The Determinants of Political Knowledge in Comparative Perspective

Transcription

1 ISSN The Author(s) The Determinants of Political Knowledge in Comparative Perspective Kimmo Grönlund* and Henry Milner Political knowledge is a powerful predictor of political participation. Moreover, what citizens know about the political system and its actors is a central aspect of informed voting. This article investigates how and why political knowledge varies between citizens. The analysis is comparative and based on data from the Comparative Study of Electoral Systems. At the micro level, the results confirm results from national surveys specifically that education explains what citizens know about politics. It is found in a contextualized analysis, however, that the effect of education varies with the country s degree of economic redistribution. In more egalitarian countries, political knowledge is less contingent on education attained than in more inegalitarian countries. Similarly, education seems to have a stronger effect in countries with majoritarian electoral systems compared to countries with proportional systems. The political knowledge of the citizenry is a key aspect of democracy. Democratic theorists from John Stuart Mill to Robert Dahl have stressed the importance of political information in democratic decision making. Since representative democracy is based on the delegation of power from citizens to representatives, the defining moment is at the time of elections: for democracy to function properly, electors need to have sufficient knowledge of the political system and the political actors in order to cast a meaningful vote. If they are to ensure that political parties and leaders are accountable for their actions, voters need information to evaluate their performance. They need to be able to compare parties commitments and manifestos in a prospective manner against their own political preferences and, in retrospect, to know enough about the parties record to give some weight to the credibility of their commitments (Manin et al. 1999, 44 6). Even though contemporary democratic political systems are egalitarian in principle (i.e. they are based on universal suffrage where every citizen has a vote), practice is less egalitarian since citizens are differently endowed with resources that can be used for political activity and influence (Verba et al. * Kimmo Grönlund, Åbo Akademi University, Social Science Research Institute, PO Box 311, FIN Vasa, Finland. kimmo.gronlund@abo.fi 386 Scandinavian Political Studies, Vol. 29 No. 4, 2006

2 1978, 1). Some normative theorists suggest that democratic systems that restrict political equality to equal opportunity to vote are deficient. Dahl (1989, 114 5) claims that being entitled to vote and take part in political decision making does not constitute political equality if these powers are unequally exercised in practice. The most basic of resources that can be used for political activity and influence is political information. Rational choice theories in particular emphasize political information as a precondition for political participation. Moreover, as observed by Downs (1957, ), the costs of acquiring the needed information are inversely related to a person s economic resources: voting is more costly to the poor in a society than to its wealthier citizens. Voting is dependent on both the citizen s preferences and his or her resources; if it were costless, only the indifferent would abstain and every citizen with a preference would vote. 1 As information costs rise, voters with weak preferences and low resources are more likely to abstain. 2 Rational choice thinkers, operating in an economic framework of analysis, tend to regard political information as similar to consumer information and the political market as analogous to product markets. Democratic theorists see a more complex relationship: political knowledge constitutes a multifaceted link between the people and the political elite. While voters need information to hold elites accountable for their actions, the elites can alter, filter or even block relevant political information and in that way influence popular opinion (Maravall 1999). The latter has traditionally led observers concerned with democratic participation to evaluate political institutions and actors critically to be worried about elites blocking or distorting, rather than fostering, the dissemination of political knowledge. The unstated assumption in this approach is that the informed democratic citizen is born rather than made. Yet the informed participating citizen is proving increasingly elusive. Empirical data increasingly point to low and apparently declining political knowledge in certain democratic countries (e.g. Bennett 1988; Howe 2002), which is linked to a decline in the traditional forms of political participation. In established democracies, voter turnout has been declining since the 1980s (Norris 2002; Franklin 2004) reflecting a decline in party membership (Schmitter 2001) and party identification (Manin 1997). 3 There is a clear parallel between this decline and a trend toward an increasing distrust of parliaments, governments and political parties (Pharr et al. 2000). Thus, paradoxically, we find ourselves in a context in which there is a greater need for political knowledge in order to evaluate the performance of parties and candidates, yet a growing tendency to distrust and disregard the traditional sources of such knowledge. 4 This makes for a potentially worrisome combination: if we are to understand and possibly address the decline in democratic participation, we need to know better how political institutions can stimulate the dissemination of political knowledge. It is to be hoped that this state of affairs will Scandinavian Political Studies, Vol. 29 No. 4,

