Introduction

Configurational theorising shifts researchers’ attention from the assessment of the “net effects” of causal variables to a more contextual understanding of the multiple possible ways in which causal conditions may combine to produce a given effect (Ragin, 2008). Configurational theorising revolves around three tenets: 1) Conjunctural causation: the effect of a single condition unfolds in combination with other conditions; 2) Equifinality: multiple configurations (or combinations) of conditions may lead to the same outcome; 3) Causal asymmetry: the causes leading to the presence of an outcome of interest may be quite different from those leading to the absence of the outcome. According to Dess et al. (1993: p. 776) “a configuration contains relationships among elements or items representing multiple domains”. Therefore, configurational theorising moves the theoretical discourse forward because it is not confined to the study of net effects (i.e., the more X, the more Y). For example, “linear regression examines the net effect of a variable on the outcome by holding other variables constant” (El Sawy et al., 2010: p. 839). Configurational theorising instead studies the holistic effect stemming from a configuration (or combination) of causal conditions. QCA is probably one of the most formalised configurational, comparative methods that relies on Boolean algebra rather than linear algebra, the logic of implication rather than covariation and conjunctural causation rather than simple interaction effects (Thiem, Baumgartner & Bol, 2016).

Theory

Configurational theorising is premised on the assumption that configurations (or combinations) of causally-relevant conditions should be linked to the outcome of interest. Since the focal unit is the configuration (rather than the individual variable), it follows that a given condition may have a different effect on the outcome depending on its combination with other conditions. This notion, in turn, fits with the idea of causal complexity. It implies that a causal condition may have opposite effects depending on its combination with other conditions, so much so that the same condition may contribute to the presence of the outcome when other conditions are present, but it may actually contribute to the absence of the outcome when other conditions are absent (Ragin, 1989). As a result, “the researcher is urged not to "specify a single causal model that fits the data best" (the standard practice using statistical techniques)” (Ragin, 2014: p. xxii). This, in turn, will spur researchers to discover multiple causal models that involve conjunctions of three or more conditions, thus moving beyond second-order or third-order interaction terms (Schneider & Wagemann, 2012).

Configurational theorising revolves around the following six principles (Ragin, 2014):

1. Sets rather than variables: standard statistical techniques are based on “variables”, that is, units that can take on a range of values so as to sort, rank or array observations relative to one another. Instead, configurational theorising is based on sets, that is, groupings that entail membership criteria and have classificatory consequences. For example, “male” is a set that invokes a group of individuals (i.e., male individuals) whereas “gender” is a variable. Likewise, “Swedish” is a set that invokes a particular group (e.g., the Swedish population), but “nationality” is a variable (Pappas & Woodside, 2021). By the same token, “degree of democracy” is a variable, but “democratic” invokes a set such as the group of “democratic countries.” Again, it is important to reiterate that sets are not simple nominal-scale classification (e.g., democratic versus not-democratic countries) because observations (e.g., countries) can vary in the degree to which they satisfy membership criteria. For example, a country can be a full member of the set (or group) of democratic countries (scored as 1.00) while another country can be a full non-member of this very same set (scored as 0.00) and yet another country can be neither in nor out of the set of democratic countries (scored as 0.5, the point of maximum ambiguity).
2. Calibration rather than measurement: standard statistical techniques are based on variables that are measured by using valid and reliable scales or indicators. Observations “are evaluated relative to one another, based on inductively derived, sample-specific statistics such as the mean and standard deviation. For example, a "high" score is well above the mean score; a "low" score, is well below the mean score. All variation in an indicator is usually treated as meaningful and taken at face value” (Ragin, 2014: p. xxiv). To this end, calibration is the process by which set membership scores are assigned to observations on the basis of external standards. For example, taking the uncalibrated variable of per-capita Gross National Product (GNP), it is possible to calculate membership in the set of rich countries by using three external standards, namely the score that would qualify a country as a full member in the set of rich countries (scored as 1.00), the score that would qualify it as a full non-member in the set of rich countries (scored as 0.00) and the cross-over point (where the country in question is scored as 0.50 because it is neither in nor out of the set of rich countries).
3. Qualitative outcomes instead of dependent variables: standard statistical techniques revolve around dependent variables, so much so that “the goal of research is to explain cross-case and/or longitudinal variation in the chosen dependent variable” (Ragin, 2014: p. xxv). Configurational theorising instead focuses on qualitative outcomes, that is, observable changes or discontinuities. For example, instead of studying longitudinal or cross-case reduction in welfare spending, analysts should first define the concept of interest (i.e., the key features of welfare state retrenchment) and then calculate the countries’ membership in the set of countries experiencing welfare state retrenchment (the outcome of interest).
4. Constructed populations rather than given populations: again, standard statistical techniques use either given populations or random samples from these populations. Instead, configurational theorising entails constructing populations in the course of the investigation, by comparing both the presence and absence of the outcome of interest (i.e., positive and negative cases).
5. Set relations rather than correlations: standard statistical analyses are based on correlations (“the more of X, the more of Y”). Correlational arguments are symmetric arguments (“if more X entails more Y, then less X implies less Y”). Configurational theorising on the other hand is based on asymmetric set relations. For example, “the assertion that “the developed countries are democratic" does not require that the not-developed countries be not-democratic. There can be many not-developed countries that are democratic, and their existence does not count against the initial claim, which is asymmetric” (Ragin, 2014: p. xxivi). Accordingly, standard statistical techniques parse matrices of bivariate correlations or their mathematical equivalents. Instead, configurational theorising uses truth tables that list different configurations of causally-relevant conditions.
6. Causal recipes rather than net-effects: the standard statistical template revolves around net-effect thinking, that is, “the net effect and statistical significance of each causal variable are based on its unique (non-overlapping) contribution to explained variation in the dependent variable” (Ragin, 2014: p. xxivii). As such, configurational theorising is about how individual conditions combine to produce the outcome of interest, thus helping analysts formulate causal recipes that will lead to the outcome of interest. These recipes can be evaluated on the basis of their theoretical (i.e., consistency parameters) and empirical importance (i.e., coverage parameters). It is worth stressing that consistency and coverage (also known as parameters of fit) “are analogous to the respective assessments of significance and strength in regression” (Misangyi et al., 2017: p. 269-270).

