Resumé

In this study we replicate American studies of discrepancy at the aggregate level between popular agreement about abstract democratic principles and the lack of their application to particular political contexts. We explain this discrepancy by adopting a research design which allows us to assess the functional relationships between respondents' evaluation of democratic norms and the major expressions contained in these norms, including the engaged democratic values.

Multivariate statistical analyses showed a systematic tendency for the democratic value component to have the greatest weight and the group evaluation component the least weight in Dane;;' formation of democratic norms. This stress pattern was replicated across a series of subgroups of Danes varying in level of political interest, amount of formal education, and extent of subjective social competence.

Introduction

In his book Who Governs? Robert A. Dahl writes, '... a common tendency of mankind — and not least of Americans — is to qualify universals in application while leaving them intact in rhetoric' (Dahl 1961, 319). Earlier American studies of democratic values (conceived here as perceptions of the desirable democracy) and of their application to more concrete situations support Dahfs generalization. Survey investigations done by Samuel Stouffer (1963), James Prothro and Charles Grigg (1960), and Herbert McClosky (1964) testify that a significant discrepancy exists between the American people's democratic values and norms (understood here to be the application of these values to

---

* The research reported in this paper was financed by grants from the Danish Social Science Research Council. The questionnaire was written by the author and by Palle Svensson, Institute for Political Science, Aarhus University. We received comments and critical remarks especially from Kirsten Just Jensen, Esther Körmendi, and Finn Madsen, the National Institute for Social Research. The National Institute for Social Research collected and processed the interview material. Steen Carl Nielsen, Institute for Social Sciences, Odense University, performed many of the statistical analyses. The data are available upon request from the Danish Data Archive, Odense University, Denmark.

particular political contexts). For example, in their investigation of democratic values in general and their application to political situations, Prothro and Grigg found that while 95 percent of the respondents agreed with the statement that ' [e]very citizen ought to have equal opportunity to influence government policy', 49 percent disagreed with the statement that '[i]n a referendum in a town only people who are well-informed about the problem decided upon, ought to have the right to vote', (Prothro & Grigg 1960, 276-94). This and other findings from American surveys of public opinion lead us to expect markedly higher support in the general population for democratic values generally than for their application to concrete political contexts.'

In the public opinion studies by Prothro and Grigg and by McClosky, the observed consensus on abstract democratic principles and the lack of consensus on the application of these principles to particular political contexts is interpreted as the lack of relationship between democratic values and norms. However, this is not demonstrated by measurement of the relationship between the degree to which individuals support the abstract democratic principles and the extent to which they agree to their application to particular political situations. McClosky writes, '... democratic ideas and rules of the game are ordinarily encountered not in pure form or in isolation but in substantive contexts that are bound to influence the ways in which we react to them'. (McClosky 1964, 376). Yet McClosky, Prothro and Grigg, and Stouffer make no attempt to measure the independent contributions of different aspects of the given substantive context, including the engaged democratic value, in relation to individual agreement or disagreement with the democratic norms.

In his book Tractatus Logico-Philosophicus, Wittgenstein remarks that a sentence is a function of the expressions which it contains (Wittgenstein 1961, 26). Consequently, a sentence which applies an abstract democratic principle to a particular political context is a function not only of the democratic principle contained but also of the keywords which capture other aspects of the substantive political context. This approach to the problem requires prior independent assessment of respondents' agreement with each of the expressions contained within the particular democratic norm items. Such a research design is rarely employed in previous American studies (see, however, Lawrence 1976, 80-100).

In this Danish study of consensus amongst the general public with respect to democratic values and norms, we replicate the American studies of the aggregate discrepancy between popular agreement about abstract democratic principles and their application to particular contexts. In addition, we extend Lawrence's study of the explanation of this discrepancy by adopting a research design that permits the assessment of the functional relationships between respondents' evaluation of democratic norm sentences and all the major expressions contained in these items, including the engaged democratic values.

A Research Design

In this section we present a research design for the study of the relationship between democratic values and their application to more concrete political situations in the Danish electorate. This research design is cast in the form of a mapping sentence in order to describe and explain the discrepancy between democratic values and norms. (The use of mapping sentences to structure social scientific research has been proposed by Louis Guttman; see for example Guttman 1971, 40-46; Levy 1976, 115-27.)