3 spur on other efforts to assemble and interpret comparative empirical data on political knowledge. The present article has several aims. First, it tests the determinants of political knowledge at the individual level in a comparative setting. This phase aims to determine whether the usual suspects (i.e. predictors of political knowledge in national surveys) also explain knowledge cross-nationally. In this analysis, socio-demographic and economic variables indeed explain variation in political knowledge, with education as the most important explanatory variable. Then, second, we bring in contextual variables in order to clarify how the country s macro structure affects the relationship between levels of education and political knowledge. The macro indicators used pertain to the distribution of the country s economic welfare and its electoral system. Our analysis suggests that these macro structural constraints influence the association between formal education and political knowledge. This finding and its implications are discussed in the final part of the article. Data: Political Knowledge as Dependent Variable We have noted the accumulating data showing an apparently growing proportion of citizens lacking the knowledge needed to act as effective democratic citizens. However, these conclusions are based on figures from within single countries, most frequently the United States. There is also a burgeoning literature linking political knowledge to voting (Junn 1991; Delli Carpini & Keeter 1996; Milner 2002). At the individual level, controlling for education and other relevant factors, knowledgeable voters turn out significantly more frequently than other citizens, whereas the politically ignorant more often remain passive at elections (see, e.g. Grönlund 2003). Yet, for all the burgeoning research, there is still little comparative work that examines the sources of political knowledge (i.e. studies in which political knowledge serves as dependent variable). To deepen our understanding of this phenomenon, we need to identify and explain cross-national differences in levels of political knowledge. Yet such comparisons present a problem as we do not yet have a standard battery of questions used in international surveys to gauge a country s level of political knowledge comparable, for example, to questions used as indicators of political efficacy or social trust. 5 Fortunately, there is one rich store of largely unexplored data that allows us to gain an insight into important differences in political knowledge. This is the data on political knowledge gathered in surveys in the Comparative Study of Electoral Systems (CSES) based at the University of Michigan (Sapiro & Shively 2002). The first CSES wave, which we use in this article, was carried out between 1996 and The data allow us to analyse the different factors that may explain differences 388 Scandinavian Political Studies, Vol. 29 No. 4, 2006

4 in political knowledge among individuals and thus begin to fill the gap in our understanding of the sources of political knowledge. The first step is to define and operationalize the concept of political knowledge. This is not a simple matter. Political information is often used as a synonym for political knowledge, but knowledge is something more since it includes a cognitive dimension. Information becomes knowledge if the person is able to process it, to distinguish correct from incorrect information (Sartori 1987, 117). It thus requires a certain level of skill. Luskin (1987), for example, refers to political sophistication, which he defines as the quantity and organization of a person s political cognitions. While it is possible to compare respondents ability to analyse the same information, this is of little use for the purposes of comparing respondents in different countries since the quality of such analysis is at least to some extent context dependent. On the other hand, we can compare individuals simply on their level of correct information since we can safely assume differences in cognitive abilities to be randomly distributed. This is the case here since the data are derived from surveys in which the total number of respondents exceeds 40,000. The type of political information needed to cast an informed vote can be placed into two categories. One type of question gets at the workings of the institutions of the political system and the rules of the game. The other type pertains to the ongoing political debate, measuring voters knowledge of the political actors, parties or persons, and their differences in relation to policies, issues or ideologies. The political knowledge questions selected by the CSES members that serve as the basis for this study generally fit this pattern, with respondents typically asked about the institutions of government of their country, as well as the leading politicians or national parties. Questions of the first type are more common overall, but there is a fair degree of variation in the use of the two types 6 and there are also, infrequently, questions about international politics. In addition, though all the questions have been closed-ended, the number of alternatives afforded to the respondents has varied: while most countries provide four alternate responses, some have used true/false questions. Regrettably, no special efforts were undertaken in the CSES to standardize the content of the political information questions used in the national surveys. Nevertheless, some effort was made to attain a measure of standardization in another dimension that of the difficulty of the questions. The CSES has tried to achieve comparability by instructing the national research teams to formulate the three political information questions in such a manner as to result in one question being answered correctly by two-thirds, another by one-half and the last by one-third of the respondents. While this had a useful constraining effect on the variety of questions asked, it could not, and was not intended to, bring about uniformity. Because of this, we do not treat the overall results by country (the mean score in Table 4) as reflecting political Scandinavian Political Studies, Vol. 29 No. 4,

