The 1994 referendum was the second Norwegian referendum on accession to the EU. In 1972, a small majority of voters (53.5 per cent) rejected membership. In 1994, 89 per cent of Norwegian voters turned out to reject the accession proposal for a second time by an even smaller majority of 52.2 per cent. The Norwegian 1994 referendum is an apposite case for examining voter competence in EU referendums because it provides a critical test of our examination of voter competence and cue-taking. European integration has been more politicised in Norway than in most other countries and all political parties have taken a clear stance on the issue (Aardal & Valen 1997; Midtbø & Hines 1998; Saglie 2000; Hobolt 2005). This high salience of the European issue implies that we would expect voters to be more certain of their own issue preferences and better equipped to judge how the ballot proposal and the reversion point relate to these preferences. Moreover, party competition on the issue ought to make party endorsements more reliable cues in the referendum. In other words, if we find that voter competence is low and that partisan endorsements provide inadequate information in this referendum, we would expect even greater problems in other EU referendums held in low-salience and low-information environments. On the other hand, if we find that voter competence is high among certain voters, this can teach us important lessons about the circumstances under which voters can act competently and the type of information necessary to achieve this. In addition to providing a ‘critical test’ of voter competence, the post-referendum survey from the 1994 referendum contains a number of questions on knowledge and information as well as the respondent's perception of party positions that allow us to examine the processing of cues. This analysis thus contributes to the (mostly American) debate on information and voting behaviour by examining the question in a European setting and by explicitly analysing how different types of information and heuristics affect voter competence.
Party endorsements and voter preferences in the Norwegian referendum
The Norwegian debate on membership had begun already in November 1992 when the Labour government applied to join the EU, and it intensified after membership negotiations began in April 1993. The proposal of membership was initiated by the governing party, Arbeiderpartiet (Labour), which with 41 per cent of seats was by far the largest party in the Norwegian parliament (Stortinget) after the 1993 election. In addition to this, the main opposition party, the conservative Høyre, also supported membership, as did the populist right-wing party, Fremskrittspartiet (Progress Party). Altogether these pro-EU parties made up 65 per cent of seats in Stortinget. Hence on the basis of this information alone, we would predict that the membership proposal would pass, but only 60 per cent of the voters who identified with the pro-EU parties actually followed their recommendation. In comparison, 76 per cent of voters who identify with the anti-EU parties voted ‘No’. EU membership was opposed by Senterpartiet, an agrarian centre party, as well as two small parties on the far left and two small centre-right parties. Hence, while Norwegian party politics is structured along the left-right dimension, the European integration issue has split the traditional blocs and created new political alliances (Aardal 1995; Saglie 2000). Figure 4 shows the location of the parties on the European and the left-right dimension, using data Ray's party expert survey.4 This figure also includes the mean position of the voters on these dimensions.
Figure 4. Party and voter positions in Norway on two dimensions. Notes: Ap: Arbeiderpartiet (Labour Party), Frp: Fremskrittspartiet (Progress Party), H: Høyre (Conservative Party), KrF: Kristelig Folkeparti (Christian Peoples' Party), RV: Rød Valgallianse (Red Alliance List), Sp: Senterpartiet (Centre Party), SV: Sosialistisk Venstreparti (Socialist Left Party), V: Venstre (Liberal Party), No party: voters who did not vote in the 1993 Storting election. Sources: EU-avstemningen 1994 and Leonard Ray's party expert survey. See Appendix.
Download figure to PowerPoint
As Figure 4 illustrates, there is no unambiguous relationship between the left-right dimension and the EU dimension. The extreme left tends to be very anti-EU, but not more so than the Centre Party. Another important finding is that voters tend to be closer to their parties on the left-right dimension than on the European dimension. This is particularly true for voters of the two largest parties, Arbeiderpartiet and Høyre, who are relatively closely aligned to their party on the left-right dimension, but far more eurosceptical than their preferred parties. As argued above, these orthogonal dimensions may potentially reduce the ability of voters to act competently since adhering to the endorsements of preferred parties based on proximities on the left-right dimension provides inaccurate information on where they are located in relation to the proposal on the European dimension. The important question is therefore not only the distance between voter and preferred party positions on the EU dimension, but also whether voters are aware of this distance. Voters are known to project their own views onto parties they support, thereby enhancing the appearance of consistency in their positions. Since the Norwegian post-referendum survey contains questions on both self-placement and the perceived position of parties on different policy dimensions, we can compare the ‘perceived’ distance between voters and parties with the ‘actual’ distance, based on party expert survey data. Table 2 shows the results for both dimensions.
