B.J.Pol.S. 36, 159-173 Copyright © 2005 Cambridge University Press doi:10.1017/S0007123406000081 Printed in the United Kingdom
Do Campaigns Help Voters Learn? A Cross-National Analysis
KEVIN ARCENEAUX*
Recent empirical studies on American elections suggest that campaigns provide voters with the necessary information to make reasoned voting decisions. Specifically, campaigns help voters learn about the electoral relevance of 'fundamental variables', such as the economy and party stances, that have been consistently shown to predict electoral outcomes. Do these findings generalize beyond the American case? This article uses cross-national survey data in order to subject this thesis to a more comprehensive test. The analysis provides further support for the hypothesis that campaigns 'enlighten' voters as the election draws near. Moreover, the article shows that some voters learn more from campaigns than others. Campaign effects are more pronounced among individuals with low political sophistication and those living in party list systems. Implications for future research are explored, suggesting a ripe research agenda using under-tapped cross-national data.
Studies of campaigns and their effects on voting behaviour have focused primarily on the American case, producing a rich set of theories and findings. Written off fifty years ago as having a minimal impact on voting behaviour, the study of campaign effects enjoyed a renaissance in the 1990s. These recent studies have developed a more nuanced view of campaign effects, eschewing early scholars' expectation that campaign messages manipulate easily persuadable voters. Increasingly, campaigns are viewed as playing an informational role, helping voters make a decision that is in line with their pre-existing preferences.1
While this new view of campaign effects is plausible, does it generalize beyond the American case, or is this another example of 'American exceptionalism' ? To answer this question, we need variation in context. Cross-national data are well suited for this purpose, providing the means to compare campaign influence across countries that have different electoral experiences. Stevenson and Vavreck utilize cross-national data to test empirically the prevailing wisdom in the American literature that campaigns are primarily informational tools used by voters.2 Consistent with American studies, they find that election outcomes are more likely to reflect economic performance in countries that have long campaigns, suggesting that campaigns help voters learn about the political relevance of the economy. These findings are compelling but, because they are based on aggregate data, they do not directly address the microfoundations underlying the American campaign effects model.
* Department of Political Science, Temple University, Philadelphia. An earlier version of this paper was presented at the 2001 meeting of the Society for Political Methodology, Emory University, Atlanta, Ga. I owe gratitude to Martin Johnson, Ken Scheve, Randy Stevenson and the anonymous referees for their helpful suggestions. I also thank Ray Duch and Randy Stevenson for sharing their cleaned and pooled version of Eurobarometer data with me. The customary caveat regarding my responsibility for errors applies.
1 Shanto Iyengar and Adam F. Simon, 'New Perspective and Evidence on Political Communication and Campaign Effects', Annual Review of Pscyhology, 51 (2000), 149-69.
2 Randolph T. Stevenson and Lynn Vavreck, 'Does Campaign Length Matter? Testing for Cross-National Effects', British Journal of Political Science, 30 (2000), 217-35.
160 arceneaux
Relying on individual-level survey data across twelve years in nine European countries, this article provides a direct test of the thesis that campaigns play an informational role in voting decisions. The article offers two contributions to cross-national campaign literature. First, it submits the micro-level assumptions underlying Stevenson and Vavreck's macro-level model to an empirical test, which by extension also subjects the micro-level model that underlies the dominant theory in the American campaign effects literature to the broadest empirical test to date.3 Secondly, the article goes one step further, using the individual-level data to explore whether individual and institutional characteristics moderate campaign effects. The findings suggest that campaigns do in fact educate citizens, helping them make more 'enlightened' decisions, and that political sophistication and electoral rules affect the considerations voters incorporate in their voting preferences. In short, some voters glean more from campaigns than others.
the role of campaigns: enlightenment rather than persuasion
Scholars in the 1940s and 1950s concluded that political campaigns had minimal effects, if any at all, on voting behaviour.4 Berelson et al. use panel survey data to show that voters make up their minds well before the campaign even begins.5 Recent research in political science has challenged this 'minimal effects' school of thought. A number of impressive studies have uncovered the important role campaigns play in US elections. While voters' pre-existing attitudes and assessment of things beyond the control of political campaigns, like the economy, have the strongest impact on voting decisions, campaigns play a major role in producing these effects. Finkel finds that when citizens change their minds during the course of a campaign it is generally in the direction of their pre-existing attitudes.6 In a laboratory experiment, Ansolabehere and Iyengar show that subjects exposed to thirty-second issue-oriented campaign advertisements were more likely to evaluate candidates on the basis of those issues.7 Campaign messages prime voters to rely on particular issues, events and candidate characteristics to the exclusion of others by making them more accessible in the minds of voters.8
Rather than persuading voters to change their minds, as earlier scholars expected, campaigns help voters make up their minds. By increasing the amount of relevant political information available to voters, campaigns help citizens cast votes in line with their
3 Stevenson and Vavreck,'Does Campaign Length Matter?'
4 Joseph T. Klapper, The Effects of Mass Communication (Glencoe, 111.: Free Press, 1960).
5 Bernard R. Berelson, Paul F. Lazarsfeld and William N. McPhee, Voting: A Study of Opinion Formation in a Presidential Campaign (Chicago: University of Chicago Press, 1954).
