Left and Right in Comparative Politics

Transcription

1 Left and Right in Comparative Politics Detlef Jahn University of Greifswald Abstract This paper takes advantage of the fact that party manifesto data are freely available. The reliable and transparent coding of party manifestos allows for the testing of theories and hypotheses. In this paper I focus on the left-right index which is important in many areas of political science. Although the RILE, developed by the Comparative Party Manifest Group in the early 1990s, has been successfully applied to many research questions, I suggest an alternative left-right index which is more flexible in identifying country and time specific changes in the left-right discourse. This index consists of a left-right core index which is closely related to a parsimonious concept of left and right deduced from political theory, on the one hand. On the other hand, the index is open to other issues which align with the core left-right statements and which change over time and space. These two elements of the left-right index make it possible to analyze changes in the left-right semantic, a factor which has been emphasized in historical and theoretical works. However, I also demonstrate that the left-right index developed in this paper is able to address additional research questions which are essential in comparative politics. Two further examples are provided: one examines the importance of the left-right dimension over time and in specific countries and the other introduces an index for ideological party cohesion. In particular, the latter index is seriously needed in order to test and work with elaborated spatial theories. Paper prepared for presentation at the conference on Mapping Policy Preferences from Texts, May 15-16, 2014 at the WZB Berlin Social Science Center, room A 300, organized by MARPOR Manifesto Research on Political Representation. Most of the data discussed in this paper can be downloaded at: 1

2 There are not many data sets like that of the party manifesto project, recently termed MARPOR project. It is unique in that it covers the analysis of party manifestos from most democratic countries from 1945 until the most recent elections. For the scientific community it is of great value that the data is free for everybody to use and that the process of data collection is transparent and responsive to the needs of the scientific community. The quality of the data is continuously improving and the methodology of using the manifesto data has become increasingly elaborate over the years. All of this makes the party manifesto data an invaluable tool for researchers and enables the scientific community to address new research questions and to test established theories by using data for many parties and countries over a long time period. In In previous work drawing on the MARPOR project and its predecessors, the data has been presented for established Western democracies (Budge et al. 2001) and for the newly emerging democracies in Central Eastern Europe (Klingemann et al. 2006). Some of the most recent work applying this data set addresses fundamental statistical challenges and provides solutions for a better use of the manifesto data (Volkens et al. 2013). In this context it needs to be emphasized that suggestions for alternative concepts and methodologies are only possible because the MARPOR project shares its findings with the scientific community. This enables discussions about the most suitable ways of addressing pressing questions in political science. One such question focused upon in this paper is concerned with the aggregation of party positions in ideological dimensions. In comparative politics the left-right dimension has strong explanatory power (McDonald and Budge 2005). Therefore, one major achievement of the MARPOR project is that it offers a right-left (RILE) scale which places the parties on a left-right continuum. The above mentioned publications therefore focus on the left-right position of the political parties in Eastern Europe, the European Union and the OECD countries. Since the positions of parties on the left-right dimension are of paramount importance for comparative politics, I will address the issue of how to use this dimension in empirical studies. The left-right dimension emerged from political discourses relevant in the 19 th and early 20 th century and has been implied in the writings of Marx and Engels, Lenin, Luxemburg, Disraeli, Green, Spencers and many others. Even more importantly, the left-right dimension is grounded in elaborated theoretical and even philosophical debates (Lukes 1990; 2003; Bobbio 1996). For conceptual and analytical reasons it is therefore essential to align empirical analysis with the theoretical studies in the field (Biezen and Saward 2008). Although this index has been very successful in many areas of research (see for instance: Klingemann et al. 1994; McDonald and Budge 2005) there are some issues which should be addressed when using the RILE. Some authors see the difficulties implied in the construction of the RILE as a reason to offer alternative indices (Bartolini and Mair 1990; Gabel and Huber 2000; Franzmann and Kaiser 2006; Keman 2007). The author of this paper shares the concerns of the above mentioned authors that the RILE may not be the best index when comparing many countries with different traditions and histories over a long period of time. In order to offer a solution to the challenges of the RILE, I have constructed an alternative left-right score (LR) (Jahn 2011) which may, in some cases, be more suitable and offers more analytical options for party ideological analysis with party manifesto data (Jahn 2010; 2014; Jahn and Oberst 2012). 2