The general mapping sentence to be used in this study is reproduced in Table 1. It states in its entirety: the extent to which respondent (x) thinks that he or others ought to have permission to carry out democratic or non-democratic acts with instrumental or expressive purpose, can be mapped onto a positive or negative norm. This mapping sentence contains three main expressions, keywords or 'facets' (A, B, and C in Table 1), namely group, act, and purpose (who? what? why?). Each expression or facet has a series of values, elements or 'structs'. Thus, the facet 'group' in the mapping sentence has the elements 'self (understood here in the broad sense as positive reference group or ingroup') and 'others' (understood here as negative reference group or outgroup'). A certain combination of elements or structs from different expressions or facets of the mapping sentence is called a value of the sentence variable or a 'structuple'. For example, we posed the following survey question to the respondents in our sample, 'Do you think that we in our society ought to let students collect signatures in order to get Denmark's aid to the developing countries expanded?' This question is itself a value of the sentence variable

'democratic norm' or a sample or structuple from the population of sentences
that combine an 'out-group' undertaking a democratic act with an instrumental
purpose.

Table 2 reports the operational mapping sentence for the sentence variable 'democratic norm' adopted in this study. This mapping sentence specifies the elements or structs used to sample the three expressions or facets employed in this study: group, act, and purpose. As examples of 'in-groups' we use farmers and workers, as instances of 'out-groups' draft refusers and students. The structs used to specify the democratic acts are collection of signatures, criticism of the government, and participation in legal demonstrations, while blocking the traffic is the struct that illustrates non-democratic acts. Finally, the element 'expressive purpose' in the theoretical mapping sentence described in Table 1 is omitted from the operational mapping sentence described in Table 2. All the five political issues or structs specified in the latter mapping sentence (get Denmark out of the Common Market; get Denmark's aid to the developing countries expanded; fight unemployment; prevent the adoption of nuclear power in Denmark; lift the ban on narcotics) refer to instrumental purposes.

These imbalances in the investigator's research design are due partly to our special interest in democratic values, partly to the scarcity of space in our survey questionnaire. Even the reduced operational mapping sentence generates 116 distinct values of the sentence variable 'democratic norm'; in other words it contains 116 possible combinations of two or three structs from the three facets. Some of these questions describe quite improbable situations and are therefore omitted. (For example, we did not ask our respondents whether they thought that farmers should be allowed to participate in legal demonstrations in order to lift the ban on narcotics.) Other possible survey questions were eliminated in order to maximize the variation in the contents of the different democratic normative items. The final result of this process of elimination was sixteen questions used to measure our respondents' tolerance toward democratic and non-democratic acts in different hypothetical situations. Appendix A reproduces the survey questions and associated predetermined answers which our respondents could give to the interviewers.

Having described in some detail the procedures used by the investigator for the purpose of constructing the survey questions about democratic norms used in this study, we now turn to the procedures used by our respondents to answer these questions. Actually, these procedures will always remain secret to us. For theoretical purposes, however, it is necessary in this study to impute to our respondents a standard procedure for their formation or calculation of democratic norms during the survey interview. This standard procedure should allow us to estimate the degree to which our respondents' democratic values, as opposed to other aspects of concrete political contexts, determine their democratic norms.

We shall impose the general linear model as our rule of transformation upon the relevant survey evidence from our respondents. Thus, we expect that a respondent's degree of tolerance with respect to a given democratic norm is a linear and additive function of his attitudes toward the given reference group, the democratic act, and the political issue involved. In other words, we assume that the appropriate rule of transformation at the level of the respondent is a linear combination of independent variables.

In order to facilitate multiple regression analysis of the formation of respondents' democratic norms in the survey interview, we have adopted a measurement procedure similar to that used in a nationwide survey of American public opinion which was undertaken by the National Opinion Research Center in 1971. (Cf. Lawrence 1976, 80-100.) The validity of this measurement procedure has been confirmed by other investigators who have shown that it is possible to measure the components of respondents' attitudes separately (see e.g. Triandis 1967, 230). In our survey questionnaire we have included questions measuring respondents' evaluation of different population groups,

their tolerance of democratic and non-democratic acts in general, their attitudes toward various political issues, in addition to their tolerance of different hypothetical political contexts which include two or three of the abovementioned aspects or facets: reference group, democratic act, and political purpose. Appendix A reproduces all of these survey questions together with the corresponding pre-determined answers which respondents could give to the interviewers. The administration of these survey questions to our sample enabled us to collect evidence not only about the degree of agreement and disagreement between democratic values and norms in the mass public (as most earlier American investigations have done), but also about the degree to which respondents' attitudes toward reference groups and toward political issues are sources of disagreement.

