Boundary Conditions

Conceptualization of Boundary Conditions

Whetten (1989) describes the function of BC as conditions that “place limitations on the propositions generated from a theoretical model. These temporal and contextual factors set the boundaries of generalizability, and as such constitute the range of the theory” (p. 492). Other authors also establish connections between BC, the generalizability of a theory, and context (e.g., Bacharach, 1989; Dubin, 1978; Edwards & Berry, 2010).

In his definition of BC, Whetten (1989) refers to a theoretical model, a term that he uses synonymously with theory, as Dubin (1978) also did. Therefore, it is important to consider the notion of theory first. The definition of theory has been debated throughout the history of the philosophy of science. Organizational research has also been concerned with its understanding of theory and the related questions of what constitutes a theoretical contribution and what makes a theory a good theory (e.g., Bacharach, 1989; Corley & Gioia, 2011; Davis, 1971; DiMaggio, 1995; Dubin, 1978; Merton, 1968; Sutton & Staw, 1995; Weick, 1989, 1995; Whetten, 1989). Our own perspective is best depicted by highlighting some elements of theory that are important for the purposes of this article. We assume that empirical reality exists objectively and is represented in theoretical models in a simplified manner. In our view, the purpose of theory is to describe, explain, and predict a part of this empirical reality; its predictions are amenable to empirical tests. Moreover, we conceive of theory as consisting of theoretical constructs (concepts, variables) and causal relationships (hypotheses, propositions) that describe the way in which the constructs interact.

Theories in the social sciences are not applicable irrespective of context. Context has been labeled an “amorphous concept capturing theory-relevant, surrounding phenomena or temporal conditions” (Bamberger, 2008, p. 839) and defined as “situational opportunities and constraints” (Johns, 2006, p. 386). ² Practically, context is multidimensional (Bamberger, 2008) and relates to various temporal and spatial conditions (Bacharach, 1989). ³ We can think of a dimension of context as a variable (or variable range) at the same level of analysis as the unit of analysis (e.g., when small- or medium-sized enterprises depict the context for firm-level research) but also as a higher-level variable in which the units of analysis are nested (e.g., when context is a country or industry in which firms operate). For the moment, we will concentrate on the first delineation to foster readability, followed later by the second. Moreover, to ease the discussion, we assume that the “context-delineating variable” is continuous (without loss of generality).

Now, let us begin developing the BC concept itself. In accordance with the understanding of theory as an empirical model that represents the real world in a simplified manner, a given theory can fit the empirical reality to differing degrees, measured by the deviation of the theory’s predictions from the real values. This accuracy does not exist in an absolute manner but depends on the empirical context. For example, while the resource-based view may accurately predict the behavior of firms operating in low- or moderate-velocity environments, it appears to break down when firms are operating within high-velocity markets (Eisenhardt & Martin, 2000). Similarly, while traditional operations management literature on the focused factory may explain the behavior of discrete-parts manufacturing firms accurately, it misrepresents the context of international process industries (Ketokivi & Jokinen, 2006). Scholars may however be able to adjust a theory structurally (i.e., by making changes and amendments to constructs and or causal relationships) such that the fit to high-velocity markets, international process industries, or any other meaningful delineation of context improves. For example, a moderating variable may be amended. Such structural changes influence the simplicity of a theory, as indicated by the plurality of concepts and relationships. Structural changes can foster the degree of cross-context applicability of a theory, which we refer to by the term generalizability.

Assuming that a theory is given, such that its structure in terms of concepts and relationships between the concepts is fixed, we can conceive of the accuracy of its theoretical predictions as a mathematical “function” that varies across contexts. We use the term function in the established sense of a relation between a set of inputs and exactly one output for each combination of inputs. Accordingly, the boundary conditions function of a given theory depicts the accuracy of theoretical predictions for any context. This is the (static) perspective on BC that we suggest to adopt in future research. Since BC, as defined here, assess the generalizability of a theory across contexts, our understanding is fully compatible with the aforementioned conceptions of Bacharach (1989) and Whetten (1989) yet more exact and explicit. Figures 1a, 1b, and 1c visualize examples of BC. Figure 1a depicts an example where the accuracy of theoretical predictions is high for low values of the context-delineating variable (e.g., accuracy is high for low-velocity markets) and gradually decreases toward high values of the context-delineating variable. Figure 1b considers a case where accuracy is high only for medium values of the context-delineating variable and decreases in both directions. Finally, Figure 1c portrays the opposite situation from the one visualized by Figure 1a.

OPEN IN VIEWER

Figure 1.

Static perspective on boundary conditions. Note: Solid lines denote (partial) certainty of the range; dashed lines denote (partial) uncertainty.

We would like to highlight that our conception of BC ties in very closely with the desirable properties of any theory, which are precisely the aforementioned features of accuracy, generalizability, and simplicity (e.g., Thorngate, 1976; Weick, 1999). Accuracy is reflected in the dependent variable inherent to the BC definition, generalizability is embedded in the cross-context perspective, and simplicity is determined by the given structure of the theory (which will later on also be regarded as variable). For this reason, we dare hope that the BC concept will henceforth lose some of its surrounding obscurity and will instead become a more integral part of theory development efforts.

To further foster conceptual clarity, it seems worthwhile to discuss the relation of BC to closely related terms. To this aim, Table 2 summarizes the definitions and descriptions of concepts used prominently throughout this study. Let us now consider the range (or domain) of a theory. Dubin (1978) defines the domain as “the territory over which we can make truth statements about the model and, therefore, about the values of the units composing the model” (pp. 134-135). We adapt this notion to our definition of BC by specifying it as all those contexts for which the accuracy of theoretical predictions is high. We would also like to distinguish between BC as defined previously and the boundary of a theory. We suggest that the boundary of a theory acts as a cutoff point (line, surface, etc.) that identifies the outer end of the range of a theory (depending on the degree to which multidimensionality of context is considered).

OPEN IN VIEWER

Table 2.

Boundary Conditions (BC) and Related Concepts.

Concept	Definition and Description	References
Theory; theoretical model (used synonymously)	Theory seeks to explain (predict) a part of the objectively existing empirical reality and is amenable to empirical tests. It represents the empirical reality in a simplified manner. Theory consists of theoretical concepts (constructs, variables) and interrelated causal relationships (hypotheses, propositions) that link these concepts to another and describe the way in which they interact.	Dubin (1978), Whetten (1989)
Context	Context refers to “situational opportunities and constraints” (Johns, 2006, p. 386); it is inherently multidimensional, comprising numerous space- and time-related factors. A single dimension of context can be regarded as a variable (range) or as a higher-level variable.	Johns (2006), Bacharach (1989), Bamberger (2008)
Boundary conditions of a theory	The boundary conditions function of a given theory depicts the accuracy of theoretical predictions for any context. As such, it describes the generalizability of a theory across contexts.	Authors’ own definition, based on Whetten (1989), Bacharach (1989), Thorngate (1976), Weick (1999)
Range of a theory; domain of a theory (used synonymously)	The range or domain of a theory denotes all those contexts for which the accuracy of theoretical predictions is high. Note that according to the uncertain view on BC, the range of a theory is fuzzy.	Adapted from Dubin (1978)
Boundary of a theory	The boundary of a theory is a cutoff line across contexts that identifies the outer end of the range of a theory. Note that according to the uncertain view on BC, the boundary of a theory is fuzzy.	Authors’ own definition
Range of a construct	The range of a construct denotes those contexts in which the construct can be applied with the same intended meaning.	Authors’ own definition, based on Bacharach (1989), Dubin (1978), Suddaby (2010)
Static perspective on BC	The static perspective on BC considers BC as a meta-theory pertaining to a theory such that the theory is given and the BC denote its cross-context generalizability, in the sense of providing an answer to the when question (Whetten, 1989). Currently, this perspective of BC as an outcome implicitly dominates BC-related research.	Authors’ own definition, based on Whetten (1989), Bacharach (1989), Dubin (1978)
Certain view on BC	The certain view on BC is a variant of the static perspective that considers BC for a given theory seemingly objectively from the perspective of an omniscient evaluator. This variant is helpful for illustration purposes, yet it fails to consider the fuzziness of the range and boundaries of a theory.	Authors’ own definition
Uncertain view on BC	The uncertain view on BC is a variant of the static perspective that considers BC for a given theory from a scholarly perspective, acknowledging the fuzziness of a theory’s range and boundary. This study shows that uncertain BC can jeopardize the validity of subsequent research when it falsely assumes a wider range of the respective theory than given, thereby calling for intensified exploration of BC.	Authors’ own definition
Dynamic perspective on BC	The dynamic perspective on BC directs attention to the process of exploring BC, which it considers as a research effort directed at making BC more certain. Moreover, it views the exploration of BC as intimately intertwined with the theory development process. It suggests that the exploration of BC can be regarded as an instrument (i.e., means) for further development of the respective theory (i.e., end).	Authors’ own definition

In his seminal analysis of BC, Bacharach (1989, p. 499) visualized how an entire theory, comprised of constructs and relationships, is constrained by its boundary, thereby raising the question if it makes sense to consider BC only with respect to an entire theory or also to the building blocks of a theory. Dubin (1978) favors the latter perspective: “The boundary-determining criteria of a model apply with equal force to the units employed and the laws of interaction among those units” (p. 126). Bacharach shares this view and further elaborates when he requests that constructs should have their “time and space parameters” (p. 502) specified. Specifically, to foster construct clarity, researchers should specify under which contextual conditions a proposed construct will or will not apply (Suddaby, 2010). Accordingly, it makes sense to think of boundary conditions of constructs. One of our later empirical examples further substantiates this view, also elaborating on the argument of Suddaby (2010) that the meaning of a construct can change from one context to another. Accordingly, we define the range of a construct as follows: The range of a construct denotes those contexts in which the construct can be applied with the same intended meaning. The boundary of the construct is then the outer edge of the contexts beyond which the meaning changes; the BC of a construct can then be conceived as a function that describes to which degree the meaning is still the same.

