Measuring country level support for European integration: A median voter approach. Christopher Prosser. University of Oxford, UK

Measuring country level support for European integration: A median voter approach Christopher Prosser University of Oxford, UK chris.prosser@politics.ox.ac.uk Abstract Scholars of European have frequently examined questions about the role public opinion toward European integration plays in domestic and European politics and policy. Existing approaches to measuring aggregate level opinion toward integration are overly simplistic and problematic. Using existing Eurobarometer data, this paper proposes a new way of estimating aggregate level support that utilises a median voter approach to the problem and estimates median position and polarisation over European integration for each country in a Eurobarometer wave. It then demonstrates why this approach is superior to existing measures by reanalysing three previously published studies. Paper prepared for presentation at the 2012 Elections, Public Opinion, and Parties Conference, University of Oxford, UK, September 7-9, 2012. 1

What do Europeans think about European integration and what impact does this have on domestic and European politics? The earliest examination of these questions was concerned with the decline of support for the European Communities during the period of eurosclerosis and its consequences for further integration. 1 This was followed by work which examined the stabilisation of attitudes at a highly supportive position towards during the 1980 s. 2 As the Cold War ended and the European project picked up speed these feelings began to cool. 3 Most recently Hooghe and Marks have argued that this shift in public attitudes towards European integration has been from a permissive consensus to a constraining dissensus. 4 Many scholars have turned their attention to explaining these attitudes: Some have focused on the role of economic performance and the perceived benefits of membership. 5 Others have examined the role social and cultural attitudes. 6 A different set of research has turned to questions of the role public opinion toward European integration plays in politics. Some have examined whether voter attitudes towards integration plays a role in domestic elections. 7 Whilst others have looked at the role of public opinion towards integration plays in 1 David H. Handley, Public Opinion and European Integration: The Crisis of the 1970s, European Journal of Political Research 9, no. 4 (1981): 335 364. 2 Ronald Inglehart, Jacques-Rene Rabier, and Karlheinz Reif, The Evolution of Public Attitudes Toward European Integration: 1970-1986, Journal of European Integration 10, no. 2 3 (1987): 135 155. 3 Mark Franklin, Michael Marsh, and Lauren McLaren, Uncorking the Bottle: Popular Opposition to European Unification in the Wake of Maastricht, Journal of Common Market Studies 32, no. 4 (1994): 455 472; Richard C. Eichenberg and Russell J. Dalton, Post-Maastricht Blues: The Transformation of Citizen Support for European Integration, 1973-2004, Acta Politica 42 (2007): 128 152. 4 Liesbet Hooghe and Gary Marks, Calculation, Community and Cues: Public Opinion on European Integration, European Union Politics 6 (2005): 419 443; Liesbet Hooghe and Gary Marks, A Postfunctionalist Theory of European Integration: From Permissive Consensus to Constraining Dissensus, British Journal of Political Science 39 (2009): 1 23. 5 Christopher J. Anderson and Karl C. Kaltenthaler, The Dynamics of Public Opinion Toward European Integration, 1973-93, European Journal of International Relations 2, no. 2 (1996): 175 199; Richard C. Eichenberg and Russell J. Dalton, Europeans and the European Community: The Dynamics of Public Support for European Integration, International Organization 47, no. 4 (1993): 507 534; Matthew Gabel, Public Support for European Integration: An Empirical Test of Five Theories, Journal of Politics 60, no. 2 (1998): 333 354; Matthew Gabel and Guy D. Whitten, Economic Conditions, Economic Perceptions, and Public Support for European Integration, Political Behavior 19, no. 1 (1997): 81 96. 6 Joseph H. Janssen, Postmaterialism, Cognitive Mobilization, and Support for European Integration, British Journal of Political Science 21 (1991): 443 68; Lauren McLaren, Public Support for the European Union: Cost/Benefit Analysis or Perceived Cultural Threat?, Journal of Politics 64, no. 2 (2002): 551 566; Adam Luedtke, European Integration, Public Opinion and Immigration Policy: Testing the Impact of National Identity, European Union Politics 6, no. 1 (2005): 83 112. 7 Clifford J. Carrubba, The Electoral Connection in European Union Politics, The Journal of Politics 63, no. 1 (2001): 141 158; Geoffrey Evans, Europe: A New Electoral Cleavage?, in Critical Elections: British Parties and Voters in Long-term Perspective (London: Sage, 1999), 207 222; Geoffrey Evans, European Integration, Party 2

