Conditions of far-right strength in contemporary Western Europe: an application of Kitschelt s theory*

European Journal of Political Research 44: 837 860, 2005 837 Conditions of far-right strength in contemporary Western Europe: an application of Kitschelt s theory* JOHN VEUGELERS & ANDRÉ MAGNAN University of Toronto, Canada Abstract. Applying the demand-side claims of Kitschelt s theory, and the expectation that electoral systems affect voter choice, this article provides an explanation of cross-national variation in support for new radical right (NRR) parties between 1982 and 1995. After discussing concepts and measures, two versions of qualitative comparative analysis (Boolean analysis and fuzzy-set analysis) are applied to data for ten West European countries. The results suggest that, in combination with electoral systems that had larger district magnitudes, NRR strength resulted from a restructuring of the space of party competition due to post-industrialism and growth in the welfare state. Convergence between major parties of the left and right was not among the combination of conditions that led to NRR success. Apart from demonstrating that fuzzy-set analysis can yield a simpler explanation than Boolean analysis, this study reveals anomalous NRR outcomes for Austria, Belgium and France. Few studies of the contemporary extreme right are comparative in nature, and fewer yet provide analysis in which the same factors or processes are considered across countries, and instances of far-right weakness are compared with instances of strength (see Minkenberg 2000: 170). This generalisation has met with some notable exceptions (Jackman & Volpert 1996; Knigge 1998; Givens 2002; Lubbers et al. 2002; Golder 2003; Swank & Betz 2003). However, these studies use multivariate methods to model the average effect of selected variables in sets of pooled, cross-national data. Even when interaction effects are included, a drawback of multivariate analysis is insensitivity to causal complexity because variables are treated as if their effects were independent of context, thus invariant across cases. Unlike factor-oriented methods that search for a single model that explains the most variance, case-oriented comparative methods allow for multiple combinations of causal conditions and their related outcomes. Historical or social-scientific outcomes are analyzed in terms of intersections of conditions, and it is usually assumed that any of several combinations of conditions might produce a certain outcome (Ragin 1987: x). Boolean analysis is a case-oriented comparative method based on the binary classification of data: cases are coded according to the presence or absence of conditions and outcomes. Similar sets of conditions can lead to Published by Blackwell Publishing Ltd., 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

838 john veugelers & andré magnan different outcomes across cases; likewise, different causal conjunctions can yield the same outcome (Ragin 1987: 78 87). Complementing Boolean analysis is another case-oriented method fuzzy-set analysis that accommodates ordinal, interval and ratio-scale data. By comparison with Boolean analysis, it allows a finer measurement of values associated with conditions and outcomes. With Boolean analysis, cases are either in or out of a set; with fuzzyset analysis, they can be positioned in the grey zone in between (Ragin 2000: 150 171). This article applies a theory derived from The Radical Right in Western Europe: A Comparative Analysis by Herbert Kitschelt (1995) a study that has received much critical attention (Betz 1996; Minkenberg 1997; Schain 1997; Steinmetz 1997; Western 1997; Karapin 1998; Zimmerman 1998), but has yet to be re-evaluated through comparative analysis. We summarise Kitschelt s theory, discuss measurement issues and make the case for an amendment that includes electoral systems. Two kinds of quantitative comparative analysis are then applied. While the either/or nature of the path dependencies in Kitschelt s theory matches the binary logic of Boolean analysis, adding fuzzyset analysis permits a comparison with findings from a case-oriented approach that is more refined. This study is limited by constraints of method and data. Only West European countries in which NRR (new radical right) parties competed in elections are analysed. Variation in electoral support for NRR parties is used to measure NRR demand. Hence, including countries without NRR parties would be inappropriate because for them the demand could not be expressed through voting. Also limiting the scope of this study is the difficulty of finding data for one of the causal conditions: partocracy. We use data on perceptions of corruption, but these only capture part of the concept. Finally, data for another causal condition, core convergence, pertain to the period 1982 to 1993. Assuming the effects of core convergence can be expected to persist for two years less than the usual length of time between elections, but at least as long as the memory of voters this study is restricted to the period 1982 to 1995. Kitschelt s theory of support for the new radical right Kitschelt s theory addresses variation in the electoral success of parties he labels new radical right (NRR). 1 These parties combine nativism in the realm of citizenship rights, authoritarianism in the realm of collective decisionmaking, and laissez-faire in the realms of social and economic policy. Unifying such ideas is a blend of individualism and authoritarianism that sets NRR parties apart from others, and has led to clashes with the libertarian left over

conditions of far-right strength in contemporary western europe 839 Table 1. Performance of the New Radical Right, 1982 1995 Election year and Country Party NRR support (%) Weaker NRR outcome Finland no NRR party Germany Republicans (and National 1983 0.0 + 0.2 = 0.2 Democratic Party in 1983, 1987 0.0 + 0.6 = 0.6 1987 and 1990) 1990 2.1 + 0.3 = 2.9 1994 1.9 Ireland no NRR party Netherlands Centre Party and Centre 1982 no NRR party Democrats 1986 0.4 1989 0.9 1994 2.5 Spain no NRR party Sweden New Democracy 1982 no NRR party 1985 no NRR party 1988 on NRR party 1991 6.7 1994 1.2 United Kingdom National Front and British 1983 0.1 National Party 1987 0.0 1992 0.0 Stronger NRR outcome Austria Freedom Party 1983 5.0 1986 9.7 1990 16.6 1994 22.5 1995 21.9 Belgium Vlaams Blok (and National 1985 1.4 Front in 1987, 1991 and 1995) 1987 1.9 + 0.1 = 2.0 1991 6.6 + 1.1 =7.7 1995 7.8 + 2.3 = 10.1 Denmark Progress Party 1984 3.6 1987 4.8 1988 9.0 1990 6.4 1994 6.4 France National Front 1986 9.8 1988 9.7 1993 12.4 Italy Northern League 1983 no NRR party 1987 no NRR party 1992 8.7 1994 8.4 Norway Progress Party 1985 3.7 1989 13.0 1993 6.3 Sources: Hossay (2002); Boothroyd (2003); Nordsieck (2003).