3 The remainder of the paper is structured in the following way. First, I summarize the rationale of the RILE as it was presented in the original article by Laver and Budge. At the end of this section, I summarize some of the shortcomings of this index which I wish to overcome with my own left-right (LR) index. Next, I present the rationale of my alternative left-right index in section 2. In the third section, I present some concepts and examples relating to how to use the LR and other ideological indices developed in the same fashion. The indices developed in this paper are not only available for individual parties but also for veto players (Jahn 2010) ( 1 The Right-Left (RILE) Index for Party Manifesto Data of the Party Manifesto Group In a recent publication, Ian Budge and Thomas Meyer (2013: 105) are surprised that I assess the RILE as inductive and that I would go with that against all the documentation from its first emergence onwards. In order to justify my judgment I wish to summarize my reading of the history of constructing and legitimizing the statements included in the RILE. I see clearly a trend from inductive procedures to ever stronger deductive claims in the publications concerning the RILE. However, before I start this analysis a remark on inductive and deductive research is in order. I wish to emphasize that deductive reasoning is not at all superior to inductive inferences. However, deductive and inductive approaches certainly have different advantages and shortcomings. Deductive approaches allow for theory testing but are not as well tailored to fit empirical data. In fact it is exactly the point that data might not fit the theory which determines the essence of the deductive approach. Therefore, a deductive index of left and right asks whether the left-right dimension is still a valid tool to analyze programmatic preferences in modern societies. A precondition for such an analysis is, of course, that left and right are clearly defined. Conversely, an inductive approach starts out from empirical observations. By summarizing the observable manifestations of left-right criteria, one looks for patterns in the data and asks if there could be an underlying dimension which may be classified as left-right. The obtained left-right dimension is then defined by the patterns of data which emerge from the analysis. The advantage of an inductive approach is that it fits the index to the data whereas the disadvantage is that it is more vulnerable to validity threats. However, most left-right indices use a combination of deductive and inductive reasoning though with a different emphasis on one approach or the other. This is certainly also true for RILE and the index introduced in this paper. The original purpose of the analysis conducted by the party manifesto group was to test between spatial models and a saliency theory. In the latter, political parties strive to increase the saliency of their favored issues. In this view, issue ownership is the center of interest rather than maximizing votes based on leftright position (Budge, Robertson, and Hearl 1987). Since the coding scheme is built in this way, some scholars question if it is even possible to retrospectively create a left-right index from the party manifesto data (Harmel, Janda, and Tan 1995; Laver and Garry 2000). However, Laver and Budge (1992) later constructed a Right-Left Index (RILE) from the PMG data. 1 There will also soon be an extensive documentation of the indices discussed in this paper for party and government positions and veto players. 3

4 Starting out from 54 statements which were developed as a result of extensive experimentation and discussion in the research group Laver and Budge (1992: 21) first collapsed categories that seemed to be closely related to each other. Even if later it was proclaimed that the RILE was based on deductive considerations and deduced from the writings of Marx and Burke (Klingemann et al. 2006: 6), this is not really accurate as the seven subcategories were created in the following way: The rationale behind the grouping of coding categories was based largely on the lessons learned from original analysis. Certain theoretical distinctions between categories persistently forced coders into awkward decisions, so these were combined. In addition, we did a large number of exploratory factor analyses to search for combinations of variables that persistently loaded together over a wide range of systems. (Laver and Budge 1992: 23) That means that the seven subcategories were not constructed having different ideologies in mind (that is, that they were not deduced in a systematic fashion from a theory of left and right) but rather that some issue areas were grouped in a more or less inductive or intuitive manner. The seven subcategories and the thirteen individual issues that were used for further analysis are summarized in Table 1. The subcategories and remaining issues constituted 20 policy categories that were used in factor analyses in order to establish a left and right dimension (first factor). 2 The factor analysis was conducted for each of the eleven countries in the study. 3 It turned out that the subcategories State Intervention and Peace and Co-Operation loaded on the left, and Capitalist Economy and Social Conservatism loaded on the right. Some items which also loaded more or less consistently either Left or Right in the individual countries were included in the common left-right scale. So in the end, the left-hand end of the scale was constructed by combining items of the subcategories State Intervention and Peace and Co-Operation in addition to the single items of Democracy, Welfare State Expansion: positive, Education: positive, and Labour Groups: positive. The right-hand end of the scale was constructed as the sum of the frequency counts for Freedom and Human Rights and Military: positive in addition to the items of the subcategories Capitalist Economy and Social Conservatism. The final scale was then calculated by subtracting the frequencies of the thirteen left items from the thirteen right items. 2 En passant and without further specification Laver and Budge (1992: 25) tell us that certain coding categories were excluded from subsequent analysis because they were very little referred to or because of their inherent theoretical ambiguity. The excluded statements were: per101 (Foreign Special Relationship: positive), per102 (Foreign Special Relationship: negative), per109 (Internationalism: negative), per302 (Centralization: positive), per304 (Political Corruption: negative), per405 (Corporatism: positive), per408 (Economic Goals), per409 (Keynesian Demand Management: positive), per415 (Marxist Analysis: positive), per507 (Education Limitation: positive), per602 (National Way of Life: negative), per604 (Traditional Morality: negative), per607 Multiculturalism: positive), per608 (Multiculturalism: negative), per702 (Labour Group: negative), and per704 (Middle Class and Professional Groups: positive). Underlined statements must have been dismissed because of theoretical reasons since they are as frequently mentioned than included issues. 3 It is not clear what are these eleven countries. Only ten of them are mentioned in Laver and Budge (1992: 27): Austria, Belgium, Britain, Denmark, France, Ireland, Italy, Luxembourg, Netherlands, and Sweden. Also the time period of the factor analysis is unclear. Later it was said that the data were investigated through factor analysis of the then existing MRG data up to broadly (Klingemann et al. 2006: 6) 4