6 Steven E. Finkel, 'Re-examining the "Minimal Effects" Model in Recent Presidential Campaigns', Journal of Politics, 55 (1993), 1-21. See also Andrew Gelman and Gary King, 'Why are American Presidential Election Campaign Polls So Variable When Votes are So Predictable?' British Journal of Political Science, 23 (1993), 409-51; Thomas M. Holbrook, Do Campaigns Matter? (Thousand Oaks, Calif.: Sage, 1996); Thomas M. Holbrook, 'Campaigns, National Conditions, and U.S. Presidential Elections', American Journal of Political Science, 38 (1994), 25-46.
7 Stephen Ansolabehere and Shanto Iyengar, Going Negative: How Attack Ads Shrink and Polarize the Electorate (New York: Free Press, 1995). See also, Craig L. Brians and Martin P. Wattenberg, 'Campaign Issue Knowledge and Salience: Comparing Reception from TV Commercials, TV News, and Newspapers', American Journal of Political Science, 40 (1996), 172-93.
8 Larry M. B artels, 'Messages Received: The Political Impact of Media Exposure', American Political Science Review, 87 (1993), 267-85; John Zaller, 'The Myth of Massive Media Impact Revived: New Support for a Discredited Idea', in Diana C. Mutz, Paul M. Sniderman and Richard A. Brody, eds, Political Persuasion and Attitude Change (Ann Arbor: University of Michigan Press, 1996).
Do Campaigns Help Voters Learn? 161
pre-existing attitudes and proclivities. Gelman and King produce an in-depth exploration of this phenomenon in presidential elections.9 They note a striking puzzle in American politics: presidential election outcomes are highly predictable on the basis of data known months before the election, while trial-heat, pre-election polls are highly variable until the election draws near. For instance, in the 1988 US presidential election, George Bush was the winner predicted by forecasters in July, while pre-election polls showed his challenger, Michael Dukakis (who would later lose to him), leading by 19 percentage points.
Gelman and King point to the presidential campaign as an explanation of this puzzle. Forecasting models rely on a few 'fundamental variables', such as economic performance and presidential popularity, to make predictions. The disconnect between trial-heat polls and aggregate forecasting models suggests that voters are not using information about these fundamental variables when answering early pre-election poll vote intention questions. The reason voters do not use this information, Gelman and King assert, is because they do not yet have accurate knowledge about these fundamental variables.
By increasing the information available to voters, campaigns help them update their beliefs regarding fundamental variables accurately and accord them weight in their voting decision. Months before the election, the average voter may have little sense of how fundamental variables relate to her vote preference. Over the course of a campaign, voters learn more about fundamental variables and, more importantly, their political relevance. For instance, a Democrat may not connect the declining economy to her vote preference in July, but by November she is reminded by the campaign that the Republican incumbent is to blame for high unemployment, affecting her vote decision. In July, her vote preference may not have reflected her attitude about the economy, but by November it is evident. For the remainder of the article, I will refer to this as the 'enlightenment hypothesis'.
cross-national study of campaigns
Thus far most of the research on campaign effects has been done with data from US elections, but there is an active research agenda studying campaigns in a cross-national context.10 Ultimately, cross-national research is necessary to gain the needed variation in real-world settings that will provide greater insights into the contours of campaign effects. While Gelman and King offer a plausible hypothesis regarding the informational role
9 Gelman and King, 'Why are American Presidential Election Campaign Polls So Variable When Votes are So Predictable?'; Christopher Wlezien and Robert S. Erikson, 'The Timeline of Presidential Election Campaigns', Journal of Politics, 64 (2002), 969-93.
10 For excellent examples, see Susan A. Banducci and Jeffrey A. Karp, 'How Elections Change the Way Citizens View the Political System: Campaigns, Media Effects, and Electoral Outcomes in Comparative Perspective', British Journal of Political Science, 33 (2003), 443-68; Shaun Bowler and David M. Farrell, Conclusion: The Contemporary Election Campaign (New York: St Martin's, 1992); David Butler and Austin Ranney, Electioneering: A Comparative Study of Continuity and Change (Oxford: Clarendon Press, 1992); David M. Farrell and Martin Wortmann, 'Party Strategies in the Electoral Market: Political Marketing in West Germany, Britian, and Ireland', European Journal of Political Research, 15 (1987), 297-318; Rachel Gibson, Michael Margolis, David Resnick and Stephen J. Ward, 'Election Campaigning on the WWW in the USA and UK: A Comparative Analysis', Party Politics, 9 (2003), 45-75; Donald Granberg and Soren Holemberg, 'Election Campaign Volatility in Sweden and the United States', Electoral Studies, 10 (1991), 208-30; R. J. Johnston and C. J. Pattie, 'Campaigning and Split-Ticket Voting in New Electoral Systems: The First MMP Election in New Zealand, Scotland, and Wales', Electoral Studies, 21 (2002), 583-600; M. Kreuzer, 'Electoral Mechanisms and Electioneering Incentives: Vote-Getting Strategies of Japanese, French, British, German, and Austrian Conservatives', Party Politics, 6 (2000), 487-804; Stevenson and Vavreck,'Does Campaign Length Matter?'
