, each causal link in our model has been theoretically justified by grounding the arguments in the existing RBV and e-commerce literature and then presenting them in a causal loop diagram, thereby demonstrating theoretical structure validity. We invited a panel of two marketing professionals from the consumer electronics industry and two scholars (one from marketing and the other from SD) to review and endorse the model structure, which is a common practice in the SD field (Sterman 2000). Parameter assessment means evalu- ating the constant variables against knowledge of the real system, both conceptually and numerically (Forrester and Senge 1980). We asked the expert panel to confirm that the constant variables used in the model were conceptually identifiable and numerically reasonable, compared to the real system. The boundary adequacy test verifies whether the important concepts are endogenous to the model. The key concepts of capabilities, resources, and performance variables (e.g., market share and profitability) were all confirmed as endogenous to the model. Table 3 lists the exogenous vari- ables (i.e., constant variables) and the endogenous variables (i.e., stocks and flow variables, auxiliary variables), the values of which are determined from within the model. Finally, the dimension consistency tests assess whether the dimensions of the right-hand side and left-hand side of the equation are internally consistent (Barlas 1996). The model passed the consistency check using Vensim (Sterman 2000).

Structure-oriented behavior tests indirectly assess the validity of the structure by focusing on model-generated behavior patterns (Barlas 1989), with the objective of un- covering potential structural flaws by observing model behaviors under different conditions. We applied three common tests of this kind: extreme-condition test, sensitivity test, and integration error test. The extreme condition test involves assigning extreme values to selected parameters and comparing the model-generated behavior to the anticipated behavior of the real system under the same extreme condition (Barlas 1996). We conducted several key extreme tests and observed the model behavior (see Table 6). For instance, if marketing percentage (i.e., percent of revenue allocated to

marketing) was adjusted to 100% and 0%, the company’s profitability in both situations dropped dramatically, as was expected. Overall, the model passed all the extreme tests. The sensitivity test determines the parameters to which the model is sensitive and verifies if the real system would exhibit similar sensitivity to the corresponding parameters (Barlas 1996). We conducted a set of sensitivity tests by observing how model-generated behaviors correspond to changes in the values of the key constructs. For instance, Appendix B shows the different model behaviors for market share when marketing resources endowment was set at different numeri- cal values while other factors were kept unchanged. All the simulation runs exhibit reasonable behavior; that is, market share expanded correspondingly as marketing resources increased, and eventually reached an equilibrium state due to the limit to growth. Finally, the integration error test assesses whether the results are sensitive to the choice of time step or numerical integrating method (Sterman 2000). When it is halved, the results are consistent; thus, the model passes the integration error test.

The behavior pattern test assesses whether the model accu- rately reproduces the major behavior patterns exhibited by the real system, such as growth, decline, and oscillation (Forrester and Senge 1980). We conducted this test by referring to the behavior of Best Buy, our case study firm. We identified a set of the retailer’s reference behavior patterns from the prac- titioner material and tested the model’s ability to reproduce them. Best Buy’s temporal behavior on front-end capabilities was plotted by analyzing the contents of its Web site, a method adopted by Zhu (2004). Instead of focusing on Best Buy’s current Web site, we analyzed the contents of the site’s archives from the past 10 years (1998–2007) using Internet Archive, a non-profit digital library of Web sites (www.archive.org). Two research assistants majoring in IS independently coded Best Buy’s archived historical Web sites; any differences between the two coders were resolved in discussion with one of the coauthors. Appendix C shows the reference mode behavior plotted using data from Best Buy with a smooth trend-line added to enhance the visual effect. Appendix D shows the front-end capability accumulation behavior over the 10 years, generated by the model. We observed a similar pattern between these two plots: a rapid building of capability in earlier years followed by a slow- down. Thus, the model is able to replicate Best Buy’s actual front-end capability accumulation process.

As another example of the behavior pattern test, we plotted Best Buy’s revenue from 1997 to 2007, collected from Standard and Poor’s Compustat Database. Revenue data were adjusted by discounting against the growth rate of the U.S. consumer electronics industry for each year (collected from the U.S. Census Bureau) in order to observe the revenue pat-

1318 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Table 6. Extreme Tests Result Summary No. Extreme Tests Parameter Setting Result 1 No investment in marketing Percentage for Marketing = 0 Passed 2 Extreme long time to build Front-end capacity Time to Develop Front-end = 1000 month Passed 3 No customer loss Loss rate = 0 Passed 4 Extreme high customer loss rate Loss rate = 100% Passed 5 Extreme low contact rate Contact rate = 0 Passed 6 Extreme high contact rate Contact rate = 100% Passed 7 Extreme low revenue per customer Normal revenue per Customer = 1

dollar/customer Passed

8 Extreme high revenue per customer Normal revenue per Customer = 10000 dollar/customer

Passed

9 Extreme high market size Total market size = 10000 customer Passed 10 Extreme low customer conversion rate Normal conversion rate=0 Passed 11 Extreme high customer conversion rate Normal conversion rate=100% Passed 12 Extreme low initial market share Initial market share = 0 Passed 13 Extreme high initial market share Initial market share=100% Passed 14 Extreme long time to adjust marketing

capability Time to Adjust Marketing Capability = 100 month

Passed

15 Extreme long time for front-end capability to depreciation

Front-end Depreciation Time = 100 month Passed

tern under the premise of zero market growth, an assumption underlying the model. Best Buy’s adjusted revenue pattern is shown in Appendix E, which indicates that Best Buy’s reve- nue stagnated and even slightly declined between 1997 and 1999, followed by strong growth afterwards. A similar reve- nue pattern was observed using the model-generated behavior; that is, revenue declined in the first year and was followed by strong growth from that point forward (Appendix F). Thus, the model is able to replicate Best Buy’s actual revenue pattern.

Overall, we rigorously performed the three sets of tests recommended for validating SD models. The results indicate that the model structure has sound face validity, that the model behavior is sensible under extreme conditions and differing sensitivity scenarios, and that the model is capable of replicating some of the major reference behavior patterns observed in a real company. Thus, the model is a reasonably valid representation of the underlying processes of capability building and performance.

Simulation Analysis

Following this rigorous validation process, we then simulate the model to address our research question: How and why can an initial e-commerce resource endowment by a tradi- tional firm improve or deteriorate the firm’s performance over time? Specifically, we experiment by varying the values

of the key resource endowment variables to understand the temporal interactions among the two resources and firm performance. We used market share as the performance indi- cator.5 A major advantage of simulation methods is the abil- ity to conduct “what if” analysis to examine how the model behavior would respond to varying initial conditions of key variables (Sterman 2000).

We concentrated on simulating two sets of managerial deci- sions, with two scenarios each for comparison, related to (1) the amount of additional resources that should be invested in traditional marketing or e-commerce and (2) if e-commerce is selected, from where the additional resources should be drawn.