5 Table 1: Individual Categories of the RILE New Category State intervention Quality of life Peace and co-operation Anti-establishment views Capitalist economics Social conservatism Productivity and technology Old categories 403 Regulation of capitalism 404 Economic planning 406 Protectionism: positive 412 Controlled economy 413 Nationalisation 501 Environmental protection 502 Art, sport, leisure and media 103 Decolonisation 105 Military: negative 106 Peace 107 Internationalism: positive 204 Constitutionalism: negative 304 Government corruption 602 Defense of national way of life: negative 604 Traditional morality: negative 401 Free enterprise 402 Incentives 407 Protectionism: negative 414 Economic, orthodoxy and efficiency 505 Social services expansion: negative 203 Constitutionalism: positive 305 Government effectiveness and authority 601 National way of life: positive 603 Traditional morality: positive 605 Law and order 606 National effort, social harmony 410 Productivity 411 Technology and infrastructure Coding categories retained intact 104 Military: positive 108 European Community: positive 110 European Community: negative 201 Freedom and domestic human rights 202 Democracy 301 Decentralization 303 Government efficiency 503 Social justice 504 Social services expansion: positive 506 Education: positive 701 Labour groups 703 Agriculture and farmers 705 Underprivileged minorities Source: Laver and Budge 1992: 24. The inductive method of constructing the RILE in combination with the addition of statements which are not theoretically connected to a left-right ideology makes the index ambiguous. In the way that half of the statements coded in the PMG-dataset refer in one way or another to the left and right makes it very difficult to interpret why some parties behave peculiarly (Pelizzo 2003). Others also saw some 5

6 inconsistencies in the RILE. Keman (2007: 79) concludes that the inclusion of aspects like military, constitutionalism, morality, law and order, democracy, and so on in a left-right scale is not only confusing, but also wrong. Including other matters of serious dispute that divide political parties and their constituencies is certainly relevant for understanding electoral politics and party systems, as well as policy-making by government. Yet, reducing these complex differences into one dimension or using simple dichotomies is bad for comparability and conceptual clarity. 4 In addition Keman points out that a one-dimensional analysis would probably not be appropriate for most Western societies. As I stressed before, the major problem of the RILE is its emphasis on inductive identification of a leftright dimension (Jahn 2011). Of course, an inductive approach fits the scores to the data set, so that we find a basic dimension in any case. However, what this dimension then means is less clear. This has also been admitted by the researchers of the MARPOR project. Referring to their left items Klingemann et al. (2006: 5-6) state: There is after all no logical or inherent reason why support for peace should be associated with government interventionism However, a theory does exactly that: it combines various issues with an inherent and logical reasoning. According to Klingemann et al. the association of the statements is only justified because these statements are mentioned together by political parties: The fact remains however that party ideologies do put them together. (Emphasis in original). Gabel and Huber (2000: 95) even question the possibility to develop and defend arguments about inclusion, exclusion, and relative importance of specific issues in order to construct a left-right scale deductively. However, a deductive scale would make interpretation much easier. It could also answer the question of whether what we conceptualize as left and right has still a meaning in practical politics today. It is therefore a bit perplexing that in later publications - step by step - ever greater deductive claims were raised concerning the RILE. Although it was stated that a priori theoretical coherence was the prime consideration (Budge et al. 2001: 26) no reference whatsoever was made to any theory in the first publications concerned with the construction of the RILE. Laver and Budge (1992) article was technical in nature and theoretical claims were not made at all. In Budge et al. (2001: 21/22) it was said that the left items refer to unspecified Marx writings and the right items are familiar from the writings and speeches of exponents like Reagan and Thatcher. Since the latter personalities are certainly not political theorists, in later writings right positions refer to Edmund Burke (Klingemann et al. 2006: 6). So far, the climax of the ex post theoretical underpinnings has been expressed by Budge (2013; see also Budge and Meyer 2013: 88-90), who claimed that the left items refer to Marx, Engels and Lenin s writings and the right items are connected with the work of Disraeli, Green, and Spencer. Were that mentioned in the original text during the construction of the RILE and if clearly identifiable arguments of the mentioned authors would be consistent with the 26 items dealing with right and left, the RILE would have been deductive. However, when we look at the history of the construction of the RILE, one can only conclude that inductive motives of finding issues which parties link together (Budge et al. 2001: 21) have been the major interest. 4 As an alternative Keman (2007) suggests a left-right index which includes only socio-economic topics (see also Bartolini and Mair (1990) for a similar approach). 6

