Is there Anything Valuable?

Is there Anything Valuable? A Spatial Proximity Model of Party Support for Electoral Reform and its Empirical Relevance across OECD Countries Damien Bol Draft paper, please do not quote without author s permission Abstract In the existing literature, values and self-interests are often opposed to explain parties support for electoral reform (Renwick, 2010). However, their respective empirical relevance has never been properly evaluated since a theoretical model of the influence of values in party preference for electoral reform is still missing. To fill this gap, this article first constructs a spatial proximity model of party support for more inclusive electoral system. Its empirical relevance is then tested and confronted to the seat-maximization model developed by Benoit (2004), which captures self-interested motivations, with the help of an unique comparative dataset of the electoral reform proposals of OECD countries since 1961. Maximum-likelihood estimations with randomeffect corrections show that the left-right ideological scale is even a better predictor of party support for more inclusive electoral system than the expected seat payoff. Previous versions of this article have been presented at various international conferences (EPOP 2010, ECPR 2010, CES 2011) and at internal seminars of the Université de Louvain, the European University Institute, and the Université de Montréal. Thanks to all the participants for their useful comments, and in particular to Pedro Riera who fully reviewed this paper a countless number of times. Université de Louvain and Belgian Scientific Foundation. 1

Values and self-interests in electoral reform The academic community seems to unanimously consider that parties are first and foremost strategic players behaving in a strongly self-interested manner (see for example the recent literature survey made by Adams, Merrill III and Grofman (2005)). Thus, it does not come as a surprise to observe that the exact same assumption is very frequently taken to explain party support for electoral reform (see the recent literature survey made by Rahat (2011) for instance). Following this perspective, Benoit (2004) developed a baseline theoretical model of electoral system choice that constitutes the starting point of many researches on the subject. It states that parties rank alternatives according to their respective expected seat payoff compared to the status-quo and then opt for the preferred one. Along the same lines, he argued that an electoral reform occurs when the parties that have the same preference are able to secure together a majority of seats within the parliamentary assembly that is in charge of voting on electoral reforms. More formally, considering s, the status-quo electoral system, and p, the proposed electoral system, the utility of the party i to support p (U ip ) is a function of its seat share in parliament (S is ) and of its expected seat share if p is implemented (S e ip ). U ip = f(s e ip, S is ) = S e ip S is (1) In parallel, some authors, although much fewer, pointed out that the behaviors of some political parties involved in electoral reform processes cannot be explained entirely by their expect gain in particular, or self-interested motivations in general. They typically argued that parties may also well support a change of electoral system because of the values they defend, or in other words, because they do honestly think a reform would be better for the people they represent. Pushing this idea further, this latter perspective sees electoral reforms as policies that are implemented, just as other economic or social policies, to achieve social goals considered as relevant. Using Strom s (1990) typology, these authors consider parties as policy-seeking actors that first and foremost seek influence over state policies. More precisely, values have been mobilized to explain the evolution of party supports for electoral reform in countries engaged in political turmoils, such as Germany during institution-building period following 1945 (Scarrow, 2001), Belgium during the democratic renewal that occurred after 1999 earthquake elections that see the Christian-democratic parties loosing their dominant position (Pilet, 2007), the United Kingdom during the expenses scandals of the mid-1990s (Renwick, Lamb and Numan, 2011), in Eastern European countries during the democratization period following the collapse of the Soviet Union (Birch et al., 2002; Renwick, 2011b), or in West European countries during the rise of Socialist parties of the first half of the 20th century (Carstairs, 1980; Penades, 2008). The most comprehensive research in this line of thinking have been certainly achieved by Renwick who traced back the behavior of all relevant political actors involved in the debate surrounding electoral reform in France, Italy, Japan and New Zealand since 1945 2

