c Consult author(s) regarding copyright matters Notice ... Theorizing (accepted... · A Systematic Framework for Multilevel Theorizing in Information Systems Research ... research

This may be the author’s version of a work that was submitted/acceptedfor publication in the following source:

Zhang, Meng & Gable, Guy(2017)A systematic framework for multilevel theorizing in information systemsresearch.Information Systems Research, 28(2), pp. 203-224.

This file was downloaded from: https://eprints.qut.edu.au/104342/

c© Consult author(s) regarding copyright matters

This work is covered by copyright. Unless the document is being made available under aCreative Commons Licence, you must assume that re-use is limited to personal use andthat permission from the copyright owner must be obtained for all other uses. If the docu-ment is available under a Creative Commons License (or other specified license) then referto the Licence for details of permitted re-use. It is a condition of access that users recog-nise and abide by the legal requirements associated with these rights. If you believe thatthis work infringes copyright please provide details by email to [email protected]

Notice: Please note that this document may not be the Version of Record(i.e. published version) of the work. Author manuscript versions (as Sub-mitted for peer review or as Accepted for publication after peer review) canbe identified by an absence of publisher branding and/or typeset appear-ance. If there is any doubt, please refer to the published source.

https://doi.org/10.1287/isre.2017.0690

https://eprints.qut.edu.au/view/person/Zhang,_Meng.html

https://eprints.qut.edu.au/view/person/Gable,_Guy.html

https://eprints.qut.edu.au/104342/

https://doi.org/10.1287/isre.2017.0690

1

A Systematic Framework for Multilevel Theorizing in

Information Systems Research

Meng Zhang

Information Systems School, Science and Engineering Faculty, Queensland University of Technology,

Brisbane 4000, Australia

Email: [email protected]

Mail: Y706C-34, QUT Gardens Point Campus, Brisbane, QLD 4000, Australia

Guy G. Gable

Information Systems School, Science and Engineering Faculty, Queensland University of Technology,

Brisbane 4000, Australia

Email: [email protected]

Mail: Y704, QUT Gardens Point Campus, Brisbane, QLD 4000, Australia

Abstract: Information systems (IS) research usually investigates phenomena at one level of analysis at a

time. However, there are complex IS phenomena that are difficult to address from such a single-level

perspective. A multilevel perspective offers an alternative means to examine phenomena by simultaneously

accounting for multiple levels of analysis. Although useful guidelines for theory development are widely

available, they give little specific attention to developing theory that is conceptualized and analyzed at

multiple levels. Multilevel theorizing or developing theory from a multilevel perspective is more complex

and involves unique challenges. To promote multilevel theorizing in the IS discipline, we focus on

addressing challenges involved in multilevel theorizing and propose a holistic framework for systematically

developing theory from a multilevel perspective. Drawing from the Organization Science and IS literature,

the proposed framework harmonizes and synthesizes previous guidelines, providing a practical basis for

conceptualizing and studying multilevel phenomena.

2

Keywords: multilevel theory; multi-level theory; multilevel research perspective; multilevel theorizing;

multilevel perspective; multilevel paradigm; level of analysis; theory building; research methods.

Introduction

Most information systems (IS) studies address a single level of analysis at a time. For example, studies may

focus only on the individual level, explaining why users adopt or use information technologies (IT) (e.g.,

Davis 1989; Venkatesh and Davis 2000), or the organizational level, explaining how firms leverage IT to

generate business benefits (DeLone and McLean 2003; Melville et al. 2004). However, examining

phenomena from a single level of analysis at a time may be incomplete and disjointed (Bélanger et al. 2014;

Burton-Jones and Gallivan 2007). As a more comprehensive approach to theory building, a multilevel

(research) perspective encourages simultaneous examination of phenomena at multiple levels of analysis

(Bélanger et al. 2014; Burton-Jones and Gallivan 2007; Kozlowski and Klein 2000; Rousseau 1985; House

et al. 1995). For example, to understand group IT adoption, researchers may consider both the individual

and the group levels, by examining how a group’s attitude toward IT adoption results from an interaction

among group members with diverse attitudes (Sarker and Valacich 2010).

A multilevel perspective is defined as an approach to theory development that considers the relevance of

multiple levels of analysis (Kozlowski and Klein 2000; Burton-Jones and Gallivan 2007; Bélanger et al.

2014). Level of analysis refers to the conceptual entity under theoretical and empirical investigation such as

individuals, groups, and organizations. Such conceptual entities may be hierarchically organized (Rousseau

1985); for example, a group may consist of individual members and an organizational department may

consist of work teams. As argued by Mathieu and Chen (2011), given levels of analysis are hierarchically

organized, many phenomena at one level are influenced inevitably by factors above or below the focal

level. Thus, multilevel theorizing explores opportunities for developing theoretical propositions that explain

phenomena spanning multiple levels of a hierarchical system (Kozlowski and Klein 2000).

A multilevel perspective to theory development offers an effective means to integrate theories that address

3

different levels of analysis (Rousseau 2011; Burton-Jones and Gallivan 2007; House et al. 1995). House et

al. (1995) argued that linking theories at separate levels helps build a coherent and integrated research field.

Burton-Jones and Gallivan (2007) maintained that theory integration across levels may create new research

opportunities. Further, a multilevel perspective represents a more natural and complete way of examining

phenomena (Burton-Jones and Gallivan 2007). For instance, individual-level studies of technology adoption

might suggest that successful technology implementation needs to ensure that the technology is easy to use,

whereas organizational-level studies of technology adoption might urge firms to implement cutting-edge

technology in order to sustain their competitive advantage. As cutting-edge technology may be difficult to

implement or use in practice, there might be contention between the implications from the individual-level

and the organizational-level studies. To resolve the situation, a multilevel perspective may integrate and

harmonize the individual-level and the organizational-level studies (Burton-Jones and Gallivan 2007).

Although several studies in IS research have benefited from adopting a multilevel perspective (e.g., Kang et

al. 2012; Sarker and Valacich 2010; Lapointe and Rivard 2005), studies that examine IS phenomena from

a multilevel perspective are scarce (Bélanger et al. 2014; Burton-Jones and Gallivan 2007). Furthermore,

existing guidelines primarily for single-level theory (e.g., Dubin 1978; Bacharach 1989; Jaccard and Jacoby

2010) paid scant attention to challenges involved in multilevel theorizing (Rivard 2014). Thus, there have

been extensive discussions over the past decades on issues related to multilevel theory development, as

depicted in Table 1 (e.g., Rousseau 1985; Klein et al. 1994; House et al. 1995; Chan 1998; Morgeson and

Hofmann 1999; Kozlowski and Klein 2000; Burton-Jones and Gallivan 2007; Bélanger et al. 2014).

4

Table 1. Snapshot of Literature Related to Multilevel Theorizing

Key Literature Primary Focus Major Limitations Rousseau 1985 Analyzed phenomena that span multiple levels of

analysis, but share the same nature of content. Did not elaborate in detail other types of multilevel models.

Klein et al. 1994 Analyzed the similarities and differences of members of a group and the implications for developing multilevel models.

Did not elaborate in detail other types of multilevel models.

House et al. 1995 Elaborated the core value of multilevel theorizing. Did not address multilevel theory development procedures.

Chan 1998 Extended Rousseau’s (1985) analysis and developed operationalization approaches for compositional models that span multiple levels of analysis, but share the same nature of content.

Did not focus on theoretical explanations of multilevel phenomena.

Morgeson and Hoffman 1999 Analyzed the nature of bottom-up processes and the implications for developing multilevel models.

Did not consider top-down processes.

Kozlowski and Klein 2000 Comprehensively addressed all aspects of multilevel theory development including theorizing, operationalization, data collection, and statistical analysis, and elaborated Morgeson and Hoffman’s (1999) conceptual analysis of bottom-up processes.

Incomplete in not considering many important types of multilevel models, especially models that involve top-down processes.

Burton-Jones and Gallivan 2007 Further elaborated Morgeson and Hoffman’s (1999) conceptual analysis of bottom-up processes and contextualized their analysis using System Usage studies.

Did not address important types of multilevel models that are a result of top-down processes.

Bélanger et al. 2014 Analyzed the role of IT or IS in supporting meaningful conceptualization of entities.

Did not address theoretical analysis of different types of multilevel models.

However, existing guidelines for multilevel theorizing in the literature are not systematic in several respects.

First, terms used to describe multilevel theorizing have proliferated, engendering confusion for readers.

Although several useful attempts have been made toward a clarification1 (e.g., Bélanger et al. 2014; Burton-

Jones and Gallivan 2007; Gallivan and Benbunan-Fich 2005; Kozlowski and Klein 2000), they did not

focus on clarifying terms that are used to describe different types of multilevel effects2. A partial list of such

terms includes: “homogeneity”, “heterogeneity”, “within-unit and between-unit variance”, “shared unit

property”, and “configural unit property”. For instance, Klein et al.’s (1994) term “homogeneity” refers to

the circumstance in which all individual members of a group have the same value regarding some attribute

or property (e.g., every member of a group has the same extent of expertise), whereas their term

“heterogeneity” is not opposite in meaning to their term “homogeneity”, but, instead, refers specifically to

the existence of “frog-pond effects” (which are explained in subsequent sections). As a further example of

1 A list of related terms is attached in Appendix A.

2 Our sincere thanks to Reviewer 2 for detailed guidance in attention to this issue.

5

existing confusion, the term “configural unit property” introduced by Kozlowski and Klein (2000) refers to

“mono-, unequally distributed effect” or “combined-effect” or both (these terms are defined in subsequent

sections). The term “shared property” introduced by Kozlowski and Klein (2000) is also seemingly opposite

in meaning to “configural unit property”. However, “shared unit property” has an utterly different

meaning, referring only to a subset of “mono-, equally distributed effect” (this term is also defined in

subsequent sections). Furthermore, existing guidelines are incomplete in terms of considering the possible

types of multilevel models, and therefore are inadequate for analyzing all kinds of multilevel phenomena in

empirical settings. While Kozlowski and Klein’s (2000) work has been most comprehensive in covering all

the aspects involved in multilevel theorizing, their typology of multilevel models lacks important types of

multilevel models that involve top-down processes. For instance, they regarded “frog-pond effects” as the

only way to account for top-down processes when influences from a higher level are unequally distributed

across members of a group, which limited the rich range of possibilities in multilevel theorizing. Lastly,

existing literature represented different types of multilevel phenomena as a set of somewhat scattered,

isolated models, but did not clarify the inherent logic that connects these models. This representation of

multilevel models may inhibit easy access to multilevel theorizing.

To encourage more multilevel theoretical and empirical studies in IS research and to address limitations in

the existing guidelines for multilevel theorizing, this paper proposes a framework for multilevel theorizing.

Drawing primarily from the IS and the Organization Science literature (e.g., Klein et al. 1994; House et al.

1995; Kozlowski and Klein 2000; Burton-Jones and Gallivan 2007; Bélanger et al. 2014), we integrate,

harmonize, and synthesize existing guidelines, thereby offering a systematic approach to conceptualizing

and examining a spectrum of multilevel phenomena. We argue that the framework is more comprehensive

in considering multilevel phenomena than anything prior, and that the framework establishes a coherent

basis for researchers to better make sense of multilevel phenomena. The framework also shows that only a

subset of possible multilevel phenomena have been explored in empirical studies. It is thus hoped that the

framework will enrich analytical toolkits available for IS research and will encourage researchers to address

untapped terrains of multilevel phenomena.