The Discrepancy Observed

The nationwide Danish survey study took place in the Fall of 1979. The National Institute of Social Research administered the interviews August 21-September 9, 1979, to 1478 Danes distributed throughout Denmark. This achieved national sample of 1478 respondents represented 76 percent of the members of the original target sample. The nationwide study was preceded by a pilot study of 135 respondents living in four municipalities on the island of Funen (Pedersen 1980, 24-43). This section reports the findings regarding the degree of discrepancy between support for democratic values and norms and confirms the expectation based upon previous American studies that the general public supports democratic values more in the abstract than in their application to concrete political contexts.

The following question was used to measure respondents' tolerance of certain democratic and non-democratic acts: 'I am going to mention certain actions which people take at times in order to protest or merely to show what they believe. For each individual action you are asked to tell me whether you generally believe that it should always be permitted, sometimes be permitted, or never be permitted: 1. to collect signatures; 2. to criticize the government, for example in a speech at a meeting or in a letter to a newspaper; 3. to participate in legal demonstrations; 4. to block the traffic.' Table 3 reports the distributions of responses to this question.

Table 3 shows that democratic majorities are tolerant with respect to the three democratic acts and intolerant with respect to the non-democratic act. In addition, a democratic consensus (defined here as a qualified majority of 75 percent — the midpoint between democratic majority and unanimity) exists with respect to three of the four acts: 65 percent of the respondents believe that collection of signatures should always be permitted, 80 percent that criticism of the government should always be permitted, 76 percent that

participation in legal demonstrations should always be permitted, and 2 percent
that people should always be allowed to block the traffic.

Table 3 also shows that a democratic consensus exists with respect to all four acts in the abstract, if we focus our attention on the percentage of respondents who prefer the answer category 'should never be permitted'. These questions, which admittedly form a very modest purposive sample from the almost endless conceivable population of democratic value questions, produce an almost complete democratic consensus about tolerance of the three democratic acts and intolerance of the nondemocratic act. If our respondents apply their democratic values consistently to their democratic norms, then these percentages cited from Table 3 indicate the minimal limits for their willingness to apply their democratic values to concrete political contexts.

A comparison between the survey evidence presented in Tables 3 and 4 allows us to observe the aggregate-level discrepancy between democratic values and norms in the Danish electorate. We notice that attitudes toward the four values of the sentence variable 'democratic norm' (structuples) which we constructed in order to measure attitudes toward the acts of collecting signatures and blocking the traffic, when these acts are engaged in more specific political contexts, show no marked difference between tolerance of act in general and in specific contexts. By contrast, tolerance of criticism of the government and of participation in legal demonstrations shows a clear tendency to diminish when applied to more concrete political contexts. Of the five questions dealing with criticism of the government, only one produces a level of tolerance equal to that of the act in general: 81 percent would permit farmers to criticize the government repeatedly. Intolerance is particularly prevalent with respect to draft refusers' repeated criticism of the government (which only 56 percent will permit) and with respect to criticism of the government because it does not want to lift the ban on narcotics (which only 44 percent of the interviewees will allow).

A similar picture emerges with respect to the seven questions that measure

respondents' tolerance of participation in legal demonstrations. Only two questions produce a level of tolerance toward particular hypothetical situations equal to that displayed toward the act in general: 78 percent of the respondents will permit farmers to hold a legal demonstration, and 79 percent will allow workers to demonstrate against the use of nuclear power in Denmark. By contrast, only 49 percent of our interviewees will permit people to demonstrate in order to get the ban on narcotics lifted, and only 62 percent will allow draft refusers to demonstrate. Finally, if one compares the percentages of respondents who believe that a democratic act should never be permitted (or that a nondemocratic act should be permitted), we observe in all sixteen possible comparisons

between Table 3 and Table 4 that the former percentage is markedly smaller than the latter. In other words, more respondents systematically refuse to permit the three democratic acts in concrete political contexts than in the abstract, and more respondents would permit the undemocratic act in specific political situations than in the abstract.

