Elections in Australia, Ireland, and Malta under the Single Transferable Vote: Reflections on an Embedded Institution

Citation preview

Page vii →

Acknowledgments

The editors would like to acknowledge the generous financial and intellectual support of the Center for the Study of Democracy, University of California–Irvine, for this project and also for allied projects on the single nontransferable vote, mixed-member proportional, and list proportional representation. Additional financial support was provided by the University of California–Riverside. Professor Grofman would also like to note the support of NSF Grant #SBR 97–30578 (Program in Methodology, Measurement and Statistics). An earlier and shorter version of the conclusion appeared as Bernard Grofman and Shaun Bowler, “STV in the Family of Electoral Systems,” Representation 34, no. 1 (winter 1997): 43–47. The editors would also like to thank the reviewers of this volume for their careful reading and many constructive suggestions.

Page 1 →

Introduction: STV as an Embedded Institution Shaun Bowler and Bernard Grofman The single transferable vote system (STV) is an important electoral system for both practical and theoretical reasons. In allowing voters to identify a rank ordering of their preferences and not just to mark an X, STV permits voters greater choice and makes possible ballot splitting to express highly differentiated preferences. In particular, it allows for the possibility of party-based voting without limiting the voter’s choices to the candidates of a single party in the way that pure list proportional representation (PR) systems do. Also, in societies that are divided along ethnic or religious lines, STV permits voters to signal strong preferences for candidates of one group while still showing some support for candidates of other groups. For these and other reasons, it has long been advocated by many, beginning with John Stuart Mill, as a tool for electoral reform. But although we regard STV as a very important system in its own right, in this volume we are not merely interested in the study of STV per se; rather, we use STV, and electoral rules more generally, as a lens through which to understand the effects of institutions as being mediated by the political and social context in which they are embedded (Farrell 1997). What makes STV a particularly useful institution to study for the purposes of this volume is that the three countries we have chosen to look at (Ireland, Australia, and Malta) provide for a kind of “natural experiment” on the impact of STV in terms of a “most similar systems” design (Grofman 1999). First, essentially the same electoral system (STV) has been employed in all three societies. Second, all three societies have had sustained historical associations with Great Britain and adapted a number of their political ideas from Page 2 →that source. It might thus appear reasonable to expect that the same electoral arrangement will have broadly similar consequences in these three settings. Yet despite the similarities among these countries, ways in which they differ prove critical to understanding the impact of STV as an embedded institution. The collection of papers contained here shows that the same electoral system can, in fact, have quite different effects under different conditions: STV in Australia is not the same as STV in Ireland, which, in turn, is not the same as STV in Malta—or, indeed, in Canada. At first glance, this statement may seem relatively trivial. Yet it is trivial neither in its implications nor in the nature of the evidence required for its substantiation. In thinking about political institutions in general and electoral systems in particular, it is often argued that certain basic effects hold across a wide variety of settings (e.g., Duverger’s law and hypothesis with respect to the consequences of electoral system types for minimal /maximal levels of party proliferation). We shall argue the case that even what are thought of as basic electoral systems effects can, to a significant extent, be conditional or contingent on other factors. But laying out that argument is far from all that this volume seeks to accomplish. First, we wish to substantiate in very specific detail the ways in which a given electoral institution, STV, operates differently in different settings. This collection of papers embodies a research design that allows us to demonstrate that the effects of electoral systems vary in meaningful ways. In choosing the disparate settings of Australia, Ireland, and Malta (and also looking at local elections under STV in Canada), we are able to compare what would, on its face, appear to be the same electoral institution across several different national settings. The resulting collection of findings tells us things that neither single-country studies nor solely analytical treatments of electoral systems can tell us and allows us to understand institutions, especially electoral institutions, as embedded within a context of actors and organizations that can shape, and even undermine, our theoretically grounded expectations about the role of institutions. Because we do focus on particular countries, we can take advantage of the knowledge to be gained from detailed case studies by observers with immense country-specific knowledge. This allows us to begin to accomplish our second task: identifying the particular institutional arrangements and/or other factors that interact with STV to modify or curtail its expected effects on party systems and on governance. Here, we shall be paying particular attention to party nomination procedures, on the one hand, and what would appear to be obscure and

irrelevant technical details of implementing Page 3 →STV that differ across the three settings, on the other. Of particular importance to the latter group are those laws that regulate the ways in which voters can complete the ballot in expressing preference orders. The contribution of this volume, then, is threefold. We contribute to the understanding of one of the major types of electoral mechanisms, preference voting, as it operates in real-world settings. Second, the papers contained here, when viewed in toto, support the argument that electoral system effects are conditional and contingent rather than being categorical and mechanical; they help begin the task of truly understanding the nature of institutional embeddedness in specific detail. Third, the evidence marshaled to support the claim for institutional embeddedness, developed from a most-similar-systems research design that bridges the more traditional approach of cross-national statistical work and case studies, can, we believe, be a useful model for other research in comparative politics. In the context of this aspect of our work, we would emphasize that this volume is but one in a series that uses a most-similar-systems design to study electoral institutions as part of a general project that was first organized by Bernard Grofman, with the continuing support of the University of California, Irvine, Center for the Study of Democracy (and with other support from the UCI Focused Research Program on Public Choice, the UC Center for German and European Studies, and the UC Center for Pacific Rim Studies). The first component of this project involved the study of the single nontransferable vote system (SNTV) in Japan, Korea, and Taiwan (Grofman et al. 1999); the second, on STV, resulted in this volume; the third involved the study of list PR in the Nordic countries and will result in a conference volume to be coedited by Arend Lijphart and Bernard Grofman; the fourth involved the study of electoral systems that mix single-member district and PR components (such as those in Germany and, more recently, Japan, Italy, New Zealand, and Russia) in a conference jointly organized by Martin Wattenberg, Matthew Shugart, and Stephen Levine that also is planned to result in a conference volume. Taken together, these studies provide a series of in-depth treatments of electoral system effects. Although effects between systems are clearly important, this approach allows a look at effects within systems as well. Because variation in effects within systems cannot be explained by reference to the given system itself, this forces us to look elsewhere for other factors to explain that variation. In doing this, we can begin to outline the limits of the effects of electoral systems themselves in part by showing how institutions are embedded within specific organizational contexts that blunt or change institutional effects. Page 4 →We turn now to a discussion of the wider literature on electoral systems to situate this collection of papers in the context of the present renaissance in electoral systems studies.

Electoral Systems and Electoral System Research Recent years have seen a great upsurge of interest in the topic of electoral reform and the study of electoral systems worldwide among both academics and the public (see, e.g., Amy 1993; Farrell 1997; Independent Commission on the Voting System 1998; Royal Commission on the Electoral System 1986; Lijphart 1986). Electoral systems are not just a topic of the day in the emerging democracies in Eastern Europe and the former Soviet Union but also have been prominent on the political agendas of established democracies. Most notably, Japan, New Zealand, and Italy have recently made dramatic changes in their electoral rules. Even within the seemingly locked-in-cement first-past-the-post electoral institutions of Britain and the United States, an increasing number of experiments in new electoral systems are taking place at a local level (in the United States)1 or at a regional or supranational level (in Britain);at the national level, lobby groups and reformers in both countries are pushing forward the agenda of electoral reform. In Britain, even aside from general discussions of electoral reform for nationwide elections, a series of debates and controversies have taken place over the shift away from first-pastthe-post elections for the European Parliament and for the new Assemblies of Scotland and Wales. The Jenkins Commission report thus promises a series of reforms for yet a third set of British elections. These pressing realworld political agendas have given fresh impetus to the academic study of electoral systems. But all of this interest begs a prior, and more prying, question: What do we expect to learn from studying electoral systems? Elections are, of course, central to conceptions of democracy in the modern era, so they are consequently important in a definitional sense. Some might even argue that elections are the key hallmark of democracy. But this does not automatically explain why political science has such a relatively well developed series of literatures

devoted to the topic of elections or why there has been such a recent upsurge of interest on the topic of electoral rules. Why, then, should we study electoral systems? The answer to this question is found in both of the major literatures on electoral systems—one largely aggregate and statistical (Lakeman 1974; Lijphart 1994; Grofman and Lijphart 1986; Rae 1967) and the other largely formal and focused at the micro level. Both share the same broad conclusion: Page 5 →Different electoral systems produce different outcomes (Riker 1982; Cox 1997). Both also deal with central normative issues in democratic theory. The aggregate and statistical literature focuses on practical questions associated with real-world elections and polities, in particular, on the effects of electoral systems on party systems. A common theme has been the examination of outcomes produced by different systems in terms of the proportionality of the seats-to-votes relationship. In normative terms this reflects a concern for the fairness of the result. Similar issues arise when we look at which system best serves the descriptive representation of women and minorities (Amy 1993; Pitkin 1967). The formal literature on the topic of electoral rules has concentrated to a great deal on the extent to which outcome can change markedly as the rules for counting votes change, even when voters’ preference orders remain fixed. From here the formal literature has gone on to address the ease with which varieties of systems may be manipulated and used by voters acting strategically (Riker 1982; Cox 1997). Thus, the short answer to the question, Why study electoral systems? is that the choice of electoral system is centrally important in answering questions about what democracy is and should be. Normative concerns are typically very close to the surface in most studies of electoral systems. The importance of the issues at stake can be seen in the debates over which electoral system to choose in the new democracies of Eastern Europe and in the prolonged discussions of electoral reform in societies such as Britain, New Zealand, and Japan. But even if electoral systems, as generally understood, are important topics to study, this only serves to prompt the question, “Why study STV?” After all, STV systems are not used by the richer and more powerful states that command so much of scholars’ attention. Here we advance two justifications. The first concerns the position of ordinal systems generally within the field of electoral system research. The second, and more important, justification considers how we understand institutions. By the very nature of the questions being addressed, the literature on electoral systems is necessarily comparative. The effects of one system are seen by comparing it with another; either in a formal or statistical sense, the typical approach is to look across systems (e.g., Lijphart 1994; Rae 1967). And to the extent that commonly shared benchmarks exist across such broad-ranging literatures, they are concerned with the performance and outcomes of the familiar Anglo-American (single-member simple-plurality) system. The other most commonly used benchmark is that of list PR. Within this setting, the class of ordinal electoral systems (alternative vote [AV], cumulative vote Page 6 →[CV], limited vote [LV], SNTV, and STV) comprise a generally neglected middle ground of study. But this does not mean that the argument for studying STV is based on some novelty value. In fact, novelty for novelty’s sake can be one of the siren songs of the study of electoral systems, luring students into intriguing exceptions and idiosyncrasies in the way elections are conducted at the expense of asking wider questions. Rather, the study of systems such as STV is important for our understanding of what electoral systems can and cannot do at a theoretical level. STV is a member of the relatively less well understood family of ordinal electoral systems. These systems present both voters and parties with a wide range of strategic options and possibilities—a range that is far wider than that presented by either the single-member simple-plurality system or list PR, the two most commonly studied systems. Indeed, it seems to be the study of the contrast between these two systems and the working out, both empirically and theoretically, of the full consequences of Duverger’s law that has preoccupied much of the recent interest in the study of electoral systems. Because of this, it should come as little surprise to find that substantial literatures exist on the ability of voters to cast tactical or sophisticated ballots under single-member districts as well as on issues of proportionality and the related issue of just how many parties each system can support.

Ordinal systems, however, allow for a much richer range of behavior than that produced by categorical systems because they allow voters much greater freedom in how to complete a ballot. Under ordinal systems, voters might be asked to list a set of preferences either in rank-order form (as under STV and AV) or simply make multiple marks for candidates they like (as under CV) or be given several votes (as under SNTV). The normative case in favor of allowing voters such freedom is set forward most eloquently by Enid Lakeman (1974). STV allows voters to choose both between and within parties and so reflects a diversity of opinions within society. Elections are for the benefit of the electors, not for the political parties or any other interests, and the electors must see to it that they get a system which seems to them adequate for the expression of their views. (Lakeman 1974, 273) For Lakeman, as for other scholars (e.g., Farrell 1997), STV offers a large degree of freedom for voters to express their views. Over and above any normative value, such freedom may also have enormous consequences for parties, not just in terms of how many parties may be produced by a variety of Page 7 →systems but also in terms of how they campaign and electioneer and how they may (or may not) fight internally. Not only, then, are answers to such standard electoral studies questions as “How many parties will be produced under system X?” a lot more uncertain under ordinal systems, but also a whole different range of questions is raised by these systems. Systems such as STV produce a much richer range of more obviously political behavior than do other electoral systems more familiar to students in this field (Bowler 1996). As such, they provide an important alternative insight into the impact of electoral systems. The stark contrast between single-member districts and list PR can help to support a very mechanical view of how electoral systems work in practice. Ordinal systems can, as the collection of papers in this volume shows, support an understanding of electoral systems that emphasizes the contingent and conditional, as opposed to categorical, nature of the relationship between electoral systems and the political behavior of both voters and parties. This range of outcomes can be seen quite clearly with respect to three of the main examples considered within this volume: The actual practice of STV differs markedly across the cases of Australia, Ireland, and Malta, exhibiting very disciplined and highly centralized parties at the one end of the continuum while exhibiting a much more fluid and internally riven political process at the other. This brings us to the second broad reason for pursuing a study of STV as a system in this way: Doing so helps us to understand the embeddedness of institutions. By contrast to the more standard approaches to the study of electoral systems, which compare across different systems, this volume adopts the somewhat different approach of making comparisons within systems. The point of this exercise is to arrive at an understanding of electoral institutions, and perhaps institutions more generally, as being embedded within a particular context. Recent years have seen a revival of interest in the study of political institutions in general (see, e.g., North 1990; March and Olsen 1989). One way of capturing this recent history of political science might be to call it “bringing institutions back in,” after the emphasis on avowedly noninstitutional explanations—such as behavioralist or Marxist ones—during the 1960s, 1970s, and on into the 1980s. For students of electoral systems, however, institutions never left! Within this one area of political science, it has always been clear that institutions have the property of shaping the incentives, the expectations, and hence the behaviors of politicians. For students of electoral systems, then, institutional effects and the idea that one should pay close attention to the formal rules of the game have always been a concern. Page 8 →Nonetheless, both in terms of the broader literature on institutions and also within the electoral systems literature, emphasis has been placed on a fairly broad-grained understanding of what institutions are and should be. Institutions are often seen, within these literatures, as irresistible forces shaping political outcomes whatever the context. But as in this volume, a comparison of broadly similar institutions across different settings can be seen to produce quite different substantive political processes. To the extent that the institutional setting is, in effect, held constant across these different cases, we must then turn to examine alternative explanations as a means of understanding the different electoral outcomes produced in them. Either we must look to a finer-grained institutional analysis—one that takes account of seemingly small differences in rules that may turn out to be quite

consequential—or we can look to see how electoral systems interact with other institutions. Alternatively, we can anchor our understanding in the historical and cultural processes in which the electoral institutions are located (see, e.g., Steinmo, Thelen, and Longstreth 1992). In either event—either by appeal to a more refined analysis of institutional features or through a development of arguments that expose the limits of focusing only on formal institutions—we can begin to advance a contextually informed understanding of institutions. The extent to which we can understand the contexts in which institutions exist determines how fully we can understand the institutions themselves; they do not, after all, exist within a vacuum. This study, therefore, bridges the aggregate, macrostatistical studies that build on cross-system comparisons and those studies that rely on very specific historical or single-case examples and often on highly formal treatments. In both of the research traditions within the study of electoral systems, electoral institutions are accorded a very prominent, indeed, central role. Implicitly this gives a central causal role to electoral institutions, which can then be held responsible for setting the constraints with which the actions of a given set of actors are determined. Within the literature that employs a cross-national statistical approach, electoral systems are often seen as overwhelmingly important institutions in and of themselves, shaping the behaviors of all actors within a given country. It is this conception, rooted pretty firmly in an appreciation of Duverger’s law, that has given electoral engineers hopes of being able to fix many problems by putting in place an appropriate electoral system in the newly emerging democracies. A number of problems, however, face the would-be engineer. No matter how strongly a given electoral system may shape the politics of a particular Page 9 →country, other factors also have an impact, and so electoral systems are unlikely to be a panacea. Further, each electoral system implies a series of trade-offs between different objectives that an electoral system might accomplish. And these trade-offs are often difficult to figure out, making it hard for one electoral institution to embody optimal trade-offs on all dimensions, even if one could agree on how to trade off competing desiderata. Under STV, for example, the price of giving voters more of a say over individual candidates may be the weakening of parties through intraparty factionalism. On the other hand, the steps that the parties take to counter these pressures may rob voters of the ability to actually choose in meaningful ways between candidates. And this seemingly small point shows up one of the main virtues of approaching the study of electoral systems in the way we do here. In principle, STV should encourage both intraparty competition and fractionalization of the party system because of two features of the system. First, STV sets up multimember districts in which voters rank-order candidates and, second, the threshold of representation is relatively low. The rank ordering of multiple candidates from the same party sets up clear incentives for candidates from the same party to try to do each other down in the interests of winning. Candidates can expect to receive votes from fellow party members who are eliminated and this, coupled with the lower chance of losing everything altogether, means that the internecine struggle is not likely to be as fierce as under systems such as SNTV. But the struggle for votes between candidates of the same party can easily be seen to promote factionalism and discontent.2 The threshold of representation (TR) defines the minimum support necessary to earn a party a seat, based on the mostfavorable-case scenario in terms of how the other parties divided up their votes. Its companion measure is the threshold of exclusion (TE), which adopts a worst-case scenario and defines the maximum support a party can gain without winning a single seat. The threshold of representation provides a necessary condition for parliamentary representation; the threshold of exclusion provides a sufficient condition for it. At some risk of generalizing too broadly, the point here is that these measures are one means of indicating a number of features of a political system, including the relative ease or difficulty with which new parties may enter the system. Compared with other systems, new parties have a relatively easier time forming and prospering under STV than under several other systems. Moreover, intraparty factional wars can readily provide the motivation to form such new parties. Taking these two arguments concerning intraparty fights and the ease of entry into the system together, we can

arrive at quite straightforward and Page 10 →readily understood predictions to the effect that STV should promote a highly fractionalized party system. Yet STV does not do this as much as the abstract studies suggest it might, which is just one of the puzzles that are raised, and addressed, by this volume. At least part of the answer to the puzzle of why we see less fractionalization under STV than we expect to see is that countervailing institutions—in particular, political parties—work to combat the incentives of the electoral framework. The picture that emerges from the studies in this book is that parties are much stronger and more resilient institutions that the electoral-engineering approach would have us expect. Because of this, we arrive at a more nuanced view of institutional effects. Current political science thinking holds that institutions matter. A priori, however, it is not always clear which institution should matter to what, and this is especially clearly shown by this comparison of STV within systems. One of the major virtues of this approach is that it allows a discussion of how institutions are embedded with particular contexts, and one of the actors to which we point as being especially important in this embedding is the political party. We can show, on the basis of this collection of papers, that different parties have responded in different ways to the incentives of the electoral system. This means that we can begin to explain cross-national patterns of variation and the sizable gap between expectations derived from analytical models and the actual practice of elections in all these nations. Specifically, although analytical models of STV predict highly factionalized parties and party systems, none of the examples show this pattern. In fact, disciplined parties comprise a central part of these nations’ political systems. Such parties are able to exist in part because they consciously adopt strategies that blunt the tendency to factionalize. This argument has quite broadranging relevance. Anchoring our argument more explicitly in the new institutional literature will allow us to draw out this relevance as well as provide a clearer distinction between organizations such as parties and the formal institutional framework within which they work. The broader-ranging difference between our conception of institutions and that of much previous work is that it allows actors to react against institutional effects. Much of the work on electoral systems accords with the broader literature on institutions and paints a picture of political actors who are largely passive in the sense that they accept the terms and conditions of those institutions even if the institutions produce outcomes that are somehow unpalatable. Actors (parties and politicians), in these accounts, seem to act almost as do consumers and firms in models of perfect markets and are price takers unable to influence or affect the setting in which they find themselves. By contrast, real-world Page 11 →actors are likely to try to work around and even manipulate the institutions under which they work. This provides an important dynamic or evolutionary component to our understanding of institutions. Although institutions do shape the incentives facing actors and, in doing so, can frustrate those actors’ ambitions, the actors can respond to and try to evade those incentives. One way in which parties may do this is by developing rules internal to the party over, for example, nomination procedures that dampen any tendency to factionalization. Alternatively, actors may change the fine print of the body of rules surrounding electoral systems that govern the conduct of elections or electioneering. For example, parties may change rules on ballot structure, campaign expenditure, or the timing of elections to give themselves advantage.

Plan of the Book This book begins with three chapters that provide an overview of preferential systems and how they operate. With this general background established, the following chapters examine the workings of STV with regard to countryspecific experiences. We first provide three chapters on Ireland, which remains the best known and the most studied of the nations that have used STV; indeed, the Irish experience has often been taken to be fully paradigmatic for its operation. Yet as the subsequent chapters indicate, there may be no single, paradigmatic example that represents this system in practice. Moving from Ireland, then, the chapters consider experiences in Australia, Malta, and Canada. Finally, a brief conclusion summarizes the main points of the book. However, rather than reviewing the chapters in the order in which they appear in the volume, we prefer to discuss them in broader theoretical terms, with a focus on three specific features of electoral systems: (1) consequences for the strength of parties (2), incentives for strategic behavior on the part of both voters and parties, and (3) the importance of seemingly narrow technical differences.

Strength of the Parties/Ability of the Parties to Control the Nomination Process

Colin Hughes’s chapter takes up the case of Australia where, he notes, we see a combination of substantial experience with STV in various forms at both the national and state levels alongside the presence of persistently strong parties. A particular concern of Hughes, also taken up in subsequent chapters, is that of the transfer of preferences and party discipline. These topics are addressed, Page 12 →for the case of Ireland, by Michael Gallagher and Michael Marsh, respectively. In principle, STV encourages intraparty competition as candidates from the same party jockey for first preference. This opens up the possibility, as Michael Gallagher demonstrates, for candidates to be defeated by copartisans and for parties to factionalize and splinter. Whether and how parties keep control of campaigns and nominations comprises a strong theme of Colin Hughes’s discussion of the various Australian experiences. It also comprises a similarly strong theme of Michael Marsh’s study of Ireland, which shows that parties continue to structure choices for voters. The possibility for party control to fray in the presence of untidy or candidate-centered vote transfers or unravel in the face of overnomination is given an additional twist in Wolfgang Hirczy de Miño and John C. Lane’s study of Malta. They find that even when Maltese parties overnominate, they still manage to retain control of the political process. Taken together, these papers provide an answer to a puzzle provided by Douglas Rae. In his 1967 book he hypothesizes that ordinal systems (such as STV) should exhibit much higher levels of factionalization than categorical systems (such as list PR or first-past-the-post) but finds, to use his own splendid words, that the “facts betray the theory” (129) and that “[his] theory is absolutely wrong” (127). In addition to displaying a commendable honesty and openness of approach, Rae offered—but did not test—a number of conjectures that might help address this puzzle. Taken together, the chapters by Gallagher, Lane and Hirczy de Miño, Hughes, and Marsh offer an answer that lies in the behavior of parties themselves. By comparison, we can see J. Paul Johnston and Miriam Koene’s fascinating look at a completely unstudied North American example of STV in Canadian local and provincial elections. These elections were often nonpartisan and, in some cases, involved voting populations of non-English-speaking recent immigrants.

Strategic Behavior by Voters and Parties Johnston and Koene’s work points up a particular concern that often surfaces in discussions of STV: the degree to which the system itself is understandable to voters. Several studies show that voters can and do understand STV in very general terms (see, for example, Lakeman 1974, 153–61; Bowler and Farrell 1991a, 1991b). Even these studies, however, sidestep some important aspects of strategic behavior under STV and, indeed, the question of the very ability of voters to engage in strategic behavior under a system as complex as STV. Two chapters, in particular, consider the strategic questions posed to voters. Analysis of strategic problems posed by first-past-the-post systems typically Page 13 →begins and ends with the decision of voters to cast a ballot for the second most preferred party rather than their first most preferred party under some conditions in elections with more than three parties. Under STV, however, a much greater and more subtle range of behaviors can be considered strategic. The chapter by Michael Laver argues that strategic behavior is, in fact, a central component in understanding how lower-order preferences are expressed. Part of his argument is that voting with an eye to coalition formation makes voters strategic in their lower-order preferences, not least because small parties under STV must try to keep the voters from expressing lower-order preferences from the large and, in particular, nearmajority parties. A different look at problems of strategizing is found in the chapter by Neal G. Jesse, which focuses on why there are fewer parties under STV than theories of electoral systems would have us expect. Previous chapters stressed discipline internal to the parties; here Jesse shows that strategic voting behavior (generally understood) can help to limit the number of parties present in the system. In part, his explanation can be read as showing that the kinds of strategic behavior found under first-past-the-post can also be found under STV provided that party strengths within a district are known by voters. In consequence, votes will flow to the larger parties Both of these chapters point out the importance of lower-order preferences and that one of the virtues of STV in allowing voters to express these lower preferences is not without consequence for the practical politics of parties. More widely, these contributions also point up the wider range of strategic behaviors that are possible under STV than under first-past-the-post.