6

A Framework for Multilevel Theorizing

Multilevel theorizing is not incompatible with prior prescriptions on theory development (e.g., Dubin 1978;

Bacharach 1989; Jaccard and Jacoby 2010; Gregor 2006; Weber 2012). For instance, notions such as

parsimony, elegance, accuracy, and generality (e.g., Bacharach 1989), and ideas about how theory should

be evaluated (e.g., Weber 2012) apply similarly to multilevel theorizing. The key difference is there are

extra considerations needed for multilevel theorizing. Consistent with prior studies (e.g., Kozlowski and

Klein 2000; Burton-Jones and Gallivan 2007), we do not examine a general process of theory development

(e.g., discuss how to formulate a research question), but focus specifically on extra considerations that must

be addressed in multilevel theorizing. We thus consider how multilevel theorizing can extend single-level

theory, by focusing on differences between multilevel theorizing and the development of single-level theory.

A framework for multilevel theorizing is proposed. The framework synthesizes the prior body of thought

(e.g., Kozlowski and Klein 2000; Klein et al. 1994; Morgeson and Hofmann 1999; Burton-Jones and

Gallivan 2007; Rousseau 1985; House et al. 1995) and characterizes an approach that guides researchers to

develop multilevel theory systematically (see Table 2).

Several caveats should be noted. First, researchers should remain open to the iterative process of theory

development. As argued by Weick (1989), theory development is not linear, but entails evolutionary

understanding and cyclical modification of a research design. Although the framework consists of a flow of

steps, it is necessary for researchers to reflect regularly on execution of prior steps and if useful to return to

prior steps until satisfied.

Further, the proposed framework focuses mainly on theory building instead of theory testing. For empirical

studies, it is important to address challenges in validating multilevel theory such as operationalization

(Chan 1998; Kozlowski and Klein 2000), sampling (Kozlowski and Klein 2000; Klein et al. 1994), and

statistical analysis of hierarchical data (Hox 2010; Snijders and Bosker 1999). However, these issues are not

covered herein.

Moreover, although multilevel theorizing can be used to develop both variance-based models and process

7

models and can be used together with both quantitative and qualitative techniques, our analysis emphasizes

variance-based modeling approach and quantitative analysis. Given such a focus, researchers may need to

consider contrasting scenarios that demonstrate the variance of constructs. For example, to analyze how

organizational culture might influence employees’ work performance, researchers may need to consider

what types of organizations rank higher on a particular dimension of organizational culture (e.g., “respect

for authority” or “security culture”) and what types of organizations rank relatively lower. In other words,

to analyze such phenomena from a variance-based perspective, empirical scenarios must be able to vary.

Lastly, although operationalization issues are not a central focus, some discussion on operationalization in

places can help to elaborate the theoretical nature of a phenomenon. In particular, given an

operationalization approach needs to conform to the theoretical nature of a phenomenon, illustrating a

complex multilevel phenomenon via its possible operationalization approaches may concretize abstract

notions with tangible, familiar empirical cases. Regardless, we caution that operationalization procedures

mentioned herein should not be interpreted as the only way to test multilevel models.

8

Table 2. Framework for Multilevel Theorizing

Step Main Activities Rationale

(1). Specify single-level theory that characterizes the focal phenomenon of interest;

• Clarify the focal phenomenon of interest with a construct or a relationship;

• Consider relevant dependent variables; • Explore whether there is a suitable, existing single-level

theory; • Remain open to new dependent variables as theorizing

unfolds.

• Theorizing is guided by the focal phenomenon of interest;

• Multilevel theorizing often starts with single-level theory (Kozlowski and Klein 2000).

(2). Identify candidate theoretical entities that might be relevant to the single-level theory;

• Consider the focal entity’s larger contexts (e.g., a family unit or an institution) and granular counterparts (e.g., individuals or project teams);

• Narrow down candidate theoretical entities based on a holistic consideration of factors such as research interest, expertise, reference theory, prior conceptual framework, analytical tools, and availability of data;

• Clearly articulate rules that define candidate entities and justify the existence of candidate entities by considering, for instance, the principles for identifying the collectiveness of the entities or how the existence of IT might support the structure of the entities.

• An entity is almost always embedded within a larger context or composed of a number of smaller components or both;

• Formal classifications of entities (such as organizational departments) may fail to capture the nuances of phenomena (Mathieu and Chen 2011).

(3). Explore zoom-in or zoom-out theorizing strategies;

• Zoom in to consider both the focal entities and their internal structures and functions;

• Zoom out to consider both the focal entities and their external contexts;

• Tentatively present possible top-down or bottom-up effects revealed by zoom-in or zoom-out strategies.

• Multilevel theorizing hinges on a deliberate shift of research attention;

• Research attention can only selectively abstract reality;

• The possibilities of selective abstraction should be systematically explored.

(4). Specify top-down or bottom-up effects.

• Analyze what kinds of top-down influence or bottom-up contribution are potentially relevant and assess the likelihood of interaction among the different kinds;

• Analyze whether the top-down influence applies equally in amount to all lower-level entities and whether the bottom-up contribution applies equally in amount from all lower-level entities;

• Analyze whether top-down influence or bottom-up contribution is targeted at a construct or a relationship;

• Specify the exact top-down and bottom-up effects and present relevant theoretical and empirical justifications;

• Consider alternative conceptualizations of top-down and bottom-up effects and comparatively assess competing conceptualizations.

• Multilevel phenomena can take complex forms; researchers must be precise about the exact phenomena under scrutiny;

• Thinking about alternative conceptualizations of multilevel phenomena can help refine thinking, spark creativity for preferred choices, and generate new research questions for future investigation (Klein et al. 1994).

9

Step 1 - Specify Focal Single-Level Theory

It is important to clarify the focal phenomenon of interest (Kozlowski and Klein 2000). This can be carried

out through specifying a focal, single-level theory. A construct, or two constructs at the same level of

analysis and a relationship between the two constructs may serve as such a focal theory3. For example, the

analysis may commence with a focus on organizational productivity subsequent to the implementation of a

business intelligence system. The researcher may focus on the influence of organizational structure (as an

independent variable) on organizational productivity.

Notably, although interesting independent variables may initially motivate a study, theorizing should be

anchored to at least one relevant dependent variable. The ultimate aim of theoretical explanation is to

answer the question of “why” through exploring alternative independent variables or antecedents (DeLone

and McLean 1992). Hence, the focal construct or relationship should be or should include at least a

dependent variable (Kozlowski and Klein 2000). Example dependent variables include: IS project success

(Rai et al. 2009) and IS avoidance (Kane and Labianca 2011). Furthermore, where a research domain is

relatively more advanced and the aim is primarily to extend an established or widely accepted theory, the

established theory or a part of it could serve as the focal single-level theory. For example, researchers may

use the three constructs of Technology Acceptance Model (i.e., Ease of Use, Usefulness, and Intention to

Use) together with the relationships among the three constructs as the focal single-level theory (e.g., Sarker

and Valacich 2010).

Although a focal single-level theory offers a locus for anchoring subsequent examinations, researchers may

change dependent variables as the theoretical investigation unfolds. One possibility is that the researcher

finds more interesting or more researchable questions that can be addressed only by replacing one or more

of the dependent variables. For example, where the researcher’s initial interest lies in studying the impact of

organizational structure on organizational productivity after the implementation of a business intelligence

system, the researcher may find that different teams tend to use the system in diverse ways and to different 3 In our view, a construct alone is also theory, with which some may disagree.

10

extent and, therefore, may argue that team productivity is a more appropriate dependent variable to

examine the impact of the system.

It is also possible that the researcher adds a dependent variable at a level separate from the original focal

level. For example, the researcher may initially study how individual system usage affects a person’s task

performance and later may find it interesting to explore how individual members’ system usage might

aggregate at the group level to influence group performance (e.g., Burton-Jones and Gallivan 2007).

Step 2 - Identify Candidate Theoretical Entities

It is crucially important to ask what theoretical entities or levels (e.g., work teams, departments) are relevant

to the phenomenon of interest (Kozlowski and Klein 2000; Klein et al. 1994; Bélanger et al. 2014). Because

a theoretical entity is almost always embedded within a larger context, or composed of a number of smaller

components, there are often several if not many entities available for theoretical investigation (Kozlowski

and Klein 2000). The researcher must consider what theoretical entities or levels are most relevant. Several

aspects can be considered regarding the relevance of theoretical entities. Choices are best made given a

holistic consideration of these aspects.

The researcher’s own personal interest or knowledge background is perhaps one of the strongest drivers for

identifying candidate entities. The availability of reference theory or conceptual frameworks can be another

source of insights; for example, social network theory stimulated empirical studies that shift level of analysis

from an individual level to a collective level by taking the network of individuals’ communication patterns

into consideration (e.g., Kane and Labianca 2011). Furthermore, given multilevel theory needs to be tested

using hierarchical data (e.g., employees’ performance grouped per their work groups and departments), the

availability of existing hierarchical data or the feasibility of collecting such data can be a driver; it may be

more productive to pursue data-rich areas.

In identifying candidate theoretical entities, it is crucial to make clear what characteristics or attributes the

theoretical entities entail (Klein et al. 1994; Rousseau 1985; Mathieu and Chen 2011; Bélanger et al. 2014).

11

Mathieu and Chen (2011, p. 615) observed that defining entities based entirely on “formal classifications,

such as designated members of a team, full-time employees on an organizational payroll, Standard

Industrial Classification codes, and alike” are inappropriate for phenomena involving special forms of

collectives, such as “ad hoc task forces”, or where there are exceptions, for example, due to “a large

number of part-time or temporary workers”. It is necessary to “define why a particular collective is a salient

grouping entity or provide clear rules for inclusion and exclusion of membership” (Mathieu and Chen

2011, p. 615). Burton-Jones and Gallivan (2007, p. 665) suggested four principles that can be used to

identify collective entities: “(1) Do the individuals consider themselves to be members of a collective (that

may, in turn, be part of a larger collective)? (2) Do the individuals recognize one another as members and

distinguish members from nonmembers? (3) Do the collective members’ activities show more tightly

coupled interdependence within the group than with others in the larger collective? (4) Do members of the

collective share a common fate (or consequence) that is not totally shared by the larger collective?” In a

similar vein, Bélanger et al. (2014) suggested that the existence of IT may be used to help identify entities.

For instance, researchers may define entities according to routines and structures supported by the

presence of IT; where a newly implemented strategic decision support system involves a chain of activities

that are required to be enacted by a cohort of organizational users, this cohort of organizational users, even

if belonging to separate organizational departments, may serve as a collective entity for theoretical analysis.

Step 3 - Explore Zoom-In or Zoom-Out Strategies

Once candidate theoretical entities are tentatively identified, researchers should direct attention away from

one focus to another. Two distinctive strategies of shifting focus can be explored. To illustrate, consider a

researcher with a camera4. The camera represents the means for the researcher to observe a complex real-

world system (see Figure 1). No matter how advanced it is, a camera can only characterize limited aspects

of the real-world system (due to, for example, limited resolution, or distortion as a result of information

loss). To describe the behavioral patterns of objects within a real-world system, the researcher may focus on 4 We thank Reviewer 2 for offering this metaphor.

12

the objects themselves and track their patterns of movement (see Figure 1[a]). Alternatively, the researcher

may (i) zoom in to observe the internal structures of objects and track both their movements and changes in

internal structures (see Figure 1[b]); or, the researcher may (ii) zoom out to observe the objects within their

external environments and track both their movements and changes in external environments (see Figure

1[c]).

Figure 1[a]. Single-Level Abstraction

Figure 1[b]. Zoom-In Abstraction

Single-Level Abstraction

Repeated Observations A Complex Real-World System

Zoom-In Abstraction


13

Figure 1[c]. Zoom-Out Abstraction

Figure 1. Possible Strategies for Conceptualizing a Complex Real-World System

This camera metaphor conveys a main difference between the development of single-level theory and

multilevel theory. Single-level theorizing requires accounting for only the behavioral patterns of the focal

entities (Figure 1[a]). In contrast, multilevel theorizing requires the researcher to zoom in with their

conceptual lens, thereby considering both the focal entities and their internal structures and functions

(Figure 1[b]), or to zoom out with their conceptual lens, thereby considering both the focal entities and

their external contexts (Figure 1[c]).

