Apparently, political representation is an elusive phenomenon. 'In spite of many centuries of theoretical effort', says Heinz Eulau, 'we cannot say what representation is' (Eulau 1967, 54). Philip Converse and Roy Pierce are equally harsh in their judgements and talk about the literature on the concept of political representation as 'riddled with confusion' (Converse & Pierce 1986, 490). Hanna Pitkin, the author of the modern classic, The Concept of Representation (1967), is also very critical when she summarizes the literature. Old conceptual pillars like authorization and accountability she finds formalistic and devoid of content: 'Neither (concept) can tell us anything about what goes on during representation, how a representative ought to act ... whether he has represented well or badly.' (Pitkin 1967, 58; Jewell 1983, 304). Notions such as that legislatures should be mirror images of populations (descriptive representation) and more lofty concepts like that of symbolic representation are dismissed by Pitkin as unrealistic and incomplete.

Heinz Eulau continues to be critical and perhaps also puzzled after having read Converse and Pierce's book Political Representation in France (1986). In a review article he still talks about the 'mystery of representation' and wants to jettison most of the ideas about influence from below that he thinks 'pervade all of the current models of representative democracy'. A different model that acknowledges the existence of party elites and the prevalence of influence from above, is Eulau's solution. In his view, 'in all

Resumé

Using interview data from studies with voters and members of parliament, both in 1968/69 and in 1985, Swedish representational democracy is assessed. The analysis focuses on four linkage processes - social representation, role representation, policy representation and anticipatory representation. The results are compared with Converse and Pierce's from their study of political representation in France. In their analysis they find the French system falling 'short of its potential for much of the electorate much of the time'. Heinz Eulau, in a review article of Converse and Pierce's book, goes one step further, branding the French democracy as elitist and run by unresponsive parties. Looking at the results from the Swedish studies, an Eulauean conclusion seems a bit drastic, while a verdict of the kind Converse and Pierce delivered for France may be too lenient. Clearly, the Swedish system is far from any model of popular representation run from below. A more elitist model, with an emphasis on policy leadership on the part of the political parties, fits most of our results much better.

modern representative democracies it is the electorate that responds in a more or less active manner to the elites' policy initiatives, thus indeed having some "power" ... to hold the elites responsible within a policy framework set by the elites rather than by the citizenry' (Eulau 1987, 210-212).

As a critique of Miller-Stokes congruence model and the application of it by Converse and Pierce in France, Eulau's alternate model has merit. Clearly, the Miller-Stokes model is individualistic. The focus is on the degree to which individual congressmen represent their constituencies, and its perspective is influence from below, from voters to representatives. At least for the French case, Eulau suggests a more collectivistic opinion formation model in which '. . . party elites are the dominant formulators and proponents of public policies . . .' (211).

Heinz Eulau's model could be viewed as cynical or realistic, depending on one's own perspective, but as a model it is not a novelty; certainly not in Europe with its tradition of disciplined and ideological parties and experience with political opinion formation from above. In an American context, however, it may seem to be a new idea. But, of course, that is not the case. Eulau's alternate model has many features in common with the responsible-party model that has been discussed in the United States at least since the 1940s.¹

Interesting as it may be, the purpose of this article is not to make statements on varied linkage models or to pursue the generic meaning of representation. My purpose is more concrete and empirically oriented. Acknowledging that representation is a multifaceted phenomenon, I will present some results from an ongoing research program on national political representation in a non-French and non-American setting. That setting is Sweden.

As a Western European nation, the Swedish political system resembles the French system more than the American. Like France, Sweden has a multiparty system with extremely cohesive legislative party groups.² But in contrast to France, Sweden has an extremely stable party system and an election procedure that puts the parties, not individual candidates, at center stage.³ Thus, the Swedish system, even more than the French, is a case where some kind of a collectivist representation model centered on the political parties clearly is a more viable alternative than various forms of individualistic representation models.

Although results from other democratic countries will be discussed most of my data originate from two Swedish representational studies, one performedin late 1960s and the other in the middle of the 1980s. In both instances the data consist of a study with members of the Riksdag and interviews with a sample of the electorate.⁴ Data from both mass and elite levels are a necessity since all models of representation have a relational

character, ultimately dealing with influence or power distributions across
levels.

Since representation is a complex phenomenon it can not be approached head-on, especially if one wants to address empirical questions. The approach has to be more restricted and gradual, studying various aspects step by step. Therefore, I have limited my analysis to four representational processes in this article, starting traditionally with social representation, moving to role representation and policy representation, and ending more speculatively with anticipatory representation.⁵

Studying social representation - the degree to which elected representatives the same socio-economic traits as the electorate - is not an entirely idle endeavor. The mirroring aspect of social representation is not without intrinsic value. But, admittedly, mirroring studies are a bit flat if they are not coupled with analyses of whether different social traits mean anything when it comes to policies and decisions in parliaments. Therefore, our study will comprise both an analysis of the extent to which the social composition of the Swedish Riksdag reflects the social make-up of the electorate, as well as an analysis of the relationship between recruitment and policy attitudes and power positions in the Riksdag.

Social representation has been called 'standing-for' representation, in contrast to 'acting-for' representation, which involves concrete actions and a will to represent (Pitkin 1967). The role-analytical approach to analyzing representation has developed this idea further. The central thought is that representatives' will to represent constituents or other groups is a crucial ingredient in any representational process. John Wahlke and Heinz Eulau, who pioneered these kinds of ideas, made a distinction between various segments of the representatives' roles, including a party-role segment, a pressure group-role segment, and a representational-role segment (Wahlke et al. 1962; Wahlke 1987). The latter segment is the one that most often has been picked up by other researchers, although the original distinction between the style of representation (trustee, delegate) and the focus of representation (district, party, pressure group etc.) has not always been maintained. However, in our analysis of Swedish legislators' role conceptions distinction will be used and we will ask what difference, if any, different role preferences make to various forms of behavior in the Riksdag.

But, as Converse and Pierce point out, the will to represent is not a 'sine qua non for judging that any representation worthy of the name is occurring' (Converse & Pierce 1986, 664). Representation, even acting-for representation,can place without an intention to be responsible to a constituency. The obvious case is when representatives have the same policy stances as their constituents and follow their conscience when acting in parliament. Thus, policy congruence between legislators and the electoratecould one route toward creating a correspondence between

legislative actions and the wishes of the people. Such a correspondence could be called decisional representation, to be distinguished from policy representation, which involves issue opinions but no actions. Notwithstandingits decisional representation, or the basic representationbond, Converse and Pierce call it, will not be studied in this context. The analysis will be confined to the easier task of studying policy representation. Since the Swedish elite and mass studies included a large number of identically worded issue questions, it is possible to measure the extent of policy representation through direct comparisons between average issue positions of legislators and voters. Thus, the measures we will be dealing with are measures of collective representation.⁶ They involve the comparison of the entire Riksdag with the entire electorate, or comparisons of legislative party groups with their respective voter groups in the electorate.

A high degree of policy congruence between leaders and voters does not guarantee decisional representation, although it is usually a good beginning. On the other hand, lack of policy congruence does not preclude decisional representation. There are other representational processes that can contributeto representation when the route through policy representationfails. most effective of those alternative routes is representation through successful use of the technique of anticipation on the part of the representatives. Anticipatory representation occurs when legislators accurately read the mind of the people and perform accordingly. Going back to the discussion above about representational roles, it is obvious that a will to represent must be present in order for anticipatory representation to work. Hypothetically, delegate role preferences should be more conducive to anticipatory representation than, for example, trustee role preferences. However, a will to represent, no matter how fervent, will not suffice if the representatives* perceptions of their constituents' attitudes are inaccurate. In other words, anticipatory representation presupposes a good elite knowledge of voter sentiments. Representatives with accurate perceptions of their constituents' opinions could, if they so choose, provide a high degree of decisional representation. Knowledge and accurate perceptionsbecome concepts. Without accurate elite knowledge of what the voters wish, there can be no successful anticipatory representation. Of course, the importance of the elite knowledge factor has not escaped previous students of political representation.⁷ Politicians' perceptions of district sentiments is one of the four corners in Miller and Stokes' diamond model of representation (Miller & Stokes 1963), and Converse and Pierce (1986, ch. 20) devote a whole chapter in their book on French representation to elite knowledge of constituency sentiments. Thus, our analysis of the extent to which Swedish legislators correctly perceive the issue opinions of the electorate and of their own voters will not break new ground. But,

hopefully, it will give insights into how powerful the tendency to wishful
thinking and perceptual distortion can be, even in a system characterized
by strong disciplined parties.

Social Representation

The occupational composition of the Swedish Riksdag has changed rather drastically since democracy and universal suffrage were accomplished in the beginning of this century. The direction of change, however, has not been in substituting members with humble social origins for members from the upper or upper middle classes. That has not occurred. The proportion of working class members in the Riksdag, for example, was about the same in 1985 (9 percent) as in 1912 (8 percent), and the proportion of managers and professionals has not decreased; it was 27 percent both in 1912 and in 1985. Instead, the most important change that has occurred is that members with a background as farmers have almost disappeared in the Riksdag and have been replaced by full-time politicians. The farmers' share of the membership has decreased from 40 percent in 1912 to 7 percent in 1985. At the same time, the proportion of career politicians has increased from 5 percent in 1912 to 28 percent in 1985.