Similarly, when considering moderating influences, which is the most established form of exploring BC, we immediately see that moderators are analyzed for specific causal relationships even when an entire theory is not widely applicable across contexts. Hence, it is sensible to think of relationships as having their own BC. ⁴ A consideration of BC related to constructs and causal relationships is also practical as it facilitates viewing the BC of a theory as the result of the BC of its components. In other words, any construct or causal relationship may be bounded in terms of its cross-context applicability such that it narrows the BC of the theory as a whole. In this vein, Edwards and Berry (2010) state: “Contingencies built into theories increase precision by specifying when and how the relationships predicted by the theory should vary and whether the theory in toto is applicable in particular circumstances” (p. 676).

Importantly, the fact that the BC of an entire theory are shaped by BC of the building blocks of theory (i.e., constructs and causal relationships) does not mean that inference from the BC of the theory to the BC of at least one of the components is possible. For example, in empirical research, model fit statistics may indicate good fit for a rather complex model in one, rather narrow context but deteriorating fit when context is viewed more broadly. Scholars may not always be able to pin down the exact causes. Moreover, the BC of an entire theory are not simply the narrowest BC of its components. Rather, it is reasonable to assume that the lacking consideration of constructs and relationships may also limit the BC of an entire theory.

In the past, the BC concept has not been used unanimously. For example, some authors have used the term BC synonymously with limitations (e.g., McNatt & Judge, 2004), a concept that is highly misused, in our view. We define limitations as theoretical and methodological imperfections of a study that do not question the validity of the respective findings substantially, at least in the respective authors’ view. Examples of limitations are low response rates in survey research, convenience sampling in case study research, or lacking empirical realism because of the utilization of student samples in experimental research. The fact alone that BC relate to a theory (rather than a study) denotes that BC and limitations are not associated logically with each other. This means, for example, that coercing researchers who conducted a perfectly rigorous case study to lay out supposed “limitations” in terms of limited generalizability of the findings across contexts is inappropriate. Rather, limited generalizability is an inherent downside of applying the case method (e.g., Gibbert & Ruigrok, 2010) and does not denote a weakness of the study but an unexplored range of the theory thus generated.

Others have used the notion of BC as a synonym for contingencies or moderators (e.g., Mayer et al., 2009; Okhuysen & Waller, 2002). We would like to make a distinction again. Moderators are an important theoretical tool for exploring BC; however, they are not the only such tool, as this study shows. Therefore, they should not be regarded as a synonym for BC. Yet other authors have referred to BC by other terms such as scope conditions (e.g., DiMaggio, 1995; Suddaby, 2010). We hope that our effort at clarifying the concepts surrounding BC (see Table 2) will help scholars use unambiguous terminology in future research.

Static and Dynamic Perspectives of Boundary Conditions

We refer to the static perspective of BC when BC are considered a meta-theory such that the theory is “given” and the BC denote its cross-context generalizability. Currently, this perspective implicitly dominates BC-related research (e.g., Bacharach, 1989; Dubin, 1978; Whetten, 1989). In the following, we distinguish two variants of the static perspective, the certain view and the uncertain view. The certain view is depicted in Figures 1a, 1b, and 1c, where an omniscient evaluator would consider the BC for a given theory in a seemingly objective manner. We believe that this variant is helpful for illustration purposes. However, it fails to regard the narrowness of the researcher’s knowledge.

We may assume that any “reasonable” theory possesses a range different from zero, that is, it is accurate in at least some contexts, as depicted in Figures 1a, 1b, and 1c. At the same time, any theory is limited in its range (Dubin, 1978). It also seems plausible that most experienced scholars will be able to conduct thought experiments to identify some extreme contexts that would be clearly out of the range of their theories, meaning that they know or can easily find out about the lack of applicability of the theory (also see Gray & Cooper, 2010). For example, in their conceptual manuscript on dynamic capabilities, Eisenhardt and Martin (2000) argued on purely theoretical grounds that the resource-based view is not applicable in high-velocity markets. We conjecture that in between the contexts that are within the range and those that are outside of the range, scholars will usually face uncertainty regarding the applicability of their theories. If a researcher possessed an explanation for when a theory is still “somewhat applicable,” then this understanding could immediately be employed to adjust the structure of the theory such that it becomes more widely applicable. ⁵ In addition, “in any study, there are an infinite number of contextual parameters to consider” (Bamberger, 2008, p. 840). For example, with respect to firms as units of analysis, contextual boundaries may be created by firm size, firm internationalization, firm age, industry, or further “spatial” context delineations. Accordingly, context delineations must be regarded as multidimensional in nature, meaning that BC may be established for one contextual dimension but not another. There will generally be at least some contextual conditions that have not yet been investigated and are difficult to anticipate. Moreover, organizational reality is vibrant and dynamic, meaning that temporal contexts also change more or less continuously. In consequence, we must assume that no BC exist that are or can be fully known (Ketokivi & Jokinen, 2006). This also means that the range and boundary of any theory contain some degree of fuzziness.

The uncertain view on BC as a variant of the static perspective considers BC for a given theory from a scholarly perspective, acknowledging the fuzziness of a theory’s range and boundary. Figures 1d, 1e, and 1f illustrate examples of situations in which scholars “know” (in the sense that prior empirical tests have not led to the repudiation of a theory) about a minimum range of a theory, also know about some out-of-range contexts, but face uncertainty in between.

The uncertainty that is typically associated with BC as an outcome calls for inquiries into how BC can be investigated. Attention is hence redirected to the process of exploring BC. We can now define the process of exploring BC as an effort directed at making BC more certain. We refer to this procedural perspective as the dynamic perspective on BC. To the best of our knowledge, neither the aforementioned uncertain view nor the dynamic perspective on BC have been explicitly analyzed before. The remainder of this section focuses on the dynamic perspective. We begin by considering BC-exploring approaches.

BC-Exploring Approaches

As before, we assume an uncertain initial understanding of the BC associated with a theory (i.e., perceived certainty regarding some within-range and some out-of-range contexts and uncertainty in between), as exemplarily depicted in Figure 2. We distinguish four approaches that explore BC further, contingent on the different starting points A, B, C, and D in Figure 2. Scholars can apply an “inside-out exploration of BC” (see Figure 2, point A), an “outside-in exploration of BC” (see Figure 2, point B), an “exploration of BC under uncertainty” (see Figure 2, point C), and an (unplanned, emergent) “serendipitous BC exploration” (see Figure 2, point D). All of these approaches are further detailed and juxtaposed in Table 3.

OPEN IN VIEWER

Figure 2.

Dynamic perspective on boundary conditions (BC) with BC-exploring approaches.

OPEN IN VIEWER

Table 3.

Juxtaposing Boundary conditions (BC)–Exploring Approaches Contingent on Their Starting Points.

Approach	Inside-Out Exploration of BC	Outside-In Exploration of BC	Exploration of BC Under Uncertainty	Serendipitous BC Explorationa
Starting point (see Figure 2)	A	B	C	D
Perspective	Theoretical	Theoretical	Contextual	Unknown (yet atheoretical insofar as the possibly limited range of the respective theory is not noticed)
Role of the theory	Primary (main focus of attention)	Primary (main focus of attention) or auxiliary (trigger to theory development)	Auxiliary (borrowing of theory is considered consciously)	Auxiliary (borrowing of theory occurs unconsciously)
Goal	Exploration of BC in the sense of developing an answer to the when question (Whetten, 1989)	Exploration of BC in the sense of developing an answer to the when question (Whetten, 1989) or theory development for context B	Deep understanding of context (e.g., development of middle-range theory pertaining to this context)	Unknown (yet by definition not BC-related)
Knowledge about theory applicability in the respective context	Known known for point A, known unknown in the contexts to be explored	Known known for point B, known unknown in the contexts to be explored	Known unknown in the vicinity of point C (which is the area to be explored)	Unknown unknown (no consideration of range and boundary of theory)
Treatment of prior literature related to the theory under scrutiny	Reflective (i.e., it is applicable in A and presumably also in its vicinity)	Reflective (i.e., it is not applicable in B and presumably not in its vicinity)	Reflective trial-and-error consideration (i.e., it could be applicable or not)	Initially: naïve trust; later: reflective
Importance of BC exploration	Identical with reaching the goal	Identical with reaching the goal or a cherished side effect	A necessary task for reaching the goal	No awareness to or interest in BC exploration
Trigger to BC exploration	Conscious decision or perplexity	Conscious decision: “pursuing failure” (Gray & Cooper, 2010)	Necessity for reaching the goal	Need to move out of dead end
Research process	Tame problem (Rittel & Webber, 1973)	Somewhat tame problem (Rittel & Webber, 1973)	Somewhat wicked problem (Rittel & Webber, 1973)	Wicked problem (Rittel & Webber, 1973)
Outcome	More certain BC in the vicinity of context A	More certain BC in the vicinity of context B and/or development of theory for context B	Better understanding of context C and more certain BC in the vicinity of context C	Valid results for the aspired goal plus more certain BC in the vicinity of the respective context

^aThe column portrays the pathway that leads from the starting point to BC exploration; other paths are described in the text body.