5 knowledge differences between them. However, the large number of respondents and countries does allow for useful analyses bringing in a comparative dimension. We are able to use the individual-level data to identify variations according to micro structural variables that is, socio-demographic and socio-economic characteristics, attitudinal variables and contextual effects. Moreover, the data allow us to analyse and compare the dispersion of political knowledge within countries controlling for contextual effects linked to differences in political systems. Of the 30 countries that participated in the first wave of the CSES, questions on political information were asked in Table 1 lists the countries and sets out relevant data on their political system as well as the size of the national sample. Among the included CSES countries, there are established democracies, new democracies and non-democracies. Moreover, there is a distinct group of post-communist countries. Among these, the Ukraine is the only one that does not yet meet the democratic criteria of Freedom House; Romania joined the group of countries with moderate political rights and civil liberties in The countries electoral systems are specified together with the effective number of political parties in their parliaments, as well as their income distribution levels (Ginis), both of which prove relevant to the analysis of the dispersion of political knowledge below. 8 The electoral systems are also dichotomized into proportional (P) and majoritarian (M) systems. In addition, whether voting is compulsory is specified since requiring citizens to vote may affect the likelihood that they will acquire political information in order to cast that vote. Results In Table 2, political knowledge is linked to several personal and contextual characteristics. 9 The table provides the shares of correct answers on the three political information items asked in the survey. As we can see, half of the 44,246 respondents gave two or three correct answers, whereas 18 percent could not answer a single question correctly. It should be noted that the don t know (DK) answers are treated in the traditional manner as incorrect answers (as strongly urged by Luskin & Bullock 2004). 10 There has been a recent debate on how DKs should be treated (Mondak 2001; Mondak & Davis 2001) as evidence mounts that respondents with certain characteristics especially women are less inclined to guess. Not discouraging guessing thus inflates men s levels of knowledge: indeed, a recent study claims that because of this, 50 percent of the gender gap is illusory (Mondak & Anderson 2004). This should be kept in mind in assessing the data in Table 2, which shows, for example, that in the group with three correct answers the difference is 13 percentage points in favour of men. We did test for variation within each country 390 Scandinavian Political Studies, Vol. 29 No. 4, 2006

6 Scandinavian Political Studies, Vol. 29 No. 4, Table 1. Information on the Countries included in the Analysis Country Political system 1 Electoral system 2 Type Compulsory voting Gini index 3 Parties 4 Respondents Election year Australia Old democracy AV M Strict , Belgium 5 Old democracy PR P Strict , Canada Old democracy SMP M No , Czech Republic Post-communist PR P No , Germany 6 Old/post-communist MMP P No , Hong Kong I 7 Non-democracy PR P No 1, Hong Kong II 7 Non-democracy PR P No Hungary Post-communist MMM M No , Israel Old democracy PR P No , Mexico I Non-democracy MMM M Weak 2.9 2, Mexico II New democracy MMM M Weak , Netherlands Old democracy PR P Not enforced , Norway Old democracy PR P No , New Zealand Old democracy MMP P No , Poland Post-communist PR P No , Portugal Old democracy PR P No , Romania Post-communist PR P No , Spain I Old democracy PR P No 2.7 1, Spain II Old democracy PR P No , Sweden Old democracy PR P No , Switzerland Old democracy PR P No , Taiwan New democracy MMM M No 2.5 1, Ukraine Post-communist MMM M No , United Kingdom Old democracy SMP M No , United States Old democracy SMP M No , Notes: 1 According to Freedom House, old democracies have met the criteria for at least 20 years. 2 AV = alternative vote, MMM = mixed-member majoritarian, MMP = mixed-member proportional, PR = proportional list system, SMP = single member plurality. 3 According to United Nations Development Programme; not listed separately for Taiwan. 4 The effective number of parties according to the Laakso and Taagepera method; not applicable for Hong Kong (Norris 2004, 84 7; Klesner 2002; Ho 1999). 5 Separate samples: Flanders 2,179, Walloon 1,960 respondents. 6 The former DDR länder and East Berlin have been coded as post-communist. 7 Less than half elected by popular vote in 1998 and 2000 (CIA); the effective number of parties is not applicable. 8 Strict in one canton The Author(s)

7 Table 2. Political Knowledge in Different Groups in the CSES Sample Share of correct answers Total (%) N All ,275 Male ,739 Female ,506 Age 40, , , , , , ,090 Education 40,767 None Incomplete primary ,386 Primary ,500 Incomplete secondary ,425 Secondary ,660 Postsecondary trade/vocational ,028 University undergraduate incompleted ,066 University undergraduate completed ,864 Marital status 39,935 Married ,094 Widowed ,724 Divorced ,071 Single ,046 Household income 35,122 Lowest ,594 Second ,428 Third ,669 Fourth ,951 Highest ,480 Voted? 41,275 No ,968 Yes ,307 Party identification? 38,096 No ,050 Yes ,046 Who people vote for makes a difference 39,459 1 It won t make a difference , , , ,743 5 It can make a difference ,277 Satisfaction with the democratic process 38,960 Not at all satisfied ,669 Not very satisfied ,666 Fairly satisfied ,150 Very satisfied , Scandinavian Political Studies, Vol. 29 No. 4, 2006