Looking at the family background of members gives the same picture. The social base of the Riksdag has not been broadened. The proportion of members brought up in working class families was roughly the same in the 1930s as in the 1980s (around 30 percent), while the percentage of members from upper middle class homes has increased somewhat, from 12 percent in 1937 to 19 percent in 1985.

If the decline of the farmers and the rise of full-time politicians is the most conspicuous change in the recruitment pattern to the Riksdag, another trend is perhaps the most politically important, at least at a symbolic level. That trend is the increased recruitment of members with a background in the public sector, although members from the private sector were still a majority in 1985.

It is an exaggeration to declare that Swedish members of parliament constitute a social elite. On the other hand, it is equally wrong to claim that they are 'an average mixture of ordinary people', as Harold Laski once asserted that British parliamentarians were.⁸ Of course Laski was wrong. British MPs, like Swedish members of the Riksdag, are not proportionally recruited from all relevant groups in society. They, like most other legislators the world, are primarily recruited from more privileged groups.

In the Swedish case age, occupation, ownership of dwelling place and
level of education were the socio-economic circumstances in which members
differed most from the electorate (see Tables 2 and 3). Compared to the

electorate, members were more likely to be middle aged, have upper middle
class occupations, own as opposed to rent their dwelling place, and have a
higher education.

Other socio-economic factors that were investigated (including gender, occupational sector, dwelling place in a home or in an apartment, religion, and father's occupation) revealed comparatively smaller differences between Riksdagen and the electorate. However, the differences that existed indicated that members as compared to voters more often were men, worked in the public sector, lived in private homes, were frequent church goers, and grew up in the middle class.

Moving away from descriptive social representation to questions pertaining the relationships between recruitment and behaviors of legislators does not necessarily mean that things get more exciting. Summarizing the predominantly American literature on the connections between recruitment and behavior of legislators in the Handbook of Legislative Research (1985), Donald Matthews reveals some disappointment when he states that

'. . . viewed as independent variables explaining the behavior of legislators
or as inputs shaping legislative systems, recruitment has been pretty much
awash' (p. 42).

Matthews negative conclusion is to a degree valid also for Sweden. Socioeconomic variables like gender, age, occupation, education etc. are not important explanatory variables for attitudes or behaviors of Swedish members of parliament. But, on the other hand, they are not totally uninteresting either. Some rather strong relationships do exist.

In order to give recruitment a fair shake we studied the relationship between a large number of social variables (eight in all) and a whole set of different issue items and ideological left-right scales (twenty-three in all) in a series of bivariate analyses both within party groups and in the entire

Riksdag. In order to avoid being confined to studying only attitudes as dependent variables, we also analyzed how the various social background variables were related to the members' formal power positions in the Riksdag.⁹

The results were fairly easy to interpret. In most cases there were no interesting relationships, either in the entire Riksdag or within the different party groups. The average correlations between the issue variables and the social variables was a mere 0.06 for the Riksdag as a whole and only between 0.02 and 0.04 within the larger parties.¹⁰ The comparative results for the relationship between the social variables and the power position index were slightly more interesting, with somewhat higher correlations, especially within party groups.

However, some of the results we found were not without relevance. For example, there existed a rather clear relationship between gender and policy position on issues. Compared to men, women members in all parties had a tendency to be more negative to nuclear power plants, more interested in spending money on environmental protection, more positive toward social welfare programs, and more conservative on moral matters.¹¹ Furthermore, terms of left-right ideological position, the female Social Democratic members were somewhat to the left of their Social Democratic male colleagues. Thus, in addition to the norm of equality, whether men or women are recruited to the Riksdag is not totally inconsequential, and may have some policy implications.

The relationship between occupation and left-right attitudes was less
pronounced in 1985, than it was in the Riksdag Study of 1969, but the basic

pattern is still the same. Professionals and full-time politicians tend to be to the left in the parties, while farmers and businessmen gravitate to the right. When it comes to workers as a group, they tend to be the most leftorientedin Riksdag. However, since most of them are Social Democrats,that reflects party influence more than class influence. Within the Social Democratic Party members from the working class did not lean to the left, on the contrary, they were slightly to the right both in 1969 and 1985.

Occupation also proved to be a relevant factor in analyzing the members' power positions, particularly within the bourgeois parties. In the Liberal and Conservative Parties, full-time politicians had the highest number of formal power positions, while businessmen and farmers had the least. The result was reversed among Center Party members. As the former Agrarian Party, in the Center Party farmers still retain more power positions than fulltime The relationships were weaker among Social Democratic members. But, ironically, within the Social Democratic Party working class members were the occupational group with the fewest number of formal power positions.

Role Representation

Swedish members of parliament are primarily party representatives. They are elected on party ballots and they almost always vote according to the party line in parliamentary roll calls. Without doubt, the parties, not the individual members, are the principal actors in the Riksdag.

The party dominance, however, does not seem to bother most members. When asked about representational focus, a clear majority of members (77 percent) viewed working for their own party's program as a very important task. Other tasks or focuses we asked about, were less popular. An errandboy ran a distant second, with 48 percent of the members asserting that solving problems for individual constituents was a very important duty for them. A district focus, working for the interests of one's own local constituency, came in third (44 percent). Focusing on representing various interest groups (wage earners, business, farmers, women etc.) was the least popular among the ones we asked about. Fewer than a quarter of the members considered interest group representation a very important task.

A representational focus on the party program does not necessarily mean that the representational style also has to be party oriented. At least in theory, it is conceivable to combine a party focus with a trustee style of representation. But, as expected, when it comes to representatives of flesh and blood, there exists a rather pronounced correlation between representational focus and representational style. Swedish members who

focus on representing the party have a clear tendency to also favor the
party loyalist role over the trustee or delegate role.

Our analysis of preferred representational styles involved both members and the electorate. Three interview questions with forced-choice alternatives used to elicit people's normative conceptions of which representational legislators ought to follow in parliament. The questions were of the following kind: How should a member vote if one's own opinion does not coincide with that of one's party? Should he go "with his own conscience or with his party? The opinion of his local constituents was, in turn, matched against party and conscience in two subsequent questions.

The three-piece question technique makes it possible to extract rankorderings use them to designate role preferences. A party loyalist is a person who puts party standpoint ahead of both own conscience and constituents' opinion, while a trustee and a delegate, respectively, ranks own conscience and voters' opinion highest.

Among members, the party loyalist role was clearly the most popular, although its dominance was not overpowering. A narrow majority of the members preferred the party loyalist role (51 percent), while 34 percent supported the trustee role and 15 percent the delegate role.¹² Thus, Edmund Burke, the most renowned among advocates of the trustee role, still has, at least in theory and when talking to political scientists, a substantial number of supporters among Swedish members of parliament.

Burkean individualism was also quite popular among the voters. A substantial 30 percent preferred the trustee role in the electorate. The trustee role was not the one most supported among voters, however. That position was occupied by the delegate role. A majority of the Swedish public (56 percent) preferred the members to assume a delegate role.¹³ Thus, the congruence between the role preferences of the members and the electorate was, indeed, very poor.

The Swedish results are not unique, either on the elite level or on the level of the mass electorate. The party loyalist role is the dominant one in many, although not all, studies on European legislators. And mass publics, both in Europe and in the United States, tend to prefer their representatives to assume delegate or trustee roles instead of the party loyalist role.¹⁴ Regrettably differing question wordings preclude more detailed comparisons countries and political systems. Role preferences are rather esoteric phenomena, and as such are very sensitive to how interview questions are phrased.¹⁵

Traditionally in Europe, the party loyalist role has been more readily accepted among socialist representatives than among centrist or conservative Non-socialist members usually tend to be more oriented toward the trustee role. Thus, to a large extent, role preferences among European lawmakers have been tied in to the left-right cleavage. In the 1969 Riksdag Study that was clearly the case, with a much stronger support for the party loyalist role among Socialist members than among Bourgeois members.

However, in 1985 much of this difference had disappeared. The party loyalist role had become more accepted, especially among non-socialist members, resulting in a weakened correlation between role preferences and the ideological left-right cleavage. For example, in 1985 the party loyalist role was as much supported among Conservative members as among Social Democratic and Communist members. Most supportive of the old bourgeois pro-individualistic and anti-party role conception were the members the centrist Liberal Party.¹⁶

With a dwindling connection between role preference and ideological position, it is an open question what the varied role convictions mean when it comes to legislative actions in the Riksdag. As Converse and Pierce put it when discussing the behavioral consequences of different representational styles, 'the proof of the pudding is often taken to be the degree of fit between verbal responses . . . and . . . actual choice behavior'. In the French case Converse and Pierce tend to view their often very weak correlations as proof of the meaningfulness of the role constructs. In Sweden, by contrast there is no doubt that the representational roles, as we have operationalized them here, are fairly uninteresting as predictors of the members' behavior or attitudes.¹⁷

Swedish members of parliament, no matter whether they suscribe to a delegate, a trustee, or a party loyalist role, do not behave much differently on roll calls or in other forms of legislative behavior. Thus, in most cases, the various role convictions serve only a decorative function. They are worn like ceremonial robes in church.

