Analysing Party Politics in Germany with New Approaches for Estimating Policy Preferences of Political Actors

German Politics ISSN: 0964-4008 (Print) 1743-8993 (Online) Journal homepage: http://www.tandfonline.com/loi/fgrp20 Analysing Party Politics in Germany with New Approaches for Estimating Policy Preferences of Political Actors Marc Debus To cite this article: Marc Debus (2009) Analysing Party Politics in Germany with New Approaches for Estimating Policy Preferences of Political Actors, German Politics, 18:3, 281-300, DOI: 10.1080/09644000903055773 To link to this article: https://doi.org/10.1080/09644000903055773 Published online: 09 Sep 2009. Submit your article to this journal Article views: 1000 View related articles Citing articles: 5 View citing articles Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalinformation?journalcode=fgrp20

Analysing Party Politics in Germany with New Approaches for Estimating Policy Preferences of Political Actors MARC DEBUS This contribution introduces the development of German party competition and coalition politics in recent decades on the one hand and the latest techniques for estimating the preferences of political actors on the other. It argues that, for reasons of social change, the preferences of the electorate were reshaped and political actors therefore had to adopt programmatic changes. In addition, the establishment of a new socialist party in the eastern German states, and since 2005 also in the western part of Germany, resulted in new coalition strategies for the traditional political parties, which are discussed between and within the parties. Researchers can apply various methods to test what programmatic position political actors adopted and whether they changed their programmatic orientation over time. The paper discusses the advantages and drawbacks of the major strategies in estimating preferences of political actors and provides an overview of the contributions to this issue of German Politics. INTRODUCTION Patterns of the party system, party competition and coalition politics in Germany have changed significantly during recent decades. 1 One major reason is the unification of western and eastern Germany and the emergence of two different party systems. While in the western part of the country the four-party system persisted with Christian Democrats (CDU/CSU) and Social Democrats (SPD) as the two larger political actors and the Liberals (FDP) and the Green Party (since 1993: Alliance 90/Greens, Bündnis 90/Grüne) as the two less represented parties, the Eastern German states developed a three-party system with three strong actors: CDU, SPD and the socialist PDS as the successor of the former governing party of the German Democratic Republic (GDR). These different constellations of party systems resulted in various coalition options and strategies for the parties. A second explanation is processes of social change 2 that led towards, first, the emergence of more mobile voters that have no or only shaky long-standing party attachments 3 and, secondly, in new conflict lines that produced new political parties such as the Greens or, more recently, Labour and Social Justice The Electoral Alternative (Wahlalternative Arbeit und soziale Gerechtigkeit, WASG), which merged with the PDS into the new party Die Linke in July 2007. 4 The emergence of the WASG and the SPD s difficulties in mobilising their core voter clientele unionised labour workers and employees during the red green government led by Chancellor Gerhard Schröder, which was in office German Politics, Vol.18, No.3, September 2009, pp.281 300 ISSN 0964-4008 print/1743-8993 online DOI: 10.1080/09644000903055773 # 2009 Association for the Study of German Politics

282 GERMAN POLITICS from 1998 until 2005, sheds light on the emergence and relevance of new issues. Moreover, the positions of voters and parties have changed on classical conflict issues such as economic and social affairs. A number of SPD members left their party most prominently the former party chair Oskar Lafontaine and became members or supporters of the WASG or the Linke. They argued that the Social Democrats changed their programmatic profile in such a way that it no longer represents the interests of lower income groups. 5 As the short description of the development of the SPD s internal struggles and the foundation of a new left-wing party in Germany indicates, German parties seem to have developed a new programmatic profile in the 1990s and in the first decade of the twenty-first century. This is not only because of the appearance of a new leftwing party, but also because of changes in the profile and preferences of the electorate. To test the existence or the degree of programmatic shifts as well as theories of voting behaviour, party competition, coalition politics and legislative behaviour, political scientists need data on the policy preferences of political actors. The aim of this special issue of German Politics is to present recent methods of content analysis and to apply them to the German case. This introduction will provide an overview of the development of party politics in Germany in recent decades. In a second step, it gives a survey of the latest approaches on estimating policy preferences of political actors. Thirdly, it introduces the articles of this volume and their contribution to the analysis of party politics in Germany on the one hand and to the methods of extracting policy preferences of political actors on the other. PARTY POLITICS IN GERMANY IN RECENT DECADES Party politics in Germany were and still are strongly influenced by the aftermath of unification, the different degrees of quality of life in western and eastern Germany and the crisis of the welfare state and its reforms. 6 The aspect mentioned first resulted in the strength of the PDS in the five eastern German states. 7 Different levels of wages, a greater share of unemployed people compared to the western part of Germany and the sense of being second-class citizens strengthened the ruling party of the GDR in elections on the local, state and federal levels. The Socialists became the third or in states like Brandenburg, Saxony and Thuringia the second largest party in eastern Germany because of their ability harness the discontent many voters felt with the new economic system and welfare state policies. 8 Since the end of the 1980s right-wing extremist parties also made use of dissatisfaction with the economic situation and issued immigration and asylum policies to win votes in elections. While none of these parties Republicans (REP), German Peoples Union (DVU) or National Democrats (NPD) received more than 5 per cent in Bundestag elections, they won parliamentary representation in a number of eastern and western Länder. 9 The REP, for instance, received a vote share of 10.9 per cent in the Baden-Württemberg state elections in 1992 and 9.1 per cent in the elections held four years later. In Saxony and Mecklenburg-West Pomerania the NPD won 9.2 and 7.3 per cent of the vote share in 2004 and 2006, respectively. Populist parties such as the Law and Order Offensive Party (Partei Rechtstaatliche Offensive, PRO) won 19.4 per cent at the