turnout and voting in European elections. 8 Recently efforts have also turned to the more direct link between public opinion European level policy output. 9 Some of these research questions examine individual attitudes towards integration, but others can only be examined at the aggregate level. These questions present scholars with the challenge of how to measure attitudes to European integration at the country level. Despite being a well-established research area and a plethora of data, existing measures of public opinion to integration have been unsophisticated and have several problems. This paper outlines these issues and proposes a new way of estimating country level public opinion towards European integration that draws on spatial models of politics to transform the existing Eurobarometer data. Previous measures Measuring country level public opinion on European integration is almost a unique problem in political science in the sheer quantity of data available. The Eurobarometer survey series has asked citizens of the European Union and its predecessors for their opinions towards integration at regular intervals for almost 40 years. 10 Almost all measures of public opinion towards European integration based on the Eurobarometer survey series (including the current proposal) are based on a question Politics and Voting in the 2001 Elections, Journal of Elections, Public Opinion & Parties 12 (2002): 95 110; Peter Mair, The Limited Impact of Europe on National Party Systems, West European Politics 23, no. 4 (2000): 27 51; Cees Van der Eijk and Mark N. Franklin, Potential for Contestation on European Matters at National Elections in Europe, in European Integration and Political Conflict (Cambridge: Cambridge University Press, 2004), 32 50; Catherine E. de Vries, Sleeping Giant: Fact or Fairytale?: How European Integration Afffects National Elections, European Union Politics 8 (2007): 363. 8 Sara Binzer Hobolt, J.J. Spoon, and James R. Tilley, A Vote Against Europe? Explaining Defection at the 1999 and 2004 European Parliament Elections, British Journal of Political Science 39, no. 1 (n.d.): 1 23; Daniel Stockemer, Citizens Support for the European Union and Participation in European Parliament Elections, European Union Politics 13, no. 1 (2012): 26 46; Peter Van Aelst and Jonas Lefevere, Has Europe Got Anything to Do with the European Elections? A Study on Split-ticket Voting in the Belgian Regional and European Elections of 2009, European Union Politics 13, no. 1 (2012): 3 25; Daniele Caramani, Is There a European Electorate and What Does It Look Like? Evidence from Electoral Volatility Measures, 1976-2004, West European Politics 29, no. 1 (2006): 1 27. 9 Dimiter Toshkov, Public Opinion and Policy Output in the European Union: A Lost Relationship, European Union Politics 12, no. 2 (2011): 169 191. 10 European Commission, Eurobarometer, 1973-2012, http://ec.europa.eu/public_opinion/index_en.htm. 3

that asks survey respondents whether they think [their country s] membership of the European Union is a good thing/bad thing/neither a good or bad thing? 11 Some previous measures of individual attitudes have used multiple questions to create an index of support for European integration, such as the measure of individual level opinion developed by Gabel. 12 This measure combines the common question with a second Eurobarometer question which asked whether they were Very much for/to some extent for/to some extent against/very much against efforts to unify Western Europe. Unfortunately however, this second question was only asked between 1978 and 1994. Similar questions, asking whether people thought their country had benefited from membership, how happy they were with the speed of integration, or how sorry they would be if the European Union was scrapped, are also faced with the problem of only being asked for part of the survey period. That most measures are based on a single question is a matter of practicality rather than design: the membership question is the only one that asks a respondent for their general attitudes towards integration that has been asked across the entire span of the Eurobarometer survey series. In the existing literature, several scales have been constructed from the good/bad/neither good or bad thing question. The three most common measures differ as to what they do with the respondents that answer a bad thing and neither a good or bad thing. The simplest, gross support, is simply the percentage of respondents who answer a good thing and does not count the other two answers. The second measure, first proposed by Inglehart and Rabier, net support, subtracts the percentage of respondents answering a bad thing from the percentage answering a good thing, and ignores those answering neither. 13 The third, which I term total net support, subtracts the percentage of respondents answering both a bad thing and neither a good or bad thing from the 11 Previously European Union would have been replaced by the relevant predecessor: the European Communities, Common Market etc. 12 Matthew Gabel, Public Support for European Integration: An Empirical Test of Five Theories, Journal of Politics 60, no. 2 (1998): 333 354. 13 Ronald Inglehart and Jacques-Rene Rabier, European Community Study 1973, n.d.; Eichenberg and Dalton, Europeans and the European Community: The Dynamics of Public Support for European Integration. 4

