JOURNAL OF COMPARATIVE POLITICS 81 COMPARISON OF ELECTORAL MANIFESTOS' ISSUE STRUCTURES IN CONTEMPORARY DEMOCRACIES – THE METHODOLOGICAL PERSPECTIVE Samo KROPIVNIK 1 ………………………………………………………………………………… From methodological perspective we address a substantive question of political science research. What are similarities and differences in issue structures in contemporary democracies? Issue structures are recognized on the basis of the content of national electoral programmes (manifestos) in the period from 1992 to 2003. In analyses, we use multivariate clustering methods first to obtain groups of countries as aggregates of manifestos and then to obtain groups of manifestos. Eventually countries are categorised drawing on the importance of four different types of manifestos in a country. The results are, whenever possible, graphically presented as line graphs, dendrograms and a galaxy. We find out that only in some countries manifestos clearly belong to one dominant type, and therefore only these countries make sense as aggregates of manifestos in comparative studies. Unfortunately for studies taking countries merely as aggregates of manifestos, the heterogeneous countries are in majority. Key words: political party program, elections, issue positions, issue domains, MARPOR, comparative studies, country profiles, Euclidean distances, Galaxy, agglomerative hierarchical clustering, k-means clustering. 1 POLITICAL PARTY MANIFESTOS AND ISSUE STRUCTURES IN SOCIETIES Political party manifestos (electoral programmes, party programmes or electoral manifestos; we use all these terms as synonyms) play a significant 1 Samo Kropivnik is Associate Professor of Social Sciences Research Methodology at the Faculty of Social Sciences, University of Ljubljana, Slovenia where he teaches courses on methodology and research methods. His research focus is on a choice of a research design in applied studies and on creative use of descriptive approaches and multivariate methods in social sciences. Since 2008 he is local academic coordinator of ECPR Summer School in Methods and Techniques. Contact: samo.kropivnik@fdv.uni-lj.si. JOURNAL OF COMPARATIVE POLITICS 82 role in electoral democracies, directly in electoral campaigns 2 and in broader sense as a constitutive part of a political system. 3 They recognize the importance of critical issues, develop a party position on them, set the course of actions a party will take if elected, unite a party internally and, last but not least, advise party activists and supporters as well as inform the general electorate. 4 In party manifestos, political and policy ideas, positions and goals are recorded, publicised and documented for analysis. Therefore comparisons among national party manifestos expose similarities and differences among countries in issue structures and even in cleavages that form the core of political science comparative studies. 5 In Europe, parliament party manifestos have been systematically collected since 1979.Their content has been coded, and the data are available for analysis (project MARPOR, previously MRP and CMP). 6 Each parliamentary party programme is characterised according to its match with a standardised set of carefully selected, precisely defined and theoretically relevant issue positions. Subsequently, position codes are merged in seven mutually exclusive and theoretically exhaustive domains that are defined in Table 1. 7 Obviously, for each document, contextual data are also available on political party, party family, country and election year. TABLE 1: ISSUE DOMAINS AND ISSUE POSITIONS (CODES) 2 E. g. David Farrell and Rudiger Schmitt-Beck (eds.), Do Political Campaigns Matter?(New York: Routledge, 2002) and John W. Kingdon, Agendas, Alternatives, and Public Policies (2nd edition) (New York: Longman, 1995). 3 E.g. Richard S. Katz, Democracy and Elections (Oxford: Oxford University Press, 1997) and Susan C. Stokes, “Political Parties and Democracy,” Annual Review of Political Science, 2 (1999), 243–267. 4 See Hans-Dieter Klingemann, Richard I. Hoffebert and Ian Budge, “A Theory of Democratic Policymaking,” in Parties, Policies and Democracy, ed. Hans-Dieter Klingemann et al (Boulder: Westview Press, 1994), 240–270; Ian Budge and Michael Laver, “Policy, Ideology and Party Distance: Analysis of Election Programmes in 19 Democracies,” Legislative Studies Quarterly, 11, 4 (1986), 607– 617; Ian Budge and Richard I. Hofferbert, “Mandates and Policy Outputs: US Party Platforms and Federal Expenditures,” American Political Science Review, 84 (1990), 111–131. See also David Robertson, A Theory of Party Competition(London: John Wiley & Sons, 1976) and Richard Rose, Do Parties Make a Difference (2nd edition)(London: Macmillan, 1984). 5 Importance of issue structures and cleavages in societies is put forward in Seymur Martin Lipset and Stein Rokkan (eds.), Party Systems and Voter Alignments (New York: Free Press, 1967). See also Giovanni Sartori, Parties and Party Systems: A Framework for Analysis, Vol. 1(Cambridge: Cambridge University Press, 1976). 6 See Andrea Volkens et al, The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR) (Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB), 2012). 