ANALYSING PARTY POLITICS IN GERMANY 283 Hamburg state elections in 2001 and became a member of the new coalition government, which, however, disintegrated due to personal conflict three years later. 10 While parties from the far right were not able to win continuous representation, the PDS and its successor party, the Linke, were able to promote the image of an eastern German interest party with a solid base of voters. Due to merger with the WASG in 2005, the welfare state reforms of the second Schröder government and the grand coalition between CDU/CSU and SPD led by Chancellor Angela Merkel (CDU), the Linke was also able to win parliamentary representation in states in the western part of Germany. 11 The new party won more than 5 per cent of the votes in Bremen in 2007 as well as in Hamburg, Hesse and Lower Saxony in 2008. Despite internal struggles, the Linke was re-elected in Hesse in the January 2009 elections. In the 2005 federal election, the strength of the PDS.Linke, which received 8.9 per cent of the vote, prevented a majority for the camp of CDU/CSU and FDP or SPD and Greens. The parties of the German two and a half party system 12 Christian and Social Democrats as well as the FDP reacted programmatically to the emergence of the Green party at the beginning of the 1980s and to the changes in the social structure from an industrial economy towards a service economy. Social Democrats copied the third way of Tony Blair s New Labour party and adopted a more centrist ideological position in the election campaign of 1998. 13 The background behind this strategy was to increase the incentives for voters from the middle class, which have no or only a weak party identification, to switch to the Social Democrats. While the share of traditional groups in the electorate that traditionally support social democratic or socialist parties has continuously decreased since the 1960s, 14 political actors from this party family have to reach out for other groups in the electorate. One possibility is a change in the overall programmatic profile, 15 which, however, carries the risk of losing support among the traditional backers of these parties. 16 In the case of Germany, it was not only the Social Democrats that underwent this programmatic change, but also the Greens as the SPD s coalition partner at the federal level. The latter accepted the participation of German military in international peacekeeping missions, which was clearly against their traditional pacifist leaning in foreign and defence policy and resulted in strong intra-party conflict between the left and right wing of the party. 17 Furthermore, the Greens in the federal government promoted reform of the welfare state more vigorously than their social democratic coalition partner. 18 During Chancellor Schröder s second term between 2002 and 2005, the red green government implemented the so-called Agenda 2010, which drastically changed Germany s welfare state and resulted in fierce conflicts inside the SPD. 19 Schröder stepped down as the party chair in 2004. Wolfgang Clement, the then acting minister of economic and labour affairs and former prime minister of North Rhine-Westphalia, faced sharp criticism during and after his term of office. Because of his suggestion not to vote for the SPD in the state election in Hesse in 2008, 20 party officials started a campaign for Clement s exclusion. Because of his many critics Clement the strongest advocate of Schröder s economic reforms decided to leave the Social Democratic party in the autumn of 2008. In contrast to the parties mentioned last, the perceived degree of programmatic change inside both Christian democratic parties and the Liberals was less from the beginning of the 1990s. While the FDP stressed its market economic profile and