In this section we have compared the univariate response distributions produced by sixteen democratic norm questions with the univariate response distributions produced by four democratic value questions. The results indicate that consensus exists in the Danish electorate with respect to the democratic values measured in this study. On the other hand, when these democratic values become engaged in particular political contexts, the respondents tend to diminish their level of tolerance: nine of the different values of the sentence variable 'democratic norm' enjoy the support of a consensus among our respondents, five are supported by a majority, and two are supported by a minority.

Thus, the data from this Danish study corroborates the conclusion from previous American public opinion studies. Because of differences in research design, we are unable to tell whether this conclusion holds more true in the American electorate than in the Danish electorate. Even Lawrence's analysis of the 1971 National Opinion Research Center survey study, which also treats reference groups and political issues as sources of disagreement between democratic values and norms, cannot be directly compared to the investigation reported here because of differences with respect to research design, method of analysis, reference groups, and political issues. With respect to the existence — as opposed to the magnitude — of a discrepancy between the level of support observed for democratic values and norms in both the American and Danish electorate, there can, however, be no doubt.

Explanation of Discrepancy by Bivariate Correlation Analysis

In pursuing the theoretical purposes of this study we have postulated that respondents calculate their answers to our democratic norm questions as functions of their attitudes toward the major expressions or facets contained in these questions, namely the involved reference group, the engaged democratic value, and the specific political purpose. Agreement between democratic value and norm at the level of the individual respondent may arise in at least three ways which cannot be distinguished empirically in this study: because the respondent always makes up his mind about the norm solely on the basis of the engaged democratic value; because he makes up his mind about the norm on the basis of all the engaged expressions or facets but attributes so much

value to the democratic act that his selection of norm always is democratic; or because he believes that it is socially desirable to select the democratic value, then recognizes the intention of the investigator in the survey questionnaire, and chooses the democratic norm in order to be consistent within the survey interview.

On the other hand, a discrepancy between democratic value and norm at the level of the individual respondent may arise because respondents evaluate the democratic norm not only on the basis of the engaged democratic value but also on the basis of their attitudes toward the given group of actors and the specific political purpose, and because they find that their personal benefit from the application of the given democratic principle is less than the subjective cost of letting a controversial group of political actors promote a disliked political purpose. In other words, value-consistency at the level of the individual respondent may conflict with group-consistency or issue-consistency or both. Such conflict may be observed in our data in the form of imperfect correlations between democratic norms and values, and in the form of positive correlations between norms and attitudes toward the groups of political actors and toward their political purposes.

In this section we demonstrate the functional relationships between democratic norms, on the one hand, and group evaluations, democratic values, and political issue attitudes on the other. We rely on a series of hypothetical situations that systematically vary the elements or structs in the three facets of the operational mapping sentence reported in Table 2. Thus, our respondents' replies to the survey questions that have been reproduced in Appendix A permit estimation of the statistical relationships between evaluation of a group, for example draft refusers, and democratic norms, as well as the sensitivity of this correlation to different combinations of democratic values and political issues. Or the investigator may focus on a democratic value, such as criticism of the government, and estimate its effect on democratic norms across a series of political actor groups and political issues.

Ideally one would want to analyze a large number of questions sampling a large part or all of the surface of the operational mapping sentence, but in practice we had to settle for a modest number of questions as illustrations of the enormous set of possible combinations. (For a description of the logic of this part of the research design, see, e.g., Ross & Smith 1968, 373-81.) Tables 5 and 6 present respondents' answers to twelve of these questions and report comparative analyses of the bivariate correlations, measured by the Gamma coefficient, between respondents' democratic norms and their group evaluations and political issue opinions.

In Table 5 we include farmers and draft refusers as examples of groups of
political actors. In addition, we use criticism of the government, participation
in legal demonstrations, and blocking the traffic as examples of democratic

and non-democratic acts. Cross-classification of these two facets thus produces six hypothetical political situations that respondents are asked to evaluate. We notice in Table 5 that respondents' evaluations of farmers apparently have relatively little effect on their democratic norms: the bivariate correlations are rather small and even negative, for example —.10 with respect to participation in a demonstration. By contrast, respondents' evaluation of draft refusers correlates markedly with their democratic norms: for example, the Gamma coefficient is .49 with respect to participation in a demonstration.