percentage answering a good thing. An immediate issue with all three measures is how they treat those who are neutral and hostile towards European integration. Is the absence of explicit support for European integration (neither a good nor a bad thing) the same as opposition to it (a bad thing?) or is absence if explicit condemnation the same as support? Or, as is the case with net support should neutrality simply be ignored? Rather than justify their choices on theoretical grounds most scholars have taken a pragmatic approach to this problem. Eichenberg and Dalton deal with this issue by arguing that empirically there is little difference between the measures as they are very highly correlated. 14 Although this is true, it is an artefact of the data and not a structural feature of the measures and the high correlation overstates the similarity between the measures. The percentage of respondents answering a good thing is generally high (mean = 56.47%), whilst those answering neither a good or bad thing is generally low (mean = 12.46%). In cases where support for European integration is low the differences between those who are neutral to integration and those opposed will have a greater impact on the calculation of the score. The importance of this point is easily illustrated with an example. In case A 60% of respondents think their country s membership of the EU is a good thing, 20% think it is neither a good or a bad thing and 20% think it is a bad thing. In case B 45% answer good, 50% neither good or bad, and only 5% bad. Finally in case C 62% answer good, 3% neither good or bad, and 35% bad. For the gross support measure the scores for A, B, and C respectively are 60, 45, and 62. For the net support measure the scores are 40, 40, and 27. For the total net support measure the scores are 30, -10, and 24. This illustration shows that the choice of measure can lead to substantial differences in perceived levels of support for integration the gross support measure ranks case C as having the highest support, whilst the net support measure suggests it has the lowest, and the total net support measure puts it in the middle. 14 They are only referring to net support and total net support, however the same is true for gross support. 5

This is not just an abstract, theoretical issue either. As figure 2, which plots the two net support measures against the gross support measure illustrates quite clearly, the relationship between the different measures is heteroscedastic: the correlation between the measures decreases and support for integration gets lower. Figure 1. Correlation between the two net support measures and the gross support good thing measure. The correlation between the two net support measures (not shown) exhibits a similar pattern. Given that, particularly recently, research into public opinion towards European integration has been driven by a concern with opposition towards the EU, it is important that we are clear what we are measuring and we do not confound our results by conflating or ignoring neutrality and opposition. 6

A median voter approach Rather than try and find answer to the question to whether respondents who think that European integration is neither a good nor a bad thing should be counted as opposition, support or ignored entirely, the approach proposed here sidesteps the problem by estimating a score on an interval level scale of support for European integration for the median respondent in each observation. Using a measure of central tendency for each observation rather than the number of respondents who give a particular answer brings measurement of public opinion towards European integration in line with the standard practice of aggregated measurement. Using the median in particular fits well with general theories of political competition, and has the additional advantage of being easily rescaled. In order to estimate the median opinion on European integration this approach reimagines the Eurobarometer question about European integration as a spatial model of social choice. Spatial models of politics have long been a feature of political science, particularly in the literature on party competition. They originated with the work of Black and Downs, who introduced and developed the median voter model respectively. 15 The detail of the various (increasingly complicated) models that have been developed since are not important here. This approach uses the basic theoretical model of electoral competition: In a two choice vote, if voters vote for the option that is closest to their own views the option that is closest to the median voter s preference, which by definition includes the majority of the electorate, will win the election. The approach here utilises the median voter idea in a different way but shares common assumptions with the approach: 1) Voters can place different options along a one-dimensional continuum. 2) A voter s preferences are single peaked. That is a voter will choose the option closest to their actual preference. 15 Duncan Black, On the Rationale of Group Decision-making, Journal of Political Economy 56 (1948): 23 34; Anthony Downs, An Economic Theory of Democracy (New York: Harper Collins, 1957). 7

3) Voters will always choose the option based on their true preferences. These assumptions are necessary for the parsimonious conclusions of the median voter model but they are clearly broken in political reality: politics is multidimensional and some voters will vote strategically. In order to estimate the median attitude towards European integration it is necessary to make similar assumptions: 1) The survey question is measuring a one-dimensional continuum. 2) Respondents will answer with the option closest to their true beliefs. Again the assumptions are simplifications of the reality: European integration is clearly a multidimensional issue. However given the anonymous and inefficacious nature of survey research for the respondent, strategic voting is likely to be less of an issue. These assumptions are simplifications but they are given here for transparency. They are assumptions shared by any survey approach, and more likely to resemble reality than the original median voter model. Rather than modelling who will win an election, the approach here uses the known distribution of respondents and responses to estimate what that median position actually is. In order to do so it utilises an approach first develop by Kim and Fording for use on party positions estimated from manifesto data. 16 The Kim-Fording method estimates the median voter position on a policy scale by arranging parties in the election in ascending order of their position on the policy scale and identifying in which party the median voter lies. The interval between the previous and subsequent parties is used to calculate the median voter position using the following formula: 16 HeeMin Kim and Richard C. Fording, Voter Ideology in Western Democracies, 1946-1989, European Journal of Political Research 33 (1998): 73 97; HeeMin Kim and Richard C. Fording, Voter Ideology in Western Democracies: An Update, European Journal of Political Research 42 (2003): 95 105. 8