Although conditions are oftentimes selected in a deductive manner when deploying configurational theorising (Park, Fiss & El Sawy, 2020), it is possible to evaluate such theories by looking at the extent to which theoretical expectations overlap with empirical results. While the focus of standard hypothesis testing is to reject (or fail to reject) the null hypothesis (or a similar benchmark), the focus of theory evaluation in the context of configurational theorising is to evaluate hunches derived from theory “by creating intersections of the Boolean expression describing the theory (T) and the empirical solution (S)” (Schneider & Wagemann, 2012: p. 305). In other words: although configurational theories sit somewhere "midway between exploratory and hypothesis-testing research" (Kent, 2005: p. 226), the design is not appropriate to be used for hypothesis-testing, but rather for the creation of "propositions" which determine the membership in certain configurations (i.e. pathways). The intersection of Theory (T) and empirical solution (S) describes the part of the theory that is supported by empirical evidence. In the intersection of the lack of theory (NOT T), and the empirical solution (S), empirical findings overlap with those cases not expected by theory. The result of this intersection suggests an extension of existing theories. The intersection of theory (T) and the absence of empirical solution (NOT S) captures those cases for which theory predicts the occurrence of the outcome, but which our solution fails to capture. Hence, it suggests a delimitation of existing theories. Finally, the intersection of the lack of theory (NOT T) and the absence of empirical solution (NOT S) “denotes a configuration of conditions that neither theory nor the empirical findings deem sufficient for the outcome” (Schneider & Wagemann, 2012: p. 305).