The evidence reported in Table 5 suggests that the relationship between group evaluation and democratic norm may be relatively constant for different democratic values, but that it varies markedly from group to group. Perhaps the correct specifying factor is the degree to which the given group is controversial in the eyes of the Danish citizenry. While farmers are well-liked, draft refusers constitute a more controversial minority group. 76 percent of the respondents have a positive or very positive attitude toward farmers, while only 26 percent have a positive or very positive attitude toward draft refusers. Thus, farmers make up an 'in-group'; draft refusers an 'out-group'. Table 5 also shows that the bivariate correlations between democratic norms and values are systematically higher than those between democratic norms and group evaluations, relatively stable between the two comparison groups, and quite similar across the three democratic and non-democratic values as well. For instance, attitudes toward participation in a demonstration correlates .75 with respect to farmers and .76 with respect to draft refusers.

Table 6 reports an analysis analogous to that underlying the results reported in Table 5, namely a comparative analysis of the bivariate correlations between respondents' democratic norms and their political issue opinions as well as democratic values. Table 6 includes as examples of democratic values criticism of the government and participation in legal demonstrations, and, as examples

of political issues, fighting unemployment, lifting the ban on narcotics, and getting Denmark out of the Common Market. Cross-classification of these elements in the operational mapping sentence thus allows the investigator to analyze respondents' attitudes toward six hypothetical political situations. This design makes it possible to compare the bivariate correlations between democratic norms and democratic values while controlling for political issues and between democratic norms and opinions about political issues while controlling for democratic values.

Table 6 shows that the bivariate correlations between democratic norms and democratic values are high across opinions about political issues ranging from .55 to .76. In addition, Table 6 shows that these correlation coefficients are systematically higher than the corresponding correlation coefficients between democratic norms and political issue opinions that range from .19 to .46. For example, the Gamma coefficient is .74 between democratic norm and the democratic value of participation in legal demonstrations with respect to the issue of the government's fight against unemployment, whereas it is only .41 between democratic norm and attitude toward the government's fight against unemployment with respect to the democratic value of participation in legal demonstrations.

To sum up, in this section we have begun to attempt to explain the discrepancy between the level of support for democratic values and norms observed in our sample of the Danish electorate. Tables 5 and 6 reported the results from bivariate correlation analyses of the evidence. This evidence was generated according to a certain language game, to use Wittgenstein's expression, namely sets of questions constructed by the investigator and corresponding sets of answers offered by the respondents. The survey questions included not only a purposive sample of sixteen values of the sentence variable 'democratic norm' but also questions measuring respondents' attitudes toward each major element or keyword contained in each of the sixteen distinct values of our sentence variable 'democratic norm'. Although it was, of course, impossible for us to follow exactly how the respondents made up their minds about the sixteen democratic norm questions, we postulated that they calculated their answers as functions of attitudes toward the three expressions or facets — group, act, and purpose — in our operational mapping sentence described in Table 2. Tables 5 and 6 confirm that this postulate apparently is empirically valid, since group evaluations, democratic values, and political issue opinions are all correlated markedly and repeatedly with democratic norms. The statistical results reported in Tables 5 and 6 suggest, with respect to the relative weight attached to the three keywords in respondents' answers to the democratic norm questions, that democratic values have the largest weight, while group evaluations and political issue opinions have somewhat smaller weights. Consequently, the discrepancy between democratic values and norms at the level of individual respondents appears to be explained by the fact that group evaluations and political issue opinions show considerable influence on the formation of democratic norms.

Explanation of Discrepancy by Multiple Regression Analysis

A statistical problem that is inherent in our interpretation of the bivariate correlation analysis described above is the implicit assumption of uncorrelated independent variables. If this assumption is invalid, then the correlation between dependent and independent variables may be due to a third variable. For purposes of solving this problem we introduce statistical controls for third and fourth variables by adopting multiple regression analysis.

Multiple regression analysis estimates the functional relationship between an independent variable and a dependent variable in the form of a regression coefficient, controlling for the effects of the remaining variables in the model. Thus, the assumption of uncorrelated independent variables in the statistical analysis is waived. Instead the ordinary assumptions underlying elementary

multiple regression analysis are introduced, such as linear, additive relationships
between dependent and independent variables as well as measurement of
variables at the interval level.