7 For detailed description of data creation process see Annika Werner and Andrea Volkens, Manifesto Coding Instructions (3 rd fully revised version) (Berlin: Wissenschaftszentrum Berlin für Sozialforschung, 2009). JOURNAL OF COMPARATIVE POLITICS 83 Domains shares are the most characteristic and most valuable feature of the dataset since when considered jointly, domains exhaustively cover existing national (country specific) policy and political issues and simultaneously offer universal comparison in time and space (e.g. among policy arenas, among countries). They have been created for this purpose and their validity has been repeatedly evaluated and confirmed in various comparative studies. 8 2 THE AIM AND THE STRUCTURE OF THE RESEARCH Drawing on the MARPOR dataset, comparisons can be carried out between national political parties competing for votes in a certain election year or even throughout longer periods with more election cycles involved. 9 More ambitiously, countries (national policy arenas) can be compared on the basis of domain shares recognized from national political parties’ manifestos. In national electoral arenas, political parties compete with each other, and thus they unavoidably react to each other. Therefore, their manifestos ideally reflect their own goals as well as their responses to initiatives of other major parties. Only when manifestos are combined in a national collection of issue positions they have the potential to comprehensively describe a country’s specifics in a domain structure. What we claim and intend to demonstrate in the paper is that in order to describe competently and comprehensively the issue structure content of electoral campaigning in any country in comparison with others, one has to consider both a) the country level, i.e., aggregated shares of domains in the analysed countries and b) the party programmes level, i.e., exact shares of domains in each party manifesto involved. The latter is required to estimate differences between programmes in a country in order to realize how the nationalization (or modernization) of party manifestos has progressed so far. 10 In our empirical study, 11 we begin with comparisons among countries as aggregates of manifestos and focus on Slovenia 12 as a referential unit whenever required. Since Slovenia became an electoral democracy only in 1990, we limit the study to the period between 1990 and 2003, the latter being the last election year for which complete MARPOR data are available at the time of writing this paper. We employ graphically presented country profiles (line graphs) and hierarchical agglomerative clustering procedures (dendrograms) to both give a general overview and to recognise similarities and differences between Slovenia and the other fifty countries that participated in the project during the period analysed. Next, in order to take 8 See e.g. Ian Budge et al, Mapping Policy Preferences: Estimates for Parties, Electors, and Governments 1945–1998 (Oxford: Oxford University Press, 2001) and Hans-Dieter Klingemann et al, Mapping policy preferences II: estimates for parties, electors, and governments in Eastern Europe, European Union, and OECD 1990–2003 (Oxford: Oxford University Press, 2006). Additional studies are documented on project website, see Universität Mannheim, MZES. "Projekt Euromanifestos", Universität Mannheim. Available at http://www.mzes.uni-mannheim.de/projekte/manifestos/ (24 March 2012). 9 Simona Kustec Lipicer and Samo Kropivnik, “Dimensions of party electoral programmes: Slovenian experience,” Journal of Comparative Politics, 4, 1 (2011), 52–75. 10 C.f. Seymur Martin Lipset and Stein Rokkan (eds.), Party Systems and Voter Alignments (New York: Free Press, 1967). 11 The data collection and a part of the analyses were completed in the framework of the research project, “Pre-election campaign and democratic evolution of state and society,” conducted at the Centre for Political Science Research at the Institute of Social Sciences, University of Ljubljana and financed by the Slovenian Research Agency (1/5/2009–30/4/2012). 12 For characteristics of electoral competition see e.g. Danica Fink-Hafner, “Slovenia,” European Journal of Political Research, 42 (2003), 1078–1086 and Simona Kustec Lipicer, “Volilna kampanja na lokalni ravni: poskus konceptualno-metodološke opredelitve na primeru Slovenije,” Lex localis, 5, 1 (2007), 105–122. JOURNAL OF COMPARATIVE POLITICS 84 differences inside countries into account, we return to original (basic, not aggregated) units of analysis and cluster individual party manifestos (almost 1300 units in the selected period) into groups (ideal types) using hierarchical agglomerative clustering methods and K-means clustering method. In a contingency table the obtained groups are split according to the manifestos’ countries of origin to estimate variation inside countries. Furthermore, we employ Euclidean distances to present relations between Slovenia and all other countries in a Galaxy, a graphical format developed for that purpose. The central (referential) country is represented as the Sun and all other countries are represented as different planets allocated around the Sun proportionally to Euclidean distances. Countries (i.e., planets) are depicted according to the issue domains structure of their manifestos, taking into account the type and the level of homogeneity, which are both recognized drawing on the results of cluster analyses of manifestos. 