When employing a zoom-in strategy, the researcher seeks to account for a lower-level entity in addition to

the focal level entity; for example, rather than consider a group (or other collective entity) as a “black box”,

the researcher may analyze group members’ different contributing roles. When employing a zoom-out

strategy, the researcher seeks to account for a higher-level entity in addition to the focal level entity; for

example, rather than consider individuals (or other entities) as isolated from external contexts, the

researcher may analyze the individuals’ received environmental or contextual regulations.

Zoom-in and zoom-out strategies can be carried out to capture two distinctive theoretical effects: (i)

bottom-up effects where a higher level emerges from or is sustained by a lower level and (ii) top-down

effects where a lower level is regulated, constrained, or reinforced by a higher level (Kozlowski and Klein

Zoom-Out Abstraction


14

2000). Each of zoom-in or zoom-out strategy may result in bottom-up (“ ⇑ ”) or top-down (“ ⇓ ”) effects or

both (see Table 3). Bottom-up effects can capture, for example, how a certain pattern of system usage by a

group might derive from its individual, lower-level counterparts (Burton-Jones and Gallivan 2007). One

well-known top-down effect is that technology acceptance behaviors may vary across diverse cultural

backgrounds (Straub et al. 1997). Researchers should tentatively identify possible bottom-up or top-down

effects that correspond roughly to one or more of the scenarios in Table 3.

Table 3. Possible Theoretical Effects Resulting from Zoom-In and Zoom-Out Strategies

Focal Single-Level Theory Strategy Theoretical Effects Captured

Focal Construct! → Focal Construct!

Zoom-In

Bottom-Up Effect Focal Construct! → Focal Construct!

⇑ New Construct! → New Construct!

Top-Down Effect Focal Construct! → Focal Construct!

⇓ New Construct! → New Construct!

Zoom-Out

Top-Down Effect New Construct! → New Construct!

⇓ Focal Construct! → Focal Construct!

Bottom-Up Effect New Construct! → New Construct!

⇑ Focal Construct! → Focal Construct!

Step 4 - Specify Top-Down and Bottom-Up Effects

Although top-down effects and bottom-up effects may coexist in one study, we examine them separately.

Top-down and bottom-up effects are broad notions that encapsulate a large number of possible theoretical

effects. It is necessary for the researcher to specify and refine the exact effects appropriate for the

phenomenon. Such specification remains one of the most challenging tasks in multilevel theorizing.

A Typology of Multilevel Phenomena

To help specify multilevel phenomenon, we propose a typology (see Table 4). In the typology, there are

eight types of top-down or bottom-up effects (labeled with Roman numbers I to VIII) and two Null types,

which represent the absence of top-down or bottom-up effects. Further, each top-down or bottom-up effect

type has two variants; researchers may consider one or a mix of these sixteen archetype models to specify

the phenomenon of interest. The typology of multilevel phenomena is cross-referenced with the literature

(see Table 5). The distinctions (i.e., “Kind”, “Amount”, and “Target”) that are used to differentiate these

15

models are briefly summarized in Table 6 and thereafter illustrated in detail. Note that the models in the

typology are archetypes and that a study may include more than one of these models. Researchers thus

should systematically explore the relevance of each type according to the specific research setting. We next

discuss how each archetype should be specified in detail.

16

Table 4. Typology of Multilevel Phenomena

Direction Type Kind Amount Target Archetype Model*

Top-down

I Mono Equally Distributed

Construct

Relation

II Mono Unequally Distributed

Construct

(Frog-pond instance: )

Relation


III Combined Equally Distributed

Construct

Relation

IV Combined Unequally Distributed

Construct


Relation

(Frog-pond instance: ) Null Top-down effects are absent.

* Dashed lines divide lower-level and higher-level constructs; XL refers to a lower-level construct (e.g., individual); XH refers to a higher-level construct (e.g., group). These models are cross-referenced with the literature in Table 5.

YL

XH

WH

YLXL

YL, n

XHf (n)

(XL — XH) YL

WH

YL, nXL, n

f (n)

(WL — XH)

YLXL

YL

f [ XH1; XH2]

f [ XH1; XH2]

YLXL

YL, n

f [ XH1; XH2]f (n)

(XL — f [ XH1; XH2]) YL

f [ XH1; XH2]

YL, nXL, n

f (n)

YLXL

(WL — f [ XH1; XH2])

17

(Continuing Table 4…)

Table 4. Typology of Multilevel Phenomena

Direction Type Kind Amount Target Archetype Model*

Bottom-up

V Mono Equally Distributed

Construct

Relation

VI Mono Unequally Distributed

Construct

Relation

VII Combined Equally Distributed

Construct

Relation

VIII Combined Unequally Distributed

Construct

Relation

Null Bottom-up effects are absent.

* Dashed lines divide lower-level and higher-level constructs; XL refers to a lower-level construct (e.g., individual); XH refers to a higher-level construct (e.g., group). These models are cross-referenced with the literature in Table 5.

YH

XL

WH

YHXH

XL

WH

YHWH

XL,1 XL,2 XL,N…

WH

XL,1 XL,2 XL,N…

YHXH

WH

f [ XL1; XL2]

YH

WH

f [ XL1; XL2]

YHXH

…

WH

f [ XL1, 1; XL2, 1] f [ XL1, 2; XL2, 2] f [ XL1, N, XL2, N]

YH

…

WH

f [ XL1, 1; XL2, 1] f [ XL1, 2; XL2, 2] f [ XL1, N, XL2, N]

YHXH

18

Table 5. Archetype Models Cross-Referenced with the Literature

Descriptor Archetype Model Direction Top-Down Bottom-Up

Type I II III IV Null V VI VII VIII Null Kind Mono Combined Mono Combined

Amount: Equal/Unequal E U E U E U E U Target:

Construct/Relation C R C R C R C R C R C R C R C R

Cross-Reference with the Literature* Klein et al. (1994):

Individuals free of group influence (pp. 200 – 201);

Mixed effects models and mixed-determinants models (pp. 221 – 223);

The group as a whole (pp. 199 – 200);

Frog-pond effects (pp. 201 – 203);

Morgeson and Hoffman (1999): Structure and function of collective constructs;

Chan (1998): Typology of compositional models (p. 236);

Kozlowski and Klein (2000): Global unit properties (pp. 29 – 30);

Cross-level direct effects (pp. 42 – 43);

Cross-level moderator (p. 43);

Cross-level frog-pond (pp. 43 – 44)

Shared unit properties (p. 30);

Configural unit properties (pp. 30 – 32);

Composition emergence (pp. 52 – 77);

Compilation emergence (pp. 52 – 77);

Burton-Jones and Gallivan (2007): Structure and function of collective System Usage construct;

* Greyed cells indicate the model archetypes are not previously discussed in the literature.

! 16!

Table!2.!Comparison!between!Multilevel!Perspective!(MLP)!and!Single@Level!Perspective!(SLP)!

Phase&of&Research& Differentiating&Characteristics& MLP& SLP&

Formulation!of!

Theoretical!Model!

Construct! (i)!Narrower!definition! ! ✓! ✓!

! ! (ii)!Broader!definition! ! ✓! ✗!

! ! Level!of!construct!(LC)! (i)!For!any!(i,!j),!LCi!=!LCj;! ✓ ✓

! ! ! (ii)!There!is!(i,!j),!LCi!≠!LCj;! ✓ ✗

! Relationship!among!

constructs!

(i)!Structural!relation! ! ✓! ✗!

! ! (ii)!Causal!relation! (a)!Single@level!causal!

relation!

✓! ✓!

! ! ! (b)!Cross@level!causal!

relation!

✓! ✗!


structurally!related!

constructs!

Functional!equivalence! ! ✓ ✗

Operationalization!of!

Theoretical!Model!

Level!of!Measurement!(LM)! If,!for!any!(i,!j),!LCi!=!LCj;! (i)!For!any!i,!LCi!=!LMi;! ✓! ✓!

! ! ! (ii)!There!is!i,!LCi!≠!LMi;! ✓! ✗!

! ! If,!there!is!(i,!j),!LCi!≠!LCj;! (i)!or!(ii);! ✓! ✗!

Test!of!Theoretical!

Model!

Data!Collection! (i)!Hierarchical!data! ! ✓! ✗!

! ! (ii)!Non@hierarchical!data! ! ✓! ✓!

! Statistical!Technique! Multilevel!analysis!technique! ! ✓! ✗!

! ! Traditional!technique! ! ✓! ✓!

! Level!of!Statistical!Analysis!

(LSA)!

If,!for!any!i,!LCi!=!LMi;! (a)!LSAi!should!be!LMi;! ✓! ✓!

! ! If,!there!is!i,!LCi!≠!LMi;! (b)!LSAi!should!be!

transformed!from!LMi!

based!on!the!relevant!

structural!relationship!in!a!

theoretical!model;!

✓! ✗!

! Content!of!Test! (i)!Existence!of!level! ! ✓! ✗!

! ! (ii)!Construct!validity! ! ✓! ✓!

! ! (iii)!Structural!relationship! ! ✓! ✗!

! ! (iv)!Causal!relationship! ! ✓! ✓!

! Result!of!Test! (i)!Presence/absence!of!(a)!

single@level!causal!relation!and!

unknown!of!(b)!structural!

relation!or!(c)!cross@level!causal!

relation!

! ✓! ✓!

! ! (ii)!Presence/absence!of!(a)!

single@level!causal!relation,!

presence/absence!of!(b)!

structural!relation,!and!

presence/absence!of!(c)!cross@

level!causal!relation!

! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

! 16!



Formulation!of!

Theoretical!Model!






constructs!



relation!

✓! ✓!


relation!

✓! ✗!



constructs!



Theoretical!Model!





Model!






(LSA)!






theoretical!model;!

✓! ✗!









relation!

! ✓! ✓!







! ✓! ✗!

19

Table 6. Summary of the Key Terms Adopted in the Typology

Purpose Key Term Definition Example Differentiate “Kind”

Mono-Effects Top-down effects are characterized as mono-effect when a higher-level entity produces one or a similar kind of influence on its lower-level entities; or when a higher-level entity produces multiple kinds of influence on its lower-level entities and each of these kinds can have an independent effect irrespective of the presence of the other kinds. Bottom-up effects are characterized as mono-effect when lower-level entities make one or a similar kind of elementary contribution to their higher-level entity; or when lower-level entities make multiple kinds of elementary contribution to their higher-level entity and each of these kinds can have an independent effect irrespective of the presence of the other kinds.

Group culture and group size both independently affect individual performance.

Combined-Effects Top-down or bottom-up effects are characterized as combined-effect, when such influence or contribution consists of multiple kinds and these kinds interact and combine to effect.

Group culture and group size interact and combine to affect individual performance.

Differentiate “Amount”

Equally Distributed Effects

Top-down effects are characterized as equally distributed if each higher-level entity has influences similar in amount on its lower-level entities. Bottom-up effects are characterized as equally distributed if the bottom-up contributions from a group of lower-level entities are the same or similar in amount.

Each member’s attitude or opinion contributes to group consensual attitude equally in amount.

Unequally Distributed Effects

Top-down effects are characterized as unequally distributed if each higher-level entity has unequal influences on its lower-level entities. Bottom-up effects are characterized as unequally distributed if the bottom-up contributions from a group of lower-level entities are unequal in amount.

Organizational culture affects some employees’ job performance more than others’ job performance.

Differentiate “Target”

Effects Targeted at a Construct

Top-down or bottom-up effects affect the variance of a construct. Organizational culture affects individual performance.

Effects Targeted at a Relation

Top-down or bottom-up effects affect the occurrence or magnitude of a relationship.

Organizational IT training affects the extent to which individual IT usage promotes individual performance.