840 john veugelers & andré magnan gender issues, multiculturalism, environmental protection and participatory politics (Kitschelt 1995: 19 20). As Table 1 shows, the electoral performance of NRR parties has varied across Western Europe. Kitschelt (1995: 5 24) assumes electoral results are produced by an interaction between supply and demand. Thus, to have a relatively strong NRR vote, a country needs a party with a NRR platform and structural conditions that make such a platform appealing to voters. Even if a segment of its electorate is receptive to right-authoritarian ideas, a country will not produce a recognisable NRR vote unless it also has a party that represents such ideas. Likewise, a country is not ripe for NRR success without the structural conditions that encourage demand for NRR parties: advanced post-industrial capitalism; convergence of the main parties of the moderate left and right; and partocracy, a patronage-based party system and political economy. Advanced post-industrial capitalism According to Kitschelt (1995: 6 16), post-industrial capitalism has shifted the old left-right axis of party competition because a new pool of authoritarian voters has emerged in opposition to the libertarian left. Demand for the NRR has been growing because jobs exposed to international competition are associated with a workday experience and personality type that encourage right-wing authoritarian demands. A society that is agrarian and industrial can only produce a form of right-wing extremism modelled on inter-war fascism. In sum, NRR parties are most likely to appear and to be electorally successful in post-industrial societies with large welfare states (Kitschelt 1995: 20). Post-industrialism refers to a relative decline in the primary and secondary sectors of the economy as the service sector grows (Bell 1974). Thus one indicator of post-industrialism is the level of employment in the service sector as a proportion of civilian employment (OECD 2002a). Variation in the size of the welfare state can be captured by different indicators, such as civilian government employment as a proportion of the working-age population, or the proportion of employment in the education, welfare and public health sectors (Esping-Andersen 1990; Huber & Stephens 2001). Due to its simplicity and validity (Swank 2002: 73), we settled on public expenditure on social protection as a proportion of the gross domestic product (OECD 2002b). Core convergence between left and right Kitschelt (1995: 11 25) claims NRR voters are not necessarily more authoritarian than supporters of other parties. The strategy of the moderate right thus

conditions of far-right strength in contemporary western europe 841 affects the electoral chances of the NRR. When the moderate right moves away from the political centre, it gains credibility as an alternative to the NRR; if it moves toward the centre, electoral space opens up for the far right. Similarly, the centripetal movement of a major party of the left can create the perception of a cartel of established left and right parties that do not present voters with real alternatives. Does core convergence denote a process or state of affairs? In arguing against the view that parties are simply the reflection of voter preferences, Kitschelt (1995: 14) emphasises that support for right-authoritarian parties depends on the strategic interaction of competing parties in the party system. Dynamic conceptions of convergence follow in Kitschelt s theoretical exposition and country-by-country case analyses. Yet, a static treatment of convergence creeps into his systematic comparison of the left-right distance in different countries, for here the data pertain to 1989 alone (Kitschelt 1995: Table 2.2; the data are from Laver & Hunt 1992). This is problematic, for if convergence denotes a process it cannot be captured by observations at a single point in time (Abedi 2002: 554). Knutsen (1998: Figure 1) which gives expert judgements of party positions in 13 European countries for 1982 and 1993, provides a more appropriate basis for estimating convergence. Like the data used by Kitschelt, that used by Knutsen shares the drawbacks of expert judgements on party positions (Mair & Castles 1997; Budge 2000). Nonetheless, they allow an estimation of change over time and were used for this study. 2 Patronage-based economy and party system Partocracy the fusion of state, party, and economic elites in politicoeconomic networks characterized by patronage, clientelism, and corruption (Kitschelt 1995: 161) creates a voter demand met by a subtype of the NRR: the populist antistatist party. Kitschelt (1995) says Austria and Italy are partocracies, but does not say if they are the only ones. This may be due to a paucity of cross-national research, for information about patronage is necessarily nebulous and sometimes sketchy (Mair 1997: 142). Nonetheless, national rankings of perceptions of the degree of political corruption in the public service are available. Partocracy and corruption are not synonymous, but this indicator seems defensible. Kitschelt s discussion of Austria and Italy stresses it is not simply partocracy that has been responsible for antistatist populism. Along with concerns over excessive state intervention and poor opportunities for political involvement, voter support for populism stems from a desire for more transparency in politics, and hence is accompanied by the perception that impropriety must be rooted out (see also Mény