3 COUNTRIES AS AGGREGATES OF MANIFESTOS In the table below (Table 2) average issue domains shares in the period from 1990 to 2003 are calculated from the MARPOR database for each country. The shares are modified so that their sum is always 100 percent in order to enable comparisons between countries. This is achieved by excluding uncoded sentences, i.e., sentences without any issue-oriented content, such as general statements, introductory remarks etc. On average, there were 8 percent of uncoded sentences in a program. The last row of the table shows the average shares for all countries combined. TABLE 2: AVERAGE ISSUE DOMAINS SHARES IN 51 ANALYSED COUNTRIES IN THE PERIOD FROM 1990 TO 2003 13 13 Extracted and recalculated from Andrea Volkens et al, The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR) (Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB), 2012). JOURNAL OF COMPARATIVE POLITICS 85 Issue domains shares for e.g. Slovenia can be easily read from the table. It is clear that the domains shown are not equally important (i.e., if a domain share in a document is taken as the indicator of importance). The economy domain and welfare and quality of life domain are overrepresented and external relations, freedom and democracy, and fabric of society domains are more or less underrepresented (according to an ideal 14.3 percent share in the case of uniform distribution). Apart from that basic description, we cannot make any conclusions regarding commonness or uniqueness of the domain structure of Slovene manifestos. Taken separately domains have no substantial meaning. Clearly, we have to compare them with the overall structure (average) and further to all other countries. Figure 1, depicting national profiles and the average profile of domains importance, is created for that purpose. The Slovene profile is red and all other profiles are black. Among the black profile lines, the thick line shows the average profile. FIGURE 1: NATIONAL PROFILES, SLOVENE PROFILE AND THE AVERAGE PROFILE OF THE IMPORTANCE OF ISSUE DOMAINS (focused graphical presentation of Table 2) Graphical presentations convey more pieces of information at the same time and enable multiple comparisons on various levels. 14 Figure 2 readily shows that there is quite a variation in national profiles and that although the Slovene profile is not very different from the average profile, its specific characteristics cannot be neglected. Its most distinctive features are a considerably higher share of welfare and quality of life domain and substantially lower shares of external relations and economy domains. Regarding all three distinctive domains, the Slovene profile is quite extreme in comparison with other countries. Only Sweden and New Zealand pay more attention to the welfare and quality of life domain; only Australia and New Zealand pay less attention to the external relations domain; and Azerbaijan alone to the economy domain (these countries are recognized as shown in Table 2). Apart from these Slovene characteristics, less obvious distinguishing features can be recognized only by more detailed and more formal analyses. Among them, direct comparisons domain by domain to determine particularities are straightforward and could be performed drawing on Table 14 Gary T. Henry, Graphing Data. Techniques for Display and Analysis (London: Sage, 1995); see also Edward R. Tufte, The Visual Display of Quantitative Information(Cheshire, CT: Graphic Press, 1983). JOURNAL OF COMPARATIVE POLITICS 86 2, in the same way as demonstrated above for the most obvious Slovene specific characteristics. However, a more productive method is to employ a measure of profiles (countries) similarities or dissimilarities according to all seven domains taken jointly. Among possible measures, Euclidean distance 15 proved ideal for the purpose as it captures resemblances and differences in profiles in a way that is similar to the human eye and mind. It pays less attention to small differences and puts more stress on large differences. For example, small differences in all seven domains will result in lower Euclidean distance between two profiles than an exact match in six domains and a noticeable difference in one domain, even if a noticeable difference is much smaller than the sum of the seven small differences in the first case. Euclidean distance precisely and realistically captures differences in profiles and expresses them as numerical values that can be used for more formal comparisons among manifesto issue domains structures in different countries. Furthermore, drawing on Euclidean distance and employing clustering methods, 16 we are able to uncover patterns, i.e., recognizable and relevant combinations that draw more attention to certain issue domains and less to others, which are typical of certain countries. In other words, we not only can realize how similar each country is to others but also can put together those