Actor-level drivers of the devil shift

Regarding individual determinants, it is essential to understand how political actors directly interact and have the possibility (or not) to build trust relationships among each other. Well beyond the ACF, theories of the policy process claim that different types of political actors behave in different ways in policy processes (i.e. state actors, political parties, interest groups, etc., Nohrstedt and Ingold Reference Moe2011). Political parties compete with each other in political decision making not only for policy reasons – i.e. because they wish to influence the policy outputs – but also for electoral reasons – i.e. because they wish to show to their constituency that they defend their interests (Besley and Case Reference Besley and Case1995; Stokman and Zeggelink Reference Stokman and Zeggelink1996; Strolovitch Reference Strolovitch2006). Similarly, interest groups follow a “logic of membership” (Schmitter and Streeck Reference Schmitter and Streeck1999) – i.e. they defend the policy beliefs of their members and compete with other interest groups for new resources or even for organisational survival. Therefore, both political parties and interest groups defend well-defined values and policy positions, and both are exposed to blameshifting by the general public and/or by their own supporters. In addition, both might want to exaggerate the power of and conflict with opponents in order to promote internal cohesion and rally their members to political action (Sabatier et al. Reference Sabatier, Hunter and McLaughlin1987).

By contrast, state actors and administrative organisations (i.e. public administrations and independent regulatory agencies) will usually advocate more moderate positions and play a more “neutral” role in policy making (Jenkins-Smith and Sabatier Reference Ingold and Fischer1994, 187). Even if they can be part of advocacy coalitions (Sabatier Reference Sabatier1987), state actors are frequently able to connect different opposing coalitions, engage in compromise finding and act as “policy brokers” within a policy subsystem (Ingold and Varone Reference Ingold2012). Because they generally advocate for more moderate positions, state actors are also less likely to experience important defeats (Sabatier et al. Reference Sabatier, Hunter and McLaughlin1987). In sum, the different roles political parties and interest groups play, as compared with state actors, are likely to affect how they are perceived in policy processes. From that we derive our first hypothesis as follows:

Independently from the formal type of policy actor, we assume that actors’ power also has an influence on the devil shift. Many policy-making processes are characterised by a broad inclusion of state and non-state actors trying to influence policy outputs by forming advocacy coalitions (Sabatier Reference Sabatier1987; Stokman and Zeggelink Reference Stokman and Zeggelink1996; Ansell and Gash Reference Ansell and Gash2008). Not all actors within a coalition play the same role, but some actors occupy a more central role within a coalition, whereas others are more peripheral and less crucial representatives of the coalition (Hojnacki Reference Henry1998; Beyers and Braun Reference Beyers and Braun2014). Indeed, leaders and policy entrepreneurs are an important resource to a coalition (Mintrom and Vergari Reference Maas and Hox1996; Sabatier and Weible Reference Sabatier and Weible2007). More powerful actors are, thus, more visible in the policy process in general and within their coalition in particular. Given this, they should be especially affected by the devil shift, meaning that they are especially likely to appear as more powerful and more evil than they actually are. Powerless actors, by contrast, should hardly suffer from the devil shift. They are by definition less important within a coalition, and their influence on the policy process, therefore, remains largely unnoticed. This strongly reduces the risk of misperception in terms of the devil shift. We, therefore, formulate our second hypothesis as follows:

Process-level drivers of the devil shift

As former studies have shown (Weible et al. Reference Weible, Siddiki and Pierce2011a), the devil shift is not only an individual phenomenon. Characteristics of the policy process also affect the overall perception of power and conflict. First, the behaviour of actors and coalitions in policy processes depends on institutional opportunity structures (Baumgartner and Jones Reference Baumgartner and Jones1991; Ostrom Reference Ostrom2005; Sabatier and Weible Reference Sabatier and Weible2007; Leifeld and Schneider Reference Leach and Sabatier2012; Fischer Reference Doreian and Mrvar2014). Actors may exploit different institutional venues in order to defend their policy beliefs and reshape the policy conflict. Venue shopping in such a context refers to “the activities of political actors that seek out a decision setting where they can air their grievances with current policy and present alternative policy proposals” (Pralle Reference Pralle2006, 26). If a policy actor is prevented from participating in a specific venue (e.g. has no access to government’s consultation or to parliamentary hearings), or if the decision rules regulating this venue are biased in favour of their opponents, then this actor may opt for an alternative policy venue (e.g. launching a popular initiative, engaging in litigation). Shopping across alternative venues is, thus, a tactical option for policy losers, disadvantaged groups or outsiders following a “conflict expansion” strategy (Schattschneider Reference Schattschneider1960). This conflict expansion strategy does not only imply to shift venues but also to expand the problem boundaries, to adopt a new frame of the policy solution and, last but not least, to “label opponents as enemies” and “encourage conflict and [the] appearance of it” (Pralle Reference Pralle2006, 16). By contrast, if policy actors engage in a conflict containment strategy or even in “quiet politics” (Culpepper Reference Culpepper2011), they meet in the same institutional venue, play repeated games and develop stable patterns of interaction. In such a collaborative policy process, trust relationships are supported by the absence of venue shopping. For instance, Leach and Sabatier (Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2005) demonstrate that interpersonal trust increases significantly if alternative decision-making venues are not activated by policy actors. Our third hypothesis, thus, reads as follows:

Second, the type of conflict, influenced by the type of policy at stake, may also play a role. In Switzerland, as in most western countries, the traditional Right–Left opposition is still the most important conflict line among the political elite (Kriesi et al. Reference Kriesi, Adam and Jochum2006b; Sciarini Reference Sciarini2014). Whereas right-wing parties as well as employers’ associations and economic interest groups defend free markets and deregulation, left-wing parties and trade unions defend more state intervention. We expect the devil shift to be stronger when actors deal with socioeconomic issues than if a process is about other issues. First, on traditional issues dealing with social or economic policy, conflict lines and coalition boundaries are well established and entrenched. Coalitions have been defending opposite basic “world views” or ideologies for decades, and are, therefore, trapped in the vicious cycle of the devil shift – i.e. the mutual reinforcement of conflict and misperception (Sabatier et al. Reference Sabatier, Hunter and McLaughlin1987). In policy processes raising newer issues and conflicts, goals and resources of the adversaries are not clearly identified yet, and thus coalition boundaries are not well established. Second, socioeconomic conflicts directly tackle actors’ material interests as they deal with the redistribution of resources between groups of actors. Other more recent conflicts deal with regulatory issues or issues related to the internationalisation of domestic politics or the openness towards foreigners. Such issues are more based on values and have less imminent material consequences for actors (Lowi Reference Lijphart1964). They are, therefore, less clearly linked to a well-defined ideological camp, which lowers the risk of the devil shift.

Nine policy-making processes

The empirical analysis is based on data on the nine most important policy processes in Switzerland between 2001 and 2006. The choice of the most important policy processes is based on an extensive expert survey among experts of Swiss politics.Footnote ¹ These nine processes include the following: the law on nuclear energy dealing with nuclear energy plants and the treatment of nuclear waste; the law on telecommunication dealing with the transition from a monopoly to a market situation within the Swiss telecommunication industry; the new law on foreigners restricting immigration from countries outside the European Union (EU) to specialists and providing integration measures; the 11th pension reform aiming at adapting the public pension scheme to new economic circumstances; the reform of fiscal equalisation and tasks distribution, representing a major reform of the competencies and financial flows in Swiss federalism; the law on the infrastructure fund establishing a fund for the financing of urgent road and railway projects; the programme of budget relief 2003 reducing public spending in times of crisis; the extension of the bilateral agreement with the EU on the free movement of persons to the 10 new EU member states; and the bilateral agreement with the EU regarding Switzerland’s participation in the Schengen/Dublin agreements.

