of an electoral system, that can accommodate more or less parties, might also be influenced by this environment. As they put it – parties minus issues equals one.
But there is another way that the number of parties and the number of dimensions is related, which is evident if we take a more data-centred perspective on the problem. If we talk strictly about how parties differentiate among themselves, then empirically speaking the first question should be, how can we see this difference? And here it should be kept in mind, that the dimensions on which parties are differentiated depends on the number of parties that we are looking at. Stoll (2010a) elaborates on this important point about the structure of political spaces being related to the number of parties that inhabit them. She emphasises that if we think from the perspective of party positions, then the dimensionality of political space is “the number of salient conflicts that are linearly independent given party positions on those conflicts, where two conflicts are independent or orthogonal if parties’
positions on one cannot be predicted using their positions on the other” (ibid., p. 409). She calls the latter the effective dimensionality.
This effective dimensionality is dependent on the number of parties that inhabit the space, as the differences between n objects can always be perfectly represented on at most n−1 dimensions (see section 2.1.2). Thus, a space that is needed to represent the differences between parties cannot have an effective dimensionality that is greater than the number of parties minus 1. This reflects the conclusion of Taagepera and Grofman (1985) noted above. In the case of two parties, the difference between them can be represented on a line – a single dimension.5 Whatever the number of issues that the parties differentiate themselves on, one side will always be associated with one party and the other side with the other party (or else there is no differentiation). All issues are aligned in the same direction and so there can be only one “visible” dimension of ideology where one party is on one side and the other party is on the other. In a similar way, all the possible differences between three parties can be perfectly represented on at most two dimensions and so forth.
parties and between parties and voters is how they perceive each other. These perceptions should matter for how parties decide to cooperate and how voters should decide to vote, should they care about political issues in their decisions at the ballot box.
The spatial notions of left and right have a long cultural and political history and therefore it is only natural they they were taken up by political scientists as a tool for making sense of politics. In their first incarnations in politics it seems that they were treated more as a categorical dimension while in political science they have been treated as a continuous dimension. But somewhere along they way it seems that it faded from attention that if we are interested in political behaviour, we should also be interested in how the relevant actors see the space. Instead, the core of this kind of thinking in political science has been built up on empirically tangent assumptions, according to a logic that follows more from a scientific and less from a phenomenal approach of political space.
In addition to studies which try to reconcile the ideas of the spatial model of political behaviour with empirical reality, we are thus left with a range of issues that remain unresolved, but which could have easier solutions if we keep in mind the distinction between a phenomenal and a scientific space.
The preference should be for inductive analyses, especially if we are interested in how people actually think about politics and not in what makes our formal models consistent or simple. If people’s or parties’ positions on issues do not unproblematically collapse into simpler dimensions, then perhaps this is simply a fact of the reality of politics. Maybe parties and maybe voters do have diverging understandings about the differences between themselves. And in line with how similarities between any kinds of objects are perceived, the ideological dimensions that we are able to see and discern do depend on the number of “objects” we are looking at and the latter in turn is related to the number of general social cleavages.
If we think of how the notion of space has been used in political science, then no matter what position we take, it is evident that there are many unresolved issues. From this more abstract level concerning the nature of political space they have carried over into many aspects of the empirical analysis of political space, which we turn to next.
CEUeTDCollection
Chapter 3
Empirical Knowledge of Political Space
In this context, position and the three dimensions were beside the point. Not, of course, that the category of space had been abolished. When I got up and walked about, I could do so quite normally, without misjudging the whereabouts of objects. Space was still there; but it had lost its predominance. The mind was primarily concerned, not with measures and locations, but with being and meaning.
