Abstract
This article explores the practical challenges one faces when combining qualitative comparative analysis (QCA) and process tracing (PT) in a manner that is consistent with their underlying assumptions about the nature of causal relationships. While PT builds on a mechanism-based understanding of causation, QCA as a comparative method makes claims about counterfactual causal relationships. Given the need to ensure alignment between the ontological understandings of causation that underlie a method and methodological practice, the different ontological foundations result in methodological guidelines that contradict each other, forcing the analyst to choose whether to be more in alignment with one or the other method. This article explores the implications of contrasting guidelines in a practical case study, where a QCA for sufficiency is followed by two PT case studies of positive cases.
This article explores the practical challenges one faces when combining qualitative comparative analysis (QCA) and process tracing (PT) in a manner that is consistent with their underlying assumptions about the nature of causal relationships. While PT builds on a mechanism-based understanding of causation, QCA as a comparative method makes claims about counterfactual causal relationships (Beach and Pedersen 2016a). 1 Given the need to ensure alignment between the ontological understandings of causation underlying a method and methodological practice (Hall 2003), the different ontological foundations result in methodological guidelines that contradict each other on issues such as types of causal claims being investigated and case selection practices. The consequence of this is that the analyst is forced to choose whether to be more in alignment with one or the other method.
The contradiction in methodological principles is best seen in relation to existing guidelines for proper case selection. This article illustrates that guidelines for proper case selection that are aligned with the type of causal claims being made in fuzzy-set QCA (fsQCA) are more restrictive than those that are appropriate in PT case studies aimed at tracing mechanisms. Existing guidelines for case selection for PT after an fsQCA for sufficiency suggest we should only select typical cases where the fuzzy score for membership in the conjunction is lower than membership in the outcome, and that we should only select cases that are only members of one conjunction (Schneider and Rohlfing 2014). 2 However, when we take mechanisms seriously when selecting cases, neither recommendation is required for successful PT case studies. As mechanisms build on a different understanding of causal relationships (the mechanistic understanding), a cause that is present (i.e., over the crisp-set threshold) should trigger the causal mechanism irrespective of whether it is in the fuzzy-set subset or not, contingent on the requisite scope conditions for it to function being present in the case. Additionally, when using PT, we can achieve control for other causes at the level of within-case evidence (unique evidence) instead of achieving control through case selection.
The contrasting guidelines created by the different ontological assumptions about causality the two methods mean that we face trade-offs when engaging in research, forcing us to choose practices that are more in alignment with one or the other methods underlying ontological foundations. However, the implication of these contrasting guidelines when combining the two methods is not necessarily flawed inferences. If we follow case selection principles more in alignment with QCA, we would simply have a smaller set of cases to select from and thereafter make generalizing inferences to only the uniquely covered cases.
Where the contrasting guidelines become problematic is regarding the types of causal claims we are making after research. PT is used after QCA to provide within-case evidence of causal relationships. However, PT only enables us to make the inference that the studied causal is present or not in the case, but it does not tell us anything about whether the cause(s) was necessary or sufficient. To make claims about necessity or sufficiency, we need counterfactual variation (Woodward, 2003), which in case studies can be achieved by transforming a single case study into a form of comparative most-similar-systems design, where we hold everything else equal but the hypothesized necessary condition, and then speculate using logical arguments about whether the outcome would have been any different had the condition not been present (Goertz and Levy 2007). But this in effect transforms the single case into a comparative case study, comparing the existing with a hypothetical counterfactual case. In contrast, PT involves tracing whether there is actual within-case process-related evidence of a theorized mechanism actually operating as predicted in the chosen case.
At the same time, the two methods do compensate for the other’s weaknesses in ways that are not widely known in the literature, but that will be explored in this article. In particular, PT provides a helpful tool for making theoretical sense of the conjunctions produced in QCA of sufficiency. As discussed in the introduction of this special issue, when there are multiple terms in a sufficient conjunction, QCA as such does not help us figure out even simple questions such as whether conditions are causal or scope conditions, nor does it help us figure out whether the terms together produce a single mechanism, or whether they act in sequence or other more complex patterns. PT is useful here for two reasons. First, at the theoretical level, after a QCA of sufficiency has found a robust conjunction, the analyst has to flesh out a theoretical mechanism that links the conjunction with the outcome for us to be able to trace it empirically using PT. In this conceptualization process, the analyst can, for example, find that a condition that was theorized individually as being causal becomes a scope condition without causal powers when it is logically combined in a mechanism where it operates in conjunction with other causes. Second, when tracing the theorized mechanism in empirical PT case studies, the analyst might find that the mechanism did not work as expected, with conditions working in sequence instead, resulting in theoretical revisions that produce a more accurate causal theory.
This article proceeds in four steps, exploring the challenges and opportunities resulting from two different understandings of causation underlying QCA and PT. This article first presents the research area used to explore the practical challenges of combining the two methods. I investigate the causes of congruence between what voters want and government positions in European Union (EU) constitutional negotiations, part of the broader phenomenon of the representation of voter views in public policies. The section develops a range of potential causes for congruence from the existing literature. Second, using a QCA-first design, I undertake an fsQCA of sufficiency. Utilized in a theory-building fashion, I investigate whether the potential causes identified in the literature form conjunctions of conditions that are together sufficient to produce congruence between voter views and governmental positions. This article only finds one conjunction that is robust: the combination of proportional representative (PR) systems and the EU being a highly salient issue in domestic politics (electoral connection).
Third, this article engages in a PT case study of two positive cases of the electoral connection conjunction. At the theoretical level, gaming through a causal mechanism for the conjunction suggests that one of the two terms should better be understood as a scope instead of causal condition, providing a better theoretical understanding of the found conjunction. Issues of case selection are then discussed, finding that a restrictive policy in alignment with QCA tenets results in some promising potential cases being rejected. The actual case studies find some evidence for the presence of the hypothesized mechanism, although when we select a nonuniquely covered case where another causal condition is also present (referendum), there is some confirming evidence that suggests that the referendum condition hinders the working of the mechanism sparked by the electoral connection conjunction.
Finally, the conclusion discusses the methodological lessons learned in practice, focusing on the need to justify case selection in terms of whether one is more in alignment with either QCA or PT and the need to make conclusions that are consistent with the types of inferences made possible by PT case studies, that is, that through our tracing of a mechanism produced by a conjunction we find that it is either present or not in a case, but where we can make no claims made about its necessity or sufficiency.
