Abstract
In cross-national research on party systems, the empirical units of analysis are often assumed to be self-evident, which can be conducive to misleading research results. This problem is particularly important with regard to party system classification, for which a methodologically rigorous approach to the units of analysis is needed. This article proposes a set of operational criteria for identifying elements that qualify for inclusion within the universe of democratic party systems among individual election outcomes and country-specific sequences of elections. On this basis, I introduce additional criteria for distinguishing between party systems and party non-systems, and among party systems evolving within the same nation-state settings. By applying these criteria to a set of 1502 national legislative elections held in the world’s democracies from 1792 to 2009, the article identifies 162 units that can be entered into a classification of the world’s democratic party systems and 21 party non-systems.
Introduction
In his seminal contribution to comparative research on political parties, Sartori (1976) defines party systems as recurring patterns of interaction among political parties. At the most easily available level of observation, party systems manifest themselves as election outcomes. But, as follows from Sartori’s definition, the reverse is not necessarily true: not every election outcome can be viewed as a manifestation of a party system. First, we have to establish whether this or that particular election can be viewed as potentially possessing party system properties. Not every election is party structured, and, if it is, the systemic nature of interaction cannot be taken for granted. Second, even if we are reasonably confident that a party system is in place, isolated election outcomes can be viewed, at best, as snapshots. What is lacking is cross-temporal continuity, without which we cannot observe the recurrent patterns of inter-party interaction. But if an election outcome is part of a sequence of elections, then the question is: Does every such sequence make a party system? And if some of them do while others do not, how does one draw the line? In order to answer these questions, we need operational criteria.
In particular, such operational criteria are essential for classifying party systems. In political science, the choice of the units of analysis is defined by research goals. For pursuing some of them, party system properties may be of secondary concern, while the properties of individual election outcomes are of primary importance. But many other research goals are such that, for them, party system properties are essential both conceptually, for building hypotheses, and empirically, for testing them. For instance, in order to study the interplay between electoral systems and party systems, which is one of the flourishing fields in political science, we have to employ analytical units to which we can ascribe systemic properties. Yet too often the units of analysis on the party system side of the equation are assumed to be self-evident, which leads to simplistic operationalizations such as taking the average effective number of parties for the whole period of observation. In fact, there are no self-evident units of analysis in research on party systems. The reason is exactly that they are reducible neither to individual election outcomes nor to their sequences. In order to enter party systems into analysis, we have to develop a tool for identifying their theoretically important types. In an earlier article (Golosov, 2011) I discussed this problem from a methodological perspective. In a parallel effort, this article takes a first step towards an empirically informed, comprehensive classification of the world’s democratic party systems. It does not provide a classification as such. For a classification to be built, we need both a method and a set of empirical objects to which the methodological tool can be applied. The goal of this article is to establish a set of operational criteria for identifying party systems among elections and sequences of elections that do not necessarily possess systemic properties, and to apply these criteria to the whole universe of democratic elections from the end of the eighteenth century to the present time. The end product is the list of units that can be entered into a classification of the world’s democratic party systems.
Individual election outcomes
Election outcomes are election results tallied by party. For a long time, the question of what kinds of election outcome could be entered into party system classification remained unproblematic. There seemed to be an implicitly assumed answer: all elections are eligible, if held in independent nation-states under conditions that satisfy common-sense requirements of democracy. For example, Blondel (1968) simply considered all ‘Western democracies' as natural sites where party systems of different types could be observed. A more methodologically self-conscious approach was developed by Sartori (1976), who disqualified some of the communist regimes and several other dictatorships by arguing, quite convincingly, that ‘one-party systems' are not really party systems, and introduced the concept of a hegemonic party system, in which multiple parties exist but no real competition is permitted. His primary focus, however, was on democratic party systems worldwide, mostly but not exclusively in advanced industrial countries. Similar perspectives were later taken by Ware (1996) and Siaroff (2000). I follow this line by excluding elections held in non-democratic conditions. By this, I do not mean that non-democratic party systems are infeasible. Quite the reverse, the existence of more than one party is an important property of the most widespread contemporary variety of non-democratic rule, electoral authoritarianism (Schedler, 2006). Such parties do interact in patterned ways and therefore they can form party systems, not even necessarily hegemonic ones. Today, there are many instances of fragmented legislatures totally controlled by an authoritarian executive that does not affiliate itself with any of the parliamentary parties or even state party preferences, if only to sustain a legitimizing claim of ‘standing above all fractions', as exemplified by Uzbekistan. What is common among the authoritarian party systems is that elections are not the pivotal arenas of inter-party interaction, which results in fundamental differences between them and democratic party systems. As a consequence, the properties of the two kinds of party system are incomparable. I am dealing only with the democratic ones. However, the universal spread of electoral authoritarianism coupled with the arrival of many new and uncertain democracies makes it imperative to use specialized tools for distinguishing between the two kinds of political regime.