284 GERMAN POLITICS played down its progressive policy positions in social, interior and justice issues, the Christian Democrats stressed their conservative profile in societal affairs by bringing issues like immigration and the Leitkultur to the top of the political agenda during their time in opposition. After becoming the strongest party in government in the grand coalition with the Social Democrats in 2005, the CDU led by Chancellor Angela Merkel implemented progressive policies in family, youth and women s policies under the direction of Minister Ursula von der Leyen (CDU). CHANGING PATTERNS OF COALITION POLITICS As already indicated in the previous section, the changes in the order of the party system resulted not only in a larger number of coalition options, 21 moreover, some party combinations such as a coalition between Christian Democrats and Greens became more likely during the 1990s. While the first black green coalition was formed in Hamburg in 2008, even in 1996 and 2001 the CDU in Baden-Württemberg had thought of the black green option as realistic. In the following, I briefly discuss the developments in coalition strategies with a focus on the three smaller parties. The FDP lost its numerically pivotal status in 1983 because the Greens won parliamentary representation at the federal level for the first time. While the results of expert surveys 22 on programmatic policy positions of German parties do not indicate a loss of the ideologically pivotal position of the Liberals due to the existence of the Greens, which were considered until the end of the 1980s by all other parties including the SPD as incapable of participating in government, the blackmail potential of the Liberals decreased significantly. After the establishment of the PDS as a parliamentary player, the chances that SPD and FDP could win a parliamentary majority decreased further. Only in 1998 did both parties win a majority of seats, but the SPD preferred a coalition with the Greens, whereas the Liberals preferred a coalition with the CDU/CSU rather than with the Social Democrats since members of the FDP s libertarian wing left the party after its decision to switch coalition partners from the SPD to the Christian Democrats in autumn 1982. 23 The emergence of the ecological movement and thus the success of the Green party had its roots in the changing social structure and the emergence of the materialist/ post-materialist conflict dimension. 24 As already indicated, in the 1980s the Greens a priori excluded themselves from the coalition formation game by preferring to work in opposition. Before 1990, however, the SPD was also not in favour of a red green coalition at the federal level. Not until their defeat in the 1990 Bundestag election did the Greens adopt more moderate positions and move away from their selfidentification as an anti-system party. 25 Without other options, the SPD now accepted the Greens as its ally against the ruling Christian-liberal government. While red green alliances at the state level during the 1980s were characterised by coalition crises and finally the collapse of these state governments, in the 1990s none of the red green coalitions which were formed gave up office early for reasons of policy conflict. 26 Yet two state governments where the SPD, Greens and FDP were involved broke down for reasons of policy conflict: the Greens left the coalition governments in Brandenburg and Bremen in 1994 because of major disagreements with their coalition partners mainly on economic issues. 27 This suggests that a traffic light coalition was

ANALYSING PARTY POLITICS IN GERMANY 285 infeasible not only at state level but also at national level during the 1990s. Yet despite the existing CDU/CSU FDP coalition the three socially progressive parties worked together to sign liberal abortion legislation into law in 1992. The PDS as the fifth parliamentary player was excluded from any coalition government at the federal level because of its history as the successor of the East German Communists. 28 Because of its parliamentary strength in eastern Germany, it was difficult to completely ignore the PDS in the coalition formation game. With the FDP and Greens often outside parliament in the five East German states, only coalitions between the CDU and SPD were possible. In 1994 and 1998, the Social Democrats in Saxony- Anhalt formed a minority government which was supported by the PDS. Also in 1998, both parties formed a coalition in the state of Mecklenburg-West Pomerania. In addition, the SPD and PDS formed a coalition government in Berlin in 2001, which was renewed after the 2006 elections. Yet before every election on the federal level the SPD rejected any cooperation with the socialists. For reasons of dissatisfaction with the economic and welfare reforms of the red green government, left-wing Social Democrats and a number of labour union activists founded the new left-wing party WASG in 2004. Before the federal elections held in September 2005, the WASG merged with the PDS to form the Linkspartei.PDS and renamed itself the Linke in 2007. 29 All other parties rejected this new party alliance as a potential future coalition partner. Since the takeover of the grand coalition in November 2005 and the strength of the Linke in a number of state elections in western Germany, however, the patterns in German coalition politics became fluid again. After the state election in Hesse in January 2008, the leadership of the SPD state party favoured the formation of a red green minority government tolerated by the Linke. This strategy was clearly contrary to the SPD s pre-electoral statement to reject cooperation with the new left-wing party and resulted in conflict within the SPD in Hesse and a call for fresh elections in January 2009. It also caused discussions in the federal party regarding its future coalition strategies. After the resignation of Kurt Beck as the party chair in September 2008, who gave the SPD in Hesse the green light for their plans to form a red green red coalition government, the new party leader Franz Müntefering preferred a complete rejection of the Linke as a coalition partner in September 2008, but then changed his mind in December of the same year and rejected the Socialists as a potential coalition partner on the federal level only. 30 This shift in the Social Democrats coalition strategy could have a significant impact not only on the outcome of the government formation process in Thuringia and the Saarland which will take place after the elections in August 2009, but also on the voting behaviour in the next Bundestag elections, one month later. Summing up the description of the development of party competition and coalition politics in the unified Germany, it turns out that the stable four-party system disappeared not only because of the strong position of the PDS in the eastern states, but also because of the demand for a clearly left-wing party which would emphasise welfare issues in western Germany since the implementation of the economic reforms of the second Schröder government between 2002 and 2005. This development led towards new coalition options such as alliances between Christian Democrats and Greens or between Social Democrats, the PDS and the Linke, respectively. To test whether (1) the described shifts in the programmatic positions of parties really existed