162 arceneaux
campaigns play, it may not generalize beyond the confines of an American presidential election.
Stevenson and Vavreck take an initial step subjecting the enlightenment hypothesis to cross-national empirical investigation using aggregated party vote shares from 113 elections across thirteen countries. In particular, they find that one type of fundamental variable - economic performance - has a stronger impact on party vote shares the longer the campaign, suggesting that longer campaign length gives voters more time to absorb campaign information and update their beliefs and weights concerning fundamental variables with greater precision. While these results are consistent with the enlightenment hypothesis, it does not directly test the microfoundations underlying Gelman and King's thesis. Are voters more likely to connect fundamental variables to their voting preferences as election day draws near? Ultimately, individual-level data are required to test this hypothesis directly.
In addition, it is likely that campaign effects are heterogenous across individuals and countries. Some voters are more likely to be aware of the fundamental variables about which campaigns aim to educate voters, and some political institutions may be more conducive to campaign 'enlightenment' than others. In particular, attention is paid to whether campaign effects vary across levels of political sophistication, institutional clarity of responsibility and types of electoral institutions. There are strong theoretical reasons in the existing voting literature to expect that these factors will moderate campaign effects. Each is discussed in turn.
Political sophistication reflects the degree to which individuals consume political information, creating gaps in campaign reception across individuals. Zaller contends that in competitive national elections (as studied here), campaign effects should be strongest among the least sophisticated.11 Campaign messages abound in these elections, trickling down to even the least attentive. Because people with low levels of political sophistication hold small stores of political knowledge, they are more likely to accept the political messages they encounter. Consequently, they should be more likely to learn from campaigns than individuals with higher levels of political sophistication.12
Powell and Whitten contend that political institutions affect the clarity with which individuals are able to attribute responsibility to the government for economic performance.13 Countries with single-party control of the government, for instance, are more likely to attract credit and blame for the economy than countries that feature coalition or minority-party governments. Anderson corroborates Powell and Whitten's clarity of attribution hypothesis with individual-level data, showing that individuals in countries with institutions that facilitate responsibility attribution are more likely to link their economic perceptions to their voting preferences.14 Yet Anderson does not entertain the role campaigns might play. Campaigns in countries where responsibility for the economy is more easily attributable to the government may more easily educate voters about the connection between the economy and their vote preference as the election draws near.
11 John Zaller, The Nature and Origin of Mass Public Opinion (Cambridge: Cambridge University Press, 1992); Zaller, 'The Myth of Massive Media Impact Revived'.
12 Pippa Norris and David Sanders, 'Message or Medium? Campaign Learning During the 2001 British General Election', Political Communication, 20 (2003), 233-62.
13 G. Bingham Powell Jr and Guy D. Whitten, 'A Cross-National Analysis of Economic Voting: Taking Account of the Political Context', American Journal of Political Science, 37 (1993), 391-414.
14 Christopher J. Anderson, 'Economic Voting and Political Context: A Comparative Perspective', Electoral Studies, 19(2000), 151-70.
Do Campaigns Help Voters Learn? 163
Electoral institutions may also condition the impact of campaigns. Countries in which voters cast a vote for a party list rather than a candidate (or some combination of party list and candidates) should be more conducive to campaign effects in terms of party-related fundamental variables (e.g., economy and ideology). The party-centric voting model as articulated by Downs and Key characterizes the voting decision as a comparison of party platforms (ideology) and performance (e.g., economy).15 Party-list systems place the party at the centre of the voting decision by specifically and unambiguously asking individuals to choose a party rather than a candidate. Candidate-based electoral institutions muddy the voting calculus by allowing characteristics of the candidates - rather than their parties -to become more prominent considerations. Consequently, campaigns in party-list systems may aid voters in learning more about the fundamental variables most important to the judgement of parties rather than candidates.
research design
Defining Theoretical Variables
The enlightenment hypothesis invokes two theoretical concepts in need of empirical definition: fundamental variables and election proximity. Gelman and King draw empirical measures of fundamental variables from the US presidential forecasting literature.16 These include the usual suspects that have been consistently found to be important in US voting behaviour, especially attitudes about the economy and partisan attachment. To be sure, attitudes about the economy are also of fundamental importance in European elections and will be included in this article's analysis as well.17 Partisan attachment is less of a factor in European elections, though. Ideology, rather, plays a more important role in voting decisions and will be included in the analysis in lieu of party identification.18
Stevenson and Vavreck define election proximity as a dichotomous threshold, noting that Gelman and King show learning begins to increase rapidly about six weeks before the election.19 The threshold in their article demarcates the length of campaign into 'long' and 'short', premised on the assumption that individuals learn more information in a long campaign. In this article, election proximity is measured as the distance of the survey from election day. While different from Stevenson and Vavreck's measure, it stands as an indicator for the same concept: the amount of information citizens incorporate into their voting decision. Gelman and King specifically show that individuals appear to have less information the further away the election. Thus, the election proximity measure used here takes advantage of the individual-level data and more closely mirrors Gelman and King's measure.