The inside-out exploration approach considers BC from a theoretical perspective concerned with the further development of the respective theory. Its goal is to explore BC as the “final phase” of theory development, in the sense of developing an answer to the when question (Whetten, 1989). Its denomination stems from the perspective of scholars who investigate the further applicability of a theory in slightly different contextual conditions, departing from “known territory” (i.e., knowing that context A is still within the range of the theory) outward to unknown territory. Thereby, scholars who use the inside-out exploration are aware of the fact that the unknown territory is unknown (it is hence a “known unknown”). The exploration of BC can occur as a conscious decision or can be triggered by perplexity. Vis-à-vis the other three BC-exploring approaches, the research process can be characterized as a so-called tame problem (Rittel & Webber, 1973). ⁶ This means that there is an explicit problem statement (i.e., willingness to further explore BC) preceding the solution of the problem and that the research process can be managed and organized as a project with clearly defined phases (e.g., as a large-scale empirical investigation) and outcomes (i.e., a better understanding of BC in the contexts adjacent to point A in Figure 2). For example, Kopelman and Thompson (1976) purposefully sought to strengthen the cross-context applicability of expectancy theory by exploring five moderator variables. Among the articles that explicitly refer to BC, the inside-out exploration approach seems to be the most widely applied (see Table 1).

The outside-in approach addresses BC exploration in the opposite manner to its conceptual sibling (the inside-out approach) by first focusing on a context in which the theory is clearly not applicable (see point B in Figure 2). The perspective of the outside-in approach is also theoretical in nature, and its goal is also to further explore BC. However, often another goal exists that is the development of new theory specific to context B (and its contextual vicinity) (see the examples in Table 1). Scholars who pursue this approach are aware of the lacking applicability of the theory in the context to be explored. Therefore, they will refrain from falsely assuming applicability of the related literature in their own context. For example, Ketokivi and Jokinen (2006) sought to understand why some factories opted to be unfocused in their production processes even when the extant literature suggested that focused manufacturing strategies are most effective. The trigger mechanism commencing the exploration of BC through this approach is best portrayed as “pursuing failure” (Gray & Cooper, 2010), that is, a conscious decision to learn from the misfit between the theory and the context of interest. The research process can be characterized as a somewhat tame problem. It is more wicked (Rittel & Webber, 1973) than the inside-out approach because new theory must be created, meaning that the problem statement is less clear and the research project does not follow a fully straightforward plan but more likely involves some feedback loops and iterations. The outside-in approach is used fairly often and has facilitated some very impactful contributions (see Table 1).

The other two approaches are fundamentally different in that they are not directly concerned with exploring the BC of the respective theory. Rather, they follow different goals and borrow from the respective theory (consciously or unconsciously). Theory is borrowed frequently in organizational research. For example, constructs are adopted (Suddaby, 2010), and specific findings of a theory are incorporated into the theorizing process elsewhere (Corley & Gioia, 2011). Our BC analysis highlights that such theory borrowing is only valid if the context to which the theory is borrowed falls into its range. Contrarily, if the context falls outside the range of the theory, borrowing it would not be valid. Given the frequent uncertainty surrounding a theory’s BC, scholars who consider borrowing theory often have to deal with unexplored BC. It appears that this issue has become more pressing over the past decades because of an increased emphasis on middle-range (i.e., relatively more context-specific ⁷ ) theorizing (Corley & Gioia, 2011; Merton, 1968; Pinder & Moore, 1979). Since we are unaware of any methodological research addressing this widespread problem, we will analyze it in considerable detail, also drawing on two empirical illustrations. The approach labeled exploration of BC under uncertainty concerns theory borrowing when scholars are aware of the unclear range of the theory that they consider borrowing (i.e., it is a known unknown), whereas the approach denominated serendipitous BC exploration refers to cases where theory is borrowed without considering the BC issue (i.e., it is an unknown unknown).

Regarding the exploration of BC under uncertainty, we assume that scholars adopt a “contextual” perspective in that they have a particular interest in an empirical context (see point C in Figure 2). The goal is to establish some form of deep understanding of the respective context (e.g., by means of developing middle-range theory pertaining to this context). The role of theory with uncertain BC is auxiliary only, that is, scholars want to borrow it but are uncertain if doing so is valid. We may assume that the exploration of its BC is a necessary task for reaching the goal; if it were different, the respective theory would presumably be dropped. As our later empirical illustration will show, this type of setup depicts a somewhat wicked research problem. The reason is that reliance on the respective literature seems necessary, yet it can never be fully trusted. Literature pertaining to the theory can only be considered in a trial-and-error mode. If all goes well, this approach results in a better understanding of context C and of the theory’s BC in the vicinity of this context. One of the empirical examples will illustrate this approach.

The final BC-exploring “approach” is serendipitous BC exploration. Here, we refer to cases in which scholars borrow theory with uncertain BC without considering the fact that their own context may not fall into the theory’s range; thereby, they may end up exploring BC more or less by chance. In other words, no initial positioning of the respective context vis-à-vis the theory occurs (as a visual illustration, the respective point D in Figure 2 is not positioned within the chart but outside of it). Initially, literature sources on the respective theory are hence relied on naïvely such that its findings are assumed to be true irrespective of context. If the respective context lies within the range of the theory (from an omniscient evaluator’s certain view on BC), then the scholars’ unawareness to BC-related issues will not have any impact, leading to valid research outcomes (whatever they are) in absence of other methodological issues. The BC of the borrowed theory will hence not be explored. If the respective context, however, lies outside of the theory’s range, then two things can happen: First, the investigation may end up in a dead end. Then, scholars will presumably “turn the entire research process upside down,” looking for anything that may have gone wrong, thereby possibly directing their attention to BC-related issues. If so, then they will explore the respective BC out of necessity (same as in the previous approach). Needless to say, such a research process depicts a wicked problem. The outcome would then be a valid result for the originally aspired goal plus a clearer understanding of the BC in the vicinity of the respective context. Second, scholars may never notice the inherent flaw stemming from the fact that they built on theory in an invalid (i.e., out-of-range) manner. They would hence pursue as planned throughout the research process (which would falsely be treated as a tame problem) and reach a result that is not (fully) valid. One of the empirical examples will illustrate a situation in which we, thankfully, ended up in a dead end, thereby triggering the serendipitous exploration of BC of a construct that we employed.

In the previous paragraph, we used the word approach with quotation marks when referring to serendipitous BC exploration. The reason is that this “approach” only happens to explore BC as an effect (if all goes well) but never aims for BC exploration. Scholars should consider the BC of, ideally, all theory that they borrow. Hence, methodologists could be tempted to dismiss unawareness of the BC of borrowed theory on prescriptive grounds. However, since virtually no construct, causal relationship, or entire theory has its BC clearly established (as per the uncertain view), we conjecture that unawareness of unclear BC will often just happen in research practice as a descriptive fact. Moreover, given the multidimensionality of context and BC, it seems to us that no strategy can exist that avoids the unawareness issue fully and certainly. By considering and illustrating this “approach,” we aim to (a) sensitize researchers to at least consider the BC of the theory that they build on ex ante (thereby trying to avoid position D in Figure 2) and (b) suggest the ex post consideration of BC as a “toolkit” for situations in which a rigorous investigation has suddenly ended up in a dead end.

The Role of Moderators and Other Theoretical Tools for Studying Boundary Conditions

Previous research that explicitly discusses theoretical “tools” for the exploration of BC is scarce. The few noteworthy exceptions that we are aware of equate BC with moderators (contingencies) (e.g., Edwards & Berry, 2010; Goldsby et al., 2013; Gray & Cooper, 2010). Scholars widely agree that moderator analysis is methodologically challenging (e.g., Aguinis, Sturman, & Pierce, 2008; Cortina, 2003; Gwowen, 2009). For the purpose of our analysis, a simple example suffices. Accordingly, we consider the influence of a single contextual moderator on a single causal relationship. Such a moderator can impact the “intercept and slope of the function that describes the relationship between” independent and dependent variable (Edwards & Berry, 2010, p. 676) and make a linear function become curvilinear or vice versa (Busse, Mahlendorf, & Bode, 2016).