In this study, the major tool for identifying democratic conditions is one of the most widely used cross-national measures of democracy, Polity IV, as updated in 2010 for the previous year. 1 There are two reasons for this choice: first, the extensive chronological scope of Polity IV, and second, its relative independence from election outcomes (Bogaards, 2007). The second aspect is important as a means of ensuring that highly concentrated party systems are not automatically rendered non-democratic, as could have happened with the measure of Vanhanen (2003). I judge the conditions to be democratic if, for the election year in the given country, the DEMOC score of the Polity IV database is no smaller than 6. The transitional score of −88, or sequences of such scores, also qualifies for inclusion if followed by the score of 6 or larger, but not otherwise. In some cases when legislatures were elected briefly before democratization and continued well into the democratic periods, such elections are included. The major exception from the rules is Columbia in 1958–73 when, while qualifying as a democracy, it did not produce meaningful election results in terms of the party structure of its assembly. I follow Polity IV by including countries with severe suffrage restrictions, such as South Africa and South Rhodesia during their apartheid periods and some of the early West European democracies. 2
Polity IV rates only independent nations. In a recent analysis, Bardi and Mair (2008) argue that party systems can also exist at sub-national levels, and, of course, this argument can be extended to colonies and other dependencies. Without rejecting this approach on theoretical grounds, I would suggest that sub-national and colonial party systems can be included within the field of inquiry only upon careful clarification of their status vis-à-vis national party systems upon which they structurally depend (Gibson and Suarez-Cao, 2010). So far, I see no sufficient reason for departing from Sartori’s (1976: 83) methodological imperative of taking only national party systems into account. When identifying independent countries, I followed Polity IV in viewing the British dominions, the Scandinavian countries in personal unions with foreign monarchs, the Commonwealth of the Philippines (1935–42), and some of the states with limited international recognition, as independent states. Of the small countries, this approach had to be extended to Iceland (1918–44) and Newfoundland (1907–34), but no states with limited international recognition were included. Indeed, the glaring gap in Polity IV is the lack of information on small countries. Since the purpose of this study is inconsistent with discounting them, I had to fill the gap by employing another major tool of democracy measurement, the Freedom House ratings. 3 The small countries were included on the condition that their average Political Rights and Civil Liberties score was 3.0 or less. 4 The pre-1973 placements of small countries within the democratic or non-democratic categories were few, because most of them achieved independence after 1973, and generally unproblematic. In particular, major constitutional reforms were considered as crucial thresholds of democratization. The two countries omitted from my analysis due to the lack of the data are the United States of Colombia (1867–85) and Liberia (1847–83). Polity IV recognizes both of them as democracies, and they reportedly did have party-structured legislatures during parts of these periods (Liebenow, 1964; López-Alves, 2000: 96–139), but no precise information about the corresponding elections seems to survive.
I also employ several additional criteria. One is the level of party development. Democratic elections are not necessarily party-structured. There are several historical instances of democratic polities without political parties, such as the nineteenth-century Orange Free State (Bryce, 2008: 266). Today, a whole set of non-party democracies can be found in the Pacific region (Anckar and Anckar, 2000). Such elections obviously had to be excluded. Moreover, all elections in which non-partisan candidates won more than a third of seats were also discounted for the reason that, in such situations, inter-party interactions yield their importance to the leading parties' interactions with independents, which does not fall under the category of party system. The second additional criterion is the level of elections, with only national legislative elections being taken into account. 5 Finally, only direct elections were included. A party system, while ultimately manifesting itself in the composition of the legislature, is rooted in the electorate, and a degree of direct correspondence between voter preferences and party representation is therefore desirable. This desideratum, however, is not necessarily fulfilled by indirect elections. The application of this criterion made me exclude several series of elections that would otherwise qualify, such as the early twentieth century Costa Rica (Lehoucq and Molina Jiménez, 2002), Norway and Sweden (Caramani, 2003). Partial elections, such as the staggered ones in the US and elsewhere, were considered separately as elections on their own.