8 in order to control for differences between more egalitarian welfare states and other countries, but found the gender gap to be present in each. It should be noted that the CSES rules discourage DKs by registering them as such only if the respondents come up with the DK alternative themselves. However, some countries chose wordings that did not discourage the DK alternative. A glance at Table 2 shows that of all the tested variables, education is most closely related to political knowledge. Neither age, party identification, the perception of one s vote making a difference nor democratic satisfaction co-vary with political knowledge to the same degree as education. Clearly, education increases political information levels; so does household income, as we can see, though the effect is weaker. As expected, moreover, voters are more knowledgeable than non-voters: only 15 percent of voters have no knowledge (i.e. answered none of the three questions correctly) compared to 27 percent of non-voters. In the next section of this paper, we investigate differences in the effect of education in different countries and what these differences may mean in terms of contextual variables, more precisely: characteristics pertaining to economic redistribution and the electoral system. Before doing so, we go beyond bivariate associations to multivariate ones in an effort to determine how well the variables in Table 2 explain differences in political knowledge. In Table 3, three logistic regression models are presented. In the first model, called individual, only the micro-structural variables are entered. Due to a very large number of missing values, the income level of the respondent s household has not been included in the regression analyses (see the number of valid cases in Table 2). Education remains the most powerful predictor of knowledge in the multiple regression, followed by age. Overall, older, better educated, employed males are the most likely to be politically knowledgeable. The Nagelkerke pseudo R 2 of 0.14 indicates that the model possesses explanatory power. Being married raises political knowledge, but this is offset by having small children. All in all, the structural variables at the micro level draw a picture of citizens who are more integrated into society as being the more politically informed hardly a surprising result. We need to be careful in comparing the second model, labelled preferential, with the first model due to the large number of missing cases. Nevertheless, the consistency is undeniable: when preferential variables are included, education still prevails as the most important independent variable, while age and gender also remain important. Of the new variables, having voted clearly increases the likelihood of possessing political information which supports the link between information and voting as does party identification. By contrast, believing that voting makes a difference only marginally increases the level of political knowledge, while satisfaction with the democratic process has no effect. The final model is a parsimonious compromise. There are sufficient independent variables to create a good fit without losing too many cases. Even Scandinavian Political Studies, Vol. 29 No. 4,

9 Table 3. The Determinants of Knowledge in the Whole Sample (Logistic Ordinal Regression) Individual Preferential Final model B SE p B SE p B SE p Education *** *** *** Age/ *** *** *** Male *** *** *** Married *** *** *** Children at home *** *** *** Widowed *** *** *** Divorced * * Unemployed ** Voted? *** *** Party identification *** *** Voting makes difference *** *** Satisfaction with democracy Threshold *** *** *** Threshold *** *** *** Threshold *** *** *** N 40,508 34,759 36,107 Missing cases (%) Model Chi 2 5,735 *** 4,966 *** 5,342 *** 2 Log likelihood 35,031 78,510 68,232 Nagelkerke R Notes: *Significant at the 0.05 level, **significant at the 0.01 level, ***significant at the level. All variables are coded into a scale from 0 to 1. Education: seven values from no education to completed university undergraduate. Age is in years, divided by 100. Male, married, children, widowed, divorced, unemployed, voted, party identification are dummies. Voting makes a difference (five values) and satisfaction with democracy (four values) are ordinals. so we are left with 13.2 percent missing cases. This is due largely to the 3,179 missing cases resulting from the inclusion of party identification. We have nevertheless kept this variable given the established fact that political knowledge facilitates citizens ability to link their policy views to their evaluations of parties and politicians. The final model has a Chi 2 of 5,342 and a pseudo R 2 of 0.15, which is reasonable for a large sample like that of the CSES. In the final model, education remains the most important independent variable, which is shown by the logit-coefficient B (2.11) and the small standard error. Consistent with the earlier models, being older and male increase the probability of being politically informed, and all of these micro-structural variables are in fact more closely associated with political knowledge than party identification and voting versus non-voting, which are nonetheless also predictors of political knowledge. Political Knowledge, Education and Economic Redistribution An often cited manifestation of the ecological fallacy is that though better educated individuals are more informed, this does not mean that better 394 Scandinavian Political Studies, Vol. 29 No. 4, 2006