Top-Down Effects

Type I: Top-Down, Mono-, Equally Distributed Effects

Researchers may consider Type I effect, when a higher-level construct directly constrains or facilitates a

lower-level construct. Kozlowski and Klein (2000) called it “cross-level direct effect”. For example,

Cenfetelli and Schwarz (2011) posited that website quality (at a website level) may directly affect users’

system usage intention (at an individual user level). The extent of such top-down influence remains

20

invariable for all users of the same website. As noted by Kozlowski and Klein (2000), multiple such top-

down relationships may coexist in a single “mixed determinants model” where multiple higher-level

constructs affect one lower-level construct, or a single “mixed effects model” where one higher-level

construct affects multiple lower-level constructs.

Rather than influence a lower-level construct, a higher-level construct may moderate the occurrence or

magnitude of a lower-level relation (Kozlowski and Klein 2000). In other words, top-down effects may be

targeted at a relation rather than a construct. For example, an organization’s IT training support might

increase its employees’ effectiveness in using the IT system (a construct) (Niederman et al. 1996); or, culture

of a society (e.g., Japan, Switzerland, and United States) might influence the extent to which a person’s

perceived social norm will affect the person’s intention to adopt technology (a relation) (Straub et al. 1997).

This distinction between “targeting a construct” and “targeting a relation” applies to all types of top-down

and bottom-up effects (see the column “Target” of Table 4).

Based on this distinction, a variant of Type I effect is thus when a higher-level construct affects a lower-level

relation. Theory of strategic IT alignment considers both the industry and the firm levels (e.g., Tallon and

Pinsonneault 2011). Where IT is aligned with a firm’s strategy, the firm can become more agile. Firm

agility, however, is most useful when the parent industry is highly volatile, because agility can help a firm

respond more quickly to change. Hence, industry volatility (at the industry level) may moderate the firm-

level relationship between firm agility and performance.

Type II: Top-Down, Mono-, Unequally Distributed Effects

Top-down effects may equally or unequally affect target entities. For example, organizational culture might

affect some employees’ job performances significantly more than others’ job performances. This might be

because new organizational members need a period of time to assimilate organizational culture, and thus

new organizational members’ job performances might be less subject to the influence of organizational

culture. Top-down effects should be characterized as equally distributed if the top-down influence from the

higher-level entities is the same or similar in amount on the lower-level entities; otherwise, such top-down

21

effects should be characterized as unequally distributed (see the fourth column of Table 4).

Notably, equal or unequal distribution of top-down effects does not suggest that the values of the construct

of the focal entities within a group must be similar or different. If organizational culture affects employees’

job performances equally in magnitude within an organization, it does not suggest that the values of the job

performance construct for all the employees of the organization must also be the same.

Different from Type I effect, researchers should consider Type II effect when suspecting that the top-down

influence applies unequally in magnitude or degree to lower-level entities. The primary differences between

Type I effect and Type II effect are illustrated in Figure 2. In particular, Type II effect emphasizes the

importance of the relative position or location of an entity within a group, such as shorter- versus longer-

term members of an organization.

Figure 2. Distinction between Type I Effect and Type II Effect

Type II effect also has two variants depending on whether it is targeted at a lower-level construct or a

lower-level relation. The first variant is targeted at a lower-level construct. What is called “frog-pond effect”

(also called “heterogeneous, parts, or individual-within-the-group model” by others [e.g., Klein et al. 1994;

Kozlowski and Klein 2000]) represents an instance of Type II effect.

Consider groups of frogs with each group tied to a pond. A frog looks relatively larger if all other frogs in

the same pond are smaller (see Figure 3). For example, Frog A looks larger in Pond 1, but smaller in Pond

2, given other frogs nearby. Frog-pond effect refers to the phenomenon that the location of a frog may

enlarge or reduce its perceived size. It is an unequal distribution of top-down influence, because the pond

of a frog (at a higher level) only makes some frogs appear larger, but not all larger or with equivalent

Higher-LevelEn.ty

Lower-LevelEn.tyLoca.on

a b c d

++ ++ ++ ++

TypeIEffect

Loca.ona b c d

+++ ++ + ++

TypeIIEffect

22

magnitude. Frog C, for instance, looks relatively small in both Pond 1 and Pond 2; the pond has little

influence on the perceived size of Frog C relative to others.

Figure 3. Example of Frogs of Different Size Situated in Different Ponds

There are practical implications. Assume larger frogs are more competitive than smaller ones. A frog’s

absolute size may not matter. Instead, Frog A is more competitive in Pond 1 but less competitive in Pond 2.

Hence, Frog A will more likely survive in Pond 1 but less likely in Pond 2. It is thus a frog’s relative size that

matters most. In other words, the extent to which a frog’s size will determine survival also depends on the

frog’s particular circumstance (i.e., its pond or its neighboring competitors). In particular, a pond can have

unequal influence on frogs: a pond may help frogs that are small in an absolute sense but are comparatively

larger than their neighboring frogs to survive (e.g., Frog A); however, a pond may not help frogs that are

both small in an absolute sense and smaller than their neighboring frogs to survive (e.g., Frog C)5.

There is little doubt frog-pond phenomena can be qualitatively examined. To examine such phenomena

5 In other words, “the pond” has an influence on the lower-level entities. In many empirical settings, “the

pond” is a human collective such as a group or an organization.

A

CB’

A’ B’B’

C’ C’

C’C’

CC

C C

CC

Pond 1 Pond 2

Size of (A, C) = (80, 40); Average Frog Size in Pond 1 = (80 * 1 + 40 * 7) / (1 + 7) = 45; Relative Size of (A, C) = (80/45, 40/45) ≈ (1.78, 0.89);

Size of (A’, B’, C’) = (80, 120, 40); Average Frog Size in Pond 2 = (80 * 1 + 120 * 3 + 40 * 4) / (1 + 3 + 4) = 75; Relative Size of (A’, B’, C’) = (80/75, 120/75, 40/75) ≈ (1.07, 1.60, 0.53);

23

quantitatively, it is necessary to introduce further assumptions. A simple assumption commonly used is that

“the unequal portion” of top-down influence can be removed by converting the value of a variable to a

value relative to its group average (e.g., the average size of all the frogs in Pond 1) (Klein et al. 1994).

Afterward, such phenomena can be specified and quantitatively examined similar to analyzing single-level

models. For example, a frog’s size can be replaced by a size calculated as its size relative to the average size

of all frogs in the same pond; this computed relative size can then be used to predict its survival chances.

This treatment assumes pond boundaries would be eliminated as if all the frogs in Pond 1 and Pond 2 were

in one hypothetical pond. The converted size of frogs can predict their survival chances (“Relative Size” in

Figure 3).

For example, a member’s influence on a group decision to adopt IT may depend partly on the member’s

relative status within the group based on things such as educational attainment or experience using IT, as

compared with the group average (Sarker and Valacich 2010). Higher-status group members’ influence will

be less affected by their group membership, because they retain high influence after joining the group. In

contrast, for group members with relatively lower status, their influence might reduce significantly because

of their group membership. Consider another example: Resource-Based View (RBV) can be interpreted as

a model where the frog-pond effect applies to both independent and dependent variables. RBV explains

how diverse firm resources generate sustained competitive advantage (Barney 1991). Rarity of resource is

important: the value of a resource depends on its total amount possessed by all the firms in the same

industry. One central proposition in RBV is that rarer resources possessed by a firm – rare relative to the

average of the amount of resources possessed by all the firms in the same industry – may generate super-

normal firm performance – higher performance relative to the firm’s industry average performance.

Other assumptions can be used for quantitatively analyzing Type II effect. Extending Resource-Based

View, Relational View can further explain how strategic alliance networks can promote sustained

competitive advantage (Gulati 1998; Dyer and Singh 1998). According to Relational View, a firm with a

better structural position within its strategic alliance network may gain greater sustained competitive

24

advantage. Quantification of such a structural position within a strategic alliance network may rely on

social network measures (see Newman 2003). Calculating the exact value of a firm’s strategic position

relative to its embedded alliance network requires a chain of complicated mathematical operations. Given

this added complexity, whether the underlying rationale of the calculation can appropriately capture the

phenomenon under examination may become difficult to assess.

It is noteworthy that conversion from Type II effect to an “ostensibly single-level” model is mainly to assist

in quantitative description of phenomena; it never changes the multilevel nature of Type II effect. Such a

process can be illustrated with the following equation:

𝑋!! = 𝑓! 𝑋!,𝑋!,… ,𝑋!

𝑋! is the value of a variable X for an individual i. N is the total number of individuals in a group. 𝑋!! is the

value of variable X for individual i relative to the values of variable X for all the individuals in the group

(i.e., 𝑋!,𝑋!,… ,𝑋!). The conversion function 𝑓! relates specifically to individual i. If subtracting the average

of a group is considered an appropriate representation, the equation reduces to

𝑋!! = 𝑋! − 𝑋!

!

!

/𝑁 = 𝑋! − 𝑋!

where 𝑋! is group average for variable X (i.e., the simplest treatment as described by previous authors [e.g.,

Klein et al. 1994]). Further, some multilevel theorists regard Xi – Xg as at a distinctive level, separate from

either the lower (to which Xi relates) or the higher level (to which Xg relates) (Kozlowski and Klein 2000).

The frog-pond effect instantiates one possible scenario of the first variant of Type II effect. In particular,

the frog-pond effect captures the scenario where top-down influences unequally affect an independent

variable; in contrast, top-down influences may also unequally affect a dependent variable6. Consider

6 Although the distinction between targeting an independent variable and targeting a dependent variable

could be applied to other types of effect, the distinction is made only for Type II and Type IV effects. One

reason is to accommodate multilevel effects that have been examined in the existing literature (i.e., “frog-

25

another scenario where members of a group have various levels of sensitivity to the influence of group

norms: longer-term members may be more inclined to follow group norms, whereas shorter-term members

may be less affected by group norms. Given the difference in the sensitivity to group norms, a group norm

(e.g., “reinforcing the value of working hard” vs. “encouraging a balance between work and life”) may

unequally affect group member’s productivity at the individual level. Hence, the causal relation from group

norm at a higher level to productivity at a lower level depends on the seniority or “location” of an

individual member within a group (which cannot be characterized by frog-pond effect). In other words,

such a causal relation is a function of i, where the parameter i specifies the individual member i.

Where the top-down effect is targeted at a lower-level relation, the second variant of Type II effect may be

relevant. For example, a student’s academic performance relative to their class may increase the student’s

confidence and may further influence the student’s learning efficiency. Hence, the relationship between a

student’s time spent in learning and the student’s future performance may be moderated by the student’s

current academic performance relative to their class. Similar to the first variant, this variant needs to

identify some “position” or “relative status” related variables. The difference is that the researcher should

focus on analyzing how those variables might moderate lower-level relationships (rather than directly

affecting other variables).

Similarly, the earlier scenario of “top-down, unequal causal influence” may also be targeted at a lower-level

relation. For example, group norm may unequally influence the extent to which a person’s frequency of

using IT affects the person’s work performance. Namely, the “power” or “effectiveness” of group norm on

individual members of a group depends on a member’s particular circumstance.

Type III: Top-Down, Combined-, Equally Distributed Effects

Type I and Type II effects both focus on analyzing “mono-effect” where every kind of top-down influence

can be independently examined. However, there might be “combined-effect” where this precondition of pond effect”). Another reason is that we consider this distinction as a minor one when it is applied to other

types of effect.

26

independence is violated. To understand “mono-effect” versus “combined-effect” (see the third column of

Table 4), consider the effects of group culture and group size on individual performance. Assume group

culture and group size would promote individual performance only when the values of group culture and

group size are between medium and strong and between medium and large respectively (see Table 7). In

this case, it makes theoretical sense to analyze their combined-effect rather than analyze them separately.

In contrast, if such interaction is absent, and either group size or group culture or both influence individual

performance separately, the effects should be analyzed as mono-effects.

