In virtually all democratic political systems, parties or party representatives compete for votes. The question of how people view the various parties - which are seen as alike and which are seen as different, and what structure, if any, undergirds those perceptions - is fundamental to understanding electoral competition in democratic societies. In this paper we will examine the structure of mass perception of political parties in six multiparty systems. Our goal is both descriptive and theoretical. Descriptively, we are interested in providing a visualizable overview of electoral competition in each of the countries. Theoretically, we are interested in understanding the relationship between party strategy and mass support in multiparty systems.

We start with the premise that there is an underlying spatial structure that guides mass perceptions of political parties. The simplest illustration of the idea of spatial structure, and one that has had great influence on how political competition is understood, is the left-right model of electoral conflict. According to this model, both parties and voters can be located on a single ideological continuum with voters choosing the party that is closest to them (Downs 1957). If this model were applicable, we would understand a great deal of the political dynamics of the society once we had located both the parties and voters in their appropriate positions. Spatial position would predict the preference of each voter across the full

Resumé

This article examines mass perceptions of political parties in Norway, Sweden, Denmark, West Germany, the Netherlands and France. For each country we construct a map in which the parties, voters, and demographic groups are located to provide a visual and spatial overview of the structure of electoral competition. Two dimensions are adequate for displaying the main ideological cleavages in each of the countries. In each case there is a strong left-right dimension combined with a more culturally defined and usually weaker second competitive dimension. In general, we find that no parties occupy the center areas of the space, which are usually dense with voters. This leads us to question the adequacy of the traditional spatial model of elections for describing competition in multiparty systems.

set of parties. In addition, the proximity of the parties to each other should be a meaningful guide to the coalitions that would form in the legislature. Such a system would be highly structured, of course, since the essential features of political conflict in the society could be summarized on a single dimension.

The basic spatial paradigm can be extended to any number of dimensions (see, for example, Davis et al. 1970). In the multidimensional model, voters and parties are located on several dimensions representing either different policies or different ideological concerns. The theoretical perspective underlying both the unidimensional and multidimensional spatial model is that people have a preference for a specific position in an ideological or policy 'space' and choose the party whose position in the space is closest to their own.

Recently, Rabinowitz & Macdonald (1989) have argued that most voters do not have specific ideological or policy preferences. Rather, they contend that voters have diffuse preferences which lead them to favor one side or the other in a policy debate. Rabinowitz and Macdonald construct a 'directional' model of voting based on this idea. In the directional model, issues for the electorate are dichotomous with voters differing in terms of both which side they favor and the intensity they feel with regard to each issue. Parties differ in terms of the direction of their stands and the emphasis they place on various issues. An implication of the directional model is that parties in multiparty systems should generally be non-centrist on at least some issue in order to attract voter support (Rabinowitz et al. 1991).

Both the traditional proximity model and the directional model are 'spatial' in the sense that voters and parties can be represented as points in a political space. They make different predictions, however, about how successful parties will behave. In general, the traditional spatial model suggests that parties should locate near large densities of voters; parties should be centrist if the voters are centrist. The directional model suggests that parties should be non-centrist and non-extreme. By observing the general relationship between party and voter positions in the spaces, we will be able to discern whether the parties tend to gravitate toward clusters of voters, as would be anticipated from classic spatial theory, or whether the center areas of the space are devoid of parties, as would be predicted by directional theory.

Prior Research

Research on party systems in Europe at both the theoretical and empirical
level has focused on the dimensional structure of the party space (Lipset
& Rokkan 1967; Inglehart & Sidjanski 1976; Converse 1966; Converse &

Valen 1971; Valen 1981; Daalder & Rusk 1972; Rusk & Borre 1976; Weisberg & Rusk 1970; Nannestad 1989). Empirical studies suggest that two dimensions are necessary to represent the basic conflict structure in most of the societies, while in some cases more dimensions may be required. Empirical analyses have consistently found that one dimension corresponds to the left-right conflict with the character of additional dimensions dependingon national setting in which the analysis takes place.

Our work will differ from most earlier efforts to assess mass perceptions of multiparty systems in that we will place individuals as well as parties in the space. Theoretical work in this area has stressed the importance of the relationship between party locations and voter distributions, yet the empirical studies in European settings to date have generally been restricted to party placements.! We shall analyze several countries - Norway, Sweden, Denmark, Germany, the Netherlands and France - and thus be in a position to place our results in a comparative context. The survey data cover a span of almost twenty years, from 1967 for France to 1985 for Norway. The reason for this is mainly variation in data availability among the countries. The spread of the time-frame does not concern us excessively as we are investigating questions that are not linked to any particular period or to any subset of electoral democracies.

Data and Methods

The data used to construct the spatial mappings are sympathy ratings of parties (and sometimes party leaders or social groups). The basic question format was developed in the United States as a 'thermometer' question where survey respondents are asked how warmly they feel toward various objects. Responses can range from 0 to 100, where 100 indicates very warm or favourable feelings and 0 indicates very cold or unfavourable feelings. This question type has been modified in various ways for use in different national settings, but in each country respondents indicate their sympathy toward political objects on a scale with a reasonably large number of response alternatives. The questions have proved to be excellent predictors of voting behaviour (see, for example, Weisberg & Rusk 1970; Valen 1981; Nannestad 1989), and thus there is no concern that they lack political meaning.