15 Anthony Downs, An Economic Theory of Democracy (New York: Basic Books, 1957); V. O. Key Jr, The Responsible Electorate (Cambridge, Mass.: Harvard University Press, 1966).
16 Gelman and King, 'Why are American Presidential Election Campaign Polls So Variable?'
17 Stevenson and Vavreck,'Does Campaign Length Matter?'
18 Christopher J. Fleury and Michael S. Lewis-Beck, 'Anchoring the French Voter: Ideology versus Party', Journal of Politics, 55 (1993), 1100-9; Ronald Inglehart and Hans D. Klingemann, 'Party Identification, Ideological Preference, and the Left-Right Dimension among Western Mass Publics', in Ian Budge, Ivor Crewe and Dennis Farlie, eds., Party Identification and Beyond: Representations of Voting and Party Competition (London: Wiley, 1976), pp. 243-73.
19 Stevenson and Vavreck,'Does Campaign Length Matter?', base the six-week threshold on Figures 7A through 7C in Gelman and King, 'Why are American Presidential Election Campaign Polls So Variable When Votes are So Predictable?', p. 442.
164 arceneaux
This is not to say that the length of campaign is unimportant. If citizens only have a week in which to glean campaign information, they are likely to know less on election day compared to those who had a full six weeks. It turns out that none of the countries with surveys conducted near an election had very long campaign lengths. The elections were unscheduled and the campaigns lasted near the six-week threshold. However, it is possible to use these individual-level data to test the mechanism that underlies Stevenson and Vavreck's aggregate-level findings. If voters learn more during long campaigns than during short ones, their knowledge should increase as the campaign continues. This implies that voters should know more, say, three weeks from the election than they do six weeks from the election.
Data and Model
This article uses individual-level data collected from Eurobarometer surveys to test the enlightenment hypothesis. The benefit of using the Eurobarometer is that for nine countries it has asked consistently worded questions regarding the two important fundamental variables investigated in this article, sociotropic economic evaluations and ideology.20 Moreover, since the Eurobarometer is administered contemporaneously in these countries, it provides needed variation in election proximity. The following model was used to test the enlightenment hypothesis (see Appendix for description of variable coding):
Vltj =/(#> + PiEt + p2Ii + PsTj + p4(Et x Tj) + p5(Ii x Tj) + 8C + a), (1)
where V7,y = vote intention (1= prefer incumbent party, 0 = otherwise) for each respondent i on each Eurobarometer j';21 Et = sociotropic economic evaluation ( — 1 = nation's economy worsened, 0 = stayed the same, + 1 = improved); = self-reported ideology (ranges from 1 to 10, 1 being the furthest from the incumbent party(ies) and 10 being the closest);22 7} = election proximity threshold (1 = six weeks or less before the election, 0 = six weeks or more before the election); C = matrix of controls for age, gender, education, and country and year fixed effects;23 8 = vector of slope coefficients associated with C;/= logistic function.
empirical findings
The results from Equation 1 are displayed in Table 1. Column 1 displays the basic model. The slope coefficients for Economy and Ideology represent the relationship between those
20 Only Eurobarometers since 1982 that had an economic evaluation question were used: 1982 (spring), 1982 (fall), 1983 (spring), 1983 (fall), 1984, 1985, 1986, 1987, 1988, 1991 and 1994. Belgium, Denmark, France, Germany, Ireland, Italy, Luxembourg, the Netherlands and the United Kingdom were surveyed.
21 The vote intention question used on the Eurobarometer is: 'If a general election were held tomorrow, which party would you vote for?'
22 The ideology of the incumbent party was dichotomously assigned on the basis of the party family to which the government party(ies) belonged. Thus, governments were either considered primarily leftist or primarily rightist. In the instance of grand coalitions where both leftist and rightist parties formed the government, the party of the prime minister was used as an indicator of incumbent party ideology (this was a frequent issue in Belgium and Italy).
23 The model was also estimated with country dummy variables interacted with the covariates, allowing these variables to have different effects across countries. The coefficients of interest and their standard errors were completely unaffected by this specification. Given the inclusion of nation and year fixed effects in the model, the introduction of country dummies-covariate interactions results in little reduction of unexplained variance.
Do Campaigns Help Voters Learn? 165
table 1       The Conditional Effect of Election Proximity on the Relationship
Between Fundamental Variables and Vote Intention, Eurobarometers 1982-94
s.e.