10 educated societies (i.e. those with the highest average years of schooling) are the best informed (see, e.g. Franklin 2004, 17). This is in part simply because knowledge is not equally distributed in all societies; the gap between the less and more educated can and does vary. The same applies to political knowledge. To make sense of differences in political knowledge we need to know about distributions, not simply central tendencies. This distinction is central to the work of Milner (2002), who uses the term civic or political literacy rather than knowledge or sophistication to compare societies: a given society s level of civic literacy reflects the proportion of its citizens with the knowledge and skills to effectively exercise that citizenship. 11 The data cited here potentially shed light on two key aspects of the discussion of civic literacy. The first aspect centres on the relationship of political knowledge to redistribution. Milner (2002, 13) argues that high civic literacy societies can be distinguished from low civic literacy societies by policies aimed at the redistribution of resources that are both material as well as non-material, the latter taking the form of a number of measures enhancing access to knowledge, including political knowledge. 12 A number of largely indirect indicators suggest a relationship between the English-speaking mature democracies more laissez-faire policy stance with regard to both material and non-material redistribution, and their apparently relatively low level of civic literacy contrasted with the high civic literacy Northern European/ Scandinavian countries who take a more active approach to material and non-material redistribution. If this analysis is correct, we should see it reflected in the dispersion of political knowledge among the educational categories in the CSES data. Table 4 provides a breakdown of the data by the proportion of correct answers by educational categories. 13 The final column shows how widely the means vary (from 31 percent correct in Israel to 64 percent in the United Kingdom) a clear reflection of the variation in the level of difficulty of the questions. The comparative variation from the overall mean in the other columns indicates differences in political knowledge dispersal by education. One caveat: since the level of difficulty varies not only among the countries, but among the questions within one country as compared to another, we need to be somewhat wary of comparing two countries by their dispersion rates. For example, in the case of two countries with the same mean, but one with three equally difficult questions, and the second with one very easy question and two quite difficult ones, there is a better likelihood that, other things being equal, those with the lowest educational levels will answer at least one question correctly in the latter case. This is where the CSES insistence that the three questions display three specific comparable and varying levels of difficulty is helpful since it should exclude just such extreme differences. Hence, in comparing means of several countries, we can be fairly Scandinavian Political Studies, Vol. 29 No. 4,

11 396 Scandinavian Political Studies, Vol. 29 No. 4, 2006 Table 4. Political Knowledge according to Education in the CSES Countries None to incomplete secondary Secondary or postsecondary University undergraduate* All ANOVA Mean N Mean N Mean N Mean N F** R 2 Australia , Belgium-Flanders , Belgium-Walloon , Canada , Czech Republic , Germany , , Hong Kong , Hungary , Israel , Mexico , , , Netherlands , New Zealand , , , Norway , Poland , , Portugal , , Romania , Spain , , Sweden Switzerland , , Taiwan , Ukraine , United Kingdom , , United States , Total , , , ,382 1, Notes: *Including respondents with an incomplete degree. **Between groups within each country, all significant at the level. Unweighted data The Author(s)

12 confident that dispersion differences reflect country rather than question variation. Differences in the dispersion of political knowledge by levels of educational attainment would appear to confirm the predicted relationship between civic literacy and policies aimed at the redistribution of resources. A quick illustration of that relationship is provided by comparing the differences between the mean political knowledge of the entire sample with that of the lowest educational grouping. From among the CSES countries, we place on one side the relatively low civic literacy English-speaking democracies identified by Milner (United States, United Kingdom, Canada, New Zealand) and, on the other, the high civic-literacy, Northern European/Scandinavian democracies (Sweden, Norway, Germany, the Netherlands). 14 Summing up the figures, we find that the lowest educational group of the Anglo-Saxon countries averages 13.3 points below their countries overall combined average of correct answers compared to 5.8 for the German-Scandinavian nations. 15 In Table 4, we see the results of an analysis of variance of the effects of education on political knowledge that was carried out for each country. The F-values for the mean knowledge levels between the three educational categories are statistically significant in each country, but, as seen in the table, the strength of the association varies. Political knowledge seems to be especially dependent on formal education in countries where income is more unequally distributed, whereas in the European welfare states political knowledge tends to be less tied to education. In Figure 1, a scatter diagram illustrating this pattern is produced. The Gini index for each country is placed on the x-axis, while the y-axis represents the F-value of the ANOVA. The larger the F-value, the more education determines political knowledge in the country. Each country s score on the two variables is visible in the diagram, as well as the classification of its political system. Figure 1 displays a strong linear association (r xy = 0.39): in countries where income is generally more equally distributed, political knowledge is not primarily dependent on formal education, whereas in countries where economic resources are more unequally distributed, education tends to drive political knowledge. Looking at Figure 1, we can see that the countries are scattered along a more or less linear pattern, with Belgium by far the most significant outlier. Indeed, if Belgium is excluded from the above, r xy rises to a high Hence the case of Belgium merits a short digression. The findings here correspond to the picture gained from the limited available data suggesting that Belgium is lower in civic literacy than other Northern European countries (Milner 2002, 60 1). Yet its level of redistribution is similar to theirs. The data provided above suggest that this may have something to do with compulsory voting, leaving the less educated less motivated to acquire information, merely casting a vote at elections. Yet their voting in large numbers means Scandinavian Political Studies, Vol. 29 No. 4,