(Renwick, 2010). Although insightful, these contributions remain either single or multiple case-studies and their subtle conclusions are thus hardly generalizable beyond the idea that values do play a role in electoral reforms. Some authors did however conduct genuine comparative analyses of party support for electoral reform. Strikingly though, they all focused one critical period in world history, namely the political liberalization and suffrage extension that occurred in many industrial democracies at the beginning of the 20th century. They thus derived context-specific explanations of the role played by values in electoral reform in stating that they took the form of a willingness of parties to adapt to a changing environment being at level of political liberties (Blais, Dobrzynska and Indriadason, 2005), economical structures (Cusack, Iversen and Soskice, 2007), or partisan competition (Calvo, 2009). To date, only one article has assessed the impact of values on party support for electoral reform freed from interferences brought by specific contexts. Relying on elite survey data of 4 modern democracies, Bowler, Donovan and Karp (2006) has examined the attitude of individual politicians concerning a change of the electoral system in their country. More specifically, just as this article, these authored assessed the empirical relevance of various motivations including values and self-interests. Their results show that politicians are first and foremost driven by their willingness of being reelected although all things being considered, the more left-wing they are, the more they tend to support electoral reform. However, the article lacks of an important consideration concerning the direction of the electoral reform. Yet, there are good reasons to think that this may increase the extent of the impact of values on party support for electoral reform. There is indeed no general agreement among specialists about a potential ideal electoral system that would be also preferable. They all have pros and cons (Bowler, Farrell and Pettitt, 2005). For example, while List Proportional representation (List PR) ensures that each social class or each minority group is represented in parliament, it tends to produce multi-party governments that undermine the capacity of voters to sanction bad politicians. Depending on one s criteria, being the governability, the inclusiveness representation, or whatsoever, a different electoral system should be supported. A spatial proximity model In order to construct a general model capturing the influence of values on party support for electoral reform, the well-known Downsian spatial proximity model of vote is adapted. All the ingredients are indeed present to do so. As stated above, besides being the rules of the games of political competition, electoral reform can be seen as a policy instrument aiming at achieving social goals. It is indeed known for a long time that the type of electoral systems in use have dramatic consequences for the conduct of politics and, more generally, for the society as a whole. From Mill (1861) to Schumpeter (1942), nearly all the classical authors on democracy have pointed out this causal relationship, while the first to give it scientific grounds was Duverger (1954). The idea of spatial proximity can thus be applied to parties to modelize their support for the status-quo or the proposed electoral system, in the sense that they will opt for the alternative that is the closest from 3

their policy position. Once gain it can be expressed formally, considering s the status-quo electoral system, p the proposed electoral system and a policy space with a single social goal that can be achieved by the electoral system G, the utility of the party i to support p (U ip ) depends on the distances between the willingness of i to achieve this social goal and the capacity of s and p to do so ((G s G i ) and (G p G i )). U ip = f(g i, G s, G p ) = (G s G i ) 2 (G p G i ) 2 (2) Defined this way, it appears that self-interests is never entirely absent from party behavior, even when it is driven by so-called values. For party, acting in line with what it has been elected for- ie. the values emphasized in its manifesto -and presenting itself as trustworthy and consistent is certainly the best option to maximize its vote share, and in turn its seat share, at next elections 1. Therefore, values on the one hand, and self-interests on the other may not be the most appropriate labels. Along this line of thinking, some authors has referred to similar motivations using the terms act-contingent and outcome-contingent (Shugart, 2008). According to them, act-contingent stands for party motivations that are driven by gains expected from the very fact of supporting electoral reforms, and are thus in line with the spatial proximity model presented above. To the contrary, motivations are said to be outcome-contingent when the expected gains are brought by the mechanical and psychological effects of the electoral system itself. They thus somehow corresponds to Benoit s (2004) seat-maximization model. For the sake of coherence and clarity, the terms values and self-interests are however used throughout this article to refer respectively to act-contingent an outcome-contingent motivations. The main challenge for assessing the empirical implications of this theoretical model consists in defining G, a social goal that is known to be potentially achieved by a change of electoral system. Also, in order to measure G i, it should echo a minimally salient issue of the political agenda. If it was defined as, say, improving sincere voting, it would be hard to know how much parties want to achieve it, since it is generally not very publicized. In this perspective, the focus is put on what is probably the most scrutinized social goal in the literature on electoral system: the inclusiveness of political representation, or in other words, the representation of citizens from all social groups in parliament including minorities and underprivileged people. This goal has the advantage to be the object of a consensus among authors, and to be so for a long time, to say that the electoral system is crucial if one wants to achieve it (see for example, Duverger (1954), Rae (1967), Lakeman (1976), or Lijphart and Grofman (1984)). Besides, producing an inclusive representation is not as consensual as it seems. The typical way to ensure it is indeed to lower down the cost of entry in parliament to allow all social groups to get represented, which in 1 It is worth noting that the Downsian spatial proximity model of vote also implicitly assumes similar party strategies. Since voters made their choice exclusively based on party policy position, they are supposed to act consistently if they want to be re-elected at next elections. 4