Seemingly Narrow Technical Differences As Gallagher cautions in his chapter, it is possible to overstate the determining effects of electoral systems. Sometimes we may remain “uncertain as to just how many features of a country’s political system can really be attributed to the electoral system.” But some differences are clearer, and others may turn on what seem to be relatively narrow and technical issues. As Ben Reilly and Michael Maley show, for example, in comparing preferential voting in single-member districts to preferential voting in multimember districts, we can see quite sizable impacts as district magnitude shifts. But such differences need not be as large or as dramatic as a shift in district magnitude, which, arguably, changes the single transferable vote system into an entirely different system altogether—the alternative vote—and vice versa. As that chapter, Hughes’s chapter, and also David M. Farrell and Ian McAllister’s chapter show, other differences can be produced by allowing voters to treat the ballot as a list system Page 14 →(the so-called ticket vote in Australia) or by altering how many preferences voters must express. Thus, for example, although Australian federal elections insist that voters write down a more or less complete preference ordering across all candidates, Irish rules and rules for several of the Australian states allow voters to mark far fewer (optional preferential voting). In consequence, we see variations in spoiled ballots, invalid votes, and, most dramatic of all, in the kinds of preference transfers that occur. As Farrell and McAllister show, the transfer of preferences across candidates in Australian Senate elections involves the massive transfer of votes between candidates overwhelmingly within the same party in an electoral equivalent to program trading on the stock exchange. By contrast, Irish elections and elections within Australian states such as Tasmania see a far more untidy transfer of preferences, with the possibility of leakage of preferences to and from candidates of other parties. This far less disciplined transfer of preferences thus sets the context for the kinds of strategic behavior considered by Laver and Jesse. Overall, as noted earlier, this volume marks one in a sequence of studies that stems from a series of meetings held in Irvine, California, that took a broadly similar approach to the study of electoral institutions. Taken together, these volumes mark a new approach to the study of electoral institutions. The intent is to apply a most-similarsystems research design in which the same electoral system is examined in different settings. Each volume in the series of studies makes the same broad point: that significant differences exist within types of electoral system. Together they allow us to better understand the differences across electoral systems. By structuring each volume in such a way as to attempt to hold the electoral system constant, we must look to other causal factors to explain differences among the cases considered. It is especially the behavior of political parties, who seek—and find—ways to blunt institutional incentives, that we find to be the major sources of explanation for cross-national variations and limits to the theoretically anticipated impact of electoral institutions.

NOTES 1. See, for example, Davidson and Grofman (1994). 2. Although these may be bad things from the point of view of the parties, we should note in passing that for advocates of STV such as Lakeman, these are potentially quite good possibilities because they would be produced from the ground up, as it were, by voters themselves deciding whether there should be organized parties at all (Lakeman 1974, 271).

Page 15 →

Overview: Preferential Electoral Systems Page 16 →

Page 17 →

Through a Glass Darkly: Understanding the World of STV David M. Farrell and Ian McAllister In general, strong PR advocates tend to be strongly in favor of the STV form of PR. It is ironic that list PR, which is the most common electoral system in Western democracies, does not have any enthusiastic champions. STV may be the theoretically optimal form of PR in the opinion of the academics, but, in practice, list PR is more attractive to established political parties and hence much more widely used (Lijphart and Grofman 1984, 6). Since Lijphart and Grofman made that statement a decade and a half ago, the political world has changed dramatically. Eastern and Central Europe and the former Soviet Union have embarked on a path of democratization involving, inter alia, the adoption of electoral systems suitable for burgeoning liberal democracies.1 Meanwhile, established democracies such as Italy, Japan, and New Zealand have taken steps to change their electoral systems, and others (e.g., the United Kingdom) look set to follow.2 With the singular (and only temporary) exception of the small state of Estonia (Taagepera 1990; Wilder 1993), none of the electoral reforms has involved the adoption of STV. In short, the past decade has borne out Lijphart and Grofman’s observation. Although it has not been adopted by any of these states, STV has still featured prominently in debates about electoral reform. It remains much written about in the academic literature as a system with positive features and has many proponents (Bogdanor 1984; Farrell 1998; Lakeman 1970; Newman 1992; Taagepera and Shugart 1989; Wright 1980). Among the positive features attributed to the system are that it permits voters maximum possible Page 18 →choice on their ballots and the restrictions it places on the power of the party bosses in determining the order of candidate election. But it is probably for these same reasons that political parties do not favor it.3 Another disadvantage of STV is that, to date, only two small countries—Ireland and Malta—have adopted it for their national lower-house elections. Questions are raised about the applicability of such an electoral system in larger, more populous countries. The fact that STV is a rare system in its application does not aid the case for its adoption. Inevitably, attention focuses on the apparently negative political consequences of STV in those systems where it is used. In fact, most critical attention is given to just one case—Ireland. In Ireland, the problems of intraparty factionalism and excessive attention to localist, particularistic concerns are attributed to politicians who must compete with each other for votes on ordinally arranged STV ballots (Bax 1976; Chubb 1963; Sacks 1976). For the most part, such arguments ignore the fact that Irish political culture is endemically localist, like that of Italy, which does not use STV, and to blame the electoral system is disingenuous (Baranski and Lumley 1990; Gallagher 1987a). They also ignore the fact that there are other STV examples at the regional and national (upper-house) level. In the United Kingdom, post-1973 regional elections in Northern Ireland have been conducted under STV, and in Australia, elections to the upper house, the Senate, also use STV. There are, therefore, many examples with which to test the consequences of STV in a range of different political environments. STV suffers from a credibility problem. It may be the case that people know of its existence, that textbooks explain—with varying degrees of clarity—its operation, and that prominent lobbyists campaign for its adoption. But to what extent is the system fully understood? For that matter, to what extent is there actually a system called STV? In this chapter, we show that depending on where you look, STV varies considerably in the forms that it takes. This has important implications for the debate about general applicability of the system as well as for the issue of its political consequences.

The Development of STV It used to be the case that the types of countries that adopted STV formed a pattern that was said to have lawlike status. As Lijphart observes: “[W]e still have a perfect social science law without any major exceptions—very rare

in the social sciences—linking political culture with forms of PR. When Anglo-American countries use PR, they always choose STV; in other countries, the Page 19 →choice is list PR” (1987, 100).4 Estonia’s decision to adopt STV in 1989 endangered this law, though only temporarily, as the Baltic state switched to list PR in 1992 (Taagepera 1990; Wilder 1993). The law was more seriously damaged by New Zealand’s decision in 1993 to adopt a mixed-member proportional system (Jackson and McRobie 1998). The United Kingdom looks set to shatter the law once and for all if the recommendations of the Labour government’s recent Independent Commission on electoral reform—which proposes a mixed-member system—are implemented (Farrell 2000).5 One reason for this “law” has been the Anglo-American tradition of direct election. Prior to New Zealand’s referendum, the members of the lower house in every Anglo-American country had been directly elected, rather than gaining their place as a result of ballot placement. In other words, voting is candidate centered rather than party centered, as is commonly associated with closed-list PR. As Sinnott (1993) observes, STV “involves a notion of the connection between the individual representative and his or her constituency that is much closer to the notion of representation implicit in the first past the post system than to the notion of the representation of parties underlying list systems” (68). In practice, that has meant a system based primarily on candidates, with parties being less central. Indeed, the party affiliation of candidates was not listed on ballot papers in Ireland until the 1965 election. For the Australian Senate, parties had been allowed to rank-order their candidates and group them together since the system began in 1949, but it was not until 1984 that party names were included on the ballot paper. The name given to this electoral system, which we are referring to as STV, has varied in the countries where it is used. STV was originally known as the “Hare system,” after the English barrister, Thomas Hare (1806–91), who is credited most (together with the Dane, Carl Andrae) with devising it in the 1850s (Bogdanor 1981; Butler 1963; Hart 1992). In Ireland and the United Kingdom, the system is referred to as “STV” (or even as “proportional representation”), whereas in Malta it is known as the “Hare-Droop” system. Matters are somewhat more confused in Australia. In Tasmania—the first state to use STV—the system is called “Hare-Clark,” the latter name after Andrew Inglis Clark (1848–1907), who, as Tasmanian attorney general, was responsible for introducing STV there. There has been a tendency to also refer to the Australian Senate system as Hare-Clark. As we will see later, this is inappropriate for a number of reasons: It is more accurately entitled “Senate-style PR” or alternatively “the Australian upper-house version of STV.” For the most part, STV has tended to be used in local authority elections. Although the evidence is sparse, it suggests widespread usage in a number of Page 20 →countries. For example, it was first used for all-Ireland local elections in 1919, before being adopted by the Irish Free State for national elections in 1922 and foisted on Northern Ireland for local and regional elections. By the mid-1920s, Northern Ireland had replaced STV with the plurality system. Other places to use STV at the local level in the earlier part of this century include Scotland, India, and New Zealand. Hart (1992, 210) refers to the use of STV in elections for Scottish education authorities on four occasions, at three-year intervals between 1919 and 1928. According to Bogdanor (1981), under the 1919 Government of India Act, STV was “proposed on an experimental basis” (133). In New Zealand, the Local Elections (Proportional Representation) Act of 1914 permitted local authorities to adopt STV. However, only Woolston Borough and Christchurch City did so, the former for the elections of 1917 and 1919 and the latter for those of 1917, 1929, 1931, and 1933. Various local elections and polls acts contained the option for STV until it was finally removed in 1966 (Royal Commission 1986, 37–38). In the United States, STV has been used at some time to elect city councils in approximately two dozen cities (Barber 1995; Engstrom 1990). The first city to adopt the system was Ashtabula, Ohio, in 1917; the last were four Massachusetts cities in 1950. The largest city to adopt STV was New York City in 1936. In most cases, however, the system has been dropped, though it is still used in Cambridge, Massachusetts, and for 32 community school boards within New York City. Among the other places where STV has been used are Canada (Alberta and Manitoba). South Africa (the Senate), and Gibraltar (for the Legislative Council) (Lakeman 1970). In 1973 it was readopted for local and regional elections in Northern Ireland and subsequently in 1979 for European Parliament elections. STV was used in 1998 to elect the new Northern Ireland Assembly established as part of the current peace process.

The Political Consequences of STV In evaluating any electoral system, there are two central questions. First, how proportional is it in translating popular votes into seats? Second, what are the effects of the electoral system on the political system, such as, for example, the numbers of parties in the legislature and the degree of governmental stability? Rae (1967) sees the first question as reflecting the “proximal,” or short-run, effects of electoral systems and the second question as reflecting the “distal,” or long-run, effects. For the most part, STV is included in the general family of PR systems in terms of its party system effects (Lijphart 1994; Page 21 →Taagepera and Shugart 1989). There are, however, disagreements over the degree of proportionality that the system provides. To some authors, this is one area in which STV can be distinguished from other PR systems. A curious feature of this debate at the academic level—at least until recently—is the inadequacy of the available tools to assess the issue of comparative proportionality (Gallagher 1991; Lijphart 1994). In particular, there appears to be considerable difficulty with including STV in any evaluation. As Lijphart (1986) has observed, many simply ignore STV altogether for this reason and instead focus on the list systems. There are two main difficulties with assessing the proportionality of STV. First, the relatively low level of district magnitude (at least as used in Ireland, which is the usual focus of attention) means that STV tends to be labeled as less proportional or—as the phrase goes—quasi-proportional (Taagepera and Shugart 1989, 207; also Katz 1984). Of course, one way around this problem is simply to ignore district magnitude and focus instead on the electoral formula. This is the approach adopted by Lijphart (1986) and also by Blondel (1969), who goes so far as to suggest that STV is the most proportional system. Second, STV is a quintessentially candidate-based system, which causes difficulties for measures of proportionality that are based on vote and seat shares for parties (Gallagher 1975; Mair and Laver 1975). Scholars have approached this problem in a number of ways. Rae (1967) makes the admission that because of this, “[i]t is not quite clear how this arrangement is likely to compare with other PR formulae” (38), and accordingly he does not attempt an overall assessment of the degree to which STV is more or less proportional than list systems. Instead, he merely concludes that “in general, the Irish formula behaves like any other sort of proportional representation. It operates quite proportionally” (111). Lijphart (1986, 175) has suggested two methods of dealing with this problem. First, it could be assumed that all voters cast a ballot for the candidates of only one party (so that there are no interparty transfers;6 alternatively, one could assume that interparty transfers cancel each other out). Second, he proposes analyzing the votes in the final round of counting (that is, after all of the transfers have taken place). Under either method, STV rules become roughly equivalent to largest remainder-Droop, which Lijphart (1986, 1994) categorizes as being intermediate between the most proportional and the least proportional system. Crucially, according to Lijphart, STV is a proportional system; it is not merely quasi-proportional. Apart from the question of proportionality, there is one other dimension of variation between STV and other electoral systems. Focusing in particular Page 22 →on the significance of intraparty transfers, Katz (1980) has drawn attention to the consequences of STV for party campaign styles. Largely on the basis of an Irish-ItalianU.K. comparison, Katz finds evidence that STV in Ireland tends to emphasize local campaigning, with intraparty rivalry, a focus on district work and local concerns, and a low importance attached to ideology and national issues. This has wider organizational and representational implications: “As predicted by the theory of electorally determined parliamentary behavior, the matters of real importance to the [Irish] deputies are constituency service, and on these matters deputies who must electioneer independently continue to act independently” (108). So far, we have assumed that STV is represented by one system; indeed, for the most part, the literature often implies that there is one system called STV. In reality, however, despite the fact that STV is a comparatively rare system (in its application in legislative elections), there are considerable variations in the way in which it is applied. The next section sets out the major patterns of variation among the different forms of STV.

The Major Forms of STV At present STV is used for major political elections in only four systems: Australia (Australian Senate; upper houses of New South Wales, South Australia, and Western Australia; Tasmanian lower house; the Legislative

Assembly of the Australian Capital Territory [ACT]), Ireland (lower and upper houses; local and European elections), Malta (legislative elections for the unicameral parliament), and the United Kingdom (Northern Ireland local elections; Assembly and European elections). The actual application of STV varies across these cases. In this section we show how the variations fall into four major groupings. Table 1 provides a summary of the four main types of STV in operation, outlining the five possible dimensions of variation. The STV systems are listed from left to right according to the date STV was first used on a wide scale: Tasmania in 1909 (since used in 28 elections), Malta in 1921 (20 elections), Ireland in 1922 (28 elections), and the Australian Senate in 1949 (20 elections). There are, first, variations in district magnitude (characteristic I). As a general rule of thumb (see also Taagepera and Shugart 1989), district magnitude needs to be at least five members to ensure a proportional result. On average, therefore, Ireland is disadvantaged, with only a minority of its constituencies having five members. And in some specific cases—notably in the ACT and Northern Territory, where there are only two seats for the Senate—Australia also falls below this threshold.7 We can expect, therefore, to find notable variations in degrees of disproportionality. Page 23 → TABLE 1. Variations in STV Systems Page 24 →Second, there are variations in what is expected of the voters (characteristic II). In Ireland and Malta, voters need only mark a first preference on the ballot paper for their votes to be counted as valid.8 In Tasmania and the ACT, by contrast, voters have to declare at least as many preferences as there are seats to be filled.9 The Australian Senate operates the most extreme form of ordinality rule, requiring electors to declare a preference for each candidate on the ballot paper (or, at least, for all but one candidate—the blank space being counted as the final preference).10 Together with compulsory voting, the requirement to complete all preferences has been seen as a major factor behind the unusually high numbers of invalid (or “informal”) votes in Australian elections (Gow 1971; Hughes 1966; McAllister and Makkai 1993). The Australian electoral law was reformed in 1984, ostensibly to deal with the problem of spoiled votes; though, as we discuss later, this reform also has important implications for the nature of party control over candidate nomination and vote choice. Since 1984, electors in Australian Senate elections can choose to vote in one of two ways. The ballot paper is divided in two. Electors can, as before, either declare a preference for all the candidates on the ballot paper, or they can vote “above the line” of division, in this case simply putting a 1 against the name of their preferred party. This latter option means that they accept the candidate ordering proposed by the party leadership and, furthermore, that they accept the party leaders’ decision on the order in which votes will transfer to other parties later in the count. Needless to say, it is the preferred option of the vast bulk of Australian voters. Third, variations in ballot paper design (characteristic III) have important implications for the candidates.11 The issue here concerns the shortcuts that voters take in filling out their ballot papers, generally referred to in the literature as “alphabetical” or “donkey” voting (Darcy and McAllister 1990). The more the voter is taxed by the system, the greater the likelihood that he or she will make use of shortcuts. Therefore, in a compulsory voting system (all Australian STV elections) and where there are minimum preference-voting requirements (to varying degrees in all Australian elections), donkey voting can be expected (Darcy and Marsh 1994; Kelley and McAllister 1984; Mackerras 1970). This problem is further compounded in Australia by the presence of a large number of non-English-speaking migrants, many from countries lacking democratic traditions. With minimal information about the electoral system, they often utilize shortcuts when completing the ballot. Page 25 →There are two ways of dealing with this problem. One is to make full use of it in a party’s strategy and list party candidates in order of importance. This is the procedure followed for Australian Senate elections, where the rank ordering of candidates is set by the party organization. This is heavily reinforced by the vote-above-theline option (ticket preferences), introduced in 1984. The introduction of ticket preferences had the effect of immediately reducing spoiled votes in Senate elections, as it was designed to do, but increasing them in House of Representatives elections, where the ticket preferences option is not used. More recently, spoiled votes have declined to about 3 percent in each election. An alternative strategy is to counteract the problem by rotating the

position of the candidates, thereby giving each a fair chance. This procedure was adopted in Tasmania in 1980 and immediately reduced the incidence of alphabetic voting (Darcy and Mackerras 1993). Ballot paper design also has important implications for the degree of leadership control over candidate nomination and, indeed, for leadership influence over candidate election. The Australian Senate represents the extreme case where, in effect, a senatorial candidate needs to pay more attention to the party leadership than to his or her voters. As Jaensch (1986) puts it: “[I]n the real world of parties and voters in Australia . . . position on the ballot paper is crucial to re-election” (58). A candidate for one of the larger parties placed high on the list is virtually guaranteed election; therefore, the senatorial campaign is really less one of trying to win the hearts and minds of the Australian voters and more one of trying to win the hearts and minds of party selectorates (Bowler, Farrell, and McAllister 1996). In all the other systems, which operate either by alphabetical placement or rotation, the party leadership has no direct influence. A fourth source of variation is provided by the counting method that is adopted (characteristic IV). In Malta and Ireland, when transferring surpluses, the calculation is based solely on the next preference marked on the ballot paper. This can have important implications in later counts, particularly when the result between two candidates is very close. Depending on which ballot papers were selected from the pile at an earlier stage in the counting process, one candidate could gain or lose. In other words, there are random effects involved in the counting process. As Gallagher’s detailed overview of Irish elections from 1922 to 1944 reveals, close results are not uncommon (Gallagher 1993b; Gallagher and Unwin 1986). One solution to this problem is to adopt the Gregory Procedure, which is used in Tasmania, the ACT, the Australian Senate (since 1984), and for elections to the Legislative Councils of Page 26 →South Australia and Western Australia. Here, instead of selecting surplus papers to be transferred, all of the elected candidate’s ballot papers (not including papers received from an earlier vote transfer) are transferred but at a fraction of their face value (Wright 1980). Finally, different practices are followed for the filling of casual vacancies (characteristic V). On the one hand, in the case of the Australian federal and state upper houses, there is the assumption that the seat belongs to a party and that therefore the filling of the vacancy should be determined by that party. On the other hand, in Malta, Tasmania, and the ACT there is the belief that the voters have decided and thus that the best way to fill the vacancy is to go back to the original count and reallocate the quota, awarding the seat to the next candidate. Ireland falls between these two extremes, preferring to ask the voters to decide again in a by-election (and thereby using the alternative vote system). Only in Ireland, therefore, does a government have to contend with a midterm test. In this aspect, the Irish system can also be said to favor the larger parties, which stand a greater chance of having their candidates elected in a majoritarian by-election.12

STV and Party Strategy It is possible to identify distinct tactics in the treatment of vote preferences in the various STV systems. This can be illustrated by a straight comparison between Irish and Australian vote management tactics, which reveals two very different patterns, each of which are held to be of maximum benefit to the party. First, there is the strategy of concentrating first-preference votes on particular candidates and letting any surpluses transfer down the line to the other party candidates. This can be termed a “plump for one” (PFO) strategy. It is the established practice in all Australian upper-house STV elections. It has been attempted unsuccessfully in Tasmania for House of Assembly elections, where voters ignored “how to vote” cards (Wright 1986, 135–6). PFO is obviously the preferred method of any politician wishing to hold onto his or her seat in a multiseat district (Farrell 1985). However, there is an important difference between a series of PFO strategies adopted by individual candidates and a centralized PFO strategy for the entire district that targets just one candidate. The latter is the practice in Australia; the former is traditionally the practice adopted by machine politicians in Ireland. The Irish practice, which operates in a system renowned for bailiwick politics, is an appropriate response by politicians to “friends and neighbors” voting behavior. In essence, the districts are carved up among the different party candidates, and Page 27 →the voters’ rank ordering varies accordingly. What this amounts to is an entirely different vote management strategy that can be termed “spread the preferences” (STP).

Irish parties have adopted STP as the principal form of vote management. As the parties have professionalized and centralized their operations, new procedures have been adopted to maximize the efficiency of the STP strategy. Party statutes have been amended, and greater use has been made of existing rules as the central leadership has increased its role in candidate nomination (Farrell 1994; Gallagher 1988; Mair 1987). The parties issue candidate cards (equivalent to how-to-vote cards) that vary the rank ordering of their candidates in different parts of the district. District polling is used to calm the nerves of those candidates being asked to sacrifice first preferences for running mates (Farrell 1993). It is generally argued in the literature that for parties aiming to maximize the numbers of seats they win, STP strategies are far more effective than PFO strategies (Gallagher 1992; Wright 1986; for an exception, see Cohan 1979).13 For instance, Gallagher (1992) is clear in his mind that STP is usually the superior strategy: “Under the assumption that no votes transfer across party lines, a party’s ideal strategy under STV is to ensure that all of its candidates have exactly the same number of votes at every stage of the count, and that when any one is eliminated, his or her votes transfer evenly among all the party’s other candidates” (482). An inappropriate distribution of the preference vote between candidates of the same party can lead to that party losing an otherwise winnable seat. The obvious question is this: If STP could result in more candidates being elected, why do Australian parties not use this strategy for Senate elections instead of PFO? Why pile up preferences if spreading them can produce more gains? There are two possible explanations. First, it could be argued that STP is really only appropriate in a system characterized by a prominence of friends-and-neighbors voting, where there is a high degree of localism in voting behavior. Such is evidently the case in Ireland, but is it in Australia? Research by Johnston (1978) suggests that there is localism in Australian voting practice. His examination of the 1974 Senate election in Victoria found “very strong evidence for the existence of friends-and-neighbors effects” (152). However, voting takes place across a whole state or territory, so the degree of localism may vary considerably. A second, alternative explanation is that STP is a risky and complex strategy. Its success depends on the size of the party vote in the district, on the degree of solidity in vote transfers, and on the actions of the other parties in the Page 28 →race. It is an experiment that can go badly wrong. This problem need not apply in the Australian case, however, for one very good reason—ticket voting. Australian parties are virtually guaranteed maximum vote solidity in the transfer of preferences between their candidates. Indeed, because of the ticket-voting system, Australian parties have an ideal situation should they want to adopt STP strategies. All they would have to do would be to change the electoral law to allow the rotation of candidate names (below the line) on the ballot paper while maintaining the ticket vote. This would ensure an extremely efficient STP vote management strategy, guaranteeing the spread of preferences across all the candidates. Consistent with the arguments set out in the introduction of this volume, the Irish and Australian cases demonstrate clearly the extent to which the institutional effects of the electoral system can be mediated by party strategy. Another interesting illustration of this point was provided by a mock-ballot STV simulation of London voters coinciding with the European Parliament elections of 1994 (Bowler and Farrell 1996). For the purposes of this simulation, London was divided into two multiseat districts, and the ballot papers contained the names of some of the key candidates running in the “real” election. Given that by definition such a simulation allowed no scope for party strategy, here was an ideal opportunity to observe the preference-transfer tendencies of this sample of voters. The patterns revealed an intriguing midpoint between the Irish STP and Australian PFO tendencies, providing “a useful benchmark of how UK voters would express their preferences if left to their own devices” (Bowler and Farrell 1996, 29). Of course, in a real STV election the voters would not have been left to their own devices; as the Irish and Australian cases demonstrate, the British parties would face a choice of strategies over how to play the system.