We re-analyzed all the bivariate correlations using statistical controls for third and fourth variables with the aid of cross-tabulation. This analysis failed to turn up any clear evidence of statistical interaction between the independent variables: the bivariate correlations showed few and unsystematic changes and hardly any changes of sign with statistical control. We also assigned sets of positive, successive integers to the answer categories from which respondents could choose their replies to our survey questions. Consequently, we believe that the theoretical mapping sentence or rule of transformation at the level of the respondent can properly be roughly expressed mathematically as a linear combination of our three independent variables. We do this in our multiple regression analyses of respondents' formation of democratic norms and treat the resulting estimated equations as our operational mapping sentences or rules of transformation at the level of the respondent.

Table 7 reports a set of standardized regression coefficients from sixteen estimated equations which express the degree of linear relation between democratic norm, group evaluation, democratic value, and political issue opinion. It also lists the corresponding squared multiple correlation coefficients which indicate the goodness of fit of these regression equations. It is seen from the table that the effect of group evaluations on democratic norms is smaller than the effect of democratic values in all six hypothetical situations in which comparison is possible. This is at variance with American studies which find that '... civil liberties attitudes in concrete situations are better predicted by attitude toward the group in question than by support for the abstract principle of free speech ..'. (Quoted from Zellman 1975, 44; see also Sullivan, Piereson & Marcus 1982, 225).

Table 7 also shows that the effect of group evaluations on the formation of the Danish respondents' democratic norms is larger for an 'out-group' (in this case draft refusers) than for an 'in-group' (in this case farmers). These results may be due to the relatively homogeneous character of Danish society. Denmark has no significant language barriers (unlike, e.g., Belgium and Switzerland), racial or ethnic distinctions (unlike, e.g., the United States), religious divergencies (unlike, e.g., Holland and Northern Ireland) or regional conflicts (unlike, e.g., England and Spain). It is therefore more difficult for the investigator (and for political leaders) to locate controversial 'out-groups' with a marked effect on the formation of democratic norms in Danish society than in, say, American society. One is reminded that the American investigations cited above introduced with great effect communists, atheists, socialists, blacks, and other negative reference groups in their survey questionnaires in order to study the relationship between democratic values and norms in the American electorate.²

Table 7 also confirms the bivariate correlation analysis showing that political issue opinions have a variable but sometimes quite strong effect on democratic norms. This effect is, however, less than the effect of democratic values on the formation of our respondents' democratic norms in all six hypothetical situations where comparison is possible. The standardized regression coefficient ranges in magnitude from .10 to .23 for opinions about political issues, but from .26 to .46 for democratic values. This evidence agrees with the finding by Lawrence that political issue opinions have less effect on democratic norms than do democratic values. He writes, 'Although both general norms and issueorientation have a large independent impact on tolerance, when their effects

are compared general norms emerge much the stronger of the two.' (Lawrence 1976, 96). This finding is also confirmed by our multiple regression analysis of the last four democratic norm questions in Table 7 which contain as independent variables group evaluation, democratic value, and political issue opinion. A comparison shows that in three of the four cases the democratic value has a markedly stronger effect on the democratic norm than political issue opinion, and in the fourth case the two effects are about equally strong. Finally, our analysis agrees with the previous multiple regression analysis of the comparative effects of group evaluations and democratic values on the formation of democratic norms; in all four cases the effect of group evaluations is markedly smaller than the effect of democratic values.

An interesting problem for further empirical work is the identification of patterns or structures in the coefficients which the multiple regression analysis produces. Such an effect coefficient may be regarded as a quantitative measurement of the emphasis, stress or weight which respondents attribute to a keyword in a sentence (or to an aspect of the context) when they decide whether to approve or disapprove of the situation described in the sentence. The meaning of a sentence depends, as is wellknown, upon the stress pattern which is attributed to the words of the sentence in pronounciation and reception. The interviewers from the National Institute of Social Research were not instructed to read the questions about respondents' democratic norms in any particular way, and they probably accentuated the keywords in these questions more or less equally. But it is beyond doubt that our respondents themselves attributed different weights to the keywords, when they perceived a question about democratic norms and combined their attitudes in order to reach an answer.