Thus, the initial dataset for this study was constructed: to include the results of all direct national legislative elections that were held in independent countries, took place in democratic conditions, and returned party-structured legislatures, within the chronological scope of 1792 to 2009. The data were derived from a variety of print and online publications. 6
Sequences of elections
While some of the patterns of inter-party interaction can be identified at the level of individual election outcomes, their recurrent nature can be revealed only by observing sequences of elections. Thus the concept of a party system necessarily involves cross-temporal continuity. From the angle of building a set of operational criteria, the first practical question is: how many elections are sufficient for including a sequence of elections within a set of units for party system classification? The answer to this question can arrive in no other form but an arbitrarily set number. A theoretical cue, however, can be taken from one of the central party theory concepts, institutionalization. Rose and Mackie (1988: 536) state that ‘a party is judged to have become institutionalized if it fights more than three national elections. A group that fails to do this is not an established political party, but an ephemeral party’. I find this approach quite appealing, but it has to be kept in mind that in most countries a sequence of three national legislative elections embraces 12 to 15 years, and it is difficult to argue that a pattern of party interactions that survives for such a long period is ephemeral. Therefore, a sequence of no less than three national legislative elections can be viewed as eligible. However, while it is true that such a sequence can take a long time, it can also be too short if the elected legislatures are subjected to premature dissolution. For example, there were three national elections in Denmark in 1920. In most instances, such series of elections are built into larger sequences and therefore unproblematic, but if isolated they defy the notion of long-term continuity in patterned interactions among political parties. Thus I complement my definition with an additional element: no less than three national legislative elections have to be held in the course of more than five years.
Of course, the eligible sequence of elections can be much longer than the above-stated minimalistic criteria prescribe. Can it be interrupted? If so, then what constitutes an interruption? The answer to this question is largely determined by my operational definition of eligible election outcomes: interruptions occur, first, when the country in question loses independence, and, second, when it ceases to be a democracy, even if elections continue to be held. Each time when this happens, after independence/democracy is restored, I register an interruption. The length of authoritarian interruptions was generally disregarded, so that they were registered for any occasion when the indicators of democracy used in this study fell below the established thresholds. In two countries, Portugal (1918) and Bangladesh (1996), I noticed brief authoritarian interludes not registered in Polity IV, and thus not qualifying as full-fledged interruptions, yet in both cases the lengths of interludes were sufficient for holding non-democratic elections. These elections were simply excluded from the corresponding sequences. The Tinoco dictatorship in Costa Rica (1917–19), while also unregistered in Polity IV, was however judged to be long enough to constitute an authoritarian interruption.
A minor issue that complicates this straightforward approach is whether an interruption occurs if the country itself becomes redefined territorially, as happened with Germany in 1990, Malaya/Malaysia in 1963, Somalia in 1960 and the United Kingdom in 1922. In all these cases, I judged that continuities in patterns of party interactions outweighed the importance of territorial transformations. Thus the German elections of 1987 and 1990 are parts of the same 1949–2009 sequence. In contrast, I considered as interruptions the events of radical regime change, even if they were not accompanied by breakdowns of democracy. This minor criterion was applied only to the establishment of the Fifth Republic in France and the Bolivarian Republic in Venezuela, and to the demise of the apartheid regimes in South Rhodesia (Zimbabwe) and South Africa. Once eligible sequences of elections are identified, it becomes clear that many individual election outcomes are irrelevant for the purposes of this study simply because they are isolated. For instance, Senegal held free elections in 2001 and 2007, but since the number of elections in the sequence does not reach three, the available two had to be discounted. By systematically applying the established criteria, I identified 1502 election outcomes suitable for further analysis, and aggregated them into 145 party sequences. Consistently with the above-stated approach, for some of the countries I registered more than one sequence, as a result of which the overall number of countries in the analysis is smaller, i.e. 111.