7 The RILE has other shortcomings which disturb the construction of a valid left-right index and limit its use for addressing other important questions of party ideology. One major problem is that the RILE treats each statement included in the RILE as equal. That means, for example, that the claim of nationalizing the industry weights as strongly left as demanding the expansion of education. Although it has often been claimed that the degree of rightness and leftness is important and that there is indeed a hierarchy of right and left statements 5, the RILE makes no distinction of how strong different left and right statements are. That can lead to the result that some mild left statements and one strong right statement makes a party radically left although there are good reasons to assume the opposite. An issue in the research of political ideologies is also whether political ideologies change over time or differ in various national contexts. The RILE treats left and right as constant over time and across countries. However, it has often been claimed that the semantic of left and right has changed over time and is dependent on country specific conditions (Bartolini 2000; Kitschelt and Hellemans 1990). As Benoit and Laver (2006: 202) demonstrate, the left-right semantic has a specific meaning for everybody; suggesting that correlating left-right positions with other policy issues in order to specify what is left and right shows that left and right means something very different in the countries under investigation. That is, the meaning of the left-right dimension is not the same across different national contexts. And even more so Benoit and Laver (2006: 203) conclude: we might extend this conclusion to speculate that the substantive meaning of left and right might also change over time, with equally deep implications for the interpretation of times series of party positions, even in one country, on some common left-right scale. Furthermore, there are some voices which doubt that the left-right dimension is still as dominant as some decades ago and that other political dimensions have increased in importance or modify the leftright dimension substantially (Kitschelt 1984; Giddens 1994). It would therefore be essential to have a left-right index which could identify the core of the left right ideology and the special features of left and right in specific countries and periods. With such an index one could also answer questions of whether the left-right semantic is in decline or if it is able to renew itself by aligning with new issues (according to the saying that there is life in the old dog yet ). Having an index which identifies changes in left and right could also focus on questions relating to which parties are left or right and why. The left-right ideology is plagued by the fact that everybody has an intuitive feeling about which party is left or right. However, often these deeply rooted feelings are based on inflexible presumptions which are not open to ideological shifts in reality. This in turn often leads to the conclusion that indices are perceived as wrong when they do not match held assumptions. More importantly, such a perspective neglects subtle and substantial shifts of party positions. Another point to mention is that party ideology is not only the position of a party on a continuum from left to right. In party research, political ideology also has other important aspects. One neglected aspect in empirical studies which obtains high importance in analytical terms is ideological party cohesion. Most scholars strongly agree that political parties as collective actors are by definition not unitary actors and that party cohesion is an important analytical and substantive issue with fundamental consequences for 5 Advocating public ownership of industries puts one far to the left; desires to have government closely regulated privately owned firms are not quite as far left. (McDonald, Mendes, and Kim 2007, 3). 7

8 theoretical modeling and empirical analysis, as well as with practical implications. Therefore, it would be an advantage if a left-right index is able to grasp other important aspects of party ideology beyond position such as the concept of party cohesion. As I will show in this paper, the LR-index can address thi question in contrast to the RILE. Finally, there are also some methodological concerns which apply to the RILE. In general one can question if factor analysis is appropriate for the number of observations and the character of the data. In particular, the second aspect raises concern. The party manifesto data are certainly not normally distributed. The RILE scores are based on counting the frequency of various statements. Such a variable is a typical case of a count variable, which normally has no normal distribution. An inspection of the party manifesto data shows that most statements are highly skewed. In this case, it is not appropriate to use a linear factor analysis but rather a nonlinear factor-analytic model based on latent class analysis (Vermunt and Magidson 2005) or alternative techniques (see below). How to Grasp the Changing Meaning of Left and Right: Towards a Country and Time Specific Left-Right Index In order to construct a left-right index which addresses the issues mentioned above, I analyze the 56 statements of the party manifesto coding by distinguishing between a core of left and right which have been systematically deduced from political theory and additional statements which are country and time specific. In order to estimate the degree of leftness and rightness, I use multi-dimensional scaling (MDS) for each statement which has been included in the LR index. That means that my index consists of three elements: a left-right core (LR_core), additional left-right issues (LR_plus), and the estimation of the degree of leftness and rightness of each issue included in the index. My starting point is the theory of left and right developed by Noberto Bobbio (1996). My approach can be briefly summarized in the following way (for details see Jahn 2011): P1: Bobbio defines the political left as linked to equality and the political right as legitimizing inequality. P2: A set of positional issues is identified as being linked to equality and inequality. C1: This set of issues is the core of left or right. P3: Not all issues are equally radical. C2: The radicalism of issues can be estimated by the use of these issues by political parties (MDS). P4: Left and right are defined in national and temporal contexts. C3: Issues which correlate significantly with the core left-right issues are country and/or period specific issues. P5: Not all of the country and time specific issues are equally radical. C4: The radicalism of issues can be estimated by the use of these issues by political parties (MDS). 8

