Qualitative comparative analysis: justifying a neo-configurational approach in management research

Introduction

In the past two decades, management research has developed a significant interest in advancing the neo-configurational approach, which “enables researchers to theorize and empirically examine causal complexity” (Misangyi et al., 2017, p. 257). Causal complexity is guided by the three principles of:

1. conjunction, which refers to an outcome occurring from the interdependence of multiple conditions (Schneider & Wagemann, 2012);

2. equifinality, which suggests the possibility of multiple pathways leading to the same outcome (Gresov & Drazin, 1997); and

3. asymmetry, which means that attributes “found to be causally related in one configuration may be unrelated or even inversely related in another” (Meyer, Tsui, & Hinings, 1993, p. 1178).

However, the dominant research tools, primarily based on correlations, are not designed to capture these three principles of causal complexity (Ragin, 1987, 2000); instead, these tools are characterized by single (and net effect logic), linear, and symmetric theorizing (Abbott, 1988; Delbridge & Fiss, 2013; Ragin, 2008). Thus, through Ragin’s (1987) seminal work, qualitative comparative analysis (QCA) has emerged and gained prominence over the years as a widely embraced research tool for the empirical exploration of the configurational approach.

Departing from the epistemological foundation of correlation-based methods that use linear algebra, QCA relies on set-theoretic relations by adopting Boolean algebra to understand interactions or unions of set memberships that take into account each case as a configuration or bundle of causal attributes. Thus, in using QCA, it is possible to compare and examine cases with different sets of causally relevant conditions to identify the decisive configurations, and thereby unravel causal complexity (Ragin, 2008).

Developed initially for case-based analysis in sociology by Ragin (1987), QCA is increasingly applied in different management research sub-disciplines (Wagemann, Buche, & Siewert, 2016) such as organization design (e.g., Grandori & Furnari, 2008), strategy (e.g., Fiss, 2011; Greckhamer, Misangyi, Elms, & Lacey, 2008), marketing (e.g., Frambach, Fiss, & Ingenbleek, 2016; Johansson & Kask, 2017), corporate governance (e.g., Bell, Filatotchev, & Aguilera, 2014; García-Castro, Aguilera, & Ariño, 2013), public administration (e.g., Federo & Saz-Carranza, 2018b; Verweij, Klijn, Edelenbos, & van Buuren, 2013), international business (e.g., Crilly, 2011; Schneider, Schulze-Bentrop, & Paunescu, 2010) and family business (e.g., García-Castro & Aguilera, 2014; García-Castro & Casasola, 2011), among others.

The key advantage of QCA over other research tools rests on its capability to capture altogether the three principles of causal complexity (Misangyi et al., 2017). First, it is primarily used to analyze how multiple, independent causal attributes are combined so that they are consistently associated with a given outcome (i.e. conjunction). Second, it helps to assess whether there are different combinations of conditions associated with the same outcome (i.e. equifinality). Finally, it explores the possibility that both the presence and the absence of attributes could be associated with the outcome (i.e. asymmetry).

However, despite the surge in the number of studies over the past two decades, demonstrating the shift toward using configurational theories to understand the complexity of organizational phenomena, there are persistent scholarly debates in the literature over the use of QCA to analyze causal complexity. On the one hand, the QCA scholarly community advocates QCA as a novel and available tool that is capable of capturing all three principles of causal complexity and is thus the preferred choice for empirically testing configurational theories (Fiss, Marx, & Cambré, 2013). On the other hand, scholars have underscored the various pitfalls of QCA, particularly when contrasting it to conventional quantitative methods. For example, Lieberson (2004) argued that there is no evidence for supplanting existing correlation-based practices by QCA because of QCA’s deterministic approach and disregard of probabilistic processes. In a similar vein, Seawright (2005) pointed out three assumptions about causal inference that QCA has failed to address: the absence of an established testing tool for nonlinear functional form, the treatment of missing variables, and the inherent implied causation.

Although Ragin and Rihoux (2004, p. 22) have previously addressed these concerns, stating that “supplanting regression analysis and related techniques is not our goal, nor is it the goal of others who advocate QCA”, that the application of QCA also involves probabilistic approaches, and that QCA carefully takes into account nonlinearity, omitted variables, and case-based causal inferences, skeptics continue to emphasize how QCA might not be useful; for instance, Tanner (2014) argued that QCA is of questionable value for policy research, while Hug (2012) highlighted the measurement error underlying the application of QCA. As a result, QCA scholars have published several papers listing best practices when conducting QCA to avoid the pitfalls (e.g., Greckhamer, Furnari, Fiss, & Aguilera, 2018; Leppänen, McKenny, & Short, 2019; Misangyi et al., 2017; Thiem, 2017), in addition to the detailed QCA methodological principles previously discussed in the literature (Ragin, 1987, 2000, 2008; Rihoux & Ragin, 2009; Schneider & Wagemann, 2012). Furthermore, alternative techniques to QCA as a configurational comparative method also have emerged (Baumgartner, 2009), wherein other scholars advocating the configurational approach have criticized some of the assumptions and generally accepted best practices of QCA (Baumgartner & Thiem, 2017). However, our goal here is not to review these prior works, but instead we aim to critically reflect and offer insights on how to justify the use of a QCA-based study for future research endeavors in the field of management.