Moderators can be identified by means of various methodologies, such as conceptual theorizing, primary empirical studies, as well as meta-analyses. First, the conceptual theorizing strategy revolves around “researchers thinking about the conditions that favor and disfavor a theory” (Gray & Cooper, 2010, p. 628), anticipating contexts under which the accuracy of theoretical predictions is relatively high or low. This approach feeds right into the second strategy, primary empirical research, which appears to be the most acknowledged means for establishing BC (Edwards & Berry, 2010; Goldsby et al., 2013; Gray & Cooper, 2010). Many empirical examples apply confirmatory approaches that develop hypotheses pertaining to the moderating effects of certain contextual variables. For example, both Dietz et al. (2004) and Gulati, Lavie, and Singh (2009) (see also Table 1) proceed in this manner. Third, meta-analyses are very potent tools that help deduce the presence of moderators from variance in effect sizes across studies (Edwards & Berry, 2010). Excellent descriptions of and prescriptions for the conduct of meta-analyses have been described elsewhere (e.g., Cortina, 2003; January 2008 issue of Organizational Research Methods). Importantly, the other two tools that we want to suggest are also not linked to any particular methodology.

We argue that the refinement of constructs and the amendment of mediators are “methodologically equivalent” tools to the amendment of moderators. We use the notion of methodological equivalence not in a statistical-methodological sense but more broadly to encompass the entire theorizing process. While theorizing is (and must be) largely driven by creativity and abductive reasoning, an underlying idea of this research is that “theorizing is inextricably tied to methodology” (Delbridge & Fiss, 2013, p. 325) and that “theory and method inevitably interact” (Van Maanen, Sørensen, & Mitchell, 2007, p. 1145). This view suggests that theorizing is a research process that is prone to methodological analyses and (at least somewhat) susceptive to methodological prescriptions.

To be clear, from the static perspective on BC, only moderators depict extant cross-context differences quantitatively. However, it is the dynamic perspective on BC that relates to the theorizing process. Therefore, we argue that similarity or dissimilarity of theoretical tools for exploring BC must not be assessed against the structural outcomes but against the effects that manifest over the theorizing process.

Edwards and Berry (2010, p. 676) demonstrate mathematically how the integration of a moderator variable increases the accuracy of theoretical predictions across contexts. At the same time, it fosters the generalizability of a theory (i.e., its range for a required level of accuracy is widened), yet it also decreases its simplicity (i.e., one additional variable for each moderator is amended to the theoretical model). Thereby, an intimate relation between the theory development process and the exploration of BC is highlighted because not only are BC explored (i.e., they become more certain than before the consideration of the moderator), but also the theory itself changes (in what we have called its structure). The two illustrative case examples resonate with this observation, thereby indicating that the process of exploring BC and the further development of the respective theory cannot fully be separated.

Recall that we view the BC of a construct as the degree to which the construct can be applied across contexts with the same meaning. Scholars who aim to explore BC may hence refine constructs to clarify their meaning across contexts so as to make the BC more certain. Of course, BC exploration is not the only reason for refining constructs; this also happens for statistical or purely theoretical reasons. We conjecture that BC-related refining of constructs will often entail splitting one rather general into multiple more specific constructs. Because of the increase in specificity of the constructs’ meanings, each of the resulting constructs can subsequently be measured more precisely, thereby fostering the accuracy of the respective theory and its generalizability across contexts. However, an increasing number of constructs means a loss in simplicity of the theory. In a nutshell, the effects of refining constructs from the dynamic perspective on BC can hence be identical with the ones of the established approach, amending moderators (see Figure 3). The first illustrative case will substantiate this evaluation further. To be clear, we do not claim that the aforementioned effects are necessary consequences of exploring the BC of a construct. For our purpose of widening the perspective on the “armory” of theoretical tools for exploring BC, it suffices that these effects depict possible and somewhat probable consequences.

OPEN IN VIEWER

Figure 3.

Theoretical tools for establishing boundary conditions (BC). Note: A simple X→Y relationship was chosen as initial theory, for the sake of simplicity.

The amendment of mediators reflects a third tool for exploring BC because it is “mediating mechanisms through which … situational or temporal” context factors “may exert their influence” (Bamberger, 2008, p. 843). In accordance with this observation, the effectiveness of the mediator tool is derived from its facilitation of theoretical insights (the “Verstehen”). The second illustrative case example will show how, in an exploration of BC under uncertainty, the uncertainty surrounding the BC of a theory was caused by its structural “coarseness.” Amending mediators meant “zooming into” the previously coarse causal relationship, thereby facilitating a much better understanding of its fit to the context of interest; the BC became more certain by amending the mediators. Because of the distinction of different causal pathways in the theoretical model, this tool makes a theory more accurate and generalizable. The associated downside is a loss in simplicity. From the dynamic perspective on BC, this mechanism is hence also methodologically equivalent to the previous two (see Figure 3).

Initiation of the Investigation

The initial purpose of the investigation was to understand the influence of hostile environments on managerial decision-making processes. At the forefront of our study, we observed that the understanding of how managers react to hostile environments when making decisions to invest in firm innovativeness was not well developed throughout the extant literature. Moreover, previous literature has purported that firms encountering hostile environments experience negative impacts on many operational functions, impeding their ability to compete and survive. However, evidence exists that not all firms fail; some are able to survive these conditions (Potter, 1994). Therefore, our research was driven by the curiosity of how some firms are able to navigate and maneuver through hostile environments and subsequently survive.

Research Design and Process

Our research process underwent four phases: initial theorizing, interaction with practitioners, construct refinement and further theorizing, and a confirmatory empirical study. As this sequence highlights, there was a setback that prevented us from proceeding directly from the initial theorizing to the confirmatory empirical study. In fact, both the second and the third phases emerged out of sheer necessity. The research process evolved in a wicked manner (see Table 3).

In the initial theorizing phase, we sought to fully understand the concept of environmental hostility and hypothesize its effects on our dependent variables. We screened the hostility literature over the past 40 years. A total of 118 manuscripts in operations management, strategic management, general management, and other related business fields were examined. Per this investigation, we found that much of the literature is anchored on the seminal works of Khandwalla (1972, 1977), defining environmental hostility as a condition of perceived threat to the organization’s primary goals and suggesting for hostility to be on a continuum with malevolent settings on one end and munificent settings on the other. Khandwalla (1977) operationalized and measured hostility based on the riskiness of the competitive environment, difficulty of staying afloat, and extent to which firms control their own destinies. However, it was perplexing to see that studies incorporated various understandings of and emphasized different aspects of hostility. For instance, whereas several studies place emphasis on external competitive pressures as depicting hostility (e.g., Lofsten & Lindelof, 2005; Nandakumar, Ghobadian, & O’Regan, 2010), others highlight resource scarcity (e.g., Luo, 1999; Stonebraker & Liao, 2006). Yet other studies blend several factors together (e.g., Baum & Wally, 2003; Nicholls-Nixon, Cooper, & Woo, 2000). Noting these different adaptations of the concept of environmental hostility, we began to wonder if perhaps there was something awkward in the way literature was approaching the construct.

Literature has overwhelmingly claimed that hostility impacts firm performance and survival negatively. Its effects on firm capabilities such as innovativeness were, however, not clear and at times contradictory across prior empirical investigations. The effects of hostility had been studied across an array of functional areas (e.g., operations, entrepreneurship, and strategy), providing varying results contingent on the audience that was surveyed. At the same time, we encountered immense difficulty in hypothesizing the expected influence of hostility on firm innovativeness. Each time the issue was considered, different arguments were salient to us, calling for positive influence on one and negative influence on another occasion. We felt stuck at this point in our investigation.

Upon continuous reexamination of the 118 articles, we discovered that the definitions of hostility depicted various combinations of four subdimensions: market decline, restrictiveness, competition, and resource scarcity; yet, operationalization of the construct occurred in a holistic manner. This led us to wonder if perhaps hostility should be disaggregated into the four aforementioned components instead of being viewed, operationalized, and measured in a holistic manner. We decided to discuss the concept with managers operating in different industry and functional contexts as a next step before continuing with our theoretical stream of research.

In the subsequent but partly overlapping phase, we approached managers and inquired what they believed “environmental hostility” entailed. Upon interviewing a panel of 20 managers from different industries, we found that not a single person had heard the expression “environmental hostility” (or hostile environments) before or could offer an explanation of what it meant. Yet once we provided the traditional definition of hostility from the extant management literature, managers’ responses began reflecting the four subdimensions that we suspected to have identified. One manager mentioned that his business was facing high levels of competition and having trouble staying afloat because of intense regulations within the industry. Another manager detailed a lack of resources that the firm was facing. Yet another discussed a downturn in markets, scarce resources, and competition as all having a severe impact on the firm. What we discovered was that all managers were facing some environmental hostility, but each manager had to deal with different combinations of hostile environments and varying severities for each form. Therefore, each manager associated the overall notion of “environmental hostility” with those hostile characteristics that were most salient to them.

This led us to conclude that across different industry contexts, the meaning of the holistic hostility construct changes. Undetected by 40 years of prior research, the range of the construct had been too narrow to allow general application across industries. Taking a holistic approach toward understanding and operationalizing environmental hostility could lead to the hazard of overlooking these different forms, which in turn could jeopardize the validity of our theoretical model. At this point, we decided that a new, author-developed scale depicting and operationalizing hostility along four subdimensions would be the best approach for understanding the different influences. We also chose to include the traditional, holistic hostility scale for purposes of comparing the results from the old and the new scale.