Implications for the Political System The examples of vote management tactics outlined previously serve to illustrate the implications of different STV systems for party strategies. Far more significant are the implications of variations in STV for the political system generally. Two questions are worth asking: Which of the STV systems is more democratic, and which is more proportional? In answering the first of these, we can again draw a distinction between the Irish and Australian

Senate systems. On the one hand, it could be argued that the Australian upper-house version of STV has more democratic outcomes than most—if not all—of the Page 29 →other STV systems. As Rae (1967, 37) points out, the low level of district magnitude generally associated with STV means that not all lower preferences are taken into account in the counting process; that is, the bulk of seats are filled after only a few counts (see also Bowler and Farrell 1991a, 1991b; Darcy and Marsh 1994). This means that with lower levels of district magnitude there are higher proportions of wasted votes. In one sense, because of its high levels of district magnitude, the Australian Senate version of STV actually comes closest to the system originally intended by Thomas Hare. This point is repeated by all authors (without exception), who argue that one of the limiting features of STV is that it cannot incorporate a large district magnitude. Indeed, Thomas Hare had envisaged all of the United Kingdom as one giant constituency! This was one issue with which his followers disagreed; to facilitate the voters, STV systems have always worked with relatively small district magnitudes. According to Hermens (1984), the “largest STV constituencies ever used” (20–21) were in the boroughs of New York between 1937 and 1947. He cites one example of Brooklyn borough in 1937 where there were 99 candidates. The ballot was more than four feet long. As a result there were large numbers of invalid and exhausted votes, and there was clear evidence of alphabetical voting. Contrary to what Hermens may believe, in fact, Australia probably holds the record for the largest STV constituencies over time as well as the largest ballot papers. For example, in New South Wales in 1987 there were 50 candidates on the ballot paper; in 1990 there were 62 candidates, in 1993 there were 66, and in 1995 there were 99. On the other hand, the Irish version of STV allows voters more choice and greater flexibility in their choices. Irish voters are not forced to declare all their preferences; there is a greater element of free choice. Furthermore, the use of ticket voting in Australian Senate elections since 1984 has given even more power to the party bosses than they would have under the most rigid of party list systems. Not only can the party bosses determine the order in which their candidates are elected, but they can also decide on how voters’ later preferences will transfer to other parties. The extent to which such a practice is consistent with the ideas of STV is highly questionable. On the second major issue—the implications of different STV systems for proportionality—the study of electoral systems has progressed far in a short period of time. Only a few years ago, Gallagher (1991) observed that “there is surprisingly little discussion of what exactly we mean by proportionality and how we should measure it” (33). Research by Gallagher and others (Katz 1997; Lijphart 1994; Taagepera and Shugart 1989) has done much to fill this gap. In particular, Lijphart’s recent study (1994) has added greatly to our Page 30 →understanding of the consequences of different electoral systems for the political system, including those of both Ireland and Malta. To maximize the comparability of our research with Lijphart’s work, we have followed the same methodological criteria. Accordingly, an electoral system is defined, in this section, as “a set of essentially unchanged election rules under which one or more successive elections are conducted” (Lijphart 1994, 13). We also focus on exactly the same time period as does Lijphart, studying all elections from 1945 to 1990. The principal points of distinction, in our case, relate to district magnitude/effective threshold and assembly size because we are dealing with only one electoral formula. Here we follow Lijphart’s rule that there should be a 20 percent change for it to constitute a new system. As Table 2 indicates, this results in 10 different STV systems.14 The 10 cases are listed in terms of increasing size of district magnitude (and decreasing effective threshold). District magnitude (DM) is defined, in the usual manner, as the average number of seats per district. In the table we see the major distinction this produces between full-Senate elections (following what are known as double dissolutions) and half-Senate elections in Australia. AustraliaDD (1–2) and Malta (1–2) also show varying tendencies, with the former reducing its district magnitude in recent years (owing to the new seats for the two territories, but note that the change is only greater than 20 percent for full-Senate elections) and the latter increasing its district magnitude.15 Lijphart (1994)—following Taagepera and Shugart (1989)—proposes a refinement of district magnitude, the effective threshold (ET), which is designed to take account of the two main thresholds in electoral systems: the threshold of inclusion and the threshold of exclusion (see the introduction of this volume). The resulting figure (seen as an estimate and therefore presented as a value to only one decimal

place) is described as a midpoint in a range between no representation and full representation. In Lijphart’s study, the STV systems have the lowest district magnitudes of all the PR systems. He argues that this is inevitable because higher district magnitudes entail large numbers of candidates and thereby “impose heavy burdens on the voters who have to rank order the candidates” (1994, 30). As we discussed earlier, the use of ticket voting in Australian Senate elections does have the important positive feature of facilitating a larger district magnitude. But the Tasmanian district magnitude of 6.64 (ET of 10.3) demonstrates that STV systems can be more proportional without experiencing the extremes of ticket voting. As indicated by the rank orders in table 2, adding the Australian Senate and state elections to the population of STV systems increases average DM and decreases average ET. The respective means for all nine STV cases (excluding Malta3) are 7.8 and 10.3. This compares favorably with an average DM of 9 and an average ET of 8.6 for the thirteen d’Hondt systems in Lijphart’s study (excluding Israel, the Netherlands, and the Euro cases) and average scores of 7.7 and 9.1, respectively, for the four other non-d’Hondt systems (excluding Israel and the Euro cases).16 Page 31 → TABLE 2. Types of STV Systems, 1945–1990 Lijphart’s study focuses on the effects of these independent variables (DM/ET and assembly size) on four main indices of disproportionality: Rae Page 32 →(1967), Loosemore and Hanby (1971), Gallagher’s (1991) leastsquares index, and Lijphart’s own measure of largest deviation. His expressed preference is for the least-squares index, although, as he shows, all four are highly intercorrelated. Again, to maximize comparability with his research, we have calculated the four disproportionality indices for our 10 cases, and these are reproduced in the appendix table.17 Furthermore, in the appendix table we also list the party system characteristics of the different STV systems.18 As Lijphart shows, lower levels of disproportionality are associated with lower effective thresholds and larger assembly sizes. Because it is not always the case that systems with large assembly sizes also have the lowest effective thresholds, one must take account of the contradictory pressures that these two variables can produce. For instance, as table 2 shows, Ireland has both the largest assembly size and the highest effective threshold. The effects of these two independent variables on disproportionality (as measured by the least-squares index) are tested in table 3. Once effective threshold is taken into account (with the single exception of New South Wales), there is a clear relationship between assembly size and disproportionality. So although double-dissolution Australian Senate STV may have the lowest level of disproportionality of all the STV systems, Ireland—arguably the more democratic of the two systems—comes a close second and, in fact, has the lowest level of disproportionality of all the systems with higher effective thresholds. TABLE 3. Average Percentages of Disproportionality, Classified by Effective Threshold and Assembly Size in Nine Electoral Systems Page 33 →

Conclusion Although it is one of the earliest and perhaps the most discussed of all forms of proportional representation, STV has been used consistently in only a small number of countries. There is no single form of STV; rather, the systems that are (and have been) used differ widely on five major characteristics, ranging from district size to ballot paper design and the methods that are used to fill casual vacancies. Perhaps most importantly, there are considerable variations in how parties seek to use the system to maximize their vote: at one extreme, the Irish parties have always sought to spread voter preferences among as wide a range of candidates as possible, while at the other extreme, the Australian parties seek to concentrate votes on one single candidate and then redistribute the preferences. Although it has been championed enthusiastically by advocates of proportional representation, and indeed even appeared at the turn of the century to be the “natural” alternative to the plurality system, STV has not been widely adopted. The most important reason halting the spread of STV has been the emergence of modern, mass political parties. When STV was being promoted almost a century ago, parties were comparatively new, and responsible

parliamentary government had not yet become responsible party government. With the continuing concentration of legislative power in the hands of parties, particularly in the postwar years, political parties have naturally sought electoral systems that preserve rather than diminish their power over the electorate. As a quintessentially candidate-based system, STV does not serve this purpose. For this reason, it is perhaps not surprising that the newly emerging democracies have generally adopted party list systems. Australia appears, at least superficially, as an exception to this pattern. The country has one of the strongest and most disciplined party systems in the world yet uses STV for the federal upper house, the Senate, and it is also used in some of the states. The reason why it has not been replaced is that the parties have been able to adapt the system sufficiently to ensure party-based rather than candidate-based voting. But even here, STV has its critics. Following the failure of successive governments with lower-house majorities to gain the balance of power in the Senate, there has been intense discussion about changing the electoral system so that the Senate result more properly reflects the result in the lower house. This was given added weight after the 1998 federal election, when a Liberal-National government was re-elected on a platform that included wide-ranging tax reform. The party holding the Senate balance of Page 34 →power, the Australian Democrats, has vowed to block the government’s tax reform legislation. What future exists for the various forms of STV? The lack of interest in introducing STV, particularly in the new democracies, suggests that it will remain the preserve of a small number of countries where it has become a permanent feature of the political landscape. Moreover, the steps that have been taken by the established democracies such as Italy to replace PR because of its alleged deficiencies do not provide a good example for countries that might otherwise seriously consider its introduction. STV will undoubtedly remain in the countries where it is well entrenched, but even here, it is likely to be placed under increasing pressure by party organizations keen to further concentrate their political power.

NOTES 1. For an earlier examination of the arguments in this chapter as well as a multivariate analysis, see Farrell, Mackerras, and McAllister 1996. 2. In France over this period, the system has changed from majoritarian (second ballot) to list PR and back again to majoritarian. 3. For a recent instance, see the report of Britain’s Independent Commission on the Voting System (1998), in which STV was judged to be too complex. 4. Another feature that the STV cases all appear to share in common is that they are round islands! We are grateful to Colin Hughes for this observation. 5. Bogdanor (1981) refers to one other example that is not consistent with Lijphart’s law. As is well known, STV was developed coincidentally by Thomas Hare (1806–91) in England and Carl Andrae (1812–93) in Denmark, neither being aware of the other’s ideas. Andrae was a government minister. In 1855 he applied his system to Danish federal elections, making this “the first occasion on which the single transferable vote system was applied to a national legislature” (105). 6. As Gallagher (1992) observes, this assumption may have some credence in Malta (and also, as we shall see, in the case of Australian Senate elections), “but it should be made clear that this does not give a full picture of the way STV can operate in practice” (480). 7. A related distinction is between those systems that have a consistent district magnitude (Malta on five and Tasmania on seven) and those that offer a range (all the rest). 8. In Ireland an X is sufficient for a vote to be counted as valid. Similarly, in the case of Australian Senate ticket voting, an X above the line is also counted. Page 35 →9. This requirement was seen as a necessary reform when it was decided to readopt STV in Tasmania in 1907. The system had been abolished in 1901 (after a brief test since 1896), one reason being because of the apparent habit of voters to plump for only one or two favored candidates (Newman 1992, 43–61). APPENDIX TABLE. Indices of Disproportionality and Party System Characteristics Page 36 →10. In reality a ballot paper that has most of the preferences completed (about 90 percent) is accepted; however, this fact is not advertised and therefore not generally known. Also not advertised is the

fact that if a voter places a 1 next to one candidate and a series of 2s next to the remainder, then this is also counted as a valid (or “formal”) vote. 11. Ballot paper design can also have important consequences for voters. The question here is this: Which is fairer, a ballot paper that sorts by candidates (Ireland), where the voter has to work harder to find the relevant party labels, or a ballot paper that sorts by parties (all other systems), where the voter has to work harder to find the relevant candidates? For further discussion, see Darcy and Marsh 1994. 12. Ireland also represents one of the few cases of a PR system that incorporates by-elections. As a result of its recent electoral reforms, New Zealand is another example. 13. For formal discussion of coordination failure under SNTV systems where parties fail to achieve an even spread of votes among the candidates in a district, see Cox 1997, chap. 13. 14. Because Malta3 is an unusual case, as a result of the contingent upper tier that was used in its 1987 election, we exclude it from the analysis. 15. The making of two new two-seater districts in the Australian territories in 1975 meant that average DM for full-Senate elections dropped from 10 to 8 (a 20 percent change). The equivalent change for half-Senate elections (when both senators in the territories are also re-elected) was from 5 to 4.25 (only a 15 percent change). The increase in 1984 in state senatorial seats from 10 to 12 and from 5 to 6, respectively, did not constitute a change of 20 percent, when allowance is made for the territorial seats (which remained as twoseaters). 16. We exclude Israel and the Netherlands from these averages because they have very large DMs owing to the fact that in both cases the entire nation is one district. 17. For comparisons with other electoral systems, see Lijphart 1994, appendix B. 18. Again, directly comparable with Lijphart 1994, appendix B.

Page 37 →

The Single Transferable Vote and the Alternative Vote Compared Ben Reilly and Michael Maley AV has been implemented at Australian federal elections in two different contexts. Australia has a bicameral Parliament consisting of a House of Representatives that is elected at least every three years and has always utilized single-member constituencies and a Senate, the members of which serve staggered six-year terms, with half of the membership facing election every three years (except when the Senate is prematurely dissolved). All of the originally federating states are guaranteed equal Senate representation and have always constituted at-large multimember electorates.1 AV has been used since 1918 to elect members of the House of Representatives and has remained largely unchanged since its introduction. Voters are required to indicate a rank ordering of preferences for all candidates. The essence of the system is that after first-preference votes have been counted, the candidate standing lowest on the poll is excluded, and his or her votes are transferred to the remaining candidates according to the preferences shown by the voters. This process of excluding the lowest-standing candidate and transferring his or her votes continues until there are only two candidates left.2 The candidate who has an absolute majority is elected. From 1919 to 1946, however, a form of AV that involved the application of the system to multimember electorates was also used to elect senators. Typically three vacancies had to be filled (though that number was sometimes made greater by casual vacancies). Separate scrutinies of the ballots were undertaken, one to fill each vacancy. The scrutiny to fill the first vacancy proceeded in the same way as a House of Representatives scrutiny. For the second Page 38 →vacancy, however, the votes showing a first preference for the candidate elected to the first vacancy were transferred to the remaining candidates according to the second preferences shown by the voters; thereafter the scrutiny proceeded in the same way as a House of Representatives scrutiny. In the scrutiny to fill the third vacancy, the votes showing a first preference for the candidates elected in the scrutinies to fill the first and second vacancies were transferred to the remaining candidates according to the preferences shown by the voters, and thereafter the scrutiny again proceeded in the same way as a House of Representatives scrutiny. Further vacancies, if any, were filled similarly to the third. The implementation of AV in single-member constituencies on the one hand and multimember constituencies on the other gave rise to systems that had substantially different political impacts, which are touched on later in this chapter. Although there has been general satisfaction with the way in which AV has operated for House of Representatives elections, it came to be widely viewed as an unsatisfactory way of electing senators (Wright 1980, 90; Odgers 1972, 46–47). STV was accordingly introduced for Senate elections by legislation enacted in 1948 and first put into operation at the Senate election of 1949. We commence by examining briefly the historical development of the various systems used in Australia. We then discuss a number of variants of the basic systems that have been implemented or mooted. Thereafter, we compare the operation in practice of Senate STV and Senate AV and of Senate STV and House of Representatives AV. In doing so, we seek to highlight ways in which the systems have evolved in response to local factors and the ways in which unintended consequences have influenced their development. We also note the degree of pragmatism that has marked the development of electoral systems in Australia.

Historical Developments The first serious proposal for applying a preferential balloting method to nationwide elections was put forward by Thomas Hare in 1856 as part of his proposal for a form of STV, The Election of Representatives, Parliamentary and Municipal (Hare 1873). Central to this proposal was the novel procedural step of asking voters to rank-order all candidates in order of their preference—a move that Rokkan described as “the great innovation” of British electoral reformers (Rokkan 1968, 15). From the outset, Hare’s proposal that votes should be transferable was heavily criticized by politicians as being Page 39 →overly complex and beyond the understanding of ordinary

voters. Hare saw it instead as a “vast augmentation of their electoral power” and characterized those who had argued that it was beyond the capacity of the average elector as being mischievous and patronizing (Hare 1873, viii). Like his mentor John Stuart Mill, Hare was not reluctant to ascribe wider benefits to this new institutional arrangement, seeing it as “bringing to the duty of voting reflection, judgement and moderation” (122) that “by the opportunity the voting papers afford of separating, distinguishing, and bringing out every form of political opinion, will give an immeasurable increase of force and strength to the representative principle” (127). It had been pointed out in 1863 in a report from Robert Lytton of the British Embassy in Copenhagen that Hare’s original scheme made elections in single-member districts all but impossible, owing to the overly high quota for election in Hare’s original scheme (McLean and Urken 1995, 45–46). The next stage in the development of the preferential vote was therefore the adaptation of Hare’s original proposal to single-member rather than multimember elections in what we would today call the alternative vote. The first documented example of such a method appears to have been carried out by Professor W. R. Ware of Harvard College in 1871, although the essentials of the system had been originally proposed (and dismissed) by Condorcet a century earlier. In a letter to Hare, Ware detailed the efficacy of Hare’s method when used for a mock election by Harvard students to select their four favorite writers (Hare 1873, 353). He also showed that the Hare method can be utilized to fill casual vacancies (and thus obviate the need for by-elections) by reexamining the ballots cast for a retiring member and distributing the second and later preferences accordingly. Ware used this example to show that “contrary to the generally received opinion, the system of preferential voting is applicable to the choice of a single candidate” (v). Although the significance of this advance appeared to be lost on Hare, it was recognized for what it was by E. J. Nanson, the professor of mathematics at Melbourne University from 1875 to 1922 and an influential campaigner for electoral reform in Australia and internationally. Following from the critiques made by Condorcet in the eighteenth century, Nanson made several telling criticisms of the Ware method relating to its lack of monotonicity and its failure to pick a Condorcet winner consistently. But he also praised the alternative vote for its simplicity and lack of susceptibility to strategic voting, which he saw as making it “extremely suitable for political elections” (Nanson 1900, 10). By the late 1890s, with the framing of the Australian Constitution virtually Page 40 →completed, Nanson’s primary concern was to influence the choice of electoral system for the new federal Parliament. It was clear that political pressures would ensure that this system would utilize single-member constituencies, for the House of Representatives at least, and Nanson was at pains to ensure that AV, which he saw as being “far superior” to other single-member systems, was chosen (Nanson 1900, 10). (He campaigned unflaggingly for the adoption of STV for the Senate, which would also have been his favored system for the House of Representatives were multimember electorates adopted.) In a series of articles in the Age and the Argus, collectively published in 1900 under the title The Real Value of a Vote and How to Get It at the Coming Federal Elections, Nanson emphasized “the absolute necessity of insisting that no candidate be elected in a single electorate unless he poll an absolute majority of the votes” and thus initially advocated the contingent vote, then used in Queensland (Nanson 1900, 10). The contingent vote essentially operates as a runoff election compressed into one round. If no candidate gains a majority of first preferences, all candidates other than the two leading contenders are simultaneously eliminated, and their votes distributed to one or the other of the top two, according to the preferences marked, to find a winner. The method, developed as a simple alternative to a two-round runoff election, was used in Queensland from 1892 to 1942 and for Democratic primary elections in the U.S. state of Alabama between 1915 and 1931. It has been used for presidential elections in Sri Lanka since 1978 and in 1996 was recommended as an alternative electoral system for the United Kingdom by the Labour Party’s Plant Commission, who (thinking they had invented a new electoral system) called it the “supplementary vote.” (Reilly, 1997a). Plans for a new Greater London Authority involve the direct election of a mayor by means of this system (Wilder, 1999). Intriguingly, Nanson claimed that “there is nothing new in preferential or contingent voting. This method was used in France and Switzerland 130 years ago”—an assertion that presumably refers to Condorcet’s writings, which Nanson was known to have read (Nanson 1900, 7). However, Nanson went on to emphasize that even

better than the contingent vote was a procedure that entailed an exhaustive use of preferences—that is, the alternative vote, a system which was, at that time, unused anywhere in the world (12). Nanson’s proposals had a significant influence on the first federal electoral bill in 1902 that provided, inter alia, for AV, with optional preference marking, in the House of Representatives and for STV in the Senate. The government’s stated aim in introducing AV was “to bring out in the most certain Page 41 →way possible the choice of the majority of the electorate” (quoted in Reid and Forrest 1989, 98). The progressive lineage of the proposed system was clear from the ensuing debates: references to John Stuart Mill, Thomas Hare, Catherine Helen Spence, E. J. Nanson, and Andrew Inglis Clark were made by both supporters and opponents of the bill. However, the conservative opponents of what were then quite radical proposals emerged victorious: Both optional preferential voting in the House of Representatives and STV in the Senate were deleted during the passage of the bill, to be replaced with the safe options of plurality and so-called block voting (i.e., plurality voting with multimember constituencies), respectively. According to Reid and Forrest, with the wisdom of hindsight it is easily seen now that the major proposals of 1902 were decades ahead of their time. . . . [I]t took seventeen years, until 1919, before the contingent (preferential) method of voting for House of Representatives with single-member divisions was acceptable; and it took almost fifty years, until 1948, before the Parliament finally accepted a brand of proportional representation for the Senate’s State-wide, multi-member electorates. In both cases the reforms came not as a result of the pursuit of principles of electoral justice, but from pragmatic considerations of party gain. (Reid and Forrest 1989,99) The first attempt at resurrecting AV came in 1906, when Protectionist Prime Minister Alfred Deakin wrote to his Labor counterpart, Chris Watson: Have you thought of the exhaustive ballot—compulsory voting for every candidate. The lowest dropped out & his votes divided until someone gets a majority. With this our candidate & yours could not do each other anything like so much harm & and one or others of us would win in several Reidite constituencies. Of course it must be compulsory voting for all candidates. It would be a great safety valve for both of us. What do you think? (Reid and Forrest 1989, 114) Watson, whose rising Labor Party was the clear beneficiary of vote splitting among the non-Labor candidates, was not interested. The government introduced another bill for AV in August 1906, but it lapsed with the Parliament’s dissolution, and a Liberal private member’s bill in 1911 met a similar fate (Reid and Forrest 1989, 115). It was not until 1918 that AV was finally introduced for federal elections to the House of Representatives, primarily to reduce Page 42 →the potential for splitting of the non-Labor vote by candidates from non-Labor parties.3 The increasing incidence of Labor candidates beating a divided field of conservatives under plurality rules prompted the Liberals to launch a royal commission into electoral matters, which recommended that AV and STV be used for elections to the House of Representatives and Senate, respectively. But the issue lay dormant until 1917, when the newly formed Nationalist Party, under the leadership of W. M. Hughes, was able to form a governing majority. Under considerable pressure from the farming lobby, which threatened to split the Nationalist Party’s vote by standing its own candidates, the Hughes government introduced AV for the House of Representatives and passed the Commonwealth Electoral Act 1918, a piece of legislation that, although much amended, remains the statute governing electoral competition at the federal level today. The introduction of AV was thus intimately related to the need to counter the possibilities of vote splitting and to encourage and reward collaboration or coalition arrangements between parties (Graham, 1962). The introduction of multimember AV for the Senate was effected by legislation enacted in the following year, 1919. The justification offered for this move was that it would impose on Senate elections the same principle governing House of Representatives elections, namely, that a candidate to be elected should be supported by an absolute majority of voters, and that it would avoid confusion that might arise from having different methods of marking the ballot paper for House of Representatives and Senate elections.4 When the history of multimember AV is reviewed, its most striking feature is the stable and predictable way in which it operated; each seat was

filled by a separate election but with the same electorate voting at each. Predictably, this resulted in a strong tendency toward outcomes in which the same party grouping won every seat. Of the 60 occasions on which a state was polled under multimember AV, 55 produced such an outcome.5 This resulted in periods of sustained and nationwide dominance by one party and produced on a number of occasions a Senate that was scarcely workable as a legislative body, let alone as a house of review. In the period from 1947 to 1950, for example, there were only three opposition senators—a leader, a deputy leader, and a whip—facing 33 government senators. This level of systemic dysfunction meant that the electoral system itself was primarily responsible for the erosion of public confidence in, and the legitimacy of, half the federal Parliament.6 The system gave virtually no encouragement for minor parties to participate in Senate elections. At House of Representatives elections during the period Page 43 →from 1919 to 1946, it was from time to time possible for party dissidents to win election as independents or as representatives of new parties, where they had a geographically concentrated base of support: Thirty-three such candidates were elected at general elections from 1919 to 1946. At Senate elections, however, the use of the states as electorates saw these pockets of support swamped and sufficed to exclude minor-party representation completely. It can readily be accepted that the replacement of AV by STV for Senate elections was justified on representational grounds. It was advocated by the incumbent Labor government as a reform that would enhance the status of the Senate, a view that has definitely been validated by subsequent experience (Uhr 1995). In the eyes of historians there is, however, no doubt that pragmatism strongly motivated the change: Proportional representation had long been advocated by those who wished to see an end to the lopsided majorities often occurring in the upper house. The circumstances under which the bill was finally passed, however, left the labor government open to criticism. Amid signs that the government was losing ground among the voters, it was suggested that labor was contriving to insure that even if it lost control of the house of representatives, it would maintain its hold on the senate. . . . This is precisely what happened, although the labor majority in the upper house was reduced to eight. (Fusaro 1966, 390) For all the high principles that had motivated the original proponents of STV and AV, the implementation of those systems in Australia ultimately reflected the political interests of the party in power at the time rather than more high-minded considerations of the value of a particular system. As we shall see, this trend has continued to the present day.

Variants of AV and STV7 The historical motivations for the introduction of AV largely dictated what is perhaps the most distinctive element of its Australian implementation: the compulsory expression of preferences. Among the advocates of AV and STV, it has regularly been stressed that voters should be able to choose how many preferences they will indicate (Lakeman 1970). This point of view has found little favor among Australian politicians, at least during their periods of service in government. Partly this reflects calculations of partisan interest. An Page 44 →additional factor, however, may have been the implications for Australia’s system of compulsory voting of allowing only a partial expression of preferences. If it were to be conceded that voters have the right to be indifferent in regard to a subset of candidates, it would seem to follow that voters have the right to be indifferent in regard to all the candidates, and the justification for compulsory voting, which has (until very recently) enjoyed bipartisan support, would thereby be called into question. Since 1918, voters at House of Representative elections have been obliged prima facie (and commanded by the instructions printed on the ballot paper) to indicate a preference ordering for all the candidates. At Senate elections, the situation has been similar: Between 1919 and 1931, electors were required to express preferences for a number of candidates equal to twice the number of vacancies plus one (or, if the number of candidates was less than twice the number of vacancies plus one, for all the candidates). At elections from 1934 until 1983, the voters’ task was made even more onerous: They were required to express preferences for all the candidates. The strong

connection between this provision and the tendency for voters to render their votes invalid when faced with long lists of candidates to be rank-ordered is illustrated in table 1. When AV was introduced for Senate elections, the invalid vote more than doubled compared with the previous election, and it remained similarly high after the introduction of STV in 1948. From 1901 to 1917, the invalid vote was less than 6 percent at 5 out of 7 elections. From 1919 to 1983, the invalid vote was less than 6 percent at only 1 out of 24 elections. Within that period, no voters were more sorely tested than those issued a New South Wales Senate ballot paper in 1974. Faced with 73 candidates, some 12.3 percent of voters, heavily concentrated in Labor constituencies, cast invalid votes. The final seat was won by the Liberal Party by a margin of 31, 736 votes—only 1.3 percent of the total formal vote in the state—and that win produced a Senate consisting of 29 Liberal/National Representatives, 29 Labor representatives, and 2 others holding the balance of power. It came to be perceived almost immediately that the high invalid vote in New South Wales may have cost the then Labor government its chance to force its legislative program and budget through the Senate, with the ultimate result being a legislative stalemate that was broken in dramatic and unprecedented circumstances by the dismissal of the incumbent government in November 1975 by Governor-General Sir John Kerr (Mackerras 1974).8 The 1974 experience thus highlighted, more than any other single episode, the possible political impact of requiring a full expression of preferences. Page 45 → TABLE 1. Invalid Votes at Senate Elections in Australia Page 46 →The main variants of AV and STV that have been mooted or introduced for federal elections in Australia have represented a response to the effects of the requirement for voters to express full preference orderings. They are optional preferential voting, various saving clauses, and ticket voting.