9 The LR_core is deductive in that it uses ten left-right issues from the party manifesto data set which are found in Bobbio s theory of left and right. I made a systematic content analysis of Bobbio s book and deduced which items of the party manifesto data fit his argument. Propositions P1, P2 and the empirical conclusion C1 are therefore purely deductive and set the groundwork for further analysis. The LR_core is a universal (across countries and time) left-right index. That is, it is stable and has no context specific elements. The core issues are specific features of left and right. Without referring to these issues as left and right, a left-right index would be meaningless. Therefore, they are in fact leftright issues over all countries and time periods. In a second step, I estimate the radicalism of each issue (P3 and C2). Although it makes sense to claim that the demand for nationalization is more radical left than the demand of market regulation, it is not possible to determine theoretically the degree to which these two issues are differentially radical. I therefore use an inductive approach to determine the degree of radicalism. Table 2 summarizes the statements that I use for further analysis of the left-right dimension of political parties in highly developed democratic industrial societies. The table ranks these statements according to their correspondence to the core of the three classical ideologies of the nineteenth and twentieth century. The theoretical reasons for using just these statements is extensively given in another publication (Jahn 2011): Table 2: Core Left/Right Statements LEFT RIGHT Scientific Socialism Liberal Conservative Nationalization (per413) Controlled Economy (per412) Economic Planning (per404) Market Regulation (per403) Welfare State Limitation (per505) Free Enterprise (per401) Economic Orthodoxy (per414) Traditional Morality (per603) Social Harmony (per606) National Way of Life (per601) Explanations: Numbers refer to the statements of the PMG data. Statements of each dimension are shown from up to down in order of degree. We can derive the following assumptions from the theoretical analysis. First, as pointed out above, there may be a hierarchy of importance of the party manifesto data according to their association with the concept of equality or inequality. Second, if the analytical conclusion that the left has a one-dimensional claim while the right is divided into Conservatives and Liberals is correct, the results should reflect this triangular constellation. However, this would also mean that the very concept of left and right is not onedimensional but refers to at least two dimensions which are made up of the three ideologies. If that is so, the endeavor of constructing a one-dimensional left-right scale can only exhibit limited success. I chose not to elaborate on this point in this article in order to keep the presentation short (for a discussion and empirical analysis of the dimensionality of left and right see for instance Marks et al. 2002; Benoit and Laver 2006). In order to construct a new index that meets all the criteria mentioned above I apply a three step approach. First, I construct a general index deduced from political philosophy which is generalizable over countries and over time (the core of left and right). Second, I use regression analysis in order to identify 9

10 country and time specific components of left and right. Finally, I sum up both components in a new leftright index. For the first step, I apply the multidimensional scaling technique. Multidimensional scaling (MDS) is a set of data analysis techniques that display the structure of distance-like data as a geometrical picture (Coxon 1982; Cox and Cox 2001). An advantage of MDS over factor analysis is that it can be applied to count variables. For the analysis, the choice of time periods and countries is highly relevant. The RILE is based on an analysis of ten or eleven countries for the period between the second half of the 1940s until around A reason for this selection other than data availability has not been provided. I decided to use the Post-Second World War data until the oil crisis in 1973 (the cut-off date is October 1, 1973). The oil crisis represents a date when the continuity of post war development was questioned to a significant degree for the first time. After this date, some countries (above all Norway and Denmark) experienced major changes in their party systems (although because of the issue of European integration). Deciding on this time period excludes the new European democracies Greece, Spain, and Portugal from the calculation of the core left-right statements. 6 With respect to the selection of countries, I include all established and democratic OECD countries in order to identify a left-right dimension. I experimented with different samples. For instance, it could be argued that the left-right semantic is mainly a European issue and is less relevant to the new world countries and Japan. However, this hypothesis is not convincing since left and right also have different meanings among European states (Bartolini 2000). 7 Figure 1 shows the plot of the derived stimulus configuration of the above introduced core statements for left, liberal, and conservative ideologies, respectively. The analysis is based on a two dimensional solution. 6 Further analysis with different time periods shows that the results are robust. However, in order to anchor the left-right scale I had to decide for a cut-off point that makes sense analytically. Determining the cut-off point was guided by substantial and methodological considerations. Substantially one could also have used 1968 as cut-off point since the left-right discourse was severely altered because of the intervention of the Soviet Union in Czechoslovakia (emergence of Euro-Communism). Another cut-off point could have been 1980 since in the early 1980s a left libertarian discourse gained momentum (Kitschelt 1994). Furthermore, liberal discourse was radicalized by the rhetoric of Margaret Thatcher and Ronald Reagan. From a methodological standpoint, the shorter the core time period, the better one can identify changes in the meaning of left and right over time. On the other hand, a short time period contains too few observations to make meaningful data analysis possible. A compromise between these two claims had to be found. This is not relevant for the analysis of the core statements which we debate here but this period determines the number of observations (a party at an election) in the following analysis of the extra-statements which is based on country analysis. Since that analysis is conducted as moving election periods the core period determines the length (or, better, the number of elections) for further analysis. From this perspective, a period from the 1940s until 1980 seems too long. On the other hand, the country with the lowest number of observations influences the regressions. That is the case for the USA which has only two observations (Democrats and Republicans) for each election. The elections from 1948 until 1972 lead to N = 14, which is still very low but acceptable in a bivariate regression. It would be 10 if 1968 was the cut-off point and 16 if 1980 was used. 7 After careful consideration I have decided that it is not appropriate to exclude the USA, Japan or other non-european countries in order to avoid ethnocentrism. I have included Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Japan, Luxembourg, New Zealand, Netherlands, Norway, Sweden, Switzerland, United Kingdom, and the United States. Results with subsets, however, have always arrived at the same triangle results between left and conservative/liberal. 10