Fundamental Variables Sociotropic Evaluation
Ideology
Control Variables Age
Female
Education
Income
Conditional Variables
Election Proximity (six weeks)
Election Proximity x Sociotropic Evaluation
Election Proximity x Ideology
Constant
Number of observations
x2
Pseudo-/?2 %Modal category % Correctly predicted % Reduction in error
042*** 0.38***
0.01*** 0.09*** -0.00 0.14***
— 0 72***
0.09*
0ii***
0.56
58,295
10,771.71*** 0.14 67.3 70.6 10.0
0.01 0.01
0.001 0.02 0.00 0.01
0.18 0.05 0.03 0.90
Note: Dependent variable = Vote Intention (1 = prefer incumbent party). *p<0.10 **p < 0.05 ***p < 0.01 for two-tailed z-test; baseline category = more than six weeks. Country and year fixed effects were also included, but are not shown here to conserve space.
variables and vote intention when the Eurobarometer was administered more than six weeks away from the election, and the interaction terms represent the relationship between these variables and vote intention when the election was six weeks or less before from the survey administration. As expected, individuals are more likely to factor their assessment of the economy and their ideology into their voting decision when the election will be held in six weeks or less.24 Yet these effects are substantively small. The remainder of the analysis will demonstrate that these effects are conditional on particular individual and institutional characteristics. In short, campaigns do more to enlighten some voters more than others.
The basic model outlined in Equation 1, was re-estimated with the relevant three-way interactions to accomplish this task. The results from these analyses are presented in Table 2. Column 1 tests the hypothesis that campaign effects should be strongest among
A dummy variable indicating whether an election had occurred a month or less before the survey was included to control for possible halo effects (see Lonna Rae Atkeson, ' "Sure, I Voted for the Winner!" Overreport of the Primary Vote for the Party Nominee in the National Election Studies', Political Behavior, 21 (1999), 197-215). The results were not substantively affected in any of the analyses.
166 arceneaux
table 2      The Moderating Influence of Individual and Institutional Characteristics on Campaign Effects, Eurobarometers 1982—94
Individual or institutional characteristic
Political Institutional Party-List
Sophistication, b      Clarity, b System, b
Fundamental variables
Sociotropic Evaluation	0.31***	0.53***	0.46***
	(0.03)	(0.02)	(0.02)
Ideology	037***	0.38***	0.52***
	(0D1)	(0.01)	(0.01)
Control variables			
Age	0.014***	0.01***	0.01***
	(0.001)	(0.001)	(0.001)
Female	0.09***	0.09***	0.10***
	(0.02)	(0.02)	(0.02)
Education	-0.00	-0.00	0.00
	(0.00)	(0.00)	(0.00)
Income	0.14***	014***	o14***
	(0.01)	(0.01)	(0.01)
Conditional variables			
Election Proximity	— 2 19***	— 1 13***	I 43***
(six weeks)	(0.46)	(0.29)	(0.34)
Election Proximity x	0.12	-0.37	-0.10
Sociotropic Evaluation	(0.15)	(0.23)	(0.16)
Election Proximity x	0.31***	011***	- 0.28***
Ideology	(0.07)	(0.03)	(0.05)
Characteristic	-0.05	_ 1 ^7***	-0.34
(see column heading)	(0.03)	(0^41)	(1.36)
Characteristic x	0.64***	0.16*	— 3 90***
Election Proximity	((ü8)	(0.09)	(0.40)
Characteristic x	0.05***	-0.06***	— 011***
Sociotropic Evaluation	(0.01)	(0.01)	(0.025)
Characteristic x	0.01	NA	— 0 34***
Ideology	(0.01)		(0.01)
Characteristic x	-0.01	0.20**	0.26
Election Proximity x	(0.06)	-0.09	(0.18)
Sociotropic Evaluation			
Characteristic x	— 0.09***	NA	0.67***
Election Proximity x Ideology	(0.02)		(0.06)
Constant	0.63	0.86	-0.02
	(0.90)	(0.94)	(0.93)
Number of observations	58,295	58,295	58,295
x2	10,744.3***	10,756.4***	11,816.3***
Pseudo-/?2	0.14	0.14	0.15
% Modal category	67.3	67.3	67.3
% Correctly predicted	70.5	70.6	70.8
% Reduction in error	9.9	10	10.7
Note: Dependent variable = Vote Intention (1 = prefer incumbent party). *p<0.10 **p< 0.05 * * *p < 0.01 for two-tailed z-test; baseline category = more than six weeks. Country and year fixed effects were also included, but are not shown here to conserve space. Standard errors are shown in parentheses below b. NA = not available.
Do Campaigns Help Voters Learn? 167
those with lower levels of political sophistication. In the American public opinion literature, Zaller's measure of political sophistication, which asks survey respondents to answer factual questions about politics, is most commonly used.25 Since the Eurobarometer surveys used in this analysis did not include such items, Inglehart's measure of cognitive mobilization is employed instead. Cognitive mobilization is closely related to political sophistication.26 Individuals who exhibit high levels of cognitive resources are better able to assimilate and act upon political information than those with low levels of cognitive resources. In fact, the cognitive mobilization index in the Eurobarometer relies on indicators that are often used as proxy measures for political sophistication.27
Political Sophistication and Campaign Effects
It does not appear that political sophistication moderates the degree to which learning about the economy occurs during campaigns.28 Across all levels of political sophistication, individuals are more likely to link the economy to their voting preference when the election is close. In contrast, political sophistication does moderate campaign effects in terms of ideology in expected ways. As Figure 1 illustrates, the link between ideology and vote preferences is stronger when the election is six weeks or less away than it is when it is more than six weeks away among individuals with low to medium levels of political sophistication. Conversely, this relationship is attenuated among individuals with high levels of political sophistication when the election is six weeks or less away.