Based on the classic combination of positional, decisional and reputational approaches (e.g. Knoke Reference Knill and Tosun1993, 30), we identified the main collective actors participating in each process (political parties, interest groups, government and administrative agencies, subnational actors and scientific institutions). Following the decisional approach, we identified the actors that took part in the different venues (e.g. expert committees, consultation procedure, parliamentary committees, etc.) of the policy processes. To this list, we added the actors holding an overall strategic position in the Swiss political system (positional approach). Finally, we checked during the first interviews conducted with the administrative actors responsible for the policy process to ensure that no powerful actor was missing (reputational approach). Data regarding reputational power, collaboration, convergence/divergence relations and issue satisfaction were gathered through 212 face-to-face interviews with individual representatives of these collective actors.Footnote ² We made sure to select the individual within an organisation best able to respond in the name of the organisation with respect to a given issue. Contrary to mail or online surveys, face-to-face interviews are supposed to result in more valid and reliable data, as the interviewer is able to immediately respond to questions from the interview partner. The interview questionnaire consisted mainly of closed questions.

Research design and operationalisation

The devil shift argument states that actors from different coalitions are likely to misperceive the power and beliefs of their opponents. Thus, to begin with, we need to define which actors belong to the same or to opposing coalitions, respectively.

As mentioned in the theroretical part of this paper, actors in a coalition do have similar beliefs and coordinate their activities (Schlager Reference Schlager1995; Knoke et al. Reference Knoke1996; Weible and Sabatier Reference Weible and Sabatier2005; Henry Reference Heaney2011). To take into account both elements, coalitions are identified in two steps. In a first step, actors are grouped together according to their profile of belief convergence and divergence. Based on a list comprising all actors participating in the process, interview partners were asked to select the actors with whom their organisation had convergent (coded as 1) or divergent (coded as −1) views regarding the policy issue at stake.Footnote ³ These subjective perceptions of actors’ convergence with other actors have been shown to be a good proxy for similar beliefs (Ingold Reference Howlett, Ramesh and Perl2011). Groups are then identified with the “balance” procedure in the computer programme Pajek (Batagelj and Mrvar Reference Batagelj and Mrvar1996). This algorithm continuously rearranges the matrix of actors until reaching an arrangement that is closest to a predefined structure with only positive within-group ties and negative between-group ties (de Nooy et al. Reference Nohrstedt and Ingold2005). Deviations from this ideal arrangement are indicated with an error term, which can be higher or lower depending on the number of groups defined by the researcher (Doreian and Mrvar Reference de Nooy, Mrvar and Batagelj2009). We selected the number of groups with the lowest error term – i.e. the number of groups that created the group structure closest to an ideal arrangement. Actors within one group are allies, whereas actors from different groups are opponents.

Because simply perceiving another actor as an ally is not enough for actors to form a coalition, the second step identifies actors within each group who cooperate at least indirectly. To that end, we again rely on our interviews with political actors. More specifically, we asked our interview partners to select actors with whom their organisation cooperated closely during the policy process at stake from the same list of actors mentioned above.Footnote ⁴ Based on the symmetrised cooperation networkFootnote ⁵ among actors within each group, we identified all cliques of actors that are cooperating with each other at least indirectly (i.e. two-cliques, Wasserman and Faust Reference Wasserman and Faust1994) with the computer program UCINET (Borgatti et al. Reference Borgatti, Everett and Freeman2002). This allowed eliminating actors that have similar beliefs as other actors in the group but that are not well integrated in the cooperation structure within the group.Footnote ⁶

Thus, actors sharing similar beliefs (step 1) and cooperating with each other at least indirectly (step 2) form a coalition.Footnote ⁷ This procedure leaves us with two to four coalitions per process. From the original 212 actors in the nine networks, 142 were considered members of a coalition, and are, therefore, included in the subsequent analysis.Footnote ⁸

The opponents appear more powerful than they actually are

The first argument of the devil shift states that actors see their opponents as more powerful than they really are. This implies that the devil shift exists if the power of actor b from coalition B is overestimated by actor a from coalition A. Formally, this can be expressed as follows:

(1) $${\rm Devil\,Shift\,if}\, {Pp_{{_{{(a \to b)}} }} \gt \mathop{{Pr_{b}}} } $$

In expression (1), Pp stands for perceived power and Pr represents real power. A test of this argument, thus, needs to compare two variables – one measuring the perception of the power of actor b by actor a $$\big( {P_{{p_{{_{{(a \to b)}} }} }} } \big)$$ and one measuring actor b’s real power $$(Pr_b)$$ . If the first is bigger than the second, there is an overestimation of the power of the opponent, which indicates a devil shift. We measure variable $$\left {P_{{p_{{_{{(a \to b)}} }} }} } \right$$ as the attribution of reputational power from actor a to actor b. In line with the theories of the policy process (Sabatier Reference Sabatier2007), the power of an actor is operationalised by its influence on the policy process at hand. Data stem from interviews. Based on the list of actors participating in a given policy process, interview partners were asked to indicate which actors had, in their view, been very influential in that policy process.Footnote ⁹ This results in a binary matrix of reputation attribution, where 0 expresses that actor a does not consider actor b as powerful, and 1 means that actor a does consider actor b as powerful.Footnote ¹⁰ Actor b’s real power (Pr) is approximated by calculating its average reputational power. This variable varies between 0 and 1 and represents the percentage of interview partners who considered actor b as powerful. The reputational measure for assessing the power of political actors has proved to be a valid and encompassing measure of actors’ power in studies on public policy, policy networks and political decision making (e.g. Knoke et al. Reference Knoke1996; Kriesi et al. Reference Kriesi2006a; Fischer et al. Reference Fischer2009; Henry Reference Heaney2011).Footnote ¹¹

As an example for the construction of this variable, we take the case of the Radical Democratic Party in the process leading to the pension reform. The average reputational power of this party, as estimated by all actors, is 0.87 (Pr). However, actors from the opposing coalition (the Left) estimated the power of the Radical Democratic Party to be 1.00 on average (Pp).Footnote ¹² There is, thus, an overestimation of power – i.e. a devil shift effect of 0.13 for this actor. Because the perceived power is larger than the real power, opponents evaluate this actor as more powerful than it really is.

The opponents appear more evil than they actually are

The second part of the devil shift argument states that actors perceive their opponents as more evil than they actually are – i.e. they see the dissimilarity between their own policy beliefs and the ones of their opponents as bigger than it really is. More concretely, the devil shift exists if actor b from coalition B perceives its policy beliefs to be more distant to the beliefs of actor a from coalition A than they really are. Formally, this can be expressed as follows:

(2) $${\rm Devil\ Shift\ if}\, {BDp_{{(a \to b)}} \gt BDr_{{(a \to b)}} } $$

In expression (2), BDp stands for perceived belief dissimilarity and BDr represents real belief dissimilarity. In order to test this argument, we again need to compare two variables – i.e. one representing perceived belief dissimilarity $BDp_{{(a \to b)}} $ and one representing real belief dissimilarity $BDr_{{(a \to b)}} $ . If the first is bigger than the second, there is an overestimation of the belief dissimilarity with the opponent, which indicates a devil shift. The measure of perceived belief dissimilarity is based on the same question that we used to identify coalitions (see above) – i.e. the question asking our interview partners to select those collective actors with whom their organisation had converging or diverging beliefs about the policy issue at stake. This information is represented as a matrix, with the value of 1 expressing perceived belief dissimilarity and the value of 0 representing perceived belief similarity.

The measure for real belief dissimilarity also stems from our interview data, but is based on another, arguably more objective, measure: we asked our interview partners to indicate how satisfied they were with the solution to a number of issues raised during the policy process at stake (four to seven issues, depending on the policy process). Actors could indicate whether they were fully satisfied (1), rather satisfied (2), rather unsatisfied (3) or not satisfied at all (4). If two actors were satisfied to the same degree with an output, we assumed that their policy beliefs were similar. If, on the contrary, two actors differed with respect to their degree of satisfaction with the output, they arguably did not have the same policy beliefs. Based on actors’ satisfaction scores, we calculated Manhattan distancesFootnote ¹³ between them and standardised this value by the theoretical maximum distance. This results in a matrix of actors’ real belief dissimilarities based on their satisfaction with the solution to a given issue. Values of 1 express maximum belief dissimilarity, whereas 0 stands for maximum belief similarity.

