Party Identification Old but Gold!

Party Identification Old but Gold! Or: How to measure different stages of the electoral process in times of growing volatility and uncertainty - a comparison of Party Ratings, Propensities to Vote and yet another two new instruments 1 Johannes N. Blumenberg 2 Thorsten Faas 3 FIRST DRAFT PLEASE DO NOT CITE OR QUOTE WITHOUT THE PERMISSION OF THE AUTHORS PLEASE COMMENT Paper prepared for the XXII World Congress of Political Science in Madrid, July 2012. 1 The paper presented is one of two papers which seek to examine different measurements of electoral determinants and electoral choice and focusses primarily on Propensities to Vote and a new instruments called Vote Distribution whereas Party Bonding the PID supplement and name giver of Party Identification Old but Gold! becomes just mentioned among the lines without further explanation and examination. 2 Johannes.Blumenberg@uni-mannheim.de, Mannheim Centre for European Social Sciences, University of Mannheim 3 Thorsten.Faas@uni-mannheim.de, School of Social Sciences, University of Mannheim

Summary Ultimately on Election Day, voting is a discrete choice among parties. For methodological reasons, but also and increasingly for substantial reasons (given the growing uncertainty and volatility of voting), electoral researchers have looked for new ways and measurements to trace the process ultimately leading to people's discrete choices. Whereas feeling thermometers have been used for this purpose for quite some time, more recently, van der Eijk et al. (2006) have also argued that we need to "rethink our Dependent Variable in Voting Behavior" and have proposed to use Electoral Utilities/Propensities to Vote. In this article, we suggest yet another possibility: Giving respondents the chance to distribute up to 10 votes (instead of just 1) among parties. Based on a multi-wave panel survey conducted in the run-up to the 2011 state election in the German state of Baden-Württemberg, we critically evaluate these different measures in terms of their respective validity and reliability. Additionally these considerations become supplemented by an alternative measurement of the classical party identification, which seeks to enlarge the analytical power on this stage on the electoral process as well. 1 Introduction One of the main interests of electoral research is to explain the actual vote choice. Why do people opt for one party and not for another, to what extent do candidates, media, networks, economy, party identification and heuristics, influence their choices and why do some people tend to switch between parties more often than others? The questions we raise are located in a very complex setting. But, when it comes to our analyses, we oftentimes have to rely on one single qualitative dependent variable: the discrete vote choice. The lack of complexity of the outcome variable is paired with the limitations of analytical methods which can be used based on it. Until the ascension of discrete-choice modelling, caused by the rapid development of computing power and statistical programs, we were bound to less powerful methods such as log linear modelling (van der Eijk et al. 2006). And even with the further establishing of discretechoice models as standard methods the problems did not vanish. Firstly, those models got their own substantive and methodical problems, which are often insufficiently recognized. (van der Eijk et al. 2006: 425). One example would be the lack of independence of the alternatives, where independence is needed to fulfil the criteria for multinomial logistic modelling (Fuchs and Kühnel 1994: 330; Thurner 2006: 195). Secondly, the limits of the dependent variable remain. Especially when it comes to questions relating to contextuality, heterogeneity, and the linking of electoral 2