11 Figure 1: Derived Stimulus Configuration Plot (Euclidean Distance Model) for the Left-Right Dimension 2 per601 1 per603 per606 0 per413 per404 per403 per414 per401-1 per412 per Left vs. Right Explanation: Nationalization (per413),controlled Economy (per412), Economic Planning (per404), Market Regulation (per403), Welfare State Limitation (per505), Free Enterprise (per401), Economic Orthodoxy (per414), National Way of Life (per601), Traditional Morality (per603), Social Harmony (per606). Number of observations = 745. The number of the per refers to the statements as mentioned in Budge et al and Klingemann et al The plot clearly shows a triangle, demonstrating that the right is divided in a conservative and liberal camp. 8 In general, all ideological statements group into the hypothesized categories. The plot shows strikingly that Controlled Economy, Economic Planning and to a very impressive degree Nationalization are the most radical left statements. Market Regulation is a more moderate left statement. In the liberal ideology, limiting the welfare state and the claim for Free Enterprise are, as predicted, the most radical right statements. The three conservative statements all fall perfectly together. The plot also demonstrates that the conservative statements constitute their own dimension and are isolated from the liberal-right statements. In terms of left and right the reference to Traditional Morality is the most radical conservative statement and Social Harmony the least radical. The analysis also shows that left items are more in opposition with the liberal than with the conservative items. Overall, the data align themselves impressively with the theoretically deducted positions, supporting the chosen deductive approach. In the next step, I will use the empirical findings from the theoretical statements for a left-right score of political parties in highly industrialized democracies. In order to construct a left-right scale of the core statements, I weight the frequency of each statement with the stimulus coordinates of the first (left-right) dimension. 9 This index is called LR_core. 8 The model needs five iterations. After this the S-stress improvement is less than The final matrix stress value is An RSQ of 0.94 shows that 94 percent of the variance in the MDS space is accounted for by the input data. If I force the analysis on one dimension, I need eight iterations and the matrix stress value increases to 0.23 (RSQ =.82). Generally, a value lower than.1 is excellent while a value between.1 and.2 is acceptable. 9 The stimulus coordinates scores for each statement are: Nationalization 2.06; Controlled Economy 1.55; Economic Planning 1.27; Market Regulation 0.38; National Way of Life -0.53; Traditional Morality -1.09; Social Harmony -0.28; Free Enterprise -1.09; 11

12 LR_core = (per413p * Sper413) + (per412p * Sper412) + (per404p * Sper404) + (1) (per403p * Sper403) + (per505p * Sper505) + (per401p * Sper401) + (per414p * Sper414) + (per603p * Sper603) + (per606p * Sper606) + (per601p * Sper601) Explanation: p = percentage of statement; S = stimulus coordinate scores of the first dimension of each statement (per). In the third step, I test for additional issues which are aligned with the LR_core. The assumption here is that the left-right dimension changes and is context specific (time and country) as suggested by Bartolini, and Benoit and Laver (P4). However, in order to be labeled left or right it needs to correlate with the LR_core (C3). In order to identify the extra-statements I use the core left-right score, as developed above, as an independent variable for each statement not included in the core left-right index. If the absolute value of the z-value (coefficient divided by standard error) is greater +/-2, I include the statement as an extra leftright issue (LR_plus). 10 Otherwise, I do not use it as a component of the left-right index. In order to obtain country and time-specific scores, I use moving time periods in each country, starting from the period of the first post-war election until the last election before October 1, Then I use the period from the second post-war election until the first election after October 1973 and so on until I reach to the last election in the MARPOR data for each country. In order to account for the non-normal distribution, I apply three different regression models: a normal OLS-regression model, a Poisson regression model, and a negative binomial regression model (Cameron and Trivedi 1998; Hilbe 2007). In some instances, count variables approximate a normal distribution. In this case a linear regression is acceptable (Hoffmann 2004: 101). More often, however, the statements' frequencies have a rapidly descending tail. This means that the appearance of these statements is a rare event. Therefore, the variables' distribution often peaks at one or two (or the respective percentage value) and then becomes much rarer at higher values. Count variables, especially when they gauge rare events, often follow a Poisson distribution (Cameron and Trivedi 1998). However, Poisson regression models are only appropriate in cases of rare events and granted that the variance more or less equals the mean. Closer inspection of the statements shows that this is not true in most cases for party manifesto data. If the variance is much larger than the mean, there is an issue of overdispersed variables. Given that the dependent variable is overdispersed, an alternative approach for rare events has to be used: the negative binomial regression (NB) model. As a rule, I used the threshold of the overdispersion parameter alpha = 1 until I switched from Poisson models to NB. In the rare cases where the dependent variable approximated a normal distribution, I used linear OLS. In addition to relying on statistical parameters, I further inspected all variables and bivariate correlations visually (histograms and scatter plots) in order to judge the appropriate model. This is also important because sometimes there are too Economic Orthodoxy -0.79; Limited Welfare State In order to make our LR index comparable with the other indices we changed the sign by multiplying the stimulus coordinate scores with I use z-values instead of significance levels since we do not work with a sample but the universe of policy statements at elections (King 1986). 12