This being said, it has to be acknowledged that even if most correlations between role preferences and various forms of behavior and preferences are dismally weak among Swedish members, the directions of the correlations usually the expected ones. For example, members in all parties supporting the individualistic trustee role show a slightly lower degree of party fidelity on roll calls than party loyalists (Holmberg 1976, 24). Also, similar to Converse and Pierce's findings for France, members in Sweden leaning toward a delegate role preference, are a little better at accurately perceiving voter sentiments than other members.¹⁸

Notwithstanding these weak correlations, it is quite clear that representation preferences are not important determinants of legislative behavior among Swedish members of parliament. As a consequence, role representation is not a very useful concept in studying political representation Sweden.

Policy Representation

In a democratic system like the Swedish one, based on party representation, a high degree of policy congruence between representatives and the electorate not an obvious result. The reason is simple, although often overlooked. Aggregation and articulation of public sentiments is not the only function of parties. Besides mirroring attitudes in the electorate, parties are also supposed to mold public opinion. Populist notions of party representation tend to stress the first function, while elitist notions lean toward the second. The important point, however, is that both functions are integrated and legitimate tasks in party-based democratic systems.

Going back to policy congruence, the connection between degree of congruence and the function the parties emphasize is readily observable. We expect policy agreement between elite and mass to be comparatively low when the parties engage in opinion formation and try to change the will of the people. On the other hand, when parties are content with mirroring existing opinions, we expect higher degrees of policy congruence. Expressed differently and in terms of policy agreement within parties, we expect parties interested in preserving the status quo to exhibit higher degrees of congruence that parties who try to change society more or less radically.

Looking at the empirical results on policy congruence in Sweden the

most immediate problem is not the relevance of the above mentioned theoretical notions. Instead, the most pressing problem is a methodological one. Before we can start discussing different levels of policy congruence we must have valid and easily interpretable measurements. Since we are dealing with collective representation, correlation measures of the individualkind by Miller-Stokes and used by Converse and Pierce in their French study, will not do. The solution has to be some kind of proximity measures that indicate the distance or discrepancy between the opinions of groups of members and groups of voters.

Three such measures are presented in Table 5. They are applied to the results from twenty identically phrased issue questions posed to both voters and members of Riksdagen in 1985. The most complicated, but also the most analytically useful of the measures is the means difference measure. It shows the average divergence of members' and voters' opinions when response alternatives on all issue items have been scaled between 1 and 5, with 3 indicating middle positions, and excluding don't knows. Theoretically, means difference measure can vary between 0.0 (perfect congruence) and 4.0 (maximum difference). The two other measures are more easily interpretable. The percentage difference measure indicates the averaged difference between members' and voters' answers to dichotomized issue question, after don't knows and persons with middle positions (3s) have been excluded. The third measure, proportion of issues displaying different majority positions among members and voters, is calculated from the results of the dichotomized item analysis. And, as for the percentage difference measure, zero (0) indicates perfect congruence and 100 maximum difference.

Compared across issues or across parties, the three measures yield roughly the same results, which, of course, is comforting. However, if we are interested in evaluating the level of policy congruence, none of the measures is very helpful per se. We need relevant reference points before we can determine whether policy agreement between representatives and voters is high or low in Sweden.

Comparison with other countries is one such obvious reference point. The regrettable drawback with that possibility is that representational studies are still an underdeveloped research area, with few studies done and a slow development of measures useful in across-country comparisons. For example, when policy congruence is measured, many studies are usually based on very few issue questions, making generalizations very hazardous. The classical Miller-Stokes study included only three issue areas, Charles Backström, in a study of collective policy congruence in the United States, used six different issues, while Converse and Pierce in their French study utilized more issue items, but only five of them were worded identically in the mass-elite surveys.¹⁹

Backström studied the Ninety-Second Congress, elected in 1970, and found an average difference of 11 percentage points, between the public's opinion and that of House members. The comparable result for Senate members was 20 percentage points. Although a bit speculative, a comparison the Swedish results in Table 5 shows the level of average policy congruence between the Riksdag and the electorate to be in the same range as the American results. The average difference between the electorate's and the members' opinions across the twenty issue questions was 15 percentage in 1985.²⁰

Converse and Pierce did not rely totally on correlational measures in their French study. They also reported some results pertaining to collective policy congruence, which makes a tentative comparison between French and Swedish congruence levels possible. A recomputation of our means difference measure to make it comparable with the standardized measure

Converse and Pierce uses gives an average value across twenty issues of 0.33 for Sweden. The average French result for the five position questions (all dealing with European integration) included in Converse and Pierce's study is almost exactly the same, 0.34.²¹

Admittedly crude as these comparisons are, nevertheless, they give us some apprehension of the level of policy congruence in Sweden. And the results imply that Swedish congruence is not very different from what has been found in other countries. Apparently, it is no higher than in political systems like the French or the American, but neither is it lower.

However, across-country comparisons are not the only possible reference point. Another obvious reference point is to compare congruence levels across time. The most evident problem with such a strategy is that the results most certainly will be affected by which issue questions are included in the comparisons. In all political systems, policy focus and disputed issue areas change over time, meaning that the number of sampled issues must be fairly large in order to avoid haphazard results.

Since the number of issues included in the Swedish Representation Studies was not forbiddingly small - twenty issue questions were included both in 1985 and in 1968/69 - a comparison should not be too risky, although only two of the twenty issues were the same and identically phrased across the years.

Given the methodological uncertainty of the comparison, it is reassuring to find a close resemblance between the measured levels of policy congruence 1968/69 and in 1985. Across the years all of our three measures of congruence show very similar results, both on the level of the entire Riksdag and on the level of the individual parties. Furthermore, the small uniform displacement that is discernable is an expected one. Many of the changes that have taken place in the Swedish political system since the late 1960s - most importantly the constitutional reform which made elections more frequent and increased the proportionality between votes cast and seats received in the Riksdag - would lead one to expect a shift toward a higher degree of policy congruence.²² And that is exactly what we find in the results. The average percentage difference between the opinions of members the electorate has decreased from 20 percentage points in 1968/ 69 to 15 in 1985.²³ Moreover, the shift is not restricted to the results for the comparison between all members and the entire electorate. It is also observable within all parties, with the possible exception of the Conservative which exhibits roughly the same level of within-party policy congruence across the years.

No doubt, comparing levels of congruence across countries or across years in the same country are relevant and a worthwhile endeavor. But if one is interested in the more fundamental and evaluative question, whether or not given levels of congruence are to be considered high or low, the

across-country or across-time comparisons are not always helpful. For example, is an average difference of 15 percentage points between members' and voters' issue opinions (the Swedish result for 1985), in an absolute sense, to be evaluated as a high or a low level of congruence?

Using the notion of random sampling, and comparing the 15 percentage points difference with the average percentage difference we would expect to find if all members of the Riksdag (349) were selected randomly, instead of through the electoral process, the answer is that an average percentage difference of 15 points indicates a fairly low level of policy congruence. The possibility of getting a 15 percentage points attitude difference between the Riksdag and the electorate, if the members were chosen by a random process, is very small, less than one in a thousand.

The impression that our results do not indicate high levels of policy congruence is strengthened when one reflects upon the fact that no less than a third of the studied issue questions, both in 1968/69 and in 1985, displayed different majority positions among members and the electorate. It is difficult to reconcile a result like that with any notions of high degrees of policy representation. A contrary evaluation seems more apt.

When empirical levels of policy congruence brought about by the representational are markedly lower than would be the case if member selection were done randomly, there is little doubt that the democratic processes that tend to produce issue agreement between elite and mass have been less successful than the democratic processes that produce attitude differences between representatives and the electorate. Looking at the results across parties and across different issue areas gives clear cues as to what processes have been at work, producing opinion differences between electors and elected. Without simplifying too much, it is obvious that a pronounced tendency to policy leadership, sometimes combined with failed opinion formation efforts on the part of the political parties, goes a long way in explaining the rather poor congruence results in Sweden.

For example, a number of the issue questions exhibiting low degrees of policy agreement are typical policy leadership issues in which the party elites - most often the Social Democrats - have long since been struggling with recalcitrant voters. Among issues of this kind are long-lasting conflicts like the ones over commercial television, the level of welfare benefits and the wage earners' funds.²⁴ On other less politicized issues -a question about the future high-tech society can serve as an example - it is quite probable that the Riksdag opinion is much more informed and therefore ahead of public opinion. The high-tech issue question exhibited by far the lowest congruence level in our 1985 Representation Study, with members being much more positive toward the future high-tech society, and the general public being more negative and reluctant toward the changes.