In case (1), if the firm decided to spend 5% of its revenue on strengthening marketing capability, it would have three bud- getary allocation choices: full endowment in traditional marketing resources (Scenario 1), full endowment in front-end e-commerce resources (Scenario 2), or splitting the endow- ment (e.g., 50/50) between traditional marketing resources and front-end e-commerce resources (Scenario 3).

5Market share is a performance indicator equivalent to revenue in our example. We also included profit as a second performance indicator, further enhancing our confidence in the simulation results. This analysis is sum- marized in Table 7 but not discussed in the text for the sake of brevity. The results on profit behavior are available upon request.

MIS Quarterly Vol. 42 No. 4/December 2018 1319

Fang et al./System Dynamics Modeling for IS Research

Figure 7. Market Share Under Scenarios on Different Investment Choices

The simulation model was run to generate these three scen- arios, together with the base scenario; namely, no additional investment. Figure 7 presents the behavior patterns of market share under these different scenarios. Under the base scenario (i.e., no additional investment), the model stayed in equi- librium with the original 25% market share. The base is the benchmark against which other simulation runs are compared. In Scenario 1, the budget was invested in traditional mar- keting activities only (S1 in Figure 7) and market share grew exponentially in the first 24 months to gradually arrive at a higher equilibrium status (about 35% market share). In Scenario 2, the budget was spent on front-end e-commerce capability-building (S2 in Figure 7); market share picked up slowly in the first 12 months and significantly lagged behind the growth path of Scenario 1. However, market share grew exponentially after month 24, and eventually outperformed Scenario 1, starting from month 48. Market share eventually achieved a state of equilibrium at about 40% share around month 120, which is 15% more than that of the base scenario and 10% more than that of Scenario 1. In Scenario 3, half the budget was spent on traditional marketing and half was spent on front-end e-commerce capability-building (S3 in Figure 7). In the beginning, Scenario 3 market share was higher than Scenario 2, but not as high as Scenario 1. Ultimately, the ma- rket share of Scenario 3 arrived at a point between Scenario 1 and Scenario 2.

Scenarios 4–6 present the material decisions about reallo- cating resources from the existing marketing function to front- end capability building. As noted earlier, e-commerce capa- bility building is a cross-functional effort requiring commit- ment and collaboration from both the marketing and IT

functions (Chandy et al. 2003; Schmidt et al. 2000). Scenarios 4, 5, and 6 simulate situations in which 0%, 50%, and 100% of the e-commerce budget was drawn from marketing resources, respectively. Scenario 4 serves as a reference pattern, simulating the situation where no e-commerce budget was drawn at the expense of existing marketing resources. Curve S4 in Figure 8 shows that market share started to increase after the first six months, grew exponentially after that, and finally reached an equilibrium at about 40% market share. Scenario 5 simulates the situation where half of the e- commerce budget was drawn from existing marketing resources. Curve S5 in Figure 8 shows that market share initially declined below the original equilibrium state for the first 30 months, followed by a relatively flat growth slope, eventually arriving at a moderate equilibrium state of about 32% share. Scenario 6, on the other hand, assumes that the entire e-commerce budget was drawn from existing marketing resources (S6 in Figure 8). Here, market share declined exponentially despite the reallocated resources being invested in front-end capability, eventually arriving at a lower equilibrium state of 18% share. Table 7 summarizes the values of the scenario variables, their respective market share patterns, and the end-of-run values of market share.

Result Discussion

The simulation results of S1–S3 allow us to extend the existing e-commerce literature by directly observing the tem- poral interaction between e-commerce resource endowment and performance. Comparing Scenario 1 (S1, traditional mar- keting resource endowment) and Scenario 2 (S2, e-commerce resource endowment), we observe in Figure 7 that S1 market

1320 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Figure 8. Market Share Under the Scenarios on Different Investment Sources

Table 7. Simulation Analysis Results

Scenario Input Values of

Scenario Variables

Performance Behavior End-of-Run Values of

Performance Variables

Market Share Profit Market Share Profit

Base: No Additional Investment

No change Equilibrium throughout. Equilibrium throughout. 25% 22,000

S1: Invest in Traditional Marketing

Additional 5% of the revenue invested in marketing

Exponential growth in the first 24 months, followed by diminishing growth. Achieve equilibrium at month 48.

Profit growth exponentially through month 24, followed by diminishing growth. Achieve equilibrium around month 48.

36% 23,750

S2: Invest in E-Commerce (resources drawn from outside marketing)

Additional 5% of the revenue invested in e- commerce

Lagged growth in the first 12 months, followed by strong growth between month 12–48. Achieve equilibrium around month 120.

Profit growth slow in the first 12 months, followed by exponential growth between months 12–48. Achieved equilibrium around month 120.

40% 26,350

S3: Invest Half in Traditional Marketing and Half in E-Commerce

Additional 5% of the revenue invested evenly between mar- keting and e-commerce

Growth between scenario 1 and 2. Equilibrium level between scenario 1 and 2.

Profit growth steady in the first 24 months, followed by diminishing growth. Achieved equilibrium around month 120.

39% 25,550

S4: Invest in E- Commerce (resources not drawn from within marketing)

5% increase in e- commerce resource endowment, none of which deducted from marketing resource endowment

Lagged growth in the first 12 months, followed by strong growth between month 12–48. Achieve equilibrium around month 120.

Profit growth slow in the first 12 months, followed by an exponential growth between months 12–48. Achieved equilibrium around month 120.

40% 26,350

S5 Invest in E- Commerce (resources partially drawn from within marketing)

5% increase in e- commerce resource endowment, half of which deducted from marketing resource endowment

Growth declined in the first 24 months, followed by relative slow growth. Achieved equilibrium around month 120.

Profit declined in the first 12 months, followed by moderately strong growth. Approached equilibrium around month 120.

31% 24,080

S6: Invest in E- Commerce (resources all drawn from within marketing)

5% increase in e- commerce resource endowment, all of which deducted from marketing resources endowment

Growth exponentially declined from the beginning. Approached equilibrium around month 96.

Profit exponentially declined from the beginning. Approached equilibrium around month 120.

18% 16,100

MIS Quarterly Vol. 42 No. 4/December 2018 1321

Fang et al./System Dynamics Modeling for IS Research

share increased exponentially, exceeding that of S2 in the first 48 months. Market share in S2 outgrew S1 afterward, even- tually arriving at a higher equilibrium state. These contrasting performance behaviors at different times are the result of the change in dominance of the temporal reinforcing effects of the marketing versus e-commerce resource endowment loops. In S1, the reinforcing loop of marketing resources endowment (R1 in Figure 5) dominated at the early stage when resource endowment in the marketing function rapidly enhanced existing marketing capability (see S1 in Figure 9), leading to strong short-term growth. This growth effect was later coun- teracted by the limit to growth caused by balancing loop B1. In contrast, in S2, the reinforcing loop of e-commerce resource endowment (R2 in Figure 5) dominated in the early stage. However, it took time to build front-end capability from scratch (S2 in Figure 9) and to influence market share by adding to marketing capability (see S2 in Figure 7), thus growth was slower in the short-term. The reinforcing effect of front-end capability increased revenue later, which in turn made more resources available for building both marketing and e-commerce capabilities and which led to superior per- formance in the long term. The time delay to build front-end capability is the reason behind the observed “worse before better” performance behavior in S2. For Scenario 3, at the beginning, R1 generates some increase in market share; later when the front-end capability starts to take effect, R1 and R2 both helps market share increase. We summarize the mech- anisms underlying the different temporal behavior in the three scenarios in Table 8.