turn decreases the chances to produce a single-party government that also undermines the strength of the governments and its ability to conduct efficient policies (Powell, 2000). Putting it crudely, it is said in the literature that PR produces a very low entry cost for new electoral contestants and in consequence more inclusive political representation than the First-past-the-post electoral system (FPTP), or more generally than all types of majoritarian and pluralitarian electoral systems 2. The picture is however more complex as many systems falls in-between these two poles such as Mixed-member electoral systems (MM). Besides, as the proposed electoral system is often rather loosely defined, let just think about the proposal of introducing some bits of proportionality 3 into the French Two-Round System (TRS) made by Fillon s commission in France in 2007, it cannot usually be calculated with precision. Also, approximating the entire term is not an easy task since party willingness to achieve a social goal and the capacity of an electoral system to do is hardly comparable. Therefore, further developments of the spatial proximity model need to be operated. In particular, the difference between G s and G p should be emphasized. U ip = (G 2 s G 2 i + 2G s G i ) (G 2 p G 2 i + 2G p G i ) = (G 2 s + 2G s G i ) (G 2 p + 2G p G i ) = (G 2 s G 2 p) + 2G i (G s G p ) (3) In isolating the term G s G p, the empirical implications of the model gets easier to derive. The sign of this difference, positive if the status-quo is more inclusive than the proposed electoral system and negative otherwise, can indeed be approximated with a limited amount of case-knowledge (see below), and even if the proposal is loosely defined 4. Further, this sign can be fixed so that of the utility of party i of supporting electoral reform is transformed into the utility of supporting the more inclusive electoral system 5 among the proposal and the status-quo (U ig ). This utility is then defined as a function of the party willingness to implement inclusive political representation G i and of whether the more inclusive electoral system is the status-quo or the proposal D g. This additional parameter of the direction of the more inclusive electoral system is indeed required to account for the specific role played by the status-quo in spatial proximity models (Grofman, 1985). U ig = f(g i, D g ) = 2G i + D g (4) 2 As a general rule, PR lowers down the entry cost of parliament for all social groups at the exceptions of geographically concentrated ethnic or linguistic minorities that may be better of under majoritarian and pluralitarian electoral systems (De Winter and Tursan, 1998). 3 Author s personal translation of the French une dose de proportionalité. This does not make more sense in the original language though. 4 The signs of G s G p and G 2 s G 2 p are always identical since the transformation of the two constitutive terms- ie. the square function -is also identical. 5 The term more inclusive electoral system stands for the electoral system that is likely to produce the more inclusive political representation. 5