13 few positive observations in order to conduct reliable calculations. In this case, statements have not been considered for the indices. 11 The statements that have been identified as extra-statements by the regression analyses have been entered in a new MDS-analysis for the respective time periods in the countries. Their stimulus scores on the first dimension 12 of these extra-statements (LR_plus) have been added to the stimulus scores of the core left-right statements (LR_core) in order to get the final Left-Right (LR) score. LR_plus = (LRX per * LRX S ) (2) Explanations: LRX = extra-statements identified through regression analysis; per = percentage of the i s statements; S = stimulus coordinate score of the i s statement. LR = LR_core + LR_plus (3) Extending the Analysis of Left and Right In this part of the paper I will give some examples of how the LR index extends the party ideological analysis and addresses important questions in current comparative politics. The issues are the different meanings of left and right in countries and over time, the relevance of the left-right dimension in various countries and periods of time, and finally the measurement of ideological party cohesion. Changing Meaning of Left and Right in Space and Time The RILE is a left-right score that contains almost half of all statements (pers) used in the party manifesto coding. This large number of items guarantees that the left-right scale has enough issues to reach a result. The disadvantage is that one is not sure which of the many included items make up the left-right position for a specific party in a specific period of time. This is a general conceptual problem which has already been addressed above by constructing an alternative LR-index. More specific is the question relating to how far the left-right dimension has to be modified because new issues arise or align with the left-right dimension. Volkens et al. (2013: 5) state: Helpful suggestions are often made about updating the scale to conform to the way issues are grouped now (at any rate in electoral surveys) adding green issues to the left for example or opposition to the EU on the right. The response to this request is negative because such suggestions run up against the consideration of continuity and cross-national comparison because: what substantive research implications can one draw from left-right movement involving different issues at different times? (ibid.). 11 This ends up in around regressions (3 regression models * ten time periods * 46 statements) for each country (in total around regressions). There are on average 10 time periods for each country. For the USA there are for instance 9 time periods: , , , , , , , , and , (N=14). 12 In some instances, I used the second dimension or changed the sign. In order to estimate which sign and dimension should be used I used the core items, especially Nationalization and Limited Welfare State as orientation. In the case of Japan it was often difficult to determine which dimension should be used. This is, of course, a clear indicator that the classical left-right items are not clear demarcation items and that therefore the left-right dimension is not very clear in Japan. The left-right dimension was sometimes second in Canada, Luxembourg, Belgium, and the Netherlands. However, in these cases it was clearly identifiable. 13