Party systems and non-systems
There is no apparent reason to believe that every sequence of elections, even if uninterrupted by a loss of independence or an authoritarian breakdown, makes a party system. Sartori (1976) wrote about ‘atomized pluralism’ in which parties do exist and compete but do not interact in a patterned way. Following this line of reasoning, Sanchez (2009: 487) introduces the concept of party non-systems, defined as ‘party universes characterized by a fundamental absence of inter-temporal continuity in the identity of the main parties'. In operational terms used in this study, a party universe – and therefore a party non-system – is a sequence of elections. Indeed, Sanchez explicitly refers to continuity in time as to an essential definitional element. In the contemporary world, the salience of party non-systems greatly increased with the arrival of many new democracies. Some of these democracies' nascent party arenas are remarkably unstable, which is true both for highly fragmented settings of post-communist East Central Europe (Lewis, 2000) and for highly concentrated settings of sub-Saharan Africa (Mozaffar and Scarritt, 2005). The definition of party non-systems offered by Sanchez makes extensive use of the notion of extra-system volatility, defined by Mainwaring et al. (2010) as the vote-share of new parties. 7
For my analysis, the importance of this conceptualization goes beyond the problem of distinguishing between party systems and party non-systems. The operational criterion of extra-system volatility can also be used for identifying separate party systems within sequences of elections. This aspect seems to be of fundamental importance for party system research. In fact, the very understanding that nation-specific sequences of election outcomes can and should be divided into subsets corresponding to individual party systems is not novel in political science. It is conventional to speak of several chronologically defined party systems in the United States. For instance, Beck (1997) finds as many as six of them. Similarly, Carty et al. (2000) discuss three-party systems of Canada. In the cited examples, party systems are distinguished by nation-specific, ad hoc criteria. A more comparatively oriented approach is proposed by Siaroff (2000), who develops several criteria for identifying party system types, applies these criteria to individual sequences of elections and registers continuity only in those cases when every election in the given sequence fits into a specific type. Not surprisingly, from this perspective party system change occurs frequently, which leads Siaroff (2000: 71) into claiming that Belgium has gone through six different party systems since World War II. The criteria used by Siaroff are mostly quantitative, involving the number of important parties and their relative balance.
Building on his earlier work on party system change, 8 Mair (2006: 67) argues against using continuous numeric variables for distinguishing among party systems. He proposes to focus instead on the changing patterns of government alternation, the emergence of new governing alternatives, and their access to office, adding, however, that ‘any application of these guidelines requires sensitivity to the peculiarities of the case’. Of course, the requirement of ‘sensitivity to the peculiarities of the case’ defies the very idea of universalistic operational criteria for party system classification. For this reason alone, Siaroff’s approach is more appealing. While to some extent Mair’s rejection of enumerating the number of party systems in well-established democracies holds the ground empirically, the major problem of Siaroff is different from what Mair’s theoretical reasoning implies. The number of important parties, their relative balance, and party system volatility are fundamental party system properties (Pennings and Lane, 1998: 4 f.) exactly because they define the patterns of government alternation, register the emergence of new government alternatives, and reflect the availability of office access to new claimants. While it is definitely true that the emergence of important new parties might not immediately translate into a new pattern of alternation, simply because the dynamics of party system development in well-established democracies is constrained by societally and institutionally induced inertia, the potential for a long-term change is already there, and this has to be recognized if we want to go beyond the surface of events. In my view, the major problem of Siaroff is that he underestimates the difference between election outcomes and party systems. Indeed, while the above-mentioned indicators are presumed to address party system properties directly, in actuality they are normally used for describing individual election outcomes. Without making such a distinction at a methodological level, Siaroff registers a specific quality of a party system only if all its components possess the same quality, which is wrong from a system theory perspective: a system is different from the sum of its parts. Yet I have to side with Siaroff in his recognition that, for distinguishing among party systems, we need universalistic operational criteria. While it is certainly true that party system change is a process involving several dimensions, some of them going well beyond the domains of electoral and coalition politics (Mair, 1989), these two domains are essential (Laver, 1989). Besides, party system fragmentation and volatility have a clear advantage of being readily measurable.
Before defining these criteria in operational terms, an important methodological issue has to be resolved. All essential measures in party system research are equally applicable at two levels, electoral and legislative. The effective number of parties, already when first formulated by Laakso and Taagepera (1979), was applied at both. The conventional measure of party system volatility (Pedersen, 1979) was normally operationalized and understood as voter volatility, but the structural properties of the Pedersen index do not exclude the possibility of its application to legislative seat distributions. Of course, the same is true about the less sophisticated measure of extra-system volatility. It can be defined either as a vote-share of new parties or as the seat-share of parties previously unrepresented in the assembly. The question is: What level is to be chosen? The answer to this question can be derived from the main body of research on party system classification. While the original focus was on vote distributions, already in the seminal work of Rokkan (1970) it shifted to seat distributions. The theoretical reason for this shift was that Rokkan explicitly related party system properties to the patterns of interaction among parties in their competition for office, and defined them in terms of coalition-building. Since then, and especially in the light of Sartori’s (1976) contribution, this emphasis on coalition politics became stronger, and even though vote-based measures remain widely in use it is now conventional to operationally define party systems on the basis of seat-shares. An additional advantage of such an approach is that the availability of the data on seat distributions is much greater. In practice, a more or less comprehensive vote-based electoral dataset can be constructed only for the post-World War II period. With vote-shares, therefore, my undertaking to go well beyond this threshold would have been entirely hopeless.