Optional Preferential Voting Optional preferential voting (OPV) is, as described earlier, identical to full preferential AV except that voters are not required to express a preference for every candidate; if they wish, they can express a preference for only one. OPV is currently used for state elections in New South Wales, where it was introduced by the Labor government in 1981, and in Queensland, where it was introduced in 1992 on the recommendation of the Electoral and Administrative Review Commission, which considered that AV as practiced at the federal level forced voters to express preferences for candidates about whom they may know little or nothing (Electoral and Administrative Review Commission 1990, 59). The rate of plumping by voters—expressing a preference for one candidate only, effectively turning the system toward a straight plurality contest—has tended to increase over time in both of these states. The decision to express preferences also appears to be closely related to the recommendations made by parties on their how-to-vote cards.9 In a survey conducted at two New South Wales by-elections in 1992, fully 75 percent of electors followed party voting directions, resulting in plumping rates of 43 percent in one district (Gordon) and 63 percent in another (Ku-ring-gai). In the Ku-ring-gai case, less than 33 percent of electors filled in all squares on the ballot paper (personal [fax] communication, C. J. Craven, State Electoral Office of NSW, 12 July 1996). In Queensland, plumping rates stood at 23 percent at the first OPV election in 1992 but were significantly higher in those cases where how-to-vote material from one of the major parties did not suggest preferences (Electoral Commission Queensland 1995, 14). There is also a clear partisan component to plumping rates that reflects the long-standing coalition arrangements between the Liberal and National parties: In both NSW and Queensland, Labor voters are considerably more likely to plump than supporters of the coalition parties (Lucy 1982, 105). OPV is probably the only form of preferential voting suited to conditions of low literacy or numeracy—such as applied, for example, in the then Territory of Papua New Guinea, which successfully used OPV for its first three mass suffrage elections between 1964 and 1975 (Reilly 1997b). Although this has been the only application of OPV on a national level to date, OPV has been widely applied at the provincial level: It was first used earlier this century Page 47 →for elections to lower houses in Western Australia and Victoria and in the Canadian provinces of Alberta, Manitoba, and British Columbia.

In all of these cases it was abandoned—for full preferential voting in the Australian states and for plurality voting in the Canadian provinces—after voters increasingly used their vote to plump. There were also several unsuccessful attempts to introduce OPV to New Zealand and the United Kingdom in the early part of this century. In New Zealand, the short-lived Mackenzie government of 1912 introduced a bill to do so, but it lapsed with the defeat of the government after only 15 weeks in office, and further attempts in 1923 and 1938 met similar fates (Royal Commission on the Electoral System 1986, 31). In the United Kingdom, the bipartisan Speaker’s Conference of 1916–17 made a majority recommendation that OPV be adopted in all non-borough constituencies, amounting to about two-thirds of all electorates. The government refused to accept this recommendation and moved a free vote on the choice of electoral system, with the majority of the House of Commons in favor of the preferential vote but with the House of Lords committed to STV. The end result was a stalemate between the two houses and the maintenance of the plurality formula—to the historical cost of both the Labour and Liberal parties. Another attempt by the minority Labour government (with Liberal support) to introduce OPV in 1930 met with the same lack of success. But support for AV in the United Kingdom has recently made something of a comeback, with the Jenkins Commission recommending in 1998 that a combination of AV and party list proportional representation replace plurality for future British elections, suggesting that some 80 percent of all seats be elected by AV in single-member districts (Independent Commission on the Voting System, 1998). In Australia, compulsory preferential AV has had two main positive effects for the non-Labor parties: It has facilitated the coalition arrangement between the Liberal and National parties by enabling them both to stand candidates in some seats without the danger of vote splitting, and it has enabled the preferences of one small party that had split from Labor, the Democratic Labor Party, to flow predominantly against Labor and hugely assist the coalition maintaining control of government throughout the 1960s. The Liberal and National Parties, having been in a more or less close electoral alliance at the federal level since 1922, have thus consistently seen OPV as being against their interests and have strongly resisted its implementation on a number of occasions. The Labor Party’s position on the matter has been far less consistent. At the federal level, Labor had until the early 1970s traditionally opposed preferential voting of any sort, seeing it (correctly) as a means by which the nonLabor Page 48 →forces were able to aggregate their vote shares to defeat Labor candidates, and had been committed to revert to plurality voting. Labor leader Gough Whitlam, who had an abiding interest in electoral systems, took a different view, supporting OPV as “perhaps the only electoral procedure in the world which allows electors to express their indifference to candidates” (Whitlam 1985, 679). During its period in office from 1972 to 1975, Whitlam’s government attempted to implement OPV, but the legislation in question was rejected by the Senate. When Labor regained power in 1983, its policy was still to introduce OPV. This position appeared to enjoy popular support: A national survey in 1979 showed that the majority of electors favored the optional version, with 72 percent for optional and only 26 percent favoring compulsory preference marking (Hughes 1990, 141).10 When Labor was returned to power in 1983, it lacked a majority in the Senate and rapidly concluded that it had little hope of forcing through a change to OPV. This proved to be fortuitous for the Labor Party; as new issues on the left of the political spectrum such as environmentalism became increasingly prominent, so the second and later preferences of voters supporting parties appealing to that element of the electorate drifted increasingly to it. At the 1990 election, where support for the Australian Democrats and Green parties reached its height, the Labor Party appeared to be the beneficiary of around two-thirds of the preferences from these parties (Papadakis and Bean 1995, 103–4). The success of Labor’s strategy in 1990 was notable not least because historically the preference distribution process had tended to assist the non-Labor parties. The Labor Party consequently changed its official policy in 1991 to support “full preferential voting for all elections for lower houses of Parliament” (Australian Labor Party 1991). Party pragmatism was the overriding motivator in each case. Thus, despite the rational arguments in its favor, the extent to which its validity is accepted internationally, and the degree of public support that it appears to enjoy, there has never been a time when OPV seemed close to being implemented for federal elections in Australia, and at the moment its prospects of being introduced seem as remote as ever. Saving Clauses Despite their reluctance to recognize a right of voters to choose how many preferences they will express,

Australian politicians have been prepared to make some concessions to the difficulties associated with marking a ballot paper. These have taken the form of so-called saving clauses, under which a Page 49 →vote marked otherwise than in accordance with the instructions printed on the ballot paper may be accepted as formal. The first such clauses were enacted in 1983, but those relating to House of Representatives elections were repealed in 1998. Different rules are therefore now used to determine the validity of votes cast for the House of Representatives and the Senate. At a House of Representatives election, a ballot paper must show consecutive preferences for all the candidates (Commonwealth Electoral Act 1918, paragraph 268[1][c]). At a Senate election, different rules apply depending on the number of candidates (Commonwealth Electoral Act 1918, subsections 270[1] and [3]). Where there are 10 or more candidates, the ballot paper is considered formal if it expresses a first preference (only by the use of the numeral 1) for one and only one candidate and has, a sequence of consecutive unrepeated numbers commencing with 1, or numbers that with changes to no more than three of them would be in such a sequence, in not less than 90 percent of the squares opposite to the names of candidates. Where there are 9 or fewer candidates, the ballot paper is considered formal if it expresses a first preference (again, only by the use of the numeral 1) for one and only one candidate and has a sequence of consecutive unrepeated numbers commencing with 1, or numbers that with changes to no more than two of them would be in such a sequence, in all the squares opposite to the names of candidates or in all those squares except one square that is left blank. Several points can be made about the Senate rules. For all their complexities, they are attempting to express a relatively simple idea: A distinction should be drawn between those voters who have attempted to express preferences for all candidates but who have made “mistakes” in numbering (whose votes deserve to be saved) and those who have not attempted to express a full preference ordering (whose votes are not worthy of salvation). This interpretation is not a matter of inference; it was explicitly stated as the rationale for the rules when they were enacted by the Parliament (Australian Senate 1983, 3210–16). Secondly, in their content, they contrast distinctly with the principle applied in most democratic polities that a vote should be counted so long as the voter’s intention is clear. At recent federal elections in Australia, ballot papers have been ruled invalid solely because the voter had used a tick in lieu of the numeral 1 to indicate a first preference. Finally, it has always been clear that the political parties never intended the savings clauses to dilute in any significant way the force of the obligation to express full preferences. Confirmation of this comes from the fact that there is no reference in the instructions on the ballot paper to those clauses; voters are still instructed to show preferences for all candidates. Page 50 →Ticket Voting The fear felt by parties regarding the possible consequences of OPV arguably has an even greater rational basis in relation to the Senate than in relation to the House of Representatives. If a voter votes for a party’s sole candidate for the House of Representatives under AV, the party will gain the full benefit of that vote. If, however, the same voter, perhaps from force of habit, votes for only one of the party’s Senate candidates, the party loses the possible benefit of the transfer of that vote as part of a candidate’s surplus. Ticket voting, introduced in 1983, constitutes the second distinctive element of the Australian implementation of STV and represents an open attempt on the part of the political parties to control the flow of their supporters’ preferences. It also may well be the only voting procedure in the world that requires parties to reveal publicly in advance of an election their full preference orderings of all their opponents, and it is therefore surprising that little systematic research into the effects of ticket voting has been seen to date (Fiji, which adopted ticket voting with an AV system in 1997, is the other major case we are aware of). The system may be summarized as follows (the provisions of the Commonwealth Electoral Act 1918 that deal with ticket voting are sections 168, 211, 211A, 269 and 272): Two or more Senate candidates have the right to request that they be grouped together on the printed ballot paper. Within 48 hours after the close of nominations, such a group of candidates may lodge with the Australian Electoral Commission a written statement setting out a preference ordering for all the candidates that they

wish their supporters to adopt. Such a preference ordering must show higher preferences for members of the group than for any other candidates. The preference ordering is known as the group’s group voting ticket. In lieu of such a statement, a group may lodge either two or three statements, giving preference orderings that, although different, must (1) show the same ordering for the members of the group and (2) show higher preferences for members of the group than for any other candidates. The right to lodge a voting ticket (or two or three voting tickets) may also be exercised by an ungrouped candidate who is an incumbent senator. The Senate ballot paper is divided into two parts, separated by a heavy black line. Grouped candidates are printed below the line in columns, Page 51 →whereas a square is printed above the line, over the column of any group of candidates that has lodged a group voting ticket. A similar square is printed above the line over an ungrouped incumbent senator who has lodged a voting ticket. A voter may vote either by numbering every candidate below the line, to indicate the order of his or her preference for them, or by marking a square above the line to indicate that he or she wishes to adopt as his or her preferences the group voting ticket lodged by the group. Where a voter adopts a group voting ticket as his or her vote, his or her ballot paper is deemed to have been marked in accordance with the preference ordering specified in the group voting ticket. Where a group has lodged two or three group voting tickets, there is still only one square printed above the line over that group. The voter is not able to choose which of the two or three group voting tickets he or she will adopt. Instead, in a case where two tickets were lodged, half of the votes cast by marking the square are deemed to follow one ticket, while the other half are deemed to follow the other; if the number of votes in question is odd, a determination is made by lot of which ticket the last ballot paper is deemed to follow. In a case where three tickets were lodged, one-third of the votes are deemed to follow each ticket, with a determination again being made by lot of the fate of the remaining ballots in cases where the number of votes in question is not divisible by three. The idea of ticket voting, oddly enough, did not come from the political parties themselves. The concept can be found in the writings of C.L. Dodgson (otherwise known as Lewis Carroll), whose contribution to social choice theory is becoming more and more widely recognized (McLean, McMillan, and Monroe 1996, 182–83; Hoag and Hallet 1926). It was brought to public attention in Australia through a submission to a parliamentary committee from Emeritus Professor A. L. Burns of the Australian National University, a specialist in international relations who had a side interest in electoral systems (Burns 1975).11 In a sense, however, the notion of ticket voting was in the air by 1983, even though it appears from an international perspective to be a most unusual practice, and from the outset it was readily accepted by the electorate. The reason for this is that the bulk of voters at Australian elections, long obliged to mark full preference orderings, faced with ballots of increasing size (see table 1), deprived until 1983 of ballot papers that showed candidates’ party affiliations, and in general either possessed of relatively strong Page 52 →party allegiances or dragged to the polls by compulsory voting and desirous of concluding the process as quickly as possible, had formed the habit of relying for guidance on how-to-vote cards. The adoption of the voting ticket scheme represented an institutionalization of this practice. There is a major difference in the simplicity of the task facing a person wishing to cast a ticket vote and that facing a person who wishes to determine his or her own preference ordering. The ticket voter has to mark but one square on the ballot, whereas the person wishing to depart from group tickets must rank-order every candidate. The net effect, not surprisingly, has been a steady increase in the use of ticket voting since its introduction. Table 2 sets out, for each state and territory, the percentage of Senate voters who have used the ticket-voting option. The second distinctive element of the ticket-voting scheme is the provision made for the lodging of multiple tickets and the handling of votes cast in accordance with such tickets. The approach adopted reflects a political compromise that the Labor government struck in 1983 to ensure that its enabling legislation would be passed by the Senate. It depended at the time on the support of the Australian Democrats, who viewed themselves as a center party and feared that their status in that regard might be compromised if they were required to lodge a single voting ticket that indicated a preference for one of the two major parties over the other. The lodging of multiple tickets by parties has been a regular event since 1983.

The effect of the introduction of ticket voting on the invalid vote was immediate and striking: Its first use cut the invalid vote rate by more than a half, and the rate of Senate invalidity at the 1996 elections was less than one-third of that at the notorious election of 1974. The introduction of ticket voting has, however, been strongly criticized by a number of participants in the political process in Australia who have seen it (correctly) as reinforcing party influence in the electoral process, in conflict with what are often seen as the original justifications for STV.12 For precisely that reason, there is not the slightest indication at the moment that any of the dominant parties in the federal Parliament have any interest in abandoning ticket voting. TABLE 2. Use of Ticket Voting (in percentages) Page 53 →

Senate STV and Senate AV The effects of the two Senate systems used in Australia have varied dramatically. As noted earlier in this paper, AV, in the multimember environment of Senate elections, reliably and predictably produced grossly lopsided results. As a result, the mechanics of Senate AV evolved scarcely at all. As can be seen from table 1, the introduction prior to the 1934 election of the requirement for full expression of preferences was followed by an increase in the invalid vote at the two succeeding elections; apart from that, there was no significant trend from 1919 to 1946 either in the invalid vote or in the average number of candidates per vacancy. Senate STV, on the other hand, is a system that has evolved over time. From 1949 to 1983, there was a relatively steady upward trend in the number of candidates per vacancy. Undoubtedly this can be partly explained by the far greater chance that minor parties have of achieving representation under STV than under Senate AV, a factor that can operate both by attracting new parties into the field and by holding out some hope of political survival to persons or factions who are tempted to break away from a major party. But other factors may also have been influential, most notably the decline in the real value of candidates’ deposits, which were left unchanged from 1918 to 1965.13 Coupled with the increases in the number of senators per state from 6 to 10 in 1949 and from 10 to 12 in 1984, the increased number of candidates per vacancy has produced a dramatic increase in the size of ballot papers. Table 3 sets out, for the elections following the dissolutions of the Senate in 1914, 1951, 1974, 1975, and 1983, the number of candidates on the ballot paper in each state. Since the introduction in 1983 of ticket voting, the operation of Senate STV has reached a new equilibrium. Although the tendency toward increasing numbers of Senate candidates has become more marked—at the four elections held in the 1990s, the average number of candidates per vacancy was greater than at any election prior to 1990—this no longer shows a tendency to increase the invalid vote but simply pushes more and more people toward the ticket-voting option, a trend with which the major parties are, and will continue to be, well satisfied. Although from time to time government supporters have suggested that the Senate voting system should be amended to limit the power of minority parties—and the proposed amendments have ranged from introducing single-member constituencies to requiring a party to poll a certain proportion of first preferences before being eligible for election—such amendments have to date not been favored by governments themselves and certainly would not be supported by the minority parties that hold the balance of power in the Senate. Page 54 → TABLE 3. Senate Candidate Numbers On the whole, no discernible trends in the functioning of Senate STV appear likely to give rise to a genuine need for amendment in the foreseeable future. It is also undoubtedly the case that the representation in the Senate of minor parties, which have held the balance of power continuously since mid-1981 (and, except for the period from 1976 to 1981, since 1962), has served to create a new and powerful role for the Senate as the voice of minority opinion in federal politics. The current political influence and legitimacy of the Senate is directly related to the use of a proportional electoral system, just as the mega-majoritarian system of multimember AV, which effectively ensured majority overrepresentation, was responsible for an appreciable loss of both political power of, and public confidence in, the Senate.

Senate STV and House of Representatives AV

In comparing the operation of House of Representatives AV since 1919 with the operation of Senate STV since 1948, it is notable that they have evolved Page 55 →in parallel. For the great bulk of the period since the introduction of Senate STV, it has proved difficult, if not impossible, for minor-party candidates to be elected to the House of Representatives, a pattern that has only very recently begun to change. This pattern probably reflects as much as anything a decision by candidates. Although the election of a minor-party candidate to the House of Representatives has always been the exception rather than the rule, prior to 1948 it was very much more difficult—in fact, virtually impossible—for such a candidate to be elected to the Senate. The introduction of STV for the Senate abruptly reversed the position, which undoubtedly led minor parties to focus their efforts to gain representation on the Senate and to run their strongest candidates there. A major similarity between the two systems has been the extent to which voters have been willing to adopt as their preference orderings the recommendations of the parties that they support. Although House of Representatives AV differs from Senate STV in not offering the voters a ticket-voting option, this does not reflect any decision in principle that ticket voting would be inappropriate for the House of Representatives but rather a view on the part of the dominant parties that it is not yet needed. The introduction of ticket voting has not been suggested for AV elections in Australia, but there is nothing in theory to preclude such a step being taken; indeed, such a procedure has been adopted for elections in the Pacific Island state of Fiji (Constitution Review Commission 1996, 707). On the whole, House of Representatives AV has evolved surprisingly little over the years when compared with Senate STV. To a certain extent this reflects the fact that, being an intrinsically simpler system, its capacity for evolution is more limited. But more importantly, its basic tendency has been to reinforce the centrist, bipolar party system that was effectively in place at the time of, and provided the political rationale for, its introduction. Senate STV, on the other hand, has tended to undermine that system, which has led the major political parties to opt for changes to the basic Senate STV model to maintain their hegemony over the electoral process.

Conclusion Australia has been prepared to engage in electoral experimentation to an unusual degree and has adopted electoral institutions that have tended to be advocated but not implemented elsewhere in the world. The system of preferential voting that enables voters to indicate the candidate ordering of their choice is a good example of this. As McLean (1987) has argued, most voters can normally rank at least some of the candidates presented to them against Page 56 →each other, and the more information a voting procedure can make use of, the better. This needs to be tempered somewhat: Voting systems that require electors to express preferences they may not in fact possess are susceptible to random effects such as the so-called donkey-vote phenomenon observed for federal elections in Australia, in which some electors simply number sequentially from 1 onward down the ballot paper. Optional preferential systems are clearly superior in this regard. Of course, if all electors express only one preference, then the resultant contest will produce political outcomes similar to a plurality contest, a state of affairs that has some serious theoretical limitations of its own. Nonetheless, it is hard to find anyone interested in electoral systems who will oppose the utility of allowing a rank ordering of candidates per se. Lijphart, who has tended to favor list PR, nonetheless argues that the facilities for transferring and accumulating votes are two innovations “that appear to work particularly well and that deserve to be recommended as models for electoral engineers elsewhere” (1994, 15), whereas McLean (1987) claims that some facility for preference ordering is one of three basic requirements of a good voting system (154). Other political scientists have argued for preferential systems such as AV to encourage “vote pooling” and centripetal electoral behavior, thus encouraging ethnic accommodation in deeply divided societies (Horowitz 1991; Reilly 1997c). Despite this bevy of heavyweight academic support, the motivation behind the introduction and continued use of preferential systems in Australia has been undeniably partisan. At the political level, the theoretical justifications for the systems in use are seldom even mentioned, let alone seriously debated. Associated with this has been a mind-set that sees the preferences of voters as resources to be marshaled by the parties rather than allocated by the voters themselves. Although the use of ticket voting for the Senate is the most palpable example of this, to focus exclusively on ticket voting invites a misunderstanding of the longer-term influences on the system. Ticket voting merely institutionalized, and thereby entrenched, the long-term trend for voters to look to parties for guidance on preferences and, apart from lowering the invalid vote, has probably had

minimal political impact. Only in Tasmania has a candidate under Senate STV been elected earlier in the scrutiny than a candidate above him on his party’s slate and then only in the early days of the system. The requirement for full preference numbering at Senate elections, which at first sight might seem to be a minor detail, has in effect led to the creation of an electoral system that, though normally classified as an STV system, functions in practice much more like a list system than do the Irish or Maltese implementations of STV. Page 57 →It seems fairly clear that the objectives put forward in the last century by Hare and the early enthusiasts for his ideas are a long way from having been achieved in their entirety in Australia. It says much for his system, however, that even when faced with the most vigorous attempts by political parties to use it to their best advantage, it is sufficiently robust to retain the support of the people and to provide them with legislatures the legitimacy of which they do not seriously doubt.

NOTES The views expressed in this paper are purely those of the authors and do not necessarily reflect the views of either the International Institute for Democracy and Electoral Assistance or the Australian Electoral Commission. 1. Under section 7 of the Constitution of the Commonwealth of Australia, each state must vote as one electorate at Senate elections “until the Parliament otherwise provides.” Since 1975, the Australian Capital Territory and the Northern Territory of Australia have also been represented in the Senate. 2. Prior to 1990, the scrutiny concluded when a candidate had received an absolute majority of votes, rather than continuing automatically until only two candidates remained. 3. It had previously been introduced, in optional preferential forms, by state Liberal parties in Western Australia (1907) and Victoria (1911). 4. Australian House of Representatives, Commonwealth Parliamentary Debates, second reading speech of Mr. P.M. Glynn on the Commonwealth Electoral Bill (No. 2) 1919, October 21, 1919, 13621. 5. Of the five deviant cases, one arose in Tasmania, a state whose electors have always tended to show a greater resistance to party discipline than their mainland counterparts, whereas the other four were associated with very close results. All five deviant cases, moreover, arose in the first decade of the system’s operation; thereafter it functioned exclusively as a winner-takes-all system. 6. A commentator on Australian federal politics once summed up the Senate as follows: “The Senate has been the more maligned of the two Houses, and in most cases the strictures passed on it have been fully justified. It became a refuge for superannuated political hacks and time servers of all parties. It consisted for many years of mostly elderly men, undistinguished in civil life, some of them with impressive military records, but with little else to commend them as statesmen. Its debates were dull, its contribution to the national political scheme negligible. . .” (Whitington 1956, 104). For a discussion of the revitalizing effect that the introduction of STV had on the Senate, see Uhr (1995). 7. There are numerous subtleties in the precise implementation of STV for Senate elections that we do not discuss here because they have not been perceived as having a significant political impact. These include the provision made for the exclusion of candidates in bulk, the combination of ballot papers with a common transfer value Page 58 →when transfers take place following an exclusion, the choice of ballot papers eligible for distribution as part of a surplus of a candidate not elected on first preferences, the precise formula used for calculating the transfer value of those ballot papers, and the use since 1983 of the Gregory fractional transfer mechanism rather than random sampling for the distribution of surpluses. For a discussion of the random sampling method that applied until 1983, see Fischer (1980). 8. The result in the 1974 Senate election in Queensland was even closer than in New South Wales: The National Party candidate won the last seat with a margin of only 4,911 votes (out of 1,032,460 formal first preferences). The Queensland outcome was not so widely perceived to be associated with invalid voting as the New South Wales result because the invalid vote of 65,941, although substantial in absolute terms, was significantly lower in percentage terms in Queensland than in other states and not so obviously correlated with the Labor vote. 9. How-to-vote cards are leaflets, typically distributed outside polling booths on election day by major

parties, that contain the parties’ recommendations to their supporters as to how they should mark their ballot papers. The distribution of how-to-vote cards on polling day is restricted at state elections in Tasmania and at territory elections in the ACT. 10. A survey conducted for the Australian Electoral Commission at the time of the 1996 federal election found that 59 percent of eligible voters agreed strongly or partly with the proposition that “the full preferential system is a good one,” whereas 35 percent disagreed strongly or partly. On the other hand, when presented with a choice between the full preferential system and OPV, 60 percent said they preferred OPV, whereas 34 percent said they preferred full preferential. See Australian Electoral Commission (1996). 11. Professor Burns also had more radical ideas that were not accepted. He was an advocate of circular ballot papers to randomize positional voting effects and even advocated—probably in jest—the printing of ballot papers on a single-sided Möbius strip of paper. He also argued strongly that preferential voting should be seen not only as a process of identifying a most-preferred candidate but also as one of vetoing a leastpreferred candidate and, more generally, that voters should be able to express preference orderings showing indifference between candidates or even showing a last preference but no first preference. 12. Before it was first used at an election, the ticket-voting scheme was subjected to a challenge on constitutional grounds: Mr. C. J. McKenzie, an ungrouped Senate candidate standing in Queensland, sought an injunction from the High Court of Australia to prevent the conduct of the 1984 Senate election in that state on the grounds that the relevant provisions of the Commonwealth Electoral Act 1918 discriminated against nonparty candidates and were unconstitutional because they involved voting for parties rather than candidates. The Court rejected the challenge and upheld the validity of the enabling provisions in the act. See McKenzie v. The Commonwealth (1985). 13. The deposit specified in the Commonwealth Electoral Act 1918 as originally enacted was $50. In 1965, this was increased to $100 for the House of Representatives and $200 for the Senate. Further amendments in 1983 increased the deposits to $250 for the House of Representatives and $500 for the Senate. In 1998 the deposits were further increased to $350 for the House of Representatives and $700 for the Senate.