13 Figure 1. Gini Indices Scattered against the F-value (between Groups ANOVA) of the Explanatory Power of Education on Political Knowledge at the Country Level. that their preferences even though less informed can still find their way into redistributive policies resulting in the low Gini. 16 The case of Belgium notwithstanding, the finding that it is in the more economically unequal countries that relative level of education has a greater impact on political knowledge is an important one. It provides a normative democratic basis for more equal income distribution. If we compare Sweden and the United States polar opposites excluding Mexico it is clear that Swedes with a low level of education potentially exercise much greater political influence than their American counterparts. As political knowledge, the presence of which increases the likelihood of political participation, becomes independent of structural constraints, we approach genuine as opposed to merely formal political equality. 398 Scandinavian Political Studies, Vol. 29 No. 4, 2006

14 Political Knowledge, Education and the Electoral System We now turn to the second aspect of the civic literacy approach germane to this article. Students of political participation have long observed that turnout is higher under proportional representation (PR) (see, e.g. Blais & Dobrzynska 1998; Lijphart 1999; Franklin 2004). PR, it is argued, leads to higher levels of turnout because it makes every vote count and because as a consequence it motivates parties to mobilize support even in geographical areas where they are weak. This is part of the explanation, but insufficient attention has been paid to the intervening effect of political knowledge since political mobilization is largely a matter of information. Gordon and Segura (1997) argue that that the electoral system, along with the party system, and legislative institutional structure affect the availability, clarity and usefulness of political information, finding that, among other things, electoral disproportionality exerts a negative influence on political sophistication. In developing this argument, Milner (2002, 84 8) points to the fact that high civic-literacy societies tend to be characterized by certain types of institutions (heretofore referred to simply as PR), a key aspect of which is a system of elections under which different partisan positions are represented proportionally in the legislative institutions and government is characterized by compromises among them, while low civic-literacy ones tend to be characterized by majoritarian electoral and political institutions. How does this claim stand up against the CSES data? The figures in Table 4 showed no statistical difference in political knowledge between the countries using proportional as opposed to majoritarian electoral systems (the breakdown is found in Table 1). Given the variation in the level of difficulty of the questions by country, we can only suggest this to indicate no difference in the actual level of political knowledge. In any case, the relationship of particular interest here is not with mean levels of political knowledge, but with its distribution. A given society s level of civic literacy is reflected less by indicators of average or overall level of political knowledge than in the comparative level of the groups lowest in resources, especially non-material or education-linked resources. From this perspective, one explanation for the higher turnout under PR is its reducing the cost of the political knowledge needed to make an informed vote for those for whom the cost is highest (i.e. those lacking in resources). This runs against the standard interpretation that first-past-the-post (FPTP) electoral systems require less knowledge on the part of the elector since they usually deliver a simple choice between keeping the incumbents in and kicking them out. Milner argues that this understanding in effect misperceives voters as living outside time and space. It is PR systems that enhance the political knowledge of those with marginal educational resources because such systems provide them with a political map that is relatively clearly Scandinavian Political Studies, Vol. 29 No. 4,

15 drawn and stable across time and space on which to plot their political paths. Since parties under PR are not subject to the volatility of FPTP, which blows up their strength when they do well and shrivels it when they do poorly, there is a relative stability to the features of the political map. Furthermore, under FPTP, which turns setbacks into routs, parties are discouraged from risking operating at levels national, regional, and local other than the one at which they are best organized. PR electoral systems thus play an important direct and indirect role in enhancing the cohesion, stability, and consistency of political parties. They make it easier for the potential voter to identify with a political party and to use that identification as a guide through the complexities of issues and actors at various levels of political activity, both over time and over space, that is, from the local up to the regional, national, and intergovernmental. By thus simplifying a complex political reality and enhancing partisanship, PR fosters political participation especially at the lower end of the education and income ladders where information is at a premium. (Milner 2004, 27) If this logic is correct, we should find in the CSES data a higher correlation between the level of education attained and political knowledge in the majoritarian countries than in the PR ones. And there seems to be a tendency toward this assumption: under PR, Spearman s r = 0.26, compared to 0.33 under majoritarian institutions. In Table 5, we examine this relationship in greater detail. Looking at the two systems independently, we can see that they are more similar than dissimilar. Education is the most important variable in both systems, but, as noted, there is a difference. The logit-coefficients indicate that education has a larger impact in majoritarian than in proportional systems. Note Table 5. The Determinants of Knowledge in Proportional and Majoritarian Systems (logistic ordinal regression) Proportional Majoritarian B SE p B SE p Education *** *** Age/ *** *** Male *** *** Married *** *** Children at home *** * Widowed *** ** Divorced *** Voted? *** *** Party identification *** * Voting makes a difference *** Threshold *** *** Threshold *** *** Threshold *** *** N 25,304 12,038 Missing cases (%) Model Chi 2 3,451 *** 26,776 *** 2 Log likelihood 36,303 2,244 Nagelkerke R Notes: *Significant at the 0.05 level, **significant at the 0.01 level, ***significant at the level. 400 Scandinavian Political Studies, Vol. 29 No. 4, 2006