supply and demand. (van der Eijk et al. 2006: 425) Thinking of multi-party systems again and the majority of Western European Party systems are multi-party systems this is a highly relevant argument. And of course, in times with high volatility when the vote choice becomes more uncertain those flaws become more prevalent. That said van der Eijk et al. (2006) have argued that we need to "rethink the dependent variable in voting behavior" and have proposed a measurement which is based on electoral utility instead of the discrete vote choice, the Propensities to Vote. 4 In our article we critically examine the Propensities to Vote and compare them to another measurement of non-ipsative preferences 5 (Party Preferences) and an instrument, which is proposed as an alternative to the former instruments which measures ipsative preferences; we shall refer to as Vote Distribution. Essentially, the decision which measurement is best suited to explain different research questions thereby has to be both an empirical as well as a theoretical question. We therefore start by reviewing the theoretical framework of utility. We argue that the use of the different measurements might be dependent on what actually shall be explained. Afterwards we present our data and our proposed instrument before we investigate the validity, reliability and limitations of all instruments in comparison. Our conclusions are based on a multi-wave panel study which was carried out in the run up to the Baden- Württemberg state election 2011. 2 Theoretical Framework Most theories of electoral choice, as van der Eijk et al. (2006) argue, distinguish implicitly or explicitly - between utilities and a discrete choice. 6 In such a framework, the final outcome is driven by a two stage process. In stage one the voter assesses the utility of each of the individual alternatives (parties which compete at the election). This stage is denoted as utility function. In the second stage the voter chooses from different alternatives relying on a decision rule. This process is visualised in figure 1. 4 The propensities to vote were presented and analysed in former articles and conference presentations of the authors as well, but the article referred here is the one which brings the ideas behind the propensities to vote together. 5 Ipsative measurements are based on a maximum sum-rule; non-ipsative measurements are (basically) unlimited. Hence, the vote choice (single vote systems) is ipsative, since if you spend your voice on one party you cannot give it to another party. At scalo- or thermometer questions you can rate all parties high or bad, without limiting each other. Van der Eijk and Marsh 2007 compare different non-ipsative instruments and investigate their relationship to the vote cast. 6 Kroh and van der Eijk 2003, van der Eijk et al. 2006 and van der Eijk and Marsh 2007 elaborate on the proof quite extensively. 3

Figure 1: The two stage model of electoral choice Source: Van der Eijk et al. 2006: 428; Van der Eijk and Marsh 2007: 5 The concept of utility 7 is used with different connotations. Whilst utility can be perceived as satisfaction, a subjective state or experience as in the modern analysis of decision making it can also refer to the hedonic quality of experience 8 (Kahneman and Varey 1991). It is necessary to distinguish the different concepts. Kahneman and Varey therefore propose: In referring to Bentham s concept we shall speak of experience utility: the hedonic quality of experience, broadly construed to include satisfaction as well as pleasure. The value associated with a particular consequence in a decision context is its preference utility. The distinction between experience and preference utility suggests another notion, predicted utility: the individual s reflective assessment of future experience utility (Kahneman and Snell, 1990). (Kahneman and Varey 1991: 128) The notions experience utility and preference utility are distinctive concepts, yet linked to each other through predictive utility. One can prefer a political party (preference utility). Voting for this party, however, does not necessarily increase one s individual happiness 7 Tillie s dissertation from 1995 can be rated as stepping stone for the further theoretical foundation of the Propensities to Vote. 8 This is the classical utilitarianism view (Bentham 1975; Mill 1989, 2006). 4

(experience utility). (Tillie 1995: 68) Following Tillie we will apply utility in the following in the sense of preference utility. 9 As can be seen in figure 1 utility is build on determinants of electoral utility and the decision rule then leads to a voting decision. The utility of each of the alternatives therefore is a central determinant of vote choice. Van der Eijk et al. (2006) argue that including utility as a dependent variable in our analysis leads to a better understanding of electoral choice. According to them utility thereby can be measured using the so called Propensities to Vote which have been part of, among other election studies, the Dutch Parliamentary Election Study since 1982. As the authors demonstrate the variable itself is well suited to measure utility and to perform further analysis. However, to replace the dependent variable in voting behaviour, as they propose, we believe further steps are needed on the development of instruments: The Propensities to Vote are imbedded in the electoral process as a rather general utility estimation. This also means that they are detached from the discrete vote choice, since Electoral Utilities are the outcome of the utility function but not as a result of the decision rule. Based on the extant literature, a number of cases are known in which voters do not vote for the party they assign the highest utility to, but decide to choose another party. However, when we come back to what we originally wanted to explain the vote choice this is a major short come. An optimal instrument which tries to explain the voting decision has to take the decision rule into account. With the Vote Distribution question we propose such an instrument. Like the Propensities to Vote this measurement is asked for all parties which take part at the election. But in contrast to existing measures, it is ipsative so that the respondents have to weight their answers on the decisions they made for the other parties. Ultimately they have to think about which party to vote for (to decide) and then rank the other parties, what means that they have to go through the whole electoral process when giving an answer. The Propensities to Vote are also focussed on a broader evaluation since it aims towards a hypothetical will ever vote for. But a measurement which is ought to be used as complement 9 Van der Eijk et al. 2006 make clear that utility in this case as prevalent in psychology also refers to a choice option, which means, that strategic considerations are also covered by the concept. This is important since from an economical point of view utility could also be limited to the affect of an actor for an outcome. (van der Eijk et al. 2006: 427) Furthermore the authors also distinguish constant and random utility. Just like the authors we believe that random utility occurs in the case of vote choice, because vote choice is not always a probabilistic function of utility and therefore the probabilistic aspect is located at stage one. 5