Fitting with Zaller's discussion of media attentiveness, political sophisticates are more likely to pick up on more subtle campaign cues, such as complex issue positions.29 This attention to more specific issues may drive out ideology as a primary deciding factor. Less politically sophisticated individuals use campaigns to learn basic features of the electoral landscape, such as defining which party matches them ideologically, whereas political sophisticates enter the campaign with a sense of the parties' ideologies relative to their own, and use campaigns to glean other types of electorally relevant considerations.
Clarity of Responsibility, Party Lists and Campaign Effects
Political institutions also play a role in shaping campaign effects. Countries in which multiple veto players are involved in policy making mute the degree to which responsibility can be clearly linked to the incumbent party, which is why these countries exhibit lower levels of economic voting.30 Consistent with the previous literature, the
25 Zaller, The Nature and Origin of Mass Public Opinion.
26 Ronald Inglehart, The Silent Revolution: Changing Values and Political Styles Among Western Publics (Princeton, N.J.: Princeton University Press, 1977).
27 See Vincent Price and John Zaller, 'Who Gets the News? Alternative Measures of News Reception and Their Implications for Research', Public Opinion Quarterly, 57 (1993), 133-64. The cognitive mobilization index developed by Inglehart includes survey items that measure frequency of political discussion and attempts to persuade other to share the respondent's political views. Price and Zaller show that while discussion measures do tap political sophistication, they typically perform poorer than factual questions at predicting 'information gaps'. Consequently, the cognitive mobilization index has a built-in conservative bias, which means evidence for the conditional impact of political sophistication on campaign effects will at worst be muted.
28 Given the possible conservative bias in the cognitive mobilization index used to measure political sophistication (see fn. 6), this null finding may reflect measurement error.
29 Zaller, The Nature and Origin of Mass Public Opinion.
30 Anderson, 'Economic Voting and Political Context'; Powell and Whitten, 'A Cross-National Analysis of Economic Voting'.
168
ARCENEAUX
Low Medium Low
Economic evaluation
- More than 6 weeks before the election
----Less than 6 weeks before the election
Fig. I. The conditional impact of political sophistication on campaign effects with respect to ideology
results in column 2 in Table 2 demonstrate that as institutional clarity of responsibility (the measure Powell and Whitten develop and report) diminishes, individuals are less likely to link the economy to their vote preference.31
However, there is no support for the hypothesis that campaigns in high institutional clarity countries are more likely to educate voters about the link between the economy and their voting preference. In fact, the model estimates suggest the contrary. As the election draws near, individuals in low institutional clarity countries are more likely to link economic attitudes to their vote preference than individuals in high institutional clarity countries. The relationship between economic attitudes and vote preference in countries with low institutional clarity becomes significantly stronger when the election is six weeks or less away, whereas the relationship is nearly flat when the election is more than six weeks away. To be sure, the relationship between economic attitudes and vote preference is stronger in high institutional clarity countries, but the campaign appears to produce little additional learning in these countries, while appearing to do so in low institutional clarity countries.
Perhaps campaigns in low institutional clarity countries focus more attention on educating voters about the possible linkage between government policy and economic
31 The institutional clarity variable is included as an additive variable (in addition to the interaction term), because not doing so would constrain the constant for the economic evaluation slopes to be the same across levels of institutional clarity. There is no reason to impose this assumption a priori. Moreover, if the assumption is false, it may result in a biased coefficient estimate for the effect of economic evaluations on vote choice. Consequently, as a general rule, all the components of an interaction term should also be included in the model as additive variables (see Leona S. Aiken and Stephen G. West, Multiple Regression: Testing and Interpreting Interactions (Newbury Park, Calif.: Sage, 1991)).
Do Campaigns Help Voters Learn? 169
performance than campaigns in high institutional countries. Such a difference in campaign tactics might reflect the decisions of strategic politicians or party organizations trying to offset the institutional features that obfuscate that attribution of responsibility for the economy to the government. Of course, this scenario is merely speculation. Future research will be required to explore this unexpected finding.
Column 3 in Table 2 shows the impact electoral institutions have on campaign effects. This study focuses on one major electoral institution: the use of party lists. Voting procedures vary greatly across countries. Some countries ask voters to choose a party, which allocates seats to candidates on a list (for example, Belgium). In other countries, voters choose a candidate affiliated with a party (for example, France). Still others fall in between (such as Germany). Countries that exclusively use lists are of interest here.32
Consistent with expectations, campaign effects are far more powerful in party-list countries in terms of the party-centric fundamental variables explored in this article. Individuals in non-party list systems who have a positive perception of the economy are 13.4 per cent more likely to prefer the incumbent party compared to individuals who have a negative perception of the economy. In contrast, individuals living in party-list systems who positively rate the economy are 27.2 per cent more likely (twice as likely) to prefer the incumbent relative to those who negatively rate the economy (see Figure 2a). However, these findings are merely suggestive because, as the confidence intervals show, these quantities are not statistically different from one another.