We focus on three underlying arguments that warrant the use of QCA in a study. First, we discuss the importance of using configurational theories that sparked the adoption of QCA. Second, we elaborate on the need to understand altogether the three principles of causal complexity, which is the logic behind the development of QCA in the first place. Third, we emphasize the role of case knowledge as the foundation of QCA-based studies. We argue that understanding the core of these three arguments provides researchers with a justification for whether or not QCA is the appropriate method to explain the phenomenon of interest.

In the next section, we discuss the three arguments justifying why a QCA research design would be warranted to examine causal complexity and to identify the stages in which QCA supports such reasons. We then describe the general steps in conducting a QCA-based study. We conclude with a future outlook for the configurational perspective and QCA.

Underlying arguments for when to use qualitative comparative analysis

Before choosing QCA, the researcher must be aware of whether the method is warranted for the study. QCA is only applicable for studies that seek to offer theoretical contributions from a configurational perspective. It will also be useful only if the intention is to unravel causal complexity relations and if it is based on case knowledge, in which the researcher seeks to observe how different attributes of the cases consistently fit together to produce an outcome. If these three goals are the purpose of the study, researchers may then choose QCA to conduct the study. To make this clear, we give further detail on each of them below.

Procedure for conducting qualitative comparative analysis

Although the process of conducting QCA has been discussed in great detail elsewhere, we argue that it is useful to provide a brief overview of the process involved when adopting a QCA-based study. The procedure for conducting QCA is made up of four general steps (see Figure 1): designing the configurational model, building the empirical data, calibrating and analyzing the data, and reporting and interpreting the results.

Once the phenomenon to be studied and the outcome are identified, the first step is to design the configurational model. Based on theory and case knowledge, the researchers need to identify the conditions that could explain the outcome of interest. It is important to adopt a configurational perspective by identifying which conditions should have joint effects, rather than net effects, on the outcome. The challenge is to maintain a balance between the number of conditions and the sample size, to minimize limited diversity, which refers to the likelihood of having unobserved configurations because of the exponential increase in logically possible configurations associated with an increase in the number of conditions.

The next step is to build the empirical data by purposively choosing the theoretically defined sample cases for the analysis (Ragin, 2008). The goal is to ensure that the chosen cases are fitted to answering the research question (Greckhamer et al., 2018). Although initially developed to find consistent relationships in sample sizes that are too large for comparative case studies but not large enough for quantitative research designs (Ragin, 1987), QCA has been applied to studies with N that ranges from relatively small (e.g., Federo & Saz-Carranza, 2018a, 2018b; Verweij et al., 2013) through medium (e.g., Grandori & Furnari, 2008; Haxhi & Aguilera, 2017) to large (e.g., Bell et al., 2014; García-Castro et al., 2013; Misangyi & Acharya, 2014) or extremely large (e.g., Campbell et al., 2016; García-Castro & Aguilera, 2014; García-Castro & Casasola, 2011). Case selection can be made by taking into account the entire population or representative sample of a population. Random sampling typically is not advised, because exceptional cases, also known as outliers or deviant cases, might be relevant in explaining the outcome (Greckhamer, Misangyi, & Fiss, 2013).

After case selection, the outcome and conditions need to be calibrated for set memberships (Ragin, 2008). QCA application has evolved from using solely crisp sets, where set membership is distinguished between full membership or full non-membership, into incorporating in the analysis fuzzy sets with more finely-grained degrees of membership (Ragin, 2000). Another QCA variant, that is multi-value QCA, also has emerged for analyzing specifically categorical variables or intermediate set memberships (Cronqvist & Berg-Schlosser, 2009). Nevertheless, the main steps for conducting QCA along the three types are similar (Herrmann & Cronqvist, 2009). The calibration technique rests on transparency when identifying theoretically or substantively based thresholds, to ensure validity and replicability of the calibration process (Greckhamer et al., 2018; Misangyi et al., 2017). Although sample-based calibration is discouraged, the properties of the sample through its frequency distribution can also be adopted in circumstances where there is no existing theoretical knowledge that can be used for calibration thresholds (Greckhamer, 2016).