Page 59 →

A Sophisticated Voter Model of Preferential Electoral Systems Neal G. Jesse Of all the theories in the field of political science, perhaps Duverger’s approaches the state of a true sociological law. Scholarly acceptance of Duverger’s dichotomous classification of electoral systems as being either plurality systems or proportional systems is widespread and prevalent. Both his law (plurality electoral laws lead to twoparty systems) and his hypothesis (proportional representation (PR)–based electoral laws lead to multiparty systems) have survived criticisms and reformulations. However, there is one group of empirical counterexamples for which Duverger’s law and hypothesis cannot account: legislatures with Single Transferable Vote (STV) Elections. A number of political scientists studying Irish politics (O’Leary 1979; Chubb 1982) as well as a scholar of party systems (Riker 1986) have come to the same conclusion: STV is an outlier to Duverger’s hypothesis. Duverger’s hypothesis predicts that legislatures employing STV (a PR-based electoral system) should be multiparty. The lower houses of Parliament in Ireland, Malta, Tasmania, and the Australian Senate have an average district magnitude (i.e., seats per district) in the postwar period of four, five, seven, and eight and an average number of legislative parties of 2.8, 2.4, 2.5, and 2.1, respectively. Of the four elected by STV, only the Republic of Ireland can be barely classified as confirming Duverger’s hypothesis. Given each legislature’s average district magnitude, theories based on this characteristic also fail to predict the small size of the STV party systems. In other words, the size of the legislative party systems under STV cannot be determined by their district magnitudes and their classification as proportional systems. The answer Page 60 →must lie in a more complicated model that takes into account the nuances of the STV electoral system. This chapter explores this fundamental anomaly to our understanding of electoral systems and party systems. The four party systems that employ STV all possess fewer parties than predicted by either Duverger’s hypothesis (Duverger 1954) or arguments based on district magnitude (Rae 1967; Sartori 1976; Taagepera and Shugart 1989). Moreover, although all four systems produce highly proportional results, the effective number of legislative parties is two in some instances (e.g., Malta and Tasmania) and two and one-half in others (e.g., Ireland and the Australian Senate). The prevailing literature on electoral systems and party systems, as well as the body of research more specifically aimed at analyzing STV, fails to provide an explanation for this apparent anomaly. In this chapter, I develop a model of sophisticated voting behavior to explain this anomaly. I add the voter into an institutional approach to explain party system outcomes. More specifically, I assume that the voter is rational and can react to information in a sophisticated manner. The inclusion of the voter as a variable leads to a dilemma. Although the party system is a national-level phenomenon, the behavior of the voters and the workings of the electoral system—the proposed determinates—are at the constituency level. The national party system results from the indirect summation of local vote totals. The crucial interaction between the parties, candidates, voters, and the electoral system occurs in these subnational electoral districts. Therefore, an explanation of the anomaly must rest in the constituencies. This ecological problem is not new to the study of electoral/party systems. It plagues the work of many scholars attempting to explain the link between electoral systems and national party systems, not least of all Duverger’s own law and hypothesis, as many authors point out (Rae 1967; Riker 1976, 1982, 1986; Sartori 1986). Duverger’s mechanical and psychological mechanisms explain how the decisions of individual voters in reaction to the plurality rule create an incentive for only two parties to compete. However, many authors demonstrate that Duverger’s logic is missing a mechanism to explain why it is always the same two parties in every district. This last condition is necessary for a two-party system. When a regional or third party is one of the two strongest parties in a constituency, Duverger’s law predicts that it should win seats. If this occurs in enough districts, it will expand the size of the legislature, for example, as in Canada and India (Rae 1967; Riker 1976). This chapter is divided into three sections. The first section provides a theoretical framework explaining the unexpectedly low number of parties in Page 61 →STV legislatures by introducing two concepts to fill Duverger’s

logical void. First, I propose a rather weak assumption that voters know the ranking of party strengths in their own constituencies even if they do not have complete certainty about the exact electoral strengths of the parties. This limited uncertainty allows each voter to anticipate the electoral competition in her district and to subsequently modify her behavior (i.e., vote in a sophisticated manner).1 This anticipation produces a district-level party system with fewer parties than complete uncertainty produces. Second, I introduce the concept of partial homogeneity of constituency preferences. I show that many districts possess a similar rank ordering of parties and are thus homogeneous in terms of party competition. Such duplication of party rankings across districts, when aggregated at the level of the national legislature, reduces the number of parties in the legislative system. Thus, the voters’ actions lead to a realization of their expectation of a two-party outcome; they behave in a strategic manner in their districts to optimize the district-level outcomes. The second section examines how limited uncertainty and partial homogeneity reduce the upper limit and the average size of the party system, bringing both much closer to two. Both assumptions are necessary conditions for a two-party system, as neither is sufficient alone. These assumptions predict results that mirror Duverger’s logic in two ways: through the reductive effects of limited uncertainty (similar to Duverger’s mechanical effect) and the sophisticated response by the voters abandoning smaller parties (i.e., the psychological effect). The third section illustrates that both assumptions are realistic. Data from the Republic of Ireland, Tasmania, and the Australian Senate lend support for the preceding model. The results of this analysis show that my extension of Duverger’s logic can explain the apparent empirical anomalies in preferential electoral systems. Moreover, the assumptions of limited uncertainty and partial homogeneity are credible and should be extendable to other electoral/party systems. Typically, the literature in the field of electoral studies points to district magnitude as the key contributor to these parameters (Rae 1967; Sartori 1976; Taagepera and Shugart 1989). Both Rae (1967) and Sartori (1976) illustrate how increasing the average district magnitude in a national legislature increases the number of parties. Rae also shows that this effect is more profound in proportional systems than in plurality systems. Taagepera and Shugart (1989) demonstrate that as district magnitude increases, the average size of the party system increases at a diminishing rate.2 Page 62 →

Explaining STV Systems: The Assumptions of Limited Uncertainty and Partial Homogeneity Two assumptions about party systems are necessary to understand why STV legislatures possess fewer parties than expected by the conventional literature and general intuition. The first is limited uncertainty, the assumption that the voters recognize the ranking of the parties in their district. This does not imply complete certainty (that voters know the exact electoral strength of each party) or complete uncertainty (that voters know nothing of the ranking or strengths of parties). This weak assumption presumes only that voters can identify the relative strengths of parties in their constituencies. Therefore, if the reality is such that for three parties, party A is stronger than party B, which in turn is stronger than party C (i.e., A > B > C), the voters are cognizant of this true ordering. This assumption is not unrealistic. Much research points out that voters do indeed perceive the relative strengths of the parties with some degree of accuracy, particularly in plurality elections (Brady and Johnson 1987; Bartels 1988; Abramson et al. 1992; Blais and Nadeau 1996). Voters in proportional systems may also have a fair indication as to the strengths of the national parties in their constituencies (Cohan, McKinlay, and Mughan 1975; Bowler and Farrell 1996). Literature examining split-ticket voting in proportional systems partially addresses this aspect of voter cognition, and such sophisticated behavior requires at least limited uncertainty (Marsh 1985b; Roberts 1988; Darcy and Marsh 1994; Bawn 1999). According to the second assumption, partial homogeneity, the ranking of the parties in any particular district is

connected to some larger distribution; in other words, each constituency is but one sample from the larger national distribution. The national ranking of the parties can be thought of as the norm. Most constituencies approximate this norm with little deviation. Therefore, a large number of districts possess a similar ranking of the parties. Some districts deviate greatly from the norm, but their number is fewer. For example, strong regional parties should produce rankings in some constituencies that deviate substantially from the national norm. If strong regional parties exist, they may increase the size of the national party system by dominating the results in a few select districts3 However, such constituencies should be rare (that is why these parties are regional and not national) when compared with the number of constituencies in which the national parties dominate the competition.4 Page 63 →This assumption of partial homogeneity is a weak but important one. It restricts the expected party system size of the national legislature to something close to the expected size of the party system in the districts. Exactly how much it reduces the average size of the party system is the subject of the next section.

Explaining Two-Party Systems As previously mentioned, partial homogeneity assumes that the district party orderings (POd) are drawn from the national party ordering (POn). Limited uncertainty presumes that voters are aware of the party orderings in their districts. The interaction of the two creates electoral competition within each district that is likely to be similar to that in other districts. Also, voters are aware that the competition in their district is one small component of the overall national competition. Thus, each voter is somewhat cognizant of the national party ordering and her local district party ordering. If voters have some certainty that POd is the reality in their district, they expect that the seat allocation within the district will approximate POd. Returning to the previous example, if POd: A > B > C > D and voters assume that the social choice5 in their district, i, mirrors POd, then the seat distribution will be such that A > B > C > D. In other words, the transitivity of the results will not differ from the transitivity of the expectations. Operationalizing this concept, each party obtains more seats than any party expected to perform more weakly than it performs. Further, all of the seats are not divided equally among a number of parties equal to the district magnitude when DM > 2.6 There are two direct consequences of this logic: Limited uncertainty mechanically limits the number of parties that win seats and psychologically produces incentives for voters to behave in a sophisticated manner. The mechanical limit occurs because of the restriction on the distribution of seats, and the psychological effect follows from the existence of a mechanical effect as voters desert weak parties and/or candidates. Given these two assumptions, every possible outcome in the district needs to be considered to calculate the average size of the district-level party system. The number of possible outcomes is dependent on two separate factors: the district magnitude and the number of parties contesting seats. For example, in a two-member district with two competing parties, four different outcomes are possible: One party wins both seats, the other party wins both seats, or two parties split evenly the seats (which can occur two ways).7 The Page 64 →effective size of the district party system is equal to one in the first two instances, whereas in the third and fourth instances it is equal to two. With the probability of each outcome occurring being equally likely and the assumption that the joint probabilities sum to one, the average district size can be calculated.8 Using this method, the resulting average district-level party system size is equal to 1.50 parties.9 Table 1 displays the average number of district parties for different district magnitudes for both complete and limited uncertainty. The former does not place the previously discussed restrictions on the possible results in a district (i.e., the results do not have to conform to the POd. Table 1 shows the reducing effect of limited uncertainty. The size of the district party system is almost one party smaller (excluding DM = 2) than the party system with complete uncertainty. Even at DM = 6, the district party system resembles a two-party system more than a multiparty system. Moreover, three parties emerge only when the district magnitude is greater than eight. The combination of limited uncertainty and partial homogeneity produces incentives for voters to behave in a

sophisticated manner. According to Duverger’s logic, voters should abandon parties without a reasonable chance of obtaining seats (Duverger 1954; Cox 1997). For example, take a three-member district (DM = 3) and add a fifth party to the above POd such that A > . . . > D > E. Supporters of party E realize that it will not win a seat in the district and furthermore that it is unlikely to obtain a large share of the seats in the national legislature (by partial homogeneity). It thus makes sense for these voters to look for strategic options.10 But what strategic options exist for the individual voter under STV? The ability of voters to list preferences for candidates and the resulting transfer of the ballot among candidates allow each voter an opportunity to vote strategically. One option is the ability of the voter to list her preferences on the ballot in such a way that anticipates vote transfers from candidate to candidate. Strategic11 behavior entails a voter listing her preferences in such a manner as to accumulate transfer votes on a candidate who has both a reasonable chance of reaching a quota of votes and is preferable to other candidates also having a reasonable chance of being elected. Suppose a supporter of party E believes that by transferring her vote to party C, a candidate from C will obtain a quota and that she prefers party C to the other candidates/parties with a chance of winning. This voter would order her preferences to transfer her ballot to C. This does not differ from sincere voting if all of E’s supporters honestly prefer party C to all of the remaining parties. However, if these same voters actually prefer party D to party C and if D does not have a reasonable chance (transfers from supporters of E will not give D a quota), then transferring to a party other than D is an example of sophisticated behavior.12 The appendix to this chapter contains an example of sophisticated voting based on this scenario. TABLE 1. Average Effective Number of District Parties with Limited Uncertainty Page 65 →Sophisticated voting occurs during the intersection of two events: an opportunity and a motivation.13 The distribution of preferences must be such that a sophisticated listing of candidates by one or more voters changes the social choice from what it would be if every voter listed her preferences in a naive manner. Once a voter has an opportunity, she must also have a motivation to change the social choice. If a change would create only a less favorable or equivocal social choice, then motivation does not exist. The focus of this work is to ask why STV party systems are smaller than predicted. Thus, the primary goal is to identify when supporters of third parties are able to behave in a sophisticated manner to obtain seats for those third parties. In the context of STV, an opportunity exists for third parties if the sum of first preferences for all of the third parties is equal to or greater than the quota.14 Opportunity condition for supporters of third parties: exists if ∑(V3 . . . Vn) > Qd, where i is the estimated district size for party j and for i, 1 = largest party, 2 = second largest, and so on to the smallest party, n; V = ∑Vi, i = 1 to n. Qd = the quota in the voter’s district expressed as a share of the total valid vote. However, this is not sufficient to produce strategic behavior. There must also be a motivation. A number of supporters of parties 3 through n equal to or greater than the quota must also prefer at least one of the same parties to both of the two largest parties. Motivation condition for supporters of third parties: exists if for some party j, Vpj > Q, where Vpj V3 . . .Vn such that for each voter in Vpj party j is Page 66 →preferred to party k, where k 1 . . . n and j ≠ k and j ≠ n. The motivation condition is crucial. Supporters of small parties need to have some expectation that they can concentrate their votes through the transfer process on a candidate or party that can win. For opportunity and motivation to intersect, a third party of size greater than the smallest party must be preferred to the two large parties by enough supporters of the eliminated smaller parties to ensure that it is not itself eliminated. Thus, supporters of small parties need to focus their transfers on a party large enough to avoid elimination (i.e., by falling into last place at any count) and that will continue to receive transfers. If such conditions do not exist, the only possible sophisticated voting is for supporters of third parties to transfer their votes to one of the two largest parties, hoping to support the election of a candidate from the more preferred party. This case arises when the combined votes of the third parties are enough to determine which of the two

large parties obtains the final seat.15 The previous example is just such a case. With only a single seat left, the condition is similar to a plurality election in a single-member district. With more than a single seat left, the condition is similar to an n-seat plurality (with n equal to the number of remaining seats) where the quota is the post, as it were, by which candidates are elected. Opportunity condition for selecting one of the two largest parties: exists if ∑(V3 . . . Vn) > V1– V2 Motivation condition for selecting one of the two largest parties: exists if for party i, where j {1, 2} , Vpj > |V1 – V2| and where Vpj V3 . . . Vn such that for each voter in Vpj, party j is preferred to party k, where k {1,2} and j ≠ k. To recapitulate the logic of this section, voters know the rankings of the parties in their district (i.e., by limited uncertainty) even if they do not know the exact electoral strengths. They also know that the ranking of parties is unlikely to deviate from that of other districts and the national norm. However, if their district is extremely deviant, they are aware of this fact. Knowing this, each voter checks whether any sophisticated behavior is possible or desirable. Typically, a third-party supporter behaves in a sophisticated manner if she sees that by transferring her vote to a strong third party and not transferring it to more preferred but smaller parties, she may be able to better the distribution of seats in her district. If such a chance is not possible, she checks to see whether a strategic vote for the more preferred of the two larger parties is available. If no opportunity and no motivation exist to behave in a sophisticated manner, the voter just marks off her true preferences on the ballot. The concepts in this section are testable. The more complete the homogeneity of the district party orderings, the more likely the district party systems Page 67 →resemble the national party system (and vice versa). Thus, if a two-party system already exists and it is duplicated in the greater majority of the districts, each district will approximate a two-party system. This limits the third-party opportunities and preserves the prevailing party system. When opportunities for third parties are fewer than large-party opportunities, the party system should contain fewer parties. When both opportunities are either high or low, the two are at odds, and an intermediate party system between two-and multiparty may be the result.16

Investigating Homogeneity and Opportunities Two of the four concepts that I have developed thus far are relatively weak assumptions, and two are a bit stronger. The former are limited uncertainty and motivation. It is not unrealistic that voters can gauge the relative rankings of the parties in their districts. Also, that some voters disagree with the seat distribution under the naive equilibrium seems quite plausible. It is the two latter concepts, partial homogeneity and opportunity, that need some investigation to show that they are plausible. Partial homogeneity seems on its face to be a bit incredible. It assumes neither that districts are uniform nor that they are radically divergent. The uniformity condition is quite unrealistic as party strength differs by region in even the most homogeneous nations. Typically, party organization or party origin determines these geographic biases (Lipset and Rokkan 1967; Katz and Mair 1994). On the other hand, it is unrealistic to assume that the party competition in the districts is completely independent of the national party competition. Parties are strong at the national level because they are strong in a large number of districts. It is impossible for a party to be competitive at the national level without being competitive in a large share of districts. The true level of homogeneity must lie somewhere in between these two extremes of full homogeneity and complete heterogeneity, that is, it must be partial. Identifying Partial Homogeneity To show that partial homogeneity is indeed realistic, I examine postwar elections in Ireland, Tasmania, and the Australian Senate. My gauge of the strength of each party is its share of the vote in each district. I treat all fourth or smaller parties as a single party for the purposes of ordering them. This usually will have no effect on the ordering and in any case overstates the Page 68 →ability of these parties to obtain a quota. This leads to more conservative estimates of homogeneity and opportunity as any overweighting, as it were, of the fourth parties is

not likely to be duplicated in most districts and thus increases both heterogeneity and the number of third-party opportunities (because they may already have a quota).17 The next step is to estimate the degree to which each district ordering is replicated in other districts. The more districts with similar orderings, the more homogeneity. Moreover, the greater the number of districts with the same ordering as the national ordering, the more homogeneity. My methodology is as follows. Table 2 gives 13 possible party orderings (PO) of four parties (A, B, C, D).18 For purposes of cross-district comparisons, I assign a label to each party in the legislature. I label Party A the strongest party in the electorate; the second strongest, party B; and the third strongest, party C. Party D applies to the sum of the following two groups: the remaining parties and the independent/other votes. As an example, in the 1992 Irish election, Fianna Fáil was the strongest party (39.1 percent of the national vote); Fine Gael, the second strongest (24.5 percent); Labour, the third (19.2 percent); and all others, the fourth. If these results were mimicked in a district, the PO in that district is A > B > C > D. One further comment on methodology is necessary. By using all non-first-three-party votes as part of the calculation for party D, I give proponents of heterogeneity their due and provide a tougher straw man for me to knock down. For example, if no fourth party possesses more votes than the third party (party C) but the sum of all the fourth party and independent votes are greater than the votes for party C, I rank D > C. Such calculations bias my estimates of homogeneity toward heterogeneous results and away from my hypothesis. This is productive in that I now cannot overstate my findings. It also means that if the evidence does point toward partial homogeneity, than surely the heterogeneous hypothesis is void. TABLE 2. List of Party Orderings Occurring in at Least One of the Three Legislatures Page 69 →Table 3 displays the homogeneity of districts across elections for three different legislatures: the Irish Dáil, the Tasmanian House of Assembly, and the Australian Senate. Note that 51 percent of the Irish districts, 64 percent of the Australian districts, and a stunning 85 percent of the Tasmanian districts all possess a single, similar ordering.19 This ordering (A > B > C > D) is congruent with the national ordering, so it is quite apparent that complete heterogeneity does not exist. In fact, over 75 percent of all districts in each country display only two different party orderings. Variation across possible POs is amazingly small. Table 4 presents the homogeneity of districts grouped by district magnitude. For the most part, district magnitude does not appear to have any predictable effect. The reason for this may lie in the fact that the distribution of party strengths is endogenous and not directly shaped by the district magnitude (although how these strengths are translated into seats is shaped by the district magnitude). Variation does occur within a given legislature, for example, the difference between district magnitudes of 5 and 6 versus 7 and 10 in the Australian Senate, but the reason for this variation is not clear. Tables 3 and 4 hide some interesting differences between the legislatures. Tasmanian districts display almost no variation. In 7 of the 12 elections, every single district possessed a similar party ordering. The Australian Senate is similar in that only three different party orderings occurred with a frequency greater than 2 percent. What is common to both legislatures is that movement in the rankings between districts is often the result of the top two parties shifting positions. This demonstrates the weakness of third parties in both of these legislatures. In Ireland, there is a bit more variation. Six different POs occurred with a frequency greater than 2 percent, and 11 different POs have occurred at least once. The almost 14 percent of cases in which the third party or another party ranked in the top two in a district (POs 3, 4, 7, 8, 10, 12, and 13) illustrates the electoral strength of Labour and some of the regional parties during the 1950s and 1960s. Labour placed in the top two in 50 districts, and other parties placed in the top two in 22 districts.20 Most of Labour’s top rankings occurred in the 1992 election, in which it had its best-ever electoral showing. The minor party surge appears in two different time periods. The first was during the 1950s and early 1960s, when a number of small parties captured seats in mainly rural districts. The second occurred in the 1987 election, when the Progressive Democrats split from Fianna Fáil and ran a very

strong campaign. Page 70 → TABLE 3. Distribution of District Party Ordering Page 71 → TABLE 4. Distribution of District Party Orderings by District Magnitude Page 72 →Cross-district differences in party orderings in Ireland are almost three times more likely to arise from the bottom two parties changing positions than from the top two. Regarding this trend, the Irish Parliament does not have some of the same characteristics as the other two legislatures. It possesses both a larger number of seats and a larger number of districts. Most of this switching of the two bottom parties can be explained by Labour’s inability to run candidates in every district in most elections. This was particularly true throughout the period from 1954 to 1977. Strong competition from other minor parties limited Labour’s impact in the countryside in the 1950s and 1960s, whereas the domination of the two large parties in the 1970s kept Labour out of the rural districts. The effect of the apparent partial homogeneity in tables 3 and 4 is to reduce the party system to a size smaller than expected. Taking the numbers presented in both tables, one can construct the average number of seats for each party.21 Given the average distribution of party orderings in the districts, the average Irish legislature would contain only 2.35 parties. The most homogeneous election (1965) had a lower figure of 2.28 parties Identifying Opportunity Opportunity also needs some investigation to prove its merit. Much literature on STV states that the opportunities for sophisticated voting should be limited at best (Bartholdi and Orlin 1991; Bowler and Farrell 1991b, 1996). Bartholdi and Orlin (1991) go so far as to say that sophisticated voting requires informational and computational abilities that the average voter just does not possess. Or in other words, even if the voters did want to vote strategically, they are not capable of doing so. Also, it is argued that the transfer process should be able to move votes to preferred parties so that misrepresenting preferences is unnecessary and/or counterproductive. Therefore, voters should be presented with few opportunities and should lack motivation anyway. The final step is to identify the rate at which voters are offered opportunities to act in a sophisticated manner. Earlier I defined two different types of opportunity: the ability to elect candidates from one large party over the other Page 73 →(i.e., large-party opportunity) and the ability to pool transfers to elect third-party candidates (third-party opportunity). Here I show that a large-party opportunity is present in almost two-thirds of all districts and that it is more likely to occur than a third-party opportunity. This result indicates that sophisticated behavior is more likely to focus on electing large parties than third parties. This should act as a limiting factor on the number of legislative parties (at the same time it should raise the number of parties above that which would occur with complete homogeneity). Table 5 displays the type and rate of opportunity in STV legislatures. It also indicates the percent of the districts in which both opportunities exist and the percent of districts with at least one opportunity. A number of similarities appear among the legislatures. First, the rate of large-party opportunities is greater than the number of third-party opportunities. The range of this rate runs from 1.3 in Ireland to 3 in Tasmania. Second, almost all districts with a third-party opportunity also have a large-party opportunity as well, but not the reverse. Third, the overall rate of opportunities and the ratio of large-party to third-party opportunities display a negative relationship. As the number of opportunities declines, the greater the chance that the voter can make sophisticated choices only between the two large parties. Tasmania illustrates this point, as it has the lowest incidence of opportunity (46.67 percent) but the highest ratio between the two types of opportunity (3 to 1). Last, and perhaps most important, an opportunity exists for sophisticated voting in a range from one-half to three-fourths of all districts. These results suggest that voters have occasion to behave in a sophisticated manner. TABLE 5. Type and Rate of Opportunities in Ireland, Tasmania, and Australian Senate Page 74 →Table 6 displays the results of an ordinary least-squares regression analysis. The effective number of legislative parties after each election is the dependent variable. The independent variables include the rate of

large-party opportunities, third-party opportunities, the incidence rate of PO #1, and the average district magnitude. Two of the independent variables have the expected sign, third-party opportunities, and rate of PO #1. Note that the average district magnitude is insignificant. It seems that having any strategic choice, even if just between the two large parties, increases the size of the party system (although third-party opportunities have a greater effect than large-party opportunities). Thus, the size of the party system increases with heterogeneity and strategic opportunities and decreases with homogeneity and a lack of opportunities.