The Research Field—Developing Candidate Causes for the Congruence Between Voter Views and Government Positions
The following develops a set of candidate causes that might explain congruence between voter views and governmental positions in EU constitutional politics. EU constitutional politics involves negotiations between governments on the transfer of sovereignty to the EU level. In this type of sensitive issue area, we should expect that voters will have incentives to keep their elected representatives in line with public opinion (Golder and Stramski 2010; Blais and Bodet 2006). As this is more a methodological than substantive article, the descriptions of causes and the outcome are however kept quite brief.
Before we turn to the presentation of potential causes of congruence, it must be noted that the common practice in QCA of sufficiency that the effects of causes are described individually but then “tested” collectively using QCA. This practice does not reflect the underlying assumptions made about conjunctional causation. As described by Ragin (2000, 2008), the underlying ontological principle of QCA is that it causes work in configurations or conjunctions. Therefore, it is problematic that most existing applications of QCA for sufficiency present each potential cause of an outcome individually without theorizing how they might work together in conjunction with each other (e.g., Samford 2010). Therefore, to ensure methodological alignment, we should think about potential conjunctions when presenting candidate causes for a QCA of sufficiency as causes might act very differently when operating in a conjunction instead of individually. However, in practice, it can be difficult to avoid theorizing individually but then testing collectively, given that our candidate causes are usually drawn from existing large-n, variable-based literatures that typically focus on the effects of individual causes. 3 To reflect this, I term the following application of QCA theory building, in that I am utilizing the method in a more exploratory fashion to detect conjunctions instead of testing for theorized patterns/conjunctions.
Candidate Causes
In broad terms, empirical research has found that public opinion has an impact on government policies (Schmitt and Thomassen 2000; Carruba 2001; Hooghe 2003; Hooghe and Marks 2009) 4 and that the presence of referendums and PR electoral systems have been found to be empirically correlated with closer congruence (Finke 2009; Hug and König 2002). Beyond these two potential causes, three more candidates are selected based on their relative prominence in the literature.
The following first discusses how the outcome to be explained (congruence) is conceptualized and measured, followed by a short presentation of the five most plausible causal conditions identified in the literature and how they were measured in fuzzy-set scores. 5 The full data set is reproduced in Table 1.
Summary Table of All Fuzzy Set Scores of the Outcome and Conditions.
Note: EU = European Union; IGC = Intergovernmental Conference.
Source: See Online Appendix.
While a crucial part of any fsQCA, it must be noted that the existing methodological guidelines for calibrating variables from existing interval scales into fuzzy-set scores do not go beyond the suggestion that theoretical and substantive knowledge should be used to anchor the qualitative thresholds of full membership, nonmembership, and the cutoff demarking set membership (Ragin 2000, 2008:86-94; Verkuilen 2005). This lack of concrete guidelines for calibration is a notable oversight that should be corrected in future work on QCA but is beyond the scope of this article. A discussion of the operationalization of the outcome and causal conditions and their calibration can be found in the Online Appendix.
Conceptualizing and Operationalizing the Outcome
Before we can test theories of congruence, it is important to have a valid measure of the outcome that captures the level of congruence between voter preferences and government positions on the issue of more/less EU integration. Existing measures either use poor indirect proxies to capture voter views 6 or include very technical institutional questions that voters cannot be expected to have an opinion on, such as whether the European Parliament’s (EP) role in the appointment of the Commission President should be strengthened. The revised measure captures voter opinion on the issue of whether they think policymaking in a given issue area is best dealt with in the national capital or in Bruxelles, which is a question that voters can be expected to hold relatively informed ideas.
The measure is constructed by comparing scores of the median voter with governmental positions. Using the same procedure as applied by Hooghe (2003) and by Wlezien (1995), the median voter position is inferred from dichotomous survey items that ask whether voters are in favor of transferring decision-making in a range of issue areas to the EU level. Governmental positions are measured using existing data sets for the relevant rounds of EU constitutional reform. The outcome condition (congruence) is then constructed by comparing scores of the median voter with governmental positions, resulting in what has been termed a “many-to-one” congruence measure by Golder and Stramski (2010). The Online Appendix details how each component is an improvement on existing measures along with how they are combined in what approximates roughly comparable policy space. While by no means perfect, the revised congruence measure is arguably closer to what we intend to measure than existing measures.
I now turn to a brief presentation of the conceptualization and operationalization of the potential causal conditions that could produce congruence based upon the existing literature. Note that in set-theoretic logic, the causes of noncongruence are not necessarily just the inverse of what produces congruence; causation is assumed to be asymmetric, and therefore the focus is on selecting candidate conditions that can potentially explain the positive outcome (congruence).
Condition 1—Level of Proportionality of the Political System
First, does the level of proportionality of a political system matter? This condition has been the main focus of comparative research that attempts to explain variations in congruence. However, there are two potentially opposing causal logics: one in which proportional representation (PR) produces congruence and one where majoritarian systems are better at representing the views of the median voter—both of which are theoretically plausible.
A number of scholars suggest that PR electoral systems produce closer congruence between voters and government policy than majoritarian systems. McDonald, Mendes, and Budge argue that PR systems are designed to “…reflect accurately vote proportions in the distribution of parliament seats, and with their comparatively large number of parties they should be able to bring electoral and parliamentary medians into alignment better than SMD [read majoritarian] systems” (2004:18). The large number of parties means that voters should be able to find a party that represents their views (Blais and Bodet 2006). The next step in the theory details how PR produces a government whose positions are congruent with voters. In the process of coalition formation in parliament, coalition theory predicts that the median party will play a dominant role in government formation (Huber and Powell 1994), resulting in a close congruence between the position of the median voter and government.
There are however also theoretical arguments, suggesting that majoritarian systems can produce closer congruence than PR systems (Blais and Bodet 2006; Golder and Stramski 2010). Duverger’s theory predicts two parties in majoritarian systems, with the resulting increased likelihood of a single-party majority government. Spatial theories of party competition suggest that in majoritarian systems, parties are expected to converge to centrist policy positions relative to the electorate. This causal logic suggests that majoritarian systems can produce governments that more closely reflect the median voter than PR systems.
Given the ambivalence of the theoretical and empirical literature, the causal effect of electoral systems is analyzed both as regards whether PR or majoritarian systems produce closer congruence between voters and governments. In practical terms, when engaging in the logical minimization process in the QCA, no assumptions about the direction of causality are made about this condition.
The type of electoral system is conceptualized here as the level of distortion of electoral rules between vote share and seat share (PR systems have a close correspondence between votes and seat shares, whereas majoritarian systems do not) and is measured using Gallagher’s index of electoral systems, recalibrated as described in Online Appendix.
Condition 2—The Salience of EU Integration in National Politics
The second causal condition investigated is the salience of the EU integration issue in national politics. Although it is often taken for granted in studies of representation that the issue investigated is salient, for political representatives to be responsive to public opinion, voters must be concerned about the issue enough to allow it to influence their vote choice in national elections (Burstein 2003; de Vries 2007; Kriesi 2007; Baumgarten and Jones 1993:22). When an issue does not attract electoral attention, the preferences of privileged elites and special interests dominate policymaking.