Table 2. Distances between voters and parties
|Distance measure||All voters||Low political knowledge||Medium political knowledge||High political knowledge|
|EU distance (ideal – actual party position)||−2.0 (2.8)||3.0 (2.3)||−2.6 (3.4)||3.1 (2.2)||−1.9 (2.8)||3.0 (2.3)||−1.8 (2.6)||3.0 (2.4)|
|EU distance (ideal – perceived party position)||−1.3 (2.0)||2.9 (2.5)||−1.4 (2.0)||2.9 (2.4)||−1.3 (2.1)||3.0 (2.4)||−1.2 (1.9)||2.8 (2.5)|
|Projection: difference between actual and perceived proximity||0.8|| ||1.4|| ||0.7|| ||0.7|| |
|Left-Right distance (ideal – actual party position)||−0.7 (1.6)||1.9 (1.3)||−0.8 (1.6)||2.1 (1.6)||−0.7 (1.6)||1.9 (1.6)||−0.6 (1.6)||1.9 (1.3)|
|Left-Right distance (ideal – perceived party position)||−0.2 (0.8)||1.4 (1.2)||−0.3 (0.8)||1.4 (1.2)||−0.3 (0.8)||1.4 (1.2)||−0.2 (0.8)||1.4 (1.2)|
|Projection: difference between actual and perceived proximity||0.8|| ||0.8|| ||0.8|| ||0.8|| |
|N||3,332|| ||891|| ||1,540|| ||901|| |
This table illustrates that the distance between parties and voters on the EU dimension is greater than on the left-right dimension. Using party expert survey data, voters are on average 2 points more eurosceptical than their preferred party and the distance is 2.8 points in absolute values. This is a considerable distance on a 10-point scale. In comparison, voters are on average 0.7 point more left-wing than their preferred party, with an absolute (mean) distance of 1.6. There is, however, no difference in the extent to which voters project their own views onto their preferred party. On both dimensions, voters on average perceive themselves to be 0.8 points closer to their party than the expert evaluation. It is interesting to note that if we divide these results by the level of political knowledge, voters with very limited knowledge of politics are far more likely to ‘project’ (report a smaller distance than the experts) on the EU dimension than voters with higher level of political awareness, but these differences are not found on the left-right dimension. This indicates that a higher level of knowledge is required to interpret partisan cues on the European dimension correctly. Following this, we can conjecture that party endorsements may be potentially misleading if people have a very limited knowledge of party politics (see Lau & Redlawsk 2001). On the other hand, the clear positions of parties on this issue (Figure 4 shows that parties are located at each end of the spectrum) may provide useful cues to voters given that they are aware of these positions and how they differ from their own position. These propositions will be tested below.
The impact of information on voter competence
In this article, voter competence in EU referendums is conceptualised as the ability of voters to choose the alternative (the proposal or status quo) that is closest to their preferences on the issue of European integration. It has been argued that information is required for voters to perform this task, but it is inherently difficult to establish absolute standards for competent voting in a specific referendum. In order to classify voters into categories of ‘competent’ and ‘not competent’ voters, we would need to observe not only their true preferences on the EU dimension, but also the location of the ballot proposal and of status quo. While we can reasonably argue that accession to the EU is more ‘pro-integration’ than staying outside the Union, we cannot precisely locate these alternatives on a scale, nor can we be certain that the reported answers on EU attitude questions represent true preferences for all voters. Yet since the value we want to observe is how voters would have voted if they were themselves fully informed about these positions, we can get an approximate idea of competent voting by comparing groups according to their level of information. By using the method of ‘well-informed proxy group comparison’ we can compare voting behaviour across types and levels of information. The assumption is that if you are very knowledgeable about European politics, you will also better equipped to choose the best alternative according to your own preferences, and you will not be mislead by elite cues that are incompatible with your own preferences (Zaller 1992; Kriesi 2005). Hence, we want to examine the extent to which the vote choice is compatible with the voter's EU preferences and how this compares across groups with different levels of information. Moreover, we are also interested in finding out to what extent endorsement heuristics can aid competent voting.