286 GERMAN POLITICS and (2) how the diverging positions of parties and/or intra-party groups on various policy areas affect aspects like voting behaviour, coalition formation or legislative activity, political scientists need data on the preferences of political actors. To extract this kind of data, various methodological approaches exist. The following section introduces the most prominent techniques and discusses their respective pros and cons. METHODS FOR ESTIMATING POLICY PREFERENCES OF POLITICAL PARTIES There are plenty of methods for estimating the programmatic positions of political actors. 31 For the sake of simplicity, the following sections differentiate between three main approaches. First, an analysis of the legislative voting behaviour of politicians enables us to locate political actors on at least one policy dimension. Second, a number of studies reconstruct programmatic positions of political parties by conducting elite or mass surveys, so that the ideological party position is measured by the mean value of party activists or likely supporters. Third, a number of studies apply a content analysis to party policy documents or speeches of politicians. The following subsections sketch the basic proceedings of these approaches as well as their advantages and disadvantages. Analysis of Roll Call Voting The most straightforward approach to measuring the programmatic positions of political actors is to analyse their voting behaviour in parliament. By doing so, one can identify the number of relevant dimensions that structure parliamentary decision-making and locate the members of parliament on the extracted policy dimensions. While the most prominent example of an application of this method is the analysis of roll call votes in the United States Congress by Poole and Rosenthal, 32 there are also a number of studies that use recorded votes data to extract political actors policy positions in the European Union parliament, 33 the European Council 34 or the United Nations general assembly. 35 Furthermore, there are analyses of recorded votes in the Fourth Republic of France as well as in the British lower house between 1841 and 1847. 36 In the case of Germany, roll call votes have been used to measure the positions of members of the Frankfurt national assembly in 1848 and 1849 on a left right scale, 37 the cohesion of parliamentary parties in the Weimar Reichstag 38 and in the Bundestag. 39 Shikano 40 used Markov chain Monte Carlo simulations 41 and estimated the positions of German state governments from 1949 on the basis of their voting behaviour in the Bundesrat. Likewise, Debus and Hansen 42 estimated the dimensionality of the Reichstag during the Weimar Republic and located each MP on the two extracted dimensions. This example shows that it is possible to analyse the internal heterogeneity of each parliamentary group and, thereby, relax the often criticised unitary actor assumption. 43 However, besides the advantages of such a research design, there are also some disadvantages in applying programmatic positions extracted from roll call data when analysing party competition or coalition politics. One reason, for instance, is that the degree of policy assertion potential for one point in time, e.g. after the coalition formation process, could not be adequately measured. This is because the programmatic position of each legislative actor would be measured

ANALYSING PARTY POLITICS IN GERMANY 287 over time, so that it is only possible to calculate policy assertiveness at the end of a term and not at the beginning. Subsequently, use of recorded votes does not allow for a measurement of the coalition s policy position independently of the respective position of the involved parties and politicians. In addition, preferences of political actors estimated on the basis of roll call data could be biased because they cover only a small share of all parliamentary votes and even more importantly roll call votes could be used to discipline MPs in crucial votes. 44 In parliamentary democracies with strongly disciplined parties, an analysis of roll call votes will result in the estimation of a conflict line between the government and opposition camp rather than an identification of policy dimensions inside the parliament. 45 In the case of the European Parliament (EP), however, Gabel and Hix can show that a strong relationship between the policy preferences of each MP and his voting behaviour in the parliament exist. 46 The authors use survey data on MP preferences and combine it with the positions of MPs on the two major dimensions of the EP, that is, the left right axis and the conflict between the stances for or against European integration. 47 Position Extraction from Elite Studies and Regular Surveys Another possibility of extracting the programmatic positions of political actors is to use survey data. In this approach, the basic assumption is that in the case of regular surveys the party supporters or, if elite surveys are used, the party leadership, reflect the ideological position of the respective party best. Well known examples of the first type are the studies by Kitschelt, where internationally comparable survey data from the Eurobarometer 48 and the World Values Study 49 were used to estimate the positions of parties in West European countries. There are also studies with data on programmatic positions of parties or more generally political actors that are based on surveys that focus on special groups like party members or party leaders, both on the middle and the top level of party organisation. 50 More recent studies on the determinants of voting behaviour also refer to data from regular surveys or election studies to place political actors and voters into one common policy space. Adams, Merrill and Grofman applied data from the American, British, French and Norwegian election studies to estimate the left right position and specific policy-area positions of parties and voters and thus to evaluate their model on voters decision making. 51 Another alternative would be to ask the members of parliament either on the subnational, national- or supra-national level about their policy positions. 52 Both accounts have advantages and disadvantages. One clear advantage when selecting regular survey data to estimate the positions of political actors would be the large number of available survey datasets, either on the national or the international level. Consider for example the Eurobarometer surveys: they have been conducted every year since 1973. Each survey includes a variable that allows for the estimation of the parties position on a single left right dimension, so that a valid and reliable cross-national analysis of ideological party placements over time is possible. Problems will arise, however, when the estimation of party positions on more than one dimension is required for evaluation of the respective hypotheses. Furthermore, another requirement for the adherence of validity and reliability would be that the content of the question that asks for the respective respondents attitude towards their policy position stays the same over time and has the same meaning in the countries included in the survey