I measure salience with data from the 1999 European Election Study, where voters are asked directly about how important they perceived the EU issue to be, whereas in the 1994 and 2004 studies, voters were only asked to rate either the most important or three most important issues for them, yet EU integration is not a top three issue for most voters (Van der Eijk et al. 1999).
Condition 3—Elite Debate on the EU Issue
For voter preferences to be represented, many scholars contend that voters must have a real opportunity to choose between different EU policies in the positions advocated by political parties (Wessels 2001:152; Golder and Stramski 2010). The theoretical claim is that when there is a strong elite consensus on EU integration-related issues, elites will have few incentives to raise the EU issue in national elections (Steenbergen and Scott 2004; Ray 2003; Van der Eijk and Franklin 2004). In this situation, “If voters are not offered a choice between different visions of Europe, whatever the differences between voters, these cannot be expressed in their choices of parties to support” (Van der Eijk and Franklin 2004:39). Based upon this, we should therefore expect that when major parties in a political system offer different choices, this will produce closer congruence.
The elite debate condition is operationalized by measuring the views toward more/less integration of the three largest parties of each member state (defined according to the most recent election results). The data for measuring party positions are two different expert surveys on the positions of political parties on European integration (less-more dimension [Ray 1999] for the Amsterdam Treaty (AT) negotiations and the 2002 Chapel Hill expert survey for the Constitutional Treaty (CT) negotiations [Steenbergen and Marks 2007]).
Condition 4—Referendums
Does the use of referendums produce congruence between voters and governments? Several scholars contend that governments that are forced to consider the ultimate judgment of the treaty by the median voter are more in line with public opinion on EU integration (Hug and König 2002; Finke 2009; Christin and Hug 2002; Steenbergen, Edwards, and de Vries 2007). If party elites align their views with voter preferences, we should expect that these demands are transmitted further through the political system, resulting in national positions that more closely reflect voter views when policies have to be ratified in a referendum.
The scores for referendums are based upon information from Christin and Hug (2002) for the AT negotiations and Closa (2007) for the CT negotiations. Whether a country was scheduled to hold a referendum was coded dichotomously, with a score of full membership (1) given to countries scheduled to hold referendums and 0 to those that did not.
Condition 5—The Strength of the Executive vis-à-vis Parliament (EU Affairs Committees)
One of the key insights of principal–agent theory as it relates to the study of parliamentary systems is that we should expect that agent (executive) shirking should decrease when the principal (parliament) possesses powerful control mechanisms (Strøm 2000; Müller and Strøm 2000; Saalfeld 2000; Finke 2009). In parliamentary systems, once a government cabinet is appointed, it can be difficult for parliament to control the government and therefore strong committee oversight can potentially be an effective means to control the government (Strøm 2000; Müller and Strøm 2000; Saalfeld 2000; Finke 2009). Assuming that parliamentary views are in line with the median voter, we should expect that stronger control/scrutiny mechanisms available to a national parliament will produce closer congruence between public opinion and government positions.
The strength of the executive is operationalized in this article by examining whether a strong EU affairs committee exists or not. The data used to measure the strength of EU affairs committees is a Conference of Parliamentary Committees for European Affairs of Parliaments of the European Union (COSAC) report from 2005 that details the relative strength of committees in each member state. Kassim, Peters, and Wright (2000) were also consulted in order to ensure that there had been no major change in the mechanisms during the earlier AT negotiations. Countries that had some form of committee were coded as being in the set of EU affairs. Differences of degree, from full to partial membership, were coded based upon the strength of binding nature of the mandate given by the committee.
The Analytical Two Step: Combining fsQCA and PT
This article employs an fsQCA for sufficiency in a more theory-building fashion, followed by a PT case study of the most robust conjunction from the fsQCA. PT is used in a theory-testing manner, with the emphasis being a structured empirical test of whether there is evidence suggesting that a hypothesized causal mechanism exists between the found conjunction and the outcome. In the terminology of the introduction, this is a mechanism-centered type design that attempts to unpack mechanisms and trace them empirically in a case study.
As discussed in the introduction of the special issue, when we take mechanisms seriously, they should be treated as middle-range theories that are expected to be present in a population of cases when the scope conditions or context that trigger and/or allow them to operate are present. Causal mechanisms are defined as a theorized system that produces the outcome through the interaction of a series of parts that transmit causal forces from X to Y (Machamer 2004; Machamer, Darden, and Craver 2000). Instead of a minimalist understanding as used by Elster and others (Elster 1998), where causal mechanisms are not unpacked, a systems understanding attempts to unpack the causal process in between X and Y into its constituent parts. Together, the parts of the mechanism can be thought of as a coherent story that explains how a cause or set of causes can produce the outcome through a causal process. Ideally, there should be what can be termed productive continuity from one part of the mechanism to the next, transmitting causal forces through the mechanism without any theoretical gaps. As will be seen below, theorizing explicit mechanisms between causes and outcomes can result in finding that logically that what we thought was a cause is actually just a scope condition that has to be present for a relationship to hold. A scope condition does not “do” anything in a mechanism-based understanding of causation; it is merely a factor that has to be present for a relationship to work (Falleti and Lynch 2009). In contrast, the most interesting parts of mechanisms that we want to “trace” empirically in PT are the activities that are the “active” part of the process linking a cause and outcome.
Adherents of the systems understanding contend that parts of mechanisms do not collapse back down onto counterfactual claims (e.g., Waskan 2011). Instead, parts of mechanisms only make sense as parts of a whole causal process or system that together transmit causal forces from a cause to an outcome. While for analytical reasons the mechanism is typically theorized by splitting it into distinct parts, by making explicit the activities that transmit causal forces from one part to the next, the ensuing PT empirical analysis has a dynamic, processual element that attempts to capture what is going on in the causal arrows. In contrast, a counterfactual-based case study would engage in speculation about what might have happened if the part had not been present, comparing the actual with the hypothetical. However, here the empirical evidence is speculation, whereas when tracing mechanisms, the goal is to produce evidence of the causal process as it actually played out in a case.
Methodologically, when tracing mechanisms, the goal is to uncover how actual causal process played out in particular cases. This means that the evidence that is relevant to trace mechanisms is arguably also a different form of evidence than that which is relevant for assessing counterfactual-based claims, where evidence is the form of the “difference” that values of X have for values of Y (Russo and Williamson 2007; Illari 2011). In PT, we are using what can be termed “mechanistic” within-case evidence, which relates to observational data on the actual operation of parts of mechanisms within actual cases. Here, it is not the counterfactual “could things have been different” that acts as evidence but the actual process-related activities that we observe in the case study.