= + 50 Where: M = Median position L = The lower end (ideological score) of the interval containing the median C = The cumulative frequency (vote share) up to but not including the interval containing the median F = The frequency (vote share) in the interval containing the median W = The width of the interval containing the median In order to use this method estimate the median position on European integration it is first necessary to assign scores for each option on an interval scale. The method proposed here is compatible with any scale but for ease of interpretation and similarity with previous measures a scale ranging from 0 to 100 is used, where 0 is the anti-integration endpoint and 100 is the prointegration endpoint. The scale assumes an even distribution of attitudes towards European integration and so the scale is divided into thirds, as shown in figure 3: the first third representing the area covered by hostile attitudes towards integration, the middle third representing neutral attitudes that are a mix of good and bad, and the final third representing positive attitudes towards integration. The survey responses are then assigned values on the scale as follows, illustrated in figure 3: 1) A bad thing is assigned to the intersections between hostile and neutral at one third of the scale. 2) Neither a good or a bad thing is assigned to the midpoint of the scale. 3) A good thing is assigned to the intersection between neutral and positive attitudes twothirds into the scale. 9

Figure 2. Attitudes towards European integration and positions assigned on the interval scale. In order to calculate the width (W) and lower end (L) of each interval the midpoint between the assigned values are calculated (shown as m1 and m2 in figure 2). The bounds and widths of each interval are shown in table 1. Interval Lower end Upper end Width A bad thing 0 41.66 (m1) 41.66 Neither a good or bad thing 41.66 (m1) 58.33 (m2) 16.66 A good thing 58.33 (m2) 100 41.66 Table 1. Lower and upper ends, and width of each interval The frequency portion of the equation is now calculated from the Eurobarometer survey results. Approaching the question of what to do with respondents who answer don t know as if the survey was an election provides an obvious solution to the problem of what to do with non-respondents: they are considered as non-voters and their abstentions not counted for the estimate of the median. The proportion of respondents answering each category is then recalculated as the proportion of respondents who answered one of the three categories. The median score then should correctly be thought of as the median score of those who answered the question. If we assume a random distribution of non-answerers than this median would be the same as the sample as a whole, but whether non-answers are randomly or systematically distributed is outside the scope of the present paper. 10

The results of the calculations for each country/time observation are illustrated in figure 3. Austria Belgium Bulgaria Czech Republic Denmark 40 50 60 70 80 Estonia Finland France Germany Greece 1970h1 1980h1 1990h1 2000h1 2010h1 1970h1 1980h1 1990h1 2000h1 2010h1 1970h1 1980h1 40 50 60 70 80 1990h1 40 50 60 70 80 2000h1 2010h1 1970h1 1980h1 1990h1 2000h1 2010h1 1970h1 1980h1 1990h1 2000h1 2010h1 40 50 60 70 80 40 50 60 70 80 40 50 60 70 80 Hungary Ireland Italy Latvia Lithuania EU Median Luxembourg Malta Netherlands Poland Portugal Republic of Cyprus Romania Slovakia Slovenia Spain Sweden UK Graphs by Country year Figure 3. Longitudinal median attitude towards European integration by country 11

The proposed median measure is highly correlated with the three previous measures discussed here: 0.9582 with gross support, 0.9790 with net support, and 0.9873 with total net support. As has already been discussed however, high correlation should not simply be taken as substitutability. The main improvement of the median measure over the previous measures is that it does not ignore or equate the neither good or bad thing and bad thing responses and produces an estimate that is theoretically meaningful. Polarisation Conceptualising the survey question as a spatial model also suggests an additional measure: polarisation of attitudes to European integration. Measures of polarisation have long been used in the literature on party-systems to explain a wide variety of phenomena, from political stability, and cabinet-survival, to the quality of a party system. 17 Similarly, scholars of American politics have long debated the extent and consequences of electorate polarisation in the United States. 18 As of yet the same is not true however of the study of European integration: although many authors have examined the role competition over European integration might play in elections, no longitudinal measure of voter polarisation over European integration has been developed. The measure proposed here seeks to fill this gap. The measure used is the weighted standard deviation of responses to Eurobarometer question on attitudes to European integration for each observation, using the same scale as the median measure, calculated as follows: 17 Giovanni Satori, Parties and Party Systems (Camb: Cambridge University Press, 1976); Paul Warwick, Government Survival in Parliamentary Democracies (Cambridge: Cambridge University Press, 1994); Russell J. Dalton, The Quantity and the Quality of Party Systems, Comparative Political Studies 41, no. 7 (2008): 899 920. 18 Gary Jacobson, Polarized Politics and the 2004 Congressional and Presidential Elections, Political Science Quarterly 120, no. 2 (2005): 199 218; Morris P Fiorina, Culture War? The Myth of a Polarized America, 2nd ed. (New York: Pearson Longman, 2006); Alan I. Abramowitz and Kyle L. Saunders, Is Polarization a Myth?, Journal of Politics 70, no. 2 (2008): 542 555. 12