to the discrete vote choice should be focussed on the current election situation to make sure the current situation becomes enlightened. We therefore hypothesize firstly that with the implementation of the ipsative character and the removal of the general term the Vote Distribution measurement will be better suited to explain the vote choice than the Propensities to Vote. Secondly, whilst the general utility evaluation also projects long-term evaluations and influences the new instrument we also expect it to be more unstable during the campaign. To make sure that with the removal of the long-term focus the new instrument does not just project evaluations we furthermore compare both Propensities to Vote and Vote Distribution with one of the established non-ipsative measurements: Party Ratings. Theoretically those Party Ratings are part of the determinants of the utility function (figure 2). Figure 2: Instruments logic at the process of electoral choice Source: Van der Eijk et al. 2006: 428; Van der Eijk and Marsh 2007: 5; allocation of instruments by us. We can derive that therefore, thirdly, Propensities to Vote and Party Ratings should correlate higher than Vote Distribution and Party Ratings and that the explanatory power of the Party Ratings towards the vote choice is the weakest from the instruments. 6

Given that our hypotheses are correct the Vote Distributions would suit as real replacement for the dependent variable vote much better than the Propensities to Vote. However, where utilities are to be explained Propensities to Vote would yield a better instrument. 3 Data and measures For our analysis we make use of a rolling Panel study 10 with a total of six panel waves which was implemented in the run-up to the 2011 Baden-Württemberg state election in Germany. It was focused on processes of opinion formation and decision making of voters in the context of the German multi-level political system. Its sample was drawn from a German non-probably online panel (YouGov Deutschland AG). All in all, 4,800 respondents participated in the panel study, 1,500 of them in all six waves (four waves pre and two waves post election). The field time of the panel survey was between November 2010 and May 2011. The state election took part at 27 th March 2011. The different instruments were implemented in the first waves of the study. However, since the Propensities to Vote and the Vote Distribution questions are very similar and we wanted to compare for possible effects between the questions, not all of the respondents were confronted with both questions: As written above our main focus was to explain processes of opinion formation. We therefore decided to present the Vote Distribution question the question which in our eyes works well to the majority of the respondents: 75% had to answer the Vote Distribution question only. 12.5% got both questions and another 12.5% the Propensities to Vote only. 11 To capture the utilities we used a question similar to the Propensities to Vote question developed by van der Eijk and Niemöller 1984 12 : If you think of all the political parties here in Baden-Württemberg, how probable is it that you will ever give your vote to the following parties at a state Election in Baden-Württemberg? The value 0 means that it is not probable at all that you will ever vote for this party at a state election. The value 10 means that is very probable that you will ever vote 10 The rolling Panel combines elements of a classical Panel Study with the characteristic of a Rolling-Cross- Section study. For further information about the rolling Panel Study Election Study Baden-Württemberg 2011 see Faas and Blumenberg 2012. 11 The number of cases for the different questions can be found in table 1. 12 This and the following questions became translated from German. Please check the appendix for the question wordings used in the survey. 7