When PT is used in a theory-testing fashion, we know both X and Y and we are able to deduce a mechanism from existing theorization. Here, the conceptualization of causal mechanisms is a deductive exercise, where using logical reasoning, we focus our analytical attention on the most plausible causal mechanism 7 that explicates how X is causally linked with Y, along with the context for its proper functioning, defined as scope conditions for the mechanism. Naturally, after empirical research, we might revise the causal mechanism because we find evidence that the mechanism did not work as we expected, meaning that PT in practice is often a back-and-forth between formulating and reformulating mechanisms and empirical testing (Beach and Pedersen 2016a).
Being forced to conceptualize explicitly the mechanism whereby a conjunction produces an outcome helps us determine whether all of the conditions together produce a single mechanism, or whether they operate in sequence, or more complex patterns as discussed in the introductory article. After we have analyzed the mechanism(s) empirically, we gain further information about how the conjunction works in practice that enables us to update our theoretical framework to better capture how the conditions work in conjunction.
Before we turn to a presentation of the fsQCA results, it is important to flag two overlooked but important implications that the different ontological assumptions about causation in QCA and PT have for case selection practices. First, after an fsQCA for sufficiency, appropriate cases that match the types of causal claims being made are cases where the fuzzy-set score for membership in the conjunction is lower than membership in the outcome and that we should only select cases that are only members of one conjunction (Schneider and Rohlfing 2014). However, PT builds on a mechanistic understanding where a set threshold of a causal concept arguably plays a different role than for fuzzy-set QCA. In fuzzy sets, a score of 0.5 is one of maximum ambiguity; whereas for mechanisms, once the membership of the case is above the threshold, a causal relationship should be triggered by the condition, contingent on the requisite scope conditions also being present. This means that if the theory is correct, the mechanism will be triggered in all cases that are members of the causal condition(s). In cases that have case scores of the condition below the threshold, the cause(s) only have trivial or nonexisting causal effects. In other words, when building on a mechanistic understanding, set thresholds are the critical distinction for determining causation or not, whereas for fsQCA, it is full membership/nonmembership that is most important (Ragin 2008:90). In practice, there can be good reasons to select cases that are not close to the threshold because there can be calibration errors that might result in us thinking a case is in the set but in reality it is not (Beach and Pedersen 2016b). Therefore, we recommend that only cases that we are very confident in being in the set of X and Y (and the requisite scope conditions) are selected, meaning that we only select cases relatively far from thresholds to ensure robust results.
Furthermore, an additional reason to only focus on the crisp-set thresholds is that, as discussed in the introduction of this article, tracing a causal mechanism in a single case using PT does not tell us anything about whether a cause (or conjunction of causes) is necessary or sufficient in a case—this type of inference requires us to have counterfactual differences that do not exist within any given PT case. Given the inability of PT to analyze necessity or sufficiency, selecting a case from fsQCA of sufficiency above the fuzzy subset line is irrelevant.
The implication of the mechanistic understanding and set thresholds is therefore that when selecting cases, any case above the crisp-set threshold of the conjunction and outcome is a positive case that can be used for tracing mechanisms (X and Y are both >0.5; Beach and Pedersen 2016a; Goertz 2016). 8 This typically results in a larger number of candidate cases than when fsQCA guidelines are utilized.
Second, existing case selection techniques suggest that when studying mechanisms, we want to select cases where we have isolated the effects of individual causes. Gerring and Seawright write that “…researchers are well advised to focus on a case where the causal effect of one factor can be isolated from other potentially confounding factors” (2007:122). They term this type of case a “pathway case.” Schneider and Rohlfing draw on this guidance in their discussion of case selection for PT when they state that we should choose “unique set” cases, where we “…focus on one term…to unravel the mechanism through which it contributes to the outcome in the case under study” (2013:566-67).
However, when we conceptualize a mechanism in PT for a particular conjunction, we are in effect isolating the effects of these causes analytically by delineating the component parts of the mechanism it produces and how they operate together. When operationalized in the form of predictions about what observable manifestations each part of the mechanism should leave, we develop as unique empirical predictions as possible, meaning that they do not overlap with those of other theories. In Bayesian logic, uniqueness relates to the probability of finding the predicted evidence with any plausible alternative explanation for finding the evidence (Beach and Pedersen 2016a; Rohlfing 2012, 2014). When translated into PT of mechanisms, we make evidential predictions for each part of the mechanism and then evaluate their certainty and uniqueness of finding the evidence based on theoretical and empirical knowledge. This has nothing to do with whether alternative theories of the cause of the outcome are present or not, unless an alternative causal theory is theoretically and empirically mutually exclusive of the cause and mechanism that we are tracing (Rohlfing 2014; Beach and Pedersen 2016a). Of course, it also means that we cannot make claims about other causes of the outcome; only that we have confirming or disconfirming evidence of our theorized mechanism linking a given cause with an outcome.
Evaluating uniqueness is in effect a form of “control” for other causes but at the evidential level of individual fingerprints of parts of causal mechanisms in operation (Beach and Pedersen 2016a). This means that we do not need to control for other causes when selecting cases. This even holds in situations where there is potentially an overdetermined outcome, defined as a situation where multiple sufficient causes are present. The classic example of overdetermination from the philosophy of science can be used to illustrate why properly conceptualized and operationalized mechanisms can be empirically distinguished from each other (Brady 2008). The example deals with a person found dead after wandering in the desert. It is found that there was poison in the person’s canteen, but there was also a hole in the canteen. In this instance, both poison (X1) and thirst (X2) are potentially sufficient causes, but only one can actually have had a causal relationship with the outcome, that is, he died of either poisoning or dehydration. If we conceptualize and operationalize mechanisms linking X1 (poison) or X2 (thirst) with his death (Y), these two different mechanisms would be expected to leave very different empirical fingerprints that could be distinguished from each other. If a poison like cyanide was the cause, the mechanism could involve first drinking the poisoned water, followed by the poison being absorbed into the blood stream, and the final part being poison inducing heart failure that produces death. This mechanism linking poison and death is at the conceptual level very different from the causal mechanism linking dehydration with death and would also leave very different empirical fingerprints that we should be able to distinguish from each other (unless the vultures have gotten there first and removed all evidence). While there might be parts of the mechanism where evidence might be accountable by both causes (e.g., heart failure can be produced by both poison and dehydration), more detailed empirical expectations should then be developed that are more unique to one causal mechanism. For instance, heart failure by poison produces certain types of scar tissue in the heart that are quite unique to this mechanism of death. Therefore, if operationalized adequately, there would therefore be no observable equivalence in the traces left by the two competing mechanisms. By looking for the predicted empirical fingerprints of the two different theories, we would be able to determine whether the person died of poison or thirst, with the other potential sufficient cause relegated to being spurious. Controlling for other causes can therefore be done at the within-case, evidence level using PT.