The next step after data calibration is data analysis. To do this, researchers need to build and analyze truth tables, referring to the number of rows representing the logically possible configurations from the given bundle of conditions (2^k where k is the number of conditions used in the analysis). Using the logic of necessity and sufficiency (Ragin, 2008), the goal here is to identify the configurations that consistently produce the outcome. Consistency and coverage scores are used to evaluate the results; the former refers to the measure of fit among the different conditions comprising a configuration yielding the outcome, while the latter refers to the empirical relevance of the configuration (Ragin, 2008). Consistency scores suggest how often the cases exemplifying the configuration produce the outcome of interest. Hence, a consistency score of 0.80 would mean that 80 per cent of the cases are showing the relationship. A consistency score of at least 0.90 is recommended for a condition to be considered necessary, meaning that the condition always needs to be present to produce the outcome. Meanwhile, a raw consistency score of at least 0.80 and a proportional reduction for inconsistency (PRI) of at least 0.65 are recommended to consider a condition or configuration to be sufficient (Greckhamer, 2016), meaning that the condition or configuration would be enough to produce the outcome. It is important to conduct a necessity analysis of individual conditions before conducting a sufficiency analysis. A frequency threshold, referring to the minimum number of observed cases representing a row, is also set during the analysis. Although one case is typically used in small-N analysis, a higher number can be set in medium-N and large-N analyses, provided that the analysis retains at least 80 per cent of the cases (Greckhamer et al., 2013).

Furthermore, as with any other research methods, QCA is sensitive to various methodological decisions when performing the analysis, and hence robustness checks are also encouraged to ensure the validity of the findings. However, an examination of the findings’ robustness needs to follow the set-theoretic logic in which the resulting necessity and sufficiency of the conditions after the checks do not offer substantively different interpretations of the findings (Schneider & Wagemann, 2012). Some ways of conducting the robustness checks include adding, dropping or changing the conditions in the model (e.g., García-Castro & Francoeur, 2016), modifying the calibration thresholds (e.g., Fiss, 2011), and exploring multiple consistency thresholds (e.g., Ragin & Fiss, 2017).

The final step in conducting QCA is to present and interpret the results. In presenting the results, researchers may opt to display the consistent results in a configuration table using Boolean formulas or the notation suggested by Ragin and Fiss (2008). An example of a Boolean formula is: Condition A+Condition B*∼Condition C → Outcome, where the plus sign (+) denotes “or”, the asterisk (*) denotes “and”, the tilde (∼) denotes absence, and the arrow (→ ) shows the causal direction. The notation by Ragin and Fiss (2008) uses “●” for the presence of the condition and “○×” for the absence of the condition. In some instances, unobserved configurations known as counterfactuals or logical remainders are present during the analysis. QCA allows researchers to perform counterfactual analysis to identify core and contributing conditions (Fiss, 2011). When considering easy counterfactuals (i.e. those that are consistent with theoretical knowledge and empirical evidence) and difficult counterfactuals (i.e. those that are only consistent with empirical evidence), parsimonious solutions are generated, which produce the core conditions. However, if only easy counterfactuals are considered, intermediate solutions are generated, which produce both core (only those conditions from the parsimonious solutions) and contributing conditions. If no counterfactual analysis is performed, complex solutions are generated. Although it is recommended that intermediate solutions are presented as results, scholars have also used complex solutions to present findings as close to the data as possible (e.g., García-Castro et al., 2013; García-Castro & Aguilera, 2014). Another best practice in conducting QCA is to perform analyses for both the presence and the absence of the outcome, to explore the asymmetry assumption underlying causal complexity. This step is particularly important to ensure that the inverse combinations of conditions resulting to the presence of an outcome are not related to the absence of the outcome.

Finally, to interpret the results, researchers will need to rely on case knowledge to make sense of the configurations that emerge from the analysis. QCA remains a qualitative exercise, and thus entails an understanding of the cases selected for the analysis. The cases could determine whether the configurations are exemplified by typical or deviant cases that can be helpful in interpreting the results.

Discussion and conclusion

Our aim in this article was to critically reflect and offer insights on how to justify the use of QCA in future research endeavors in the field of management. We have done this by critically analyzing three arguments as to why QCA would be warranted for use in a research study. First, we discussed the need to assume configurational theories to build and empirically test a causal model of interest. Second, we explained how the three principles of causal complexity are assumed during the process of conducting QCA-based studies. Third, we elaborated on the importance of case knowledge when selecting the data for the analysis and when interpreting the results. We contend that these three arguments need to be the underlying goals of the research to justify choosing QCA as the method.