At the current stage of my analysis, the task is similar to what has been done at the previous stage: to provide an operational definition for significant interruptions of continuity between individual election outcomes. The difference is that this time the lines have to be drawn within uninterrupted sequences of elections, which requires additional criteria. I follow Sanchez (2009) in using extra-system volatility as one of such criteria. But I use it not only, nor even primarily, for singling out party non-systems, but prima facie for defining party systems. At the operational level, I view every election outcome in which previously unrepresented parties jointly gain no less than 25 percent of seats as starting a new party system. The threshold is obviously arbitrary, and it cannot be otherwise, but some underlying reasoning needs to be revealed. When establishing the exact value of the threshold, I wanted to avoid a situation in which all instances of party system transformation would be confined to the rare events of party system collapse (Dietz and Myers, 2007), which is generally understood as the farthest-reaching variety of the phenomenon but not its only variety. Consider a sequence of two elections, in one of which assembly seats are equally divided between two parties, while in the other a new party takes a good share of seats from one of them. If this share is 50 percent, then one of the previously duopolistic parties disappears, which makes for a party system collapse. If this share is 20 percent, then the second party’s share is still larger than the share of the new one, which indicates the lack of fundamental party system change. Yet if it is 25 percent, then the new party achieves parity with the second one, which is sufficient for concluding that the within-system balance has changed profoundly. Upon running such simple exercises on several hypothetical election outcomes corresponding to major party system types (in the cited example, it was a two-party system), I arrived at a conclusion that the 25 percent threshold was optimal.
When dealing with extra-system volatility, I treated as new parties only those organizations for which organizational continuity with the parties or coalitions previously represented in legislatures could not be traced. If it could, changes of party labels were disregarded. Even though tracing some of the continuities was a challenging task, in most cases it was not prohibitively difficult. 9 I also confronted a minor problem of taking into account the ‘others' category (when collecting the data, I made every possible effort to minimize it, yet its complete elimination proved to be beyond my means) and, especially, non-party legislators. The appropriate solution was found in systematically counting them as new parties. While with regard to independents this goes against the self-evident argument that the rate of survival enjoyed by them in a party-structured assembly can be high, note that this study is focused on party system properties. From this perspective, the continuous survival of individual non-party legislators is irrelevant or even, arguably, instability-inducing. At the same time, I have to recognize that this approach slightly overstates the rates of extra-system volatility in countries using plurality electoral systems because of these systems' emphasis of candidates over parties (Brancati, 2008).
The extra-system volatility criterion allowed me to identify discontinuities within sequences of elections. In most cases, such discontinuities resulted in the emergence of new stable patterns, so that no new instances of extra-system volatility followed in the course of three or more elections. Such sequences were considered as new party systems. In some other cases, however, extra-system volatility of more than 25 percent resurfaced in the course of three or more consequential elections. Such sequences were considered as party non-systems. There were also instances in which extra-system volatility singled out one or two elections, normally at the beginning of the given sequence or at its very end. Such elections, not numerous enough to form either a party system or a party non-system, were ascribed to a separate category dubbed fringe elections. Their location at the beginning or at the end sides of election sequences bears heavily on their systemic role. If placed at the outset, they are likely to be associated with founding elections that often produce unstable patterns of inter-party competition (Bogdanor, 1990). But, if placed at the end, they are likely to be the first elections either in a new party system or in a new party non-system, even though the possibility of retaining the fringe status cannot be theoretically excluded. Of the 1502 elections that emerged as eligible cases from my previous analysis, 1364 fell into the category of party systems, 92 into the category of party non-systems and 46 into the category of fringe elections.