These simulation results provide a tractable theoretical explanation for reconciling the inconsistent findings on the performance outcomes of e-commerce resource endowment in the extant literature. The results suggest that performance resulting from e-commerce resource endowment may improve in the long term, but suffer in the short term because it takes time to build e-commerce capabilities. Thus, earlier inconsis- tent results may simply be due to the timing of the empirical investigation. If firm performance is observed a long time after e-commerce is deployed (e.g., over 5 years), a positive relationship is more likely (Hulland et al. 2007; Zhu 2004). On the other hand, if firm performance is observed in the early stage of e-commerce deployment, such as in the late 1990s, a negative relationship is more likely (Coughlan et al. 2001; Han and Noh 1999). Thus, our SD model adds to the e-commerce literature by revealing that e-commerce resource endowment will improve long-term, but not short-term, per- formance. This contrast in the short-term versus long-term effect of e-commerce resource endowment demonstrates the intriguing implications that SD can yield.

Scenarios 4–6 take a step further in accounting for the inherent tension in the process by adding a negative link

between e-commerce resource endowment and marketing resource endowment in Figure 5 (i.e., the addition of loop B2). In Scenario 4, building front-end capability did not drain any resources from existing marketing capability. The equi- librium state of the market share was maintained while the front-end capability was built. When the front-end capability started to generate positive effect (see S4 in Figure 10), marketing capability increased, attracting more customers and generating higher revenue, which means more resources were available to invest in traditional marketing and front-end capability. Both R1 and R2 loops (in Figure 5) generated exponential growth until the balancing loop B1 (limit to growth) started to dominate. The market share approached an equilibrium state at a higher level.

In Scenario 5, where half of the e-commerce resources were drawn from existing marketing resources, we observed that market share declined from the initial state of equilibrium in the base scenario then grew exponentially between months 24 and 60 (see S5 in Figure 8). This decline was an immediate side effect of e-commerce resource endowment, due to the dominating effect of reinforcing loop R1 in a vicious way. Fewer marketing resources caused revenue to fall, making even fewer resources available for subsequent resource endowment in marketing and e-commerce. Only after the front-end capability started to build up and generate revenue (see S5 in Figure 10) did both R1 and R2 start to have a positive effect, improving firm performance.

Scenario 6, where all the e-commerce resources were drawn from existing marketing resources, represents a business failure (i.e., exponential decline of market share and profit). Loop R1 dominated in a vicious way, generating exponential decay of the market share (see S6 in Figure 8). When front- end capability started to function and created a temporary rebound of marketing capability between months 12 and 24 (S6 in Figure 10), the customer base had already dropped so much that the positive effect of R1 and R2 was too low to rebound. The system was dominated by the balancing loop B1 that kept market share at a low equilibrium level. With this stagnated low-level market share, resource endowment remained low, hence the low level of marketing capability. We summarize the mechanisms underlying the different tem- poral behavior in the three scenarios in Table 9.

Our findings reveal why draining existing resources to build e-commerce capability might undermine firm performance: the positive performance effect from investing in emerging capabilities is often postponed, but the negative effect of cutting resources from critical existing functions occurs immediately. A significant and immediate deterioration of firm performance would be negatively path-dependent, decreasing financial resources that could otherwise endow the

1322 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Figure 9. Marketing Capability Under the Scenarios on Different Investment Choices

Table 8. System Behavior and Dominant Loop Analysis for Initial Level of Investment Scenario Performance Behavior Dominant Loop Identification Remarks S1 Exponential growth

followed by equilibrium R1 shifts to B1 Tipping points occurs when domin-

ant loop change from R1 to B1. S2 Slow growth followed by

exponential growth R2 (with delayed effects), shifts to R1, R2 With delayed effect, market share

of S2 grows slowly at the begin- ning. Later, R1, R2 work together to achieve a higher equilibrium

S3 Exponential growth followed by slow growth and then approaching equilibrium

R1 (weaker than S1) shifts to R1, R2 (weaker than S2)

Market share of S3 is in between of scenario 1 and scenario 2.

emerging capability-building process. Only those firms that minimize potential resource erosion in existing critical functions will be likely to enjoy the long-term performance benefits of emerging capabilities. Thus, our findings stress that the initial level of resources directed to e-commerce (and to traditional marketing channels) will be an important trade- off decision that will determine the success or failure of e- commerce initiatives.

Contributions, Limitations, and Future Research

Key Contributions

This study contributes to the IS field in five ways. First, it makes a novel contribution by introducing system dynamics

as a modeling tool for extending variance theory from the system perspective. We establish that, unlike the variance- based models commonly used in IS research, SD specializes in developing theory based on the system logical structure by understanding the dynamic causal structure underneath complex and dynamic system behaviors. It stresses the theo- retical logic that “structure determines behavior” and focuses on examining the circular and temporal aspects of causal structure (Forrester 1961; Sterman 2000). Our introduction of SD is novel in that it is distinguished from past SD appli- cations, which build SD models on empirical observations or logical speculation (Abdel-Hamid 1988; Amitava 2001; Black et al. 2004; Dutta and Roy 2005; Quaddus and Intrapairot 2001), rather than on the existing theory. Other SD appli- cations (in business fields such as general management and operations management) that build on existing theory do not question assumptions about the logical structure underlying an existing theory as a way of theory development (e.g., Größler

MIS Quarterly Vol. 42 No. 4/December 2018 1323

Fang et al./System Dynamics Modeling for IS Research

Figure 10. Market Capability Under the Scenarios on Different Investment Sources

Table 9. System Behavior and Dominant Loop Analysis for Source of Investment Scenario Performance Behavior Dominant Loop Identification Remarks S4 Slow growth, followed by

exponential growth, eventually approaching equilibrium

Dominant effect from R2 (with delayed effects),and then from both R1 and R2, and then from B1

Both reinforcing loops (R1 and R2) work positively to improve market share.

S5 Slow reduction followed by slow growth

Dominant effect from R1 (in a vicious way) and then from both R1 and R2 (in a positive way)

When R2 starts to work in a posi- tive way, R1 changed from taking vicious effect to virtuous effect.