In the next phase, we began constructing a new scale to capture the effects of hostile environments, containing questions reflecting the four aforementioned subdimensions of hostility. The questions were developed by observing and triangulating the information provided to us by practitioners during our previous encounter. The survey contained five questions per hostility form in order to have enough information to test the homogeneity of items within each latent construct (Harvey, Billings, & Nilan, 1985). Using a Likert-type scale, respondents were asked to indicate the level of hostility that their organization is facing during the present time period. We followed all suggestions of Schwab (2004) for increasing face and content validity.

Subsequently, we drew on contingency theory and the prior literature on hostile environments to hypothesize the impacts that each of the four forms of hostile environments has on innovativeness, thereby also distinguishing between product innovativeness and process innovativeness. For example, we posited that competition would have a positive impact on product innovativeness, while contrarily, resource scarcity would have a negative impact. There were multiple occasions where we (correctly) predicted opposed influences, which endorses our assessment that “environmental hostility” is not a unidimensional construct and confirms that indeed, we could not have deduced any meaningful hypotheses at the outset of this investigation.

In the final phase, we sent out our survey and collected data surrounding the four forms of environmental hostility, the two forms of innovativeness, and firm performance. Following a similar approach to previous hostility studies conducted in the same arena, our target audience was U.S.-based operations managers and directors working directly with the production process. We focused on small- and medium-sized firms and collected the data across eight different industries. We tried as much as possible to model our new data collection approach after what had been previously conducted throughout the extant literature. That way, any differences detected could not be readily attributed to differences in sampling procedures.

Results

The large-scale statistical data confirmed our preliminary findings that had indicated that hostility pressures are not homogeneous across industries. For example, when compared to other industries, aeronautical and food/beverage firms reported on average higher exposure to restrictiveness whereas construction firms portrayed on average greater exposure to severe market decline. We analyzed the data using structural equation modeling. The results indicated that each of the four hostility forms impacts firm performance negatively but with varying strengths and significance levels. Moreover, we discovered that the four forms of hostile environments had partially opposed direct effects on product and process innovativeness, ranging as expected from positive to negative. We tested not only the aforementioned model but also numerous alternatives, most importantly one containing the traditional holistic hostility scale. This model returned nonsignificant results.

The results confirmed that the disaggregation of the environmental hostility construct was indeed necessary for studying the impacts of hostile environments on innovativeness. The traditional construct was associated with an overly narrow range that prevented it from being applicable across different contexts because managers interpreted the scale relative to the hostility form(s) that their firm was facing, thereby leading respondents to attach meaning to the traditional concept that best matched their respective context. The dispersion of the extant literature shows that the BC of the construct were initially very uncertain. In retrospect, the concept must be viewed as too unspecific to be employed in any context, precisely because it can be infused with various meanings. The illustrative case explored the BC of the construct and made them more certain by identifying this shortcoming. By having replaced the hostile environments construct with the four constructs market decline, restrictiveness, competition, and resource scarcity, our theoretical model has become more widely applicable and more accurate yet less simple.

Initiation of the Investigation

Practical interest in the management of LSPs triggered our investigation. At the outset, managers of shipping organizations, which are the customers of LSPs, had pointed out to us that on average, LSPs are hardly innovative. LSP-related research later confirmed that this capability is frequently lacking (Wagner, 2008). The reasons were, however, not apparent. Perplexingly, the literature also provided evidence that LSPs operate in innovativeness-fostering environments and that they obtain a number of benefits from being innovative. Therefore, an explanation of LSPs’ low innovativeness appeared very useful for corporate practice since it would likely allow the practitioners to derive measures for fostering their innovativeness.

We did not look at LSPs’ innovativeness from the lens of a grand theory but rather used accumulated empirical evidence (Shrivastava, 1987) on the innovativeness-survival relationship as our theoretical basis. Previous research has shown that firms’ ability to compete in the marketplace depends on innovativeness (Damanpour & Gopalakrishnan, 2001) and that indeed, research and development activities as an input (Esteve Pérez, Sanchis Llopis, & Sanchis Llopis, 2004) and innovation as an output (Cefis & Marsili, 2006; Damanpour & Gopalakrishnan, 2001) facilitate survival. Correspondingly, “few issues have been characterized by as much agreement among organizational researchers as the importance of innovation to organizational competitiveness and effectiveness” (Wolfe, 1994, p. 405). At the outset of this investigation, we thus derived a working hypothesis from the theory on the innovativeness-survival link according to which firms that strive intensively for innovativeness benefit from improved abilities to develop new products, improve processes, or adapt to environmental necessities such that ultimately, their chances of survival increase. Therefore, according to prior theory, we would expect LSPs to regard innovativeness as key to their survival and strive intensively for innovation. However, counter-stream to this working hypothesis and the aforementioned evidence, many LSPs survive without any mentionable innovativeness (Wagner, 2008).

We were hence facing uncertainty with respect to the question if the prior theory holds in our context or not (see Table 3). Because of this misalignment between the predictions of the theory and the situation in our context of interest, we suspected that industry (i.e., belonging to the LSP industry or not) might have to be added as a moderator to the relation between striving for innovativeness and survival (see Figure 4). However, no reasons were apparent that could explain this possible moderation effect.

Research Design and Process

As the theoretical literature on the innovativeness-survival relationship did not yet depict or explain any industry moderation effects, it stood in apparent conflict with our prior knowledge on the LSP context. Therefore, we were worried that the range of this theory might not stretch far enough to include our context for reasons that we did not understand. Due to the seemingly special nature of the LSP context, we were generally very cautious about the extent to which our research was informed by prior literature throughout this investigation. Therefore, it seemed logical to apply a highly exploratory research design.

Our investigation was motivated by an expectation that (a) the LSP context surrounding innovation would be special and (b) there would be differences between LSPs and other service providers. These two factors are closely intertwined but point in slightly different directions, with the former calling for open-minded and unbiased exploration of LSPs’ innovativeness context, whereas the latter suggests for cross-context comparisons. These factors also required different data to be analyzed, with the former needing qualitative factors characterizing the LSP context and the latter requiring quantitative comparisons. Over the course of the investigation, we reviewed the related context-specific and context-unspecific literature, analyzed the distinctiveness of LSPs and logistical services vis-à-vis other service providers and other services, studied innovation management at the levels of the firm and the innovation process, and computed and compared various input-, output-, and outcome-related innovativeness indicators. When our research question was finally answered to our full satisfaction, we had conducted four distinct studies that built on each other in a stepwise fashion to facilitate a comprehensive explanation. The empirical database for each study was compiled in Germany; therefore, the legal and cultural context could not confound the results. The research process was rather wicked.

We began first by exploring LSPs’ context by means of a highly interactive workshop series in which we discussed the applicability of extant innovation management concepts at “typical” LSPs with seven managers from four logistics service providers. Over the course of the workshops, the practitioners increasingly directed attention to the processes of innovation generation at LSPs. All practitioners agreed that overall, LSPs are only marginally innovative. Among the participants, we discussed seven extant, supposedly context-independent innovation process models that we refined to match the LSP context. From the practitioners’ assessments of applicability at their firms, three LSP-related variables emerged (high customer dependence, high dispersion, adverse innovation mentality of personnel), which needed to be integrated into a refined, LSP-specific model of how innovation generation processes evolve. These variables had obstructing rather than facilitating effects within innovation processes, which later led us to conclude that innovation management within LSPs tends to be particularly costly.

Having found some support for our ex ante expectation of a specific context at LSPs, we began a literature review on innovation management at LSPs that employed content analysis. Our intention was to better understand the peculiarities of LSPs’ firm- and process-level innovation management. Upon investigation, contextual boundaries were discovered to be slightly blurry since, for example, logistics departments within industrial firms may occasionally act so independently that they can be regarded as “internal LSPs.” After careful consideration of its applicability, we found it possible to categorize the data with a useful framework from the general innovation management literature (Adams, Bessant, & Phelps, 2006). The categorized prior literature helped us understand some characteristic features of LSPs’ innovation context, such as an inherent difficulty to innovate the most basic elements of logistical services (i.e., acts of transportation, handling of goods, and warehousing), but it did not fully explain the differences between LSPs and other service providers. Thus, further studies were required.

Next, we conducted an in-depth study of 13 cases of LSPs’ firm-level innovation management, aiming to establish a link to each firm’s respective internal and external context. Six context factors were identified. Three of them reflected and hence confirmed the factors from our initial study, another three were not identified therein (organizational size, importance of technology, and growth potential). With the help of some data from a statistical yearbook, it was possible to conclude that those additional factors were relevant at the level of the individual firm only. This means that there was some variance at the level of the individual firms that does not have an average or typical value that characterizes the context of LSPs as a whole.

At this point, we concluded that a view focused on LSPs, only with relatively indirect cross-context comparisons based on implicit understandings of other contexts, would not be sufficient to fully answer the research question. Therefore, another study was amended that compared input-, output-, and outcome-related innovativeness indicators of LSPs and other service providers, using a high-quality secondary database. Significant and substantially meaningful differences between LSPs and other service providers were identified and explored. The central conclusion was that for LSPs, innovation efforts are as beneficial as for other service providers but more costly to conduct. This finding triggered the refinement of the initially suspected moderation model into a moderated mediation model, as depicted in Figure 4.