The effects for ideology are more stunning (see Figure 2b). As the election approaches ideology weighs much more heavily in the voting decisions of citizens in party-list systems. At the extremes, individuals who are closest to the incumbent party's ideology {Ideology =10) are only 11.6 per cent more likely to prefer the incumbent party than those who are furthest from the incumbent party's ideology {Ideology = 0) in non-party list systems.33 In party-list systems, those ideologically closest to the incumbent party are 64.9 per cent more likely to prefer the incumbent party than those who are ideologically most distant from the incumbent party.34 Moreover, in countries that do not use the party list, ideology appears to be less reflected in vote preference as the election draws near. This finding may be the result of candidate-based considerations becoming more prominent in the voting decision as the election campaign warms up and focuses on candidate personalities.
Does Learning Increase During the Campaign?
Another implication of Gelman and King's micro-level is that learning should increase during the campaign as the election draws closer. Stevenson and Vavreck test this implication by focusing on campaign length - voters are able to learn more in longer
Closed, flexible and open-list systems were defined as 'party-list systems'. See Gary W. Cox, Making Votes Count: Strategic Coordination in the World's Electoral Systems (Cambridge: Cambridge University Press, 1997), pp. 45-7.
33 In the case of coalition governments, it would be more accurate to say that individuals who are closest to the ideology of the dominant coalition partner are more likely to prefer one of the parties in the coalition. The singular term, party, is used here in the interest of expository simplification.
34 This comparison of extremes is reported to demonstrate a point: in non-party list systems, even extreme changes in ideology have small affects on vote propensities.
170
ARCENEAUX
Not Party-List System
Party-List System
0.6-
5=* 0.4-c
CD
E
I 0.2-
CD
fe o-l
-0.2-
-1
-1— r~
1 -1
Economic Evaluation
More than 6 weeks before the election
----Less than 6 weeks before the election
(a) Sociotriopic economic evaluations
Not Party-List System
Party-List System
~~ More than 6 weeks before the election ---Less than 6 weeks before the election
(b) Ideology
Fig. 2. The conditional impact of party-list electoral systems on campaign effects
campaigns. As discussed earlier, there is less variance in the campaign length of the countries studied here, but the availability of individual-level data allow us to test the mechanism the produces the aggregate finding Stevenson and Vavreck observe: voters' learning increases throughout the campaign. So, if voters begin paying attention to a campaign six weeks before it is held, they should possess more political information as
Do Campaigns Help Voters Learn?
Ill
table 3       The Relationship Between Fundamental Variables and Vote Intention When the Election is Three Weeks Away, Eurobarometers 1982—94
s.e.
Fundamental Variables Sociotropic Evaluation
Ideology
Control Variables Age
Female
Education
Income
Conditional Variables
Election Proximity (three weeks)
Election Proximity x Sociotropic Evaluation
Election Proximity x Ideology
Constant
Number of observations
X2
Pseudo-/?2 %Modal category % Correctly predicted % Reduction in error
042*** 0.38***
0.01*** 0.09***
-0.00
q^4###
— \ iq***
0.21**
0.22***
0.55
58,295 10,740.61*** 0.14 67.3 70.6 10.0
0.01 0.01
0.001 0.02 0.00 0.01
0.25 0.08 0.04 0.90
Note: Dependent variable = Vote Intention (1 = prefer incumbent party). *p<0.10 **p<0.05 ***p< 0.01 for two-tailed z-test; baseline category = more than three weeks. Country and year fixed effects were also included, but are not shown here to conserve space.
each week passes. To test this, analysis is restricted to those countries with elections three weeks or less away, which is half as close as the six-week threshold.35
Table 3 shows that campaign effects magnify when the election is three weeks away.36 Individuals who think the economy has improved are 25.3 per cent more likely to prefer the incumbent than those who believed it has worsened when the election is less than three weeks away. The same type of individuals were 20.5 per cent more likely to prefer the incumbent when the election is less than six weeks away, netting nearly a 5 per cent gain as the election moves three weeks closer. Furthermore, when the election is three weeks away, individuals who are ideologically close to the incumbent party (Ideology = 7.5) are 50.6 per cent more likely to prefer the incumbent party than those who are ideologically further away from the incumbent party (Ideology = 3.5).37 Similarly situated individuals are only 30.6 per cent more likely to prefer the incumbent party when the election is six weeks away. These findings suggest a dynamic learning process leading up to the election.
The same substantive conclusions are reached if a continuous variable indicating the number of weeks prior to the election within the six-week time frame is used instead of the three-week threshold dummy variable.