The spatial mappings are constructed based on two assumptions: (1) the more favorably a respondent evaluates a party, the closer the individual should be located to the party; (2) there is common agreement across respondents about the positions of the parties and the importance of the various dimensions. These assumptions are no more stringent than those of other standard statistical procedures. For example, regression, logit,

factor, and discriminant analysis all rely on the twin assumptions that the
basic model is 'correct' and that individuals respond according to identical
criteria.

The structural scaling proceeds in three stages. In the first stage, the lineof-sight (Rabinowitz 1976, 1986) is used to assess the similarity between the political parties. In the second stage, the similarity information is used to construct a mapping of the parties using ALSCAL, a non-metric multidimensional scaling routine (Young et al. 1978). In the third stage, individuals are located in the party space based on their preferences for the various parties, again using a non-metric multidimensional scaling procedure.

Approaches similar to ours have been criticized for recovering spaces of low dimensionality that are difficult to interpret (Budge & Farlie 1978; Knutsen 1989). While these criticisms have some merit, they also serve to point out the advantages of the approach. One of the real strengths of the method is in providing a reasonable overview of party competition in a space of low enough dimensionality for visual examination. This is only practical in one or two dimensions. Thus, while it is possible to analyze solutions of three or even more dimensions using this method, the bias is towards lower dimensional solutions.

Another characteristic of our approach is that the scaling is based exclusively the sympathy people express toward the political parties. No external criteria such as ideology or social or political attitudes are used to locate the parties in the space. While this necessarily leaves the dimensions of the space open to interpretation, the advantage of the approach is that researchers do not impose their view of the structure of the party system by selecting specific variables for analysis. Appropriate explanatory variables be introduced later as aids to interpreting the structure, but the spatial configuration reveals the interrelation between parties, exclusively in terms of how voters evaluate them.

The spaces recovered using this methodology reveal how the mass public perceives and structures the main components of electoral competition. No other procedure commonly used to analyze political systems has the capacity display this structuring. In addition, when individual points as well as object points are located in the space, it is possible to observe the relationship between voters and parties. In this way, the mappings can provide an important means for evaluating theories of electoral competition.

Norway

We will begin our empirical analyses with an investigation of the Norwegian

party system. This analysis will be somewhat more extensive than for any of the other countries. Selecting Norway for a more thorough treatment was suggested by our reasoning that a more detailed analysis of one case would be heuristically useful.

Figure 1 is a map showing the seven major parties in Norway in 1985 and the Party leaders. In each country when data for leaders were available, they were scaled along with the political parties. This provides a far sounder statistical basis for the spatial configurations as it doubles the number of points scaled.²

The space has been rotated so that the horizontal dimension corresponds as closely as possible to the left-right positioning of the parties as perceived by the mass public. In Norway, as in most of the countries included in the analysis, respondents were asked to locate the parties on a left-right scale. Since we have this independent perceptual information, it is possible to determine analytically how well the spatial structure corresponds to the left-right placement of the parties obtained directly in the interviews. This is done by regressing the left-right perceptual data on the spatial structure locations. In the Norwegian case the fit is good, with the multiple correlation between the structure and mean left-right perceptions of the parties 0.97. The adjusted R squared value is 0.90.

Looking at Figure 1, we see that the parties and leaders are widely dispersed on both dimensions of the space. In general, the leaders lie close to their respective parties. The only two parties tightly grouped are the Christian and Center parties. Consistent with the left-right orientation of the space, the Socialist Left, Liberal and Social Democratic parties lie in the left half of the space, while the Christian, Center, Conservative and Progressive parties lie in the right. The vertical dimension separates the parties into three groupings with the Christian, Center and Liberal parties defining the high end of the dimension, the Socialist, Social Democratic and Conservative parties the middle area, and the Progressive defining the bottom.

It is interesting to note that the leaders of all the parties except the Liberals (Dørum) are perceived to be more centrally located than their parties. This is most striking in the case of the Socialist Left Party where its popular leader, Kvanmo, is located close to the Social Democratic Party and its leader, Brundtland. Nevertheless, popular or centrally located leaders did not substantially enhance the electoral fortunes of their parties in 1985 (Listhaug 1986).⁶

Figure 2 shows the joint distribution of parties and voters. The parties are located exactly as in Figure 1, and the voter positions are indicated by dots. The voter locations are determined on the basis of the entire party and leader configuration shown in Figure 1, but the leaders are omitted to avoid cluttering the figure. We will follow the general practice of omitting the leaders as they are usually of less interest than the parties and greatly complicate the figures.

The most striking characteristic of the joint distribution is that the vast majority of voters are located internal to the party positions. The density of voters is greatest in the middle area of the plot. In particular, there is high density in the area bounded by the quadrilateral formed by the Liberal, Christian, Conservative and Social Democratic parties. From the perspective of traditional spatial theory a party positioned in the center of that quadrilateral would appear to be extremely strong electorally, but no party occupies that region of the space. This absence is consistent with directional theory.