= ( ) + ( ) + ( ) + + Where:,, are the number of respondents ( votes ) answering bad thing, neither good or bad, good thing respectively.,, are the scores assigned on the interval scale to the bad thing, neither good or bad, good thing positions respectively. And, is the weighted mean attitude towards integration, calculated by: = + + + + The results of these calculations are illustrated for each country in figure 4. Unsurprisingly given that they are calculated from the same data, the polarisation and median measures are highly correlated (-0.8529). The polarisation measure should be considered an alternative measure of public opinion towards European integration and not an additional measure. Caution should be used if both were used in statistical analysis as collinearity may adversely influence the results. The choice of appropriate measure will depend on the theory that motivates the analysis. The results in figure 4 illustrate the value of this additional measure, compare for example the results of the median and polarisation for the Netherlands and Luxembourg, which show an increase in polarisation over the last 20 years, whilst the median measure has relatively stable at a highly supportive position. 13

Austria Belgium Bulgaria Czech Republic Denmark 5 10 15 Estonia Finland France Germany Greece 5 10 15 5 10 15 5 10 15 5 10 15 Hungary Ireland Italy Latvia Lithuania EU Polarisation 5 10 15 Luxembourg Malta Netherlands Poland Portugal Republic of Cyprus Romania Slovakia Slovenia Spain 1970h1 1980h1 1990h1 2000h12010h1 1970h1 1980h11990h1 2000h1 2010h1 1970h11980h1 1990h1 2000h12010h1 Sweden UK 1970h1 1980h1 1990h12000h1 2010h1 1970h11980h1 1990h1 2000h1 2010h1 Graphs by Country year Figure 4. Longitudinal polarisation over European integration by country 14

Replicating previous research In order to demonstrate the effect that choice of measurement has on the outcome of statistical analysis and the utility of the new measures, three pieces of previously published research that have used one of the existing public support measures will be retested using all three of the previous measures and the new median and polarisation measures. The three pieces of research tested use different statistical techniques and different subsets of the public opinion data. The first two use public support as an independent variable to explain turnout at European elections, and European level policy output respectively. The third uses public support as a dependent variable and examines the effect of economic performance on changes in support for integration. In Citizens Support for the European Union and Participation in European Parliament Elections Stockemer uses the net support measure for each country at the time of a European election to test the impact that country level support for European integration has on turnout at European Parliament elections. 19 Stockemer finds that turnout is higher when there is more public support for integration. Table 2 shows the results of the replication of Stockemer s analysis. The first notable result is that if the gross support good thing measure had been used than the main finding of the study would not be statistically significant. Using the other measures the finding is still significant, and the results are reasonably similar between the different models. In terms of model fit (measured by R 2 and Bayesian information criterion (BIC) statistics, the new median measure offers a modest improvement to the existing approaches, providing the first piece of evidence in support of the new measures. 19 Stockemer, Citizens Support for the European Union and Participation in European Parliament Elections. 15

Gross support model Net support model (original) Total net support model Median EU support model EU Polarisation model Public opinion measure 0.198 0.147** 0.106* 0.445** -1.723** (0.117) (0.059) (0.572) (.178) (0.791) Compulsory voting 24.649*** 24.31*** 24.507*** 24.01*** 23.540*** (5.337) (4.44) (5.315) (5.04) (5.368) Electoral system type 1.553 0.63 0.774 0.447 0.614 (2.173) (3.97) (3.983) (3.855) (4.051) Simultaneous elections 5.032** 4.99*** 4.970** 5.05** 4.980** (2.173) (2.15) (2.186) (2.16) (2.194) GDP per capita -0.0003** -0.0003** -0.0003** -0.0003** -0.0002* (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Small state 9.973** 9.77* 9.962** 10.07** 9.808** (4.485) (5.05) (4.384) (4.20) (4.300) Closeness of the electoral race -0.054-0.055-0.057-0.065-0.072 (0.137) (0.125) (0.134) (0.132) (0.132) National-level turnout 0.355** 0.344** 0.356** 0.362** 0.379** (0.144) (0.150) (0.143) (0.143) (0.138) Eastern Europe -16.144*** -17.02*** -15.625*** -15.71*** -17.123*** (5.318) (5.09) (5.378) (5.29) (5.227) Constant 20.161 26.38* 29.507** 2.947 48.7389*** (16.610) (15.17) (12.834) (19.99) 11.747 N 118 118 118 118 118 R 2 0.7298 0.7343 0.7312 0.7387 0.7346 Bayesian information criterion 870.6601 868.5817 869.5665 866.5015 869.8584 Table 2. Results of Prais-Winsten AR(1) regressions examining the link between public support for European integration on turnout at European Parliament elections. For full details of the data other than the Median EU support measure see Stockemer (2012). 16