Taken together, this means that when studying mechanism, we are (usually) able to isolate the workings of individual causes through the conceptualization and operationalization process, enabling us to analytically distinguish whether a particular mechanism at the evidential level was present irrespective of other causes also being present. The implication of this point is that we do not need to isolate causes through case selection in order to trace a mechanism linking it with Y. 9 Therefore, while the logic of controlling for other potential causal conditions is relevant when assessing the magnitude of mean causal effects of individual causes in cross-case comparisons that assess counterfactual causal claims, it is not relevant for the study of mechanisms within cases because we can analytically isolate the workings of particular mechanisms irrespective of the presence of other causes in the particular case by evaluating the uniqueness of evidence. In other words, we do not need to select “uniquely covered” or “pathway” cases as suggested by the literature. Naturally, there can be cases where we do not have access to relatively unique evidence, which means we cannot confirm with any confidence that the theorized causal process was operative in the case. But this should not lead us to attempt to compensate by “controlling” for other causes through case selection because this “control” at the level of case selection does nothing to increase our ability to confirm that a causal process was operative in a case as theorized because it is not within-case evidence of the actual operation of the parts of the mechanism.
Therefore, if we select cases after an fsQCA for sufficiency based on case selection guidelines that are in alignment with the mechanistic ontological assumptions that are appropriate for tracing mechanisms in PT, I contend that we can select any case within the set of the conjunction and the outcome, irrespective of whether other causes are present or not. This has the benefit of increasing the number of potential cases for tracing mechanisms, which is particularly welcome when we are operating with relatively small, bounded populations. Further, given the sensitivity of mechanisms to contextual conditions, selecting cases both that are uniquely covered and those that are not enables us to investigate whether it is the same theorized mechanism linking X and Y is present in both sets of cases.
fsQCA Results
Five causal conditions were investigated for sufficiency using fsQCA to see whether there were subset relationships with congruence. Table 2 illustrates the results of a truth-table analysis. In order to ensure that choices about consistency thresholds did not impact upon results, I conducted the analysis using both a more restrictive (0.85) and a more permissive choice of threshold level (0.75; for more on robustness checks, see Skaaning 2011). The results were not very different, although the conjunctions were simpler in the more permissive threshold model. As the goals of this article are more methodological than substantive, the simpler conjunctions found using the permissive threshold are utilized given that it is easier to reconceptualize a conjunction as a causal mechanism when there are fewer conditions in a given conjunction.
Different Causal Pathways to Congruence.
Note: • = causal condition (present); EU = European Union.
There was one main pathway, or conjunction of causal conditions, that produced congruence. Note that the electoral channel pathway uniquely covers several cases, whereas the referendum pathway overlaps in all but one case with the electoral channel. Both pathways two and three had very weak coverage. Note further that the results do not explain what factors produce noncongruence. 10 Ten cases in all were covered by the 3 conjunctions of the 14 cases of congruence.
The most important conjunction that produced congruence was an electoral channel pathway composed of the EU issue being salient in PR systems. High salience was expected to produce congruence, but the finding suggests that salience matters only when it is coupled with a PR electoral system. This is intuitively plausible, as we should not expect that an electoral system that produces more representative government (PR) is by itself enough to produce congruence in an issue. Voters need to be concerned enough about an issue that it can impact upon their calculations of which party to vote for or not for candidates to be responsive to public opinion.
PT Case Studies of Two Positive Cases
Is the electoral channel conjunction a causal or spurious relationship? The role of PT in a mechanism-centered design is to assess whether there is evidence of an actual causal relationship by tracing mechanisms empirically using detailed within-case analysis of its component parts. The following will first conceptualize and operationalize a theory test of a proposed mechanism linking the electoral connection conjunction with the outcome, followed by two theory-testing PT case studies of positive cases of the conjunction. In the population of the electoral conjunction, a false positive might result if only one case is selected, as the relationship might have been unique to the chosen case; whereas in other cases, it is spurious (situation 1). Second, one might also infer that because we found a given mechanism in two uniquely covered cases that it is the same mechanism linking cause and outcome in cases that are not uniquely covered (situation 2). However, given the sensitivity of mechanisms to context, we should ideally also repeat the PT of the mechanism in nonuniquely covered cases to see whether the presence of other causes matters for which mechanism (if any) link X and Y. We would therefore want to do at least two PT studies of the uniquely covered cases and then two of non-uniquely covered. But for practical reasons, only two case studies were undertaken; one of a uniquely covered case, the second of a nonuniquely covered.
A Theorized Electoral Channel Causal Mechanism
The conjunction whereby congruence can be produced operates through high issue salience as channeled through elections. The conjunction of conditions was PR and high issue salience. Figure 1 illustrates a plausible electoral channel mechanism that builds on existing theorization (see The Research Field—Developing Candidate Causes for the Congruence Between Voter Views and Government Positions section). The top part of the figure illustrates the theorized parts of the mechanism, while the bottom section depicts observable implications that can be used to test whether the mechanism was present or absent in a case. In the theorized mechanism, entities are underlined, whereas activities are in italics.
Electoral channel causal mechanism.
The four parts of the mechanism deal with the role of elections and issue salience in producing congruence. When I started theorizing a plausible mechanism for the electoral connection conjunction, it became clear that issue salience cannot logically be a cause in a mechanistic sense because it was difficult to see what it could actually do beyond acting as a scope condition that makes other causes relevant. In mechanism-related terms, a cause “triggers” a process, meaning that it actually “does something.” The distinction between a causal and scope condition is therefore vital when theorizing mechanisms, whereas this distinction is not important in QCA, given that in counterfactual-based claims, an absent necessary scope condition would have the same result as a missing necessary causal condition.
Viewed as a scope condition, unless an issue is salient, politicians have few incentives to take into consideration the views of voters because we should expect they will concentrate on achieving alignment on the most electorally salient issues. Instead, it is a form of background or scope condition that has to be present for a PR election to matter.