As an example, we again take the case of the Radical Democratic Party in the process of pension reform. Based on issue satisfaction, the dissimilarity score between the opponents (the Left) and the Radical Democratic Party is 0.45 (BDr). However, the Left, on average,Footnote ¹⁴ perceives belief dissimilarity to be 0.89 (BDp). Thus, there is a difference of 0.44 corresponding to the evil dimension of the devil shift for this actor. Because the perceived belief dissimilarity is bigger than the real belief dissimilarity, opponents evaluate this actor as more evil than it really is.

Independent variables

Our theoretical arguments imply that important drivers of the devil shift can be located at the level of the actor as well as the process. Actor-level variables are measured as follows: first, actor types are measured by simple dummy variables indicating whether an actor is a political party, an interest group or a state actor (i.e. government and its public administration units). In the regression models, state actors serve as the reference category, together with a residual group of cantons (i.e. subnational entities) and scientific experts. Second, the models include a reputation variable measuring the extent of actors’ power. This variable also stems from interviews and has been explained above. It indicates the percentage of interview partners who considered an actor powerful.

Process-level variables are listed in Table 1. First, as an indicator of the degree of venue shopping, we calculate the extent to which non-state actors participated in the preparatory phase of the legislative process. The preparatory phase is the most decisive one in Switzerland, and it comprises several subphases (Kriesi Reference Knoke, Pappi, Broadbent and Tsujinaka1980; Sciarini Reference Sciarini2014). Some of these subphases are formally closed to non-state actors, but in practice they are open to non-state actors on a selective basis. To identify the various venues of the pre-parliamentary phase, we rely on official documents regarding a given policy process, and on a preliminary interview with the civil servant responsible for the respective policy process. During our interviews, we submitted the list of venues of the preparatory phase to our interview partners and asked them to indicate in which venues their organisation did actively participate.Footnote ¹⁵ To evaluate the degree of venue shopping, we calculated the proportion of non-state actors that participated in the venues of the pre-parliamentary phase for each policy process. Although this measure admittedly does not inform about the specific pattern of venue shopping for each policy actor, it is an acceptable proxy of the varying extent of venue shopping across processes, and one that is in line with the idea of conflict expansion mentioned in the theoretical section. In the policy process regarding the law on nuclear energy, the case with the lowest degree of venue shopping, only 28% of the non-state actors participating in the whole process participated, on average, in the venues of the pre-parliamentary phase of the policy process. By contrast, 48% of the non-state actors participated, on average, in the pre-parliamentary venues of the policy process regarding the infrastructure funds. Second, a dummy variable indicates whether a policy process deals with a classical socioeconomic Left–Right conflict (1) or with any other type of conflict (0). Typical examples of a socioeconomic Left–Right conflict are those regarding the pension scheme reform or the programme of budget relief 2003.

Table 1

Process descriptives

Actor and process-level determinants of the devil shift

In order to test our hypotheses, we ran multilevel models with random intercepts (Steenbergen and Jones Reference Steenbergen and Jones2002; Gelman and Hill Reference Fischer, Fischer and Sciarini2006; Bliese Reference Bliese2012).Footnote ¹⁷ The dependent variable measures the amount of devil shift at the actor level. We estimated two separate models for both dependent variables – i.e. the power and the evil dimensions of the devil shift.Footnote ¹⁸ The intraclass correlation coefficient (Bliese Reference Bliese2012) suggests that there are differences at the process level in both models: 10% with respect to the power dimension and 27% with respect to the evil dimension (see Table A.1). Further, a comparison of the fit (log likelihood ratios) (Steenbergen and Jones Reference Steenbergen and Jones2002) shows that the hierarchical null model allowing for varying intercepts across processes performs better than the nonhierarchical null model; the related difference between the log likelihood ratios was significantly different from 0. This suggests that there are indeed important differences to be explained at the process level. Note that this also holds for the power dimension, despite the fact that there was no devil shift at the aggregated level according to Table 2.