In Public Opinion and Policy Output in the European Toshkov demonstrates that there is a link between public support for European integration (measured by the EU average gross support good thing measure for each half-year) and the volume of policy output generated by the European Commission. 20 Toshkov finds that when public support for integration is high there is a high volume of policy produced, which falls as support decreases. He also finds that the reciprocal relationship between policy production and public support is much weaker. The results of the replication analysis, shown in table 3 below do not change any of the substantive findings of the link between public support for integration and policy output. Again however the new measures outperform the existing approaches in terms of model fit, with a substantial decrease in the BIC statistic for the models run with the new measures. The substantial difference between the polarisation and median measures also suggests a possible different explanation to that offered by Toshkov: it is possible that the link between public opinion and policy output is better explained by polarisation over European integration rather than absolute levels of support. In Post-Maastricht Blues Eichenberg and Dalton examine the impact of macroeconomic performance on public opinion towards European integration (measured as net support in half years for eight countries: Belgium, Denmark, France, Germany, Ireland, Italy, the Netherlands, and the UK). 21 Eichenberg and Dalton s main finding is that factors that had previously been found to be important economic growth, inflation, unemployment, and trade concentration had decreased in importance following the Treaty of Maastricht: all except inflation were no longer statistically significant, and the substantive effect of inflation had decreased. Table 4 shows the results of the replication with one exception, in both the median and polarisation models (and the total net support model) trade concentration (Intra-EU export %) is now statistically significant. 20 Toshkov, Public Opinion and Policy Output in the European Union: A Lost Relationship. 21 Eichenberg and Dalton, Post-Maastricht Blues: The Transformation of Citizen Support for European Integration, 1973-2004. 17

Gross support model (original) Net support model Total net support model Median EU support model EU Polarisation model Public support Public support t-1 0.627*** 0.677*** 0.696*** 0.709*** 0.677*** (0.118) (0.120) (0.120) (0.119) (0.120) Public support t-2 0.164 0.074 0.082 0.048 0.031 (0.116) (0.119) (0.119) (0.118) (0.121) Legislative output t-1 0.104** 0.137** 0.196** 0.063** -0.008 (0.042) (0.061) (0.077) (0.023) (0.006) Legislative output t-2 0.032 0.045 0.058 0.021-0.002 (0.044) (0.062) (0.080) (0.024) (0.006) Constant 7.892** 5.769* -2.932 13.445*** 3.727*** (3.754) (3.319) (2.770) (4.641) (1.260) Legislative output Public support t-1 0.024 0.114 0.050 0.150-2.657 (0.316) (0.225) (0.177) (0.587) (2.417) Public support t-2 0.785** 0.533** 0.396** 1.229** -5.191** (0.312) (0.225) (0.175) (0.579) (2.438) Legislative output t-1-0.095-0.099-0.090-0.067-0.935 (0.113) (0.114) (0.114) (0.115) (0.115) Legislative output t-2-0.094-0.107-0.094-0.073-0.118 (0.118) (0.117) (0.118) (0.119) (0.116) Constant -12.737 4.891 23.976-58.552 126.413 (10.09) (6.223) (4.096) (22.751) (25.422) Number of Observations 69 69 69 69 69 R 2 Public Support 0.7262 0.6727 0.7205 0.7051 0.5727 R 2 Legislative output 0.2132 0.2168 0.2066 0.1832 0.2006 Bayesian information criterion 885.1988 934.4112 968.75 807.4372 609.6484 Table 3. Results of Vector Autoregressive models (VAR) comparing public support for European integration and policy output. For full details see Toshkov (2011). 18