That policy leadership and a will to change existing circumstances are

relevant factors is also indicated by the fact that the lowest levels of withinpartycongruence exhibited by the parties on the wings of the dominant left-right dimension, i.e. within the Communist and Social Democratic Parties to the left and within the Conservative Party to the right. On average, policy congruence was higher within the Center and Liberal Parties, the 'middle parties' in Swedish politics. A reasonable explanation for this U-shaped congruence pattern is that parties located toward the wings of a policy dimension have to stress policy leadership and opinion formation more than centrist parties. Parties at the wings more often find themselves in situations where they have to try to influence public opinion, either promoting or checking change.

Obviously a center-periphery explanation of this kind is dimension specific and has to be tested on different policy dimensions, not just on the left-right cleavage. However, the dominance of the left-right dimension in Swedish politics makes such a task difficult, since most issues are more or less tapped into the left-right cleavage structure. But thanks to the rather recent emergence of a green policy dimension, cross-cutting the old leftright it is possible, even in Sweden, to test the center-periphery hypothesis on more than one policy dimension.²⁵

Nicely U-shaped, the curves in Fig. 1 lend support to the notion that
centrist parties on different policy dimensions tend to be closer to their
voters' opinions than parties toward the wings.

The presence of a U-shaped pattern is clearly discernable, both on the two highly interrelated left-right policy dimensions (the social welfare dimension and the public vs.. private sector dimension) and on the green dimension.²⁶ The Center Party, located toward the middle on the two leftright areas and toward the ecology wing on the green dimension and the Social Democrats, placed toward the left wing on welfare and public vs. private sector issues and toward the middle on the green dimension are the pivotal cases. Consequently, it is reassuring to notice that the congruence across the dimensions for both these parties are the expected ones, even in an absolute sense. The Center Party displayed higher average levels of congruence between elite and mass on left-right issues than on green issues, while Social Democrats exhibited exactly the opposite pattern, with more agreement on environmental issues than on left-right issues.

Impressive as they are, it must not be forgotten, however, that our results only pertain to one year and one specific study. The support for the centerperipheryhypothesis less unequivocal if one looks at other Swedish and international studies.²⁷ For example, in the 1968/69 Representation Study, the Conservative Party - not the middle parties - was the party closest to its voters on public vs. private sector issues, while on social welfare issues Social Democrats and Communists displayed higher degrees of policy congruence than any of the three Bourgeois parties, whether they were

centrist or not. Thus, the U-shaped curves for within-party congruence
levels in the Swedish Representational Study of 1985 are not necessarily
generalizable. They may have been a historic coincidence.

Changing perspective from average levels of congruence, which disregards what policy directions elite and mass opinions differ, to a study of how members and voters are located vis-å-vis each other on the leftright gives us yet another opportunity to analyze the centerperiphery from a slightly different angle.

Contrary to the evidence from most American and European representational which emphasize the polarization of party elites compared party voters on left-right issues, the Swedish system in 1968/69, was characterized by a systematic displacement of each member group to the left of its voters.²⁸ The ideological distances between the positions of members and voters varied systematically with the left-right locations of the parties. Communist and Social Democratic members were, on the average, more to the left of their voters than members of the middle parties, who, in turn, were more to the left of their voters than the Conservative members. Thus, in the late 19605, Swedish politics was distinguished by a system with left-leaning elites in all parties. Jörgen Westerståhl, metaphorically, the system to a field of corn with all ears billowing to the left (Westerståhl & Johansson 1981, 87).

That metaphor is not applicable to Sweden anymore, however. A quite different system has emerged during the 1980s. A model with more ideologically party elites than party voters - much like the typical American result - has replaced the left-leaning model from the 19605.

In the 1985 study, socialist members are still to the left of their voters, while bourgeois members now are to the right of their voters. Moreover, the average ideological distances between party elites and voters have changed too. The Conservative Party no longer reveals the closest resemblance voters' and members' left-right locations. That distinction has been taken over by the Center Party, thereby changing the structure of within-party elite-mass distances into a very distinct center-periphery model, with more elite-mass similarity among centrist parties than among the parties toward either wing of the left-right dimension. Accordingly, Westerståhl has changed his metaphor, now comparing the system to an open fan, with the socialist members displaced to the left and the Conservative to the right (p. 87).

Rather dramatic shifts in the ideological positions of both members and voters are behind these changes in the relationships between elite and mass in Sweden. Acknowledging that the results for the studies in 1968/69 and in 1985 to a large extent were based on different sets of left-right issue items, and hence cautioning us against too many detailed comparisons,

nevertheless, the over-all scope and direction of the changes can hardly be
doubted.29

Comparing the late 1960s and the middle 1980s Swedish opinion has moved to the right, most noticeably among legislators of the Bourgeois parties, but also among voters, both in the Socialist and the Bourgeois bloc. Even Socialist legislators shifted their average positions a bit to the right, if we can believe the exact measurements. The most important effect of these changes is an increased ideological distance between the two blocs in Swedish politics, especially in parliament, substituting the model with leftleaning elites for a model with more clearly polarized party elites than party voters.³⁰

The pre-eminence of the new elite-polarized model is further underlined if we do not restrict the analysis to the relationship between legislators and ordinary voters. Segmenting the electorate according to degree of political involvement shows the pervasiveness of the elite polarized model. Clearly, what John May has called 'the special law of curvilinear disparity' of opinion structures within parties, is not a fitting description of our results (May 1973).

Compared to party activists and party members, representatives were

not ideological closer to voters, as May's law would have it. On the contrary, they were further away. Moreover, within all parties, except the Center Party, members' left-right views were more representative of the positions of the party activists than of ordinary party voters' positions. Thus, at least with the rather broad definition of party activists that we have used (comprising in all 6 percent of the electorate), Swedish members of parliamentare a moderating force balancing more extreme views of party activists. Instead, the model that suggests itself is one of policy leadership and opinion pull toward the ideological wings, not toward the center, on the part of the members.

The elite polarized model is also visible, although barely so, when we look at subjective ideological positions and how people place themselves on a left-right scale. As in France and in many other West European countries, Swedish members of parliament tend to place themselves further to the left than the voters. Referring to their French results, Converse and Pierce quote André Siegfried, and talk about a mystery and a specifically French cultural phenomenon (Converse & Pierce 1986, 129, 132, 778). Perhaps it is a mystery, but it most certainly is not a cultural phenomenon appearing only in France. The tendency of political elites to place themselvesfurther the left than the general public has been discovered in many

European countries, including - besides Sweden- Denmark, Holland, Italy and West Germany.³¹ However, looking at Converse and Pierce's results, it is probably true that the elite-mass differences in left-right self-placementsare in France than in most other countries. The French differences were certainly more pronounced than those we found in Sweden.

The Swedish results show a slightly bimodal member distribution with the highest peak just off center to the left. The comparable curve for the electorate is nicely bell-shaped and shifted to the right, with the modal point in the center.³² Within parties, the over-all left-oriented displacement of the members occurred in all parties, with the exception of the Conservative On the average, Conservative members placed themselves slightly to the right of where their voters put themselves; thereby giving the results in Fig. 5 some resemblance to the elite polarized model.

The tendency of subjective left-leaning among members of the Center and Liberal parties, as compared to their voters, is a most interesting result, since we know from our attitude measurements that 'objectively' members of the Center Party, and especially the Liberal Party, were on average, located to the right of their voters. Moreover, this subjective left-leaning tendency vis-å-vis the voters was not something that members were unaware of. On the contrary, members in all parties, including not only the middle parties, but also Socialist parties as well as the Conservative Party perceived their voters to be located to the right of their own positions.³² Thus, to the extent members experienced any ideological demands or opinion pressure from their own voters, they tended to perceive it as coming from the right. Obviously, this result is most noteworthy for the three Bourgeois parties, since non-Socialist members perceived an ideological right-turn among their own voters that did not exist. Instead, in reality, compared to the attitudes of their representatives in the Riksdag, there existed a left turn among Bourgeois voters in 1985. The results for the Socialist members are less notable. It is not an anomaly like the result for the Bourgeois parties. Ideologically, Socialist members were clearly to the left of their voters in 1985 and that is also how they saw the situation themselves.

Anticipatory Representation

Knowledge of voter sentiments, quite obviously, plays a crucial role for representatives trying to anticipate popular reactions and follow the wishes of their voters. Intent to follow the will of the people is of a limited use if what people want is misunderstood. Knowledge of voter attitudes is less crucial for legislators not bent on popular representation, but it is not wasted on them either. In a democracy, no matter what role preference elected representatives subscribe to, knowledge of voter sentiments is always an important and often useful commodity. At least members inclined to follow the trustee or party loyalist role prescriptions could use their knowledge of voters to help themselves get re-elected.