To confront this adapted spatial proximity model to Benoit s (2004) seat-maximization model, some transformations needs to be made. As a reminder, Benoit stated that the utility of party j to support a proposed electoral system U ip is a function of the expected seat payoffs compared to the status-quo S e ip S is. This model should be also express in terms of the utility of the party i to support the more inclusive electoral system among the status quo and the proposal (U ig ). As suggested by Colomer (2004), when the direction of the reform is towards more inclusiveness, the expected seat share of a given electoral reform proposal is a negative linear function of the the proportion of seat received at the preceding elections. The rationale behind this idea is that, under exclusive electoral systems, small parties are supposed to be afraid not to win any parliamentary seat while large parties expect to obtain these precise seats as a supplement. Therefore, the utility of the party i to support the more inclusive electoral system among the status-quo and the proposal is a function of its seat share at the latest elections S is and, just as for the spatial proximity model, of the direction of the more inclusive electoral system (D g ). U ig = f(s is, D g ) Measurements and hypotheses = S is S is + D g = 2S is + D g (5) As previously mentioned, the adapted spatial proximity model is focused on the party support for more inclusiveness in political representation. The central parameter is G i, which corresponds to how much the party i wants to achieve a particular social goal. Depending on how it is defined, several hypotheses can be derived. The first one relates to party favorable position towards social groups such as minorities or underprivileged citizens is positively associated to their support to more inclusive electoral system. The best way to ensure that they are politically considered is indeed to include them into the decision-making process, and therefore to lower the entry costs in parliament. To operationalize party position towards social groups, the data gathered by the Manifesto Research Group (MRG) (?) are computed through the scaling method developed by Lowe et al. (2011) that allows creating aggregated indicators of party position that are both more reliable and more in line with recent linguistic researches on salience and positioning. They suggested to compute the logged ratios of favorable and unfavorable MRG items relating to an aggregated concept 6. In this article, the 6 items directly relate to this issue are selected, together with an extra items concerning political authority. As an inclusive political representation also has the consequence of undermining the government s strength and its ability to conduct efficient policies. The table 2.1 reports the description of these items and the direction they take in the aggregated indicator. The following hypothesis can be derived: 6 The exact computation of party position developed by Lowe et al. (2011) is log( positive items + 0.5) log( negative items + 0.5). 6

Table 1: MRG items selected to construct the aggregated indicators of party (favorable) attitude towards social groups Favorable items Unfavorable items Labor Groups: Positive Labor Groups: Negative Favorable reference to labour groups, working class, unemployed, Abuse of power by trade unions; support for trade unions, and good treatment of employees otherwise as the positive item, but negative Farmers: Positive Political Authority: Positive Support for agriculture and farmers Favorable mention of strong government, any policy aimed specifically at benefiting these including government stability, and manifesto party s competence to govern, Middle Class and Professional Groups: Positive and/or other party s lack of such competence Favorable reference to middle class, professional groups such as physicians or lawyers ; old and new middle class Underprivileged Minority Groups: Positive Favorable reference to underprivileged minorities who are defined neither in economic nor in demographic terms e.g. the handicapped, homosexuals, immigrants, etc Non-economic Demographic Groups: Positive Favorable mention of, or need for, assistance to women, the elderly, young people, linguistic groups, etc ; special interest groups of all kinds 7

H1: Social groups. The more favorable the position towards social groups, the more likely its support to more inclusive electoral system Another set of hypotheses can be derived with the help of ideologies. Formally, ideologies relate to number of values that are articulated together to form a coherent pool of norms on how the society should be organized. According to political theory literature, they are directly linked to electoral systems (Katz, 1997). Each of them indeed corresponds to a specific vision of what is a good governance, and as a consequence, of what is a good way of organizing elections. In this sense, the left-wing ideology- ie. socialism -, promotes the inclusion of all the citizens in the decision-making process as a prerequisite of the ultimate goal of equality of treatment and of chances. In this perspective, left-wing parties should promote more inclusive electoral system to ensure that each vote is counted in the final election s results, and that each social class is represented in parliament. On the opposite, the right-wing ideology- ie. liberalism -put efficiency and governability at the forefront of the collective decision-making. For the reason stated above, right-wing ideology thus entails the promotion of less inclusiveness (Schumpeter, 1942). To operationalize party ideology, two different measures of the MRG dataset are chosen. First, the items selected by Laver and Budge (1992) to capture party position along the left-right scale broadly are taken and then computed through Lowe, Benoit, Mikhaylov, and Slava s (2011) scaling method. These items ranges from market regulation to anticolonialism or law and order. Second, a formal measure of traditional party families is used. In line with the argumentation made above, a multinomial variable differentiating a right-wing (made of conservative, christian and radical populist parties) and a left-wing family (composed by the Socialists, the Communists, and the Green) is constructed, all other parties serving as reference 7. The following hypothesis can thus be tested: H2A: Ideologies (scale). The more left-wing the party, the more likely its support to more inclusive electoral system H2B: Ideologies (family). Socialist and communist parties are more likely to support more inclusive electoral system Finally, to assess the empirical relevance of values and self-interests on party support for electoral reform, a last hypothesis needs to be derived from the transformed seat-maximization model. This operation is straightforward and does not deserve long explanations. In line with the model presented above, it can be measured by the overall seat share of the party at the preceding elections (T 1 ), as an indicators of their expected seat payoffs to the more inclusive electoral system among the status-quo and the proposal. H3: Seat-maximization. The smaller the seat share of the party, the more likely its support to more inclusive electoral system 7 This typology is a classic of the studies on party competition, see for example Adams et al. (2006). 8