The location of various demographic groups appears in Figure 3. In general, we have classified respondents by age, education, income, objectivesocial (based on occupation), subjective social class and religion. (A detailed description of the coding of the categories appears in the appendix.) When the location of a demographic group is very close to the population center, denoted '*', we have not plotted the group to avoid too many overlapping labels. The rationale for the set of demographic variables that we use is found in the theory of cleavages. The formation of parties and the voting choice of citizens in most Western European democracies

are seen as determined by the structure of the cleavages of class, culture and geography (Rokkan 1970). Class and - in most countries - religion fuel the left-right conflict and make this the most decisive dimension in partisan evaluations and voting choice.

We have also included the mean position of each of the party support groups based on the reported vote of each individual. As we would expect, these groups are far more dispersed in the space than the demographic groups as they reflect people's political preferences directly. While we would expect the demographic groups to be located more centrally than the explicit voting groups, we would still anticipate some real differences between the positions of the demographic groups. The more the politics is rooted in the demographic structure of the society, the more the demographic will show intergroup distances comparable to those of the political groups. In the Norwegian case, an appropriate comparison is the distance between the various demographic groups relative to the distance between the supporters of the Social Democratic and Conservative parties.

In Figure 3 we see that the most distinctive groups in Norway are based on class. Those who have traditional working class jobs (symbol 'w' in the figure) are separated from the traditional middle class (m+) and farmers (f). The mean of those in the new middle class (m-), composed largely of non-professional, white-collar occupations, lies close to the center. Selfidentifiedclass

identifiedclassshows even more spatial dispersion with the self-identified working class (w') distinct from the self-identified middle class (m'). The distance between the two self-identified class groups is over 60 percent of the distance between the two main voting groups, a sign of strong classbasedparty in the society. The farm group is also very distinctive, lying adjacent to the location of the supporters of the Center Party. This link is natural because the Center Party is the agrarian party in Norway, but the extent to which the set of all farmers and the supporters of the Center Party overlap is striking all the same.

The vertical dimension of the configuration should not, in our view, be given a single factor interpretation. The general structure we recover is quite similar to that of earlier spatial studies of Norwegian politics (Converse & Valen 1971). In these studies the second dimension was interpreted as largely religiously based. The group dispersions here suggest it is partly religious, with those who attend church (r+) fairly high on the second dimension. But it appears to be more generally a cultural dimension which incorporates elements of center-periphery and urban-rural as well as religiosity. This reflects the overlap of these factors in influencing Norwegian politics.

In Norway, the basic legislative coalitions in the first half of the 1980s were the Social Democratic Party and the Socialist Left Party on the one hand, and the Conservative, Center and Christian parties on the other. The Progressive Party has also normally been part of the bourgeois camp, while the position of the Liberals has been rather unstable. This is consistent with the mapping we have obtained. In addition, the map shows that the bourgeois coalition is dispersed, with some distance separating the Progressives from the Christian and Center parties. This suggests a high potential for stress within the bourgeois grouping. Stress has been evident in post-1985 politics where, on some issues, the Christian-Center pair (more often the Center Party) has deserted the Conservatives, and on other issues the Progressives have broken ranks. This has created a situation in which the Social Democrats have been able to control the government even though the socialist parties have been in the minority.

The Norwegian case illustrates the general approach we shall follow. For each country we will present a mapping of the joint party-voter distribution and a mapping of the major demographic and party voting groups. We will then briefly discuss the structure of the two figures in each country. With regard to the joint plot, we will be most interested in the question of how the location of the parties corresponds to classic spatial strategic theory. We saw in the Norwegian political system that for the most part the parties were more peripheral than the voters, and the population center was not occupied by any party. With regard to the group plot, we will be most interested in the degree of separation between the demographic groups. In

Norway we observed that there were fairly strong class divisions and that
religiosity had a real bearing on spatial position.

Before leaving the Norwegian case, it is useful to contrast the structure of our results with those of Knutsen (1989) because it illustrates the strengths and weaknesses of the differing approaches. In a specific attempt to separate supporters of the various parties using discriminant analysis, Knutsen recovered four dimensions in Norway. This reflects the fact that it is possible to discriminate between the supporters of the three parties lying high on the second dimension based on demographic and attitudinal characteristics of the respondents.

We would argue, however, that the Knutsen (1989) results do not present an accurate picture of political competition in Norway. For example, the party which is closest to the center of the political space is the Progressive Party, and the supporters of the Christian and Center parties are farther apart than the supporters of the Labor and Conservative parties. Such a structure lacks validity because it is clearly antithetical to the coalition behavior of the parties. In general, the correlation between the interparty distances, as estimated in the Knutsen analysis, and those based on party sympathy is an extremely low value of 0.06. Thus, while the discriminant analysis approach is useful for distinguishing among party supporters, the sympathy mappings provide a more meaningful sense of the structure of electoral competition in the society.

Sweden

For many social scientists, Scandinavia is synonymous with Sweden. It is not possible to take the truth value of that proposition to task here, but with regard to the party system Sweden is clearly the deviant case. It is the only country in Scandinavia that has not yet experienced a substantial break-up of the old five-party model, and it is the only country where the Social Democrats are still the predominant political force (although they were in opposition from 1976 to 1982). Earlier analyses of the Swedish party system have demonstrated a strong unidimensionality in which the left-right orientation dominates voters' perceptions of the parties (Sårlvik 1974, 1976). We have replicated this finding for the 1979 election (see Rabinowitz et al. 1991). In 1982, when two new parties were included in the survey, we obtained a two-dimensional solution. This is displayed in Figure 4a.