– Aldous Huxley,The Doors of Perception
Thinking about the similarities or differences between anything, including parties, might well be impossible save in spatial terms, but this does not mean that there is only one way to do this. The spatial framework that we use can either be our own creation or created from within the processes that we are interested in. The former, scientific, approach is common if we are trying to create systems of meaning to explain phenomena that are themselves inherently “meaningless”, i.e. not animated by mental representations that are created within. In this case we presume a certain spatial structure to represent a phenomenon and judge it by the usefulness for whatever purpose we have in mind. The latter, phenomenal, approach is appropriate if we want to understand human perception or judgement. If we study party politics, then what matters is how parties themselves or voters perceive and understand politics. Thus, all else being equal, a phenomenal approach, one that does not prescribe or presume any kind of a political space, but seeks to capture as best as possible how this space is represented for the actors involved, should be preferred in party research.
Thinking about parties, and consequently measuring party politics, has in a way been unwittingly stuck between these two approaches. On the one hand we have the left-right spatial metaphor, which has deep and broad cultural roots and is related to the vary basic ways we understand the world or even to how our personalities are built. This basic line of division does have a phenomenal reality.
But much of our thinking about parties, especially the general spatial model, has been built around this metaphor without problematising its nature and without fully accounting for its limitations.
CEUeTDCollection
Even if it does exist in the minds of people, it is amenable to change over time and across different contexts and the political difference between parties can be evaluated across numerous other quality dimensions. It both does and does not make sense to assume the existence of such an overarching political dimension.
Even if in the analyses of party politics there is a strong sub-current that is focussed on political space as it might occur to the actors that are involved within that game, much of the empirical applications of measures of party politics are working with a version of the left-right dimension. This reflects the dominant currents in the spatial theory, but is also the fact that the left-right dimension is a simple solution. It is much easier to work with a unidimensional measure that can to some extent be assumed. All of this follows the logic of the “scientific” analysis of political space. But how would a phenomenal analysis of party politics look like, one which would apply some of the main methods that have been used elsewhere for the analysis of such space? It turns out that it would not be much more complicated, and would in many ways be more meaningful and revealing.
When in the previous chapter we focused on how to think about political spaces and how this has carried over into politics and political science, then the focus here is on how party politics has been actually measured. This chapter zooms in on the empirical analysis of party politics, with a particular focus on the manifesto data set as this has been overwhelmingly the primary source of information about the political profiles of parties and the spatial measures of party position. In the last part of the chapter we take a step back and suggest a way of measurement that follows from the phenomenal approach to thinking about party politics, which is more in line with how we should conduct an analysis of political space according to the theory of conceptual spaces. It introduces the pairwise measurement of party difference, both in the context of the manifesto data set as well as other data sources. The chapter ends with outlining a framework of comparison for the traditional spatial approaches for the measurement of party politics and the pairwise approach that will guide the research design for the rest of the chapters of this work.
3.1 A Plethora of Sources
We can think about political space in whichever way we want to, but what matters in the end is how we actually measure it, what information these measures contain and how useful they are in explaining something that we are interested in. Empirical data about political spaces can come from various sources, all with their relative advantages and shortcomings (for an overview, see
CEUeTDCollection
Volkens 2007). In broad terms we can divide them in two – ones that have asked someone’s opinion about a party and ones that have tired to infer the position of a party from its behaviour, i.e. the behaviour of its politicians. With regard to the former, there are mass surveys, which have asked people to locate parties usually on a left-right dimension (e.g. Schmitt et al. 2015), but rarely other dimensions, and expert surveys (Benoit and Laver 2006; Bakker et al. 2015), which have asked mostly political scientists to characterise parties on various dimensions. Some have also asked party members their opinion about political positions (e.g. Weßels 2011) and there have been various voting advice applications – software that tells people which party they match to in terms of their political preferences – where experts and party members have cooperated in determining the positions of parties on certain pre-selected issues (e.g. Trechsel and Mair 2011; Krouwel, Vitiello, and Wall 2012).