Conclusion In this chapter, I develop an explanation for why STV legislatures possess fewer parties than the prevailing paradigm predicts. My argument rests on two main assumptions: (1) Voters can identify the ranking of the each local party (limited uncertainty), and (2) the ordering of the parties at the district level is not random but rather a bit predictable (partial homogeneity). The first assumption is credible, and I give evidence that the latter is true. With these assumptions, I posit that voters behave in a sophisticated manner to achieve the best possible social outcome in their districts. I show that opportunities exist for voters (with motivation) to act in a sophisticated manner. Moreover, voters are presented with greater opportunities to transfer votes toward the two large parties than toward smaller parties. TABLE 6. Ordinary Least-Squares Regression Analysis of the Effective Number of Legislative Parties (Ns) Page 75 →The prevalence of partial homogeneity reduces both the upper limit and the average number of parties in each district. When combined with more large-party opportunities than third-party opportunities, these two factors provide strategic incentives for voters to concentrate their transfers on the large parties. Such optimizing, rational behavior leads to a reduction in the legislative party system size. Thus, my model adds to Duverger’s model of the connection between electoral systems and party systems. Through the addition of two assumptions, it explains the apparent empirical anomalies produced by one type of preferential voting. Moreover, the model should be generalizable to other preferential systems. APPENDIX: AN EXAMPLE OF SOPHISTICATED VOTING If the voter in the scenario presented on page 65 lists her preferences sincerely, she marks them in this order: E > D > C > B > A. Assume that this voter is a member of the following preference profile (34 votes; DM = 4; quota = 7 votes; parties A and B have two candidates each) and that she votes sincerely: 13 voters: A > B > C > D > E 12 voters: B > C > A > D > E 4 voters: C > B > D > A > E 3 voters: D > C > E > B > A 2 voters: E > D > C > B > A (including the voter in our discussion) Parties A and B win one seat apiece, leaving their remaining candidates with six and five votes, respectively. Party E’s candidate is eliminated because he has the fewest votes, and the two votes transfer to the candidate from Party D. Party C’s candidate is eliminated, and the four votes transfer to Party B. Party B wins a second seat, leaving two surplus votes to be transferred. One vote transfers to Party A and the other goes to Party D, pushing Party A over the quota.22 The final social choice is two seats for Party A and two seats for Party B. However, if the voter transfers her vote in a sophisticated manner, she can positively alter the final social choice. She knows that parties E and D likely do not have enough votes between them to win a seat (i.e., their first

preferences are not likely to sum to a quota). She estimates that if all of the minor parties (C, D, and E) combine their votes on a single candidate, it is possible to reach a quota and gain a seat. The most likely target is then the party with the most first preferences. From her limited uncertainty, the voter believes that Party C is the party to support, and she attempts to keep its candidate from being eliminated. Page 76 →If this voter sophisticatedly marks her ballot such that E > C > D > B > A, she can change the outcome of the first example. After Party A and Party B each win a seat, E is eliminated; one vote (from the voter in our discussion) transfers to Party C, and one vote (from the other E supporter who is uninformed) goes to Party D. Party D’s candidate is eliminated, and the four votes transfer to Party C. Party C wins a seat, and its remaining surplus elects the second candidate from Party B. The final outcome is one seat for Party A, two seats for Party B, and the fourth seat to Party C.

NOTES The author would like to acknowledge the helpful advice of George Tsebelis, from which this chapter benefited. 1. For the purpose of clarity, I use female pronouns to refer to voters and male pronouns to refer to party candidates. 2. This nonlinear result suggests that something other than just district magnitude is acting on the number of parties. Later in this chapter I propose how the possible distribution of seats within a district inherently limits the number of parties. 3. Of course, this need not be the case if the regional parties fare quite poorly. The Scottish Nationalist Party (SNP) typically has greater electoral support in Scottish districts than the Conservatives but rarely more support than Labour. Thus, it gains relatively few seats in the House of Commons. Poor showings can be the result of either the electoral rule (e.g., plurality in the example just given) or just the lack of strong electoral support for a third party (e.g., SNP in the south of England). 4. It can be the case that the entire national party system is just the aggregation of strong regional parties. One can think of the French Third Republic or the United States during Reconstruction as examples of such a party system. However, such party systems are more rare than the type I outline in text. 5. Social choice refers to the collective outcome determined by the aggregation of the individual actions—in this case, the allocation of seats to the parties based on the choices of voters and the transfer process. 6. This rules out complete equality of results. For example, in a three-member district (DM = 3) with three parties such that POd : A > B > C, possible outcomes are A = 3 or A = 2 and B = 1 but not A = B = C = 1, A = 2 and C = 1, or any other similar result. In even-membered districts greater than DM = 2, the top parties may divide the seats equally, for example, A = 2, B = 2, and C = 0. Two-member districts are exempted from the restriction mentioned in the text as a division of seats always results in a number of parties equal to DM gaining seats. It is assumed that in DM = 2 districts, a third party received no seats; thus, the result A = 1, B = 1, C = 0 actually occurs. Complete equality of results assumes that all of the parties are equally strong and thus there is radical uncertainty of the relative party strengths. 7. The calculations can also be thought of as binomial distributions in which a single party wins either 2, 1, or 0 seats. The number of ways each event can occur is calculated as n!/(n – r)!r!, where n = the district magnitude and r = the number of parties winning seats. With replacement, the formula is k[n!/(n – r)! r !], where k = the number of parties competing. Page 77 →8. Assuming equal probability is reasonable. Starting with the contrary assumption that events are not equally likely, a short calculation shows that the most likely probability for each event is equal to every other event. Allowing each probability to fluctuate from its maximum (pi = 1) to its minimum (pi = 0), with ∑pi = 1, means finding the joint density function (JDF) of the probabilities. For the continuous function of two variables, f(x, y), in the two-dimensional set A, the JDF = (p((X, Y)) A) = ∫∫f(x, y)dydx. Inserting the binomial distribution, the equation reads JDF = ∫xf(x)dx, substituting p, JDF = ∫pdp + c, which is equal to (1/2) p2 ∫c. Inserting the maximum and minimum values, 1/2(1)2—1/2(0)2 = 1/2 = p. The multinomial distribution has the same outcome, p = 1/n. Thus, the most likely distribution is one in which each possibility is equally likely. The JDF of the example in the test is then calculated as ∑(pi × Ndi)/n (Ndi

= effective number of parties for seat distribution, i, which works out to (1/4)(1) + (1/4)(1) + (l/2)(2) = 1.50. 9. Note that these calculations need be calculated with a number of parties equal to the district magnitude. Adding fewer parties reduces the resulting figure, and any greater number of parties does not change the resulting figure. 10. Duverger (1954) points to just such a pattern of voter strategy in his discussion of Britain at the turn of the century. British voters realized that the Liberal Party was slipping nationally into third place behind the Conservatives and the nascent Labour Party. Voters then abandoned the Liberals at the district level owing to this perceived national weakness. 11. I am using the words strategic and sophisticated synonymously as applied to voting or behavior. Each defines when a voter marks preferences for candidates in a manner that differs from her true preference ordering—in other words, when she marks on her ballot one or more lower preferences instead of one or more higher preferences. 12. This is especially true if transfers to party D lead to an unfavorable social choice, for example, the elimination of the candidate from party C and the seat being won by a less preferred party. 13. The logic is that of Downs (1957), but perhaps the best, and most recent, work is that of Blais and Nadeau (1996). 14. In the analysis that follows in text, for simplification I assume that voters with similar preference orderings behave in similar manners. Voters with like orderings can be thought of a group of voters who will transfer their votes in blocs. This produces groups of pivotal voting blocs rather than single pivotal voters. 15. The third parties do not necessarily need enough votes to push a large party over the quota. When only two candidates are left, whoever has the largest number of votes will obtain the final seat, regardless of whether the quota is reached or not. Thus, the third-party votes only need be large enough to push one party’s candidate over all others. 16. The predicted relationship between the three independent variables, homogeneity, large-party opportunity, and third-party opportunity, and the dependent variable, effective number of parties, is negative, negative, and positive, respectively. 17. This would run counter to my argument, as the increase in third-party opportunities should produce a higher number of parties. 18. There are, of course, 4! or 24 possible orderings of the four parties without replacement (i.e., the same party does not occupy two positions). I include only the 13 that actually occurred in the data set. Page 78 →19. Similar to others within the same legislative election, not similar to districts in elections to the other two legislatures of course. 20. Labour actually was the strongest party in three districts, and the other parties led six different districts. 21. The actual calculus is a bit complicated. First, you find the incidence rate of each combination of PO and DM. You also find the number of seats each party (by rank ordering, first largest, second largest, and so on) would win in each district given the binomial distribution of possible outcomes (mentioned earlier). You take the ratio of the total seats per party divided by the total number of seats (with the same DM) to find what percent of those seats each party is most likely to win. You then multiply the appropriate party percentage by PO incidence rate. The sum of these products is each party’s share of the total seats. Then use the calculus for finding the effective number of parties (Taagepera and Shugart 1989) to find the expected size of the legislature. As an example, say that the occurrence of PO #1 (i.e., A > B > C > D) in districts with three seats accounts for 5 percent of all of the districts. Given the binomial distribution in districts where DM = 3, party A should receive 2 seats; party B, 0.75 seats; and party C, 0.25 seats; or 67, 25, and 8.3 percent, respectively. These figures are multiplied by 0.05 (incidence rate), producing 0.0335, 0.0125, and 0.00415, respectively, which is each party’s share of the total seats derived for these districts. The sum of these figures provides a national total. 22. The process of determining where surpluses are transferred is quite complicated. To summarize, of the nine ballots Party B possessed, two-thirds listed Party A as the next choice, while one-third listed Party D. With only two votes remaining in the surplus, Party A obtains −1.34 (0.67 × 2) and Party D obtains −0.67 (0.33 × 2). 1 rounded these off to one vote apiece. Without rounding, the advantage for Party A is even more obvious than in the example in the text.

Page 79 →

STV in Ireland Page 80 →

Page 81 →

The (Relatively) Victorious Incumbent under PR-STV: Legislative Turnover in Ireland and Malta Michael Gallagher Many criteria may be used to evaluate electoral systems, from the familiar traditional yardsticks such as accuracy of representation and government stability to social choice criteria such as consistency or monotonicity. In this chapter we shall look at one relatively neglected output, namely, turnover and rates of reelection of incumbents, and examine the record of PR-STV on this criterion in the two countries that employ it to elect their national parliament. As with several other criteria used to judge electoral systems, turnover is usually seen as neither a goal to be maximized nor an evil to be avoided. Turnover is a commodity of which one can have too much or too little; excessive rates of turnover will mean a loss of experience and expertise in the legislature, maybe weakening the institution, whereas too little may lead to frustration, ossification, and an unresponsive legislature. We are concerned here not with a normative evaluation of PR-STV but with an examination of how it performs on this dimension. Of course, turnover is not determined solely by the electoral system. Other factors, most obviously electoral volatility, can also be expected to have a major impact on the reelection chances of incumbents, and to a limited extent the data will allow us to test the impact of some of these factors for the cases we cover. But whatever other factors play a role, there are many reasons to expect electoral systems to have a strong impact on the extent of turnover. In this chapter we shall examine the significance of the various threats to an incumbent’s position. How vulnerable to defeat are incumbents in the first Page 82 →place, and when defeat comes, is it more likely to result from interparty or intraparty competition? What are the main sources of turnover among members of a parliament? Conversely, is a nonincumbent wanting to enter a parliament best advised to target a fellow party incumbent or an opposing party incumbent for defeat or just wait around until an incumbent calls it a day or dies?

Turnover and Reelection Rates Before looking in more detail at the likely impact of electoral systems on turnover, we should clarify what is meant by the term turnover. Turnover could be construed as relating to (in practice, it is the obverse of) a number of different phenomena: 1. The proportion of deputies1 who, having been elected at one election, are reelected at the next election 2. The proportion of deputies who, having been incumbents at the end of the lifetime of a parliament, are reelected at the ensuing election 3. The proportion of outgoing deputies who, having been incumbents at the end of the lifetime of a parliament and having contested the ensuing election, are reelected 4. Unity minus the proportion of deputies who are first-term members These four options conceptualize turnover in different ways and produce different figures for exactly the same set of data (see Table 1). No doubt a case could be made for any of the four as the best way of measuring reelection rates; the view taken in this chapter, as argued later, will be that the third measure is the most appropriate for our purposes. But what should be beyond dispute is that these four interpretations of turnover are measuring quite different things and that comparisons between countries are meaningless unless the same measure is used in each case. Of the three main comparative studies of turnover, two (Katz 1986, 97; Matland and Studlar 1995, 5) use the first measure. The other, despite its aspirations as a comparative volume, appears to have allowed individual

contributors to use whichever of the second or third measures they preferred, thus ruling out any possibility of making systematic comparisons across countries (Somit 1994, 18n. 2).2 The fourth option seems to have been used only in some U.S. research (see Polsby 1968, 146, where turnover is understood in terms of the number of first-term members elected). Measuring turnover in terms not of rates per election but rates per year (the approach of Hibbing 1988) does not solve the problem; apart from the fact that this will reflect the frequency of elections as much as what happens at individual elections, it merely raises the question of which of the four measures outlined previously should be used as the basis for this calculation. Page 83 → TABLE 1. Reelection Rate, According to Four Different Measures, for Hypothetical Data The view taken in this chapter is that the third measure is the most useful, certainly in the context of a discussion of the effects of electoral systems. It is the only measure that reports accurately the rate at which those incumbents who seek reelection succeed or fail in that aim. It does not conflate the three completely different reasons why those elected at election E may not be reelected at election E + 1: 1. Page 84 →Death or resignation during the lifetime of parliament E, resulting in replacement by another MP (member of parliament) 2. Not contesting election E + 1 (owing to voluntary retirement, failure to secure reselection as a party candidate, party term limits to ensure rotation of parliamentarians, or legal term limits) 3. Defeat at election E + 1 If our interest were in the functioning or “institutionalization” of legislatures, then this conflation might not matter; we might be concerned only with the actual rate of change in the personnel in parliament, no matter what the cause. But given that our interest here is in the impact of electoral systems, we should not confuse the different reasons for turnover. A deputy who dies three months after election E will not be reelected at election E + 1, but this is not a consequence of the electoral system employed at election E + 1.

Turnover and Electoral Systems Electoral systems will probably have an impact on turnover rates, but what kind of impact? Will PR systems lead to more or to less turnover? And among PR systems, will those allowing voters to express preferences for individual candidates lead to more or to less turnover than those based on nonpreferential lists? The fullest comparative study of turnover tests the hypothesis that turnover will be lowest when individual candidates play the greatest role and highest when the role of the party is greatest (Matland and Studlar 1995, 7–8). The rationale for this is that when voters can express preferences for individual candidates, candidates build up a “personal vote,” as it were, and this protects incumbents when their party’s fortunes take a downturn and also makes parties reluctant to deselect them as candidates. Thus, the authors’ hypothesis is that turnover will be highest in nonpreferential list PR systems, followed in descending order by preferential list PR systems, PR-STV, SNTV (single nontransferable vote), and finally SMP (single-member plurality) systems. The conclusion drawn (15) from testing this model on the data is that the electoral system does not have a statistically significant effect on turnover. However, if the model is plausible, its operationalization is questionable. In particular, it is hard to sec why preferential list PR systems and PR-STV are placed nearer to the party-dominated end of the spectrum and SMP is placed at the candidate-dominated end. Arguably, SMP and nonpreferential list PR systems should be grouped together, because in each, party and candidate(s) Page 85 →are inseparable as far as the voter is concerned: a voter who wants to vote for the party must accept the candidate(s) put forward by the party, in the order in which the party presents them, whereas a voter who wishes to reject a (or the) specific candidate presented by the party cannot do so without also rejecting the party (Carey and Shugart 1995). Under the other three systems (preferential list PR systems, PR-STV, and SNTV), in contrast, voters can make a choice within parties or without regard to parties. They can remain loyal to their party and yet vote against its incumbents (by giving their preferential vote to a nonincumbent). Incumbents, under such systems, are doubly vulnerable; they are at risk of

being unseated both by a candidate from another party and by one of their own running mates (a running mate is another candidate of the same party, standing on the same ticket). In contrast, under SMP and nonpreferential list PR systems, only a swing against the party can lead to the electoral defeat of an incumbent. Thus, an equally plausible hypothesis would be one that envisages turnover being highest in preferential list PR, STV, and SNTV and lowest under nonpreferential list PR and SMP. We shall not try to test the impact of district magnitude, one of the factors suggested by Matland and Studlar (1995, 8), who hypothesize that increasing district magnitude is associated with more seat changes because even small changes in votes will lead to changes in seats. District magnitude in Ireland has varied from three to nine over the period, but the periodic changes in constituency configurations make it hard to construct a realistic test of the hypothesis. In Malta, there is even less variance; of the 85 constituencies used in elections over the period, the great majority have returned five members, and the rest have returned six members. In fact, the terms of the hypothesis relating to turnover could be questioned: although it is no doubt true that the chances of at least one seat changing hands are greater in a 20-seat constituency than in a single-seat constituency, it may not be true that the number of seats changing hands in a 20-seat constituency is likely to be more than in 20 single-seat constituencies considered together, because under an SMP electoral system, especially with a multiparty system, small changes in votes can lead to large and unpredictable changes in seats. We might thus expect more overall turnover in small constituencies; indeed, turnover in Canada (with single-member constituencies) is relatively high, and at the first 10 French elections of the Fifth Republic, the number of defeated incumbents was lowest at the only election in which multimember constituencies were employed, in 1986 (Ysmal 1994, 206). Page 86 →

Two Political Systems: Ireland and Malta A few basic details of the two political systems under discussion will help set the context. Ireland, like almost all other European countries, has a parliamentary rather than a presidential system; its president, though directly elected, has very few powers (fuller details of the political system can be found in the various chapters of Coakley and Gallagher 1999). Its lower house of Parliament, Dáil Éireann, is elected for five-year terms, though the timing of elections is at the discretion of the prime minister (the Taoiseach), and on average elections have occurred at intervals of about three years since the independent Irish state was founded in 1922. The Dáil currently has 166 members (TDs). Since 1932, three political parties have dominated Irish politics in a reasonably stable pattern (see Table 2 for the picture over the period covered in this chapter). The largest party since that date has been Fianna Fáil, which usually wins around 40 to 45 percent of the votes at elections; in the European Parliament (EP), it was allied with the French RPR (the Rassemblement pour la République, the Gaullists) until the RPR split in 1999, since when Fianna Fáil has been part of a group including the more Eurosceptic wing of the RPR and the Alleanza Nazionale of Italy. The second-largest party at every election since 1932 has been Fine Gael, allied in the EP with the Christian Democratic group; its average electoral strength is around 30 percent. The other durable party has been Labour, a member of the Socialist group in the EP, whose average strength is around 11 percent. Up until 1989 all governments were either single-party Fianna Fáil administrations or coalitions containing Fine Gael and Labour (and sometimes one or more of the smaller, transient parties). In 1989, Fianna Fáil took part in a coalition government for the first time, with the Progressive Democrats (members of the Liberal group in the EP). After the 1992 election, when Fianna Fáil support dropped to less than 40 percent for the first time since 1927, Fianna Fáil joined a coalition with Labour, and the increasingly unpredictable nature of Irish government formation was shown two years later, in November 1994, when for the first time ever a change of government took place without an election. Labour withdrew from its coalition with Fianna Fáil and took part in a new three-party government along with Fine Gael and Democratic Left, a small left-wing party. At the June 1997 general election, this government was defeated and replaced by a fresh coalition between Fianna Fáil and the Progressive Democrats. TABLE 2. Electoral Fortunes of Irish Political Parties, 1927–97

Page 87 →Malta is in effect a two-party system, one of the purest in the world. Virtually all the votes at every election are won by the Maltese Labour Party (MLP) or the Partit Nazzjonalista/Nationalist Party (PN); no other party has won a seat since independence in 1964. The MLP corresponds to the European Socialist or Social Democratic tradition; the PN, to the Christian Democratic. Like Ireland, Malta has a parliamentary system, with its Parliament elected for five-year terms. Elections tend to be well spaced out, with only eight in the first 35 years after independence. Its lower house is much smaller than Ireland’s, reflecting its smaller population (around three hundred and fifty thousand, compared with Ireland’s 3.7 million).

STV in Ireland and Malta In Ireland, PR-STV has been used for all parliamentary and other public elections since 1922 (Sinnott 1999). It is constitutionally prescribed (Article 16.2) and thus cannot be changed without a referendum. Two attempts have been made to replace it by the system of single-member plurality, or first-past-the-post, in 1959 and 1968. On each occasion the proposal was defeated by direct vote of the people in a referendum: in 1959 by 52 to 48 percent and in 1968 by 61 to 39 percent. Since 1948, all constituencies (districts) have returned either 3, 4, or 5 members, and average district magnitude has been about 4; at the 1997 election, for example, there were 41 constituencies from which 166 TDs were returned. In earlier years there were larger constituencies; constituencies of 7, 8, and 9 members from 1923 to 1933 and of 7 members from 1937 to 1944. Proportionality is high,3 and there does not seem to be much concern, or reason for concern, about the degree of government stability. When PR-STV is criticized Page 88 →in Irish political debate, it is mainly on the grounds that it exposes incumbents to an excessive risk of defeat, especially to running mates, and therefore imposes an unhealthy mode of behavior on TDs—unhealthy both for them personally and for the political system as a whole. Some of the arguments lying behind this criticism will be tackled in this chapter. PR-STV has been employed for every election in Malta since 1921 (Hirczy de Miño and Lane 1996). Since 1976, the Maltese Parliament has contained 65 regularly elected members, who are elected from 13 5-member constituencies; this marks an increase from 55 in 1971 and from 50 at the first postindependence election in 1966. Only in 1971, when 5 of the constituencies returned 6 members, have anything other than 5-member constituencies been employed—although owing to the use of adjustment seats (discussed later), at the 1987 and 1996 elections, 4 5-seat constituencies were in effect turned into 6-seaters. On these two occasions, the size of Parliament increased to 69 members. Differences between Ireland and Malta Three features of STV—two institutional, one party systemic—in Malta distinguish it from STV in Ireland (see Farrell, Mackerras, and McAllister 1996 and Farrell 1997, ch. 6, for discussion of the different versions of STV). The first, the provision of adjustment seats in certain circumstances, came about as a result of the outcome of the 1981 election, when the PN won an overall majority of votes (50.9 percent) yet the MLP won a majority of the seats (34 out of 65). This was due to the way the constituency boundaries had been drawn by the MLP government, an arrangement the PN alleged had been a gerrymander. The PN boycotted Parliament in consequence. A new rule was then introduced, according to which if a party won a majority of votes but only a minority of seats, it would receive additional seats so as to give it an overall majority. This provision had to be invoked at the next election, in 1987, when the PN again won 50.9 percent of the votes but again lost 34 to 31 in seats to the MLP. Accordingly, its four “best losers” were awarded seats to give it a 35–34 majority, and the size of Parliament grew to 69. The same provision was activated in 1996, when the MLP won a majority of the votes but a minority of the seats. A second feature that distinguishes STV in Malta and Ireland concerns the procedures for filling vacancies that arise between elections. In Ireland, vacancies are filled in the same way as in Britain, through byelections—indeed, this is one of the many ways in which British practices were virtually Page 89 →automatically adopted in the independent Irish state after 1922. Although the use of by-elections in a PR electoral system could

lead to anomalies, in practice by-elections have not produced particularly anomalous outcomes (Gallagher 1996). In Malta, in contrast, vacancies are filled through “casual elections” (elezzjoni kazwali), though confusingly these are usually referred to in English as by-elections. At these “casual elections,” no fresh vote takes place; the count is based on the ballot papers of the resigning or deceased MP.4 The procedure is the same as that employed in Tasmania, where it is referred to as the “countback” or “count-on” method (Newman 1992, 132–38; O’Connell 1983,45). This means that in Malta candidates can stand in two constituencies; if elected in both, they resign one seat (as required by law), knowing that at the resultant casual election another candidate of their own party will be elected. Consequently, many Maltese candidates stand in two constituencies, either as a sign of status (it seems to have become de rigueur for party leaders to secure election in two constituencies) or as a kind of insurance policy or, in the case of lesser-known candidates, to boost their chances of picking up the crumbs when a vacancy arises or simply because redistricting has split their base between two constituencies. Quite apart from these, it is not unknown for a leading candidate to switch to another constituency and thereby in effect displace one of his own party’s incumbents in that constituency. In 1987 the 173 candidates were responsible for 242 candidacies, and in 1992 most candidates stood in more than one constituency: the 162 candidates were responsible for 246 candidacies (Schiavone 1992, 615). When candidates are doubly elected, they immediately resign one seat, and another candidate of their party is elected at the casual election. In this chapter, we shall treat these cases as if the doubly elected candidate had never stood in the constituency from which he5 immediately resigns; that is, the candidate who wins the seat through the casual election is treated as if he or she had been elected at the general election, as is in effect the case. In Ireland, in contrast, this complication does not arise; in the period covered, only one candidate was doubly elected, in September 1927. The third difference between Malta and Ireland was described earlier as being party systemic. It refers to the willingness of voters in Ireland, and the complete unwillingness of their Maltese counterparts, to vote on other than a strict party basis. Maltese voters express preferences for the candidates of their favored party, and then stop (see chapter 9). Irish voters, however, are more likely to give lower preferences to another party’s candidates after voting for their own or, indeed, not to vote on a party basis in the first place; for Page 90 →example, some vote on the basis of locality, so they might give their first preference to a Fianna Fáil candidate from their area and their second to a Fine Gael candidate from the same area. At the 1997 Irish election, Fianna Fáil’s internal transfer solidarity was only 68 percent, and Fine Gael’s was 64 percent (Gallagher 1999, 138–39). The relevance of this for turnover is that for Irish parties the concept of overnomination is meaningful; in some circumstances, too complex to be discussed here, running too many candidates might cost a seat. Very often, a party organization comes to the conclusion that if it is aiming to win n seats in a constituency, n is the optimal number of candidates to run, and it rarely runs more than n + 1. In Malta, in contrast, there is no such thing as too many candidates; no matter how fragmented the party vote is, the party can be confident that when vote transfers are made, virtually all of the votes will remain within the party fold. Hence, whereas even the largest Irish parties rarely run as many candidates in a constituency as there are seats at stake, the Maltese parties usually run many more than this—the record was 19 by the MLP in the 5-member district 2 in 1998. Therefore, we should expect the risk of intraparty defeat to be higher for Maltese incumbents than for their Irish counterparts, because the former are always accompanied on the ticket by a host of rivals for the party’s seats, whereas Irish incumbents might not be. At the 1997 election in Ireland, 7 percent of the Fianna Fáil incumbents who contested the election, 26 percent of Fine Gael incumbents, and 90 percent of Labour incumbents were on a party ticket that did not contain any nonincumbents. Scope and Sources In Ireland, the 24 elections of the period from 1927 to 1997 are covered. This omits the first two postindependence elections, which were held in 1922 and 1923. The 1922 election, held shortly before the outbreak of civil war, was marked by highly constrained competition: there were only 176 candidates for 128 seats, and 38 of the 128 TDs

were returned without a contest. The 1923 election saw a large increase in the size of the Dáil, to 153 seats (an increase of almost 20 percent, which is the level at which Lijphart [1994, 13] suggests that a change in assembly size should be regarded as bringing about a new electoral system), and this, coupled with extensive redistricting, makes it relatively hard to establish causes of exit from and entry to the legislature. For the record, it may be stated that 36 of the 128 TDs elected in 1922 did not contest the 1923 election, a further 18 were defeated, and 74 were reelected. The Page 91 →source for the period from 1948 onward is officially published results, which contain details of interelection by-elections. For the 1923–44 period, for which there are no official figures, Gallagher 1993b has been used. In Malta, the first eight postindependence elections (1966–98) are included in the study. The data set for Malta involves only about a seventh as many outgoing deputies as in Ireland. The main printed source for Maltese elections is Schiavone’s (1992) comprehensive study of the period from 1849 to 1992, which contains not only the outcome in every constituency at every election but also full details of changes between elections caused by the deaths and resignations of members, details that are needed to identify the incumbents at the time when an election is called.6 In addition, John Lane has created an invaluable data bank on Maltese elections, including a complete list of the 3,364 election candidacies of the period from 1921 to 1998, as well as detailed results from every constituency.7

Turnover in Multimember Constituencies Incumbent deputies may suffer one of three basic fates at an election: reelection, retirement, or defeat. Two of these may in turn be sub-subdivided as follows: 1. Reelection 2. Retirement Retirement and replacement by another deputy of the same party Retirement and replacement by a deputy of a different party Retirement and disappearance of the seat owing to redisricting 3. Defeat Defeat to another candidate of the same party Defeat to a candidate of a different party Defeat owing to redistricting In theory it should be a simple matter to assign each outcome to one or another of these categories. However, although it is clear enough whether a deputy was reelected, retired, or was defeated, it is not always so clear to which of the subcategories of retirement or defeat an outcome should be assigned. Clear outcomes occur where a constituency (district) remains unchanged from one election to the next and at least n – 1 of the n incumbents are reelected. If one incumbent does not contest the election, or contests and is defeated, Page 92 →then it is obvious that that incumbent has been replaced, either through defeat or retirement, by whoever is the new deputy. Matters become slightly more complicated when two (or more) incumbents are replaced by newcomers, which creates the problem of determining which newcomer replaced which incumbent. The approach taken here is to assume that if an incumbent of party A disappears (i.e., retires or is defeated) and a newcomer of party A is elected, then the incumbent was replaced by the newcomer in question. In Scenario 2 below, for example, the turnover is treated as two cases of intraparty defeat rather than two cases of interparty defeat, with newcomer B

unseating incumbent A and newcomer A unseating incumbent B, which would be an implausible interpretation of events given the reasonable stability in party support in Ireland and the very high stability in Malta. When the strengths of the parties remain unaltered and the only change is within parties, turnover is best seen as intraparty.