The horizontal left-right dimension shows the five old parties located in the familiar rank order with the Communists farthest to the left, then the Social Democrats, and at a considerable distance the three bourgeois parties - the (agrarian) Center Party, the Liberals and the Conservative

Party. The two new contenders, the Environmental Party and the Christian Party, are located in the middle of the left-right axis. These parties, however, are perceived as extreme on the vertical dimension, where the Liberals and the Center Party are also close to the new contenders.

It is difficult to give a precise interpretation to the vertical dimension, but it seems reasonable that cultural aspects play a role as the Center Party and the Liberals, as well as the two new parties have advocated views in opposition to the more narrow economic ideologies of the Social Democrats, the Conservatives, and - to some extent - the Communists. A smallest space analysis of the 1968 election by Sårlvik (1976) also derived two dimensions. In his analysis, the poles of the second dimension were defined by traditional moral groups like the temperance movement and the free church at one end and Vietnam demonstrators on the other end. In our analysis, the environment protest and religious fundamentalists are joined at the same end of the dimension. In assessing the dimensionality of the Swedish data, we should also keep in mind that the new parties are extremely small and did not obtain parliamentary representation: the Christians polled 1.9% and the Environmentalists 1.6% in 1982.

Figure 4a also illustrates some of the coalition problems of the bourgeois
parties, as the Conservatives especially are perceived as distinct from the

Liberals and the Center Party. This is reflected in their inability to form a viable three-party government in Sweden. The most recent demonstration of this was the formation of a government by the Liberals and the Center Party after the majority coalition broke apart in 1981.

Turning our attention to the interplay between the party and voter distributions, we find that the density of voters is greatest internal to the parties. Again, no party is located in the center area of the space. This pattern is quite consistent with directional theory.

In the Swedish case we did not have access to measures of religiosity and class identification; this makes the demographic analysis less complete than for Norway. In Figure 4b the groups are plotted with the voters for the parties. Class is an important cleavage among the Swedish electorate as working-class persons tend to be closer to the Social Democrats and the old middle class closer to the bourgeois parties, especially the Conservatives. farmers are closest to the Center Party, while the new middle class is fairly central. Union members are even closer to the Social Democrats the working class, a reflection of the importance of the labor unions in Sweden in mobilizing the Social Democratic vote.

Denmark

The Danish party system is more fragmented than the Norwegian or the Swedish. This is primarily a result of the electoral system where a 2 percent popular vote for a party translates into a mathematically correct representation. But, as Vogt (1988) has shown, other factors contribute as well. Low state subsidies for political parties do not give established parties the edge that they enjoy in a number of other countries, and access to freetime television for even the smallest political party also stimulates the formation of new parties. These institutional factors facilitated the erosion of the established five-party system in the 1960s and early 19705. The 1973 election was a watershed event with the five old parties (Communists, Social Democrats, Radicals, (agrarian) Liberals and Conservatives) receiving only 62 percent of the vote., and the two new parties of distinctively protest character, the Progressive Party and the Center Democrats, receiving 16 percent and 8 percent, respectively (Borre 1985). Through the last half of the 1970s and early 1980s, the Danish system has shown some tendency to move toward stabilization and renewed support for the old parties (Goul Andersen 1986). The data we present here are from the 1979 election, which can be seen as fairly 'normalizing' in that the old parties gained and the new parties lost. The Social Democrats performed especially well, approaching their strength of the 1950s and the first half of the 19605.

Figure 5a demonstrates the existence of a relatively clear left-right ordering of the parties.⁴ On the far left, three parties occupy the space: the traditional Communist Party, the Left Socialists which split from the Socialist People's Party in 1968, and the Socialist People's Party which from its start in 1960 has been the main contender on the far left in Danish politics. The Social Democratic Party occupies a fairly central position in the space not very far from the Radicals. The earlier analyses of Rusk & Borre (1976), based on the 1971 and 1973 elections, showed the Radicals much closer to the bourgeois block. However, Nannestad's (1984, 1989) scaling for the 1979 election also places the Social Democrats and the Radicals fairly close together on the left-right dimension. The movement in the spatial position of the Radicals during the 1970s is consistent with its shift in coalition behavior as the party moved away from the bourgeois parties and closer to the Social Democrats. It might seem counterintuitive that the Center Democrats occupy a position to the right of the Conservatives 1979, but this finding is also corroborated by Nannestad (1989), who shows that the party has moved from a distinct center position towards the right from 1973 to 1979.

Our interpretation of the second, vertical dimension follows that of Rusk

& Borre (1976) and Nannestad (1984), who see this as a distrust dimension
on which the Progressive Party is particularly distinct.

The distribution of the voters in Denmark is somewhat different from the pattern in Norway and Sweden in that the Social Democrats are located near the center of the space. Compared to the Norwegian and Swedish social democratic parties, the Danish party is known as ideologically moderate and has compromised more with bourgeois parties (Madsen 1984; Esping- Andersen 1985). The centrist positioning of the party, however, has not led to success at the polls. From the traditional spatial perspective it constitutes something of a paradox that the more extreme social democratic parties in Sweden and, to a slightly lesser extent, in Norway, have been more successful in elections.