Jointly, the results of the four, internally consistent studies facilitated deriving a holistic explanation of the lack of innovativeness experienced by LSPs compared to other service providers. The holistic view allowed us to pin down exactly what was special about the LSP context and why.

Results

Prior theory on the innovativeness-survival link strongly emphasized benefits associated with innovativeness, whereas it spoke hardly about costs. Our initial working hypothesis derived from this literature was hence that firms should choose suitable antecedents (A) (e.g., deliberate investments into innovativeness as an organizational capability) so as to enjoy the benefits (B) associated with innovativeness (e.g., the abilities to construct innovative products and implement efficient processes) in order to ultimately foster survival (S) (i.e., A → B → S). The seeming contradiction that many LSPs survive without striving for or demonstrating innovativeness (S_LSPs = high, B_LSPs = low, A_LSPs = low) motivated us to explore the LSP context to innovation in depth and suspect that industry might be an important moderator to the aforementioned causal chain. The exploration of the LSP context led us to conclude that costs also matter as an additional mediator between innovativeness-related efforts and survival (i.e., A → C → S). Mostly by serendipity, the structure of our secondary database allowed us to posit that the benefits for LSPs and other service providers are similar (i.e., the A → B relation features a similar “effect strength”), whereas innovation efforts are inherently more costly for LSPs than for other service providers (i.e., the A → C relationship features a “stronger effect” for LSPs) because of certain characteristics of logistical services, LSPs, and the environmental conditions under which LSPs operate. If we look at firm survival in terms of “not going bankrupt,” it is immediately clear that costs affect survival negatively, whereas the benefits associated with innovation influence it positively. Therefore, the overall model that we derived can, in abstract terms, be regarded as a competitive mediation model (Zhao, Lynch, & Chen, 2010), which is also a moderated mediation model (Hayes, 2013) because the A → C relationship is moderated by the industry context.

This explanation may sound simple, but that is primarily because it is an ex post explanation. Bear in mind that the range of the prior theory on the innovativeness-survival link was unclear. In other words, it was unclear in which contexts the implication “if you want to survive, then you should aim to be innovative” holds. There are many industry contexts in which innovativeness is indeed so obviously necessary (e.g., high-tech firms or Internet firms) that the associated benefits presumably overcompensate for the costs. In these contexts, the theory as a whole can be viewed in a simplified manner as A → S, with an overall “positive effect” of A for each context within the range. However, this “coarse” theory becomes an oversimplification outside of its range where both of the mediation paths need to be viewed and balanced against each other (Busse et al., 2016). In this manner, the investigation clarified the BC of the initial theory and developed the theory further. These two processes were even so intertwined that it is difficult to disaggregate them afterward. The refined model, which is depicted in Figure 4, is more generalizable and more accurate than the original model yet less simple, as the theoretical-methodological analysis suggested (see Figure 3).

In line with our related statement in the theoretical-methodological analysis section, it was not the amendment of the moderator to the structure of the theory that made the difference in the BC exploration. The necessity to amend this moderator seemed apparent from the outset of this investigation so as to maintain fit with the innovativeness-survival theory, but it was impossible to make sense of (i.e., theorize on) this “required” moderation effect. Rather, it was the amendment of the competing mediating pathways between the independent and the dependent variable that allowed us to develop a theoretical rationale for the influence of the industry context, as per the earlier analysis of this theoretical tool.

Critical Evaluation of Exploring Boundary Conditions

We argue that BC deserve more attention from methodological scholars and applied researchers than they receive to date. Our first argument for exploring BC relates to the interdependence between BC exploration and theory development that this study has identified. Both illustrative cases have highlighted how by exploring the BC associated with a theory, this theory can be developed further. In the hostile environments case, the assessment of the cross-context applicability (i.e., the BC) of the extant “environmental hostility” construct led us to differentiate between the four forms of hostile environments. This enabled us to develop a better measurement instrument and in turn hypothesize the partly opposed effects that the four forms of hostile environments have on product and process innovativeness. Similarly, in the LSP case, the uncertain BC surrounding the innovativeness-survival relationship triggered our in-depth exploration of the LSP context and the comparison with the contexts of other service providers. Consequentially, we noticed the relatively high costs associated with innovativeness for LSPs, which in turn allowed us to zoom into the prior theory and differentiate between two competing mediation paths (the benefits path and the cost path) between innovativeness-related efforts and survival. This perspective substantially elevates the role of BC from a mere amendment to theory, as implicitly portrayed to date, to an integral part of the theorizing process. The study has also identified that BC exploration efforts typically increase generalizability and improve accuracy but reduce simplicity of the respective theory (see Figure 3).

Second, we argue that BC exploration fosters validity of organizational research, a primary methodological goal. Recall our illustration of serendipitous BC exploration. The basic problem at the outset of this “approach” is that scholars are unaware of the narrow or uncertain ranges of the theory that they borrow (see Table 3). At present, organizational research produces rather narrow theory while scholars are incentivized to claim wide ranges of the theories they develop, which makes BC-related validity problems very likely.

Our observation related to the widespread development of narrow-in-range theory stems from a shift that has happened in organizational research over the past few decades. The field has moved away from the development of grand theory and toward middle-range theory (Corley & Gioia, 2011). Middle-range theories feature a higher level of contextualization than grand theories (Pinder & Moore, 1979; Weick, 1989). Their prime advantage is that at a given level of simplicity, their lower levels of generalizability permit higher levels of theoretical accuracy (Thorngate, 1976; Weick, 1999). Middle-range theorizing was conceived of as an instrument for the development of grand theory (Pinder & Moore, 1979), yet it has become the most established form of research and proven its merits, standing on its own accord (Corley & Gioia, 2011).

Regarding incentives, established legitimacy standards in organizational research incentivize scholars to seemingly “objectify” contextual information (Johns, 2006). Thus, quantitative researchers tend to (over)emphasize the generalizability of their research, thereby stripping it of contextual connections and obfuscating its BC (Johns, 2006). In the same vein, qualitative researchers tend to remove much of the rich information on the contextual embeddedness of their research, for example to reach established page limits (Pratt, 2008). Even in qualitative research, the honest presentation of actual research processes together with the encountered challenges is rare (Gibbert & Ruigrok, 2010).

For the aforementioned reasons, we are concerned that theory borrowing may often occur in invalid manners, without proper attention to BC. In this respect, it also needs to be noted that new theory is easier to publish than, for example, theory that has its range “trimmed” and that new theory fosters a scholar’s prestige relatively more (Gray & Cooper, 2010). It seems that organizational research as a field must change its attitude and the incentives with respect to BC exploration.

Our third argument in favor of exploring BC is that this type of research contributes to mitigating the research-practice gap. Scholars have discussed for some time that a gap exists between organizational research and practice (Aguinis, Boyd, Pierce, & Short, 2011; Kieser, Nicolai, & Seidl, 2015; Rynes, Bartunek, & Daft, 2001; Shapiro, Kirkman, & Courtney, 2007) because scholars and practitioners live in different worlds, which are ostensibly separated by a chasm (Corley & Gioia, 2011; Nicolai, 2004).

We posit that the primary-empirical exploration of BC can mitigate the research-practice gap because the process draws attention to the descriptive features of contexts and their richness (Weick, 2007), such that cross-context differences are investigated and explained. This makes the respective theory more easily transferable and applicable by practitioners. For example, in the hostile environments case, the distinction between the four forms of hostility allows practicing managers to consider specifically those forms that matter the most to them. Likewise, in the LSP case, the concluding model highlights to practitioners what exactly the specific features of their contexts are that act as barriers to the development of innovativeness, thereby also cautioning them not to follow any supposed “best practices” that are in fact not applicable in their context. This benefit of exploring BC corresponds with a mitigation of the so-called knowledge-transfer problem within the research-practice gap (Shapiro et al., 2007; Van De Ven & Johnson, 2006).

Closely intertwined with the aforementioned benefit of developing the respective theory further, exploring BC may often also mitigate the so-called knowledge-production problem associated with the research-practice gap (Shapiro et al., 2007; Van De Ven & Johnson, 2006), a problem that denotes that scientific ideas and knowledge are essentially unimportant to practice. The BC exploration process “zooms into” the contexts of interest, encouraging scholars to interact with practitioners and pay attention to what they have to say. Thereby, scholars move out of their zones of comfort and critically reassess their prior perspectives (i.e., the theory with uncertain BC). The effect of BC exploration on the further development of the respective theory will hence most likely entail the incorporation of (more) real-life phenomena, as opposed to “scholars struggling to find holes in the literature” (Hambrick, 2005, p. 124).

Some downsides associated with the exploration of BC are worth mentioning. First, opportunity costs for authors should be considered. Scholars who channel their time and energy into BC-exploring research forego investing the same resources into other endeavors. This essentially means that the degree to which a theory calls for having its BC explored and the degree to which promising research alternatives are available will determine the likelihood that BC-exploring research efforts take place. Second, BC-related publications increase the complexity of organizational research by promoting yet another type of study that calls for scholars’ attention, amid large amounts of publications that are already difficult to oversee (Gray & Cooper, 2010), thereby creating opportunity costs also for readers. However, to realize the potential benefit of fostering validity, such publications are required. Finally, it must be conceded that BC exploration is a challenging task. We dare hope that the insights and suggestions in this article mitigate the degree of wickedness associated with BC-related issues substantially, thereby lowering difficulty-related barriers to BC exploration. To further this aim, the next section synthesizes our analysis into 12 practical suggestions.