288 GERMAN POLITICS sample. 53 Otherwise it would not be possible to extract valid and reliable comparable data on party positions over time. A more theoretically driven criticism concerns the assumption that the respondents mean positions reflect the parties true ideological position. It could be the case that the respondents perception of the respective parties position is correct. However, the possibility also exists that political parties in some policy areas of minor importance have programmatically different positions than their voters. The last mentioned aspect would be of less relevance for an analysis of voting behaviour, but would have considerable implications when analysing patterns of party competition and coalition politics. Expert Surveys There are a number of clear advantages, but also some disadvantages to using expert surveys for estimating the policy positions of political actors. Most expert surveys estimate the positions of parties on an overall left right dimension only. 54 In a more recent expert survey, Huber and Inglehart criticised one-dimensional approaches and called for the inclusion of other policy dimensions. 55 However, besides the fact that they invited political experts to name the most important dimension and to add a second one if necessary, in their analysis of party positions Huber and Inglehart returned to the single left right axis. The reason was that in all countries included in their sample an overwhelming majority of respondents identified the socio-economic left right dimension as the most important one, whereas there was no unanimity when identifying the second most important area of conflict. Beside the Laver and Hunt and the Benoit and Laver expert surveys 56 only studies by Warwick and by Marks et al. provide party policy positions on more than a general left right dimension. 57 The expert surveys conducted by Laver and Hunt and Benoit and Laver include a fixed set of policy dimensions for each country. Not only the positions of parties, but also the dimension saliencies for each relevant party are measured. This research design, however, has been criticised by Warwick, who argues that the large number of detailed policy areas given in the Laver and Hunt survey do not allow for an estimation of a general picture of party competition. 58 According to Warwick, the Laver and Hunt categories are overly detailed and too specific for handling them as broader policy dimensions. Warwick refers to them as a set of issues rather than a set of dimensions. 59 On that basis, Warwick applied factor analysis to the Laver and Hunt data. The result was that three dimensions a socio-economic left right one, a libertarian authoritarian social policy dimension and a new politics reflecting materialist post-materialist conflict structure party competition. When conducting a new expert survey, Warwick takes these implications into account and asks for the party s position and salience only on these three dimensions. 60 Having described the advantages of expert surveys, there are also a number of problems that arise when using expert survey data to analyse party competition and coalition politics over time. The first aspect is of a methodological nature. One may ask who are experts in locating parties on policy dimensions and, furthermore, what should be done when only a few experts respond. 61 Another and more important problem is that expert surveys are temporally stable and cannot account for variations in the respective party policy positions as well as for potential changes in the partyspecific dimension saliencies. If we take the results of the Laver and Hunt as well as

ANALYSING PARTY POLITICS IN GERMANY 289 the Benoit and Laver studies seriously, the programmatic orientation and issue saliency of parties did indeed change between the time periods in which both expert surveys were conducted. Consider, for example, the economic policy position of the German Greens. While they were ranked as a left-wing party in 1989, before the 2002 federal election the Greens were located alongside the SPD, which also moved towards a more liberal position. This is hardly a measurement error, according to the development of the German Greens and Social Democrats since the beginning of the 1990s. An alternative that accounts for the specific time-period positioning of parties would be to request the experts to position parties not only for one, but for several points in time. Such a design was applied in the case of Austria, where the party s position was estimated both retrospectively and prospectively. 62 While one might question whether an estimation of future positions of parties makes any sense, the remaining results do not show any significant moves of the Austrian parties on the left right dimension. This could be indeed a true result, but it could also be the case that even for experts on party politics it would be too complicated to estimate the position of parties in 1991 when asked about the situation in 1975. For this reason, such expert surveys have to be interpreted with caution. An alternative to data collection via surveys, regardless of whether they are based on the voting age of the population, some sort of elite or expert opinion, can be found in an analysis of policy documents. I will give an outline of such approaches in the next subsection. Hand-Coded Analysis of Policy Documents In comparison to the other alternatives mentioned here, the main advantage of an analysis of policy documents can be seen in the high degree of their availability. Before an election, nearly every party or party alliance publishes a programme for government in which its goals for the next legislative period are outlined. In a nutshell, election manifestos have the advantage that they are published before each election. Moreover, because election programmes normally have to be passed by a party congress or at least by a wider group of party elites, they should reflect more or less the mean of all intra-party groups weighted by their importance. Another aspect is relevant: the programmatic statements inside such pre-election programmes can be used as a starting point for future coalition negotiations and as a point of reference for the policy assertiveness in a coalition government formed later. 63 Generally speaking, two possibilities allow for an empirical analysis of policy documents. 64 They are, first, an approach that is based on a manual coding scheme. Such a procedure is associated with the work of the Manifesto Research Group (MRG), which has been known since 1989 as the Comparative Manifesto Project (CMP). 65 The other variant is more recent from a chronological perspective and is a computer-aided procedure which, however, still requires the programming of a dictionary that contains some a priori defined signal words. 66 The work of the MRG and the CMP, respectively, resulted in the largest and most complex database that includes party saliencies on 56 policy issues. Despite the large number of 3,018 covered election manifestos from 54 countries since 1945, the approach of this research group was criticised for both theoretical and empirical reasons. First, the MRG s coding instructions are based on saliency theory. 67 This theory assumes that parties formulate issue saliencies rather than policy positions in