Determinants of the power dimension of the devil shift

Table 3 presents the regression results for the power dimension of the devil shift. The model in the second column includes only actor-level predictors; process-level variables are added to the model in column three.Footnote ¹⁹ Results provide mixed support for our H1 that political parties and interest groups are more affected by the devil shift than state actors. On one hand, the extent of devil shift regarding political parties is significantly higher than the reference category. On the other hand, although the coefficient for interest groups is positive, it fails to reach statistical significance. Results are even less conclusive with respect to H2, as powerful actors are not seen as more powerful than they are. Whether an actor is powerful or not has no influence on whether opponents overestimate the power of that actor.Footnote ²⁰ Unlike what is suggested by the aggregate results presented in Table 2, Table 3 shows that there are some significant process-level effects on the power dimension of the devil shift. Both variables on the process level have a significantly positive effect. On one hand, the higher the level of venue shopping in a policy process, the more actors tend to overestimate the power of their opponents. On the other hand, the devil shift is stronger in policy processes that give rise to a classical socioeconomic conflict than in processes that give rise to another type of conflict. In sum, controlling for the individual-level factors helps to uncover the process-related differences in devil shift on the power dimension. We can, therefore, corroborate H3 and H4.

Table 3

Determinants of the devil shift on the power dimension

Unstandardised regression coefficients.

Estimates in bold are statistically significant at p<0.05.

Determinants of the evil dimension of the devil shift

Table 4 presents regression results concerning the influence of actor- and process-level variables on the second dimension of the devil shift – i.e. the tendency of actors to overestimate the evilness of their opponents.

Table 4

Determinants of the devil shift on the evil dimension

Unstandardised regression coefficients.

Estimates in bold are statistically significant at p<0.05.

Results support H1. As expected, both political parties and interest groups are more affected by the devil shift than state actors. This is in agreement with the fact that state actors are less inclined to defend their own preferences and would rather play a neutral role and work towards a compromise solution for the policy issue at stake. The devil shift can also be observed for powerful actors; in line with H2, the more powerful an actor is, the more belief dissimilarity is overestimated by its opponents. This is probably due to the fact that powerful actors have a higher standing within their coalition, and are thus the main targets of opponents’ criticism. Besides differences across types of actors, we expected differences across types of policy processes. Results of the multilevel analyses that appear in the second column of Table 4 lend support to only one of the two process-level hypotheses. According to our results, actors overestimate conflict with their opponents, especially in policy processes dealing with socioeconomic issues. In contrast, the devil shift does not apply to policy processes that give rise to other types of conflicts, such as the one regarding the desired level of international openness of the country (e.g. free movement of persons with the EU or Schengen–Dublin agreement). This yields support to H4. Presumably, the devil shift is especially strong in socioeconomic issues, because the same coalitions have been opposing each other for decades. This result highlights the dangerous consequences of the devil shift phenomenon: actors consider each other as more evil than they really are, which may lead to a lack of both collaboration and trust between coalitions. This again is highly likely to reinforce the misinterpretation of information on the opposing coalition and thereby to further strengthen the devil shift, thus leading to a vicious cycle. In contrast to what we observed for the power dimension, venue shopping does not affect the evil dimension of the devil shift. This suggests that the second dimension of the devil shift does not depend on the institutional context. For the evil dimension of the devil shift, we, thus, have to reject H4. In sum, our results tend to demonstrate that the power dimension and the evil dimension of the devil shift relate to different causal factors. Whereas the evil dimension is more sensitive to actor-level factors, the power dimension is more sensitive to process-level factors. The only factor that has a consistent effect across both dimensions is the type of conflict, with socioeconomic conflicts leading to more devil shift on both the power and the evil dimensions.