Tentative Prescriptions for Exploring Boundary Conditions

Based on the answers to our research questions, we developed 12 tentative suggestions on how BC should subsequently be treated by authors, reviewers, and editors given the current state of theoretical-methodological knowledge. We encourage other scholars to posit their own advice for refining these suggestions further.

The previous section has identified valuable benefits of exploring BC, namely, fostering the further development of the respective theory, increasing the validity of organizational research, and mitigating the research-practice gap. This means in essence that BC should be regarded not just an amendment to theory but an integral part of the theorizing process. Accordingly, we offer our first suggestion:

Suggestion 1: General attitude toward boundary conditions: Scholars should regard boundary conditions as a core element within organizational research and as a valuable research topic of its own account.

Unfortunately, prior research has given BC various labels and has employed the notion of BC to refer to all kinds of things. Against this background, it is unsurprising that BC were not taken as seriously in recent years as this study suggests should have been the case. One goal of this research was to foster conceptual clarity by means of developing an accurate understanding of what BC are and how they relate to other concepts (see Table 2). Therefore, we suggest:

Suggestion 2: Conceptual clarity: Future discussions on BC should employ clearly defined and unambiguous terminology that reflects the latest theoretical-methodological research on BC.

By definition, BC are intricately linked with context that is inherently multidimensional (e.g., a firm possesses a certain size, age, level of internationalization, national origin, etc.; see Bamberger, 2008). However, since handling possible interactions between multiple dimensions of context from a BC perspective is undoubtedly very difficult, it seems commendable to consider the influence of a single dimension of context at a time. That is, for practical reasons, we advocate a view on BC as a univariate function that depicts the influence of a single context-delineating variable (see e.g., Figure 1). Accordingly, scholars should design their BC exploring research such that systematic influences of context can be traced back to a single context-delineating variable. This is easier said than done because of the aforementioned multidimensionality of context. It may however be possible to eliminate specific dimensions of context. For example, in the LSP case, our entire data stemmed from Germany. Therefore, the country context could not act as a confounder (yet we could not judge its importance either; see Bamberger, 2008). Had we compared the German LSPs with Chinese telecommunications firms, it would have been much more difficult or even impossible to determine the influence of the industry context. Thus, we suggest:

Suggestion 3: Isolation of contextual influences: To facilitate BC exploration, scholars should aim at employing research designs that isolate the influence of contextual dimensions from another.

Often, dimensions of context are not simply given but can be delineated by scholars. Researchers will usually have some leeway in this respect. Since BC exploration is associated with much effort, as this study has shown, it makes sense to channel resources into those dimensions of context across which lacking generalizability seems at least imaginable or even likely, thereby adopting a polar-type logic (Eisenhardt, 1989) or a logic of pursuing failure (Gray & Cooper, 2010). Our previous analysis related to the research-practice gap suggests practical relevance as another criterion to be considered. The LSP case, as well as the studies of Eisenhardt and Martin (2000) and Ketokivi and Jokinen (2006), highlight that industry, for example, can be a highly relevant contextual variable. By exploring BC along industry membership as a practically relevant context-delineating variable, the aforementioned examples ensured that the theory that had its BC explored also became more context-specific and thus relevant.

On the downside, the LSP case also illuminated challenges related to context delineation. Most importantly, boundaries of the higher-order industry context itself were defined unclearly. That is, we deliberated long (and rather inconclusively) if we should regard independently acting logistics departments as LSPs or not. Moreover, subcategories existed within the LSP context. For example, from the perspective of practitioners, locally operating courier delivery firms, inland shipping firms, and forwarders running a large logistics network differ substantially. We decided against directly adopting these practical categories within our case analysis, but the practitioners’ differentiation triggered theoretically important analyses.

To conclude, context delineations entail numerous challenges that we suspect are largely inherent to the BC topic; they should receive more attention in future research. We suggest:

Suggestion 4: Context delineations: Scholars should carefully ponder how to delineate contexts. To the extent that this is possible, contexts should be delineated such that (a) theoretically meaningful insights from the BC exploration can be expected, (b) practically relevant context delineations are chosen, and (c) context boundaries are unambiguous. Despite their best efforts, scholars should prepare themselves for some uncontrollability of the context delineation issue.

It is imaginable that researchers deliberately seek to probe the BC of a theory, without any particular trigger. However, in light of the opportunity costs associated with exploring BC, it seems more likely that on most occasions, BC exploration is set in motion by specific triggers. Moreover, we argue that various forms of perplexity can often trigger not only the exploration of BC as such but even the understanding for how the respective theory can be further developed.

In the LSP case, it was puzzling to us that innovativeness was depicted in the literature as a key component to survival while LSPs are hardly innovative, especially given the presence of seemingly innovation-favorable conditions. Similarly, while we were initially unaware of the uncertain BC surrounding our independent variable in the hostile environments case, it was still perplexity that made us “dig deeper,” namely, fragmented literature, contradictory empirical results, and our own inability in deducing any convincing hypotheses regarding the effects of the holistic hostility construct on the two forms of innovativeness. Unexplainable results and fragmented conversations throughout the extant literature are tell-tale signs that something surrounding the research process is going unnoticed. Perplexity in terms of variance in effect sizes across studies can even be viewed as the trigger to the meta-analysis variant of the amendment of moderators. Therefore, we suggest:

Suggestion 5: Perplexity triggers: Scholars should consider perplexity, for example, in the form of fragmented literature, contradictory empirical results, outliers, real-world paradoxes, and investigative dead ends, as a trigger to the exploration of boundary conditions.

Scholars who pursue the exploration of BC under uncertainty apply a contextual perspective by definition (see Table 3). For the other deliberate approaches, that is, the inside-out and the outside-in approaches, we advocate that scholars are also attentive to specific real-world contexts and try to understand them very deeply. The rationale is that the benefits associated with exploring BC largely depend on contextually driven theorizing directed at the reasons for context specificity. Attention to rich features of context facilitates the discovery of important new insights that foster theory development. Paying attention to specific features of contexts and to juxtapositions across contexts were also the decisive elements for resolving the initial perplexity in both the LSP and the hostile environments cases, thereby complementing the previous suggestion.

Recall that in the LSP case example, firms were the units of analysis while context was defined as a higher-order variable in which the units of analysis were nested, namely, industry. The individual firms differed not just by the context-delineating variable but also by numerous other variables that co-varied with the context-delineating variable, thereby creating both within-context variance and between-context variance. Some of the other variables depicted noteworthy mean values at the level of the context. The LSP context was hence characterized by (on average) high dependence of the LSPs on their customers, high dispersion, and innovation-adverse mentality of personnel. Interestingly, the respective variables were particularly important for our concluding explanation in the LSP case. We suspect that any noteworthy differences in features that are initially regarded as merely “descriptive” may often represent good starting points for understanding the reasons why a theory may be relatively less applicable in one context than another. In other words, it seems plausible that the relatively larger the between-context variance of “other” variables becomes in relation to within-context variance, the more likely these variables will be meaningful for the BC exploration goal. However, it is also clear that they do not have to be associated with differences in causal effects. Hence, we suggest:

Suggestion 6: Attention to contexts during the exploration of boundary conditions: Scholars who are in the process of exploring boundary conditions should be attentive to specific real-world contexts and look at the respective theory from within those contexts. When context is a higher-level variable, variables that portray relatively much between-context variance vis-à-vis within-context variance may be particularly important within the BC exploration.

Whenever BC are explored under uncertainty, prior theoretical literature must be treated carefully. By relying on theory outside of its range, researchers can jeopardize the validity of their own research. The research design must therefore be at least somewhat exploratory. In the LSP case, an exploratory research design was chosen that was supposed to generate a deep understanding of the specificity of the LSP context. Even in the hostile environments case (a serendipitous BC exploration), the initially planned confirmatory research design had to be suspended until the aforementioned perplexity issues were resolved, by means of attention to context. At the same time, the results of the exploration of the uncertain context must be compatible with prior research (e.g., in terms of the employed concepts) such that the reasons for or against the applicability of the respective theory are understood and that knowledge can accumulate in the long run. The LSP case highlights the inherent tension with its dualistic need for in-depth exploration of a specific context, plus comparison of this context with other contexts. A close look at the chronological sequence of the four studies conducted for the LSP case reveals that after a first exploration of the LSP innovation context, the next study that was completed was the study that featured a cross-context comparison (although this study was started last). While there were some idiosyncratic reasons for this order, we also felt that too much of an emphasis on the open-ended context exploration ceased being instrumental at some point. We conjecture that the same could have happened with an overly strong emphasis on the cross-context comparisons. Therefore, it seems appropriate to intertwine the two facets, right from the beginning, in similar future research designs. Thus, we suggest:

Suggestion 7: Proceeding in boundary condition exploration under uncertainty: Scholars who explore boundary conditions under uncertainty should treat the respective theory very carefully, in a trial-and-error mode. Moreover, they should aim at both exploring the context of interest in depth and undertaking cross-context comparisons. Nevertheless, they should prepare themselves for a rather wicked research process comprising multiple iterations and setbacks.