S6 Exponential decay Dominant effect from R1 (in a vicious way) and then from B1

R1’s strong negative effect reduced customer base; then B1 starts to dominate the system and equilibrate at a low level.

et al. 2008; Sastry 1997). In contrast, the SD approach we demonstrate here builds on well-established variance theories by explicitly examining the underlying logical structure and extending them from the alterative, system perspective. This new approach is particularly timely as our field matures and begins to work with well-recognized variance theories. For instance, Burton-Jones et al. (2015) demonstrates at a general level that the system perspective can be used to add theo- retical insights to the IS success model. The SD modeling approach we introduce here could extend it further by reformulating the IS success model for simulation analysis from the system perspective, and subsequently observe and analyze dynamic behavioral patterns through model simu- lation. Similarly, extant IT adoption/post-adoption research, where there are known recursive relationships between technology beliefs and usage behavior (Kim and Malhotra 2005), can be remodeled using the SD approach to provide a richer understanding of the phenomena.

Second, our case study contributes to the e-commerce litera- ture by reconciling inconsistent findings observed in the existing research on e-commerce resource endowment that primarily draws on the variance perspective. By building and simulating an SD model based on the existing variance-based e-commerce research, we develop our theoretical under- standing of the complex temporal interactions between e- commerce resource endowment, traditional marketing resource endowment, and firm performance from the alter- native, system perspective. The insight gained from the simulation reconciles the inconsistent findings about the per- formance impact of e-commerce resource endowment in the existing literature (Hulland et al. 2007; Zhu 2004; Zhu and Kraemer 2002). In doing so, it moves our understanding toward a dynamic theory from the system perspective on why and how different initial levels of e-commerce resource endowment can result in drastically different performance outcomes in a nonlinear, counterintuitive manner.

1324 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Third, as demonstrated through the e-commerce case, our study makes an important contribution to IS research by providing comprehensive guidance on how to systematically build, validate, and simulate SD models. As discussed earlier and summarized in Table 2, the limited SD applications in IS research have not demonstrated a systematic process for model building and validation (Abdel-Hamid 1988; Abdel- Hamid and Madnick 1989; Dutta and Roy 2005; Michael et al. 2003). The model-building approach in these prior works is mainly based on logical speculation or case study, and thus is limited in extending theory from an existing body of literature. Moreover, most of these SD works reside at the conceptual level, without developing and validating simula- tion models. Those that do have not subjected the developed model to a rigorous validation process, a stage crucial for establishing the validity of SD models before simulation analysis. Drawing on and consolidating the original SD liter- ature, we address these methodological limitations by offering IS researchers a systematic guide on building, validating, and simulating SD models for extending variance theory.

Fourth, our study contributes to SD research by extending its applicability to the IS field. SD was originally used to model macro social systems with a long-term policy orientation (Barlas 1996; Meadows 1980). Correspondingly, its applica- tion in the business discipline mostly revolves around long- term to mid-term organizational policy issues characterized by stable, yet evolving, changes such as quality improvement (Sterman et al. 1997) and innovation diffusion (Repenning 2002). In contrast, most IS phenomena are fast-changing and focused on the short- to mid-term. As such, one cannot assume that SD is readily applicable to the IS context. By demonstrating that SD modeling can be used to further develop IS theories from the system perspective, we extend the applicability of the SD modeling approach to this rather different discipline. This is particularly pertinent as the IS field embraces the big data era. Big data accelerates change in business environments, making its related phenomena particularly volatile and dynamic. The SD approach intro- duced here could be a valuable tool for understanding a more rapidly changing business environment in the big data era.

Finally, the study shows that SD models can be an effective micro-management lab for practitioners to create scenarios that challenge managerial intuitions and sort the causal mechanisms underlying complex dynamic firm behavior (Morecroft 1984; Sterman 2000). Simulation runs based on validated SD models empower practitioners to conduct “what if” analysis using different initial values for important busi- ness variables. Doing so can bring to life the consequences of possible policy change and inform strategic business decisions (Probert 1982).

Limitations and Future Research

As with all research, our study has some limitations that call for future research. First, we only demonstrate how SD can be used to further develop Type IV theory (for explanation and prediction). This is reasonable, given the prevalence of Type IV theory in IS research; however, future research should grasp the opportunity to discuss how we can use SD to advance other types of theories, such as design theories that are arguably more unique to IS research (Gregor 2006; Hevner et al. 2004).

Second, our study focuses on extending variance theory by using SD from the system perspective. This is justified by the dominance of variance theory in IS research, but it is note- worthy that process theory is another logical structure that has received some scholarly interest in the field (Markus and Robey 1988; Parc et al. 2008; Shaw and Jarvenpaa 1997). Thus, to enhance the value of SD as a theory development tool, future research may investigate how one can extend process theory from the system perspective using SD. While we believe that a theory-based approach to building an SD model from an existing process theory is viable, we conjec- ture that the underlying assumptions of process theory in relation to SD should be carefully examined. For instance, while process theory often describes sequential events that lead to probabilistic (not necessarily deterministic) outcomes (Markus and Robey 1988), SD describes models using vari- ables (i.e., state and flow variables) that often imply deter- ministic outcomes (Sterman 2000). Thus, to extend process theory using SD, one might need to address such issues as conversion between events and variables and reconciliation of probabilistic and deterministic causal relationships. Given the distinctions of process theory, it would be highly valuable for future research to develop systematic guidance and demon- strate how SD can complement process theory from the system perspective.

Third, our study demonstrates the use of SD through one case. A multi-case approach may improve the generalizability of our approach. Nevertheless, we believe that the SD approach demonstrated in our study is widely applicable to many IS areas that investigate phenomena that feature feedback and time delays, such as IT adoption/post-adoption, human– computer interactions, and other strategic IS matters. This is because, like the e-commerce case in our study, many IS phenomena are open systems, characterized by continuous interactions with their environment and among components within the system (Burton-Jones et al. 2015). In other words, because of our interest in systems per se, the system perspec- tive is a natural fit for IS (Lee 2004). Thus, future research is encouraged to extend this SD approach, which models the system perspective, to more IS areas.

MIS Quarterly Vol. 42 No. 4/December 2018 1325

Fang et al./System Dynamics Modeling for IS Research

Fourth, future research may enhance the calibration and validation of the SD model by leveraging emerging informa- tion technologies, such as big data. The validation process elaborated in our study stems from the original SD literature, which assumed limited availability of data for model valida- tion (behavioral test). As data become increasingly abundant through the diffusion and application of information tech- nologies such as the Internet and big data, future research can capture a large volume and variety of data for model cali- bration and validation.