In the terms of the discussion in the introductory article of the special issue about the constellations of possible combinations of conjunctions, I therefore theorize that there is a sequential occurrence of the two individually necessary parts of an unncessary but sufficient cause (INUS) conditions in the form:
When the EU issue is salient in national elections, it is theorized that politicians will strive to not stray too far from what voters want in major questions like the transfer of national sovereignty to the EU (part 1 of the mechanism). Here, we should expect to find evidence that shows that the EU issue was debated in the most recent election campaign, but even stronger evidence of part 1 would be finding that voters cast their ballots on the basis of EU issues, captured in postelection surveys. For each empirical manifestation of parts of a mechanism, we need to evaluate whether they are unique and/or certain, which determines the types of inferences we can make when we find them/do not find them (Bennett 2014; Beach and Pedersen 2013). Recall that when evaluating uniqueness of an empirical prediction, we are discussing all plausible alternative explanations for finding the evidence itself instead of just one “competing” alternative explanation of the outcome.
First, finding that the EU issue was debated is a relatively certain empirical prediction, in that it is hard to argue that elections matter for congruence on EU issues if no one is discussing EU-related issues in an election. It is also relatively unique, for why would the issue be actively debated unless voters actually cared about the issue in the run-up to an election. As regards the fingerprint of postelection surveys, while there could be a degree of ambiguity in survey responses, the fingerprint is not certain as voters could feel that the EU issue is important but not the top issue determining who they vote for. Furthermore, voters might not recognize that an issue like “the economy” actually is significantly affected by the EU issue. The fingerprint would be, if found, relatively unique, for why would voters answer that the EU was important for their vote if they did not feel that it was?
The next three parts deal with the impact of the electoral system. Part 2 captures the first causal effects of PR, where the theorized better representation of the full range of voter views results in the election of parties representing the full range of public opinion. Empirical fingerprints of this part would be that voter views on the EU issue are consistent with party choice and that a range of parties reflecting the spread of voter views are elected to parliament. Both fingerprints are relatively certain, in that we have to find them for PR to matter, but they are not very unique. Just because there is a diversity of views expressed does not mean that it is PR that is causing them to be represented.
Parts 3 and 4 focus directly on the representation of the median voter’s views. Part 3 theorizes that in PR, we should expect that the median voter’s views will be represented by what can be termed the “median parliamentary party” (MPP). Part 4 postulates that in the resulting parliament—composed of many parties—the MPP should play a pivotal role in forming a governing coalition. We should expect to see empirically that the median voter’s attitudes toward the EU are reflected in the median party and that the subsequent government actually carries out its campaign promises in relation to EU positions. While both are certain predictions, they do not enable strong inferences about the link between voters and parties. Here, stronger evidence would be from internal deliberations within the MPP that provides evidence suggesting they were following the positions of the median voter (i.e., their core voters). However, while party records might shed light on this, it is almost impossible to gather this form of evidence. Participant interviews might be used, but it would be very difficult to determine whether the respondent is telling the truth or telling a political correct story along the lines “of course we listen closely to what our core voters want and represent these positions.”
The final outcome is congruence. Here, we should expect to see that congruence is highest in issues that have been debated extensively in elections and afterward in parliament.
Case Selection
For the PT theory test in the article, I have selected Portugal in the 1996 to 1997 (AT) negotiations and Ireland in the 2002 to 2004 (CT) negotiation as positive cases of the electoral channel mechanism. The scores for cases are illustrated for the electoral channel conjunction in Figure 2. 11 As discussed earlier, Schneider and Rolfing (2013, 2004) make the argument that after an fsQCA for sufficiency, we should only select as typical cases those above a secondary 45-degree line drawn within the set of conditions and outcome, where membership scores in the conjunction are greater that membership scores for the outcome. While each of the chosen cases in the following fulfills this requirement, this is not strictly necessary when engaging in PT. As discussed above, when we are interested in studying mechanisms, there should be causal mechanisms linking the conjunction with the outcome in all of the cases that have scores in the conjunction and outcome above 0.5, meaning that AUS96, IRL96, and NLR 96 would also be positive cases for PT, whereas they would be deviant cases in Schneider and Rohlfing’s terminology. Furthermore, because we can isolate the workings of individual causes at the level of evidence when tracing mechanisms, we do not need to take into consideration scores on other causes, meaning that while IRL96 and IRL03 also are members of another conjunction, they still are good positive cases if it is possible to develop empirically unique observable manifestations of parts of the mechanisms.
XY plot for case membership in the electoral connection conjunction.
In this article, I chose to select P96 as a uniquely covered positive case to make sure the mechanism works at all in the population of positive cases. In order to test the effects of the presence of other conjunctions, I chose IRL03 as a case that is a member of both the electoral connection and referendum conjunctions. I find that while the electoral connection worked in the P96 case, in the IRL03 case, the effects of the referendum conjunction shone through more strongly, suggesting that when both are present, the effects of referendums on congruence trump those of elections.
The Electoral Mechanism and Portugal in the 1996 to 1997 Negotiations?
Do we find evidence of the mechanism in the Portuguese case? 12 We should expect to find evidence that elections and voter concerns about EU issues impacted governmental position taking. If the first part of the theorized mechanism existed, we should find that EU issues were debated in the most recent parliamentary election prior to the IGC and that EU issues polled as being important for voters. There is however not much evidence that EU issues played a significant role in the October 1995 parliamentary elections, with the notable exception being the rhetoric of the populist People’s Party (PP, formerly the conservative CDS), which attempted to appeal to voters employed in declining economic sectors who were worried about the effects of future EU enlargements and more broadly about the impact of competition within the Single Market (Goldey 1997:247).
Do we find evidence of part 2, where we should see diverse views represented by a diversity of party positions on the EU? The election results and party positions on the EU issue are shown in Table 3.
The October 1995 Portuguese Parliamentary Elections Results and Party Positions.
Note: Party positions (PP) on seven-point scale, with 1 = strongly opposed to European integration and 7 = strong in favor. Party salience of European Union (EU) issue on five-point scale, with 1 = European integration no importance and 5 = European integration most important issue for party.
Source: Party positions and party salience of EU (Ray 1999).
Eurosceptic views on both the left and right were relatively well represented in the number of seats gained by the PP and the Left CDU parties (30 seats or 13% of seats). In the Ray expert survey (1999), the PP is coded as holding a quite negative view toward the EU (2.7), whereas the CDU is even more anti-EU (2.6). More moderate views held by a majority of the voters were well represented by the two major parties (PS and PSD).
Part 3 focuses more directly on the correspondence between the views of the median voter and the MPP. Looking at the election, we see that the two main moderate parties dominated the results, in particular the PS which was four seats short of having a parliamentary majority. Both parties held more pragmatic, pro-EU views, neither favoring federalism but only moderate increases in integration (Table 3; Marinho 2002). The views held by the two main parties match nicely with broad trends in Portuguese public opinion toward the EU. In the Eurobarometer conducted at the start of the IGC in 1996, 77% of Portuguese were polled as being in favor of the EU (European Commission 1996). 45% were polled as viewing the EU as a “good thing,” whereas only 10% stated it was a “bad thing” (European Commission 1996). However, the moderate and pragmatic views of a majority of Portuguese voters are better captured by responses on questions regarding whether they favored more integration. For example, there was a relatively low level of polled voters who favored a EU government, where only 41% were in favor and 38% were against in comparison to the EU15 average (54% in favor, 31% against; European Commission 1996). However, while providing some evidence for the existence of part 3, it does not enable us to discern whether representation of views is actually occurring as theorized in the mechanism.