countries that are very similar to each other: countries in a group have to be as similar as possible and groups (types) as different (unique) as possible. Because this is a problem of optimization, both statistical and conceptual criteria affect the final solution and we understand the involvement of conceptual criteria to be an advantage. Thus, we can create a relevant typology of party manifesto structures, associate each country with the most appropriate type, and realize a country position in a type (group) as being more central, more peripheral or anywhere in the middle. In the case of agglomerative hierarchical clustering, a country membership and position can be understandably presented in the form of an agglomeration tree (dendrogram), i.e., a popular graphical presentation of a clustering procedure and result. To our knowledge, there is no other single approach or multivariate method that conveys that many relevant pieces of information in so condensed and straightforward form. In the dendrogram below, countries are clustered according to their average issue domains structures on the basis of squared Euclidean distance (large differences become even more important) and the Ward method (balanced clustering criteria that respect group homogeneity and between-groups differences). 15 See e.g., Richard A. Johnson and Dean W. Wichern, Applied Multivariate Statistical Analysis (London: Prentice Hall, 1992). 16 Regarding the method description and clustering notions we refer to ibid. JOURNAL OF COMPARATIVE POLITICS 87 FIGURE 2: AGGLOMERATIVE HIERARCHICAL CLUSTERING TREE OF COUNTRIES The dendrogram shown in Figure 2 yields the following: First, it is reasonable to distinguish from two to five types of countries. However, three types seems to be the most balanced solution, as two are too superficial, four are too close to three or five, and five adds another small group to the previous four groups, among which one is already small. Second, one group is always the same. The most stable group is composed of thirteen countries at the bottom of the dendrogram. Since all countries have joined the group at a low level of dissimilarity, the group is very homogeneous. Only on a very low level of dissimilarity (which is irrelevant) can we recognize four subgroups. JOURNAL OF COMPARATIVE POLITICS 88 Third, Slovenia is in the most stable and homogeneous group. Inside the group, it most resembles Finland and New Zealand, followed by three subgroups of countries, the first including Luxemburg, Ireland, Portugal and Malta, the second France, Norway and Sweden and the last Germany, Iceland and Cyprus. Fourth, next from the bottom to the top of the dendrogram shows the least stable group, which becomes a group in the case of the three-group solution and splits in two subgroups in the case of the five-group solution. Clearly it is less homogeneous. The members comprise eleven countries from Israel to Albania (from the bottom of the dendrogram up). Fifth, the largest group includes twenty-seven countries from Estonia to Japan (listed from the top down) and splits into two unequal subgroups in the case of the four- or five-groups solutions. It is a large and a heterogeneous group. To enable recognition of the character of each type (group), typical (ideal or average) profiles indicating the importance of issue domains in each group are presented in the graph below. The average profile of all countries is included as well. FIGURE 3: PROFILES OF THREE TYPES OF COUNTRIES AND THE AVERAGE PROFILE Differences between the groups and the average as well as differences among the groups are modest; however characteristics of the types can be recognized. The most stable group (the one including Slovenia), presented by the red profile, is the most specific because of the highest and most exposed average share of the welfare and quality of life domain in party manifestos, accompanied by lower shares of the fabric of society, freedom and democracy and political system domains. Clearly, for the countries in the “red” group, welfare and quality of life domain topics are the most important issue. The least stable group, presented by the green profile (including eleven countries in the middle of the dendrogram), is characterized by the highest share of two domains, namely the freedom and democracy domain and the fabric of society domain. In contrast to the first group, welfare and quality of life issues are the least important for the “green” countries. The largest group of countries, presented by the orange profile, is close to the average (obviously, due to the size) and is specific only because of the highest (but not very eye-catching) interest paid to the political system topics. JOURNAL OF COMPARATIVE POLITICS 89 4 COUNTRIES AS COMPILATIONS OF MANIFESTOS The classification of countries on the basis of average shares of issue domains provides fundamental information for comparisons among countries but ignores differences inside countries, i.e., the level of nationalization or modernization of party manifesto structures. It is important to determine whether all party manifestos are similarly structured and therefore almost identical to the average structure