It is always difficult to judge degrees and that is particularly true when
the subject matter is knowledge of voter opinions - or perceptual accuracy

as it is usually called in the American literature - where to date very few empirical studies have been done.³⁴ Thus, any evaluation of the results from our Swedish study must be tentative, especially, since an unequivocal assessment probably is unattainable. For, as it turned out, Swedish members proved themselves to be fairly good at knowing what their own party's voters wanted, but to be considerably less knowledgeable when it came to perceiving the opinions of the entire electorate.

Nine different issue questions were studied and the average percentage of members who correctly could pinpoint the majority positions of their own party's voters proved to be 78 percent.³⁵ The corresponding percentage for how accurately members perceived the sentiments among all eligible voters was only 60 percent. The outcome is not unexpected. Given what we know from psychology, it is not surprising that the more ambiguous and heterogeneous opinions of the entire electorate proved to be more difficult to perceive accurately than the more familiar and homogeneous attitudes of the members' own party's voters.

That the degree of ambiguousness of the stimulus objects really played a role for how accurately the Swedish members perceived the opinions of the voters is very evident. Restricting the analysis to the perceptions of their own party's voters, a clear pattern emerges proving that members did much better on old, highly politicized left-right issues than on newer, less politicized and non-left-right issues. Perceptual accuracy was correlated with degree of saliency of the issues. Furthermore, perceptual accuracy was also correlated with the degree of homogeneity of voter opinions. Contrary to results from some American studies, Swedish members were best at perceiving homogeneous voter sentiments.³⁶

Turning away from the stimulus objects to the observers, our search for explanations of perceptual accuracy was pretty much a disappointment. As was the case in France, individual factors having to do with the legislators' experiences, power positions or connectedness with various information networks were very weakly related to the degree of voter knowledge.³⁷ Independent variables such as age, leadership position within party, media exposure, education, contacts with individual voters or local party organizations for no more than a few percent of the variance in the summary accuracy index we constructed from the individual perception questions. Disturbing as it is, one can but conclude that a whole array of rational knowledge formation-processes was not working very effectively when it came to Swedish members' perceptions of their own voters' opinions.

One of the main reasons why the normal information processes had difficulties producing correct knowledge, was that their effects were counteracted by another observer-related factor that proved to be much more powerful. That factor was the strong tendency on the part of the

members to project their own opinions onto their voters.³⁸ For members in agreement with the opinions of their voters, wishful thinking of this kind, led to highly inflated levels of perceptual accuracy, while for members at odds with the sentiments of their voters, projecting their own opinions resulted in faulty perceptions.

The Swedish results - both in 1968/69 and in 1985 - bear out the existence of these relationships very clearly. Among members in attitude agreement with their own voters, on average 90 percent perceived the voters' majority positions correctly, while among members with different opinions than their voters, only some 25 percent were accurate in their perceptions. A substantial 25 percent of the variance in our summary accuracy index was accounted for by a summary opinion agreement measure.

The strength of the projection tendency becomes even more evident if one performs a kind of effect analysis using members' own opinions and voters' opinions as independent variables, and members' perceptions of voters' opinions as the dependent variable. Such an analysis addresses the question whether reality (= voters' opinions) or wishful thinking has a stronger impact on members' knowledge of voter sentiments. The outcome of such an analysis based on the 1985 Swedish Representation Study is presented in Table 6. The result is a bit disconcerting.

As it turned out, both factors played an independent role, but reality proved to be less decisive than the process of wishful thinking.³⁹ The results fit a pure wishful thinking model - where members' own attitudes totally determine their perceptions - much better than a pure information-driven model in which the reality (== voters' opinions) structures the perceptions of the members.

One way to explain and at the same time make excuses for these results is to refer to the fact that members are only humans. Psychological research has shown that all human beings have a tendency to distort their perceptions in certain ways.⁴⁰ It is not easy to grant such an excuse, however, when one reflects upon the negative effects wishful thinking has in a representative democracy.

The consequences of perceptual distortions are to a certain extent paradoxical a democracy, since they cause information channels between elite and mass to function badly when they should work at their best (among members at odds with their voters), and to function most effectively when they are least needed (among members in agreement with their voters). Furthermore, among voters, wishful thinking tends to make perceptions of policy differences with preferred parties smaller than they actually are. Hence, impulses to switch party diminish. Among representatives, wishful thinking tends to blur members' perceptions of policy differences vis-å-vis the entire electorate and their own voters. The result is fewer reasons to engage in opinion formation trying to change the policy attitudes of the voters, and fewer reasons to vote differently on parliamentary roll calls. Thus, the projection tendencies of both voters and members tend to bring about an exaggerated notion of harmony within parties and an exaggerated notion of conflict vis-å-vis political opponents.

Wishful thinking is an irrational force. There can be little doubt that in the
ideal model of a representative democracy, wishful thinking is something
undesirable that should be kept at the lowest possible level.

Political Representation Swedish Style

Converse and Pierce's final verdict on the quality of representation in France is rather harsh. The last sentence in their monumental book reads: '. . . however delightful a representation instrument the French system may be for those equipped to play upon it, it seems to fall short of its potential for much of the electorate much of the time' (p. 786). Heinz Eulau considers this verdict to be too lenient. In his review article in Legislative Studies Quarterly he interprets Converse and Pierce's results as indicating a '(French) model in which ideologically stagnant legislative party elites are the dominant formulators and proponents of public policies and are quite immune to popular preferences, insofar as such preferences exist all in a discerningly articulated form' (Eulau 1987, 211).

Converse and Pierce see a French system lacking in popular representation.Eulau one step further, branding the French democracy as elitist and run by unresponsive parties. Looking at the results from the Swedish studies, an Eulauean conclusion seems a bit drastic, while a verdict

of the kind Converse and Pierce delivered for France may be too mild. Without doubt, the Swedish system is far from any model of popular representation run from below. A more elitist model, with a clear emphasis on policy leadership on the part of the political parties, fits most of our results much better. To be fair though, we have not, as yet, analyzed the basic representation bond (decisional representation) in the Swedish context. The probability, however, that such a study would lead to any other conclusions that the ones we can draw already, based on the analyses of the four linkage processes, is very small.

Perhaps our results for the first linkage process - social representation - do not put Sweden toward the bottom of the heap of Western democracies. Certainly there are nations, like the Unite States and most of the countries in the European Community, which exhibit even more exclusive recruitment patterns of elected officials than Sweden. For example, the Riksdag has more female members and members originating from the working class than most parliaments in the Western world. But, nevertheless, the social recruitment to the Riksdag is rather narrow, with a clear over-representation members from more privileged groups in society. Moreover, looking back fifty or sixty years, to the beginning of democracy in Sweden, recruitment seems not to have become less exclusive. The policy effects of the biased recruitment system are clearly mitigated, however, by the fact that socio-economic characteristics of members in most cases are unrelated or no more than faintly related to their policy attitudes. A cautious conclusion be that the social representation of the Swedish Riksdag could be improved upon, but that it is not a catastrophe as it is.

Role representation, our second linkage process, must be judged a failure though, if one expects Swedish representatives to subscribe to voter oriented role convictions. As a matter of fact, very few members adhered to the delegate role, the most populist of the representational roles. Instead, as is to be expected in a party democracy like the Swedish one, a majority of members preferred the party loyalist role, with a sizable minority supporting the trustee role. Interesting as this may be on a theoretical or ideological level, the crucial point, however, is that members' representationalrole turned out not to mean much when it came to legislative behavior. Most correlations between role convictions and relevantforms behavior or attitudes were dismally small in our Riksdag studies. Thus, on the practical level, what was in the minds of the representativesconcerning role convictions meant very little. Swedish members, when asked, pose as trustees, delegates or party loyalists, but it is of almost no practical consequence. Therefore the conclusion is quite unequivocal. The name of the game in Sweden is party representation which in practice means that the party loyalist role is of importance. Besides

this obvious faet, normative role representation is not a viable linkage
process in the Swedish context.

Policy representation, our third linkage process, is conceptually closer to decisional representation than either social representation or role representation. it is potentially more important. Members in agreement with their voters only have to vote their views in order to effectuate a high degree of popular representation. Admittedly, the simple words 'only have to vote their views' pass over a whole set of complex processes within legislatures; but still, policy representation is more proximate to decisional representation than many other more indirect forms of representation.

Granting that the level of policy congruence between members of the Riksdag and the electorate has increased somewhat in the 1980s compared to the late 19605, it is nevertheless true that the Swedish congruence level, in absolute terms, is not very impressive. A Riksdag chosen by a totally random process, without resorting to the elaborate procedures of candidate nominations, election campaigns and popular voting, would almost always exhibit higher degrees of policy congruence than the ones we found in our empirical studies in 1968/1969 and in 1985. The probability of getting a 15 percentage points average opinion difference between elite and mass, as we found in 1985, if members were chosen randomly, is very small, less than one in a thousand.

Analyzing the Swedish results across issues and parties, there is little doubt what processes have been at work, producing policy differences between members and voters. A strong tendency to policy leadership on the part of the elites is the most important process, sometimes combined with outright party unresponsiveness and with poor showings when it comes to opinion molding. Most of the issue questions exhibiting poor congruence levels were either long-standing policy leadership issues in which party elites - most noticeably the Social Democratic elite - have been unable to sway voter opinion, or new less politicized non-left-right issues on which it is reasonable to suspect that the representatives simply were ahead of a less-informed electorate.