16 however that the standard error is higher in majoritarian systems, which signals a more uniform impact for education in proportional systems. To get a clearer picture of this relationship, given the variety of electoral systems, we need a classification that is not simply dichotomous: PR and non-pr. One such method for doing so that is increasingly used in recent comparative analysis focuses on the proportionality of the outcome of elections (Lijphart 1999). There are several indices that can be used in such an operation. One can use an index for disproportionality like those of Rae (1967, 84 5) and Gallagher (1991, 38 40), or alternatively, the effective number of parties (Laakso & Taagepera 1979). There is, however, a high correlation between the different methods (Anckar 2002; Norris 2004). We have chosen to make use of the latter (i.e. the effective number of parties as a proxy for the level or proportionality of the electoral system) and the relationship is set out in Figure 2. Given the low number of cases, the linear association in Figure 2 is not significant (r xy = 0.10). However, it becomes significant (r xy = 0.49) when Belgium, the clear outlier in Figure 2 as in Figure 1, is excluded. As expected, thus, as outcomes become more proportional to the popular support attained by political parties, political knowledge becomes less dependent on formal education. As in Figure 1, the countries are scattered along a more or less linear pattern, with Belgium showing significantly higher dispersal effects of education there than in countries with, in this case, similar degrees of proportionality. The complex bi-national character of Belgium s consociational political arrangements account for its extremely large number of effective parties and, thus, its outlier position in one dimension, but on the other dimension further investigation on the sources of Belgium s unexpectedly high level of dispersion of political knowledge among educational categories is clearly in order. Returning to Table 5, we should note the effects of party identification, which increases political knowledge significantly, but more in majoritarian than in proportional systems. While initially puzzling, this relationship in fact helps us understand the mechanism at play. It is reasonable to suppose that political knowledge questions asking respondents to distinguish among parties and policies (e.g. Which of these parties is the greatest supporter of governmental regulation of the economy? ) were more likely to be answered correctly by someone with a low level of education who identifies with a particular party than someone who does not. We know further that party identification is higher under PR, the reason being that, by providing a stable political map, PR fosters party identification. It thus makes sense that party identification is more salient for distinguishing the politically knowledgeable among the poorly educated under majoritarian than under proportional systems. Further study of this relationship is certain to prove productive in deepening our understanding of the effect of institutions upon political knowledge and political participation. Scandinavian Political Studies, Vol. 29 No. 4,

17 Figure 2. The Effective Number of Parties Scattered against the Explanatory Power of Education on Political Knowledge. Concluding Remarks The preceding discussion raises important methodological questions about the kind of political knowledge questions posed. Does asking questions that measure voters knowledge of the political actors, parties or persons, and their differences in relation to policies or ideologies, not veer dangerously close to the tautological when explaining the relationship between political knowledge and various forms of political participation? If so, to avoid this, should we perhaps confine ourselves to questions that get at the workings of the institutions of the political system and the rules of the game, as well as international institutions and politics? Were we to do so, moreover, it would 402 Scandinavian Political Studies, Vol. 29 No. 4, 2006