290 GERMAN POLITICS their election programmes. Therefore, most of the 56 MRG and CMP categories only include information about favourable mentions and no negative emphasis on various policy issues. Only 12 of the 56 categories were coded bipolar, so that one is able to distinguish between negatively and positively formulated issue saliencies in the respective election manifestos. 68 In an empirical test, Budge 69 finds evidence for the assumptions of saliency theory: only one negative coded category has a similar share of coded quasi-sentences 70 as the respective positive category. 71 However, saliency theory and thus the party positions estimated by the MRG and CMP came under attack. Laver argues that parties indeed formulate positions in their programmes and apply different saliency weights for each policy issue independently of the respective positions. 72 Besides these more theoretically oriented reviews, a number of methodological and empirical points of criticism also appeared. To start with the aspect mentioned first, it is quite obvious that purely manual coding will result in problems of both validity and reliability. One major requirement for a successful cross-national comparative study is the aggregation of the information level on both dependent and independent variables, so that comparisons between countries or over time are possible. Therefore, it is a necessary evil to restrict the number of policy categories. A number of problems arise from this requirement. Besides existing and well-founded coding instructions, 73 it seems possible that coders in all participating countries could have different meanings for the classification of quasi-sentences into one of the 56 categories. In an experimental study, Mikhaylov, Benoit and Laver 74 show that even trained coders of the CMP project allocate quasi-sentences not to the same or even correct policy category, but rather to various others of the 56 remaining policy categories. The last-mentioned argument gains importance when thinking about findings on equivalence problems in comparative social science studies on the one hand and the coding and policy-issue scheme of the MRG/CMP, which is stable for the whole time period, on the other. 75 Consider, for instance, the meaning of the categories multiculturalism: positive and multiculturalism: negative in the case of the Netherlands. While at least until the end of the 1970s those categories include the party issue saliencies on the degree of pillarisation, since then issues of immigration policy will be considered under the term multiculturalism. This results in the first-mentioned category being considered until the 1970s as covering more right-wing or conservative statements and the other as including the more progressive, anti-pillarisation issues. Since the 1980s, however, the meaning of both categories in terms of the opposite between progressive and conservative points of view should have completely turned around. Therefore, the content and ideological direction of some of the 56 categories can change over time and the manifesto data does not allow for a more detailed analysis of each category. Furthermore, problems will arise when working with the general left right party placement provided by the manifesto dataset. The simple left right index is established by a subtraction of a priori defined left policy categories from right labelled issues. 76 Yet, the MRG and CMP coders did not keep in mind that maybe a policy area can switch from a left to a more right meaning during the time period since 1945. To analyse party and coalition politics in a more sophisticated way, a number of studies tried to extract specific policy-area positions of political parties using the MRG/CMP