36 Interactions with individual and institutional characteristics were also analysed using the three-week threshold, and similar results were obtained. It was not possible to assess the impact of party-list systems due to perfect colinearity.
37 These numbers reflect one standard deviation below and above the mean of Ideology (5.5).
172 arceneaux
Both the availability of information and voter attentiveness is likely to increase during the final stretch of the election, until it reaches its height on election day.
conclusion
This study demonstrates that political campaigns play an important role in helping citizens to make an enlightened voting decision on election day. The cross-national survey data employed here provide the broadest test of the enlightenment hypothesis to date, generating further evidence that campaigns have more than a minimal effect on voting behaviour. All things being equal, as the election approaches, fundamental variables are given more weight in the voting decision. Consistent with Gelman and King's findings, the upswing in voter learning occurs six weeks out from the election and increases monotonically. These findings also corroborate the micro-level assumptions in Stevenson and Vavreck's study, showing that citizens increasingly incorporate the economy in their voting preferences as the election draws near. Interestingly, though, campaign effects appear to be more powerful in terms of ideology. Perhaps this reflects the ongoing attention individuals give to economic matters in their everyday lives. Thus, campaigns can only marginally move citizens' weight of economic attitudes in their voting decision. Future research should be aimed towards unravelling this puzzle.
Moreover, campaign effects are conditioned by both individual and institutional characteristics. Campaigns appear to generate learning about ideology among those with the least information, individuals with low levels of political sophistication. This is consistent with studies of information processing in political science, especially Zaller's. Individuals with high levels of political sophistication are more likely than those with lower levels of sophistication to be already aware of the parties' ideological stances before the campaign begins. Interestingly, there are no differences in terms of the economy. Perhaps, the pervasiveness of the economy in political considerations reaches across the information gap-
The most powerful conditional effect was whether a country used party-list electoral rules. The magnitude of this conditional effect makes sense in light of the fact that the fundamental variables studied in this article (economic performance and ideology) are important considerations in party-based voting. Since party-list systems specifically induce voters to focus on parties (rather than candidates), these variables should be given more weight by citizens in these countries. Intriguingly, these data suggest that these fundamental variables are less reflected in voting preferences in non-party list countries as the election approaches. It is possible that campaigns in these countries focus citizens' attention to more candidate-centric considerations.
There was also an unexpected finding, which should be explored in future research. Institutional clarity of responsibility was found to moderate campaign effects regarding economic attitudes, but in the unexpected direction. Voters appear to become more 'enlightened' in countries that have less institutional clarity. Individuals in these countries may be responding to the strategic decisions of candidates trying to compensate for institutional factors. With responsibility for economic outcomes unclear, voters have more to learn over the course of the campaign.
All together, these findings provide implications for a lively research agenda. Further attention should be directed to the role individual and institutional characteristics play in moderating campaign effects. An eye should be turned to going beyond this article, exploring the psychological mechanisms that underlie these relationships.
Do Campaigns Help Voters Learn? 173
appendix Variable Coding Rules
Vote Intention: 0 = vote for a non-incumbent party, 1 = vote for an incumbent party Ideology: 1 = extreme opposite ideology (left/right) of incumbent party(ies) to 10 = identical ideology of incumbent party(ies)
Sociotropic Evaluation: — 1 = national economy worsened over past year, 0 = stayed the same, +1 = improved
Education: Age at which R finished education, 1 = less than 14, 2 = 15, ... , 9 = 22, 10 = still studying
Income: 1 = lowest quartile, ... , 4 = highest quartile
Age: Age in years
Female: 0 = male, 1 = female
Election Proximity: 0 = survey administered more than six weeks before an election, 1 = survey
administered six weeks or less before the election Political Sophistication: 1 = low sophistication, 2 = medium low sophistication, 3 = medium high
sophistication, 4 = high sophistication Economic Adversity: 0 = respondent employed, 1 = respondent unemployed
Clarity of Responsibility: Index developed by Powell and Whitten measuring the degree to which institutions make attribution of responsibility for economic outcomes more or less clear. Higher values indicate institutions make responsibility attribution less clear. Ranges from 0.1 to 3.2.
Party List System: 0 = not a party-list system, 1 = party-list system
appendix table Al    Variable Summary
Country	Sample size	Vote intention	Sociotropic evaluation	Ideology	Age	Female Education		Income
Belgium	11,194	0.29	0	5.62	43	0.51	5	2
Denmark	11,009	0.33	0	5.73	44	0.51	4	3
France	10,980	0.28	- 1	5.65	42	0.52	4	3
Germany	11,303	0.36	0	5.44	43	0.52	3	2
Ireland	11,032	0.30	- 1	5.36	41	0.51	3	3
Italy	11,612	0.38	0	5.50	42	0.51	1	3
Luxembourg	3,641	0.31	0	5.65	42	0.48	4	3
Netherlands	11,258	0.36	0	5.33	42	0.53	5	2
United Kingdom	14,779	0.31	0	5.74	43	0.52	3	3
Means: vote intention, ideology, age and female Medians: sociotropic evaluation, education and income