Gross support model Net support model (original) 19 Total net support model Median EU support model EU Polarisation model GDP -2.417 0.643-8.92322-1.94538 0.381363 (5.425) (7.867) (10.739) (2.941) (0.622) Inflation -0.996*** -1.217*** -1.38597*** -0.44824*** 0.06445* (0.266) (0.377) (0.486) (0.149) (0.037) Unemployment -0.315-0.300-0.77553-0.23065 0.022616 (0.334) (0.474) (0.619) (0.177) (0.031) Intra-EU export % 0.103 0.137 0.331998* 0.107037** -0.02252* (0.100) (0.154) (0.194) (0.055) (0.012) UK membership referendum 13.743*** place":"ca28.39851*** 28.39851*** 11.71735*** -0.73498** (3.682) (5.657) (6.875) (2.8) (0.333) Danish SEA referendum 16.119*** 20.547*** 28.32927*** 9.721005*** -0.16634 (3.902) (5.45) (7.206) (2.173) (0.45) UK -18.897*** -37.898*** -35.7629*** -11.9466*** 2.685206*** (2.893) (4.942) (5.354) (1.777) (0.272) Denmark -10.847*** -26.130*** -22.7122*** -7.28495*** 2.80089*** (3.452) (4.801) (5.933) (1.932) (0.236) Ireland 8.446* 5.176 15.24333 3.167969 0.258332 (4.404) (6.979) (9.707) (2.568) (0.79) Italy 13.917*** 17.199*** 27.32992*** 7.085301*** -1.65066*** (2.739) (4.478) (5.445) (1.515) (0.403) Netherlands 16.286*** 19.285*** 27.68886*** 5.777824*** -1.47564*** (2.455) (4.028) (4.816) (1.411) (0.413) German unification 3.645 4.827* 6.245372 1.470166-0.58756*** (2.248) (2.933) (4.052) (1.107) (0.301) Constant 56.858*** 43.352*** 18.09958 63.44801*** 11.4596* (8.268) (12.003) (15.236) (4.419) (0.643) N 504 504 504 504 504 R 2 0.5569 0.4075 0.3037 0.8019 0.9395 Table 4. Linear regression with panel-corrected standard errors showing the effect of macroeconomic performance on public support for European Integration, for details see Eichenberg and Dalton (2007).

This indicates that Eichenberg and Dalton may have been premature in arguing that the importance of trade concentration no longer played an important role in influencing opinion towards integration. Additionally once again the model fit is substantially improved by the use of the median and polarisation measures. 22 Conclusion That the replication of previous research does not result in the dramatic overturning of previous research should not be confused with an endorsement of the previous measures. As was clearly demonstrated in each of the replications the new measures proposed here substantially improve model fit, suggesting that the new measures are a much better estimate of public opinion than previous methods and result in a reduction of error in statistical analysis. The results here suggest that the main danger of using the previous methods is committing a type II error and incorrectly rejecting a hypothesis. This is best demonstrated by the findings that the gross support model of Stockemer s study was not statistically significant and the results suggesting Eichenberg and Dalton may have incorrectly rejected the influence of trade concentration on public opinion. The median and polarisation measures proposed here offer theoretically meaningful and demonstrably superior approaches to previous measures of public opinion on European integration. The theoretical paradigm behind the approach also suggests possible routes for further research into public opinion on European integration, particularly the determinants and role of public polarisation over European integration. Scholars of European integration are fortunate in quantity of data available to research. This paper hopes to play a small part in ensuring that this data is used in the most effective way possible. 22 Panel Corrected Standard Errors regression does not allow for the calculation of a BIC statistic. Here only R 2 is used to evaluate model fit. 20

Acknowledgements I would like to thank Daniel Stockemer, Dimiter Toshkov, and Richard C. Eichenberg for providing their data and assisting me with the replication of their work. 21

References Abramowitz, Alan I., and Kyle L. Saunders. Is Polarization a Myth? Journal of Politics 70, no. 2 (2008): 542 555. Van Aelst, Peter, and Jonas Lefevere. Has Europe Got Anything to Do with the European Elections? A Study on Split-ticket Voting in the Belgian Regional and European Elections of 2009. European Union Politics 13, no. 1 (2012): 3 25. Anderson, Christopher J., and Karl C. Kaltenthaler. The Dynamics of Public Opinion Toward European Integration, 1973-93. European Journal of International Relations 2, no. 2 (1996): 175 199. Black, Duncan. On the Rationale of Group Decision-making. Journal of Political Economy 56 (1948): 23 34. Caramani, Daniele. Is There a European Electorate and What Does It Look Like? Evidence from Electoral Volatility Measures, 1976-2004. West European Politics 29, no. 1 (2006): 1 27. Carrubba, Clifford J. The Electoral Connection in European Union Politics. The Journal of Politics 63, no. 1 (2001): 141 158. Dalton, Russell J. The Quantity and the Quality of Party Systems. Comparative Political Studies 41, no. 7 (2008): 899 920. Downs, Anthony. An Economic Theory of Democracy. New York: Harper Collins, 1957. Eichenberg, Richard C., and Russell J. Dalton. Europeans and the European Community: The Dynamics of Public Support for European Integration. International Organization 47, no. 4 (1993): 507 534.. Post-Maastricht Blues: The Transformation of Citizen Support for European Integration, 1973-2004. Acta Politica 42 (2007): 128 152. Van der Eijk, Cees, and Mark N. Franklin. Potential for Contestation on European Matters at National Elections in Europe. In European Integration and Political Conflict, 32 50. Cambridge: Cambridge University Press, 2004. 22