Table 7. Example of Combined-Effect due to Interaction between Group Culture and Group Size

Group Size

Large Medium Small

Group Culture

Strong Promoting Individual Performance

Effect Absent Medium Effect Absent

Weak Effect Absent Effect Absent Effect Absent

Another example of a top-down mono-effect is that the higher quality of an organization’s email system

may improve proficiency in communication of all its employees. Organizational culture may also affect

employees’ communication proficiency. Regardless, if organizational culture does not influence how the

email system might improve communication proficiency, the effect of the organizational email system on

communication proficiency should still be analyzed as a mono-effect.

Thus, top-down effects are characterized as mono-effect when a higher-level entity produces one or a

similar kind of influence on its lower-level entities; or when a higher-level entity produces multiple kinds of

influence on its lower-level entities and each of these kinds can have an independent effect irrespective of

the presence of all the other kinds. Top-down effects should be characterized as combined-effect, when

such influence or contribution consists of multiple kinds and these kinds interact and combine to effect.

Where there are multiple top-down, mono-effects or combined-effects in a single study, they should be

independently analyzed one by one.

Although the extent of mono-effects or combined-effects can be empirically tested, we suggest that before

committing to combined-effects researchers should have a strong theoretical explanation to justify the

27

presence of interaction; theory construction should be guided by interesting theoretical explanations, not by

theory testing (Weick 1989). For example, one such explanation might be that the influence of

organizational culture on the relationship between group system adoption and group performance may be

different for firms of various sizes. Organizational culture represents the core values and beliefs of an

organization, which are translated into organizational policies and routines. Organizational culture might

differ from one organization to another: there might be an “enhancing” culture where employees are often

inclined to follow the guidance of top management (i.e., culture is “strong”), a “chameleon” culture where

employees quickly adapt to the culture of the organizational unit assigned (i.e., culture is “medium”), or a

“countercultural” culture where employees are emotionally unrelated to the core values and beliefs of their

own organization (i.e., culture is “weak”) (Ravishankar et al. 2011). Further, for larger firms organizational

policies and routines may be designed to be more consistent with organizational culture and therefore

organizational culture may have a stronger influence on the relationship between group system adoption

and group performance, whereas for small to medium firms organizational policies and routines may not

comprehensively cover all aspects of employees’ organizational life and therefore the interaction effect

between organizational culture and organizational size may be less significant. Given this theoretical

explanation, omitting organizational size when studying how organizational culture affects a lower-level

relation, or, conversely, omitting organizational culture when studying how organizational size affects a

lower-level relation may lead to an incomplete understanding.

Researchers should consider Type III effect when simultaneously analyzing multiple kinds of top-down

influence. More specifically, one should consider Type III effect instead of Type I effect, if and only if the

multiple kinds of top-down effects cannot be separately analyzed; or, such effects cannot be decomposed

into multiple Type I effects. This scenario exists when interaction occurs; the differences between Type I

and Type III are illustrated in Figure 4.

28

Figure 4. Distinction between Type I Effect and Type III Effect

Further, Type III effect is characterized as “equally distributed” in the sense that the magnitude or degree

of the combined effect applies equally to all the lower-level entities. This suggests that once the combined

effect is identified – denoted as 𝑓 𝑋!,𝑌! – the combination of multiple higher-level variables (e.g., group

culture 𝑋! and group size 𝑌!) can be regarded as a single higher-level variable [e.g., 𝑍! = 𝑓 𝑋!,𝑌! ] that

affects lower-level entities. The model can then be further specified in the form of Type I effect. Similar to

Type I effect, Type III effect can be targeted at a construct or a relationship; thus, Type III effect also has

two variants, both of which are similar to Type I effect, but take possible interaction effect into

consideration.

Type IV: Top-Down, Combined-, Unequally Distributed Effects

When analyzing multiple kinds of top-down influence, researchers should consider Type IV effect if the

combined effect of the multiple kinds of top-down influence does not apply equally to all the lower-level

entities within a grouping. Again, similar to the previous three types, Type IV effect can have two variants

depending on its target (a construct or a relationship). Also similar to Type II effect, both the “frog-pond

effect” scenario and the “top-down, unequal causal influence” scenario of Type IV effect exist. We briefly

discuss the “frog-pond effect” scenario.

Assume that a “center” of cohesiveness exists in a medium-to-large sized group, where some members are

core while others are relatively peripheral. It might be that the previous combined effect of group culture

and group size does not apply equally to every member, but, instead, gradually reduces its power from core

members to peripheral members. That is to say, for members near the “edge” of a group, high values of

both group culture and group size cannot promote their performance.

Higher-LevelEn.ty

Lower-LevelEn.ty

TypeIEffect(onekind)

TypeIEffect(twokinds)

TypeIIIEffect(twokindswithinterac.on)

29

It is clear that the analysis of “unequal distribution of top-down influence” for Type IV effect shares much

conceptual similarity with the analysis of Type II effect. Specific to Type IV effect is the need to examine

how multiple kinds of top-down influence interact and combine. The second task, similar to Type II effect,

is to articulate appropriate assumptions that can enable quantitative specification and subsequently to

specify the model similar to a form of single-level model. For example, we might use a new variable called

“relative culture-size”, which is defined by

𝐷! − 𝐷! ×𝑓 𝑋!,𝑌!

In this equation, 𝑓 𝑋!,𝑌! represents the “absolute” combined effect for group culture and group size (see

Table 7). 𝐷! denotes the “distance” to the center of a group for member i. 𝐷! denotes the distance from the

edge of a group to its center. When taking “unequal distribution” into consideration, we might submit:

relative culture-size may promote individual performance.

Source of Top-Down Effects

In ending this section on top-down effects, we note that Klein et al. (1994) urged researchers to consider

possible “sources” of multilevel phenomena. By “source”, they refer specifically to plausible, candidate

theoretical explanations. What sources of top-down effects exist in the IS discipline? There exist several

well-known top-down processes such as cultural influence and group norms on IT use behaviors (e.g., Rai

et al. 2009; Bock et al. 2005), as well as institutional influences such as top management support and size of

firm on IT implementation, IT project success, and IT acceptance. To consider different kinds of top-down

influence, the researcher can also think of diverse IT features. For example, where an integrated Enterprise

System is used, employees may need to use separate features such as planning and reporting functionality

for business analysis, and accounting functionality for managing financial resources. There are plenty of

opportunities for examining possible synergies or contradictions that may derive from using separate parts

of a larger system and investigating how such synergies or contradictions might affect individuals, work

groups, and departments.

30

Bottom-Up Effects

Bottom-up effects also have four generic types, each type having two variants; the two variants respectively

consider bottom-up effects that are targeted at a construct and a relationship. When analyzing bottom-up

effects, we draw primarily from Morgeson and Hofmann’s (1999) idea of the “structure” of “collective

constructs” and elaborate Kozlowski and Klein’s (2000) conceptual analysis of emergence. As argued by

Morgeson and Hofmann (1999), some phenomena about collectives (e.g., groups consisting of individual

members) may originate theoretically from cognitions, behavioral processes, or characteristics of their

lower-level counterparts. They call the conceptual abstraction of such phenomena “collective constructs.”

An example is group performance where the performance of a group may be constituted by the sum of

individual performances. The “structure” of a collective construct refers to “its lower-level ongoings and

events and the interaction of ongoings and events” that give rise to the collective construct (Morgeson and

Hofmann 1999).

Identification of bottom-up effects must make clear the structure of the collective construct (Morgeson and

Hofmann 1999). Put differently, it is only the conceptual articulation of a construct’s lower-level entities

and the interactions among its lower-level entities that can offer a legitimate foundation for theorizing or

conceptualizing bottom-up effects. Bottom-up effect describes “the manner in which lower-level properties

emerge to form collective phenomena” (Kozlowski and Klein 2000, p. 15).

Before identifying bottom-up effects, how do we know which two or more constructs at separate levels

could be theoretically linked by bottom-up effects? For instance, before knowing group performance is a

sum of individual performances, why should we analyze the relationship between group performance and

individual performance? Morgeson and Hofmann (1999) offered one suggestion. They refer to the function

of a construct as the outcome or consequence of the construct in its nomological network. For example,

both individual and group system usages can increase performance (Burton-Jones and Gallivan 2007). We

can thus say that individual and group system usages are functionally equivalent because they lead to the

same outcome in their nomological network. As argued by Morgeson and Hofmann (1999), if two

31

constructs are functionally equivalent, they may be theoretically linked through bottom-up effects. Given

the function of collective constructs can suggest potential relevance of constructs at separate levels, it can be

useful to consider the function of constructs before analyzing bottom-up effects. With regards to this issue,

Burton-Jones and Gallivan (2007) offered extensive guidelines on how to analyze and conceptualize system

usage at both individual and group levels.

Our analysis of bottom-up effects draws primarily from Morgeson and Hofmann (1999) and Kozlowski and

Klein (2000). In particular, Type V to Type VIII effects further elaborate the continuum of emergence

from compositional to compilational proposed by Kozlowski and Klein (2000). Compositional emergence is

based on an underlying model where “the type and amount of elemental content – the raw material of

emergence – are similar for all individuals in the collective,” whereas compilational emergence is based on

an underlying model where “either the amount or type of elemental content is different, or both the

amount and type are different” (p. 62) (see Figure 5).

Figure 5. Comparison between Compositional and Compilational Emergence

For example, consider a team of athletes standing in a line and passing basketballs from one end to the

other end. For compositional emergence, there is only one type or kind of task that is performed in a group

(e.g., passing basketballs). And every member of the group also contributes equal amount to the outcome of

the collective task (e.g., the number of basketballs passed within a period of time). In contrast, for

compilational emergence, there are multiple types or kinds of tasks that are performed in a group; or, there

is only one type or kind of task that is performed in a group, but group members contribute unequal

amounts to the outcome of the collective task. For example, where members of a team serve different roles

in a competition, the success or failure of the competition derives from a compilation emergence process

Composition Process

Emergence

Compilation Process (Amount Different)

Emergence

Compilation Process (Type Different)

Emergence

Compilation Process (Both Different)

Emergence

32

(i.e., type of contribution is different). Consider another example of compilation emergence where one task

is performed and the amounts of contribution are different. Where a natural group (i.e., selected from a

natural population) is instructed to try their best to move basketballs from one point to another point

(assuming there is no differentiation of roles), the total number of basketballs transferred within a period of

time by the group is likely contributed unequally by the group members. This is because in a natural

population there is often differential level of skill or fitness; therefore, it is likely that some members of the

group would transfer more basketballs than other members.

Kozlowski and Klein’s (2000) “type” and “amount” of elemental content are aligned with our notions of

“mono-effect versus combined-effect” and “equally distributed effect versus unequally distributed effect”.

In particular, the distinction between mono-effect and combined-effect applies similarly to bottom-up

effects. As example of a bottom-up combined-effect, a group’s decision performance might be influenced

by multiple kinds of lower-level factors such as members’ communication or interaction style, and their

individual work performance. If different communication styles differentially influence the effectiveness of

individual member’s work performance on the group’s decision performance, this effect is better analyzed

as a bottom-up combined-effect. Bottom-up effects should be characterized as mono-effect when lower-

level entities make one or a similar kind of elementary contribution to their higher-level entity; or when

lower-level entities make multiple kinds of elementary contribution to their higher-level entity and each of

these kinds can have an independent effect irrespective of the presence of all the other kinds. Bottom-up

effects should be characterized as combined-effect, when such contribution consists of multiple kinds and

these kinds interact and combine to effect. Where there are multiple bottom-up mono-effects or combined-

effects in a single study, they should be independently analyzed one by one.

Moreover, the distinction between equal and unequal distribution also applies similarly to bottom-up

effects. Bottom-up effects should be characterized as equally distributed if the bottom-up contributions are

the same or similar in amount from the lower-level entities; otherwise, such bottom-up effects should be

characterized as unequally distributed.