for this party at a state election. With the numbers between you may gradate the probability. The only difference is that, as the study was carried out in the run-up to a state election, we decided to limit the probability to the state level as well. Other than that the long-term component and the connotation of probability are included. If respondents assign high numbers to multiple parties no probabilities on the electoral choice can be derived, however, high values for one party with lower values for other indicate in that direction. To capture the Party Ratings another non-ipsative instrument was used. This question was asked to all respondents. What do you think of speaking generally of the political parties here in Baden- Württemberg? Please describe using a scale of -5 to +5 again. What do you think of [list of parties] The question is one of several standard questions which can be found in most questionnaires on voting behaviour. With the range of -5 to +5 it also allows for negative ratings. However, for the following analysis the scores were recoded to range from 0 to 10 so that all three compared questions got the same ranges. As argued above, the Vote Distribution was measured as an ipsative instrument. The respondents were asked to think about a fictive situation when they had 10 instead of 1 vote at the state election. They then had to distribute the votes between the presented parties. This triggers the vote decision process since the respondents most likely want to give most votes to the party they want to vote. They therefore have to bring the parties into an ranking and decide on the relations within the ranking. Voters who are undecided will likely have relatively high and similar scores for two parties, where a clear preference is stated via a very high score for one party. In difference to the Propensities to Vote for this question also a residual category for other parties is needed in cases where not all parties are listed, because otherwise respondents would be forced to give false answers. The wording of the Vote Distribution question was: Let's assume that you were allowed to distribute a total of 10 votes among the political parties at the Baden-Württemberg state election at 27 th March 2011. How would you distribute those votes? 8

If you want to give your entire votes to one party, please type 10 in the field next to the party. If you want to distribute your votes to more than one party please type the designated value in the fields next to the parties. For example 5 at party A, 3 at party B and 2 at party D. The sum of votes must equal 10. If you do not want to give a vote to a party, please type 0. Before we come to the deeper analysis of the association between the instruments 13 and their analytical power we have to consider a technical aspect as well: Presumably especially the Vote Distribution might have been very challenging to the respondents. Although there are already technical ways to display sums automatically we were not able to implement such in our study. Therefore the respondents had to calculate the sums on their own. This might lead to two different problems: Firstly some respondent might miscalculate or, because they do not find the alternatives sufficient, do not want to distribute all their votes (which would be legal in most real election situations). Secondly respondents could just skip the question because they find the task too difficult or too time-consuming. Of course the best question for a questionnaire does not help if it leads to high non-responses. In table 1 we therefore first compare the amounts of undesired respondent s behaviour. As the vote distribution question was carried out as open question so that respondents presumably left out questions if they wanted to distribute no vote to a specific party only a total score (not separately for each party) is presented which sums up the mean of undesired behaviour at this question (non-response and wrong scores). Table 1: Amount of undesired respondent behaviour (wave 4) Vote Distribution Propensities to Vote Party Ratings Mean item non-response 8.1% 1.7% 2.1% N 2,316 653 2,845 We can see that the task to distribute the votes produced a clearly higher amount of undesired respondent s behaviour than Propensities to Vote and Party Ratings. However with 8% of nonresponse the ratio is still in an acceptable range taken the task difficulty. With 44 seconds in median it also took the respondents around 10 seconds longer to fill out the page with the Vote Distribution than the page with the Propensities to Vote (Party Ratings 37seconds). 13 The question of Party Bondings is based on Garry 2007 and reads as following: In Germany some people feel more close to some political parties than to others even though they do not always vote for them. How is that with you? Generally speaking, how close do you feel to the following parties? 9

4 Vote Distribution, Propensities to Vote and Party Ratings in Comparison In the latter we hypothesized that the Vote Distributions are better suited to explain the vote choice than the other instruments, Propensities to Vote and Party Ratings show a closer correlation with each other and that Propensities to Vote are the most stable of the three instruments. We first compare the connection between the instruments and the vote choice. In table 2 the percentages of high scores 14 (score 6 and higher) corresponding to the planned (wave 4 was pre-election) vote choice are displayed. Table 2: Percentages of High Scores corresponding to planned vote choice (wave 4) Vote Distribution Propensities to Vote Party Ratings Greens 91.09% 80.04% 70.52% CDU 94.77% 86.18% 87.17% SPD 93.89% 75.61% 70.48% FDP 97.45% 79.92% 77.98% Left Party 96.76% 90.41% 84.93% As we can see the Vote Distribution consequently performs better at this comparison than the two other instruments. This is no surprise, since with a score of 6 on Vote Distribution it is not possible to give more votes to another party. Remembering that the Party Ratings are part of the determinants of utility the lower percentages here make sense and fulfil the expectations. However, because of the limitations of high scores in this application we decided to run another analysis to confirm our hypothesis. Tables 3 and 4 display the predications of logit regressions where the different instruments (all parties) were used as independent variables. 14 Here ipsative instruments play out another advantage. Whilst we can say that with a vote distribution score of 6 it is certain that a person will vote for a party (simply because the voter cannot give another party a higher score) a high score itself does not indicate a vote choice at the non-ipsative instruments. Van der Eijk et al. defined a utility score of 6 as being high and compared the percentage scores with the percentages for the vote distributions from their sample (van der Eijk et al. 2006: 435). This threshold score is used by van der Eijk and Marsh 2007 as well. Whilst it would be possible to calculate a breaking point score on which the highest score can be defined we aim for comparison and therefore use six as well benefiting from the advantage that this score can be used for all of our measurements. Since not all of our respondents were questioned with the same instruments, we do not carry out the same analysis van der Eijk et al. 2006 use but just compare the accordance of a high score with the vote choice. 10