The SD model demonstrated for the case of e-commerce resource endowment is of limited scope, due to strict assump- tions applied for the sake of brevity in demonstrating SD modeling. For instance, the current study’s simplified model includes only two capability-building processes, traditional marketing and front-end e-commerce. Future researchers could look deeper inside the firm to uncover more sophisti- cated causal structures. For instance, research could consider other important capability-building processes, such as manu- facturing or supply chain management as well as their associated IT-enabled capabilities (Banker et al. 2006; Dong et al. 2017; Rai et al. 2006). In fact, there is great research potential in including both front-end and back-end (i.e., downstream and upstream) e-commerce capabilities in the same model in order to test their interactive behavior and performance implications, because empirical data for these processes are difficult to obtain simultaneously. Future research could also further analyze the constructs in the cur- rent model to develop a more in-depth understanding of the inner workings of these capability-building mechanisms. For instance, marketing capabilities can be further broken down into market analysis, brand management, and sales manage- ment (Day 1994). Similarly, front-end e-commerce capability can be categorized into informing, transacting, and custom- izing (Zhu 2004). Each of these sub-capabilities may have distinct temporal effects that require different resource endowments.

To conclude, the ultimate purpose of this study is to introduce system dynamics as a modeling tool for extending variance theory, which prevails in the maturing IS field, from the system perspective. The study fulfils its purpose by intro- ducing SD and demonstrating the SD approach for theory development through a case study of e-commerce resource endowment. In so doing, we make a contribution by demon- strating how to develop theory using SD to extend the variance-based studies. We also demonstrate the efficacy of the SD modeling in offering insights that reconcile incon- sistent findings in IS research. We call for more future research to apply SD to enrich existing theories or develop new theories in the IS discipline from the system perspective.

Acknowledgments

The authors are grateful to the senior editor, the associate editor, and the anonymous reviewers for their invaluable guidance and insightful comments. The first and third authors would like to thank the School of Management at Fudan University for the inspirational introduction to the fields of Information Systems and Systems Dynamics, and the Systems Dynamics group at the University of Bergen for research training. The first author also thanks the Information Systems area group at Ivey Business School, Western University, for research training. The work described in this article was partially supported by the National Science Foundation of China (Projects 71571155, 71522002, and 71371076) and the General Research Fund from the Research Grants Council of Hong Kong (Project CityU 11502116).

References

Abdel-Hamid, T. K. 1988. “The Economics of Software Quality Assurance: A Simulation-Based Case Study,” MIS Quarterly (12:3), pp. 395-411.

Abdel-Hamid, T. K., and Madnick, S. E. 1989. “Lessons Learned from Modeling the Dynamics of Software Development,” Com- munications of the ACM (32:12), pp. 1426-1438.

Akkermans, H., and van Helden, K. 2002. “Vicious and Virtuous Cycles in ERP Implementation: A Case Study of Interrelations between Critical Success Factors,” European Journal of Infor- mation Systems (11:1), pp. 35-46.

Alvesson, M., and Sandberg, J. 2011. “Generating Research Questions through Problematization,” Academy of Management Review (36:2), pp. 247-271.

Amitava, D. 2001. “Business Planning for Network Services: A Systems Thinking Approach,” Information Systems Research (12:3), pp. 260-283.

Ashby, W. R. 1956. An Introduction to Cybernetics, London: Chapman & Hall Ltd.

Banker, R. D., Bardhan, I., and Asdemir, O. 2006. “Understanding the Impact of Collaboration Software on Product Design and Development,” Information Systems Research (17:4), pp. 352-373.

Barlas, Y. 1989. “Tests of Model Behavior That Can Detect Struc- tural Flaws: Demonstration with Simulation Experiments,” in Computer-Based Management of Complex Systems: Interna- tional System Dynamics Conference, P. M. Milling and E. O. K. Zahn (eds.), Berlin: Springer-Verlag.

Barlas, Y. 1996. “Formal Aspects of Model Validity and Valida- tion in System Dynamics,” System Dynamics Review (12:3), pp. 183-210.

Barney, J. 1991. “Firm Resources and Sustained Competitive Advantage,” Journal of Management (17:1), pp. 99-120.

Black, L. J., Carlile, P. R., and Repenning, N. P. 2004. “A Dynamic Theory of Expertise and Occupational Boundaries in New Technology Implementation: Building on Barley’s Study of CT Scanning,” Administrative Science Quarterly (49:4), pp. 572-607.

1326 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Burton-Jones, A., McLean, E. R., and Monod, E. 2015. “Theo- retical Perspectives in Is Research: From Variance and Process to Conceptual Latitude and Conceptual Fit,” European Journal of Information Systems (24:6), pp. 664-679.

Capron, L., and Hulland, J. 1999. “Redeployment of Brands, Sales Forces, and General Marketing Management Expertise Following Horizontal Acquisitions: A Resource-Based View,” Journal of Marketing (63:2), pp. 41-54.

Chandy, R. K., Prabhu, J. C., and Antia, K. D. 2003. “What Will the Future Bring? Dominance, Technology Expectations, and Radical Innovation,” Journal of Marketing (67:3), pp. 1-18.

Churchman, C. W. 1979. The Systems Approach and Its Enemies, New York: Basic Books.

Coughlan, A., Anderson, E., Stern, L., and El-Ansary, A. 2001. Marketing Channels, Upper Saddle River, NJ: Prentice Hall.

Davis, J. P., Eisenhardt, K. M., and Bingham, C. B. 2007. “Devel- oping Theory through Simulation Methods,” Academy of Management Review (32:2), pp. 480-499.

Day, G. S. 1994. “The Capabilities of Market-Driven Organiza- tions,” Journal of Marketing (58:4), pp. 37-52.

Dierickx, I., and Cool, K. 1989. “Asset Stock Accumulation and Sustainability of Competitive Advantage,” Management Science (35:12), pp. 1504-1511.

Dong, C. Y., Fang, Y., and Straub, D. 2017. “Managing Institu- tional Distance between Collaborating Firms—The Role of Adaptive Interorganizational Systems,” Information Systems Research (28:2), pp. 309-331.

Dutta, A., and Roy, R. 2005. “Offshore Outsourcing: A Dynamic Causal Model of Counteracting Forces,” Journal of Management Information Systems (22:2), pp. 15-35.

Eberlein, R., and Peterson, D. 2000. “Understanding Models with Vensim,” in Modeling for Learning Organizations, J. Morecroft and J. D. Sterman (eds.), New York: Productivity Press, pp. 339-349.

Fang, Y., Wade, M., Delios, A., and Beamish, P. W. 2007. “Inter- national Diversification, Subsidiary Performance, and the Mobility of Knowledge Resources,” Strategic Management Journal (28:10), pp. 1053-1064.

Forrester, J. W. 1961. Industrial Dynamics, Cambridge, MA: MIT Press.

Forrester, J. W. 1968. Principles of Systems: Text and Workbook, Cambridge, MA: Wright-Allen Press.

Forrester, J. W. 1971. “Counterintuitive Behavior of Social Systems,” Technology Review (73:3), pp. 53-68.

Forrester, J. W., and Senge, P. M. 1980. “Tests for Building Con- fidence in System Dynamics Models,” in System Dynamics, A. A. Legasto, J. W. Forrester and J. M. Lyneis (eds.), Amsterdam: North-Holland.