To examine the effects of information on competent voting, we distinguish between three types of information: cue reception, cue knowledge and EU knowledge. The ‘cue reception’ indicator simply measures individual exposure to partisan endorsements. This scale thus measures the extent to which people have been exposed to and can identify the endorsements of parties, but it does not measure whether individuals identify party positions correctly. The ‘cue knowledge’ indicator measures whether individuals can correctly identify the positions of parties. This scale is created by calculating the distance between the voters' perception of party positions and their actual position (according to the expert surveys): the smaller the distance, the greater the knowledge of party positions and the higher the score on the scale. The last scale measures detailed knowledge of the EU. This EU knowledge scale is created as a summated ‘political awareness’ scale by calculating correct answers to factual knowledge questions on the EU, supplemented by subjective knowledge questions (Zaller 1992; Converse 2000).
By examining how these types of information interact with vote considerations, we can arrive at a better understanding of how information and heuristics influence voter competence. However, two potential concerns should be addressed. The first concern is that the three measures may capture the same thing and are so highly correlated that any distinction is meaningless. While this is a valid concern, the conceptual distinction between the three measures is very important when examining competence. Moreover, although the indicators are positively correlated – as we would expect – this correlation is very low (below 0.2) in all cases,5 and the inter-item reliability of each scale is very high.6 A closer look at the three indicators also reveals that the ‘EU knowledge’ group seems to capture what is generally referred to as political ‘sophistication’ or ‘awareness’ (Zaller 1992). A second potential problem is that these measures – particularly the EU knowledge measure – are endogenous to our dependent variable. In other words, one could argue that the more people support the integration project, the more likely they are to acquire information about the issue. While this is a valid argument, the empirical evidence shows no correlation between EU knowledge and support for European integration, and only a very low correlation between cue-taking and support. Hence, we have grounds to believe that these indicators are capturing three distinct types of information processing that could have markedly different effects on competent voting.
To test the impact of different types of information on voter competence, we can thus compare the behaviour of voters who differ only in the amount and types of information they possess. Our expectation – as outlined above – is that both detailed knowledge of the EU politics and knowledge of party positions will make issue preferences more important in determining the vote choice. However, we do not expect that simply receiving party cues will have the same effect since such party endorsements may be potentially misleading without a basic level of information. To test these propositions, I have specified three logit models including interaction terms for each specific information type and the variables that are expected to be most significant in determining the vote: EU preferences, preferred party endorsement and (perceived) distance to the voter's preferred party. The dependent variable is ‘Yes’/‘No’ vote in the referendum.7 The variable on individual voters' EU preferences – or ‘ideal points’ on the European integration dimension – is based on a question where respondents are asked to place themselves on a 10-point EU-attitude scale. They are also asked to place each of the parties on the same scale, and this allows us to calculate the perceived distance between the respondent and his or her preferred party. The preferred party was determined on the basis of the respondent's vote in the last Storting election (in 1993).8 Party endorsements are coded as a dummy variable depending on whether the respondents' preferred party recommended a ‘Yes’ or a ‘No’ vote. In addition, the model includes controls for age, income, region, left-right ideology and government satisfaction.9 To test the hypothesis of how the voters' utility calculus varies depending on their knowledge, the model includes two interaction terms for the effect of knowledge and party endorsement and knowledge and preferences.