The policy leadership interpretation is further strengthened when one takes a closer look at how party elites and voters are located on the allimportantideological dimension. For, as it turned out, it was the parties most bent toward policy leadership at either wing of the dimension who displayed the lowest levels of congruence. On the left, Socialist members were more radical than their voters, while on the right, the Liberal and Conservative Parties were more conservative than their supporters. The Center Party, located close to the middle of the left-right dimension, was the party that exhibited the highest degree of issue agreement between members and voters. Thus, a rather clear policy leadership model with party

elites being more ideologically polarized than party voters characterizes the
Swedish representational system in the 1980s.

Interestingly, the model looked different in the 19605. In the Representation of 1968/69 each party elite group was displaced to the left of its voters, with the Socialist members clearly more so than Bourgeois members. One way to reconcile this finding with the policy leadership model is to acknowledge that two of the Bourgeois parties, the Liberal and the Conservative Parties, have changed both their political views (they have moved to the right) and their style of conducting politics since the 19605, meaning that they now have become less populist and more interested pursuing their own policies, even if it sometimes means pulling ahead of their voters. On the Socialist side, there is no need to impose any such notion of changing political leadership styles. Policy leadership from above has always been a generic way of pursuing politics among Social Democratic and Communist Party elites, confirmed by the fact that Socialist legislators were clearly displaced to the left of their voters both in 1968/69 and in 1985.

One critical way of summing up our findings on policy representation in Sweden is to acknowledge that the level of attitude congruence between elite and mass is not especially high, and that if the explanation in terms of policy leadership is correct, it is not an altogether satisfying one on a normative level. Policy leadership does not only imply being ahead of voters; it also implies opinion formation and bringing the voters along. Despite increased levels of issue voting among voters and more resources devoted to election campaigns, it is obvious from our results that Swedish parties could do better to assure higher levels of within-party opinion agreement. At the risk of committing sacrilege, one is tempted to remind the parties that if everything fails and voters refuse to be swayed, there is always the possibility of changing party policies. Low levels of policy congruence in a system with more and more issue voting citizens, not only implies dangers for the individual parties, it also implies dangers for the system as a whole, to which the increased levels of distrust in parties and politicians among Swedish voters bear witness.⁴¹

The fourth and final linkage process, anticipatory representation, presupposesa degree of knowledge of public opinion and voter sentiments on the part of the representatives. The results from our study of perceptual accuracy among Swedish members of parliament are not promising on this score. It is doubtful whether the members of the Riksdag measure up to an acceptable standard. In the Riksdag Study of 1985, the average percentageof who could accurately indicate the majority position in the entire electorate on nine investigated issue questions was 60 percent. At first glance, that might seem to be a decent result. But when one realizes that the average would have been 50 percent if every member had known

nothing and just guessed what the majority opinions were, it is quite
obvious that our result of 60 percent correct answers is not impressive.

The members did better though, when asked about their own party's voters. Understandably feeling more at home with that task, the average percentage of members with accurate perceptions rose to 78 percent. However, when one looks a little closer at that result, it is uncertain how genuine the members' knowledge really was. For, as our analysis showed, the members' perceptions of their voters" opinions were to a very limited degree influenced by various rational processes of knowledge formation. Instead, a more irrational perceptual process proved to have been more effective. That process was the very pronounced tendency on the part of the members to engage in wishful thinking. The representatives tended, to a large extent, to project their own views onto their voters. Since most members were in policy agreement with their voters, the tendency to project inflated rather than deflated the accuracy results.

Mildly put, the results of our study of perceptual accuracy among Swedish members of parliament are less than reassuring. It is hardly an exaggeration to allege that anticipatory representation cannot be in particularly good shape in Sweden, given the state of members' knowledge about voters.

Being at the end of the road second thoughts make themselves felt. Maybe our appraisal of political representation Swedish style has been too critical. Perhaps all the elitist characteristics that we have observed in the Swedish system are inherent ingredients in any functioning representive democracy. Regrettably, the shortage of comparative results from most other Western democracies makes such a question difficult to answer. The only thing we can hope for is more research on political representation in the future. Miller and Stokes set the ball rolling thirty years ago when they did their American Representation Study in 1958. Comparative research on political representation was an integral part of their vision. They threw a gauntlet down. It is time for political scientists across the world to pick it up and start getting the job done.

---

NOTES

1. The responsible-party model is not overlooked by either Converse & Pierce or Eulau, see their discussions in Converse & Pierce (1986, 499-500, 698-709) and Eulau (1967 and 1987, 196-198).

2. The average party cohesion on roll calls in the Swedish Riksdag has been above the 95 percent mark at least since 1925, when the present system with non-secret votes was inaugurated, see Bjurulf (1972) and Holmberg (1976).

3. Members of the Swedish Riksdag are elected in multi-member districts on party ballots using a system of strict porportional representation. The essence of the system is that individual members are nominated by their parties and that voters choose between parties, not between candidates.

4. The Member Studies, both in 1969 and in 1985, were done in collaboration with the Swedish Election Studies Program. The 1969 Riksdag Study was directed by Bo Särlvik, assisted by Sören Holmberg and Per-Anders Roth. Professor Warren Miller and Professor Aage Clausen from United States participated in the planning of the study. Sören Holmberg and Peter Esaiasson directed the Member Study in 1985, with Bo Särlvik taking part in the planning stages. The Member Study in 1969 involved personal interviews with the members of the second chamber of the Riksdag (response rate 97 percent). In 1985 we used a mail survey, sending the questionnaire to all members of the now unicameral Riksdag (response rate 97 percent). The surveys of the electorate comprised some 3000 personal interviews, both at the national election of 1968 and at the national election of 1985. For more details about the Member Studies see Särlvik (1969), Roth (1970), Holmberg (1974) and Holmberg & Esaiasson (1988).

---

4. The Member Studies, both in 1969 and in 1985, were done in collaboration with the Swedish Election Studies Program. The 1969 Riksdag Study was directed by Bo Särlvik, assisted by Sören Holmberg and Per-Anders Roth. Professor Warren Miller and Professor Aage Clausen from United States participated in the planning of the study. Sören Holmberg and Peter Esaiasson directed the Member Study in 1985, with Bo Särlvik taking part in the planning stages. The Member Study in 1969 involved personal interviews with the members of the second chamber of the Riksdag (response rate 97 percent). In 1985 we used a mail survey, sending the questionnaire to all members of the now unicameral Riksdag (response rate 97 percent). The surveys of the electorate comprised some 3000 personal interviews, both at the national election of 1968 and at the national election of 1985. For more details about the Member Studies see Särlvik (1969), Roth (1970), Holmberg (1974) and Holmberg & Esaiasson (1988).

5. For a discussion of various forms of representation and an overview of the behaviorally oriented literature on representation and legislatures, see Esaiasson and Gilljam (1986).

6. Robert Weissberg (1978) has coined the measurement technique collective representation, be distinguished from dyadic representation, which involves the issue congruence between individual representatives and their constituents. Converse and Pierce (1986, 507-510, 595-604) use the terms absolute representation or absolute congruence in contrast to relative representation. The latter form of individual representation is usually measured by correlation coefficients. For a critique of the use of correlations in the study of congruence see Achen (1977).

7. For an excellent overview of the literature on perceptual accuracy see Clausen (1977).

8. Cited from Loewenberg, Patterson & Jewell, eds. Handbook of Legislative Research (1985, 18).

9. A power position index was constructed based on a summary of each member's number of formal power positions in the Riksdag and in his party group, for details see Holmberg & Esaiasson (1988, ch. 6).

10. The coefficients are squared non-linear etas. A majority of the squared correlations was within the range of 0.00 to 0.03.

11. The results are not suprising in the Swedish context. They look the same among the public. In Sweden, wornen are slightly more green on environmental issues, slightly more pro spending on welfare issues, and slightly more moralistic, see Holmberg & Gilljam (1987, ch. 11).

12. Among members, party standpoint won by 56 percent against 25 percent when pitted against constituents' opinion, with 19 percent dks and other answers. Against conscience, party standpoint was supported by 49 percent, with 31 percent in favor of own conscience and 20 percent giving other answers. Own conscience, in turn, beat voters' opinion (58 vs. 28 percent) with 14 percent giving other answers.

13. Among voters the pairwise results were: party vs. constituents, 22 percent against 61 percent, with 17 percent dks; party vs. conscience, 27 percent against 62 percent, with 11 percent dks; and, finally, constituents vs. conscience, 57 percent against 31 percent, with 12 percent dks. There are rather clear patterns in the way people answered the three questions. Among persons who gave substantial answers to all three questions, the percentage of intransitive rank-orderings was only 7 percent. Thus, a large majority of the electorate held logically ordered sets of role preferences. The comparable proportion of members with intransitive rank-orderings was also low, only 7 percent, see Granberg & Holmberg 1988, ch. 4.