842 john veugelers & andré magnan 1996: 118 122). Indeed, in analysing the rise of Italy s Northern League, Kitschelt (1995: 175) writes: [I]t was primarily the overwhelming concern with the corruption and inefficiency of the partocratic state and party machines that rallied voters around the new party. Perceived corruption is thus a key feature of partocracy. Though measurement is difficult (Mény & Rhodes 1997: 102), since 1995 Transparency International has conducted annual surveys of the perception of corruption among thousands of business executives who work internationally (Heidenheimer 1996). Our analysis uses scores for countries on Transparency International s (2003) index of perceived corruption, which runs from 0 (completely corrupt) to 10 (completely clean). Consistent with Kitschelt, the two exemplary cases of partocracy Austria and Italy appear among those countries with the highest levels of perceived corruption. However, also in this group are Belgium (5.96) and France (6.77), so cross-nationally the potential for antistatist populism may be greater than implied by Kitschelt s analysis. The electoral system Beyond the factors considered so far, Kitschelt (1995: 58) asks if NRR success also depends on the nature of the electoral system: Moderate thresholds of representation in PR [proportional representation] systems cut both ways: they discourage splinter parties, but they encourage the formation of new parties that can reckon to represent significant voter constituencies. The constraining effect of electoral laws really applies to first-past-the-post plurality laws because even in a majoritarian system with runoffs party proliferation may be encouraged if parties treat the first round of voting as a show of relative party strength that then influences the bargaining between parties before the runoff among the lead contenders. Turning to evidence, Kitschelt (1995: 58) examines the connection between electoral systems and NRR results and concludes there is only a modest correlation between electoral laws and strength of the extreme Right. Against this conclusion, two arguments may be advanced. First, re-analysis of Kitschelt s data suggests the correlation is not moderate, but strong. Table 2 sorts countries by NRR strength and whether they have a system with a proportional representation threshold below 4 per cent. Countries with a plurality system, a majoritarian system or a system of proportional representation with a threshold of 4 per cent or more are grouped together as they set higher

conditions of far-right strength in contemporary western europe 843 Table 2. Electoral systems and NRR outcomes Strict proportional system no yes NRR outcome weak SW UK GE SP IR NL strong AU FI FR DK NO IT BE Note: Strict proportional systems are those with a representation threshold below 4%. Source: Kitschelt (1995: Tables 2.3 and 2.4). barriers to parliamentary representation. Such barriers are intended to discourage multipartism (thus unstable coalition government) as well as support for extremist parties. The only anomalous instances are Austria, France and the Netherlands; otherwise, the close fit between NRR success and Kitschelt s classification of countries according to their electoral system suggests the relationship merits further investigation. Second, experts on the effects of electoral systems have redefined their subject so as to include not just the electoral formula (e.g., majoritarian, proportional or semi-proportional), but also the district magnitude (the number of members elected per constituency or other voting district; Lijphart 1994: 10). Proportionality (similarity between parties share of the vote and their share of the seats in parliament) increases with district magnitude that is, as the number of seats per voting district increases (although this does not hold for a plurality system where larger districts can be associated with less proportionality; see Farrell 2001: 21 22). Also affecting proportionality is uppertier correction. The intended effect of allocating seats at the district or national level is compensatory: to make representation more proportional than allocation based on election results at the district level alone. Proportionality is thus a function of: lower-tier district magnitude, the median number of parliamentary seats allocated to electoral districts in the country; and upper-tier correction, the percentage of seats allocated outside the lower tier (Amorim

844 For Boolean analysis, causal conditions are treated as dichotomous variables: a value of 0 indicates the absence or weakness of a condition, while a 1 indicates its presence or strength. To classify cases by the outcome, we used a threshold of 3 per cent in mean support for the NRR. When calculating the mean of NRR support for a country, only elections in which an NRR party was a competitor were included. To classify cases according to their welfare state effort (WSE), we divided them using the median of all countries mean public expenditure on social protection as a percentage of GDP for the period 1982 to 1995 (26.2 per cent). Similarly, the median of countries mean level of services as a percentage of civilian employment over the period 1982 to 1995 (66.2 per cent) was used in coding countries according to the size of their service sector (SSS). To calculate the degree of convergence between major left and right parties (CONV), the distance between left and right parties in 1993 was subtracted from the distance between left and right parties in 1982. Following Knutsen (1998: 86), who suggests change in the distance between left and right parties must exceed 0.4 units to be significant, we classified countries such that any change over 0.4 units was coded as 1, while the other cases were coded 0. To divide countries according to their level of corruption (CORR) we used the data for 1995 because these fall within the period covered by the analyses and data for earlier years are unavailable. Inspection of the data for 1995 to 2000 lends confidence to our use of data from 1995 alone since there is little variation over this time period. Corruption perception scores underwent slight inflation over time, but for most countries this change was not dramatic and it hardly affected the rankings. Countries with scores below 8.4 (the median) were coded 1 (more corrupt) while the rest were coded 0 (less corrupt). We coded countries according to their lower-tier district magnitude (DM) using the mean district magnitude of elections in which the NRR was present. How high a district magnitude must be to qualify a system as proportional is a matter of judgement. We used a mean district magnitude of 6 to distinguish between countries. Likewise, it is difficult to say precisely how much upperjohn veugelers & andré magnan Neto & Cox 1997: 157 158). Variation in these two features of electoral systems affects the likelihood that casting one s vote for a party outside the mainstream will be perceived by voters as useful, not wasted. As an amendment to Kitschelt s theory, the difficulty of achieving parliamentary representation is therefore included in our analysis. Golder (2003b) was the source of the data on district magnitudes and upper-tier correction used here. Boolean analysis