European Commission. Eurobarometer, 2012 1973. http://ec.europa.eu/public_opinion/index_en.htm. Evans, Geoffrey. Europe: A New Electoral Cleavage? In Critical Elections: British Parties and Voters in Long-term Perspective, 207 222. London: Sage, 1999.. European Integration, Party Politics and Voting in the 2001 Elections. Journal of Elections, Public Opinion & Parties 12 (2002): 95 110. Fiorina, Morris P. Culture War? The Myth of a Polarized America. 2nd ed. New York: Pearson Longman, 2006. Franklin, Mark, Michael Marsh, and Lauren McLaren. Uncorking the Bottle: Popular Opposition to European Unification in the Wake of Maastricht. Journal of Common Market Studies 32, no. 4 (1994): 455 472. Gabel, Matthew. Public Support for European Integration: An Empirical Test of Five Theories. Journal of Politics 60, no. 2 (1998): 333 354.. Public Support for European Integration: An Empirical Test of Five Theories. Journal of Politics 60, no. 2 (1998): 333 354. Gabel, Matthew, and Guy D. Whitten. Economic Conditions, Economic Perceptions, and Public Support for European Integration. Political Behavior 19, no. 1 (1997): 81 96. Handley, David H. Public Opinion and European Integration: The Crisis of the 1970s. European Journal of Political Research 9, no. 4 (1981): 335 364. Hobolt, Sara Binzer, J.J. Spoon, and James R. Tilley. A Vote Against Europe? Explaining Defection at the 1999 and 2004 European Parliament Elections. British Journal of Political Science 39, no. 1 (n.d.): 1 23. Hooghe, Liesbet, and Gary Marks. A Postfunctionalist Theory of European Integration: From Permissive Consensus to Constraining Dissensus. British Journal of Political Science 39 (2009): 1 23. 23

. Calculation, Community and Cues: Public Opinion on European Integration. European Union Politics 6 (2005): 419 443. Inglehart, Ronald, and Jacques-Rene Rabier. European Community Study 1973, n.d. Inglehart, Ronald, Jacques-Rene Rabier, and Karlheinz Reif. The Evolution of Public Attitudes Toward European Integration: 1970-1986. Journal of European Integration 10, no. 2 3 (1987): 135 155. Jacobson, Gary. Polarized Politics and the 2004 Congressional and Presidential Elections. Political Science Quarterly 120, no. 2 (2005): 199 218. Janssen, Joseph H. Postmaterialism, Cognitive Mobilization, and Support for European Integration. British Journal of Political Science 21 (1991): 443 68. Kim, HeeMin, and Richard C. Fording. Voter Ideology in Western Democracies, 1946-1989. European Journal of Political Research 33 (1998): 73 97.. Voter Ideology in Western Democracies: An Update. European Journal of Political Research 42 (2003): 95 105. Luedtke, Adam. European Integration, Public Opinion and Immigration Policy: Testing the Impact of National Identity. European Union Politics 6, no. 1 (2005): 83 112. Mair, Peter. The Limited Impact of Europe on National Party Systems. West European Politics 23, no. 4 (2000): 27 51. McLaren, Lauren. Public Support for the European Union: Cost/Benefit Analysis or Perceived Cultural Threat? Journal of Politics 64, no. 2 (2002): 551 566. Satori, Giovanni. Parties and Party Systems. Camb: Cambridge University Press, 1976. Stockemer, Daniel. Citizens Support for the European Union and Participation in European Parliament Elections. European Union Politics 13, no. 1 (2012): 26 46. Toshkov, Dimiter. Public Opinion and Policy Output in the European Union: A Lost Relationship. European Union Politics 12, no. 2 (2011): 169 191. 24

de Vries, Catherine E. Sleeping Giant: Fact or Fairytale?: How European Integration Afffects National Elections. European Union Politics 8 (2007): 363. Warwick, Paul. Government Survival in Parliamentary Democracies. Cambridge: Cambridge University Press, 1994. 25