With the relatively central location of the Social Democratic Party, the spatial structure in Denmark is more in keeping with the traditional spatial model than the structure in either Norway or Sweden. There is, however, substantial population density within the partial ellipse formed by the full set of parties in which no party is located. The pattern in Denmark is thus somewhat ambiguous as to its general theoretical implications. Of the three Scandinavian countries examined, it is the only one that does not directly challenge the traditional spatial model.

In Figure 5b we have plotted the demographic and voting groups in the party space. Both objective and subjective class are quite dispersed in the space. As in Norway, objective working class and middle class (w and m) are more centrally located than class identification (w' and m'). Religion shows a strong dispersion with the religiously-active closer to the main bourgeois parties and the non-religious located between the Social Democrats the Socialist People's Party.

Germany

The German party system is much simpler than the Scandinavian systems. The sister parties CDU and CSU on the right, and the SPD on the left form the main governing alternatives. The liberal party, FDP, has been alternating its support between the CDU/CSU and the SPD. In recent years the Greens have emerged as a fourth force in German politics. This party is partly concerned about environmental questions, but is also very anti-establishment and anti-authority (Miiller-Rommel 1985).

The dimensional analysis of the parties (Figure 6a) clearly shows the dichotomous nature of the German system with the CDU and CSU together on the right and the SPD on the left. The FDP is located fairly close to the CDU/CSU as it is seen as part of the same governmental coalition. The second dimension is probably an ecological dimension with the Greens as

the extreme party. It is obvious that the Green Party is also far to the left and hence distinct from the Swedish Environment Party which had a much more central position on the left-right scale. The density of voters is highest near the main parties (the CDU/CSU and the SPD) and in the center of the space, where no party is located. Thus, the pattern of mass and party locations in Germany, like those in Norway and Sweden, appears more consistent with the predictions of directional theory.

Figure 6b shows that the CDU/CSU is most closely supported by farmers, religious people, persons above 65 years of age, and the old middle class. The groups that are closest to the SPD are the non-religious and union members. It is interesting to see that the youngest age group and the highly educated are located somewhat between the SPD and the Greens, but much closer to the SPD. Class does not seem to differentiate very much among the voters in Germany. This is consistent with the finding of Inglehart (1987, 1297) that German class-voting is at a comparatively low level.

The Netherlands

Until the mid-19605, politics in the Netherlands could be described by the

concept of pillarization or verzuiling. That means that the main political parties were organized around specific segments of the society as defined by religion and class. The main pillars were represented by three religious parties - the catholic KVP, the calvinist ARP and the Dutch reformed CHU - and two secular parties - the social democratic PvdA for the working class and the liberal VVD for the middle class (Andeweg 1982, 15-17). From 1945 through to the 1963 election, these parties received 85 percent or more of the vote. In 1967 the electoral share of the big five fell by almost 10 percentage points, and in 1971 (the election where our data were collected) it reached a low of 71.9 percent (Andeweg 1982, 18). The decline was mostly in supp>ort for the religious parties, in particular the KVP, and reflected the effects of long-term secularization as well as shorttermpolitical Later, in the 1970s and in the early 1980s, the system stabilized somewhat (Eijk & Niemoller 1985, 350). The volatility of Dutch politics and especially the large number of parties contesting elections is due largely to the extremely liberal electoral law whereby a party which receives 0.67 percent of the vote gains representation in parliament.

Our dimensional analysis of the 1971 election includes 13 parties, one less than the number that actually elected members to the parliament.⁵ The structure of the political space in the Netherlands reveals the strong leftright of the parties. The three main religious parties (KVP, ARP and CHU) are all located on the center right, with the two fundamentalist calvinist parties (SGP and GPV) even further to the right. The main secular party, PvdA, is slightly left of center and is almost overlaid by the D66. Further to the left are three minor parties: PPR, the left-wing offshoot of the KVP; the pacifist PSP; and the Communists (CPN). DS7O, which left the PvdA in 1970 as a protest against new left influence in the social democratic party, is located more to the right and also lower in the space. The liberal party, VVD, is located on the center right but is more distinct on the second dimension.

On the vertical dimension, the farmers' party (BP) forms the opposite pole. The BP is not a traditional agrarian party, but has strong Poujadist elements of anti-tax protest and anti-bureaucracy attitudes (Lijphart 1974, 236). Since the BP is so distinct in this dimension, it is tempting to label the vertical axis a protest dimension - although the fit for some of the smaller parties that could also be subsumed under the protest label is not very good. Earlier dimensional analyses of Dutch data based on members of parliament, also produced two-dimensional solutions, but it was difficult to give a substantive interpretation to the second dimension (Daalder & Rusk 1972).

The location of the parties in Figure 7a gives a good picture of the main
coalition alternatives in the Netherlands, with the cluster of the religious
parties at the center of power and the PbdA and VVD vying for influence

(Daalder & Rusk 1972,183). The distribution of voters is spread out in the center of the space, mainly in the area between the PvdA, VVD and the KVP. From the standpoint of traditional spatial theory, it is remarkable that no party is located in this area. Thus, the Dutch case also appears to favor the expectations of directional theory.