Scenario 1 Party A Incumbent defeated Party B Incumbent defeated Party C Newcomer elected Party D

Newcomer elected

Categorization: Two interparty defeats (whether A to C and B to D or A to D and B to C may be judged in particular cases but is irrelevant for present purposes). Scenario 2 Party A Incumbent defeated Newcomer elected Party B Incumbent defeated Newcomer elected Categorization: Two intraparty defeats (A and B). Scenario 3 Party A Incumbent defeated Newcomer elected Party B Incumbent defeated Party C Newcomer elected Categorization: One intraparty defeat (A) and one interparty defeat (B to C). Scenario 4 Party A Incumbent defeated Newcomer elected Incumbent defeated Party B Newcomer elected Page 93 →Categorization: One intraparty defeat (A) and one interparty defeat (A to B). Which incumbent sustained the intraparty defeat and which the interparty defeat may be judged in particular cases but is irrelevant for present purposes. Scenario 4, in which two incumbents of party A lose (one to a newcomer of party A and the other to a candidate of party B), is treated as one intraparty defeat and one interparty defeat, as indicated. It should be noted that this differs from the approach of Katz (1986), who, in his analysis of turnover, says: “If at least one nonincumbent was elected in a constituency, then all defeats of incumbents of the same party were regarded as ‘intrapartisan defeats’ ” (99). This seems to mean that he treats the defeats of both incumbents of party A in scenario 4 as intraparty defeats, even though only one incumbent was ousted by a newcomer of party A, an approach that exaggerates the extent of intraparty defeats. Redistricting poses a problem, especially (though not exclusively) when it comes to retirements, in that there is sometimes room for dispute as to by whom retiring deputies have been replaced when the configuration of constituencies changes. In Ireland, during the time period covered in this chapter, significant redistrictings took place before the elections of 1948, 1961, 1969, 1977, and 1981, and the map covering the greater Dublin area, which accounts for about a quarter of Dáil seats, was extensively redrawn on some of these occasions. Best efforts have been made, by comparing maps of the constituency configurations, to estimate the replacement of a retiring deputy in these circumstances. On occasions, the size of the Dáil was increased—by 12 percent in 1981—so a

number of seat gains are attributable to the creation of new seats. Even when the overall number of seats does not change, the number in particular regions of the country may change (for example, between 1973 and 1977 the greater Dublin region gained five seats, while the rest of the country lost a seat), so specific seat gains and losses may still be attributable to the effects of redistricting. In Malta some degree of redistricting took place before every election, and the size of Parliament increased by 10 percent in 1971 and by a further 18 percent in 1976, so a significant number of newcomers were able to gain election without ousting incumbents. In addition, the expansion (in 1987 and 1996) and contraction (in 1992 and 1998) of four constituencies owing to the adjustment seats is treated here as a functional equivalent of redistricting. Unfortunately, there are still some cases that are impossible to categorize and for which the cause of defeat, or the destiny of the seat of a retiring Page 94 →deputy, is best treated as uncertain. This is much more common in Malta than in Ireland.

The Fates of Incumbents Table 3 shows the fates of incumbents for Irish elections over the period from 1927 to 1997, and table 4 presents the Maltese data. Although reelection rates are a little higher in Ireland and defeat is a little more common in Malta, the similarities between the two are striking. Perhaps the two most important points established are first, that by far the most common fate for an incumbent in both countries is reelection and second, that defeat is a more common source of turnover than retirement. TABLE 3. Fate of TDs at Elections in Ireland, 1927–97 TABLE 4. Fates of MPs at Elections in Malta, 1966–98 Page 95 →To take the first of these, around three-quarters of outgoing deputies and around four-fifths of those deputies who actually stand for reelection are reelected. Clearly, most incumbents in both countries are well able to fend off most of the threats to their position. Although systematic comparative figures are hard to come by, this figure seems to be at the higher end of the scale of reelection rates, judging by what data are available. Matland and Studlar use the first of the four measures we outlined earlier (the percentage of deputies who, having been elected at one election, are also returned at the next), and their figures rank Ireland fifth and Malta eleventh of the 25 cases they cover. The incumbency return rates for the elections covered are 77.2 percent in Ireland and 70.7 percent in Malta, compared with an average of 68.9 percent for all the countries covered (Matland and Studlar 1995, 36; cf. the figures in Katz 1986, 98, and Somit 1994, 12). Variation among the main parties is not great, as tables 3 and 4 show, but TDs of smaller parties (classed along with independents as Others) are clearly less likely to be reelected.8 And neither do the figures vary hugely from election to election. In both countries, the reelection rate at the first election in the period covered is the lowest of any, and the rates of both defeat and retirement are the highest. This can be seen as part of the process of the “institutionalization” of the legislatures (cf. Polsby 1968), as those whose commitment to a political career is halfhearted are weeded out. Once this had happened, and it happened early, both parliaments were peopled by deputies whose priority was staying there. The average figures for the main parties in each country vary somewhat more than the aggregate figures, but not greatly, and there is no linear trend over time. The range for Fianna Fáil during this period covers a trough of 77 percent in 1992, when the party fell to a 65-year low of support, and a peak of 95 percent in 1969,9 when a partisan redistricting of constituencies (no longer possible because the task of redistricting was handed to an independent commission in the late 1970s) ensured that only 3 of the party’s 60 contesting TDs lost their seats. For Fine Gael, the lowest reelection rate (63 percent) came in 1977, when the party slipped heavily in popular support, and the peak was in 1981, when a major gain in votes combined with a large increase in the number of seats in the Dáil to generate a reelection rate of 97 percent. In Malta the MLP’s range is from 72 percent in 1987 to 95 percent in 1971; the PN’s, from 68 percent in 1987 to 88 percent in 1976. The high figure Page 96 →for the MLP in 1971 is due to a large increase in its vote (nearly 8 percent, the largest change for any party since independence) coupled with an increase in the size of Parliament,

and the PN’s peak in 1976 can also be attributed to a major increase in the size of Parliament. The low figures for both parties in 1987 can be explained only in terms of an upsurge in intraparty defeats, as we shall see.

Causes of Defeat: Interparty or Intraparty? Incumbents, then, are usually reelected despite the multiple threats to their position. When they do lose, why do they lose, and which of those threats proves the most potent? In particular, are incumbents more likely to lose to running mates or candidates of another party? The view that intraparty competition looms large as a source of incumbent defeat features in the arguments of critics of PR-STV in the Irish context, whose view is that the degree of competition within parties compels parliamentarians, for fear of losing their seats otherwise, to concentrate unduly on “home-style” activities, that is, serving the needs of their constituents so as to retain their personal popularity, to the neglect of their national parliamentary responsibilities (see, for example, Boland 1991 and FitzGerald 1991). Exactly the same has been suggested for the United States, where King (1997) argues that precisely because incumbents are so vulnerable, at least in their own minds, they “go to prodigious lengths to protect themselves,” resulting in hyperresponsiveness, nonstop electioneering, and malfunctioning of the political system (49, 179). The other side of the case is that the PR-STV system has little to do with the high volume of demands on TDs to undertake constituency work, which would most probably continue to exist whatever the electoral system. Carey (1996), observing that Costa Rican legislators engage extensively in constituency work even though they are barred from seeking reelection, queries the causal link between particularistic activities and personal electioneering (198). In addition, a group established to review the Irish constitution observed that there will inevitably be “a conflict of interest between incumbent legislators who want as much security and as little turnover as possible, and voters who want legislators to be as responsive as possible, and have only the threat of unseating them to ensure this” (Constitution Review Group 1996, 57). If TDs did not feel vulnerable to defeat, some degree of accountability and answerability would be lost. Page 97 →Carty (1981) concluded that for Fianna Fáil and Fine Gael candidates in the period from 1948 to 1973, “fellow partisans constitute the single most important source of competition” (115). However, the word single in this phrase is significant; Carty’s point is that more seats are lost to running mates than to the nominees of any other specific party. When one groups all other parties together, a different picture emerges: Carty’s tables (115–16) show that for every party, more seats are lost to other parties’ nominees than to running mates. Looking at the sources of defeat over the 1927–97 period, table 5 shows that most defeated TDs are the victims of challengers of other parties rather than a running mate. Sixty-one percent of defeats have been interparty ones compared with 34 percent resulting from intraparty competition.10 Variations from election to election over the 70-year time span are pronounced, with especially high rates of intraparty defeat in 1933, February 1982, and 1938 and of interparty defeat in 1973, 1943, and 1957, but there is no consistent pattern over time; one cannot say that intraparty competition is becoming more salient, or less salient, as a source of defeat. Interparty variation is more significant. It is clear from table 5 that defeat by a running mate is not a possibility that need keep Labour or minor-party incumbents awake at night.11 For Fianna Fáil and Fine Gael, though, the threat is real; considering these parties together, the proportion of defeats at the hands of running mates (46.6 percent) is almost as high as the proportion inflicted by opposing party candidates (47.1 percent). In Fianna Fáil, running mates are clearly the main source of danger to an incumbent, and they pose a significant, if secondary, threat to Fine Gael incumbents. TABLE 5. Defeated TDs at Elections in Ireland, 1927–97 Page 98 →Even for the main parties, though, the extent of the threat constituted by running mates should not be exaggerated. On average over the last 24 elections, Fianna Fáil and Fine Gael together have had 120 outgoing TDs when the election is called, and of these 10 have retired, 93 have been reelected, and 17 have been defeated. Of those defeated, 8 have been defeated by a candidate of another party, 8 have been defeated by a running mate, and

1 has lost his or her seat owing to redistricting. Viewed objectively, most TDs seem to have little to worry about—although, of course, most of them do worry. Summing up this discussion of the threats to incumbents in Ireland, it is apparent that the direction of the threat varies according to party. In Fianna Fáil the main threat is from one’s running mates, though other parties’ candidates are capable of causing more than a little damage; a Fine Gael TD also needs to keep his or her eyes peeled in all directions, with other parties posing the greater threat; and TDs of the smaller parties can concentrate on fighting the enemy without, as none of any significance lurks within the party fold. The Irish and Maltese patterns have hitherto been very similar, but in this area they differ greatly. As table 6 shows, for a Maltese MP, running mates constitute by far the greatest threat. What is most striking about Maltese turnover, in fact, is the extreme rarity of defeats of MPs by a candidate of the other party. Of the 12 MPs who suffered this fate in the 1966–92 period, 7 lost in the 1966 election; 6 of these were MPs of smaller parties that disappeared after independence. Since then, there are only five cases, at seven general elections, where one can safely conclude that a MP lost his or her seat to a candidate of another party, and even two of these were special cases.12 Finally, it may be noted that the pattern in Tasmania, the other small island that uses PR-STV to elect its Parliament, is midway between those of Ireland and Malta. On average, between two and three times as many defeats of Tasmanian members of the House of Assembly are sustained to running mates as to candidates of other parties (Newman 1992, 255; Hughes, ch. 8 in this volume). TABLE 6. Defeated MPs at Elections in Malta, 1966–98 Page 99 →

Explaining Turnover Rates Merely to measure turnover is not straightforward, so explaining it will not be easy either. We have identified three features of the Irish-Maltese pattern: (1) a similar rate of defeat for incumbents in both countries, (2) a much higher rate of interparty defeat in Ireland, and (3) a much higher rate of intraparty defeat in Malta. The only systematic attempt to explain turnover rates cross-nationally is that made by Matland and Studlar (1995), who suggest eight factors as being likely to influence turnover: electoral system, district magnitude, fixed-term parliaments, redistricting, prestige of legislature, competitiveness of elections, socioeconomic development, and the country’s religion. Because overall turnover rates, as opposed to the reasons for turnover, differ little between Malta and Ireland, there is not much between-country variance to explain. Four explanations could be suggested for the different causes of turnover in the two countries studied. First, voting behavior seems to be much more stable in Malta than in Ireland (Zanella 1989). Volatility, as measured by Pedersen’s index, averages only 4.6 percent at the 8 Maltese elections we are considering, compared with 10.2 percent at the 24 Irish elections covered.13 Average volatility measured at the constituency level was 16.3 percent at the 1997 Irish election compared with only 3.9 percent at the 1998 Maltese election. Thus, it is not surprising that interparty turnover is much lower in Malta. Second, and relatedly, Malta has a two-party system, whereas Ireland has a multiparty system; in Ireland more parties stand to gain or lose seats in each constituency. Third, the explanation for the higher rate of intraparty turnover in Malta might also be related to the two-party system: if a desire for change in representation cannot be, or is not, reflected in a switch of voting support to another party, it must express itself by voting for another candidate within the chosen party. Fourth, as we have mentioned, Maltese MPs are always accompanied on the ballot paper by many nonincumbent running mates, unlike Irish TDs, who usually have few or no nonincumbent running mates. The run of elections in Malta is too small to permit refined analysis, but the 24 Irish elections in our data set allow for some hypotheses to be tested. Six plausible hypotheses would be the following: 1. Page 100 →The longer the amount of time that has elapsed since the previous election, the higher will be the number of retirements. The rationale for this is obvious.

2. The longer the amount of time that has elapsed since the previous election, the higher will be the number of defeats. The rationale is that as time passes, the composition of the electorate changes ever more, and the number of continuing voters inclined to vote differently can be expected to increase. 3. The greater the amount of volatility at an election, the higher will be the number of deputies defeated. Again, the rationale is obvious. 4. The larger the party, the more likely it is that defeats of its incumbents will be due to intraparty turnover. The rationale is that if a party has several incumbents in a constituency, it is more likely that one of them will become electorally vulnerable, whereas for a small party with only one incumbent, the personal support of the incumbent is likely to be a significant factor, making less likely the prospect of the incumbent being ousted by a running mate. 5. Turnover will be related to a party’s gains or losses of votes at an election; as a party’s performance worsens, a higher proportion of its incumbents will be defeated, and a higher proportion of the defeats of its incumbents will be interparty. Clearly, we expect a party losing votes to lose seats and to lose them to other parties; 6. Likewise, the better a party fares, not only will a lower proportion of its incumbents will be defeated, as hypothesis 5 states, but in addition a higher proportion of the defeats of its incumbents will be intraparty. We expect a party gaining seats to suffer few incumbent defeats, with the probability that many of these defeats are to running mates rather than to other parties’ candidates. We shall test these hypotheses through simple correlation analysis (multiple regressions were tried, but because of the small N—only 24 elections—there was no equation in which more than one independent variable was significant). The results are presented in table 7. 1. This hypothesis was confirmed. The longer the life of a parliament, the greater the number of incumbents who retire at the end of that parliament. The Pearson correlation coefficient (r) between months since the previous election and the percentage of incumbents retiring is 0.57 (significant at .01 level—see part I of table 7). Page 101 → TABLE 7. Correlations between Independent Variables and Aspects of Turnover, Irish Elections Page 102 →2. This hypothesis was only weakly confirmed. The longer the life of a parliament, the greater the number of contesting incumbents who are defeated. The r between months since previous election and the percentage of contesting incumbents who were defeated is 0.35 (not statistically significant—see part I of table 7). One of the assumptions on which this hypothesis was based is false: there is little relation between the degree of volatility at an election and months elapsed since the previous election (r = 0.11, not statistically significant). 3. This hypothesis was confirmed. When there is a lot of change in voting behavior, a lot of incumbents are defeated. The r between volatility at an election and the percentage of contesting incumbents who were defeated is 0.49 (significant at .05 level—see part I of table 7). 4. This hypothesis is strongly confirmed. Intraparty defeat is a much greater danger for incumbents of large parties than for incumbents of small parties. The r between the vote received by a party and the proportion of its defeats that are intraparty is 0.72 (significant at .001 level—see part II of table 7). The relationship is almost certainly even stronger than this, given that the many small parties have been grouped into collected “Others.” 5. The first part of the hypothesis is confirmed. As a party’s vote goes up, its proportion of incumbents who are defeated goes down (r = –0.35—see part II of table 7). The second part of the hypothesis is only weakly confirmed. Increase in a party’s vote is related to the proportion of its defeats that are interparty (r = – 0.25—see part II of table 7), but the relationship is not strong. Given the extreme infrequency of intraparty defeats among the smaller parties, it makes sense to examine this also for just the largest two parties, Fianna Fáil and Fine Gael (see part III of table 7). Both aspects of the hypothesis are now strongly confirmed. The r between change in the vote and the percentage of contesting incumbents who were defeated is –0.68 (significant at .001 level). Likewise, the r between change in the vote and the proportion of

defeated incumbents whose defeat was interparty is –0.44 (significant at .01 level). 6. This hypothesis is weakly confirmed. When a party does well, a higher proportion of its incumbent defeats are intraparty rather than inter-party. The r between change in the vote and the proportion of defeated incumbents whose defeat was intraparty is only 0.21 (see part II of table 7). Again, we should examine this also for just the largest two parties. The hypothesis is now strongly confirmed. The r between change in the vote and the proportion of defeated incumbents whose defeat was intraparty is 0.54 (significant at .001 level—see part III of table 7). Page 103 →

Retiring TDs Having looked at the fate of defeated TDs, we will consider briefly what happens to those seats vacated by TDs who do not contest. The number of retiring TDs in Malta over the period is too small (only 36) to make analysis worthwhile, so we will consider only the Irish figures (see table 8). As indicated earlier, the retirement category can be broken down as follows: Retirement and replacement by another TD of the same party Retirement and replacement by a TD of a different party Retirement and disappearance of the seat due to redistricting The contrast with table 5 is apparent. Whereas most defeated TDs are supplanted by opponents, most retiring TDs are replaced by someone else of their own party. About two-thirds of seats held by retiring TDs of the two main parties are retained by the party, indicating that these are, if not quite party seats, at least seats that reflect a stable body of support for the party. For the smaller parties the picture is quite different, reflecting their general lack of a durable and sizable base of support that can be relied on to produce a seat for the party almost regardless of its candidate. It would be an exaggeration to see these smaller parties’ seats as being fiefdoms, as it were, of individual TDs whose support is mainly personal, but clearly the support base of minor-party TDs combines a significant personal element along with the purely party vote, which helps explain why these TDs are in so little danger of being ousted by a running mate. TABLE 8. Retiring TDs at Elections in Ireland, 1927–97 Page 104 →

Incoming Members of Parliament Having considered political life from the perspective of incumbents, we might now ask how things look from the viewpoint of a nonincumbent who wants to be elected to a parliament. Do newcomers become deputies mainly by ousting incumbents of their own party, by winning a seat from another party, or by filling the shoes of a retiring deputy of their own party? We can identify five ways in which one might become a deputy at a general election:14 1. 2. 3. 4.

Replacing a retiring deputy of one’s own party Ousting an incumbent deputy of one’s own party Ousting a deputy of another party Taking a seat from another party without ousting a deputy, that is, replacing a retiring incumbent of another party 5. Gaining a seat owing to redistricting As with defeated and retiring deputies, there can be room for argument about how to classify particular instances of nonincumbents being elected. Matters become particularly complicated when incumbents do not actually retire but simply move constituencies, change parties, or both. Rather than double-count all the changes that take place

when incumbents shift rather than actually bow out of politics altogether, most of these cases are treated as uncertain.15 Page 105 →Table 9 demonstrates that no one route to Parliament dominates in Ireland. A plurality of victorious nonincumbents16 win a seat by ousting a deputy of another party, but over a fifth oust a deputy of their own party, and nearly as many step into the shoes of a retiring deputy of their own party. Around a ninth replace a retiring deputy of another party, and a smaller number pick up a seat owing to redistricting, which was an especially significant factor in 1981, given the increase in the total number of seats from 148 to 166, and in 1948, when the increase was from 138 to 147. There is a certain amount of interparty variation. For the smaller parties, as we would expect from our discussion earlier, it’s not common to replace a retiring deputy of one’s own party, and it is virtually impossible to oust an incumbent running mate. Success for a candidate of Labour or one of the smaller parties almost invariably comes by taking a seat from another party: either ousting an opponent (the more common method) or filling the vacuum when a rival-party TD retires. TABLE 9. Incoming TDs at Elections in Ireland, 1927–97 Page 106 →For candidates of the two main parties, there are richer pickings to be had by targeting a seat already held by the party than by eyeing a seat held by another party. In both cases, candidates are slightly more likely to succeed by ousting a running mate who is an incumbent than to replace a retiring party TD. Looking at the four party groupings in the table, it is apparent that the likelihood of entering the Dáil by replacing or ousting a deputy of one’s own party is strongly related to party size. Obviously, when a party holds more seats, there are more retiring or contesting incumbents to replace or oust. Major-party hopefuls operate in a relatively target-rich environment compared with smaller-party hopefuls, who have only seats currently held by other parties to aim at. This does not necessarily mean that candidates of large and small parties can adopt different campaigning styles. It might seem that small-party candidates would be best advised to fight their campaigns on the basis of policy, seeking to attract voters away from rival parties, whereas candidates of larger parties should compete on the basis of a personalistic or localistic appeal, because their best bet is to oust a running mate or inherit a seat from a fellow party TD who is standing down. However, in reality nearly all candidates, whether of large or small parties, need to attract a certain personal vote and to make themselves sufficiently appealing to attract lower preferences from supporters of other candidates in the form of vote transfers, so in practice all candidates could be expected to campaign in much the same way, mixing a policy-based and a personal approach, though it does not appear that any research into this specific area has yet been conducted. In Malta, again as we would expect from the earlier discussion, the most common route to Parliament consists in ousting an incumbent of one’s own party (see table 10). Taking a seat from an opponent was a significant means of entering Parliament only in 1966. Redisricting—in practice the expansion of Parliament—was also important in 1971 and 1976.

A Problem of Incumbency? A number of contributors to a comparative study of turnover discuss the “problem of incumbency” (Somit et al. 1994). Unfortunately, a lack of clarity as to what is meant by turnover, and a lack of rigorous comparisons between chapters, make it hard to tell whether there is a problem. In Canada, the high level of turnover is seen as a problem for the political system (Laponce 1994, 138), though the reader has difficulty in ascertaining just what that level is. The Japanese reelection rate of 85 percent is described as “relatively low” (Reed 1994, 283), yet the lower German rate of reelection (presumably around 75 percent, given that there is 25 percent turnover at each election on average) is seen as a problem because it is too high (Boll 1994, 181; Somit 1994, 15). Indeed, incumbency per se is described at one point as “a problem or even a threat to democracy as a whole” because “men and women are always reluctant to relinquish power,” and there is a warning about “Caesarism” (Boll and Römmele 1994, 20). Page 107 → TABLE 10. Incoming MPs at Elections in Malta, 1966–98

Page 108 →Much of this concern seems to be misplaced. High or low rates of incumbent reelection may be identified as a problem as a means of avoiding having to grapple with larger problems (this point is made by Boll 1994, 182–83). In Ireland, where, as we have seen, incumbent reelection rates are above average, the concern is not about too much incumbency but too little incumbent security, the argument being that the very real fear of losing one’s seat, especially to a running mate, adversely affects the behavior of deputies, turning their attention from national to local issues (see “Causes of Defeat” earlier in this chapter). The average Irish TD or Maltese MP engaged in constituency work to retain his or her seat, sorting out defective street lights in Maam Cross or a social welfare problem in Marsaxlokk, would be surprised to find him- or herself accused of Caesarism. The argument that it is too difficult to unseat incumbents and that not enough incumbents are defeated at elections for the good of the political system, has never, to this writer’s knowledge, been aired in Ireland.