conditions of far-right strength in contemporary western europe 845 tier correction (UPPER) is needed before the proportionality of an electoral system can be deemed significant. Under the assumption that corrections of less than 5 per cent can be considered more or less trivial, we divided the countries so those with an upper-tier correction of more than 5 per cent were coded 1, while those with a correction of less than 5 per cent were coded 0. The coded data were sorted into combinations of values on the outcomes, NRR weakness and NRR strength. This resulted in a truth table with all logically possible combinations of conditions and outcomes. The combinations were analysed using the Quine algorithm of the fs/qca 1.1 software (Ragin & Drass 2003). 3 Comparative studies rarely include enough cases to represent the full range of logically possible combinations of causal conditions. Yet, analysis must proceed with caution when making simplifying assumptions about combinations of causal conditions that are not represented empirically (Ragin 2000: 139). Combinations of causal conditions for which there are no empirical instances (i.e., those rows of the truth table for which there are no corresponding empirical cases), but that are implied by the solution may be included as simplifying assumptions only if theoretically justified. Logical reduction of the truth table produced an initial solution for causal sufficiency (no necessary causes were found): NRR = wse sss CORR DM UPPER + wse SSS conv corr DM upper + WSE sss CONV CORR dm upper + WSE SSS conv corr DM UPPER + WSE SSS CONV CORR DM upper 4 We assessed the simplifying assumptions of the initial solution through a thought experiment: If this combination of causal conditions were empirically represented, could we justifiably expect it to result in a positive outcome? According to Kitschelt s theory, higher welfare effort, a larger service sector, core convergence and more corruption should lead to NRR strength. Thus, for the first term of the initial solution (wse sss CORR DM UPPER), an allowable simplifying assumption is WSE sss CORR DM UPPER. Using simplifying assumptions allows researchers to modify the original truth table by setting cell values to don t care (represented in the table by ) where appropriate (Ragin 2003). Logical reduction of the modified truth table (Table 3) yields a solution for causal sufficiency (no necessary conditions were found):

Table 3. Boolean analysis, modified truth table* Size of Lower-tier district Upper-tier Welfare effort service sector Convergence Corruption magnitude correction NRR weakness NRR strength 1 1 - - 1 1 DK, SW 1-1 1 - - FR 0 0 1 0 0 0 UK 1 1 1 0 1 0 NL 1 1 1 1 1 - BE 0 0 1 1 0 1 GE - 1 0-1 - NO - - 0 1 1 1 IT - - 1 1 1 1 AU Note: *Modifications are the result of including simplifying assumptions based on the directional expectations that high welfare effort, a large service sector, core convergence, corruption and either a large district magnitude or a large upper-tier correction should contribute to NRR strength. 846 john veugelers & andré magnan

conditions of far-right strength in contemporary western europe 847 NRR = SSS conv DM + WSE CONV CORR + CORR DM UPPER + WSE SSS DM UPPER While the initial solution specified five causal pathways to NRR strength, this solution specifies four. Apart from adding simplicity, this solution brings out the dissimilarity of Norway, a country with a stronger NRR party but no core convergence. Otherwise, the effects of welfare effort or post-industrialism are broadly consistent with Kitschelt s theory. In combination with other conditions, corruption was an element of NRR success in Austria, Belgium, France and Italy. And, consistent with our amendment to Kitschelt s theory, more permissive electoral systems (through district magnitudes or upper-tier correction) belonged to the combinations of conditions leading to NRR success in five of seven cases: Austria, Denmark, Italy, Norway and Sweden. Fuzzy-set analysis Fuzzy-set coding is an essentially interpretive process that must ensure a fit between membership scores and the theoretical content of the concepts used (Ragin 2000: 165 166). While Boolean analysis calls for binary coding, with the fuzzy-set approach the cases usually have partial membership scores between 0.0 and 1.0. These are linked to words describing the degree of membership in the causal condition. For our fuzzy-set analysis, membership in the set of countries with a more successful NRR party hinges on the mean of electoral support for NRR parties between 1982 and 1995 (Table 4). As with the Boolean analysis, calculation of this mean is based only on results from elections in which there was a NRR party. Full membership in the set of countries with a strong NRR party is set at mean electoral support of 15.0 per cent or greater; full non-membership at 0.9 per cent or lower. The crossover point (the point at which a case becomes more in than out of the set) is set at 3 per cent mean electoral support. Scoring for welfare-state effort (WSE) is based on public expenditure on social protection as a percentage of GDP. Full membership in the set of countries with a high welfare effort is set at a mean for 1982 to 1995 of 35.1 per cent or greater. This level approximates the highest one-year level of public expenditure between 1982 and 1995 for Sweden, a leading example of the social-democratic type of welfare state (Esping-Anderson 1990). Full nonmembership is assigned to cases that fell on or below a mean of 10.0 per cent

Table 4. Fuzzy-set coding scheme: Empirical indicators and their translation into fuzzy set scores More or Neither in More or Fully in Mostly in less in the nor out less out of Mostly out Fully out Theoretical the set the set set of the set the set of the set of the set concept Empirical indicator 1.000 0.751 0.999 0.501 0.750 0.500 0.251 0.499 0.001 0.250 0.000 Strong NRR Mean electoral support 15.00 6.00 14.90 3.00 5.90 2.90 2.00 2.80 1.00 1.90 0.00 0.90 outcome for the NRR (1982 1995) (%).* Advanced post- Welfare effort: mean of 35.10 25.10 35.00 20.10 25.00 20.00 15.00 19.90 10.10 15.00 10.00 industrial public expenditure capitalism on social protection as a percentage of GDP (1982 1995). Size of service sector: 75.00 62.60 74.90 50.10 62.50 50.00 37.60 49.90 25.10 37.50 25.00 services as a percentage of civilian employment, mean (1982 1995). Core convergence Convergence: left-right 2.01 1.01 2.00 0.41 1.00 0.40 0.21 0.39 0.01 0.20 0.00 between left and party distance (1982) right parties minus left-right party distance (1993). Patronage-based Corruption: corruption 0.00 0.01 2.50 2.51 4.99 5.00 5.01 7.50 7.51 9.99 10.00 party system and perception index political economy score (1995). Proportionality of District magnitude: log 1.60 1.20 1.59 0.81 1.19 0.80 0.40 0.79 0.01 0.39 0.00 the electoral of mean DM for system electoral years (1982 1995).* Upper tier correction: 50.00 25.00 49.90 5.10 24.90 5.00 2.60 4.90 0.10 2.50 0.00 mean for electoral years (1982 1995).* Note: *based only on elections in which NRR parties competed. 848 john veugelers & andré magnan