ANALYSING PARTY POLITICS IN GERMANY 291 data. 77 By referring to multivariate statistical methods, in particular techniques of dimension reduction, the problems of the manifesto dataset structure become evident. To estimate positions of parties in more than one policy dimension, it is necessary either to include the manifestos from all elections in one country or to transpose the dataset. When applying the first procedure, problems arise when interpreting the extracted principal components. The findings in the first MRG publication 78 have to be treated with caution, because the stepwise elimination of minor important policy dimensions by applying factor analysis results in barely comprehensible principal components. In the last-mentioned study, only the first extracted principal component can easily be interpreted, mostly as a socio-economic left right dimension. Very recent research stresses the point that studies that are based on the CMP dataset treat the extracted positions of parties as their exact location in a given policy space. Therefore, Benoit, Laver and Mikhaylov developed a statistical technique that estimates error terms for each of the 56 CMP categories. 79 The application of this technique to prominent studies in political science shows that after including an error term, the relevance of the ideological variable based on the CMP dataset increases. Computerised Text Analysis An alternative to manual coding of party policy documents is the computer-based coding of political texts. 80 This approach has a number of advantages compared with the MRG/CMP procedure, but there are still problems. In contrast to the MRG/CMP, the basic principle of the computer-aided approach is not to sort sentences into one policy category by intelligent reading. Instead, the goal is to identify signal words ex ante, which were a priori defined as left and liberal or right and conservative. It is also possible to give words a neutral meaning for one dimension. 81 Thus the relevant policy dimensions have to be determined beforehand in order to assure a more theory-induced and less exploration-orientated analysis as in the publications of the MRG/CMP. Although by applying the dictionary procedure policy positions rather than issue saliencies can be measured and the possibility of human errors decreases due to the increased role of computer programming, there is still the possibility that the respective coder labels some signal words incorrectly, so that the results (partially) do not reflect reality. In addition, one has to create a codebook for each required language so that native speakers with knowledge of both the language and the ideological background of each word are required. This decreases the chances of cross-national, comparative analysis of policy positions, e.g. the analysis of similar or deviating positions of specific policy-area positions of parties belonging to similar ideological families. Another critical point is that replications of studies based on the dictionary procedure will only be possible if the dictionaries are available. If these word lists are not provided by the respective publications or available from internet sources, a correct replication will not be possible. 82 In their contribution to this special issue, König and Luig introduce a new dictionary-based approach to textual analysis. German LexIconSpace (G-LIS) identifies the positions of political parties and governments by including information on the legislative context. On this basis, G-LIS considers not only the orderings on five conflict lines but also the salience of each conflict dimension when extracting the dimensional solution ex ante. G-LIS makes use of the content of legislative bills and searches for the same words in election

292 GERMAN POLITICS manifestos and government declarations, so that political actors can be located in a multi-dimensional policy space. In addition, G-LIS calculates an error term for each estimated position because it applies Bayesian statistics. Alternative methods that do not make use of a dictionary are software programmes like Alceste. 83 Based on a matrix of word frequencies, Alceste, for instance, relies on a characterisation of dimensions extracted by applying principal components. Likewise, to the fully computerised approach Wordfish, which is described below, Alceste requires a manual selection of those parts of the political text that deal with the policy area one is interested in. Fully computer-aided methods of content analysis such as the Wordscores approach developed by Laver, Benoit and Garry 84 and Slapin and Proksch s Wordfish technique 85 are advancements of the semi-manual dictionary approach. The main advantage of both approaches is that the position estimation is left completely to computer algorithms. Therefore, potential problems associated with the dictionary procedure do not arise. The basic idea of both techniques is to compare the frequency distribution of words from different texts and to conclude a specific policy-area position of a text on the basis of the differences in the share of used words inside the set of analysed political documents. They differ, however, in one decisive aspect. Wordscores compares the word frequencies to so-called reference texts and assigns document scores based on the similarity to these references. Wordfish, by contrast, implements a parametric word scaling model to estimate document positions and does not require choosing reference texts. To clarify the differences between both techniques, I briefly describe Wordscores and Wordfish below. Wordscores compares the relative word frequency of a text whose programmatic position is known to the word distribution of a text of the same character whose position is unknown. Laver, Benoit and Garry refer to these two sorts of documents as reference texts and virgin texts, respectively. 86 In a nutshell, the position of a virgin text changes if the frequencies of some signal words, which were not ex ante determined, goes up or down. This implies that the required assumption by using Wordscores (as well as Wordfish) is that political actors do not use words randomly. Instead, in order to include ideological signals 87 in election manifestos parties will mention some types of words more frequently and others less frequently or even never. To show their hostile position towards raising taxes, liberal parties, for instance, often use the word tax or taxes in connection with a decrease of the tax burden. This approach and the theory behind it is not far away from saliency theory and the coding scheme of the Manifesto Research Group, but does in contrast enable us to analyse both the policy positions and issue saliencies of political parties. To be more precise, the Wordscore technique can be subdivided into the following theoretical and methodological steps. First, as was the case in the dictionary approach or the Pappi and Shikano principal component analysis of the CMP data, one has to identify a priori the most important policy dimensions for the party systems one is interested in. The second step is more methodologically oriented and crucial for the further proceeding as well as the stability of the results. One has to search for the reference texts of a political actor, which are used for estimating the positions of political actors given in the virgin texts. Reference texts have to fulfil the following characteristics: to obtain valid results, the selected texts should be of the same character as the