Table 3 reports the logit results. These results suggest that the type and the amount of information that people have affect the way in which they decide in the referendum. We can observe that the interactions between EU preferences and cue knowledge and factual EU knowledge are significant, but that this is not the case for cue reception. In other words, both factual knowledge of the EU and knowledge of party position will make issue preferences a more important determinant of the vote, whereas simply receiving cues makes no difference. In addition, the data illustrate that knowledge of partisan positions mediates the extent to which party recommendations affect behaviour (and the sign is negative) and makes it less likely that voters follow the recommendation of their preferred party (again, the sign is different from the main effect). The results thus seem to indicate that, as expected, both ‘encyclopaedic’ information and knowledge of partisan cues can enhance voter competence (since they augment the effect of issue preferences), whereas exposure to endorsements has no effect. The fit of each of the models is very good with a pseudo R2 of approximately 0.60 and over 87 per cent of cases correctly predicted by the models (with a distribution of the dependent variable of 46 per cent ‘Yes’ voters, according to the survey data). However, logit coefficients do not provide intuitive information about the absolute magnitude of these differences across types of information. Also, the logit models do not give us a straightforward way of comparing between different levels of information. Hence, to aid the interpretation and understanding of these information effects, I calculated the predicted probability of voting ‘Yes’, given 0.5 standard deviation change in each of the significant predictor variables, for each of the information types (cue reception, cue knowledge, EU knowledge) across three information levels (low, medium, high10). The results are reported in Table 4.
Table 3. The effect of information on voting behaviour
|Independent variables||Model 1: Reception of cues||Model 2: Knowledge of cues||Model 3: Knowledge of the EU|
|EU attitudes (‘ideal point’)||0.85**||0.06||0.53**||0.10||0.76**||0.08|
|Distance to party (EU)||0.05||0.10||0.07||0.08||0.06||0.05|
|Factual EU knowledge||–||–||–||–||−0.01||0.04|
|Cue reception* attitudes||0.01||0.02||–||–||–||–|
|Cue reception* endorsement||−0.09||0.06||–||–||–||–|
|Cue knowledge* attitudes||–||–||0.12**||0.02||–||–|
|Cue knowledge* endorsement||–||–||−0.16**||0.05||–||–|
|EU knowledge* attitudes||–||–||–||–||0.06**||0.01|
|EU knowledge* endorsement||–||–||–||–||−0.07||0.08|
|% correctly predicted||87|| ||89|| ||89|| |
|McFadden's R2||0.58|| ||0.59|| ||0.60|| |
|N||2,832|| ||2,832|| ||2,832|| |
Table 4. Probability of a ‘Yes’ vote by information categories and levels
|Explanatory variables||Model 1: Reception of cues||Model 2: Knowledge of cues||Model 3: Knowledge of the EU|
|Impact of change in variable (%)||Resulting change in ‘Yes’ vote (95% C.I.)||Impact of change in variable (%)||Resulting change in ‘Yes’ vote (95% C.I.)||Impact of change in variable (%)||Resulting change in ‘Yes’ vote (95% C.I.)|
| EU preferences||33**||46 to 80% (76–83%)||33**||46 to 80% (76–83%)||35**||46 to 81% (78–84%)|
| Party endorsement|| 7**||46 to 53% (51–55%)|| 7**||46 to 53% (51–55%)|| 7**||46 to 53% (51–55%)|
|Low information level|
| EU preferences||30**||42 to 72% (63–78%)||26**||41 to 62% (53–72%)||26**||46 to 72% (67–78%)|
| Party endorsement|| 8**||42 to 50% (45–54%)||10**||41 to 51% (47–55%)|| 5**||46 to 52% (48–56%)|
|Medium information level|
| EU preferences||34**||48 to 83% (78–87%)||33**||52 to 84% (79–89%)||34**||49 to 82% (77–88%)|
| Party endorsement|| 7**||48 to 56% (53–59%)|| 7**||52 to 59% (54–62%)|| 8**||49 to 57% (53–60%)|
|High information level|
| EU preferences||32**||46 to 78% (72–84%)||36**||37 to 73% (62–77%)||39**||42 to 28% (18–38%)|
| Party endorsement||5*||46 to 51% (46–56%)||5||37 to 42% (37–48%)||7*||42 to 49% (44–53%)|
In Table 4, the EU preference variables were increased half a standard deviation, ceteris paribus, which is roughly equivalent to a 11/2 point increase in EU positive attitudes on a 10-point scale. The results show that for all groups of voters, this change in EU preferences has a greater effect on the probability of voting ‘Yes’ than party endorsements (and also greater than the effect of any other predictor included in the model). This suggests that voter competence may be relatively high, given that our normative expectation is that voters should vote on the basis of their issue preferences. However, voting on the basis of issue preferences in a competent manner requires information. Since our assumption is that high levels of detailed knowledge of the EU enable voters to identify correctly the choice that maximises utility, our ‘proxy group’ for competence is the group of voters with high levels of knowledge of the EU. The easiest way to compare across groups is to illustrate these predicted probabilities graphically.