conditions of far-right strength in contemporary western europe 849 of public expenditure on social protection. The qualitative crossover point is set at 20.0 per cent of expenditure on social protection, the spending level of the liberal type of British welfare state under the Thatcher government during the 1980s. Scoring for the size of the service sector (SSS) is based on services as a percentage of civilian employment. We set the crossover point at 50.0 per cent of the mean of service sector employment as a proportion of total civilian employment for 1982 to 1995. Full membership was given to cases that had 75.0 per cent service employment or greater (a level near the maximum in 1995 for the Netherlands, the country with the highest scores), while full nonmembership was set at 25.0 per cent service employment or less. These qualitative anchors provide interpretive simplicity because coding is symmetrical around the crossover point of 50.0 per cent, the level beyond which a society is truly post-industrial because service sector employment outweighs employment in the primary and manufacturing sectors combined. Fuzzy-set scores for convergence between left and right parties (CONV) were derived from the scores calculated for the Boolean analysis. Knutsen s (1998: 86) criterion of significant change in party distance over time a change of 0.4 units or greater provides the crossover point for membership in the set of countries exhibiting core convergence between 1982 and 1993. We set full membership at 2.01 units of difference or greater, and full non-membership at no change (0.0), or negative change (i.e., core divergence). Countries displaying either divergence or no change in party distance could not have partial membership in the set of countries exhibiting core convergence. In scoring for perceptions of corruption (CORR), we noted that Transparency International s index is designed such that a score under 5.0 indicates a country is perceived as more corrupt than clean, while a score over 5.0 means it is perceived as more clean than corrupt. We transposed this index such that a score of 0.00 qualified a case as fully in this set, 5.0 was the crossover point and a score of 10.0 would qualify a case as fully out of the set. Changes in district magnitude (DM) are more significant at lower values. For example, the 10-unit difference between a majoritarian or first-past-thepost system with a district magnitude of 1 and another with a district magnitude of 11 is more meaningful than the difference between systems that have district magnitudes of 80 and 90. Hence, we took the log of mean district magnitudes before converting them into fuzzy-set scores (Golder 2003a: 445). Cases were assigned full membership at a logged district magnitude of 1.6 or greater (equivalent to a raw district magnitude of 40), and full non-membership at a logged value of 0.0 (equivalent to a raw district magnitude of 1 a fully non-proportional system ). The crossover point was a logged district magnitude of 0.80 (corresponding to a raw district magnitude of 6.3).

850 john veugelers & andré magnan We took the mean of upper-tier correction (UPPER) to establish the set of countries with high upper-tier correction. Full membership was assigned to countries with an upper-tier correction of 50 per cent or more, full nonmembership to cases with no upper-tier correction. To distinguish between countries with relatively trivial correction and others with a level that made the electoral system significantly more proportional, the crossover point was set at 5.0 per cent. Table 5 summarises each country s fuzzy-set membership scores. The property space constituted by countries membership in the outcome and causal conditions was analysed using the software fs/qca 1.1 (Ragin & Drass 2003). We began with the most restrictive analysis by excluding all simplifying assumptions. An adjustment of 0.10 units was included to allow for measurement imprecision and uncertainty over the establishment of qualitative benchmarks (Ragin 2000: 225). Excluding all simplifying assumptions, we found the following combination of causal conditions was usually sufficient for explaining NRR success: NRR = WSE SSS conv corr DM. Table 6 shows the levels of membership in this set. A measure of this solution s consistency with an argument of causal sufficiency can be calculated by dividing the sum of consistent cases membership scores in the causal condition (i.e., the scores for cases that do not violate the argument of sufficiency) by the sum of all cases membership scores in the causal condition (Ragin 2003). For the first fuzzy-set solution, the consistency score is 0.71, which reveals the existence of significant anomalies. The fit between solutions and empirical observations is addressed by coverage: the extent to which a combination of causal conditions is empirically represented by a smaller or larger proportion of cases (Ragin 2003). Only three of the ten countries (Denmark, Norway and Sweden) have membership scores above 0.50 in the combination of causal conditions specified in this solution, meaning these three countries represent the only good empirical instances of this set of causal conditions. Though conditions in the other seven countries may have been consistent with the argument of sufficiency (because their membership in the cause is less than or equal to their membership in the outcome), they are not well matched to this set of causal conditions. The coverage score for this solution is 0.34, a relatively low value that suggests this combination of causal conditions covers only 34 per cent of the total membership in the outcome. Next, we re-analysed the property space allowing for simplifying assumptions based on Kitschelt s theory and our expectation that a more permissive electoral system would have encouraged NRR success. 5 Under these conditions, the following combination of causal conditions is usually sufficient for predicting NRR strength (no necessary conditions were found):