14. Converse & Pierce's (1986) French results and Barnes (1977) results from Italy show a strong support for the party loyalist role among legislators. However, there are European studies where the trustee role is the dominant one among representatives; for the Dutch and Danish cases see Daalder & Rusk (1972) and Damgaard (1982), respectively; see also Cayrol, Parodi and Ysmal (1976) for French results slightly different from Converse & Pierce's. Among American legislators, the trustee role clearly is the most popular (inter aha Converse & Pierce 1986, ch. 21). On the mass level, the popularity of the delegate role is not surprising. It is less evident, however, why the trustee role usually draws more support among voters than the party loyalist role. But that has been the case not only in our Swedish study but also in American (Cain, Ferejohn & Fiorina 1987), British and French studies (Converse & Pierce 1986). Barnes (1977) Italian study provides a counter-example, though. Among Italian voters the party loyalist role was preferred to the trustee role.

---

14. Converse & Pierce's (1986) French results and Barnes (1977) results from Italy show a strong support for the party loyalist role among legislators. However, there are European studies where the trustee role is the dominant one among representatives; for the Dutch and Danish cases see Daalder & Rusk (1972) and Damgaard (1982), respectively; see also Cayrol, Parodi and Ysmal (1976) for French results slightly different from Converse & Pierce's. Among American legislators, the trustee role clearly is the most popular (inter aha Converse & Pierce 1986, ch. 21). On the mass level, the popularity of the delegate role is not surprising. It is less evident, however, why the trustee role usually draws more support among voters than the party loyalist role. But that has been the case not only in our Swedish study but also in American (Cain, Ferejohn & Fiorina 1987), British and French studies (Converse & Pierce 1986). Barnes (1977) Italian study provides a counter-example, though. Among Italian voters the party loyalist role was preferred to the trustee role.

15. The outcome of our study of role preferences among Riksdag members in 1969 could serve as an indication of how sensitive results can be to differences in research design. In 1969, the trustee role, clearly drew more support than the party loyalist role (57 percent vs. 35 percent, with the remaining 8 percent favoring the delegate role). We think that the rather striking differences in role preferences when compared to the results of the 1985 study (34 percent trustee, 51 percent party loyalist, and 15 percent delegate) to a large degree were brought about by changes in the technique of data collection. Although, the question wording was basically the same, the data in the 1969 Riksdag Study were collected through personal interviews, while the 1985 study used mail questionnaires to collect answers to survey items. Our suspicion is that personal interviews have a tendency to elicit more self-assertive answers (vote conscience of party line) than mail questionnaires. But, of course, a part of the difference between the results of 1969 and 1985 could be attributable to system changes in the Swedish parliamentary system. It is not inconceivable that trustee role convictions were somewhat more common in the old smaller second chamber of the Riksdag (233 members), than in the new larger unicameral Riksdag (349 members).

16. Results showing Conservative members being more party oriented than members of centrist Liberal parties is not unprecedented in European representational role studies. The same pattern has been visible in France (Converse & Pierce 1986), Italy (Barnes 1977) and Denmark (Damgaard 1982).

17. Converse & Pierce (1986) analyze the relationship between role preferences and fidelity of party voting and knowledge of district sentiments (683 ff.).

18. Converse & Pierce (1986, 693) show correlations in the range of +0.07 and +0.14. The associations are as weak, if not weaker in Sweden.

19. Miller & Stokes (1963), Backstrom, (1977, 419) and Converse & Pierce (1986, 223-225). of the identically phrased position questions in the Converse-Pierce French study dealt with one and the same issue area, European integration.

20. Backstrom may have computed his percentages differently than we did in the Swedish study. We excluded don't knows from the percentage base; Backstrom may not have done that (p. 420).

21. Converse-Pierce (1986) standardized the attitude differences between representatives and the electorate by dividing the differences by the standard deviations of the mass distributions (pp. 963-964). The average French value of 0.34 is an estimate computed from the graphic bars for Deputies in Fig. 19.1 (p. 598) of Converse & Pierce's book. No exact results are given in the figure.

22. Other changes that point in the same direction are the increased coverage of election campaigns in the media (Asp 1987) and the increased level of issue voting in the electorate (Holmberg & Gilljam 1987, ch. 12).

23. The average means difference went down from 0.7 in 1968/69 to 0.4 in 1985, while the proportion of issues displaying different majority positions among members and voters decreased from 35 percent in 1968/69 to 30 percent 1985, for further details see Holmberg & Esaiasson (1988, ch. 4).

24. On all these three issues Social Democratic members were clearly to the left of their voters, refusing to commercialize TV, preferring higher levels of social welfare benefits and supporting the rather unpopular wage earners' funds (Holmberg & Esaiasson 1988, ch. 4).

25. The new green policy dimension is still poorly constrained among Swedish voters, especially among people with low education (Holmberg & Asp 1984, 456-461).

26. On the individual level, both in the general public and among members, the welfare dimension and the economically oriented public vs. private sector dimension are correlated with each other. Nevertheless, it is possible to view them as two different, although not orthogonal dimensions. On policy dimensions in Swedish politics see Holmberg (1974,1975), Sårlvik (1976), Linden (1976), Petersson (1977) and Holmberg (1981).

---

27. The center-periphery hypothesis is upheld in Swedish local government studies on policy congruence from both the 1960s and 1970s (Stromberg 1974,116 and Westerståhl & Johansson 1981, 85-87). Some of the results from Thomassen's (1976, 152) Dutch Representation Study also lends support. However, Barnes', (1977, 119) Italian findings do not fit in, as do not some of the results from various European Community countries reported in Dalton (1985). Russel Dalton advances an alternative hypothesis to ours. According to Daltort, we should expect centrist parties to exhibit lower, not higher, levels of congruence than parties toward the wings. The reason being that '. . . a leftist or rightist party image . . . facilitates a clearer representational linkage between voters and parties.' (p. 292).

28. In the 19605, there were other Swedish studies that showed roughly the same result, see Stromberg (1974). For international results see McClosky, Hoffman and O'Hara (1960), Eldersveld (1964), Backstrom (1977), Barnes (1977), Thomassen (1976) and Dalton (1985).

29. Two of the left-right issue items were the same across the studies. Separate analyses of these two questions, one dealing with the level of social benefits and the other with the degree of government influence over private business, support the conclusions drawn in the text.

30. Methodological problems, especially acute on the mass level, like regressions toward the mean and non-attitudes showing up as flip-flops on the issues are phenomena that, in theory, can produce a polarized model of the kind we found in 1985. This danger has been minimized, however, by excluding dks (they have not been given middle positions on the issues). An even more stringent safeguard against the non-attitudes problem is to exclude all persons with middle opinions on the issue (3s) as we have done in some of the previous analyses of the level of congruence. Another safeguard would be to avoid working with summed attitude indices. A simple analysis of the number of issue questions that were characterized by opinions of members being to the right of the opinions of voters convincingly prove that the polarized model in 1985 is not a methodological artifact. The results in the table below show the proportion (%) of left-right issue questions exhibiting average member opinions to the right of average voter opinions. Party 1968/69 1985 vpk 7 0 s 7 0 c 20 58 fp 33 75 m 33 100 Entire Riksdag 20 17 Number of left-right issue questions 15 12

31. See Barnes (1977), Kristensen (1980), Schmitt (1984), van de Geer & de Man (1974), Dalton (1985) among others.

32. People who place themselves at 5 ('neither left or right') are most likely a mixed lot, with persons who should be dks thrown in among the true centrists. Analyzing the level of political knowledge within the various party groups, persons placing themselves at 5 tend to be somewhat less knowledgeable than persons putting themselves offcenter. differences are not big, though. Thus, I would not agree with Converse & Pierce's (1986) drastic French solution: '. . . it may be both legitimate and fitting to erase mentally most of the central mode of the distribution as non-combatant, leaving a curve which is nearly biomodel' (p. 129).

33. The results for how the members placed themselves in Fig. 5 can be compared to these results indicating how the members on average placed their own party's voters on the left-right scale: vpk 2.11; s 3.76; c 6.00; fp 6.47; and m 7.88.

---

34. Clausen (1977) reviews the mainly American literature on perceptual accuracy.

35. Members were asked to perceive the majority opinions of their own party's voters in the whole of Sweden. A district-level analysis is of limited value in this context. Within-party opinions of voters as well as of members on nationally politicized issues do not vary significantly across districts in Sweden.

36. Sigel & Friesema (1965); for European results that support the conclusion that homogeneity and saliency of the stimulus object are linked to perceptual accuracy see Brand (1972), Converse & Pierce (1986, ch. 20) and Clausen, Holmberg & deHaven- Smith (1983). For more details on the 1985 Swedish Study see Holmberg & Esaiasson (1988, ch. 5).

37. Converse & Pierce (1986, 656-657) state: '. . .we have looked at a variety of plausible factors related to the politician's situation . . . Working across . . . more than a dozen variables, we find nothing that accounts for as much as 4 percent of the variance in either of our summary knowledge factors.'