The political philosophers who discussed the concept of democratic consensus before the advent of systematic empirical investigations assumed that ordinary citizens should and would place the full weight on the engaged democratic value. The empirical portion of this assumption has not been verified. A more plausible assumption states that a respondent's distribution of emphasis or stress on the keywords of a sentence depends upon the centrality of these keywords to him, taken here to mean the amount of time he has invested in thoughts or worries about the given phenomenon.

Our study indicates that there are robust patterns or structures of effect coefficients in the rules of transformation employed by Danish citizens to answer our sixteen democratic norm questions. In all ten comparisons available, the group component has a markedly lower effect coefficient than the democratic value component. In addition, in nine out of ten comparisons the issue component has a lower effect coefficient than the democratic value component. Finally, our multiple regression analysis of our respondents' answers to the last four democratic norm questions reported in Table 7 indicates

that the group component is weaker than the issue component with respect
to coefficients of effect on the formation of democratic norms.

In short, the evidence from our national survey (and the data from the local pilot study of 135 respondents) suggest that Danes attach the highest priority to the engaged democratic value, less priority to the engaged political issue opinion, and least priority to the engaged group evaluation. Moreover, this pattern or structure of priorities was found to be quite robust in different subgroups of our Danish national sample as well.

We replicated the multiple regression analyses reported in Table 7 separately for respondents who discussed politics and for those who did not; for respondents with seven years of formal education or less and for those with more than seven years of formal education; and for respondents with low, moderate, and high degrees of subjective social competence.³ These personal background attributes were selected for further statistical analysis from the much larger set available in the survey questionnaire, because they were the ones which correlated the most with respondents' democratic values and norms within the political, social, and psychological spectra of background variables. The results of these multiple regressionsanalyses are not reported here in tabular form, but they show that the fit to the structure of effect coefficients (and the goodness of fit between regression models and the data measured by the square of the multiple correlation coefficient)⁴ found in the sample of the Danish electorate as a whole was replicated both among respondents with low and high levels of formal education and among respondents with low, moderate, and high degrees of subjective social competence. Only among respondents who discussed politics was the structure of effect coefficients described in Table 7 significantly more pronounced (and the goodness of fit between the regression equations and the data significantly higher) than among respondents who did not discuss politics, but these differences were of borderline statistical significance at the .05 level.

Yet the main impression is the great similarity with which respondents in different subgroups of our national sample combine their attitudes toward different aspects of hypothetical political situations in order to reach their answers to our democratic norm questions. In particular, the well-known correlation in the American literature and systematic, strong correlation in the Danish survey data between formal education and democratic norms must be due in the Danish case to the correlation between formal education and democratic values, not to any differential accentuation of democratic values in the calculus of answers to our sixteen democratic norm questions among subgroups of poorly and well-educated respondents.

Conclusion

In summary, we have shown that there exists a discrepancy between the democratic norms and the democratic values held by a national sample of adult Danes. In addition, we have shown that at the aggregate level this discrepancy takes the form of a systematic tendency to support democratic values more in general than in particular political contexts. The explanation for these observed phenomena is, we have asserted, that people take all aspects of a particular hypothetical situation into account when they calculate their answers to a given democratic norm question. They do not consider the democratic values alone, as political philosophers would have it, but instead they emphasize the democratic value involved most, the political issue less, and the group in question the least.

This pattern or structure of effects is reminiscent of the concept of legal hierarchy. The democratic acts of political participation and protest are decided by the Danish constitution, the political issues are settled by Danish law which must conform to the constitution, and the various subgroups in the Danish population must obey the law. In analogous fashion, the typical Danish respondent who forms an answer to a democratic norm question attaches the highest priority to the democratic value contained in the question, less priority to the political issue, and least priority to the population group. Consequently, there will be a systematic tendency for the observed level of support for democratic norms to be lower than the observed level of support for democratic values, if the typical level of support for the group or the typical level of support for the political issue, or both, is lower than the typical level of support for the democratic value. This is almost bound to be the case not only in Denmark but in constitutional democracies generally, since democratic values enjoy the support of the constitution, political issues the support of parliamentary party groups, and population subgroups the support of citizens depending upon the degree to which they are perceived as promoting or menacing the values of these citizens.

---

NOTES

1. For additional examples see Stouffer 1963, 31; McClosky 1964, 366-67. Evidence from an American nationwide survey conducted by the National Opinion Research Center in 1971 also shows markedly higher support in the electorate for democratic values generally than for their application to more concrete political contexts, see Lawrence 1976, 80-100. Data from Israel which confirm that democratic values are more popular than democratic norms, may be found in Simon & Mann 1977, 283-92.