Gallivan, M. J., Eynon, J., and Rai, A. 2003. “The Challenge of Knowledge Management Systems—Analyzing the Dynamic Processes Underlying Performance Improvement Initiatives,” Information Technology & People (16:3), pp. 326-352.

Garcia, R., Calantone, R., and Levine, R. 2003. “The Role of Knowledge in Resource Allocation to Exploration Versus Exploitation in Technologically Oriented Organizations,” Deci- sion Sciences (34:2), pp. 323-349.

Garud, R., and Kumaraswamy, A. 2005. “Vicious and Virtuous Circles in the Management of Knowledge: The Case of Infosys Technologies,” MIS Quarterly (29:1), pp. 9-33.

Georgantzas, N. C. 2003. “Tourism Dynamics: Cyprus’ Hotel Value Chain and Profitability,” System Dynamics Review (19:3), pp. 175-212.

Gregor, S. 2006. “The Nature of Theory in Information Systems,” MIS Quarterly (30:3), pp. 611-642.

Gregor, S., and Klein, G. 2014. “Eight Obstacles to Overcome in the Theory Testing Genre,” Journal of the Association for Infor- mation Systems (15:11), pp. i-xix.

Größler, A., Thun, J. H., and Milling, P. M. 2008. “System Dynamics as a Structural Theory in Operations Management,” Production and Operations Management (17:3), pp. 373-384.

Grover, V., and Lyytinen, K. 2015. “New State of Play in Information Systems Research: The Push to the Edges,” MIS Quarterly (39:2), pp. 271-296.

Grover, V., Lyytinen, K., Srinivasan, A., and Tan, B. C. 2008. “Contributing to Rigorous and Forward Thinking Explanatory Theory,” Journal of the Association for Information Systems (9:2), pp. 40-47.

Han, K. S., and Noh, M. H. 1999. “Critical Failure Factors That Discourage the Growth of Electronic Commerce,” International Journal of Electronic Commerce (4:2), pp. 25-43.

Hevner, A. R., March, S. T., Park, J., and Ram, S. 2004. “Design Science in Information Systems Research,” MIS Quarterly (28:1), pp. 75-105.

Hoffman, D., and Novak, T. 2000. “How to Acquire Customers on the Web,” Harvard Business Review (78:3), pp. 179-188.

Hulland, J., Wade, M. R., and Antia, K. D. 2007. “The Impact of Capabilities and Prior Investments on Online Channel Com- mitment and Performance,” Journal of Management Information Systems (23:4), pp. 109-142.

Kim, S. S., and Malhotra, N. K. 2005. “A Longitudinal Model of Continued Is Use: An Integrative View of Four Mechanisms Underlying Postadoption Phenomena,” Management Science (51:5), pp. 741-755.

Klir, G. 1991. Facets of Systems Science, New York: Springer Science & Business Media.

Kogut, B., and Zander, U. 1996. “What Firms Do? Coordination, Identity, and Learning” Organization Science (7:5), pp. 502-518.

Kraemer, K. L., and Dedrick, J. 2002. “Strategic Use of the Internet and E-Commerce: Cisco Systems” Journal of Strategic Information Systems (11:1), pp. 5-29.

Lane, D. C. 1999. “Social Theory and System Dynamics Practice,” European Journal of Operational Research (113:3), pp. 501-527.

Lawrence, B. W. 1997. “The Black Box of Organizational Demography,” Organization Science (8:1), pp. 1-22.

Lee, A. S. 2004. “Thinking About Social Theory and Philosophy for Information Systems,” in Social Theory and Philosophy for Information Systems, L. Willcocks and J. Mingers (eds.), Chichester, UK: John Wiley & Sons, pp. 1-26.

Luna-Reyes, L. F., and Andersen, D. L. 2003. “Collecting and Analyzing Qualitative Data for System Dynamics: Methods and Models,” System Dynamics Review (19:4), pp. 271-296.

MIS Quarterly Vol. 42 No. 4/December 2018 1327

Fang et al./System Dynamics Modeling for IS Research

Luna-Reyes, L. F., Zhang, J., Ramon Gil-Garcia, J., and Cresswell, A. M. 2005. “Information Systems Development as Emergent Socio-Technical Change: A Practice Approach,” European Journal of Information Systems (14:1), pp. 93-105.

Markus, M. L., and Robey, D. 1988. “Information Technology and Organizational Change: Causal Structure in Theory and Research,” Management Science (34:5), pp. 583-598.

Martinsons, M. G., and Martinsons, V. 2002. “Rethinking the Value of IT, Again” Communications of the ACM (45:7), pp. 25-26.

Meadows, D. H. 1980. “The Unavoidable a Priori,” in Elements of the System Dynamics Method, J. Randers (ed.), Cambridge, MA: The MIT Press, pp. 23-57.

Meadows, D. H., Meadows, D. L., Randers, J., and Behrens III, W. W. 1972. The Limits to Growth, Washington, DC: Potomac Associates.

Michael, J. G., Jim, E., and Arun, R. 2003. “The Challenge of Knowledge Management Systems: Analyzing the Dynamic Processes Underlying Performance Improvement Initiatives,” Information Technology & People (16:3), pp. 326-352.

Miller, D., and Shamsie, J. 1996. “The Resource-Based View of the Firm in Two Environments: The Hollywood Film Studios from 1936 to 1965,” Academy of Management Journal (39:3), pp. 519-543.

Morecroft, J. D. W. 1984. “Strategy Support Models,” Strategic Management Journal (5:3), pp. 215-229.

Parc, G., Bourdeau, S., Marsan, J., Nach, H., and Shuraida, S. 2008. “Re-Examining the Causal Structure of Information Technology Impact Research,” European Journal of Information Systems (17:4), pp. 403-416.

Penrose, E. 1959. The Theory of the Growth of the Firm, Oxford, UK: Basil Blackwell.

Peteraf, M. A. 1993. “The Cornerstones of Competitive Advan- tage: A Resource-Based View,” Strategic Management Journal (14:3), pp. 179-191.

Porter, M. E. 2001. “Strategy and the Internet,” Harvard Business Review (79:3), pp. 63-78.

Priem, R. L., and Butler, J. E. 2001. “Is the Resource-Based ‘View’ a Useful Perspective for Strategic Management Research?,” Academy of Management Review (26:1), pp. 22-40.

Prietula, M., Carley, K., and Gasser, L. 1998. Simulating Organi- zations: Computational Models of Institutions and Groups, Cambridge, MA: MIT Press.

Probert, D. E. 1982. “System Dynamics Modeling Within the British Telecommunications Business,” Dynamica (8:11), pp. 69-81.

Qian, Y., Fang, Y., and Gonzalez, J. J. 2012. “Managing Informa- tion Security Risks During New Technology Adoption,” Computers & Security (31:8), pp. 859-869.

Quaddus, M., and Intrapairot, A. 2001. “Management Policies and the Diffusion of Data Warehouse: A Case Study Using System Dynamics-Based Decision Support System,” Decision Support Systems (31:2), pp. 223-240.