38. Of course, projection tendencies are nothing unique to Swedish members. They have been found in many elite studies, see Clausen (1977) and Converse & Pierce (1986, ch. 20).

39. As Converse & Pierce (1986, 653) conclude for their French data, a persuasion hypothesis (members picking up as their own the opinions of the voters) is not a credible counter-hypothesis to the projection hypothesis which appears to be supported by our Swedish results.

40. Early voting studies also implied the existence of projection tendencies in the electorate, Berelson, Lazarsfeld & McPhee (1954). For a recent study on projection tendencies among both American and Swedish voters, see Granberg & Holmberg (1988).

41. Trust in parties and politicians has been on the decrease among Swedish voters since systematic measurement on the subject started in 1968, for more details see Holmberg & Gilljam (1987).

REFERENCES

Achen, C. H. 1977. 'Measuring Representation: Perils of the Correlation Coefficient'.
American Political Science Review 21, 805-815.

Asp, K. 1987. 'Rikspolitik och kommunalpolitik - valrörelsernas utrymme i Svensk dagspress
1956-1985'. Folkstyrelsens villkor SOU 1987:6.

Asp, K. 1986. Mäktiga massmedier. Studier i politisk opinionsbildning. Stockholm: Akademilitteratur.

Backström, C. H. 1977. 'Congress and the Public. How Representative is One of the Other',
American Political Science Quarterly 5, 411-435.

Barnes, S. H. 1977. Representation in Italy: Institutional Tradition and Electoral Choice.
Chicago: University of Chicago Press.

Berelson, B. R., Lazarfeld, P. F. and McPhee, W. N. 1954. Voting. Chicago: University of
Chicago Press.

Bjurulf, B. 1972. 'Från minoritetsparlamentarism till majoritetskoalition. En studie av riksdagens
1925-1938', Statsvetenskaplig tidskrift 75, 125-188.

Brand, S. A. 1972. 'Councillors, Activists, and Electors: Democratic Relationships in Scottish
Societies' in Patterson, S. C. and Wahlke, J. C. eds., Comparative Legislative Behavior:
Frontiers of Research. New York: John Wiley and Sons.

Cain, B. F., Ferejohn, J. and Fiorina, M. P. 1987. The Personal Vote. Constituency Service
and Electorate Independence. Cambridge, Mass.: Harvard University Press.

Cayrol, R. Parodi, J.-L. and Ysmal, C. 1976. 'French Deputies and the Political System',
Legislative Studies Quarterly 1, 67-99.

Clausen, A. 1977. 'The Accuracy of Leader Perceptions of Constituency Views', Legislative
Studies Quarterly 2, 361-384.

Clausen, A. and Holmberg, S. 1974. 'Legislative Voting Analysis in Disciplined Multi-Party

Clausen, A., Holmberg, S. and deHaven-Smith, L. 1983. 'Contextual Factors in the Accuracy
of Leader Perceptions of Constituents Views', The Journal of Politics 45, 449-472.

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

Daalder, H. and Rusk, J. G. 1976. 'Perceptions of Party in the Dutch Parliament', in
Patterson, S. C. and Walhke, J. C. eds., Comparative Legislative Behavior: Frontiers of
Research. New York: John Wiley and Sons.

Dalton, R. J. 1985. 'Political Parties and Political Representation. Party Supports and Party
Elites in Nine Nations', Comparative Political Studies 18, 267-299.

Damgaard, E. 1982. Partigrupper, representation och styring. Köpenhamn: Schultz.

Esaiasson, P. and Gilljam, M. 1985. 'Forskning kring parlament och parlamentariker',
Statsvetenskaplig tidskrift 88, 331-41.

Eulau, H. 1967. 'Changing Views of Representation', in de Sola Pool, 1., ed., Contemporary
Political Science: Toward Empirical Theory. New York: MacGraw-Hill.

Eulau, H. 1987. 'The Congruence Model Revisited', Legislative Studies Quarterly 12, 171—
214.

van de Geer, J. P. and de Man, H. 1974. Analysis of Responses to Issue Statements by Members
of the Dutch Parliament. Leyden: Department of Data Theory. University of Leyden.

Granberg, D. and Holmberg, S. 1988. The Political System Matters: Social Psychology
and Voting Behavior in Sweden and the United States. Cambridge, England: Cambridge
University Press.

Holmberg, S. 1974. 'Riksdagen representerar svenska folket' Empiriska studier i representativ
demokrati. Lund: Studentlitteratur.

Holmberg, S. 1975. 'Partipolitiska konfliktdimensioner i riksdagen', Statsvetenskaplig tidskrift
78, 213-231.

Holmberg, S. 1976. Legislative Behavior in Sweden: Party Cleavages and Party Cohesion.
Göteborg: Department of Political Science. University of Göteborg.

Holmberg, S. 1981. Svenska väljare. Stockholm: Liber

Holmberg, S. and Asp, K. 1984. Kampen om kärnkraften. En bok om väljare, massmedier
och folkomröstningen 1980. Stockholm: Liber.

Holmberg, S. and Gilljam, M. 1987. Väljare och val i Sverige. Stockholm: Bonniers.

Holmberg, S. Gilljam, M. and Oskarson, M. 1988. Valundersökning 1985. Teknisk rapport.
Stockholm: Statistiska centralbyrån.

Holmberg, S. and Esaiasson, P. 1988. De folkvalda. En bok om riksdagsledamöterna och den
representativa demokratin i Sverige. Stockholm: Bonniers.

Jewell, M. E., 1983. 'Legislator- Constituency Relations and the Representative Process',
Legislative Studies Quarterly 13, 303-337.

Kristensen, O. 1980. 'Den politisk-bureaukratiske beslutningsproces som medvirkende årsak
till vaeksten i den offentlige sektorn', Nordisk Administrativ Tidskrift 61, 80-102.

Linden, M. 1976. 'Political Dimensions and Models, of Political Party Sympathy: Empirical
Studies of Swedish Attitude Data', European Journal of Political Research 4.

Loewenberg, G., Patterson, S. C. and Jewell, M. E. eds., 1985. Handbook of Legislative
Research. Cambridge, Mass.: Harvard University Press.

May, J. O. 1973. 'Opinion Structure of Political Parties. The Special Law of Curvilinear
Disparity'. Political Studies 21, 131-151.

McClosky, H., Hoffman, P. J. and O'Hara, R. 1960. 'Issue Conflict and Consensus among
Party Leaders and Followers', American Political Science Review 54, 406-427.

Miller, W. E. 1964. 'Majority Rule and the Representative System of Government', in
Allardt, E. and Littunen, Y., eds., Cleavages, Ideologies, and Party Systems: Contributions
to Comparative Political Sociology. Helsinki: Westermarck Society.

Miller, W. E., and Stokes, D. E. 1963. 'Constituency Influence in Congress', American
Political Science Review 57, 45-46.

Petersson, O. 1977. Väljarna och valet 1976. Statistiska centralbyrån.

Pitkin, H. 1967. The Concept of Representation. Berkeley: University of California Press.

Roth, P. 1970. Members of the Riksdag Study 1969: Documentation Bulletin and Interview
Questionnaires. Goteborg: Department of Political Science. University of Goteborg.

Schmitt, H. 1984. 'Zur Links-Rechts Polarisierung in Mittlerer Fuhrungsschicht und Wählerschaft
10 Westeuropäischen Parteinsystem', in Falter, J., et al., eds., Politisene
Willensbildung und Intressenvermittlung. Opladen: Westdeutscher Verlag.

Sigel, R. S., and Friesma, P. H. 1965. 'Urban Community Leaders Knowledge of Public
Opinion', World Politics Quarterly 18, 881-895.

Strömberg, L. 1974. Väljare och valda. En studie av den representativa demokratin i kommunerna.
Statsvetenskapliga institutionen.

Särlvik, B. 1969. Representationsundersökningens forskningsprogram. Relationer mellan väljare
valda på rikspolitisk nivå. Göteborg: Statsvetenskapliga institutionen.

Särlvik, B. 1976. Mapping the Party Space: Distances, Evaluations and Ideological Perspectives.
Department of Political Science. University of Göteborg.

Thomassen, J. 1976. Kiezens en gekozen in een representatieve demokratie. Alphen aan den
Rijn: Samson.

Wahlke, J. C, Eulau, H. Buchanan, W. and Ferguson, L. C. 1962. The Legislative System:
Explorations in Legislative Behavior. New York: Wiley and Sons.

Wahlke, J. C. 'Legislative Behavior, 1967-1968', Legislative Studies Quarterly 12, 215-225.

Weissberg, R. F. 1978. 'Collective vs. Dyadic Representation in Congress', American Political
Science Review 72, 535-547.

Westerståhl, J. and Johansson, F. 1981. Medborgarna och kommunen. Studierav medborgerlig
aktivitet och representativ folkstyrelse. Stockholm: Ds Kn 1981:12.