There is some evidence for Part 4. The previous government (PSD) had taken moderate positions in the preparatory discussions prior to the IGC; positions that were continued after PS formed a new government after the election, with the notable and not surprising exception being that PS supported introducing new EU-level provisions on employment, echoing the sociodemographic profile of its voters (Marinho 2002). This shift in position suggests an impact of the views of the median voter as represented by the new government.
Finally, was there closest congruence in issues that voters were most concerned about? This was difficult to assess due to a lack of information. One piece of anecdotal evidence suggests that PS’ prioritization of employment matched the priorities of voters. When asked by Eurobarometer what issues they felt were most pressing to be dealt with by the IGC, employment and the fight against crime were the two highest priorities (European Commission 1996).
Concluding, there is some evidence that the mechanism as theorized actually was present, making us somewhat more confident in the presence of a causal relationship exists in the case than before we engaged in the tests. However, the material available enabled only quite weak confirming inferences at best for some of the parts (especially part 1), meaning that some degree of caution is warranted in the strength of causal claims about the presence of the mechanism we make in the case.
The Electoral Mechanism and Ireland in the 2002 to 2004 CT Negotiations?
Finding a causal mechanism between the electoral connection conjunction and congruence in the Portuguese case does not necessarily mean that the same causal mechanism links X and Y in the other cases that are members of the conjunction and outcome (positive cases) nor does it enable us to infer that the same mechanism linked X and Y when other causes were present. Due to potential equifinality at the level of mechanisms, there might be multiple causal paths between X and Y due to contextual factors. To enable stronger claims about the presence of mechanisms across the population of the electoral connection conjunction and to ensure that other causes do not dominate it when they also are present, I engage in a second theory test using another positive case that is not uniquely covered by the electoral connection conjunction. Both causes in theory could potentially operate at the same time either in synergy through one mechanism or in parallel (see the introduction for more on different possibilities). However, I find evidence that suggests that the operation of the referendum cause resulted in the crucial part 4 of the electoral connection not working as theorized. By selecting a nonuniquely covered case, I was able to detect empirically how the two mechanisms impact on each other, where one dominated the other. This highlights an overlooked methodological benefit of tracing mechanisms in nonuniquely covered cases, enabling us to discern how multiple causes work together in particular cases—in this case, one was dominated by the other.
Do we find evidence in the Irish case of voter views being translated into governmental positions? The case deals with congruence during the negotiation of the Constitutional Treaty from February 2002 until June 2004. In Ireland, a parliamentary election was held in May 2002, only several months after the EU started the negotiations of the new treaty. Yet as I will develop below, there is more evidence suggesting that while the government was responsive to voter views, this was primarily due to the looming ghosts of past and future referendums. In the Irish case, a referendum had been held in June 2001 in which a majority of Irish voters had rejected the outcome of the previous round of negotiations (the Treaty of Nice), although in the context of very low turnout (Gilland 2002; Hobolt 2009:186-91; Garry, Marsh, and Sinnott 2005). In a second referendum held in October 2002, a majority approved the Treaty of Nice because they were cognizant that they were not just voting for the Treaty of Nice but instead about whether Ireland should remain in the EU or not (Hobolt 2009). Therefore, in light of the failed referendum, the Irish government was very aware in 2002 and 2003 that during the new negotiations, it would have to champion positions that would enable the new treaty to be ratified in a coming referendum (Vergés Bausili 2004).
First, while there is some evidence supporting part I of the electoral channel mechanism, with the backdrop of the failed ratification of the Treaty of Nice 11 months earlier, we should expect to see that the EU issue was salient for voters, with parties debating the issue and voters perceiving the issue as important in the parliamentary elections. Here, there is mixed evidence. Voters did feel that the EU issue was moderately important (average score of 2.06 [SD 0.78] on a 1–4 scale, with 1 being very important) in the 2002 Irish National Election Survey.
Yet the EU was not a salient issue in the actual electoral campaign, with the most important debated issues being the economy and political corruption (Garry 2004:156-57; Gilland 2002). Indeed, following the failed referendum in June 2001, one would expect the European issue to be very prominent in the discussions, but it was the “dog that did not bark” in the 2002 election campaign (Gilland 2002; Laffan and O’Mahony 2008:83-84). An explanation of the lack of salience of the EU in the election can be that as the outgoing government had promised another referendum, many felt that EU-related issues were better discussed in the coming referendum and not in the parliamentary election (Gilland 2002; Laffan and O’Mahony 2008:83-84).
Turning to part 2, we should expect to see diverse views represented by parties elected. Table 4 shows that the average score of voters across parties on the question of more/less integration was relatively pragmatic, favoring some further integration but only in a limited fashion. In the Eurobarometer survey for spring 2002, 78% of Irish respondents replied that membership of the EU is a “good thing” (European Commission 2002, p. 22). Irish respondents are below the EU average when asked whether they favor an EU constitution, with 55% favoring it in Ireland in comparison to 63% EU average (European Commission 2002, p. 67). When we compare voter views with the party positions, we find that for the two major parties (Fianna Fail and Fine Gael) held relatively pro-EU views that overlap with what voters wanted. In addition, when we look at the electoral manifestos of the two major parties, we find further evidence that Fianna Fail was relatively pragmatic on EU issues, opposing moves to renationalize or make the EU more intergovernmental and in favor of keeping the blocking minority option in decision-making (Fianna Fail, 2002). The Fine Gael’s manifesto was more pro-integration, in particular favoring directly electing the Commission and strengthening EU competences (Fine Gael 2002). In relation to Fine Gael, we therefore see a slight disconnect between the polled views of their voters in the May 2002 election and the more pro-integrative positions in the election manifesto.
Party Positions, Voter Preferences, and Vote Shares in the 2002 Irish Parliamentary Elections.
Note: Party positions on a seven-point scale, with 1 = strongly opposed and 7 = strongly in favor of European integration. Salience for party on four-point scale, with 1 = European integration no importance and 4 = great importance. Voter views on a 10-point scale, with 0 = too far, 10 = push integration further.
Source: party positions and salience = Chapel Hill 2002 expert survey and average voter = Irish National Election Survey 2002.