Table 3: Correct classifications in wave 1 (logit regressions; dv: planned vote choice) Greens 15 CDU SPD FDP Left Party Vote Distribution 96.74% 97.91% 96.89% 98.42% 98.16% Propensities to Vote 90.94% 94.82% 94.64% 95.93% 96.49% Party Ratings 86.70% 91.47% 87.57% 95.14% 94.09% Table 4: Correct classifications in wave 4 (logit regressions; dv: planned vote choice) Greens CDU SPD FDP Left Party Vote Distribution 94.83% 97.82% 95.95% 98.01% 97.69% Propensities to Vote 90.63% 91.46% 89.17% 95.63% 97.29% Party Ratings 83.43% 90.14% 83.22% 94.83% 94.03% The classification tables thereby confirm the result above. Vote Distribution predicts (and therefore is connected to) the planned vote choice consequently better than the other instruments. Nonetheless the real advantage of the alternative measurements can be seen in their power to explain alternative votes and voting under highly competitive circumstances. Table 5 displays the results of a logit regression with the answer on an alternative vote choice (post-election) as dependent variable. 16 Table 5: Correct classifications in wave 4 (logit regressions; dv: alternative party choice (wave 5)) Greens CDU SPD FDP Left Party Vote Distribution 68.07% 86.34% 78.89% 91.49% 87.50% Propensities to Vote 63.41% 83.70% 80.80% 90.94% 92.75% Party Ratings 66.55% 84.83% 78.88% 92.01% 87.67% Even though the classifications are weaker under this circumstance, the Vote Distribution scores at least for the big parties still higher than the other instruments. We therefore see our first hypothesis confirmed: Vote Distribution is better suited to explain voting behaviour than the other instruments, because it mirrors the voting behaviour directly to multiple parties. 15 In Germany usually CDU and SPD are denoted as the two big parties. However, at the state election in 2011 the Greens became second strongest party of the parliament and now head a coalition between Greens and SPD. Therefore the Greens count as big party In this context. 16 Amount of correspondence between second highest score and alternative vote choice also have been tested, giving the same result. 11

Table 6: Correlation between Party Ratings and Vote Distribution/Propensities to Vote in wave 1 Vote Distribution Propensities to Vote Greens 0.664 0.822 CDU 0.753 0.839 SPD 0.563 0.677 FDP 0.525 0.687 Left Party 0.591 0.797 All correlations are highly significant. Table 7: Correlation between Party Ratings and Vote Distribution/Propensities to Vote in wave 4 Vote Distribution Propensities to Vote Greens 0.632 0.740 CDU 0.764 0.799 SPD 0.582 0.661 FDP 0.510 0.713 Left Party 0.584 0.731 All correlations are highly significant. We furthermore hypothesised that if the Vote Distribution really is located after the second step of the election process there must be a higher connection between the Propensities to Vote and Party Ratings than it is between Vote Distribution and Party Ratings. In tables 6 and 7 the results of pairwise correlations between the different instruments are displayed. Here especially for the smaller parties the expectations are fulfilled. High Propensity to Vote scores are connected to high Party Ratings. This is the same for the Vote Distributions but much weaker than for the latter. Still the values of correlations between Vote Distribution and Propensities to Vote are higher than those between Vote Distributions and Party Ratings. We see our third hypothesis to be confirmed. All three measurements take different places in the process of electoral choice. But as replacement of the vote choice the ipsative Vote Distribution works best. Table 8: Correlations between Vote Distribution and Propensities to Vote Correlation Greens 0.796 CDU 0.848 SPD 0.752 FDP 0.650 Left Party 0.684 All correlations are highly significant. 12