Cumulative party system change
This, however, did not finalize my effort. Party system change can occur rapidly, as in the case of a party system collapse. But slow, gradual change is also very common (Pennings and Lane, 1998: 5). In the long run, evolution can entirely transform a party system. This happens in a cumulative way: novel party system properties gradually emerge, gain strength and slowly replace old properties. This is why I dub this modality of party system transformation ‘cumulative change’. Of course, it cannot be captured by the indicators of extra-system volatility. For this end, a more subtle tool is needed. In search for such a tool, it was only natural to revert to the notion of fundamental party system properties, of which the number of parties has for long been viewed as the first on the list. An additional advantage of using this particular property is that its operationalization, in the form of the effective number of parties, is readily available. Of the mathematical formulae offered in the literature for calculating the effective number of parties, I select one featuring several advantages over the earlier proposed alternatives, especially as a measure of party system fragmentation in highly fragmented and highly concentrated party settings (Golosov, 2010). The formula is defined as follows:
Here, si and s1 are the sizes of the i-th and the largest components expressed as their decimal (absolute) shares of the total; sigma stands for summation from the largest (1) to the smallest (x) component.
The initial step at this final stage of my undertaking was to calculate the effective numbers of parliamentary parties for all elections. The immediate result did not bring much difference to the results obtained with extra-system volatility: there were leaps in the effective number of parties, many of them coincidental with the arrival of new parties. In order to capture cumulative change as such, additional steps were needed. While omitting party non-systems and fringe elections, I concentrated on those party systems that comprised no fewer than 12 elections (on the assumption that cumulative change is a long-term process) and broke them into sequences of six elections each. The number was established as simply twice as many elections as required for registering an eligible sequence. Then, for each of these sequences, I calculated the median effective numbers of parties and compared the resulting values. If a given median value differed from the previous one within a factor of no less than 1.25 (that is, if it was at least 1.25 times larger or smaller), I registered an instance of cumulative change. Theoretically, the most difficult part was to establish the dividing points between so understood party systems, because the notions of cumulative change and clear-cut divides are scarcely consistent with each other. In practice, however, it was always possible to identify those election outcomes that most visibly departed from their immediate predecessors. For instance, the median effective number of parties in Austria in 1986–2002 was 1.4 times as large as in 1966–83. In the course of both periods, the largest leap in party system fragmentation occurred in 1994, when it rose to 3.3 against 2.6 in the previous election. Then 1994 was taken as a starting point for a new party system. Given these complexities, however, it is commendable to keep in mind that the breaking points established by the criterion of cumulative change are much softer than those set by the previously introduced criteria.
An illustration: The case of Belgium
For the convenience of readers, I summarize major steps in identifying the units for party system classification in Figure 1 . These steps can be illustrated with the case of Belgium, one of the countries that passed through several party systems in the course of their democratic histories. Indeed, democracy in Belgium is one of the longest-standing in the world: the Polity IV DEMOC score first reaches a value qualifying Belgium for inclusion, i.e. 7, already for 1853, to be retained up to 1912; it is 8 for 1913–18, 9 for 1919–29, 10 for 1930–2006 (with an interruption for the 1939–43 period when Belgium temporarily lost its independence), and 8 again for 2007–9. The earliest national legislative elections in the course of this period took place in 1854. The sequence comprised 40 elections, some of them partial, before the 1939 interruption occurred. First, I tested this very long sequence of elections for extra-system volatility. Throughout the entire period 1854 to 1892 no extra-system volatility occurred, as all parliamentary seats were divided between the Catholic and Liberal parties. The Socialists arrived in 1894, but their share of seats did not reach the 25 percent threshold. After that, and up to 1939, extra-system volatility never exceeded 5 percent.