The plot of the demographic and voting groups (Figure 7b) shows that the religious groups are fairly close to their parties, with the notable exception of the Dutch reformed church. The socioeconomic groups are more central, but with low income groups, workers and persons with little education located between KVP and the PvdA. The highly-educated, the high-income groups, and the middle classes are drawn towards the DS7O and the VVD.

France

In his seminal article on political distances in multiparty systems, Converse (1966) showed that the French parties of 1958 could be described in a twodimensionalspace the horizontal axis was the left-right dimension and the vertical axis was defined by the religious factor. Our analysis of

French parties employing data from 1967 does not easily replicate this tinding. Figure 8a demonstrates that the familiar left-right pattern is recordedalong horizontal axis, but the religious divide is not apparent in the recovered configuration. In 1958 it was primarily the MRP that defined the religious pole in the space at the same time that the party was perceived as fairly centrist on the left-right continuum. The MRP was formed after the war by catholic resistance leaders and in the first years had considerable success partly because it was supposed to be de Gaulle's party and partly because it was not associated - like some of the parties in the center and on the right - with the Third Republic or the Vichy regime (Converse & Pierce 1986, 24). The party declined through the 1950s and 19605, mainly because de Gaulle created his own party and the MRP no longer filled the role of a strong anti-Communist force. The party also drifted to the right, and in 1965 the MRP presidential candidate received the support of anti- Gaullist conservatives (Converse & Pierce 1986, 31).

While the left-right fit between mass perceptions and spatial position is excellent (multiple R = 0.94), there are several small discrepancies which place the MRP somewhat further to the right and the PSU somewhat further to the left than the perceptual data do. In the case of the MRP, this difference can be accounted for by the electoral deterioration of the

party, which made the MRP marginal in the minds of French voters.

The Radicals were also losing competitiveness as a party and are thrown
somewhat outside on the second dimension, which seems to separate the
competitive from the non-competitive parties.

The density of voters is highest in the area of the quadrilateral formed by the SFIO, Radicals, MRP and UNR. The pattern of high-voter density in this center area with no party in the center region provides yet another case that is consistent with the predictions of directional theory and contrary to the predictions based on the traditional proximity model.

The relationship between the voting groups and the parties (Figure 8b) might be a bit confusing because a number of the parties were involved in electoral alliances so that the supporters of the party did not cast their votes directly for the party. The most important alliance was the FGDS on the left which united the Socialists (SFIO), the Radicals and some groups of independent voters. On the center right, the Democratic Center (CD) united the MRP aad the groups that had supported Lecanuet for president in 1965.⁶ It is striking that the voters of the Federation (FGDS) as well as the supporters of the tiny left-socialist PSU are located close to the Socialists. Communist voters are located between the PCF and the SFIO. Both the voters of the Democratic Center and the Gaullists are fairly central.

Religion is the most dispersed cleavage in the French space with the nonreligious near the SFIO and the religiously active located closer to the UNR and MRP but at a considerable distance from both. Class is less dispersed than religion, with the working class closer to the leftist parties and the old middle class slightly towards the right. Union members are located between the SFIO and the PCF, which demonstrates the strong communist influence on French labor in that period.

Conclusion

We have examined the spatial structure underlying electoral competition in six Western European democracies. Our goal in performing this analysis was twofold. First, we were interested, in a purely descriptive way, to see how competition was structured in each of these democracies. Second, we were interested in the implication of the results for theories of mass-elite interaction.

Structure of Party Competition

In general, the party structures show a remarkable degree of similarity. In
each country it is possible to represent the broad structure of electoral

competition in a two-dimensional space, with one of the dimensions strongly related to the left-right axis and the other taking on a different character in different countries. This pattern suggests that there is a fairly simple general structure to citizens' perceptions of the party system which should provide a clear sense of which parties are likely to join coalitions together and what impact, in a generic left-right sense, a vote is likely to have.

The results also suggest that there is quite a lot of differentiation beyond basic left-right distinctions. In every one of these six countries the parties are widely spread on the second dimension, and any attempt to scale them without the second dimension would have met with failure. In general, there is a strong cultural character to the second dimension which suggests that it summarizes several dimensions of cleavage which a more fine-grained analysis could separate. Nevertheless, the cultural cleavages overlap enough for two dimensions to provide a meaningful representation of party competition. There is also a general tendency for parties with distinct protest character to be located at an extreme on the second dimension. This is true of the Progressives in Norway and Denmark, the Greens in Germany and the BP in the Netherlands.

In terms of the general cleavage pattern, we find consistent class differentiation the spaces. This is distinctly weaker in Germany, where age and educational differences show more spatial differentiation. In every country (with the possible exception of Sweden where no measure was available) religiosity has a bearing on spatial position. In France and Germany both class and religious differentiation parallel the left-right dimension. In Norway and the Netherlands these differences tend to show up on the second dimension, but even in these countries, religious parties are perceived to be on the right or center right. Thus, the extent to which the features of class and religion overlap differs across societies. In no country are the class or religious distances as large as the distances between the main party voting groups. This is consistent with the expectation that cleavages would not be fully polarized in pluralist democratic polities.

Theoretical Implications

We had a second interest in examining these spatial configurations: we were interested in the implications of the structural results for theories of interparty competition. The basic thrust of the class Downsian spatial model is that parties compete for ideological locations with high density of voters. In a multiparty system, we might expect parties to be fairly dispersed but we would also expect that wherever there were large numbers of voters in a position to be courted, some party would do the courting.