that credibly represents a country, or manifestos demonstrate significantly different structures and consequently the average is nothing more than an artificial construct that doesn’t truly represent a country. In order to take the differences inside countries into account, in the second step of the study we return to original (basic, not aggregated) units of analysis and cluster individual party manifestos (almost 1300 units in the selected period) into groups (ideal types) using hierarchical agglomerative clustering method in the same way as we did before using countries as units of analysis. FIGURE 4: AGGLOMERATIVE HIERARCHICAL CLUSTERING TREE OF PARTY ELECTORAL PROGRAMMES JOURNAL OF COMPARATIVE POLITICS 90 The dendrogram above shows that the four-group solution is the most balanced and reasonable. Two- and three-groups solutions are too superficial since it is obvious that quite different groups are still joined together, five is too close to six, and the six-groups solution appears to be too particular since only a few manifestos split from two of the previous four groups, and the levels of dissimilarity are quite low. Additionally, the four- group solution has been confirmed by K-means method, as the largest drop in Ward criterion function arises when four groups replace three groups. Further, the solution (i.e. the classification of manifestos into four groups) has been optimized by K-means method. Average importance profiles of the issue domains of each group are presented in the graph below to enable recognition of each type’s character in the same way as in the case of countries (Figure 3). The average profile of all manifestos (nearly identical to the average of countries) is also included. FIGURE 5: PROFILES OF FOUR TYPES OF PARTY MANIFESTOS AND THE AVERAGE PROFILE Differences between groups and the average as well as differences among groups are now far more substantial, and type characteristics can be recognized more clearly. Regarding their interpretation, three profiles are the same as profiles of groups of countries shown in Figure 3. As before, the red profile represents the group of manifestos with the highest average share of welfare and quality of life domain. As the Figures 3 and 5 show both red profiles appear to be almost the same, which indicates that the group of manifestos highly resembles the group of countries. Notably, the “red” group is the largest one, containing 488 documents. The green profile is also very much the same as shown in Figure 3, i.e., it is characterized by the highest share of two domains, namely the freedom and democracy domain and the fabric of society domain. However, noticeably, in the case of clustering manifestos, the type has a more distinctive character since differences from the average profile and from the profiles of other groups is now far more obvious. The group contains 235 party programmes, which makes it a middle-sized group. Although the third, orange group of manifestos does not stand out in Figure 3, it also becomes far more distinctive in Figure 5 as the group’s interest paid to the political system domain topics is now more distinguishing. However, with 162 documents, it becomes the smallest group. The fourth, blue profiles new and typical only of manifesto clustering. Notably, the blue group is the second largest group, containing 408 JOURNAL OF COMPARATIVE POLITICS 91 manifestos, so it cannot be overlooked in any case. Its main characteristic is the highest share in the economy domain, but in general, the blue profile is slightly less distinctive and closer to the average. In the contingency table below (Table 3), the obtained four groups of manifestos (columns) are split according to the origin countries of the manifestos (rows) both to estimate differences in manifesto structures inside countries and to categorize countries according to in-country characteristics of manifesto structures. The cells contain row percentages and the total is shown in the last row. Countries (rows) are arranged according to the three types of countries previously established (Figures 2 and 3). Country type is marked in front of country name in a consistent colour, as shown in Figure 3: 1/in red for welfare and quality of life country type; 2/ in green for freedom and democracy plus the fabric of society country type; and 3/ in orange for political system country type. The country’s name is written in a colour consistent with its in-country characteristic type of manifesto structure (same colours as in Figure 5). Also the column headings in Table 3 are shown in the same colours: red for welfare and quality of life manifesto type; green for freedom and democracy plus fabric of society manifesto type; orange for political system manifesto type; and blue for economy manifesto type. If a country cannot be categorized in a single category because it has two characteristic types of manifestos, both colours are used (half of the name is in one colour, and the other half is in the other colour). However, the first part of the name is in the dominant characteristic type colour. If a country cannot be categorized because of the lack of distinctive characteristics, its name is printed in