Asking party members about their political preferences would of course come closest to under-standing the political profile of a party, as in the case of mass and expert surveys about parties we are looking at the impression of “external” actors about parties. However, if we are looking at voting behaviour, it would be crucial to know not so much how parties see themselves, but how voters perceive the conceptual space that is used to differentiate between parties. Regardless of how we evaluate these various sources of survey information, they all share one fundamental shortcoming – they are temporally limited, because they are only available for one time point and cannot be retrospectively constructed, as it does not make sense to ask people now about the political profile of a party in the past (Meyer 2013, p. 32). Even if the person did exist at that moment in the past, time has diluted and eroded the memory. Thus, for more extensive analyses that span across time, party behavioural data should be more feasible.
Behavioural sources of information can in turn be broadly divided into two – voting behaviour in parliament and the way that parties have articulated their political profiles in manifestos or other documents. If we look at how members of different parties either vote together or not, it is possible to infer what the locations of these parties are on an ideological dimension. This has been extensively studied through what is called roll-call voting in the United States (Poole and Rosenthal 2000).
These are votes for which it is known how each member of the legislature voted. However, this tends to work only in contexts where we have two parties and no coalition governments, in which case voting will reflect coalition-opposition status and not actual ideological position (Meyer 2013, p. 31).
Thus, party manifestos are, at least as far as analyses across countries and time are concerned, the most well suited for the analyses of political space. They do have their problems, some of which will
CEUeTDCollection
be discussed below, but for certain analyses simply no other data is or could be available.
Party manifestos are a unique source of information about party politics in several respects. They are regular statements of party policy that are attributable to the whole party (Budge and Farlie 1985, pp. 272-273; Robertson 1976, p. 72) and the only message that is in the full control of the party (ibid., p. 12). It is a more or less uniform genre for the expression of political preferences across settings, at least in comparison to other possible textual sources of data. Manifestos are not only the most direct single source of data about the overall political profile of a party, they are also the only source of data that is available for any kind of analysis of party politics that stretches out into the past and covers numerous countries and time periods. It is therefore not surprising that the latter are widely used for both human and computerised analyses to retrieve information about the political profiles of parties.
In the case of computerised analysis of manifestos as text, word frequencies are used to determine the positions of parties on a dimension. Such approaches are fast and reliable, because computers are able to process textual data fairly quickly and always do it in the same way. However, of the two computerised solutions that have been suggested in political science, one still requires human input in the form of benchmark texts that have to be defined and which determine the dimensions that are analysed (Laver, Benoit, and Garry 2003) and the other is limited in the sense that parties have to be analysed across elections (thus assuming that political space does not change over time) and can be done so on only one dimension (Slapin and Proksch 2008) (thus effectively assuming that there are no more relevant dimensions).
As far as the current state of research is concerned, human content analysis of manifesto data is still the most realistic and feasible source of data about party politics, at least for more extensive analyses. There have been several projects over the years, which have performed content analysis of party manifestos or other documents, like the Euromanifestos Project (Schmitt et al. 2016) or the Regional Manifestos Project in Spain (Alonso, G´omez, and Cabeza 2013), in order to determine where parties stand on issues or ideologies, but the one that by far stands out, that has been and is likely to be the most prominent source of data for the analysis of party politics, is the manifesto data set (Volkens et al. 2015a; for the most comprehensive overviews of this long-running project, see Budge et al. 2001; Klingemann et al. 2006; Volkens et al. 2013). The single measure for party left-right positions that has been used by far the most in empirical research – the RILE left-right index (outlined in Laver and Budge 1992) – comes from this data set, as well as numerous potential alternatives that have more often been suggested than actually used.
CEUeTDCollection
Regardless of the nature of the analysis that we aim to conduct, we are unlikely to be able to get around the manifesto data set, if for no other reason, then simply because of its scope and availability.
The empirical analyses below will also use this data set. The following sections will give a longer overview of this data, as well as the various measures of party position and difference that have been devised from it, as this is thede facto state of the art in the vast majority of party research.