Table 5. Fuzzy-set membership scores for analysis of NRR strength Size of service Lower tier district Upper tier NRR support Welfare effort sector Convergence Corruption magnitude correction Austria 1.00 0.77 0.60 0.84 0.29 0.92 0.60 Belgium 0.70 0.77 0.87 0.82 0.32 0.57 0.00 Denmark 0.75 0.87 0.83 0.00 0.07 0.56 0.73 France 0.88 0.78 0.77 0.84 0.30 0.30 0.00 Germany 0.08 0.63 0.62 0.50 0.19 0.00 1.00 Italy 0.82 0.64 0.63 0.00 0.70 0.70 0.72 Netherlands 0.08 0.82 0.89 0.85 0.13 1.00 0.00 Norway 0.80 0.69 0.86 0.42 0.14 0.63 0.32 Sweden 0.59 0.92 0.85 0.04 0.12 0.66 0.58 United Kingdom 0.00 0.62 0.81 0.88 0.14 0.00 0.00 conditions of far-right strength in contemporary western europe 851

852 john veugelers & andré magnan Table 6. Fuzzy membership in the causal expression passing sufficiency,simplifying assumptions excluded WSE SSS conv corrupt DM NRR Strength Austria 0.16 1.00 Belgium 0.18 0.70 Denmark 0.56 0.75 France 0.16 0.88 Germany 0.00 0.08 Italy 0.30 0.82 Netherlands 0.15 0.08 Norway 0.58 0.80 Sweden 0.66 0.59 United Kingdom 0.00 0.00 Consistency 0.71 Coverage 0.34 Note: Results based on the probabilistic criteria that each causal condition is usually sufficient (0.65) for predicting membership in the fuzzy set NRR strength. Alpha = 0.05 and fuzzy adjustment = 0.10. 1 Austria 0.9 France Membership in the set of countries with NRR strength 0.8 0.7 0.6 0.5 0.4 0.3 0.2 Belgium Italy Norway Denmark Sweden 0.1 Germany Netherlands United Kingdom 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Membership in the causal expression WSE SSS conv DM + WSE SSS CORR DM Figure 1. NRR strength against the maximum of the causal expressions, including simplifying assumptions

conditions of far-right strength in contemporary western europe 853 Table 7. Fuzzy membership in causal expressions passing sufficiency, simplifying assumptions included WSE SSS Maximum of Causal WSE SSS CORRUPT Expressions NRR conv DM DM (adjusted scores) Strength Austria 0.16 0.29 0.29 1.00 Belgium 0.18 0.32 0.32 0.70 Denmark 0.56 0.07 0.56 0.75 France 0.16 0.30 0.30 0.88 Germany 0.00 0.00 0.00 0.08 Italy 0.63 0.63 0.63 0.82 Netherlands 0.15 0.13 0.15 0.08 Norway 0.58 0.14 0.58 0.80 Sweden 0.66 0.12 0.66 0.59 United Kingdom 0.00 0.00 0.00 0.00 Consistency 0.74 0.94 0.72 Coverage 0.40 0.33 0.47 Partitioned coverage 0.14 0.07 Note: Results based on the probabilistic criterion that each causal condition is usually sufficient (0.65) for predicting membership in the fuzzy set NRR strength. Alpha = 0.05 and fuzzy adjustment = 0.10. NRR = WSE SSS conv DM + WSE SSS CORR DM This solution can be factored to become: NRR = WSE SSS DM (conv + CORR) This summarising statement suggests two paths to a strong NRR outcome between 1982 and 1995, each of which combines a large welfare effort, a large service sector and a high district magnitude with either corruption or the absence of core convergence. As Table 7 shows, the consistency score for the causal solution as a whole is 0.72, while the first and second terms of the solution have consistency scores of 0.74 and 0.94, respectively. The solution as a whole covers approximately 47 per cent of cases outcome scores, whereas the two terms of the solution cover 40 per cent and 33 per cent, respectively. Thus, the causal path represented by the first term WSE SSS conv DM has somewhat more explanatory weight than the second term: WSE SSS CORR DM. The total coverage of this

854 john veugelers & andré magnan solution is also considerably larger than for the most restrictive solution (47 versus 34 per cent). Partitioned coverage each term s unique contribution to the total coverage accounts for any overlap between each term s coverage and isolates whatever portion of the coverage is the unique contribution of the term in question (Ragin 2003). The partitioned coverage score for the first term is 0.14; the second term s partitioned coverage score is half as large, at 0.07. These results provide further evidence that the set of causal conditions represented in the first term of the solution is causally more important than the set represented by the second term. Furthermore, four countries provide good empirical instances of the set of causal conditions WSE SSS conv DM (Denmark, Italy, Norway and Sweden), but only one country (Italy) is a good empirical instance of the causal combination WSE SSS CORR DM. As the two terms of the causal solution are joined by the fuzzy-set operator + ( or ), a country s membership in the causal expression as a whole depends on each case s maximum of its membership scores in each term. Austria s score in the set of countries with a large welfare effort, a large service sector, no convergence between left and right parties and a high district magnitude (the first term) is 0.16, while its membership score in the set of countries with a large welfare effort, a large service sector, corruption and a high district magnitude (the second term) is 0.29 (see Table 7). The logical or, which joins the two terms, requires that we take this case s maximum of the two scores as its membership in the causal solution as a whole (WSE SSS DM (conv + CORR)). Though the second term of the solution better explains Austria s membership in the set of countries with NRR strength than the first term, Austria remains a poor example of either of these terms because for both its membership score is below 0.50. Two other cases also were not explained well by this solution: Belgium and France. Otherwise, Figure 1 shows that the combination of conditions contained in this solution provides a good fit for three cases of NRR weakness (Germany, the Netherlands and the United Kingdom) and four cases of NRR strength (Denmark, Italy, Norway and Sweden). 6 Discussion We have assessed the claim that demand for NRR parties has been spurred by post-industrialism, partocracy and core convergence. Moreover, we have tested the notion that smaller, non-mainstream parties like those of the NRR gain from permissive electoral systems that set low barriers to representation. The main findings from this study come from the fuzzy-set analysis: stronger