33

Type V: Bottom-Up, Mono-, Equally Distributed Effects

Type V effect is the simplest form of bottom-up effects. Researchers should consider Type V effect when

analyzing one kind of bottom-up elementary contribution and anticipating that this kind is contributed

equally in magnitude by the lower-level entities. Like equal or unequal distribution of top-down influence,

equal or unequal distribution of bottom-up contribution does not suggest that the values of the construct of

the focal entities within a group must be similar or different. Further, researchers should consider the first

variant of Type V effect if the bottom-up contribution is also targeted at a higher-level construct. An

example is the formation or dissolution of a virtual community (Ransbotham and Kane 2011). Arguably, if

every member of the community is equally important, a member’s action to join or leave the community

should contribute equally to the community size.

Notably, for Type V effect, only the effective parts of bottom-up contributions count. For example, only the

slowest member within a team limits the total time required for the team to trek through a jungle, although

members’ capabilities may differ. While some members may trek relatively faster, not all of their efforts are

effectively used; only the parts that equal to the contributions devoted by the slowest member are effectively

contributing to the team outcome as a whole.

Type V effect encompasses what Kozlowski and Klein (2000) called “a shared unit property”. The values

of a lower-level construct for members of a group may converge to a single value describing a higher-level

construct; such a higher-level construct is called “a shared unit property”. Common examples include a

collective’s attitudes and opinions. Members of a group may have their own attitudes toward accepting IT

(e.g., Sarker and Valacich 2010; Kang et al. 2012). Where their attitudes are highly consistent (e.g., all

strongly agree or strongly disagree), it may be appropriate to use the consensual value of members’

attitudes to represent the value of a higher-level construct – for example, a group’s attitude toward IT

(Sarker and Valacich 2010). Hence, for a shared unit property, the value of the higher-level construct is

equivalent to (rather than a sum of) the value of the lower-level construct.

The distinction between “targeting a construct” and “targeting a relation” also applies to bottom-up effects.

34

One or more lower-level constructs might emerge at a higher level, contributing to a higher-level construct

or affecting a higher-level relation. For example, group culture (a higher-level construct) might be a

composite of all the group members’ personal characteristics such as personalities and work habits (a lower-

level construct). The communication network of a team, which is comprised of members’ communication

preferences at the individual level (a lower-level construct), might influence the extent to which the team’s

adoption of decision support systems will increase the team’s decision efficiency (a higher-level relation).

As another example, rather than directly affect a group’s attitude toward IT, individual members’ attitudes

might emerge at the group level, affecting the extent to which the group’s use of IT will increase the

group’s performance. If members of a group unanimously accept IT, the group’s use of IT might more

strongly increase the group’s performance.

Notably, when members of a group have the same values regarding a construct, in many cases, what is

more important or theoretically intriguing is to consider the group as a whole and thereby to examine the

emergent higher-level construct or relation, because those lower-level members are identical regarding the

construct of interest. In fact, if this occurs, researchers do not have a good way to examine the lower-level

construct because the values of the construct do not vary within a group. Conversely, researchers often care

about the lower-level construct when the values of the construct do vary within a group and therefore may

consider the lower-level construct in the specification of multilevel effects (which are discussed in Type VI

effect and Type VIII effect).

Type VI: Bottom-Up, Mono-, Unequally Distributed Effects

When focusing on one kind of bottom-up contribution and this kind is contributed unequally in magnitude

by the lower-level entities, researchers should consider Type VI effect. The first variant is targeted at a

higher-level construct. The critical difference of this effect from Type V effect is that the bottom-up

contributions are unequal in amount according to some quantifiable assessment7. For example, when 7 Although equally distributed and unequally distributed bottom-up effects may need different aggregation

or measurement approaches, the distinction is drawn to analyze the quantitative differences of the bottom-

35

group members perform “additive tasks” such as making screws, the performance of the group may be a

sum of individual task performances. However, individual performances may vary due to the difference in

skill and experience (McGrath 1984). Moreover, an industry barrier that restricts new participants may

arise from the scarcity of resources held by the existing participants (e.g., bandwidth in a

telecommunication industry) (Porter 1998). The more valuable and more difficult to acquire the resources

are, the less likely the barrier will be overcome. Since existing industry participants may have different

portions of such resources, their contributions to the industry barrier will likely vary quantitatively.

For Type VI effect, a higher-level construct may be a sum of a lower-level construct (e.g., addition of

individual work performance) or may capture a pattern of a lower-level construct. For example, the IS use

of a group may be represented as a pattern of the members’ frequencies of using IS (Burton-Jones and

Gallivan 2007). Table 8 illustrates various situations of Type VI effect, contrasted with Type V effect.

up contributions between different phenomena, rather than to typify aggregation approaches. For instance,

“an additive [aggregation] model” suggested by Chan (1998) may be applied to both Type V and Type VI

effects (e.g., see the previous example of “the size of a community”). Further, it is important to differentiate

phenomena based on their theoretical nature (e.g., are the bottom-up contributions equally distributed or

unequally distributed among members of a group), instead of aggregation approaches. Given there are no

universal ways to measure “bottom-up equal effects” versus “bottom-up unequal effects”, we recommend

researchers to describe the quantitative similarity or difference of bottom-up contributions among members

of a group, and to use their judgment in deciding the appropriate measurement approach for a particular

empirical setting.

36

Table 8. Comparison between Type V Effect and Type VI Effect

Type V Effect Type VI Effect

Example (1) Task Performance (2) Shared Attitude (3) Task Performance (4) Majority Attitude (5) Patterned Usage

Level 2

Level 1

The conceptual relationship between lower-level and higher-level constructs can be represented using more

complex combination rules such as social network measures. For instance, to conceptualize a group’s IS

avoidance, Kane and Labianca (2011) aggregated IS avoidance values of the group’s members. In the

aggregation, members’ IS avoidance values are weighted by their “centrality”, which assesses a person’s

closeness to the center of the communication network of the group. The ties in a communication network

represent interpersonal communication. The rationale is that if a person is more central in terms of

communicating to more members or more frequently, the person’s IS avoidance behavior should be more

representative of the group and therefore should have a greater weight.

Researchers should consider the second variant of Type VI effect, if lower-level contributions emerge at a

higher level, affecting the occurrence or magnitude of a higher-level relationship. For example, a group’s

majority attitude or the attitude of the group’s opinion leader toward adopting IT might emerge at the

group level to affect the causal relationship between the group’s IT usage frequency and the group’s work

performance.

Type VII: Bottom-Up, Combined-, Equally Distributed Effects

Type VII effect should be considered when a higher-level construct is affected by multiple kinds of lower-

level elementary contributions and each kind of contribution comes equally from all the lower-level entities;

and these kinds of elementary contributions interact and combine to affect higher-level entities.

Consider the first variant of Type VII effect, which is targeted at a higher-level construct. It is common for

members of an IT project team to work on different but interconnected modules of a larger IT system.

Observations

10

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

20

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

20

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

Observations

10

Observations

50

1L

1H

Observations

Yes

2L

2H

No

Observations

Yes

No

3L

3H

Observations

10

20

Observations

70

Observations

Yes4L

4H

No

Observations

Yes

No

Observations

10

5L203040

Observations

10

5H203040

37

Group research suggests that team members may need to share a common knowledge background such as

programming language and program design patterns in order to collaborate effectively. Communication

research further suggests that communication skills such as verbal communication skill and the skill in using

formal language to represent complex notions may also influence the efficiency of team collaboration.

Although team members may have differing extent of expertise or communication skill, it is only the

overlapping knowledge and the minimum level of communication skill of a team that can contribute to its

cooperative capability (see Table 9). This is because for highly cooperative tasks only the shared part of

knowledge background may actually get used and the efficiency for transmitting messages across an entire

group depends on the person who has the poorest communication skill. Further, knowledge background

and communication skill can be complementary, because greater communication skill can facilitate

knowledge sharing and a larger set of overlapping knowledge can make communication tasks simpler by

reducing the amount of information communicated. Hence, group cooperative capability can be a fit

between a team’s shared knowledge background and minimum communication skill.

Table 9. Example of Type VII Effect

Level 2: Group Cooperative Capability (as a match between shared knowledge and minimum communication skill)

Level 1: Individual Knowledge Background and Individual Communication Skill (because only shared or minimum parts are effectively contributing to a higher level, the contributions are equally distributed for a particular kind.)

Researchers should consider the second variant of Type VII effect, if lower-level contributions emerge at a

higher level, affecting the occurrence or magnitude of a higher-level relationship. In the previous example,

rather than contribute directly to the team’s cooperative capability, members’ knowledge background and

communication skill might emerge to moderate the causal relationship between the number of members in

the team and the work performance of the team. In other words, allocating more human resources to the

Knowledge Background

Shared

Communication Skill

Minimum

Shared Knowledge

Minimum Communication Skill

Cooperative Capability

Low

Low

Low

High


Shared

Communication Skill

Minimum

Shared Knowledge



Low

Low

Low

High


Shared

Communication Skill

Minimum

Shared Knowledge



Low

Low

Low

High

38

IT project will increase the team’s performance if and only if the team’s shared knowledge background fit

its minimum communication skill.

Type VIII: Bottom-Up, Combined-, Unequally Distributed Effects

When focusing on multiple kinds of bottom-up contributions and each kind is also contributed unequally

by the lower-level entities, researchers should consider Type VIII effect. Similar to Type VII effect, these

kinds of elementary contributions are also expected to interact and combine in some way.

Consider the first variant of Type VIII effect, which is targeted at a higher-level construct. Group

performance may result from a combination of different types of tasks; for each type of task, individual

performances may vary significantly. For instance, team members of an IT project may perform additive

tasks, the outcomes of which can be aggregated (such as programming), and problem-solving (non-additive)

tasks, the process and outcome of which cannot be separated (such as collectively thinking about novel

solutions for a new problem encountered). Given individual differences, members’ performances on either

task may vary.

Given the complexities involved in capturing such underlying emergence process, there is no well-accepted

solution for Type VIII effect. Ideally, the conceptualization of Type VIII effect needs to consider the real

coordination process in which the temporal patterns of different types of tasks and the interaction among

group members are considered. Burton-Jones and Gallivan (2007) pointed to the relevance of considering

the timing and temporal patterns of system usage behaviors in conceptualizing group usage. These aspects

may help researchers construct appropriate measurement models to specify Type VIII effect. Furthermore,

insights from qualitative research and theory about collective behaviors such as coordination theory may be

used to construct measurement models of Type VIII effect.

Researchers should consider the second variant of Type VIII effect, if lower-level contributions emerge at a

higher level, affecting the occurrence or magnitude of a higher-level relationship. For example, a group’s

communication network may moderate the relationship between the group’s IT usage and the group’s

work performance. This communication network may emerge from an integration of two kinds of network

39

structures: the group’s personal communication network (e.g., face-to-face meetings and casual contacts)

and the group’s email communication network. Because each individual possesses a unique position in a

network, the contributions of individual members to the network structure are also quantitatively different.

Hence, the contributions from individual group members to the group’s communication network are

different in both kind and amount.

It is worthwhile to discuss the notion of “configural unit properties”, given it has been espoused in several

IS studies (e.g., Kane and Borgatti 2011; Kane and Libianca 2011). The term “configural unit property”

used by Kozlowski and Klein (2000) refers to a higher-level construct (or, a construct regarding a collective

of individuals) that is contributed by different kinds of lower-level contributions, or the same kind but

different amounts of lower-level contributions. Thus, configural unit properties are formed often due to one

or more of the effects from Type VI to Type VIII, and they capture an “array, pattern, or configuration of

individuals’ characteristics within a unit” (Kozlowski and Klein 2000, p. 30) such as some stable pattern of

system usage frequency as in Type VI effect, a group’s cooperative capability that represents a combination

of diverse group knowledge and skills as in Type VII effect, and a group’s communication network

structure as in Type VIII effect.