18 have the added benefit of making it easier to establish a battery of political knowledge questions that could be used in cross-national surveys. While the prospect is a tempting one, we are not persuaded that research into political knowledge is best furthered by adopting such an approach. Harder though it would be to design questions usable comparatively that test citizens knowledge of national political actors and policy differences, eliminating this dimension would undercut a basic objective of this research. While it is useful to compare people in different countries capacity to act as world citizens, this is not enough. We need to be able to compare their capacity to act as citizens of their own country since that is where, through elections, they can regularly exercise that citizenship. And for this purpose, knowledge of institutions is insufficient. As we noted at the outset, for democracy to function properly, those entitled to vote need to have sufficient knowledge of political parties and leaders to compare their commitments against their own political preferences, and to know enough about their records to give some meaningful weight to the credibility of their commitments. In so designing our research we are comparing not simply the political knowledge of individuals, but also the capacity of societal institutions to foster that knowledge. This is as it should be: institutions can be changed through political choices. A finding that Swedes are more politically knowledgeable than Americans is in itself not especially useful; but a conclusion that certain Swedish institutions make it easier for Swedes than Americans to cast an informed vote is, potentially, of far greater use, especially for the citizens of nations in the process of designing or redesigning their democratic institutions. NOTES 1. Political scientists have tended to pay insufficient attention to the importance Downs gives to information costs. For example, Green and Shapiro (1994, 47) state Downs voters paradox to be that individuals are asked to sacrifice time and transportation costs on behalf of a public good, the election of a particular candidate or party, but this is to put undue importance on the opportunity costs of the act of voting since, for economists, the cost of getting to the product and services market is normally exogenous. The real paradox, for Downs, is that benefits to individuals from informing themselves are outweighed by the costs they must bear in doing so (Downs 1957, 246). 2. Empirical research confirming this aspect of Downs theory can be found in, e.g. Berelson et al. (1954); Verba and Nie (1972); Verba et al. (1978); Lipset (1981); Blais (2000); Norris (2004). 3. The literature reveals that political knowledge facilitates citizens ability to link their policy views and choices to their evaluations of politicians and parties. Better-informed citizens can more easily identify with the parties and politicians whose policy positions are most consistent with their own views (Delli Carpini & Keeter 1996, 251 8). 4. Indeed, there is a clear link between the two. For example, a recent French survey showed clearly that the less politically informed respondents were far more frequently critical of the entire political system than those who were more informed (Abrial & Denni 2004). 5. Nevertheless, political knowledge questions compare favourably with questions related to political efficacy and trust, which tend to ask people to express an opinion on something that would never have come into their minds, and to which the natural answer is that Scandinavian Political Studies, Vol. 29 No. 4,

19 depends. In the case of trust, for example, when a finding of X percent of respondents in country Y answer yes to the question Do you believe most people can be trusted? leads to an assertion that X is the level of trust in country Y, we have to simply presume that the question has the same meaning in the different cultural contexts. With political knowledge questions with right and wrong answers, the potential distortions can be taken into account in interpreting the answers. 6. A gleaning of the surveys used in the old democracies shows about ten question equivalents that regularly occur. They fit into the categories identified as follows: Questions testing knowledge related to institutions/regime: 1. (Longest) time between general elections; 2. Number of seats in (lower house) of parliament; 3. (Maximum) term of head of government. Questions testing knowledge-related actors and policies: 1. Name of foreign minister; 2. Name of minister of finance; 3. Name of speaker of (lower house) of parliament; 4. Which of these parties is the greatest supporter of governmental regulation of the economy? 7. In Japan only two questions were used. Since the dependent variable is treated as an index of political knowledge and consists of the arithmetic mean of correct answers, Japan is excluded. 8. The Gini index measures the extent to which the distribution of income (or consumption) among individuals or households within a country deviates from a perfectly equal distribution. It varies between 0 and 100, where 0 represents perfect equality and 100 perfect inequality. The data are derived from UNDP (2004). 9. The analyses pertaining to the whole CSES sample were carried out with data weighted by the appropriate weight. See the CSES codebook for details regarding the dataset weight: sample. 10. Their several tests of the treatment of don t know in the measurement of political information are based on the 1988 American National Election Study. The results suggest that people who answer don t know might actually be less informed than people who give a wrong answer. 11. The term knowledge invites the... inference that one person reading five books is the equivalent of five reading one. Hence, despite its being used by many scholars working in this field, it is nevertheless inferior to a term which carries a clear inference that the variable under investigation cannot be compounded or stocked. The concept of literacy serves this purpose extremely well. In normal usage, one is either literate or one is not (Milner 2002, 54). 12. Milner s figures for 15 mature Western democracies indicate a very high correlation between the Gini indicator of income equality and indicators related to non-material redistribution, e.g. reading comprehension levels, especially at the lower levels of education (Milner 2002, 14). 13. For the three countries, Hong Kong, Mexico and Spain, with two election studies in the first wave, the samples have been unified since no considerable differences between the two elections were found in a separate analysis. 14. Belgium and Australia are excluded from Milner s analysis because of compulsory voting. 15. Howe (2004) presents a parallel argument using data from comparable political knowledge questions asked in recent surveys in low civic-literacy Canada and the high civic-literacy Netherlands. Overall levels of political knowledge are similar, as is the distribution by age group, but one difference between them emerges: proportionally fewer Dutch respondents know nothing or next to nothing than is the case among Canadians. 16. Yet at what cost in terms of efficiency? While the Scandinavian Gini coefficients remained stable (and low) during the recession of the early 1990s, Belgium s rose from low to average... during this period. From... Belgium s notoriously high debt levels, we infer that high social-spending policies aimed at reducing income inequality are more fragile in low civic-literacy countries where the high turnout leads governments to raise public debt to pay for the social programs. Unlike the Scandinavian countries, they do not have recourse to the kinds of sophisticated corporatist trade-offs possible in the high civic literacy countries (Milner 2002, 176 7). 404 Scandinavian Political Studies, Vol. 29 No. 4, 2006