black. In-country characteristics of manifesto type are recognized by row percentages to describe the relative impact of a manifesto type in a country. Cells containing high row percentages are filled with colours. Red, orange and yellow are used to indicate an absolute dominance (a majority) of a certain type of manifesto on different intervals: red is used for extremely high shares, making all other types irrelevant (70 percent or more); dark orange is used for very high shares on a slightly lower interval, making all other types hard to affect the country type (60 percent to 70 percent); and yellow is used for high shares on an even lower interval that allows other types to be relatively influential (50 percent to 60 percent). All other shares representing the relative prevalence of a certain manifesto type in a country are coloured grey if they are at least 10 percentage points higher than the total. The final recognition of in-country characteristics of manifesto type, i.e. the categorization of countries, is based on a subjective estimation of the importance of a manifesto type in a country, which draws on both the absolute supremacy of a type on different levels and on the relative over- presence of manifesto types (one or more). JOURNAL OF COMPARATIVE POLITICS 92 TABLE 3: COUNTRIES BY TYPES OF MANIFESTOS Table 3 reveals the following information: First, starting from previously established groups of countries, the first group marked as 1/and characterized as the welfare and quality of life domain group proved to be very well founded and perfectly stable. For the second time, all thirteen countries are categorized as a group, and the dominance of the welfare and quality of life issue domain is in all cases absolute and on a high scale at mostly over 70 percent (red cells). The only less convincing case is Ireland, where we observe a 50/50 split between the group specific domain and the economy domain. The second group, marked as 2/ and labelled the freedom and democracy and the fabric of society domains group becomes less definite and less stable since only three out of eleven countries are categorized in the same group on the basis of a convincing absolute dominance of the characteristic two domains (orange cells).One is included on the basis of their relative prevalence (grey cell); four are split between the type and other types, and the final three are categorized in another group (one, i.e., GDR clearly in the welfare and quality of life domain group with 86 percent of characteristic domain JOURNAL OF COMPARATIVE POLITICS 93 manifestos). In the case of the third group, marked as 3/ and named the political system domain group, the evidence is even weaker. Only two countries can be categorized in the same group, one on the basis of the absolute prevalence of political system domain manifestos and the other on the basis of their relative over-presence. Among the others, seven are split between this type and another type, sixteen are categorized in one of the other three types, and two remain uncategorized (there is not even a relatively higher segment of any type). Second, the welfare and quality of life group as well as the freedom and democracy and fabric of society domains group show new members. The first of the two includes more new countries, and as a rule they enter with an absolute dominance of the characteristic domain type (red, orange or yellow cells), and only two are split between manifesto types. The second group has fewer new countries and typically they join with merely relative over-presence of the type (grey cells). The political system group has no new members, but it loses a great number of countries due to their relocation into the other three groups. The fourth group, the economy domain group, materializes from the previous freedom and democracy and fabric of society domains group of countries and from the weakening political system domain group. The economy domain type of manifesto prevalence is absolute (orange and yellow cells) as well as relative (grey cells). Third, twelve countries cannot be categorized in a single type since party manifestos belong to two significant types. Additionally, two countries cannot be classified since their manifestos express no pattern. These results do not occur when countries are classified on the basis of an average manifesto structure (so they are overlooked). Fourth and finally, based on previous points, it is safe to conclude that a categorization of countries drawing on the classification of party manifestos yields better results than a classification of countries on the basis of their average manifesto profiles. The results of both procedures are reasonably similar only in the case of the welfare and quality of life group, although this group gains new country members with clear crucial domain dominance, which were misclassified when the countries were clustered. Additionally, when countries are categorized on the basis of the influence of in-country types of party manifestos, countries of the same type can be distinguished and presented according to the strength of their link to a type, i.e., according to the level of prevalence of a manifesto type, which is expressed in Table 3as different cell colours. Moreover, countries can be categorized as split between types, which is a unique but very realistic feature. Similar is true for uncategorized countries. 