So far we left out to compare the stability of the instruments. We argued that the Party Ratings most probably are the most unstable and the Propensities to Vote the most stable of the instruments. This is because Party Ratings and Vote Distribution both have a much higher situational character in their question wording as well as in their theoretical assumptions. Since the instrument Propensities to Vote also inherits long-term components (in the utility function) we expected it to be more stable. Table 9: Stability of the instruments between wave 1 and wave 4 (3 respectively 5 for Party Boundings) VD PTV PR PB (3) PB (5) Greens No Difference 50.57% 41.25% 34.81% 39.09% 35.57% Difference <2 69.27% 61.87% 65.46% 64.64% 60.92% CDU No Difference 65.33% 50.82% 41.57% 50.42% 51.28% Difference <2 77.29% 70.16% 69.85% 72.44% 72.18% SPD No Difference 53.30% 35.25% 32.24% 37.18% 35.62% Difference <2 70.63% 56.83% 61.27% 63.16% 61.64% FDP No Difference 73.42% 46.63% 40.39% 51.27% 52.74% Difference <2 85.24% 66.83% 67.80% 71.99% 73.56% Left Party No Difference 76.65% 58.45% 47.41% 58.16% 55.50% Difference <2 87.97% 75.85% 68.82% 72.28% 72.94% Based on a first comparison between the scores of wave 1 and 4 we were not able to confirm this hypothesis. While Party Ratings, as expected, is the most unstable instrument, Vote Distribution is more stable than Propensities to Vote. Reasons for this might once again lie in the ipsative characteristic of the instrument. High scores for the preferred party leave less space for variation among the others. This is different for the other scales where free variations are possible. Further investigation of the theoretical assumption is therefore needed. 5 Discussion It became clear that those three instruments cannot be used interchangeable but complementary. Each of the instruments represents a specific place within the process of electoral choice. Dependent on the research question it is advised to use a different instrument: Questions which explicitly try to model the first stage of the electoral choice process, namely the utility function, work best with the Propensities to Vote. Their non-ipsative character paired with a relative openness to different determinants of party support fosters this usage most. Nonipsative instruments of party evaluation like the Party Preferences are part of the determinants of party utility and come therefore into play even before the first stage. Using them as 13