Our article primarily contributes to configurational research by reinforcing the importance of QCA-based studies, while we underscore these three arguments that have now been taken for granted when choosing QCA as the research method. We argue that it is important to reflect on these arguments to have the appropriate research design. In the true spirit of the configurational approach, we contend that the three arguments we have presented above are necessary; however, each argument is insufficient to warrant a QCA research design.

The strength of QCA lies in how it can integrate the three arguments altogether. As QCA analyzes joint effects rather than linear relationships or the net effects of conditions, in contrast to correlation-based methods, it builds a better configurational model based on case knowledge while assuming the principles of causal complexity. Previous research has shown that QCA overcomes the limits of structural equation modeling (SEM), that is argued to capture multiple interactions; yet, SEM fails to capture the necessity or sufficiency of conditions to produce an outcome (Tho & Trang, 2015). Moreover, QCA helps create more finely grained typologies, since it segregates cases on the basis of their attributes that are associated with a specific outcome (Fiss, 2011), which is in stark contrast to cluster analysis that merely looks commonalities among variables. Misangyi and colleagues (2017), however, contend that one of the prospects for QCA is to be a complementary approach to other research methods. Indeed, Ragin and Rihoux (2004) have strongly argued that QCA is not meant to replace any existing and established research methods. Rather, QCA should be viewed as a way to provide alternative theorizing and empirical testing of a phenomenon that researchers want to explain. Despite the difference between the epistemological and methodological traditions of QCA and those of other research methods, the scholarship has in fact now shifted toward using QCA not only to compare findings with results from regression analyses (e.g., Fiss, 2011; García-Castro et al., 2013; Huang & Huarng, 2015; Meuer, Rupietta, & Backes-Gellner, 2015) but also to reinforce grounded theorizing from traditional comparative cross-case studies using qualitative data (Aversa et al., 2015; Bromley, Hwang, & Powell, 2012; Mol & Birkinshaw, 2014). We concur with this future outlook for QCA, in which useful insights may be generated by combining QCA with other methods to enhance our understanding of the complexity of the phenomenon that researchers want to study (Misangyi et al., 2017).

Nevertheless, QCA is still a work in progress. There are facets underlying the process of conducting the research method needing to be developed. For instance, in building the configurational model, it is challenging to account for all the conditions, or at least the most salient ones, that can explain the outcome of interest. This issue is reflected in the fact that there are several published large-N studies even in top management journals showing relatively low overall solution coverage scores, which might suggest that the configurational model is not adequately explaining the outcome. A question that we want to raise is whether it would also be useful to establish a minimum threshold for overall solution coverage scores to minimize alternative explanations.

Moreover, the challenge of data calibration has also persisted as an issue in QCA-based studies. Although QCA scholars have stressed the importance of theoretical and substantive knowledge in data calibration, the use of data-specified calibration techniques (such as percentiles, data distribution, and rank order as thresholds) in published studies in top management journals (e.g., Ganter & Hecker, 2014; Greckhamer, 2016) has deviated from this best practice. This can be attributed to nonexistent qualitative anchors for set memberships, which scholars argue can still be acceptable when conducting QCA provided that researchers are transparent about the calibration decisions (e.g., Greckhamer, 2016; Thiem & Duşa, 2013; Verkuilen, 2005). However, such calibration techniques raise questions about the subjectivity of the thresholds and the sensitivity of the results to slight changes in the calibration decisions. Thus, we still have to ponder how to standardize data calibration, since it is a crucial step in conducting QCA-based studies.

Finally, despite the growth in the application of QCA in several sub-disciplines of management research, there are notable sub-disciplines that are yet to produce QCA-based studies. For instance, we have still not seen studies in operations research, finance or psychology that have adopted QCA. A question that we can raise here is why these sub-disciplines continue to distance themselves from using QCA. It would be a rich ground for future research in these sub-disciplines to explore the neo-configurational approach using QCA.

In conclusion, QCA was developed as a method for empirically examining causal complexity, and eventually emerged as the pillar of the neo-configurational approach. Although relatively new in management research, with its first application only in the 2000s, QCA has become a promising tool for management scholars. As a result of its infancy as a research method, QCA is not without its limitations; however, the growing scholarly interest in QCA has put pressure on the continuous development and improvement of its application for management research. For this reason, the number of scholars advocating QCA has also increased over the years, and we expect this trend to continue in the future. Because of this outlook, we hope that, although parts of it are prescriptive, this article will contribute to the field, as we offer insights for future scholars who wish to adopt QCA in their research endeavors to enable them to consider a neo-configurational approach and justify the use of QCA.