Intraparty Competition and Party Cohesion Finally, we might ask whether the incidence of turnover, and especially the constant presence of intraparty electoral competition in the larger parties, has an adverse impact on party cohesion. A priori, we might expect intraparty competition to have a negative effect on party unity. In theory at least, candidates can establish their own electoral base and can defy party policy and instructions with impunity, and members of a parliament may be inclined to attach more importance to constituency pressures than to the party whip when deciding how to vote in the legislature. Yet when the evidence is examined, a Page 109 →clear causal link between preferential voting and party disunity proves hard to find (Katz 1980, 31–34; Marsh 1985b, 375). PR-STV, of course, allows not just a choice among candidates within a party list but a choice among candidates without regard to party. This might weaken party unity still further, with party labels becoming less meaningful than in preferential list systems. Some have argued exactly this. Katz (1980) studied the impact of the electoral system on parties and the party system in Ireland, Italy, and the United Kingdom—though in the Irish case many of his conclusions were based on responses from just 28 political actors, including TDs and constituency officials of the parties (125). He argues (107–8) that in Ireland, PR-STV has produced parties that are not internally cohesive. In a similar vein, and seemingly basing his claims largely on Katz, Blais (1991) speaks of “strong evidence that the single transferable vote leads to a weaker party system” and maintains that PR-STV “is detrimental to the development of a responsible party system” (248–49). Taagepera and Shugart (1989) say that “if strength of party organization is desired, STV is inappropriate, because . . . list PR (even with preference voting) . . . gives far more leeway to party elites in deciding who the party’s representatives may be” (28). Just why list PR with preferential voting (as in Denmark, Finland, or Switzerland) can be expected to give more power to party elites than does PR-STV is not made clear. The implication, although it is not spelled out, is that in many open-list systems there is at least some default ordering that the voters are unlikely to disrupt—though in practice, elites in some of these cases, such as Finland and Switzerland, do not seem to exercise any more control than they would under PR-STV (Marsh 1985b). One obvious difficulty for those who maintain that the Irish parties are not cohesive is the fact that in legislative votes, party deputies almost invariably vote en bloc. Deviations from party solidarity are very rare and are met with a draconian response, typically expulsion from the parliamentary party. Fianna Fáil, indeed, has a rule that any of its TDs who even abstain on a measure, never mind vote against the party line, automatically incur expulsion from the parliamentary party.17 In Malta, too, MPs of the two parties are not known for their readiness to cross party lines in parliamentary voting. Both Katz and Blais argue that solidarity of party voting in parliament, although conventionally seen as a standard test of party cohesion, is not appropriate for the Irish case. Katz maintains that policy matters simply aren’t important to TDs, who for the most part are perfectly happy to vote on specific issues in whatever way the party leadership demands. TDs’ priorities lie elsewhere: Page 110 →The apparent unity of Irish parties is born of lack of interest in policy, lack of substantial differences in the policy orientations of the parties, and the realization that parliamentary cohesion is necessary to the survival of cabinet government. . . . the matters of real importance to the deputies are constituency services, and on these matters deputies who must electioneer independently continue to act independently. (Katz 1980, 108)

Similarly, Blais (1991) says that PR-STV’s adverse effect on party unity and cohesion “is mitigated by the greater importance attached to constituency service over policy, so that parties remain superficially united” (249). However, this rather savors of special pleading in the face of an inconvenient finding. When parliamentary party discipline in a number of European countries appears to be weakening (Gallagher, Laver, and Mair 1995, 52), it is difficult to justify dismissing the exceptionally high rates of solidarity to be found in Ireland and Malta as evidence of only “superficial unity.” It may, of course, be true that not all members of the parliamentary parties have exactly identical views on every issue, but the same is true of any parliamentary caucus in any country. An army that marches forward in a phalanx rather than scattering in all directions is more sensibly described as united than divided, even—or perhaps especially—if some members of the phalanx feel privately that they would prefer different tactics. If party solidarity in parliamentary voting in Ireland and Malta were significantly lower than the European average, we might plausibly attribute this to the effect of PR-STV. When the evidence does not run in this direction, we are forced to the conclusion that PR-STV does not lead to incohesive parties.18

Conclusion Incumbents at elections in Ireland and Malta under PR-STV do not suffer a particularly high rate of defeat. Incumbents are not always victorious, but they usually are, though not to the extent that any call for term limits has yet been heard. Reelection, despite the variety of threats from running mates as well as opposing party candidates, is the norm, with only about a fifth of those deputies who contest an election being defeated. Of Irish deputies who lose their seats, about the same proportion lose to a party running mate as to a candidate of another party; not surprisingly, defeat to a running mate is a serious risk only for major-party deputies. For incumbents of Ireland’s largest party, Fianna Fáil, the main threat comes from within party ranks; for other incumbents, Page 111 →the main danger lies without. Incoming deputies, on the other side of the coin, have a variety of targets at which to aim, and about as many enter the Dáil by succeeding or ousting a TD of their own party as by taking a seat from another party. In Malta the balance is quite different. Defeats to opponents are rare, and by far the most common cause of defeat is at the hands of a running mate. The differences between the Irish and Maltese patterns shows that PR-STV per se does not necessarily lead to specific outcomes. The consequences of this for the broader political system have been much debated, in Ireland particularly. Critics of PR-STV point to the constant internecine warfare within parties and to the large volume of constituency work undertaken by deputies in an attempt to remain ahead in popularity of running mates. It is also alleged that the intraparty competition that is inevitable under PR-STV leads to incohesive and disunited parties. Others believe that the demand for constituency work in Ireland, and the need to respond to it to some degree, would be much the same under any electoral system, and they see no signs of lack of cohesiveness in the Irish parties. What is certainly the case is that PR-STV is not associated, in either Ireland or Malta, with an above-average rate of defeat of incumbents.

NOTES Earlier versions of this chapter were presented at the conference “Elections in Australia, Ireland and Malta under the Single Transferable Vote,” Laguna Beach, California, December 14–15, 1996, and at the International Political Science Association Congress in Seoul, August 17–21, 1997. Thanks are due to participants in both meetings for their comments. Many thanks also to John C. Lane for his unflagging willingness to provide information and advice on Maltese political data. 1. The terms deputies, parliamentarians, MPs, and legislators will be used interchangeably. In Ireland the lower house of Parliament is Dáil Éireann, and its members are known as TDs. In Malta, members of Parliament (the House of Representatives, or Kamra tad-Deputati) are generally known as MPs. 2. The relevant footnote is not very clear, and it is possible that contributors’ options lay between the first and third measures. 3. In other words, there is a close relationship between the share of first-preference votes won by parties and the share of seats they receive (the 1997 result, shown in table 2, was exceptionally disproportional). As is often pointed out, comparing these two variables does not fully capture the concept of proportionality under PR-STV given that the allocation of seats is (quite validly) affected by the transfer of lower preference votes

as well as by first preferences. 4. Details are given in Part 3 of the Thirteenth Schedule of the General Elections Act, 1991. Page 112 →5. No woman has ever been doubly elected, and, indeed, Malta is notable for the very low number of women MPs elected at all (see Lane 1995). 6. Although this book is entirely in Maltese, it is accessible to those non-Maltese speakers (such as the present writer) who arm themselves with a Maltese-English dictionary. 7. This can be found at . 8. Despite their complete lack of success, those classed as Others in Malta have supplied nine outgoing MPs. Of these, eight were elected for other parties at the last preindependence election in 1962 and were outgoing in 1966; the ninth was elected for the MLP in 1987 but left the party in 1989 and formed a new party, Alternattiva Demokratika, under whose label he stood and lost in 1992. 9. These figures refer to reelected deputies as a percentage of those contesting the current election, that is, the last column in tables 3 and 4. 10. All these figures are calculated by the author. The figures for the comparable cases are very similar, though not always absolutely identical, to those presented by Katz (1980, 75) and Carty (1981, 115). 11. Only two Labour TDs have been ousted by a running mate (in 1943 and 1954), and only six TDs of other parties, the most recent in 1951, have suffered the same fate. 12. Given their rarity value, these five cases could be listed. In 1971 in district 10, Carmelo Refalo (PN) lost to Anglu Camilleri (MLP). This is in fact the only case in the whole data set where the five incumbents stood, under the same label, in the same district and where four were reelected, with the defeated incumbent being replaced by a newcomer of another party. In 1976 in district 7, two MLP incumbents, Paul Carachi and Joseph P Sciberras, were defeated, and they were replaced by an MLP newcomer and a PN newcomer, Antoine Mifsud Bonnici (PN); in other words, there was one intraparty change and one interparty change. Also in 1976 in district 8, Alfred Baldacchino (MLP) was defeated, and two PN newcomers were elected. Although an MLP newcomer was also elected, this is probably best seen as a special case of an interparty change in that Baldacchino had been elected as a PN MP in 1971 but switched to the MLP in 1974, so the PN was merely regaining its seat, as it were. In 1981, Mario Felice (PN) lost his seat in district 9 to Leo Brincat (MLP). The fifth, and the second special case, was referred to earlier (note 8). It concerned Wenzu Mintoff, elected in 1987 in district 2 for the MLP, who left the party and formed the Alternattiva Demokratika. In 1992 he contested district 2 under that label and lost his seat to the MLP. 13. Pedersen’s index is calculated simply by adding the gains of the parties that gained seats and the losses of the parties that lost seats and dividing the result by 2 (see Pedersen 1979). The Maltese average is distorted by the high volatility (15.2 percent) at the first postindependence election; over the next seven elections it averaged only 3.6 percent. 14. In addition, one can become a deputy by winning a by-election in Ireland or a casual election in Malta. We will not consider this further here because the focus is on the outputs of PR-STV. 15. These difficult cases are few in number, especially in Ireland, so even a different classification would not significantly change the overall pattern. Page 113 →16. Victorious nonincumbents are not all new TDs; a minority are former TDs regaining a seat held previously. 17. Of course, it might be argued that such a rule could be evidence of an underlying lack of cohesion in that if the party was really united, it wouldn’t need such a rule to keep its TDs in line. However, it should be borne in mind that the rules of the parliamentary party are made by party parliamentarians themselves, and this rule represents a discipline by which they have chosen to be bound. It is difficult to imagine members of a genuinely disunited party imposing this rule upon themselves. Besides, virtually every European party’s parliamentary caucus has rules outlining sanctions to be imposed upon maverick members. 18. For a slightly different argument that arrives at a similar conclusion, namely, that PR-STV can go hand in hand with a stable party system, see Bowler and Farrell (1991b).

Page 114 →

Candidate Centered but Party Wrapped: Campaigning in Ireland under STV Michael Marsh The STV electoral system in multimember constituencies provides the immediate institutional context for election campaigns. In this system, voters indicate their choice by ranking candidates, who are listed in alphabetical order on the ballot paper, in order of preference. Provided that there are more candidates than there are seats available, all candidates in a constituency are thus put in competition with one another. However, the system also provides for support to be transferred between candidates. For this reason, competition is not absolute, because candidates may benefit from alliances under which some candidates exhort their supporters to express subsequent preferences for other candidates. Hence, although the electoral system provides incentives for candidate-centered campaigns, it also can provide incentives for alliances between candidates. Two elements of the political culture are critical in any understanding of the nature of those campaigns. The first is the existence of partisanship, at least in the sense that the party of a candidate serves as an important voting cue. The second is the existence of localism, under which the local roots and reputation for constituency service of the candidate also provide an important voting cue. It is the weight of these two elements, and the sometimes uneasy interrelationship between them, that gives Irish campaigns their particular character. This chapter begins by examining the evidence for the importance of these two elements, both individually and in combination. It then moves on to show how election campaigns are structured by the need to emphasize both partisanship and localism and to manage the potential conflict between the two. Page 115 →

Partisanship There is no doubt that voters are influenced by the party labels of candidates. (Indeed, given the existence of a fairly stable party system over the past 70 years, it would be remarkable if it were otherwise.) The evidence is quite clear. First, surveys testify to the existence of large numbers of voters who either admit that they think of themselves as being Fianna Fáil (FF), Fine Gael (FG), Labour, or whatever (Marsh 1985) or who admit to feeling close to one or other of the parties (Sinnott 1995). Although such data do not indicate that identification with any particular party is durable, they do testify to the existence of support for parties as such. Two aspects of partisanship can be distinguished. The first, party identification, denotes a long-term emotional identification with a party. The second, party voting, signifies simply that people are voting for a party rather than for candidates. Although the first implies the second, the second does not require the first. Party identification has probably declined in recent years. The Euro-barometer series of data on party attachment indicates that numbers feeling close to parties has declined over the period since 1973 (Schmitt and Holmberg 1995; Mair and Marsh 1999). Nevertheless, the parties themselves remain strong. There is normally relatively little change in the amount of support won by each party from election to election. Volatility, measured by the socalled Pedersen index, is often relatively low (Marsh 1985; Mair 1986), and the interelection constituency-level correlations between first-preference votes cast for candidates of particular parties are high. Table 1 shows the correlations between levels of FF, FG, and Labour support in the 1982–97 period. This indicates that groups of candidates from the same parties, at the national or constituency level, get similar levels of support, regardless of personnel, all of which serves to support the argument that party is an important voting cue. Moreover, those correlations remain at much the same level whether between adjacent elections or between pairs of elections separated by one or two other elections. The average correlation between adjacent elections is 0.73 for FF, 0.79 for FG, and 0.79 for Labour. The average correlation between all pairs of elections in the table is only a little lower: 0.63 for FF, 0.72 for FG, and 0.71 for Labour—not significantly different.

The lack of any general pattern of decay is similar to that observed by Converse (1964) in relation to attitude changes in panel studies. He explains it by positing a model in which a minority remain stable and the rest respond randomly. An alternative interpretation is a response-error model (Aachen 1975), in which each observation only approximates the true position. Each of these models is consistent with the data in table 1, but the latter seems more plausible, especially if we see response error as introduced by short-term influences, such as campaign effects and candidates. Page 116 → TABLE 1. Selected Constituency-Level Correlations in Party Support, 1982–97 Page 117 →The evidence for some decline in party voting comes from transfers. This topic has been dealt with at length elsewhere (Gallagher 1978, 1979; Sinnott, 1995). It is sufficient here to say that transfer patterns indicate that most people who give their first-preference vote to a candidate of one party go on to give their subsequent preference to another candidate from that same party when such candidates are available; that is, they appear to vote for parties rather than candidates. However, the structuring power of the party label is not quite as strong as it has been in previous years, with the party solidarity of transfers within the two largest parties declining noticeably in the last two elections (Gallagher 1993a, 1999). Of course this does not necessarily mean that party is any less important as a determinant of the first-preference vote, but it does indicate that the overall preference set of voters is less determined by party than has previously been the case. In that sense at least, party is less important.

Candidates: Localism and Constituency Service There is also evidence that candidates get a significant personal vote. Although interelection correlations in the levels of party support may be high, there is little sign of any pattern of uniform swing (Gallagher 1990, 1993a). On the contrary, the normal pattern of swing is one of considerable diversity, perhaps a symptom of candidates’ having personal followings that allow them to insulate themselves from adverse national trends or magnify positive ones. Table 2 shows the change in the FF vote at constituency level over the last four pairs of elections and makes it clear that gains in some places are out-weighed by losses in others, even in 1992 when the national vote dropped by more than 4 percent. These nonuniform swings are unlikely to reflect realignments in partisan choice, given the patterns observed in table 1. Again, they are more consistent with nonrecurring short-term influences on the vote, such as variations in the quality of candidates on offer. TABLE 2. Change in Fianna Fáil Constituency-Level Vote Percentage in Adjacent Elections, 1987–97 Page 118 →Survey evidence offers further support for this personal vote thesis. Since 1977 opinion polls have asked voters to choose among various options to indicate the most important influence on their vote. Although such questions have been rightly criticized for failing to separate the choice of a party from the choice of candidates from that party, the answers suggest consistently that the candidate matters. Yet what is it about the candidate that matters? A multivariate analysis of candidate performance at elections between 1948 and 1982 indicated that although party was a major factor, the political experience and personal characteristics of candidates also carried considerable weight (Marsh 1987). Important characteristics are generally those most consistent with a demand for deputies who can fulfill a brokerage role —that is, those who can get things done on behalf of individuals and the constituency. Further evidence suggests this role should be played on behalf of people in a specific locality. As has been said already, transfer patterns generally testify to the structuring power of party, but to the extent that transfers do not follow party lines, they suggest that some people are using a different decision heuristic, such as locality within the constituency. One study has shown a small but significant effect of locality on transfers (Marsh 1981). Moreover, several studies have identified the importance of locale on a candidate’s firstpreference vote (Sacks 1976; Parker 1984). Such a preference may be put down to familiarity but also to a desire that a deputy should represent a particular area within the constituency rather than the constituency as a whole (or more likely some other area within it). The sort of candidate vote suggested by the evidence assembled to date is a combination of the partisan and the personal. This assessment is not simply an analytic distinction made by academics but also the conventional wisdom of political organizers and candidates, and thus it influences the election campaigns they become involved

in. These campaigns have three elements. First, there is the national, essentially partisan campaign, run by party headquarters and senior party figures, that seeks (subject perhaps to some constraints) to maximize the party’s vote. Secondly, there is a nationally directed campaign in the constituencies. This seeks to organize the constituency campaigns with a view to maximizing the party’s representation in the Dáil. Thirdly, there is the local campaign itself, where the party is brought into direct contact with the voter through the candidates and the candidate employs his or her personal attributes to the full. Page 119 →

The National Campaign It is sometimes suggested that national campaigns are a relatively new phenomenon. Mair (1988), for instance, argues that “none of the parties ran full-blooded and intensive campaigns in the 1950s and 1960s” (111). Manifestos were considered unnecessary, and if they were drawn up, this was done at the last minute, and such proposals could be ignored by the local campaigns.1 This can be attributed to two things. First, the parties adhered to a fatalistic view about party loyalties that far exceeds anything claimed by adherents of the Michigan School of electoral research.2 Second, the parties lacked the organizational or financial resources to run such a campaign even if they wished to do so. Thus, Chubb (1970) comments that the small and comparatively underdeveloped resources of central offices left candidates dependent on their own efforts. If partisanship is seen as conditional, then parties need to make efforts to persuade people to their side. Fianna Fáil appeared to come round to this view following an electoral defeat in 1973 that ended 16 years of unbroken rule (for a discussion of the factors underlying such change, see Mair and Marsh 1999). The party put considerable national effort into its 1977 campaign (Farrell and Manning 1978). The apparent success of its strategy, which secured for the party an overall majority of votes, provoked other parties to follow suit, and FG gained a huge increase in its vote at the next election in 1981. Henceforth, the national organizations of all parties have proselytized to the limit of their resources (and often beyond). The contrast between the style of modern and traditional campaigns might be overstretched. For instance, writing rather earlier, Chubb (1970) notes that national campaigns were waged by leaders and through posters, with the leaders making tours of the country and major rallies in regional centers serving as centerpieces for the campaigns. Gallagher (1981) argues that subjects such as inflation and employment have always been staples of political argument, “politicians being loath to leave idle any stick with which their opponents might be beaten. . . . The real issues of Irish politics during the 1960s and 1970s, as during the 1950s, were relatively trivial ones like competence, credibility, personalities and the styles of political leaders” (272–73). Moreover, parties also issued programs and plans. In the absence of any systematic study of campaign rhetoric over the years, it is hard to say how different in content are modern campaigns. However, there is no doubt that national campaigns are now much more professionalized. They tend to be planned well in advance; employ marketing and public relations consultants to advise on presentation, Page 120 →brand image, and areas of the electorate to target; and make considerable use of advertising, especially in newspapers and billboards and even cinemas. Although every effort is also made to use television and radio to best advantage, the legal restrictions on access mean that parties must contend with interviewers, presenters and rival politicians to get their message across. The party leader and other senior figures travel the country, encouraging local activists but also providing photo opportunities for the national campaign. To some degree, the nature of the national effort depends on finance. In the absence of laws on the financing of political parties until 1997 (Laver and Marsh 1999), there was little hard data on what parties have raised and from whom. All parties combined spent around £2 million in 1992 and £2.5 million in 1997 (Holmes 1999), about 80 pence per potential voter. Major parties have accumulated significant debts as expenditure has become more substantial in recent years. A degree of direct public finding and controls at future elections may make life a little easier for the parties but will not lead to a new influx of cash to fund ever more expensive national campaigns. In fact, the limitations of what parties spend could actually serve to depress expenditure. Although all campaigns seek to mobilize the faithful, campaigns vary in respect to the structure of the competition

between parties. One factor determining the nature of competition is the set of alternatives for government. Sometimes coalition deals are signed prior to the election, and parties recommend their supporters to give lower preferences to coalition partners. This was done in 1973, for instance, when Labour and FG agreed terms for government. In 1982, the agreement was more tentative, but for both Labour and FG the main adversary was FF. As long as governments were either single-party FF or FG/Labour coalitions, this pattern of competition was natural, at least at the national level. Indeed, such adversarial competition between the two alternative governments fits the Westminster tradition of Irish politics. However, in the last few elections the lines of competition have sometimes been more complex. Fianna Fáil’s decision to make itself available for coalition, first with the Progressive Democrats in 1989 and then Labour in 1992, has sundered the traditional mold, notwithstanding the fact that the 1997 election was for the first time fought between two alternative coalition governments. Such multilateral competition has so far not appeared to make for less adversarial campaigning.3 Fianna Fáil ran a special series of advertisements attacking Labour in the last days of the 1992 campaign yet then agreed a program for government with them afterwards. However, if the current open structure of competition continues, it will be interesting to see whether this makes for less adversarial campaigning in the long run. Page 121 →An alternative reason for particular competitive strategies is electoral. Parties can focus their campaigns on parties with which they share a significant potential vote. Recent research has suggested that wide scope exists for such competition in the Irish electorate as relatively high proportions of voters express a willingness to support more than one party. The figures in table 3 come from the European election study of 1994 (van der Eijk and Franklin 1996) and are based on responses to a question that probes the likelihood that a voter will ever cast a first-preference vote for a particular party.4 Each column of the table shows the extent to which the potential electorate of a party—those who say they would consider voting for that party—is shared with other parties. So for instance, 16 percent of FF’s potential support is shared with Sinn Fein, 45 percent with the Progressive Democrats (PDs), and 56 percent with FG. What is most striking here is the extent to which there is a wide overlap in the electorates of all major parties, including FF and FG. Possibly the electoral system contributes to this. However, it suggests that governmental consideration rather than electoral considerations have guided patterns of competition at the national level. If electoral considerations were the guide, competition would simply be all-against-all.

National Management of Local Campaigns The national party has a strong interest in ensuring that local campaigns are efficiently run, that is, run to maximize the numbers of party deputies elected. This does not necessarily involve ensuring that the local campaign is run in strict accordance with the national one. It does require a high-quality field of party candidates as well as that the party fields the right number of candidates and that those candidates organize their joint campaigning so as to maximize the number of seats won. The problem for national parties is that in trying to ensure these things and failing, things may be made worse than they otherwise would have been. TABLE 3. Potential Support and Overlap between the Parties, 1994 Page 122 →In any electoral system involving local constituencies, it may well be that national messages need to be adapted slightly for local consumption. As campaigns become nationalized in coverage, it is risky for parties to make different promises in different places for fear of some inconsistency being exposed on prime-time television. There can certainly be differences of emphasis, but it may be something of a fine line between the two things. National parties make posters, advertisements, and general campaign literature available to the local campaigns, so providing a common look and feel to the enterprise. However, what is more important is the management of the personnel. The problem for parties in this respect is the difference of interest between local candidates and the national party on some matters and the fact that decisions on candidate selection and tactics are traditionally made locally (Marsh 1981; Gallagher 1988). First, there may be differences on the matter of how many candidates should be run (Cohan, McKinlay, and Mughan 1975). The national party will want to nominate at least as many candidates as there are seats it might win. However, the assessment of likely vote is never certain. In a case where the party

has one seat and thinks it might win two, it will wish to nominate a second candidate. The incumbent deputy might be happy to take a running mate where two seats are certain but may fear to do so where they are not, because two people would then be fighting for the same seat and the incumbent might lose what seemed to be a sure thing. Given, too, that local party organizations are dominated commonly by supporters of the incumbent, only one candidate may be nominated in the absence of the national party’s input. National parties have sought to impose their will on local parties in such situations. Fine Gael, for instance, under FitzGerald in the early 1980s, sought to impose a running mate on all “quota sitting” TDs in the hope of extra seats. Garvin (1970) suggests that FF had traditionally been better than rival parties at managing its candidacies. His evidence is that FF had achieved a better seats/candidacies ratio than the others. As table 4 indicates, this remains true when analysis is done of the 1948–97 era with FF’s mean ratio at 1.60 much better than FG’s 2.03 (t-value = 4.67; p |.0001). However, there is evidence that FG has tightened up, bringing its ratio much closer to that of FF, supporting the argument that nominations have been under stricter control since that time. The 1981–97 mean for FG is 1.79, lower than the 2.21 for 1948–77 (t-value = 3.04; p