It is also useful to compare and contrast configural unit properties with frog-pond effects. Both configural

unit properties and frog-pond effects are used to capture certain “differences” of the members of a group.

They both deal with the scenario where members of a group need to be conceptualized as dissimilar for

some justifiable reasons, instead of assuming that members of a group are similar or converge over time.

However, the two different effects have critical differences. First, their origins are quite different. Configural

unit properties exist because of bottom-up contributions by constructs at a lower level, whereas frog-pond

effects exist because of influences from a higher level. And for frog-pond effects, it is not always clear which

construct at a higher level caused frog-pond effects; we may know the importance of “relative size”, but

may not know which higher-level construct(s) made “relative size” more important than “absolute size”.

Second, frog-pond effects characterize only an instance of the scenarios where top-down influences are

40

unequal in amount to lower-level members of a group; in contrast, configural properties characterize both

the scenario where bottom-up contributions are unequal in amount from lower-level members of a group

and the scenario where bottom-up contributions have multiple kinds interacting and combining to effect.

Source of Bottom-Up Effects

What are the sources of bottom-up effects in the IS discipline? Because many phenomena in IS are studied

at an individual level and some are studied at a group level, there exist many opportunities to use these

studies to examine their emergent, higher-level counterparts. For example, individual IT implementation,

adoption, usage, and continued usage may provide potential opportunities for examining how these

individual behaviors emerge at a group, departmental, or organizational level.

Null Type

Is top-down or bottom-up effect always relevant? If a higher-level entity does not affect a lower-level entity

or conversely a higher-level entity does not originate from a lower level, top-down or bottom-up effect may

be absent. We refer to these two situations as Null types.

The top-down Null type is called “individuals free of group influence” by Klein et al. (1994). Higher-level

membership (e.g., group, departmental, and organizational membership) may barely affect constructs or

relationships at a lower level. For example, employees’ negative affectivity and managers’ early career

progress are not affected by their organizational membership (Klein et al. 1994).

Lower-level entities may have little contribution to a higher-level entity; or, such contribution is not clearly

identifiable. Kozlowski and Klein (2000) designated the term “global unit properties” to refer to constructs

that both originate at and manifest at the collective level. These constructs usually entail objective and

descriptive measures and do not manifest obvious lower-level origins. An example is group size, which

describes a group, but does not have lower-level emergent processes or properties. For group functions such

as marketing and purchasing, how individual members’ functions are combined and emerge at a group

level may not be clearly identifiable; group functions may thus be regarded as having no lower-level origin.

41

Summary of Typology

The typology classifies diverse types of top-down and bottom-up effects along two dimensions: (i) mono-

versus combined-effect and (ii) equal versus unequal distribution. Top-down or bottom-up effects can be

characterized by one or multiple of the four types, based on the distinction made between mono-effect and

combined-effect and the distinction between equal distribution and unequal distribution. Each type of top-

down or bottom-up effect has two variants according to its target. Taken together, theorizing with zoom-

out and zoom-in strategies should thus consider at least 16 archetypes of multilevel model (2×2×2×2).

Ignoring any distinction drawn here would not exhaustively consider all possible multilevel models.

Undoubtedly, no matter how granular the distinctions of multilevel effects are, finer distinctions are always

possible. Although the typology proposed herein is intended to encompass all multilevel effects described in

the existing literature, researchers may encounter multilevel phenomena that belong to a special instance of

a multilevel effect in the typology or that are not covered in the typology. If this occurs, researchers need to

appropriate the ideas here for new scenarios encountered or to extend the typology toward a broader range

of multilevel phenomena.

Note that the distinction between “mono-effect” and “combined-effect” herein is similar to the distinction

between “direct effect” and “interaction effect” in the statistical analysis of single-level models. We did not

adopt the labels “direct effect” and “interaction effect” to avoid possible confusion. The term “direct effect”

is also used by Kozlowski and Klein (2000) to refer to models that include a causal relationship that begins

from a higher-level variable and is targeted at a lower-level variable. The term “combined-effect” defined

herein refers specifically to a subset of “interaction effects” – namely, only cross-level relationships.

Notably, although the decision to analyze mono-effect or combined-effect depends mainly upon the nature

of the phenomenon, a phenomenon may include both mono- and combined-effects. Researchers thus need

to decide further on which aspects of the phenomenon to focus. If a problem domain is relatively nascent,

immature, or under-explored, the researcher may opt to study mono-effects by focusing solely on one kind

of top-down influence or bottom-up contribution. For example, researchers may study how individual use

42

of spreadsheet software might emerge at a group level thereby increasing the group’s work productivity, or

how individual use of an email system might form a pattern at a group level thereby improving the group’s

communication efficiency (Burton-Jones and Gallivan 2007). If a problem domain has matured to a point

where isolated kinds of top-down influence or bottom-up contribution are conclusively examined, the

researcher may consider analyzing combined-effects. For example, researchers may study how a group

attitude (e.g., skeptical, resistant to change, and thus inclined to maintain existing technologies vs. proactive

and positive toward creativities and novel technologies) might form as a result of both the attitudes of a few

elite members and the consensual attitude of majority; or how diverse features or functions of Enterprise

Systems (e.g., operational features vs. strategic features) might be used together to improve organizational

performance.

Furthermore, pure equal versus unequal distributions (a dichotomy) are arguably rare in empirical studies;

there is more likely a degree of equal or unequal distribution. It follows that researchers usually need to

devise a threshold to distinguish equal distribution and unequal distribution in empirical studies. However,

there is no standard threshold or simple way to make such a decision; researchers need to exercise their

best judgment and full discretion based on theoretical and empirical considerations (Burton-Jones and

Gallivan 2007; Kozlowski and Klein 2000).

A further point deserves some clarification. Although we used the term “effects” to encompass both “top-

down influences” and “bottom-up contributions”, they have critical differences. In particular, top-down

influences characterize cross-level, causal relationships. In contrast, bottom-up effect types characterize

“structural relationships” that link a higher-level construct to its lower-level counterparts – for example,

how group work performance is composed of work performances of individual group members. Where a

higher-level construct is not directly observable, such structural relationships may be used to operationalize

the higher-level construct, through observing the values of its structurally linked lower-level variable.

Alternative Conceptualizations

Researchers should also be attentive to possible alternative conceptualizations of phenomena (Klein et al.

43

1994). Although different types of top-down and bottom-up effect are separately analyzed, it is possible to

include multiple types in a specific context (e.g., Burton-Jones and Gallivan 2007). Furthermore,

distinguishing one type from another is not always straightforward. Researchers need to justify their choices

both theoretically and empirically. It can also be useful to present all the meaningful conceptualizations and

comparatively examine each. In doing so, researchers may be able to refine their thinking, spark creativity

for preferred choices, or generate new research questions for future investigation (Klein et al. 1994).

Implications for Information Systems Research

Multilevel theorizing is primarily driven by the desire to characterize phenomena with richer explanation

and more realistic representation. Information systems research encompasses a diversity of often isolated,

but increasingly complex research topics. We believe multilevel theorizing can offer IS researchers many

unexplored opportunities for dealing with these research topics. The proposed framework seeks to help IS

researchers better leverage such opportunities.

In particular, a single research domain, such as studies that investigate the success or effectiveness of IT, is

often examined at separate levels of analysis (Petter et al. 2008; Seddon et al. 1999). Researchers may use

the framework in this paper to develop novel theoretical insights through linking fragmented pieces of

studies within a single research domain. For example, researchers may investigate whether or how increase

in individual productivity after implementing IT might lead to increase in performance at the departmental

level as well as the organizational level. And what factors might affect the emergence of individual-level IS

success? Further, for separate research domains, researchers may also explore how diverse areas of studies

at separate levels might inform understandings of the focal phenomenon. For example, researchers may

examine how organizational structures and governance policies might affect IT implementation success or

individual IT use behaviors after adoption of IT. Researchers may also explore how individual interactions

within a team might emerge to affect the performance of using healthcare systems or business intelligence

systems at the team level and at the organizational level. Although IS phenomena usually span multiple

levels of analysis, many of such multilevel research questions have never been answered or explored in the

44

IS discipline. This opens a great range of opportunities for entering into an untapped domain of multilevel

phenomena.

Conclusions

Admittedly, multilevel theorizing is a broad subject. Our discussion in this paper is limited in several ways.

First of all, we adopted a positivist view. Although this is largely consistent with previous authors in this field

(e.g., Burton-Jones and Gallivan 2007; Kozlowski and Klein 2000), multilevel theorizing is not

incompatible with other schools of philosophical view. Furthermore, we did not address unique challenges

that may arise from qualitative approaches to multilevel theorizing. Nonetheless, we argue that the

framework developed in this paper has general value to both quantitative and qualitative researchers.

Theorizing or analyzing phenomenon quantitatively may also help qualitative researchers deepen their

understanding of phenomenon, even if they may not quantitatively test their model in empirical settings.

For instance, researchers who follow qualitative approaches also need to use conceptual frameworks to

make sense of empirical settings or to organize qualitative evidence (e.g., Lapointe and Rivard 2005).

Lastly, multilevel theory development is an emerging field. Given existing tools or statistical techniques deal

mainly with linear modeling of phenomena, complex models of multilevel phenomena can be challenging

to analyze or test. Our efforts to conceptualize diverse forms of multilevel phenomena only make some first

steps toward a comprehensive solution. There is need for future research to address how to analyze and test

ever-increasingly complex types of multilevel phenomena.

Acknowledgments

We sincerely thank the senior editor and the two anonymous reviewers for their insightful and constructive

comments. This paper derives from the first author’s doctoral dissertation; we would like to thank the two

dissertation examiners, Kalle Lyytinen and Andrew Burton-Jones, for helping the authors refine the ideas

presented in the paper. We also thank Joerg Evermann for commenting on an earlier version of this paper.

This research is financially supported by Australian Research Council (ARC) Discovery Project Grant

45

titled Towards Engineering Research Systems: Systematic Modeling of Behavioural Scientific

Research Methods (DP150101022). We thank Karen Stark for research assistance. Errors are solely ours.

References

Bacharach, S. B. 1989. Organizational Theories: Some Criteria for Evaluation. Academy of

Management Review 14(4) 496–515.

Barney, J. 1991. Firm Resources and Sustained Competitive Advantage. Journal of Management 17(1)

99–120.

Bélanger, F., Cefaratti, M., Carte, T., and Markham, S. E. 2014. Multilevel Research in Information

Systems: Concepts, Strategies, Problems, and Pitfalls. Journal of the Association for Information

Systems 15(9) 614–650.

Bock, G. W., Zmud, R. W., Kim, Y. G., and Lee, J. N. 2005. Behavioral Intention Formation in

Knowledge Sharing: Examining the Roles of Extrinsic Motivators, Social-Psychological Forces, and

Organizational Climate. MIS Quarterly 29(1) 87–111.

Burton-Jones, A., and Gallivan, M. J. 2007. Toward a Deeper Understanding of System Usage in

Organizations: a Multilevel Perspective. MIS Quarterly 31(4) 657–679.

Cenfetelli, R. T., and Schwarz, A. 2011. Identifying and Testing the Inhibitors of Technology Usage

Intentions. Information Systems Research 22(4) 808–823.

Chan, D. 1998. Functional Relations Among Constructs in the Same Content Domain at Different Levels

of Analysis: a Typology of Composition Models. Journal of Applied Psychology 83(2) 234–246.

Davis, F. D. 1989. Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information

Technology. MIS Quarterly 13(3) 319–340.

DeLone, W. H., and McLean, E. R. 1992. Information Systems Success: the Quest for the Dependent

Variable. Information Systems Research 3(1) 60–95.

DeLone, W. H., and McLean, E. R. 2003. The DeLone and McLean Model of Information Systems

Success: a Ten-Year Update. Journal of Management Information Systems 19(4) 9–30.