5 APPLYING THE RESULTS – THE GALAXY To employ features described in the last paragraph and provide a clear, conclusive picture of the manifesto structures of Slovenian parliamentary parties in comparison with those of other countries, we depict a galaxy i.e. a graph developed for that purpose. The central (referential) country, in our case Slovenia, is shown as the Sun and all other countries are shown as different planets allocated around the Sun proportionate to Euclidean distances. Countries (planets) are depicted according to their manifesto domain structures, taking into account the type and the level of homogeneity, both of which are recognized drawing on clustering results for the manifestos. The countries (planets) are coloured according to their prevailing types of manifesto structures as follows:  red for the welfare and quality of life domain group; JOURNAL OF COMPARATIVE POLITICS 94  green for the freedom and democracy and fabric of society domains group;  orange for the political system group;  blue for the economy domain group;  grey for the two uncategorisable countries. If there is more than one characteristic issue domain, a country name is written in the colour of the other, as a rule the less characteristic one. Different shapes of planets represent different levels of manifesto structure homogeneity in a country, which can also be taken as an indicator of the strength of the country’s categorization:  a circle is used for the highest homogeneity (absolute dominance of a single manifesto structure type with an over 70 percent share);  a triangle stands for high homogeneity (absolute dominance of a single manifesto type with a 60 percent to 70 percent share);  a square represents modest homogeneity (absolute dominance of a single manifesto type with a 50 percent to 60 percent share and the absence of any other over-presented type);  a rhombus is used to indicate modest variety (relative prevalence of a single manifesto type with a share at least 10 percentage points higher than the total and the absence of any other over-presented type;  a star represents a clear split in manifesto types (two types of manifestos are characteristic for a country). FIGURE 6: THE GALAXY WITH SLOVENIA AS THE REFERENTIAL COUNTRY Drawing on the Galaxy shown in Figure 6 and comparing Slovenia with other countries, one can recognize Slovenia as a country with a clear dominance of parliamentary party manifestos that favour the welfare and quality of life JOURNAL OF COMPARATIVE POLITICS 95 domain. Most Slovenian party manifestos, more than 70 percent, are classified as emphasizing the welfare and quality of life domain (red circle). Therefore, Slovenia can be categorized as a country with a homogenous party manifesto structure that is strongly associated with the welfare and quality of life issues (a strong and unmistakable case of group membership). Countries that are most similar to Slovenia are Luxemburg, Great Britain and New Zealand. All three are strong cases of belonging to the welfare and quality of life domain type (red circles). The next two similar countries are Finland and Netherlands, both belonging to the same welfare and quality of life manifesto type, but the characteristic domain prevalence is lower (red triangles). The further from Slovenia we move, the less welfare and quality of life type countries and the more heterogeneous countries we find. Readers can interpret other countries categorisation and their level of similarity with Slovenia in the same manner as the graph speaks for itself. 6 CONCLUSION The clustering approach has enabled us to compare party manifestos as both individual documents and country aggregates. It has also provided a method to estimate the level of similarity between units of interest on both levels of comparison and in general. From methodological perspective, the strong points of the applied approach can be summarized as follows: First, political party manifestos and countries are analysed according to all seven issue domains simultaneously in a multivariate manner- the opposite would be to consider each domain separately. Second, Euclidean distance represents a very realistic view of the level of similarity or difference between units (countries or manifestos), which is close to what we understand as distance in everyday life. In addition, clustering algorithms use Euclidean distance in an easy to understand way and produce vivid graphical outputs, which make the research results both comprehensible and convincing. The opposite would be to use latent concepts and rather abstract notions of covariation in multidimensional space, which are difficult to comprehend for less empirically oriented scholars. Third, in our quest for the best classification we can choose among different suggested solutions from more general to more precise (less types equals less in-group homogeneity and more differences among groups, while more types equals less differences among groups and more homogeneous groups). In