Identifying the units for party system classification (criteria in ovals)
Therefore no interruptions caused by extra-system volatility can be registered, which led me to test the second possible reason for discontinuity: cumulative change. For the first six elections held in 1854–63, the median effective number of parliamentary parties was 1.7. For the following sequences, it was 1.8 (1864–72), 1.5 (1874–84), 1.6 (1886–96), 2.0 (1898–1908) and 2.5 (1910–25). There were only four elections in 1929–39. The only instance when the effective number of parties increased more than 1.25 times was between the 1886–96 and 1898–1908 periods. When looking at the dynamics of party system fragmentation within these periods, I found that the largest leap occurred in 1898 when the Socialists ultimately consolidated their position as a major parliamentary party. Then I divided the 1854–1939 sequence of elections into two party systems, Belgium 1 (1854–96) and Belgium 2 (1898–1939). 10
After the 1939–43 interruption, there was a sequence of 20 national legislative elections starting with 1946. Throughout this period, extra-system volatility exceeded 25 percent only once, in 1981 when the major parties of the country split along ethnic lines. This left me with two uninterrupted sequences, one of which (1981–2007) consists of only eight elections and therefore does not have to be tested for cumulative change, while the other (1946–78), with its 12 elections, can be divided into two 6-election subsets, 1946–61 and 1965–78. Indeed, the test revealed that the median effective number of parties between these periods increased from 2.3 to 3.7, much more than within the factor of 1.25. The largest leap occurred in 1965, when the unprecedented successes of the People’s Party for Freedom and Democracy and Volksunie propelled the effective number of parties to 3.3 against 2.5 in the previous election. Thus, after World War II, the country had three party systems: Belgium 3 (1946–61), Belgium 4 (1965–78) and Belgium 5 (1981–2007). The fifth system remained in place in 2009.
The end product
The end product of this study is presented in Table 1 , which contains information about all party systems and party non-systems operating in democratic conditions for the period 1792 until 2009. Overall, there are 183 of them, 162 party systems and 21 non-systems. 11 For each of these units, the table provides a minimum of descriptive information. First, they are attributed to this or that country, with numeric indices attached to the names of countries where two or more systems or non-systems are registered. Second, the number of elections in each of them is reported, varying from 3 in many to 78 in the 1854–2008 party system of the United States. Third, the table provides information about the timing of the earliest and most recent elections, as of 2009, in the sequences constituting each of them. It has to be mentioned that, thus defined, the endpoints of the extinct party systems or non-systems do not exactly correspond to the actual time of their termination, which is either the year of the loss of democracy/independence or, for the cases of extra-system volatility or cumulative change, the year of the next elections. Fringe elections are not included. For instance, the actual year of the termination of the 1949–2003 Israeli party system was 2006, when extra-system volatility of more than 25 percent occurred. Both in 2006 and in 2009 fringe elections can be registered. It is quite likely, however, that upon the extension of the empirical scope of inquiry to the future elections, the 2006–9 elections will lose their fringe status by starting a new party system, i.e. Israel 2. In the fourth column, each of the units is characterized by its current status, or reason for disappearance if any. Here, the party non-systems are divided into two categories, the existing and the extinct ones as of 2009. The word ‘existing’ alone indicates that the given party system was still in place in 2009, while for the extinct party systems the table reports reasons for extinction. They coincide with some of the operational criteria established in this study: authoritarian exit, loss of independence, regime change, extra-system volatility and cumulative change. Finally, in order to provide a minimum of substantive information about the properties of the units, for each of them the average effective number of legislative parties is reported. 12
The democratic party systems and party non-systems of the world, 1792–2009
Sources: See note 6.
Conclusion
This study did not seek to provide a party system classification, and it does not. Unlike Lijphart (1994) and many other scholars, I do not think that the effective number of parties alone provides a sufficient basis for party system classification. This is all too clear from the data presented in the table. The mean effective numbers of parties reported for the 1952–84 party system of India, obviously a predominant-party system, is 1.69, while for the 1854–2008 system of the United States, a clear-cut instance of bipartism, it is 1.77. However, now the road is open to applying the earlier developed methodological tool (Golosov, 2011) to the available set of empirical objects, which will make the second step in my research programme. Two concluding remarks are in order.
First, the operational criteria developed in this study are sometimes based on arbitrarily set thresholds, which may require refinement. Yet it is clear that, however refined, such thresholds will remain, on the one hand, arbitrary, and, on the other, necessary. This is inherent in classification procedures. What is important is to avoid using situational criteria, specifically developed to accommodate certain cases or sets of cases. From this perspective, the universalistic approach taken in this study is a remedy even if the ultimate solution is not possible. Second, as any classification of dynamic objects, party system classification cannot be done once and forever. The extinct party systems and non-systems will not change, but the existing ones, even in continuous democracies, can be subject either to rapid transformation, as a result of extra-system volatility, or to cumulative change. Therefore even with a perfect tool for party system classification, the empirical procedures elaborated in the study will have to be repeated as long as the phenomenon of party system remains in the present, not just in history. The advantage of the proposed set of operational criteria is that for any research task that has to be pursued with the use of party system classification, the units of analysis can be easily identified in a uniform way. This promises the comparability of results, which is one of the most important desiderata in cross-national political research.
Footnotes
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.