Radigan, J. 2001. “Best Buy Is a Clicks-and-Mortar Believer,” CFO.com (http://www.cfo.com/article.cfm/2994890/c_2984283/ ?f=archives; accessed May 2008).

Rai, A., Patnayakuni, R., and Seth, N. 2006. “Firm Performance Impacts of Digitally Enabled Supply Chain Integration Capa- bilities,” MIS Quarterly (30:2), pp. 225-246.

Randers, J. 1980. Elements of the System Dynamics Method, Waltham, MA: Pegasus Communications.

Repenning, N. 2002. “A Simulation-Based Approach to Under- standing the Dynamics of Innovation Implementation,” Organi- zation Science (13:2), pp. 109-127.

Richardson, G. P., and Pugh, A. L. 1981. Introduction to System Dynamics Modeling with Dynamo, Cambridge, MA: Produc- tivity Press.

Richmond, B., Peterson, S., and Lyme, P. V. 2006. “An Academic User’s Guide to Stella,” System Dynamics Review (50:2), pp. 217-220.

Rodrigues, A. G., and Williams, T. M. 1997. “System Dynamics in Software Project Management: Towards the Development of a Formal Integrated Framework,” European Journal of Infor- mation Systems (6:1), pp. 51-66.

Rust, R. T., Moorman, C., and Dickson, P. R. 2002. “Getting Return on Quality: Revenue Expansion, Cost Reduction, or Both?,” Journal of Marketing (66:4), pp. 7-24.

Sabherwal, R., and Robey, D. 1995. “Reconciling Variance and Process Strategies for Studying Information System Devel- opment,” Information Systems Research (6:4), pp. 303-327.

Saini, A., and Johnson, J. L. 2005. “Organizational Capabilities in E-Commerce: An Empirical Investigation of E-Brokerage Service Providers,” Journal of the Academy of Marketing Science (33:3), pp. 360-375.

Sastry, M. A. 1997. “Problems and Paradoxes in a Model of Punctuated Organizational Change,” Administrative Science Quarterly (42:2), pp. 237-275.

Shannon, C. 1948. “A Mathematical Theory of Communication,” The Bell System Technical Journal (27), pp. 379-423 and pp. 623-656.

Schmidt, I. D., Döbler, T., and Schenk, M. 2000. E-Commerce: A Platform for Integrated Marketing; Case Study on U.S. Retailing (Volume 4), Münster: LIT Verlag.

Shaw, T., and Jarvenpaa, S. 1997. “Process Models in Information Systems,” in Information Systems and Qualitative Research, A. S. Lee, J. Liebenau, and J. I. DeGross (eds.), New York: Springer, pp. 70-100.

Smith, K. G., and Hitt, M. A. 2005. Great Minds in Management: The Process of Theory Development, Oxford, UK: Oxford University Press on Demand.

Sterman, J. D. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World, Boston: Irwin/McGraw-Hill.

Sterman, J. D. 2001. “System Dynamics Modeling: Tools for Learning in a Complex World,” California Management Review (43:4), pp. 8-25.

Sterman, J. D., Repenning, N. P., and Kofman, F. 1997. “Unanti- cipated Side Effects of Successful Quality Programs: Exploring a Paradox of Organizational Improvement” Management Science (43:4), pp. 503-521.

Straub, D. 2012. “Editor’s Comments: Does MIS Have Native Theories” MIS Quarterly (36:2), pp. iii-xii.

1328 MIS Quarterly Vol. 42 No. 4/December 2018

Fang et al./System Dynamics Modeling for IS Research

Telser, L. 1961. “How Much Does It Pay Whom to Advertise?,” American Economics Review (51:2), pp. 194-195.

Von Bertalanffy, L. 1968. General Systems Theory: Foundations, Development, Applications, New York: George Braziller Inc.

Von Bertalanffy, L. 1972. “The History and Status of General Systems Theory,” Academy of Management Journal (15:4), pp. 407-426.

Von Bertalanffy, L. 1973. “The Meaning of General System Theory” General System Theory: Foundations, Development, Applications (Rev. Ed.), New York: George Braziller Inc., pp. 30-53.

Warren, K. 2002. “Operationalizing the Impact of Competence- Building on the Performance of Firms’ Resource Systems” in Systems Perspectives on Resources, Capabilities, and Manage- ment Processes, J. Morecroft, R. Sanchez, and A. Heene (eds.), Amsterdam: Pergamon, Elsevier Science Ltd., pp. 41-55.

Weber, R. 2003. “Editor’s Comment: Theoretically Speaking,” MIS Quarterly (27:3), pp. iii-xii.

Wernerfelt, B. 1984. “A Resource-Based View of the Firm,” Strategic Management Journal (5:2), pp. 171-180.

Zhu, K. 2004. “The Complementarity of Information Technology Infrastructure and E-Commerce Capability: A Resource-Based Assessment of Their Business Value,” Journal of Management Information Systems (21:1), pp. 167-202.

Zhu, K., and Kraemer, K. L. 2002. “E-Commerce Metrics for Net- Enhanced Organizations: Assessing the Value of E-Commerce to Firm Performance in the Manufacturing Sector,” Information Systems Research (13), pp. 275-295.

About the Authors

Yulin Fang is a professor and Director of the MSc. Business Infor- mation Systems Program in the Department of Information Systems, City University of Hong Kong. He obtained his Ph.D. at the Richard Ivey School of Business, Western University, Canada. His research is focused on digital innovation, knowledge management, social media, and e-commerce. He has published in major IS and business journals, such as MIS Quarterly, Information Systems Research, Journal of Management Information Systems, Journal of the Association for Information Systems, Strategic Management

Journal, Journal of Operations Management, Journal of Manage- ment Studies, and Organizational Research Methods, among others. e has served as an associate editor for MIS Quarterly and a senior editor for Information Systems Research and Information Systems Journal.

Kai H. Lim is Chair Professor of Information Technology Innova- tion and Management and Director of Research and Ph.D. Programs at the Information Systems Department, City University of Hong Kong. His research interests include e-Health, cross-cultural issues related to information systems management, IT-enable business strategy, e-commerce, social media, mobile commerce and human– computer interactions. He has served as a senior editor for MIS Quarterly (for two terms) and on the editorial board of Information Systems Research and Journal of the Association for Information Systems. His research has appeared in MIS Quarterly, Information Systems Research, Journal of Management Information Systems, and Journal of the Association for Information Systems. Prior joining CityU, he was on the faculty of Case Western Reserve University and the University of Hawaii. He has won numerous teaching and research awards, and is one of the top-ranking teachers in CityU’s EMBA program. He has conducted executive training in Beijing, Guangzhou, Shanghai, and Hong Kong. He is also an Honorary Professor at Fudan University, China.