conditions of far-right strength in contemporary western europe 855 support for the NRR resulted where (1) a permissive electoral system allowed for the electoral expression of right-authoritarian demands created by (2) post-industrial societies with (3) large welfare states and either (4a) higher corruption or (4b) an absence of core convergence. Contrary to Carter (2002), but consistent with other studies (Jackman & Volpert 1996; Golder 2003; Swank & Betz 2003), NRR support tended to be higher in countries with systems that were more proportional. Furthermore, we can pinpoint which aspect of electoral systems matter. In combination with other conditions identified by the fuzzy-set analysis, larger district magnitudes helped the NRR, whereas variation in upper-tier correction had little effect. 7 A country in which there was a radical-right demand among voters did not produce a stronger NRR vote in the absence of an electoral system with higher district magnitudes. In combination with other conditions, NRR support tended to be stronger in advanced post-industrial countries with higher levels of welfarestate expenditure and service-sector employment. This finding lends support to Kitschelt s (1995: 13 19) contention that new political preferences within the electorate have been restructuring the space of party competition. It also undermines Swank and Betz s (2003: 239) claim that a stronger welfare state depresses NRR support by dampening the negative socioeconomic effects of globalisation. Kitschelt (1995: Figure 1.2) posits that, if accompanied by advanced post-industrial capitalism and core convergence, patronage-based politics will encourage medium support for right-authoritarian parties and relatively strong support for populist anti-statist parties. Two countries with higher levels of perceived corruption (Belgium and France) had relatively successful rightauthoritarian parties, while two others (Austria and Italy) had relatively successful parties of the populist anti-statist type. Another country with higher levels of perceived corruption (Germany) had a relatively weak welfarechauvinist party. However, in the Scandinavian countries (Denmark, Norway and Sweden), the NRR did relatively well even though perceived corruption was weak. These findings are consistent with Kitschelt s claims about the effects of partocracy on NRR demand, and they confirm the utility of his innovative distinction between subtypes of the NRR. Contrary to Kitschelt (1995) and Abedi (2002), similarity between the main parties of the left and right did not provide an opportunity for the NRR, whose success instead tended to be stronger in countries where, in combination with other conditions identified by fuzzy-set analysis, core convergence was absent or weak. This finding lends support to the alternative view that extreme-right parties have benefited from core divergence in party systems. According to Ignazi (1992), core divergence was triggered when conservative parties shifted to the right while guarding their left flank against encroachment by potential

856 john veugelers & andré magnan competitors. A centrifugal dynamic opened the way to a politics of outbidding. Less encumbered than the conservative right, the extreme right could voice the most radical promises without any strategic hindrance (Ignazi 1992: 20). Our first suggestion for future research flows from a Weberian conception of the relationship between history and the social sciences. We believe the solutions from fuzzy-set analysis should be treated as abstractions not to be confused with historical developments as they actually unfolded during the 1980s and 1990s. In this light, a fuzzy-set solution is an ideal-type to be used as a heuristic, an unambiguous summary against which to bring out the individuality of historical cases. Solutions should thus be treated as theories ought to be treated: not as ends in themselves, but as logically consistent statements whose value resides in their usefulness as tools. Using our fuzzy-set results to pose questions of specific cases provides one direction for future research. Sometimes a way of moving forward is to resolve the contradictions in the truth tables created by Boolean analysis. Strictly speaking, a Boolean truth table assumes an unambiguous output, an outcome of either 0 or 1 for every causal combination in the empirical data (Ragin 1987: 113). Contradictions occur when a combination of causal conditions co-exists with both positive and negative cases of the outcome. Table 3 would have contained a contradiction if a set of similar causal conditions had led to an ambiguous outcome, if there had been a row with empirical instances of both NRR weakness and NRR Strength. When contradictions do exist, they are resolved by introducing new causal variables that reassign the ambiguous instances into separate causal combinations. The absence of contradictions in Table 3 is reassuring from the viewpoint of explanation for it means that positive and negative instances of the outcome in question have been successfully disentangled by variation in the causal conditions selected for the analysis. It also eliminates one possibility among lines of future research, however. A different possibility is to seek a better explanation for anomalous cases. After fuzzy-set analysis, the countries fell into three groups. The first group had low levels of NRR support and low district magnitudes or levels of corruption (Germany, the Netherlands and the United Kingdom); the second had medium-to-high levels of NRR support and higher levels of corruption (Italy) or medium-to-high welfare spending (Denmark, Norway and Sweden); the third group includes countries that like the second group had medium-tohigh levels of NRR support, but whose NRR parties performed considerably better than the solution would lead us to expect (Austria, Belgium and France). Put differently, during the period 1982 to 1995 one pathway led to NRR weakness and two pathways led to NRR strength. Our fuzzy-set solution provides a good fit with the NRR results for the countries in groups one and two, but a poor fit with the NRR results for countries in group three.