defining the best solution, i.e., one that is subjectively considered the most balanced, we can (and we usually have to) apply additional, contextual and theoretical criteria (e.g., the minimum size of average differences in percentage points that we understand as a difference and don’t neglect in interpretations). The use of contextual and theoretical principles together with statistical indicators leads to more convincing results. Fourth, in our case the clustering results on the level of manifestos are used to categorize countries, drawing on our understanding of the size and the meaning of structures of in-country manifesto types. Fifth, in line with the prevailing Euclidean space based analyses, the Galaxy (Figure 6) vividly summarizes the results and enables a focused comparison of a selected country with all other countries without any falsification. The opposite would be, e.g., any kind of projection of multiple space into two dimensions. JOURNAL OF COMPARATIVE POLITICS 96 In more substantial manner, regarding past, present and future political science manifesto studies of comparisons among countries, the most important conclusion is that in some countries manifestos belong to one dominant type, and therefore these countries make sense as aggregates of manifestos. These countries are authentically represented by an average issue domains structure. On the contrary, in other countries, manifestos (parties) are clearly split between types. Therefore, analysing such countries as aggregates of manifestos doesn’t make sense because the structures of their manifestos are too different. Moreover, an average structure inadequately represents actual issue structure. Unfortunately for studies taking countries as aggregates of manifestos, the latter, i.e. the heterogeneous countries are in majority. REFERENCES Budge, Ian and Michael Laver. “Policy, Ideology and Party Distance: Analysis of Election Programmes in 19 Democracies.” Legislative Studies Quarterly, 11, 4 (1986): 607–617. Budge, Ian and Richard I. Hofferbert. “Mandates and Policy Outputs: US Party Platforms and Federal Expenditures.” American Political Science Review, 84 (1990): 111 – 131. Budge, Ian, Hans-Dieter Klingemann, Andrea Volkens, Judith Bara and Eric Tanenbaum. Mapping Policy Preferences: Estimates for Parties, Electors, and Governments 1945–1998. Oxford: Oxford University Press, 2001. Farrell, David and Rudiger Schmitt-Beck (eds.). Do Political Campaigns Matter? New York: Routledge, 2002. Fink-Hafner, Danica. “Slovenia.” European Journal of Political Research, 42 (2003): 1078–1086. Henry, T. Gary. Graphing Data. Techniques for Display and Analysis. London: Sage, 1995. Johnson, A. Richard and Dean W. Wichern. Applied Multivariate Statistical Analysis. London: Prentice Hall, 1992. Katz, S. Richard. Democracy and Elections. Oxford: Oxford University Press, 1997. Kingdon, W. John. Agendas, Alternatives, and Public Policies (2 nd edition). New York: Longman, 1995. Klingemann, Hans-Dieter, Andrea Volkens, Judith Bara, Ian Budge and Michael D. McDonald. Mapping policy preferences II: estimates for parties, electors, and governments in Eastern Europe, European Union, and OECD 1990–2003. Oxford: Oxford University Press, 2006. Klingemann, Hans-Dieter, Richard I. Hoffebert and Ian Budge. “A Theory of Democratic Policymaking.” In Parties, Policies and Democracy, edited by Hans-Dieter Klingemann, Richard I. Hoffebert and Ian Budge, 240 - 270. Boulder: Westview Press, 1994. Kustec Lipicer, Simona. “Volilna kampanja na lokalni ravni: poskus konceptualno-metodološke opredelitve na primeru Slovenije.” Lex localis, 5, 1 (2007): 105–122. Kustec Lipicer, Simona and Kropivnik, Samo. “Dimensions of party electoral programmes: Slovenian experience.” Journal of Comparative Politics, 4, 1 (2011): 52–75. Lipset, Seymur Martin and Stein Rokkan (eds.). Party Systems and Voter Alignments. New York: Free Press, 1967. Robertson, David. A Theory of Party Competition. London: John Wiley & Sons, 1976. JOURNAL OF COMPARATIVE POLITICS 97 Rose, Richard. Do Parties Make a Difference (2 nd edition). London: Macmillan,1984. Sartori, Giovanni. Parties and Party Systems: A Framework for Analysis, Vol. 1. Cambridge: Cambridge University Press, 1976. Stokes, C. Susan. “Political Parties and Democracy.” Annual Review of Political Science, 2 (1999): 243–267. Tufte, R. Edward. The Visual Display of Quantitative Information. Cheshire, CT: Graphic Press, 1983. Universität Mannheim, MZES."Projekt Euromanifestos", Universität Mannheim. Available at http://www.mzes.uni-mannheim.de/projekte/ manifestos/ (24 March 2012). Volkens, Andrea, Onawa Lacewell, Pola Lehmann, Sven Regel, Henrike Schultze and Annika Werner. The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR). Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB), 2012. Volkens, Andrea, Onawa Lacewell, Pola Lehmann, Sven Regel, Henrike Schultze and Annika Werner. The Manifesto Document Collection. Manifesto Project (MRG/CMP/MARPOR). Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB), 2012. Werner, Annika and Andrea Volkens. Manifesto Coding Instructions (3 rd fully revised version). Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB), 2009.