Ying Qian is an associate professor at the School of Management, Shanghai University, China. She obtained her Master’s and Ph.D. degrees from the University of Bergen, Norway, where she developed expertise in system dynamics and applied it to research in the information systems field. Her research interest lies in e- commerce and crowdfunding, especially from the dynamic perspective.

Bo Feng is the Dean and Chair Professor at the School of Business, Soochow University, China. She obtained her Ph.D. in Management Science from Northeastern University, China. Her research interests include electric vehicle operations, crowd-funding, air cargo capacity management, and interfaces between IS, OM, and Mar- keting. Her research has been published in Omega, European Journal of Operational Research, and Transportation Research Part C, among others.

MIS Quarterly Vol. 42 No. 4/December 2018 1329

RESEARCH NOTE

SYSTEM DYNAMICS MODELING FOR INFORMATION SYSTEMS RESEARCH: THEORY DEVELOPMENT

AND PRACTICAL APPLICATION Yulin Fang

Department of Information Systems, College of Business, City University of Hong Kong, Kowloon Tong, HONG KONG {[email protected]}

Kai H. Lim Department of Information Systems, College of Business, City University of Hong Kong,

Kowloon Tong, HONG KONG {[email protected]}

Ying Qian Department of Information Systems, School of Management, Shanghai University,

Shanghai CHINA {[email protected]}

Bo Feng Department of Management, School of Business, Soochow University,

Soochow CHINA {[email protected]}

Appendix A Comparing SD with Other Systems Simulation Methods in IS and Management

SD differs from other simulation methods commonly used in the IS and management field, such as agent-based modeling (ABM) (Axelrod 1997; Carley 1992; Epstein 2006; Lomi and Larsen 2001), discrete-event (DE)/process-centric modeling (Banks et al. 2005; MacDougall 1987; Zeigler et al. 2000), Monte Carlo method (Fishman 1995; Kroeses et al. 2014), and genetic algorithm (Bruderer and Singh 1996; Zott 2002). The theoretical logic of SD, ABM, and DE is explanation, while Monte Carlo and genetic algorithms focus on optimization. Herein, simulation with explanatory theoretical logic can be a powerful tool for specifying and extending existing theories. Both ABM and DE are well-known and commonly used system simulation methods for theory development in the IS and management fields; SD is distinguished from them in important ways.

The ABM method focuses on how a phenomenon emerges and evolves in an adaptive system (e.g., bilateral collaborative network) where multiple agents interact with and adapt to the actions of other agents. The typical purpose is to simulate a large number of autonomous agents that interact with each other, within a simulated environment and observe emergent patterns from their interactions. The common research question is often framed as: How does interaction among agents give rise to a phenomenon?

The DE method simulates a process system (e.g., a queuing system) consisting of a discrete sequence of events in time. Unlike the structural theory of SD, the theoretical base of DE is process theory. This simulation tool is typically used to evaluate strategies for system operations as well as to predict system performance. The research question for DE modeling commonly is: How will the system perform if the activity, event, or process changes?

MIS Quarterly Vol. 42 No. 4—Appendices/December 2018 A1

Fang et al./System Dynamics Modeling for IS Research

Regarding system theory development in the IS field, the three simulation methods, SD, DE, and ABM differ in terms of system level, scope, time duration, change continuity, and basic mathematical model. We review and compare the three simulation methods in Table A1.

Table A1. Comparison of Simulation Methods for Theory Development

Aspect System Dynamics Agent-Based Model Process-Centric

(Discrete-Event, DE) System

classification

Complex feedback system Complex adaptive system Process system

Theory base Structural theory Behavioral theory Process theory

Typical purpose Examine how initial conditions of

a system affect the stability of

the system over time

Test what occurs after agents

interact and how a phenomenon

emerges and evolves

Evaluate strategies for

operating a system or

predicting system performance

Research focus Modeling a wide range of

feedback effects with delayed

and circular causality

Modeling interactions among

intelligent agents

Modeling one or more

stochastic events

System level Strategic level All levels Operational and tactical levels

System scope Aggregated individuals/

homogeneous

Individual/heterogeneous Individual/heterogeneous

System key

elements

Stocks and flows Agents, actions Entities, activities, and queues

System duration Long-term and mid term Short-term to mid-term Short-term

System change Continuous Discrete Discrete

Appendix B Sensitivity Test of Model Behavior on Market Share

A2 MIS Quarterly Vol. 42 No. 4–Appendices/December 2018

Fang et al./System Dynamics Modeling for IS Research

Appendix C Reference Model Behavior: Best Buy’s Front-End E-Commerce Capability (1998–2007) (Trend Line Added)

Appendix D Model-Generated Behavior: Front-End E-Commerce Capability Accumulation Over 10 Years

MIS Quarterly Vol. 42 No. 4—Appendices/December 2018 A3

Fang et al./System Dynamics Modeling for IS Research

Appendix E Reference and Model Behavior: Best Buy’s Revenue (1997–2007)

Appendix F Model-Generated Behavior: Revenue Over 10 Years

References

Axelrod, R. 1997. “The Dissemination of Culture a Model with Local Convergence and Global Polarization,” Conflict Resolution (41:2), pp. 203-226.

Banks, J., Carson, J., Nelson, B., and Nicol, D. 2005. Discrete-Event System Simulation (4th ed.), Boston: Pearson. Bruderer, E., and Singh, J. V. 1996. “Organizational Evolution, Learning, and Selection: A Genetic-Algorithm-Based Model,” Academy of

Management Journal (39:5), pp. 1322-1349.

A4 MIS Quarterly Vol. 42 No. 4–Appendices/December 2018

Fang et al./System Dynamics Modeling for IS Research

Carley, K. 1992. “Organizational Learning and Personnel Turnover,” Organization Science (3:1), pp. 20-46. Epstein, J. M. 2006. Generative Social Science: Studies in Agent-Based Computational Modeling, Princeton, NJ: Princeton University Press. Fishman, G. S. 1995. Monte Carlo: Concepts, Algorithms, and Applications, New York: Springer. Kroeses, D. P., Taimre, T., and Botev, Z. I. 2014. Handbook of Monte Carlo Methods, New York: John Wiley & Sons. Lomi, A., and Larsen, E. R. 2001. Dynamics of Organizations: Computational Modeling and Organization Theories, Cambridge, MA: MIT

Press. MacDougall, M. H. 1987. Simulating Computer Systems: Techniques and Tools, Cambridge, MA: MIT Press. Zeigler, B. P., Praehofer, H., and Kim, T. G. 2000. Theory of Modeling and Simulation: Integrating Discrete Event and Continuous Complex

Dynamic Systems (2nd ed.), New York: Academic Press. Zott, C. 2002. “When Adaptation Fails An Agent-Based Explanation of Inefficient Bargaining Under Private Information,” Journal of Conflict

Resolution (46:6), pp. 727-753.

MIS Quarterly Vol. 42 No. 4—Appendices/December 2018 A5

Copyright of MIS Quarterly is the property of MIS Quarterly and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.