Mass-elite analyses such as we have presented here have been performed
in the US political system using candidates in place of parties (Poole &

Rosenthai 1984; Rabinowitz 1978). In these analyses the voters tend to occupy the center areas of the space while the candidates tend to be peripheral. This pattern is difficult to accommodate with traditional spatial theory, yet the theory is so widely accepted it is important to consider why the pattern might occur. In the US a three-part explanation has been developed that retains consistency with the basic spatial model. First, the two established parties are drawn to one side or the other of the left-right distribution by the pull of relatively extreme activists and the past traditions of the party. Second, the institutional structure makes the formation of new parties so difficult that there is no need to worry about the entry of a competitor in the center. Third, once a candidate is colored by a party label, the candidate is no longer free to maneuver to an optimal spatial position.

To the extent that this explanation is valid, it suggests that in systems where many parties compete, parties will be freer to maneuver to enhance their appeal to voters. Thus, in multiparty systems the center should not be vacant. Yet the same empty-center pattern is evident in every one of the multiparty systems we have examined: large areas in the central region of each of the spaces are rich in voters and devoid of parties. Even in the case of Denmark, the single country which appears to best fit the traditional spatial model, the distribution of parties runs contrary to the model. And the party with the most explicitly centrist orientation in the Danish system, the Center Democrats, fails to achieve anything like a central position. Nannestad (1989), who analyzed all of the Danish elections from 1971 to 1979 using a method similar to ours, remains generally favorable to the traditional spatial model. Yet he concludes his book with a reflection on the inexplicable nature of party positions based on the idea of rational party strategies.

Daalder (1984) has stressed the difficulty of making sense of the center in European party systems. He quotes Duverger (1964) who argues that the center cannot exist in politics because all issues are essentially bipolar. The directional theory is consistent with Duverger's premise and seems to provide greater insight into the structure of the party systems we have analyzed. The thrust of directional theory is that parties must provide some clear issue appeal in order to be electorally successful, and thus the center regions of the mappings should contain no party. It is beyond the scope of this paper to pursue these questions further, but the theoretical importance of the analysis we have performed should be underscored. Combined with the research reported in the US, these results provide a strong indictment of the traditional spatial model and indicate a clear need for a theory of mass-elite interaction that will better explain party competition in democratic Perhaps directional theory can play that role.

---

NOTES

* Svein Åge Relling provided superb research assistance on this project. Data from the following studies have been used: Forse hungsgruppe Wahlen (Mannheim), German Election Panel Study, 1983 (ICPSR 8452); Felix Heunks, M. Kent Jennings, Warren E. Miller, Philip C. Stouthard and Jacques Thomassen, Dutch Election Study, 1970-1973 (ICPSR 7261); Philip E. Converse and Roy Pierce, French National Election Study, 1967 (ICPSR 7372); Soren Holmberg, Swedish Election Study 1982; Henry Valen, Norwegian Election Study 1985; Ole Borre et al., Danish Election Study 1979. The datasets were made available by the Interuniversity for Political and Social Research, the Swedish Social Science Data Service, the Danish Data Archives, and the Norwegian Social Science Data Services. Neither the original principal investigators nor the data archives bear any responsibility for our use of the data.

1. A notable exception is Nannestad 1989.

2. This increases the constraint on the solution by approximately a factor of four since constraint is proportional to the number of pairs of objects scaled. We have systematically the parties by themselves in countries with seven or more parties to observe the effect of including leaders in the plot. The basic structural results are quite similar in all cases as is the fit of the solutions. Hence, the greater reliability achieved by including leaders appears to create no distortions in the party spaces.

3. The pattern of internal location of party leaders should not be generalized as different results emerge in other settings (figures not shown in this paper).

4. The tiny (single-tax) Justice Party is excluded from the analysis because prior studies have shown that this party is very unstable and moves around in a circular fashion in the space (Pedersen et al. 1971; Nannestad 1984).

5. We have omitted the NMP, the retailers' party, due to scaling problems. The inclusion of this party tended to produce a three-dimensional solution as the party seemed to be perceived on a dimension of its own.

6. There was also an alliance on the far right (AR) which was not identified in the coding and could not be included in our analysis (for details, see Converse & Pierce 1986, 32-34).

REFERENCES

Andeweg, R. B. 1982. Dutch Voters Adrift. Leyden University. Department of Political
Science.

Borre, O. 1985. 'Denmark', in Crewe, I. & Denver, D., eds., Electoral Change in Western
Democracies. London: Croom Helm.

Budge, I. & Farlie, D. J. 1978. 'The Potentiality of Dimensional Analysis for Explaining
Voting and Party Competition', European Journal of Political Research 6, 203-231.

Converse, P. E. 1966. 'The Problem of Party Distances in Models of Voting Change', in
Jennings, M. K. & Zeigler, L. H., eds., The Electoral Process. Englewood Cliffs, N.J.:
Prentice Hall.

Converse, P. & Pierce, R. 1986. Political Representation in France. Cambridge, Mass.: The
Belknap Press of Harvard University Press.