Figure 5 shows that EU preferences have the most significant impact on voting for the ‘proxy group’ of voters with high levels of factual information; yet information about party cues produces similar patterns of voter behaviour. If a voter is well-informed about party positions on the EU, then issue preferences will matter more and party endorsements less (partisanship is insignificant for the high information group in model 2). This similarity of patterns of voting behaviour across levels for these two types of information suggests that knowledge of party cues can be used as a reliable substitute for detailed knowledge of the EU. On the other hand, we do not detect this variation across levels when we look at ‘cue reception’, suggesting that simply being exposed to cues is not going to make one better equipped to vote competently. This confirms our expectation that a certain level of knowledge about party positions is required in order to use party endorsements as effective heuristics.
The models presented in Tables 3 and 4 estimate the effect of different information types separately, but, in reality, some people have knowledge of party cues as well as detailed information about the EU, whereas others will lack both detailed knowledge and an understanding of party positions. To analyse how this combination of information effects influences voting behaviour, we can simulate the impact of a change in EU preferences on the probability of voting ‘Yes’ (as in Table 4) for a combination of the two information types at each information level (low, medium, high). Table 5 reports the predicted impact of a change in EU preferences for each category, including the 95 per cent confidence intervals (in brackets) and the number of respondents in each cell.
Table 5. Impact of EU attitude change on the probability of voting ‘Yes’, across information categories (percentages)
|Knowledge of party cues||Factual EU knowledge|
|Low||20 (15–24)||27 (25–29)||35 (26–42)|
|N = 186||N = 288||N = 64|
|Medium||26 (21–31)||36 (33–38)||60 (47–74)|
|N = 220||N = 601||N = 127|
|High||26 (13–33)||26 (21–31)||*|
|N = 49||N = 222||N = 56|
As expected, Table 5 illustrates that the people who lack knowledge of both cues and the EU are least likely to rely on their EU preferences, but this group only constitutes about 10 per cent of people who voted. The impact of EU preferences increases the higher people score in the factual knowledge category, whereas the evidence is more ambiguous for the ‘cue knowledge’ category. These findings suggest that extensive EU knowledge makes attitudes more relevant than cue knowledge. While there appears to be little difference between the medium and high ‘cue knowledge’ groups when combined with factual knowledge, there is a considerable difference between the medium and high ‘factual knowledge groups’ when combined with cue knowledge. Hence, as suggested in other studies on the use of heuristics, political sophisticates are also better equipped to use cues effectively. Yet, both types of information appear to enhance competent issue voting: the people who rely most on their issue preferences are those people with high knowledge of the EU and medium/high knowledge of party positions (for the group of voters with the highest scores in both information categories, their votes are perfectly predicted by their ‘ideal points’).
In sum, these comparisons of voting behaviour across types and levels of information provide important insights into our understanding of voter competence in EU referendums. The findings suggest that while issue preferences matter to all voters, they are more important to the well-informed voters. Moreover, we find that voters can act competently without detailed knowledge of the EU by relying on party endorsements, if they have a basic knowledge of party positions on the EU. However, the argument that the mere presence of elite cues enhances reasoned voting is not supported by the data. The results in Tables 3 and 4 indicate that high exposure to the position of elite cues is not going to make voters better equipped to act on the basis of their preferences.