Theory Updates/Extensions

Configurational theories have recently been updated with the use of set-theoretic multi-method research (Oana, Schneider & Thomann, 2021). In other words, configurational theorising becomes stronger when researchers add within-case evidence to bolster their understanding of causality. In addition, configurational approaches can be amended by qualitative, ex post follow-up research to go into more detail, especially with regard to the identified non-predicted cases (e.g., by means of hold-out samples). Set-theoretic, multi-method research is a formalised process of identifying the best available cases for within-case process tracing in a given data set. Cases are classified as either typical, deviant or individually irrelevant. Subsequently, the best-matching pairs of cases are identified in order to perform comparative analyses mimicking the Most Similar or Most Different case study design (George & Bennett, 2005). For example, the comparison of typical cases with other typical cases may provide useful inferences about the generalisability of underlying causal mechanisms. On the other hand, the comparison of typical cases versus deviant cases may provide useful inferences about omitted conjuncts (i.e., single conditions) or conjunctions (i.e., combinations of conditions). Likewise, the comparison of typical cases with the individual irrelevant cases may provide useful inferences about the causal properties of mechanisms (Oana, Schneider & Thomann, 2021).
Another extension is Necessary Condition Analysis (NCA). Championed by Dul and colleagues, NCA is a data analysis technique based on necessity logic that can be applied either with linear algebra (as in regression) or with Boolean algebra (as in QCA) (Dul et al., 2010). Yet another extension is the study of temporal dynamics. While original formulations focused on time-related conditions, procedural variables, non-commutative sequences of conditions and the like (Fischer & Maggetti, 2017), more recent formulations have identified three distinct approaches aimed at tracking configurations over time: 1) multiple time period, single configurational analysis: the analyst performs one single configurational analysis by splitting cases into different time periods and then performs the analysis using a single truth table; 2) multiple configurational analysis for different time periods: the analyst performs multiple configurational analysis for the same sample of cases for different time periods; 3) fuzzy-set ideal type analysis: this approach calculates the cases’ membership score in different truth table rows and identifies which cases score more than 0.5 in specific rows. “The configuration in which a case has a membership of >0.5 is the ideal type it represents (sic) When calculating the case’s ideal type for different periods in time, a researcher can analyse how cases move (or not) over time in the property space”, that is, the truth table (Verweij & Vis, 2021: p. 105).

More recently, scholars have developed a trajectory-based configurational approach that conceptualises configurations dynamically, so that they express different development stages. Accordingly, scholars can now track qualitative variations occurring within single cases over time in order to show how cases can switch from one configuration to another configuration, thus tracking their trajectories, that is, their sequential movement over time (Pagliarin & Gerrits, 2020). Other approaches refer either to panel data or to time series. Whereas panel data approaches “provide some diagnostic tools to assess a set-theoretic consistency and coverage both cross-sectionally and across time” (Garcia-Castro & Ariño, 2016: p. 63), time series approaches incorporate time series variations (Hino, 2009).

Applications

Originally formulated within the Political Science and Sociology disciplines (Rihoux & Marx, 2013), the theory has been applied to other fields such as Business and Management (Harms, Kraus & Schwarz, 2009), International Relations (Ide & Mello, 2022), Sustainability (Meng, Yan & Xue, 2018), Marketing (Pappas, 2018), Education (Snelson-Powell, Grosvold & Millington, 2016) and Information Systems (Park, El Sawy & Fiss, 2017). Early publications appeared in leading Sociology journals. For example, Amenta et al. (1992) used the crisp-set version of the theory to study under what conditions the Townsend movement succeeded or failed to seek pensions for the aging population (Amenta, Carruthers & Zylan, 1992). Likewise, Roscigno and Hodson (2004) used the configurational approach in concert with quantitative methods to allow “for the examination of unique configurations of organisational and interpersonal dynamics that either diminish or exacerbate collective and individual resistance” (Roscigno & Hodson, 2004:p.15).  Around that time, new publications appeared in the field of Business and Management, especially facilitated by the Journal of Business Research (Stokke, 2007) and in the area of International Business (Pajunen, 2008). Fiss (2007) argued that research on organisational configurations has been hindered by a mismatch between theories and methods, thus endorsing the configurational approach as a viable alternative (Fiss, 2007). Afterwards, Fiss (2011) complemented these early contributions by showing the relevance of the configurational approach for typology theorising and further developed the notion of core and peripheral conditions, depending on the strength of the evidence for a causal relation with the outcome of interest (Fiss, 2011). Around this time, configurational theorising was also introduced within the Information Systems field (El Sawy et al., 2010). More recently, many publications have appeared within the social sciences that draw either on quantitative data (Park, El Sawy & Fiss, 2017; Covin et al., 2020; Pappas, 2018), qualitative data (Aversa, Furnari & Haefliger, 2015; Iannacci & Cornford, 2018) or both (Mattke et al., 2021; Bouncken et al., 2020). Table 1 summarises an exemplary collection of such empirical contributions from different fields of research within the social sciences.