ANALYSING PARTY POLITICS IN GERMANY 293 one whose position is unknown. Thus, when one is interested in the specific policy-area position of an election manifesto, it would be best to use election manifestos as reference texts. 88 These sorts of documents are very similar in terms of their structure as well as the words used. The risk of obtaining invalid results by using Wordscore will increase if, for instance, one tries to estimate the position of an election manifesto when using a speech of a politician as a reference, because the word structure in both texts will be less homogeneous. One critical factor is the allocation of policy-area positions of political parties to the selected reference documents. When assuming that election manifestos are the best choice for reference texts due to their wide coverage of policy issues and that the Laver and Hunt and Benoit and Laver expert surveys are reliable sources for reference scores, 89 the party positions extracted in both expert surveys must be allocated to the correct party platform. This is quite simple in case of the more recent expert survey. Benoit and Laver asked the experts to locate the respective parties on each policy dimension on the basis of the last general election. However, such a procedure was not applied in the Laver and Hunt study. The experts ranked the parties in their country in the first part of 1989. 90 As will be discussed in the concluding contribution to this volume, an imprecise allocation of a reference score to a political text will result in biased virgin text scores. Despite discussions on the standardisation method for Wordscores estimates 91 and on the method itself, 92 a study by Klemmensen, Hobolt and Hansen shows the robustness of the method by analysing Danish election manifestos and government speeches between 1945 and 2005. 93 The extracted policy positions of Danish political actors correlate strongly with estimates by expert surveys and the CMP dataset. 94 The more recent Wordfish approach developed by Slapin and Proksch comes very close to the wordscoring technique, but has some decisive differences. First and most important, the selection of reference texts is not required. On the basis of an item response modelling technique that is only based on word frequencies, Slapin and Proksch show that their estimates of German party positions in the time period between 1990 and 2005 correlate strongly with other measures that are based on expert surveys, CMP data or Wordscores analyses of German election manifestos. 95 The latter is particularly the case for the overall left right dimension, the economic and foreign policy domain, but does not hold for the societal policy dimension. To extract specific policy-area positions of political actors, researchers have to separate parts of the respective political document that belong to a specific policy area. Only those parts will be included in the set of word frequencies and, therefore, in the Wordfish analysis. While the advantage of Wordfish over Wordscores is that no reference texts or scores are required, 96 one challenge for both techniques is the generation of scores specific to policy areas. Laver, Benoit and Garry proposed a policy-blind approach. Policy-specific positions are not generated by changing the text input (e.g. foreign policy manifesto sections for foreign policy estimates) but by changing the reference scores assigned to the texts. Slapin and Proksch, in contrast, have argued for a substantive selection of texts. Foreign policy scores should only be based on word frequency distributions in the foreign policy sections of the manifestos. 97 This therefore requires a manual allocation of paragraphs or even sentences to a specific policy area. The latter requires not only the knowledge of the respective document s language, but

294 GERMAN POLITICS also the development of a classification scheme for specific policy areas. Slapin and Proksch refer to a dictionary-based analysis of German party competition 98 and use the headings of each chapter in German election manifestos to allocate them to the economic or foreign policy dimension. The remaining parts of the manifestos reflect the societal policy dimension according to Slapin and Proksch. 99 While this is quite a straightforward approach and may explain the missing congruence with the results of other studies on the positions of German parties on a social policy dimension, studies that apply the Wordfish technique to extract the policy positions of German state manifestos developed a more detailed concept, which identifies each paragraph as a coding unit. 100 Furthermore, comparative analyses of policy documents by applying Wordfish would result in installing a similarly large research group as in the initial coding of party manifestos by the Manifesto Research Group. One would need a group of coders for each country or language that allocates the respective unit (sentence, paragraph or chapter) to the correct or at least closest policy area. OUTLINE OF THE SPECIAL ISSUE As the description of the various approaches to content analysis has shown, every method in this field has advantages and drawbacks. The contributions to this volume therefore do not apply a single method since no one believes that there is a gold standard 101 in analysing policy preferences of political actors. The respective articles, by contrast, make use of the already existing CMP dataset or apply the CMP coding scheme to the 2005 election manifestos of German parties. In addition, the recently developed techniques Wordscores and Wordfish are applied, as well as dictionary-based techniques of content analysis. To be more specific, the essays in the first part of the special issue apply the CMP dataset to derive policy positions of German parties and present recently developed techniques of content analysis. The paper by Linhart and Shikano uses the CMP dataset to apply directional and proximity models to extract policy preferences and their intensity for all German Bundestag parties between 1949 and 2005. Slapin and Proksch present their very recent technique Wordfish and discuss guidelines on how to process linguistic information for researchers interested in using the technique, focusing specifically on German texts since the federal election of 1961. The contribution by König and Luig relies on a dictionary-based technique of content analysis. In contrast to context-free approaches, the authors use the keywords of German federal legislation and cleavage connotations to estimate the party and governmental positions from the 4th to the 16th legislative term. The innovation of their so-called G-LIS approach is that the analysis considers both a portfolio- and an ideology-specific perspective. They use Bayesian statistics to identify the dimensionality and positions of governments and parties in each portfolio space. On that basis, one is able to compare the distances between these positions in the same space over time, while previous studies had to make strong assumptions when anchoring different policy spaces. The contributions in the second part of this volume analyse the preferences of political actors in Germany on different levels of party competition. Müller describes and analyses the programmatic heterogeneity of German parties by taking into account