conditions of far-right strength in contemporary western europe 857 Research that builds on ours should search for conditions that yield a better explanation of NRR outcomes in Austria, Belgium and France. The findings from our fuzzy-set analysis are not overly complicated. Indeed, in this study the fuzzy-set analysis provides a clear advance over Boolean analysis. It is based on a coding system that better captures the nondichotomous nature of the data, yet also provides a solution with two combinations of causal conditions instead of the four produced by Boolean analysis. Fuzzy-set analysis may be superior to Boolean analysis not only because it does better justice to non-categorical data, but, paradoxically, it can also produce simpler explanations. For this study at least, concern about the sheer complexity of the set-theoretic models, whether classical or fuzzy (Schrodt 2002: 453) is unwarranted. Acknowledgments A preliminary version of this article was presented at the Joint Sessions of the European Consortium for Political Research, Grenoble, France, 6 11 April 2001. The support of the Social Sciences and Humanities Research Council of Canada is gratefully acknowledged, as are comments by G. Capoccia, G. Dierickx, A.J. Heidenheimer, H. Kitschelt, J. Myles, J.S. O Connor, C. Ragin, A. Widfeldt and two anonymous reviewers. Responsibility for the data, analyses and interpretations rests with the authors alone. Notes 1. Though different labels and definitional criteria have been assigned to the NRR, there is wide agreement about which parties actually belong to this party family (see Ignazi 1992; Betz 1994; Jackman & Volpert 1996; Karapin 1998; Schain et al. 2002; Golder 2003; Swank & Betz 2003). A possible exception arises with Finland, which Kitschelt (1995: Table 2.4) ranks among countries with a stronger NRR party. Elsewhere (Kitschelt 1995: 49 and Table 2.1), however, he follows others in excluding the Finnish Rural Party from the NRR category. 2. Our analysis of core covergence uses the same parties as Kitschelt (1995: Table 2.2), except for Greece and Portugal, which had to be dropped due to a lack of comparable data. 3. Using the Quine algorithm is equivalent to using the Ragin algorithm, applying veristic criteria and setting to 1 the minimum number of cases represented by a causal condition for it to be considered a necessary or sufficient cause. Veristic criteria require that every instance of a set of causal conditions produce the outcome. An alternative to the Quine algorithm is the agglomerative algorithm, which allows for the application of probabilistic criteria to the analysis of necessary and sufficient causes. Using probabilistic criteria lets a researcher set benchmark proportions (e.g., 0.65) that combinations of causal com-

858 john veugelers & andré magnan binations must meet to pass a test of sufficiency. For analysis involving a small number of cases, the veristic approach of the Quine algorithm is usually more appropriate (Ragin 2000: 138 139). 4. Capital letters show the presence of the causal condition, and lower-case letters show its absence, e.g., CORR means high perceived corruption, while corr means low perceived corruption; = means results from, while + stands for the logical condition or. 5. The following simplifying assumptions were made in our analysis: (1) WSE SSS conv CORR DM upper; (2) WSE SSS CONV CORR DM upper; (3) WSE SSS CONV CORR DM UPPER. WSE SSS conv CORR DM upper already conforms to these criteria as it builds on the result of the most restrictive solution (WSE conv corr DM UPPER). 6. We carried out an additional fuzzy-set analysis that withheld directional expectations about the effect on NRR support of a stronger welfare state. Explanation suffered as a result: membership scores did not change and consistency scores were almost unchanged, while coverage scores weakened. 7. Our analysis therefore yields somewhat different findings from those of Golder (2003: 461), who finds that both district magnitudes and upper-tier correction matter. References Abedi, A. (2002). Challenges to established parties: The effects of party system features on the electoral fortunes of anti-political-establishment parties. European Journal of Political Research 41(4): 551 583. Amorim Neto, O. & Cox, G.W. (1997). Electoral institutions, cleavage structures and the number of parties. American Journal of Political Science 41(1): 149 174. Bell, D. (1974). The coming of post-industrial society. New York: Basic Books. Betz, H.-G. (1994). Radical right-wing populism in Western Europe. New York: St. Martin s Press. Betz, H.-G. (1996). Review of The Radical Right in Western Europe: A Comparative Analysis by Herbert Kitschelt. Political Science Quarterly 11(1): 716 717. Boothroyd, D. (2003). United Kingtom Election Resuits. Available online at: www. election.demon.co.uk. Budge, I. (2000). Expert judgements of party policy positions: Uses and limitations in political research. European Journal of Political Research 37: 103 113. Carter, E.L. (2002). Proportional representation and the fortunes of right-wing extremist parties. West European Politics 25(3): 125 146. Esping-Anderson, G. (1990). The three worlds of welfare capitalism. Princeton, NJ: Princeton University Press. Farrell, D.M. (2001). Electoral systems: A comparative introduction. Houndmills: Palgrave. Givens, T.E. (2002). The role of socioeconomic variables in the success of radical right parties. In M. Schain, A. Zolberg & P. Hossay (eds), Shadows over Europe: The development and impact of the extreme right in Western Europe. New York: Palgrave Macmillan, pp. 137 158. Golder, M. (2003a). Explaining variation in the success of extreme right parties in Western Europe. Comparative Political Studies 36(4): 432 466. Golder, M. (2003b). Data available online at: http://homepages.nyu.edu/~mrg217. Heidenheimer, A.J. (1996). The topography of corruption: Explorations in a comparative perspective. International Social Science Journal 48(3): 337 351.