Turning to part 3, more skeptical attitudes were not well represented in the parliament elected in 2002, although their vote share increased (Gilland 2002), with the Greens and Sinn Fein only receiving 11 seats of the 166 total seats despite securing 11.3% of the votes. This can partially be explained by the way in which PR functions in Ireland, with voters selecting multiple candidates in prioritized order that then determine which party wins a seat.
While voters decided whom to vote for primarily based on their positive evaluation of the economy (Garry 2004:158), the victory of Fianna Fail ensured the return of a pragmatic, pro-EU government to power that broadly reflected the views of Irish voters. The key question then is whether the elected government actually implemented its campaign promises, as expected in part IV. When we compare the electoral manifesto with the policies adopted in the Convention and later IGC, we see that Irish government followed a very cautious line in the negotiations, being opposed to increasing EU competences in sensitive areas like foreign policy and justice and home affairs (Vergés Bausili 2004:6). 13 Yet, there is evidence that this cautious line was more due to the concerns about upsetting the coming referendum in October 2002 and thereafter about securing ratification in another referendum of the new treaty being negotiated (Vergés Bausili 2004:6). This suggests the conclusion that while the conditions for the electoral connection conjunction to function were present, there is more evidence that suggests that congruence was produced by governmental fears of holding referendums instead of the impact of voter views through the electoral connection. In other words, based on the available evidence, causal priority should be assigned to the referendum condition instead of the electoral connection conjunction because key parts of the electoral connection mechanism did not function when the referendum condition was also present.
Conclusions
The two PT case studies provided contrasting evidence of the importance of the mechanism linking the electoral connection conjunction with congruence. In the P96 case, there was some evidence that the mechanism worked as theorized. In the IRL03 case, when another conjunction was present (referendum), there was evidence that the electoral connection mechanism was dominated by the mechanism spawned by the need to hold a referendum. Whether this same result holds in other nonuniquely covered cases should then be investigated, especially as the history of the use of referendums in Ireland would suggest that this might be a relatively particular context, and that in other cases where both are present, the electoral conjunction might play a more significant role.
The inferences enabled by the PT case studies are therefore the following. First, when only the electoral connection was present, there is some evidence that it was linked with congruence through the theorized mechanism. However, the within-case analysis of the P96 case does not enable us to conclude that the causal conditions were either necessary or sufficient—to do this we would need to engage in some form of comparative, counterfactual analysis that would transform the within-case into a cross-case analysis either by analyzing two cases in parallel using a most-similar-systems design or engaging in counterfactual analysis (Beach and Pedersen 2016a). Second, when the referendum conjunction was present, it trumped the effects of the electoral connection, suggesting that it is important to investigate nonuniquely covered positive cases to investigate how different causes work together.
Additionally, given the potential of equifinality at the level of mechanisms, we should be very cautious about making cross-case inferences about causal relationships to all eight of the cases that are members of the electoral connection because there can be omitted causal or scope conditions that impact on the functioning of the mechanism, as we saw with the referendum causal condition.
Conclusions
This article explored the challenges involved in employing an analytical two-step combination of QCA and PT when we take seriously their underlying ontological assumptions regarding causality. QCA builds on a counterfactual understanding, whereas PT views causation in mechanism-based terms. If we attempt to keep our methodological guidelines in alignment with the underlying assumptions about causation, the result can be conflicting recommendations. In the following, I briefly review the three primary methodological lessons learned when the two methods are utilized together in a QCA-first two-step analysis.
The first lesson regards case selection. If we adopt case selection guidelines that are in alignment with QCA, we end up with more restrictive selection principles that only enable inferences to cases that are uniquely covered by a given conjunction. In this article, I explored the benefits of adopting guidelines more in line with PT, where having selected one positive case that was not uniquely covered by the electoral connection conjunction I found that when the referendum conjunction was also present that there was evidence that key parts of the electoral connection mechanism were impacted by the presence of the referendum condition, resulting in the mechanism not working as theorized in the Irish case. Therefore, studying nonuniquely covered cases using PT enables us to probe (1) whether one cause is dominated by another, or whether they work together in some fashion, and (2) whether mechanisms linking the causes work in parallel, together, or even against each other, as discussed in the introductory article (Beach and Rohlfing 2016).
Second, it is important to only make the types of inferences about causality that are in alignment with the underlying ontological understanding of causation of a method. Whereas QCA for sufficiency enables claims about potential sufficiency of a cause or conjunction of causes (it is still potential, given that QCA only analyzes associations but not actual causation), PT only enables us to make within-case inferences about causation based on tracing mechanisms. Indeed, based on a PT analysis, we are only able to claim that we have confirming or disconfirming evidence of the operation of the theorized causal mechanism. There can even be multiple different mechanisms that have different effects on the outcome sparked by the same cause (Illari 2011). This means that it is important for us to be very careful in specifying that PT only provides us with evidence of the workings of the mechanism we are tracing, not other mechanisms.
PT also does not enable claims about necessity nor sufficiency given the lack of counterfactual variation at the within-case level. Therefore, even after we have done a PT case study after a QCA for sufficiency, based on the PT results, we cannot claim that a given conjunction is actually a sufficient cause in the case; only that we have within-case mechanistic evidence of a mechanism linking the cause to the outcome. To claim sufficiency would require that we engage in a counterfactual thought experiment, where we logically assess a hypothetical counterfactual where the cause and mechanism were not present. But this only provides speculative evidence instead of providing actual within-case evidence of how causal processes played out in a selected case.
Further, PT only enables within-case inferences about causation, and given the risk of equifinality at level of mechanisms, we have to be very cautious about making cross-case inferences based on PT findings unless we have assessed a number of cases. In my analysis, given that I only found evidence of the conjunction in one uniquely covered positive case, I would not make a strong generalizing inferences that the electoral connection was a causal relationship in the other four uniquely covered positive cases because there might be unknown contextual factors that result in different mechanisms linking the same cause with outcome in the other cases, nor would I infer to nonuniquely covered cases.
Third, the use of PT after a QCA for sufficiency illustrated that PT is a useful adjunct to QCA for reasons that have not been widely known. I showed that by logically gaming through mechanisms linking a conjunction and an outcome and then tracing them empirically, we develop a better understanding of how a conjunction produces an outcome. In particular, forcing ourselves to “game through” each step of the causal process whereby a set of causal conditions can be linked to an outcome forces us to focus our attention on whether each of the conditions actually are causal in the sense that they “do something” or whether they are merely scope conditions that have to be present for a mechanism to function. In my analysis, I found that one condition was better thought of as a scope condition (issue salience) that has to be present before elections can produce congruence. In terms of a sequence of conditions discussed in the introduction of the special issue, it is prior to the occurrence of an election.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: from the Danish Social Science Research Council, Grant number 275-05-0236.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