Borrowing theory is an established practice in organizational research. However, the theoretical-methodological analysis has shown that this practice is only valid if theory is borrowed within its range. More specifically, any construct, causal relationship, or theory must only be borrowed within its range. On the other hand, since context is inherently multidimensional, so is the range of any borrowed (or “target”) theory. Hence, it is clear that no safeguards can exist that provide any guarantee for within-range borrowing. Assuming that the BC associated with the theory that may be borrowed are unclear, the best (and in our view, very potent) advice seems to be that “researchers [should be] thinking about the conditions that favor and disfavor a theory” (Gray & Cooper, 2010, p. 628) when they consider borrowing it. Accordingly, we offer our next suggestion:

Suggestion 8: Systematic scrutiny of borrowed theory: Scholars should systematically think about the range of the theory that they borrow, including all constructs and causal relationships, to avoid out-of-range borrowing, focusing on the theoretically most salient context delineations.

An important motivation to this study was the premature equalization of BC with moderators. The theoretical-methodological analysis from the dynamic perspective on BC and from a theory development perspective has shown that moderators are one, but only one, tool in the theoretical “armory” for exploring BC. The study has identified the refinement of constructs and the amendment of mediators as theoretically equivalent tools; further tools may exist. Moreover, the two illustrative cases, both of which have employed newly suggested tools, have shown how considering the armory as comprising only moderators would obstruct the exploration of BC and restrict it to those settings where moderators are needed. To be clear, we cannot prescribe when which kind of tool from the armory will be needed. Our point is simply that the armory contains at least three tools that can foster BC exploration, not just one. Hence, we suggest:

Suggestion 9: Tools for exploring BC: Scholars who are in the process of exploring the boundary conditions of a theory should consider (a) refining constructs, (b) amending mediators, and (c) amending moderators as theoretical tools for exploring boundary conditions and widening the applicability of a theory. Other tools may also exist.

The ultimate three suggestions refer to the role of reviewers and editors. Because of the benefits associated with exploring BC, we suggest that reviewers and editors push authors to explore or at least consider BC along salient dimensions. However, the sheer effort required for exploring BC along one dimension, together with the multidimensionality of context, should caution reviewers and editors not to overburden authors. Aiming for complete exploration of any theory’s BC represents a life-long research program rather than the achievable content of a single manuscript. Therefore, we suggest:

Suggestion 10: Requesting the consideration of boundary conditions: As a general rule, reviewers and editors should request the consideration of boundary conditions along salient dimensions. However, they should be careful not to overburden the authors beyond what can be achieved within a single study.

Weick (1979, 1999), building on Thorngate (1976), has popularized the notion that usually, a theory cannot be simple, accurate, and generalizable at the same time. ⁸ Given a certain level of simplicity, increases in accuracy come at the price of decreases in generalizability and vice versa. Nevertheless, the current institutional environment incentivizes authors to claim degrees of generalizability of their theoretical models that the models often do not possess. The unjustified equalization of BC with limitations that we pointed out contributes to this bad practice. Given that authors are necessarily biased, we call for reviewers and editors as objective instances in improving these practices. Based on the theoretical-methodological analysis in this study, we suggest:

Suggestion 11: Evaluation of portrayed boundary conditions: Reviewers and editors should regard (a) clearly delineated boundaries as an indicator of rigor, (b) admittedly unexplored boundary conditions as a signal of honesty that is not be judged negatively, and (c) apparent overstatements of generalizability as an indicator of lacking rigor.

Finally, we would like to point out an aspect surrounding BC exploration that resonates with the issue of publication bias and the file drawer problem. Consider the case of a scholar who suspects a context factor to be an influential moderator, tests it empirically, and finds its influence insignificant. According to the established (yet already contested) practice, it is quite likely that this result would never be published. However, from the perspective of BC, it contains very valuable information in that it highlights generalizability with respect to the tested context dimension, thereby making the BC more certain. Based on this reasoning, we offer our final suggestion:

Suggestion 12: Receptiveness to lacking influence: From the boundary conditions perspective, reviewers and editors should be more receptive to discussions of lacking influence of variables in theoretical models.

“Boundary Conditions” of Boundary Conditions

This section depicts some “when questions” related to our analysis. The most important question from our own perspective is, When and in what way does our BC analysis depict any insights and suggestions that differ from theory development? Indeed, when we look at BC exploration in isolation, it operates in the same way as theory development. For example, it can employ numerous well-known methodologies and theoretical “tools” (see Figure 3). That is, the exploration of BC truly is a specific form of the theorizing process. However, we would like to point out two specific features. First, from the static perspective on BC, the outcomes differ. While ordinary (i.e., what-, how-, and why-related) theory development produces a causal model that explains (predicts) a real-world phenomenon (i.e., it deals with developing a theory that fits the data), the outcome of BC exploration reflects a form of meta-theory that denotes the cross-context generalizability of a given theory (i.e., it describes the accuracy of the fit between a given theoretical model and various empirical contexts). Moreover, this kind of meta-theory has to deal with an issue of unresolvable uncertainty, as per the uncertain view on BC and given the multidimensionality of BC.

Second, from the dynamic perspective on BC, we see that BC exploration (i.e., an investigation which makes BC more certain) is not only intimately intertwined with the theory development process but even serves as a means to the end of (further) theory development. Accordingly, our study suggests that BC have a dualistic nature in that they are both an amendment to theory and a means for theory development. Therefore, we believe that searching for the difference between BC and theory development leads us in the wrong direction. BC exploration is a form of theory development that complements the “classical” form, thereby helping to exploit its full potential.

Another important question is, When is the exploration of BC particularly important or urgent? Because of the associated opportunity costs, it seems to us that perplexity drivers are decisive here (see Suggestion 5). Contested terrain between theories also calls for in-depth exploration of BC. With respect to the question, When is the exploration of BC particularly complicated?, we suspect that there may be two causes. We believe that a delineation of context as a higher-level variable complicates matters. Similarly, the different approaches for BC exploration differ in their wickedness, which is a measure of complicatedness (see Table 3). Regarding the question, When should which approach be chosen? (see Table 3), we advocate avoiding serendipitous BC explorations by means of systematically considering BC of all borrowed theory (see Suggestions 8 and 10). The choice between exploration of BC under uncertainty and the other two approaches is determined by a scholar’s perspective (i.e., theoretical vs. contextual). Choosing between the inside-out and the outside-in approach is facilitated by the researchers’ goal and a certain “spirit of adventure.” Finally, concerning the question, When is which tool from the “armory” most appropriate?, we believe it depends fully on the specific theoretical context. Similarly, we do not see any connection between the approaches (see Table 3), tools (see Figure 3), and the many domains through which BC have to be explored (see Table 1). In other words, while our “BC theory” may not yet be very accurate, the conception is simple and hopefully quite generalizable (Thorngate, 1976; Weick, 1999).

Limitations of this Study and Future Research on Boundary Conditions

We feel obliged to point to four limitations of this study. First, we departed from the assumptions that reality exists objectively, independently from theory, can be mapped in theoretical models with constructs and relationships between them, and that researchers are limited in their cognitive abilities but strive for “objective” theoretical models. It may be useful to complement this onto-epistemological perspective on BC with alternative perspectives. Second, we think of this study as a theoretical-methodological analysis of BC, with some empirical illustrations and tentative first suggestions for future research. Correspondingly, we did not intend for this research to become an all-encompassing literature review surrounding BC. Third, we must concede that our illustrative cases only explore BC for theory in the sense of accumulated empirical evidence (Shrivastava, 1987). The simple reason is that we could only report on examples that we conducted thus far. Thankfully, this limitation does not seem to jeopardize the validity of our theoretical-methodological analysis. Finally, to be clear, we make no claim to have answered any of our research questions comprehensively or definitively. Rather, our hope is that this article contributes to consensus shifting (LeBreton, 2014) with respect to BC and hopefully triggers additional future research on BC-related issues.

This study points to six topics as valuable opportunities for future methodological research on BC. First, it would be informative if scholars who managed to establish BC for a highly respected theory shared their insights and experiences with the methodological community. Second, there may be room for further development (especially expansion) of the BC-related toolkit (i.e., the approaches and tools for exploring BC). Third, future research could further develop BC-guided insights on the issue of validity, specifically regarding the validity of theory borrowing. Such research could possibly generate new prescriptions for one of the most important manifestations of methodological rigor. Fourth, studying BC from a multilevel perspective (e.g., Klein, Dansereau, & Hall, 1994) may deepen the analysis conducted here. Fifth, methodologists have studied construct development from theoretical and statistical perspectives. This research suggests that valuable methodological prescriptions may yet be derived from developing a better understanding of the ranges of constructs. Sixth, venturesome theorists and methodologists could consider the question if any meaningful “context theories” (Bamberger, 2008) exist and what they look like. It seems to us that such theories would have to describe systematic influences of context on multiple theories if they were to become more than a collection of context-specific findings. Dedicated research streams on, for example, start-up firms or business in emerging economies inspire a spark of hope that such theories are imaginable.