14 Addressing the latter question first: does the RILE really allow for the identification of an unambiguous identification of the left-right dimension? Since the RILE contains so many issues, one simply does not know what makes the left-right position in country i at time t. It could be that it was military negative and peace and nationalization in country k at time x, but democracy, education, and internationalism in country k at time y, and another combination in country m at time x and again another at time y. The different issue bias is also implicitly given in the RILE. Only it is hopelessly hidden and implicit in the lavishly constructed RILE. However, let s turn to the empirical question. Are green and EU issues part of the left-right dimension? In the LR-index these issues are not included in the LR_core. In this way, the RILE and LR index agree. However, the LR index also contains extra-statements and therefore the question is whether the green and EU issues are taken on board in the LR index. Concerning environmentalism I use per416 (Anti-Growth Economy: positive) and per501 (Environmental Protection: positive). Using these two variables as independent variables and LR_core as dependent variable gives this assumption some support. In most countries green issues are becoming part of the left ideology. This is so, for example, in Australia since 1996, Austria since 1994, Germany since 1987, Italy since 1992, New Zealand since 1999, Sweden since 1991, and Switzerland since In other countries green issues are aligned to the left over the whole period of analysis (Canada, the Netherlands, and Norway). However, there are exceptions. In Belgium and France green issues belong to the ideological right and for Iceland, Ireland, Luxembourg, the United Kingdom and the United States green issues do not belong to the left-right dimension at all. I obtain an even more diverse picture when looking at EU membership (per108 and per110 that is European Integration: positive and negative). For most countries the pro-eu position is aligned with the right and anti-eu sentiments align with the left. First of all, this is contrary to the assumption expressed above by Volkens et al. However, in Belgium it was vice versa from 1946 to 1974 and the left only integrated an anti-eu stand starting in In Austria, Finland, Ireland, Norway, and Switzerland, European integration is not an issue of the left-right dimension at all. In the Netherlands, EU membership was neutral until the last two elections (2006 and 2010) when anti-eu positions have gone together with the right ideology. Even if pro- or anti-eu positions were engaged, it was so seldom that the EU issue divided the left and right in a way that both ideologies aligned with different positions simultaneously. However, this was the case in Belgium from , in Denmark , in France in 1978, in Italy in 1979 and in 1992, in Luxembourg , in Portugal for the whole period, in Spain since 2008, and in Sweden since The example of Sweden in Table 3 shows how extra-statements are aligned to the left-right dimension over time. 14

15 Table 3: LR_Plus Statements in Sweden in the Elections from 1982 to 2010 per* Variable Name** per101 foreign special + Right Right per102 foreign special - per103 anti-imperialism + per104 military + Right Right Right Right Right Right Right Right Right per105 military - Left Left Left Left Left Left Left per106 peace + Left Left Left Left Left per107 internat + Left per108 europe + Right Right Right Right Right Right per109 internat - per110 europe - Left Left Left Left per201 freedom-hum rights + Right Right Right Right Right Right Right Right Right per202 democracy + Left Left Left Left Left Left Left Left Left per203 constitut + per204 constitut - per301 decentral + per302 central + per303 gov-admin efficiency + Right Right Right Right Right Right Right Right Right per304 pol corruption - per305 pol authority + per402 incentives + Right Right Right Right Right Right Right Right Right per405 corporatism + per406 protectionism + Left Left Left Left Left Left per407 protectionism - per408 economic goals Left Left per409 keynesian demand + per410 productivity + Left per411 infrastructure + per415 marxist analysis + per416 anti-growth econ + per501 environmental protection + Left Left Left Left Left Left per502 culture + per503 social justice + Left Left Left Left Left Left Left Left Left per504 welfare + Right Left Left Left Left Left Left Left per506 education + Right Right Right Right Right Right Right per507 education - per602 nat way life - per604 trad moral - per605 law and order + Right Right Right Right Right Right Right Right Right per607 multicult + per608 multicult - per701 labour + Left Left Left Left Left Left Left Left Left per702 labour - per703 farmers + Right Right Right per704 middle class + per705 minority groups + Left Left Left per706 non-economic groups + Left Left Left Left Left Explanations: * per refers to the items of the CMP; ** refers to the variable labels of the CMP (see Budge et al. 2001; Klingemann et al. 2006). 15

16 This analysis shows that the left-right dimension is a living dimension and that it continually integrates new issues. This integration is not uniform in all the countries or over time. However, this finding is not new. Bartolini (2000) has shown how left and right were expressed differently in different countries and over time and the classic example is probably the debate about mass strikes between Rosa Luxemburg and Karl Kausky in the German labor movement which spread with different emphasis all over Europe and beyond. The Relevance of the Left-Right Dimension over Time and Space Identifying the relevance or importance of an ideological dimension is essential for many party ideological analyses such as in spatial analysis (Benoit and Laver 2006; Warwick 2006; Tsebelis 2002). However, in order to measure the importance of the left-right dimension, it is also essential to estimate the meaning of left and right in a society and period of time. There is an easy way of identifying the importance of the left-right dimension in countries over time. This can be done by summing up the percentage score of the left and right statements and dividing this number by all statements mentioned. In this way I get a simple percentage value for the left-right dimension (LR_imp). This index shows the strength of the affinity of the left-right dimension over time and in different countries. This score can be compared to the importance of the core left-right statements (LR_core_imp) and the importance of the extra-statements (LR_plus_imp). Comparing both importance scores gives us further insight into the changing meaning of left and right across time and space. Table 4: The Importance of Core and Additional Left-Right Issues over Time and Countries LR LR_core LR_plus LR LR_core LR_plus LR LR_core LR_plus Australia Austria Belgium Canada Denmark Finland France Germany Iceland Ireland Italy Japan Luxembourg Netherlands New Zealand Norway Sweden Switzerland United Kingdom United States Total