2. A comparative study of the weight of different components in American, German, and Japanese students' attitudes toward complex stimulus persons (varying with respect to race, occupation, religion, and nationality) found marked cross-national differences. For example, the strongest component in the United States was race, but in Germany and Japan it was occupation, see Triandis, Davis & Takezawa 1965, 540-51.

---

3. Length of formal education was measured by the interviewees' answer to the following question, 'How much formal education do you have?' Frequency of political discussion was measured by respondents' answer to this question, 'How oftendoyou discuss politics with others?' Amount of subjective social competence was measured by an index of agreeing or disagreeing responses to the following five statements: i don't like giving orders to others', 'I often find it difficult to be free and easy with other people', 'I dislike having to talk in front of a group of people', 'I doubt whether I would make a good leader', and 'I would rather not have responsibility for others'.

4. The coefficients in the right-hand column of Table 7 express the ability of the estimated regression equations to explain the variance of the respondents' answers to our democratic norm questions. These coefficients show that we can at best explain only one third of the variance and typically about one fifth. Several technical factors which do not concern the validity of the general linear model, may help to reduce its explanatory power, such as the dichotomous nature of the dependent variable and inadequate precision with respect to measurement of respondents' positions on the four variables. Previous research on political attitudes indicates, however, that even the best models will probably fail to acquire a degree of explanatory power in excess of fifty percent. This is shown, for example, in the research done on Americans' candidate preference in presidential elections — attitudes which respondents do not think about for the first time when they are interviewed. There respondents' candidate preferences are explained by a linear combination of several partisan attitudes which are located most closely to candidate preference in the postulated causal field, see e.g. Stokes 1966, 19-28.

REFEFENCES

Dahl, R.A. 1961. Who Governs? Democracy and Power in an American City. New Haven: Yale
University Press.

Guttman, L. 1971. 'Social Problem Indicators', The Annals of the American Academy of
Political and Social Science, 40-46.

Lawrence, D. 1976. 'Procedural Norms and Tolerance: A Reassessment', American Political
Science Review 70, 80-100.

Levy, S. 1976. 'Use of the Mapping Sentence for Coordinating Theory and Research: A Cross
Example', Quality and Quantity 10, 117-25.

McClosky, H. 1964. 'Consensus and Ideology in American Polities', American Political Science
Review 58, 361-82.

Pedersen, J.T. 1980. 'Sammenhængen mellem demokratiske værdier og normer belyst ved en fynsk
undersøgelse', Økonomi og Politik 54, IA-A'i.

Prothro, J.W. & Grigg, CM. 1960. 'Fundamental Principles of Democracy: Bases of Agreement
and Disagreement', Journal of Politics 22, 276-94.

Ross, J. & Smith, P. 1968. 'Orthodox Experimental Designs', 373-81, in Blalock, H.M. and
A.B. Blalock (eds.), Methodology in Social Research. New York: McGraw-Hill Book
Company.

Simon, R.J. & Mann, K. 1977. 'Public Support for Civil Liberties in Israel', Social Science
Quarterly 58, 283-92.

Stokes, D.E. 1966. 'Some Dynamic Elements of Contests for the Presidency' American
Political Science Review 60, 19-28.

Stouffer, S.A. 1963. Communism, Conformity, and Civil Liberties. Glouchester, Mass:
Peter Smith.

Sullivan, J.L., Piereson, J. & Marcus, G.E. 1982. Political Tolerance and American Democracy.
Chicago: The University of Chicago Press.

Triandis, H.C. 1967. 'Toward an Analysis of the Components of Interpersonal Attitudes', in
Sherif, C.W. and M. Sherif (eds.), Attitude, Ego Involvement, and Change. New York:
John Wiley and Sons.

Triandis, H.C., Davis E.E. & Takezawa 5.1., 1965. 'Some Determinants of Social Distance
among American, German, and Japanese Students', Journal of Personality and Social
Psychology 2, 540-51.

Wittgenstein, L. 1961. Tractatus Logico-Philosophicus. London: Routledge and Kegan Paul.

Zellman, G. 1975. 'Antidemocratic Beliefs: A Survey and Some Explanations', Journal of Social
Issues 31, 31-53.