dependent variable to explain vote choice is therefore not recommendable. However, they build an important factor in many voting models and the ratings of parties are an interesting field on its own. Thirdly we proposed to use an instrument of Vote Distribution. As analysis demonstrated this instrument is best suited to model the second stage of the process, the real voting decision. For research questions which aim to explain vote choice or the voting decision process this might therefore be the most favourable instrument to implement. 6 References Bentham, J. (1975) Einführung in die utilitaristische Ethik. Klassische und zeitgenössische Texte. München: Beck. Faas, T. and Blumenberg, J.N. (2012) Die Vermessung der Dynamik: Eine rollierende Panelstudie im Vorfeld der baden-württembergischen Landtagswahl 2011 (forthcoming.) Fuchs, D. and Kühnel, S. (1994) 'Anmerkungen zum Nutzen des Rational-Choice-Ansatzes in der empirischen Wahlforschung', in M. Kaase and H.-D. Klingemann (eds.) Wahlen und Wähler. Analysen aus Anlass der Bundestagswahl 1990, Opladen: Westdeutscher Verlag, pp. 305 364. Garry, John (2007) 'Making party identification more versatile: Operationalising the concept for the multiparty setting', Electoral Studies 26(2): 346 358. Kaase, M. and Klingemann, H.-D. (eds.) (1994) Wahlen und Wähler. Analysen aus Anlass der Bundestagswahl 1990. Opladen: Westdeutscher Verlag. Kahneman, D. and Snell, J. (1990) 'Predicting Utility', in H.J. Einhorn and R.M. Hogarth (eds.) Insights in decision making. A tribute to Hillel J. Einhorn, Chicago: University of Chicago Press, pp. 295 310. Kahneman, D. and Varey, C. (1991) 'Notes on the psychology of utility', in J. Elster and J.E. Roemer (eds.) Interpersonal comparisons of well-being, Cambridge [England] ;, New York: Cambridge University Press, pp. 127 163. Mill, J.S. (1989) On liberty;. With The subjection of women ; and Chapters on socialism. Cambridge ;, New York ;, Port Chester [etc.]: Cambridge University Press. Mill, J.S. (2006) Utilitarianism. Stuttgart: Reclam. Thurner, P.W. (2006) 'Modelle diskreter Entscheidungen', in J. Behnke, T. Gschwend, D. Schindler and K.-U. Schnapp (eds.) Methoden der Politikwissenschaft. Neuere qualitative und quantitative Analyseverfahren, Baden-Baden: Nomos, pp. 193 204. Tillie, J. (1995) Party Utility and Voting Behavior. Amsterdam: Het Spinhuis. van der Eijk, C., van der Brug, W., Kroh, M. and Franklin, M. (2006) 'Rethinking the dependent variable in voting behavior: On the measurement and analysis of electoral utilities', Electoral Studies 25(3): 424 447. 14

van der Eijk, C., Marsh, M. (2007) Don t expect me to vote for you just because I like you, even if you do make me feel warm inside Paper presented at Annual meeting oft he American Political Science Association, Chicago. Appendix Original Question Wordings Vote Distribution Einmal angenommen, Sie könnten bei der baden-württembergischen Landtagswahl am 27. März 2011 insgesamt zehn Stimmen auf eine oder mehrere Parteien verteilen. Wie würden Sie diese Stimmen dann verteilen? Wenn Sie einer Partei alle Ihre Stimmen geben wollen, dann tippen Sie bitte die Zahl 10 in das Feld neben der Partei. Wollen Sie mehrere Parteien berücksichtigen, so tragen Sie die jeweilige Zahl in das Feld neben der Partei ein. Zum Beispiel 5 bei Partei A, 3 bei Partei B und 2 bei Partei D. Die Summe der Stimmen muss 10 ergeben. Möchten Sie einer Partei keine Stimme geben, so tragen Sie bitte die Zahl 0 ein. Propensities to Vote Wenn Sie einmal an die verschiedenen politischen Parteien hier in Baden- Württemberg denken, wie wahrscheinlich ist es für Sie, dass Sie die einzelnen Parteien jemals bei einer Landtagswahl in Baden-Württemberg wählen werden? Der Wert 0 bedeutet dabei, dass es für Sie sehr unwahrscheinlich ist, diese Partei jemals bei einer Landtagswahl zu wählen. Der Wert 10 bedeutet, dass es sehr wahrscheinlich ist, dass Sie diese Partei jemals bei einer Landtagswahl wählen werden. Mit den Werten dazwischen können Sie die Wahrscheinlichkeiten abstufen. Party Ratings Was halten Sie einmal ganz allgemein gesprochen von den politischen Parteien hier in Baden-Württemberg? Bitte beschreiben Sie dies wieder mit einer Skala von -5 bis +5. Was halten Sie von Party Bounding In Deutschland fühlen sich viele Menschen einigen Parteien mehr verbunden als anderen, auch wenn sie diese nicht immer wählen. Wie ist das bei Ihnen: Wie verbunden fühlen Sie sich ganz allgemein gesprochen mit den folgenden Parteien? 15

Distribution of highest and second highest score among the instruments (wave 4; 3 for Boundings) Vote Distribution 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hvd1 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hvd2 16

Propensities to Vote 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hptv1 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hptv2 17

Party Ratings 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hpr1 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hpr2 18

Party Bounding 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hpn1 50 40 30 Percent 20 10 0 0 1 2 3 4 5 6 7 8 9 10 hpn2 19