Converse, P. E. & Valen, H. 1971. 'Dimensions of Cleavage and Perceived Party Distances
in Norwegian Voting', Scandinavian Political Studies 6, 107-152.

Daalder, H. 1984. 'In Search of the Center of European Party Systems', American Political
Science Review 78, 92-109.

Daalder, H. & Rusk, J. G. 1972. 'Perceptions of Party in the Dutch Parliament', in Patterson,
S. C. & Wahlke, J. C, eds., Comparative Legislative Behavior. New York: Wiley.

Davis, O. A., Hinich, M. J. & Ordeshook, P. C. 1970. 'An Expository Development of a
Mathematical Model of the Electrical Process', American Political Science Review 64, 426-48.

Downs, A. 1957. An Economic Theory of Democracy. New York: Harper & Row.

Eijk, C. van der & Niemoller, K. 1985. 'Netherlands', in Crewe, I. & Denver, D., eds.,
Electoral Change in Western Democracies. London: Croom Helm.

Esping-Andersen, G. 1985. Politics Against Markets. Princeton, N.J.: Princeton University
Press.

Goul Andersen, J. 1986. 'Electoral Trends in Denmark in the 1980's', Scandinavian Political
Studies 9, 157-175.

Inglehart, R. 1987. 'Value Change in Industrial Societies', American Political Science Review
81, 1289-1303.

Inglehart, R. & Sidjanski, D. 1976. 'The Left, the Right, the Establishment, and the Swiss
Electorate', in Budge, I. et al., eds., Party Identification and Beyond. London: Wiley.

Knutsen, O. 1989. 'Cleavage Dimensions in 10 Western European Countries: A Comparative
Empirical Analysis', Comparative Political Studies 21, 495-534.

Lijphart, A. 1974. 'The Netherlands: Continuity and Change in Voting Behavior', in Rose,
R., ed., Electoral Behavior: A Comparative Handbook. New York: Free Press.

Lipset, S. M. & Rokkan, S., eds. 1967. Party Systems and Voter Alignments. New York: Free
Press.

Listhaug, O. 1986. 'The Norwegian Parliamentary Election of 1985', Electoral Studies 5, 79-83.

Madsen, H. J. 1984. 'Social Democracy in Postwar Scandinavia', Unpublished Ph.D. dissertation,
University: Department of Government.

Muller-Rommel, F. 1985. 'Social Movements and the Greens: New Internal Politics in
Germany', European Journal of Political Research 13, 53-67.

Nannestad, P. 1984. 'Dimensioner i vælgernes opfattelse af partisystemet', in Elklit, J. &
Tonsgaard, 0., eds. Valg og vælgeradferd. Århus: Politica.

Nannestad, P. 1989. Reactive Voting in Danish General Elections: 1971-1979. Århus: Aarhus
University Press.

Pedersen, M. N. et al. 1971. 'Party Distances in the Danish Folketing 1945-1968', Scandinavian
Political Studies 6, 87-106.

Poole, K. T. & Rosenthai, H. 1984. 'U.S. Presidential Elections 1968-1980: A Spatial
Analysis', American Journal of Political Science 28, 282-312.

Rabinowitz, G. 1987. 'A Method for Ordering Object Pairs Consistent with the Multidimensional
Model', Psychometrika 41, 349-373.

Rabinowitz, G. 1978. 'On the Nature of Political Issues: Insights from a Spatial Analysis',
American Journal of Political Science 28, 793-817.

Rabinowitz, G. 1986. 'Nonmetric Multidimensional Scaling and Individual Difference Scaling:
The Spatial Representation of Similarities Data', in Berry, W. & Lewis-Beck, M., eds.
New Tools for Social Scientists. Beverly Hills, CA: Sage Publications.

Rabinowitz, G. & Macdonald, S. E. 1989. 'A Directional Theory of Issue Voting', American
Political Science Review 83, 93-121.

Rabinowitz, G., Macdonald, S. E. & Listhaug, O. 1991. 'New Players in an Old Game',
Comparative Political Studies (in print).

Rokkan, S. 1970. Citizens, Elections, Parties. Oslo: Universitetsforlaget.

Rusk, J. G. & Borre, O. 1976. 'The Changing Party Space in Danish Party Perceptions, 1971-1973',
Budge, I. et al., eds., Party Identification and Beyond. London: Wiley.

Sårlvik, B. 1974. 'Sweden: The Social Bases of the Parties in a Developmental Perspective',
in Rose, R., ed., Electoral Behavior. A Comparative Handbook. New York: Free Press.

Sårlvik, B. 1976. 'Mapping the Party Space: Distances, Evaluations, and Ideological Perspectives',
1976 IPSA World Congress, Edinburgh.

Valen, H. 1981. Valg og politikk. Oslo: NKS-Forlaget.

Vogt, G. 1988. 'Politiske verdier og partisystemer: En komparativ undersøkelse av Danmark
og Norge', Unpublished thesis, Department of Political Science, University of Oslo.

Weisberg, H. F. & Rusk, J. 1970. 'Dimensions of Candidate Evaluation', American Political
Science Review 64, 1167-1185.

Young, F. W. et al. 1978. 'ALSCAL: A Nonmetric Multidimensional Scaling Program with
Several Differences Options', Behavioral Research Methods and Instrumentation 10, 451—
453.