46

Dubin, R. 1978. Theory Building, Free Press, New York, US.

Dyer, J. H., and Singh, H. 1998. The Relational View: Cooperative Strategy and Sources of

Interorganizational Competitive Advantage. Academy of Management Review 23(4) 660–679.

Gallivan, M. J., and Benbunan-Fich, R. 2005. A Framework for Analyzing Levels of Analysis Issues in

Studies of E-Collaboration. IEEE Transactions on Professional Communication 48(1) 87–104.

Gregor, S. 2006. The Nature of Theory in Information Systems. MIS Quarterly 30(3) 611–642.

Gulati, R. 1998. Alliances and Networks. Strategic Management Journal 19(4) 293–317.

House, R., Rousseau, D. M., and Thomas-Hunt, M. 1995. The Meso Paradigm - a Framework for the

Integration of Micro and Macro Organizational-Behavior. Research in Organizational Behavior 17

71–114.

Hox, J. 2010. Multilevel Analysis: Techniques and Applications, Taylor and Francis.

Jaccard, J., and Jacoby, J. 2010. Theory Construction and Model-Building Skills: a Practical Guide

for Social Scientists, Guilford Press, Guilford Publications Inc., New York, US.

Kane, G. C., and Borgatti, S. P. 2011. Centrality-IS Proficiency Alignment and Workgroup Performance.

MIS Quarterly 35(4) 1063–1078.

Kane, G. C., and Labianca, G. 2011. IS Avoidance in Health-Care Groups: a Multilevel Investigation.

Information Systems Research 22(3) 504–522.

Kang, S., Lim, K. H., Kim, M. S., and Yang, H.-D. 2012. A Multilevel Analysis of the Effect of Group

Appropriation on Collaborative Technologies Use and Performance. Information Systems Research

23(1) 214–230.

Klein, K. J., Dansereau, F., and Hall, R. J. 1994. Levels Issues in Theory Development, Data-Collection,

and Analysis. Academy of Management Review 19(2) 195–229.

Kozlowski, S. W. J., and Klein, K. J. 2000. A Multilevel Approach to Theory and Research in

Organizations: Contextual, Temporal, and Emergent Processes. Multilevel Theory, Research, and

Methods in Organizations, K. J. Klein and S. W. J. Kozlowski (eds.), 3–90.

Lapointe, L., and Rivard, S. 2005. A Multilevel Model of Resistance to Information Technology

47

Implementation. MIS Quarterly 29(3) 461–491.

Mathieu, J. E., and Chen, G. 2011. The Etiology of the Multilevel Paradigm in Management Research.

Journal of Management 37(2) 610–641.

McGrath, J. E. 1984. Groups: Interaction and Performance, Prentice-Hall, Englewood Cliffs, US.

Melville, N., Kraemer, K., and Gurbaxani, V. 2004. Review: Information Technology and Organizational

Performance: an Integrative Model of IT Business Value. MIS Quarterly 28(2) 283–322.

Morgeson, F. P., and Hofmann, D. A. 1999. The Structure and Function of Collective Constructs:

Implications for Multilevel Research and Theory Development. Academy of Management Review

24(2) 249–265.

Newman, M. E. J. 2003. The Structure and Function of Complex Networks. SIAM Review 45(2) 167–

256.

Niederman, F., Beise, C. M., and Beranek, P. M. 1996. Issues and Concerns About Computer-Supported

Meetings: the Facilitator's Perspective. MIS Quarterly 20(1) 1–22.

Petter, S., DeLone, W., and McLean, E. 2008. Measuring Information Systems Success: Models,

Dimensions, Measures, and Interrelationships. European Journal of Information Systems 17(3) 236–

263.

Porter, M. E. 1998. Competitive Strategy: Techniques for Analyzing Industries and Competitors,

Free Press, New York, US.

Rai, A., Maruping, L. M., and Venkatesh, V. 2009. Offshore Information Systems Project Success: the

Role of Social Embeddedness and Cultural Characteristics. MIS Quarterly 33(3) 617–641.

Ransbotham, S., and Kane, G. C. 2011. Membership Turnover and Collaboration Success in Online

Communities: Explaining Rises and Falls From Grace in Wikipedia. MIS Quarterly 35(3) 613–627.

Ravishankar, M. N., Pan, S. L., and Leidner, D. E. 2011. Examining the Strategic Alignment and

Implementation Success of a KMS: a Subculture-Based Multilevel Analysis. Information Systems

Research 22(1) 39–59.

Rivard, S. 2014. The Ions of Theory Construction. MIS Quarterly 38(2) iii–xiv.

48

Rousseau, D. M. 1985. Issues of Levels in Organizational Research: Multi-Level and Cross-Level

Perspectives. Research in Organizational Behavior 7 1–37.

Rousseau, D. M. 2011. Reinforcing the Micro/Macro Bridge: Organizational Thinking and Pluralistic

Vehicles. Journal of Management 37(2) 429–442.

Sarker, S., and Valacich, J. S. 2010. An Alternative to Methodological Individualism: a Non-Reductionist

Approach to Studying Technology Adoption by Groups. MIS Quarterly 34(4) 779–808.

Seddon, P. B., Staples, S., Patnayakuni, R., and Bowtell, M. 1999. Dimensions of Information Systems

Success. Communications of the Association for Information Systems 2(20) 1–61.

Snijders, T., and Bosker, R. J. 1999. Multilevel Analysis: an Introduction to Basic and Advanced

Multilevel Modeling, Sage Publications Inc., Thousand Oaks, US.

Straub, D., Keil, M., and Brenner, W. 1997. Testing the Technology Acceptance Model Across Cultures: a

Three Country Study. Information & Management 33(1) 1–11.

Tallon, P. P., and Pinsonneault, A. 2011. Competing Perspectives on the Link Between Strategic

Information Technology Alignment and Organizational Agility: Insights From a Mediation Model. MIS

Quarterly 35(2) 463–486.

Venkatesh, V., and Davis, F. D. 2000. A Theoretical Extension of the Technology Acceptance Model:

Four Longitudinal Field Studies. Management Science 46(2) 186–204.

Weber, R. 2012. Evaluating and Developing Theories in the Information Systems Discipline. Journal of

the Association for Information Systems 13(1) 1–30.

Weick, K. E. 1989. Theory Construction as Disciplined Imagination. Academy of Management Review

14(4) 516–531.

A1

Appendix A. List of Concepts in a Multilevel Perspective

The notion of level is central in a multilevel perspective (Rousseau 1985). Miller (1978) defines levels as

qualitatively distinguishable entities of a system, such as cells, organs, individuals, and societies. Levels can

be conceptualized differentially for particular study purposes and they are often embedded in certain

larger, higher-order systems, such as individual levels, group levels, and organizational levels embedded

within a societal system (Kozlowski and Klein 2000).

Organizational researchers differentiate three “types” (as opposed to instances) of levels for methodological

consideration, including “the level of construct”, “the level of measurement”, and “the level of

statistical analysis” (e.g., Kozlowski and Klein 2000). Kozlowski and Klein (2000, p. 27) referred to

“the level of construct” as “the level at which it [the construct] is hypothesized to be manifest in a given

theoretical model.” Rousseau (1985, p. 4) defined the level of construct as “the focal unit to which

generalizations are made.” Instead of “level of construct”, some also prefer to use “level of theory” to

refer to the level at which the proposed theory is manifested (e.g., Klein et al. 1994). Given that theory can

include multiple constructs, the level of theory for a particular multilevel theory may point to more than

one level (Kozlowski and Klein 2000). Level of measurement and level of statistical analysis are related to

theory testing; they are not addressed in this paper.

As conventionally used in the IS and the management literature (e.g., Lapointe and Rivard 2005; Drazin et

al. 1999), level of analysis often refers to “level of construct” as used by multilevel theorists such as

Kozlowski and Klein (2000) and Rousseau (1985). Another term that deserves clarification is “unit of

analysis”. Some equate unit of analysis with level of analysis (e.g., Petter et al. 2008), whereas others use

“unit of analysis” in a slightly different sense, referring to the subject of a study (e.g., Yin 1994).

Some researchers differentiate theory and models, arguing that models are simplistic representations of

theory, whereas theory typically includes more substance and detailed explanations than models (e.g.,

Kozlowski and Klein 2000). Multilevel theory thus refers to theory that “entail(s) more than one level of

A2

conceptualization [of constructs] and [statistical] analysis” (Kozlowski and Klein 2000, p. 79). A necessary

condition for theory to be multilevel is that it includes theoretical constructs that manifest at multiple levels.

A multilevel model refers to specification of the relationships among the constructs in a given multilevel

theory (Kozlowski and Klein 2000). This definition is broader than Rousseau’s (1985) notion of “multilevel

model”, which refers more specifically to a type of model with functionally equivalent relationships among

constructs at multiple levels of analysis (Burton-Jones and Gallivan 2007; Kozlowski and Klein 2000). A

cluster of terms is used to differentiate types of multilevel models (e.g., Kozlowski and Klein 2000;

Rousseau 1985; Chan 1998; House et al. 1995; Klein et al. 1994). These types of models differ in terms of

criteria such as “whether there is a causal relationship across levels”.

Development of multilevel theory differs from multilevel analysis. Multilevel analysis refers more

narrowly to statistical analysis of hierarchical data for theory testing (Kozlowski and Klein 2000). Multilevel

analysis employs statistical techniques such as Hierarchical Linear Modeling (Hofmann and Gavin 1998).

References

Burton-Jones, A., and Gallivan, M. J. 2007. Toward a Deeper Understanding of System Usage in

Organizations: a Multilevel Perspective. MIS Quarterly 31(4) 657–679.

Chan, D. 1998. Functional Relations Among Constructs in the Same Content Domain at Different Levels

of Analysis: a Typology of Composition Models. Journal of Applied Psychology 83(2) 234–246.

Drazin, R., Glynn, M. A., and Kazanjian, R. K. 1999. Multilevel Theorizing About Creativity in

Organizations: a Sensemaking Perspective. Academy of Management Review 24(2) 286–307.

Hofmann, D. A., and Gavin, M. B. 1998. Centering Decisions in Hierarchical Linear Models: Implications

for Research in Organizations. Journal of Management 24(5) 623–641.

House, R., Rousseau, D. M., and Thomas-Hunt, M. 1995. The Meso Paradigm - a Framework for the

Integration of Micro and Macro Organizational-Behavior. Research in Organizational Behavior 17

71–114.

Klein, K. J., Dansereau, F., and Hall, R. J. 1994. Levels Issues in Theory Development, Data-Collection,

A3

and Analysis. Academy of Management Review 19(2) 195–229.

Kozlowski, S. W. J., and Klein, K. J. 2000. A Multilevel Approach to Theory and Research in

Organizations: Contextual, Temporal, and Emergent Processes. Multilevel Theory, Research, and

Methods in Organizations, K. J. Klein and S. W. J. Kozlowski (eds.), 3–90.

Lapointe, L., and Rivard, S. 2005. A Multilevel Model of Resistance to Information Technology

Implementation. MIS Quarterly 29(3) 461–491.

Miller, J. G. 1978. Living Systems, McGraw-Hill, New York.

Petter, S., DeLone, W., and McLean, E. 2008. Measuring Information Systems Success: Models,

Dimensions, Measures, and Interrelationships. European Journal of Information Systems 17(3) 236–

263.

Rousseau, D. M. 1985. Issues of Levels in Organizational Research: Multi-Level and Cross-Level

Perspectives. Research in Organizational Behavior 7 1–37.

Yin, R. K. 1994. Case Study Research: Design and Methods, Thousand Oaks, CA.

Documents

c Consult author(s) regarding copyright matters Notice ... Theorizing (accepted... · A Systematic Framework for Multilevel Theorizing in Information Systems Research ... research