Data To test these hypotheses, an unique comparative dataset of party support for various electoral reform proposals has been gathered. This dataset is superior to other comparative datasets used in the existing literature 8 for two reasons. First, the units of analysis are the individual parties and not the governments. The politics of coalition is indeed said to sometimes to be relevant for the shaping of party support for electoral reform. For example, Renwick, Hanretty and Hine (2009), who analyzed the 2005 Italian reform, pointed out that the support of Forza Italia -ie. the biggest party of the ruling coalition- to the re-implementation some sorts of PR for the election of the lower house of the national parliament may be explained its willingness to secure its leading position into the rightwing coalition, which was at that time threatened by its ally Alleanza Nazionale. Second, in only considering electoral reforms that have been eventually adopted, the comparative datasets used in the existing arbitrary exclude many relevant cases. Many failed attempts of electoral reforms are indeed blocked at the very last stage of the decision-making process by non-partisan actors. For example, in 2011, the proposal to adopt Alternative Voting (AV) for the election of the British House of Common was blocked by the population itself that rejected the proposal. Yet to make sure these supports remain measurable, three restrictive case selection criteria are taken. First, only OECD countries are scrutinized, yet for the entire period of their membership to date. Because the international organization put strong conditions concerning political and economic liberties for the accession of its membership, it defines a set of homogenous countries both in terms of democratic functioning and the dimensions of political competition. These two aspects are indeed baseline criteria for the various models presented above. Second, only proposals that have been at one point on the political agenda of the government either because they were drafted by a committee specifically appointed by the government, be the subject to a referendum, or submitted to parliament by (one of the) government parties are included. As a matter of fact, electoral reforms are technical issues that in most cases do not attract much of the media attention, and which in turn do not give many incentives for politicians to express any kind of support. For example, the proposal to add an upper tier of representation to the Norway PR electoral system, which was drafted by the the Committee for Foreign and Constitutional Affairs in 1988 remains largely absent from the public agenda, and politicians in consequence, did not make any public intervention on the subject (Aardal, 2002). Because the support of parties for these minor proposals are unknown, they cannot be included in the analysis even if they would have surely brought dramatic consequences for the conduct of politics, and of the society as a whole. Third, the focus is strictly put on wholesale replacements of the electoral formula in use for the election of the lower house of the parliament (Katz, 2005). In other words, only the 8 Comparative datasets have been for example used by Boix (1999), Andrews and Jackman (2005), Colomer (2005), Cusack, Iversen and Soskice (2007), Shugart (2008), or Calvo (2009). 9

Table 2: Electoral reform proposals in OECD countries Country Year Status-quo Proposal Outcome Inclusive Austria 1989 List PR MM (Comb.) Failure Less Belgium 2001 List PR MM (Comb.) Failure Less Canada 1979 FPTP MM (Comb.) Failure More Canada 2004 FPTP MM (Comb.) Failure More Czech Rep. 2001 List PR MM (Comb.) Failure Less France 2007 TRS MM (Ind.) Failure More France 1985 TRS List PR Success More France 1986 List PR TRS Success Less Hungary 2011 TRS FPTP Success Less Ireland 1968 STV FPTP Failure Less Ireland 1996 STV MM (Comb.) Failure Less Italy 1993 List PR MM (Comb.) Success Less Italy 2005 MM (Comb.) List PR+ Success More Japan 1994 SNTV MM (Ind.) Success More Mexico 1996 MM (Ind.) MM (Comb.) Success More Netherlands 2002 List PR MM (Comb.) Failure Less New Zealand 1986 FPTP MM (Comb.) Failure More New Zealand 1993 FPTP MM (Comb.) Success More New Zealand 2011 MM (Comb.) AV Failure Less Portugal 2008 List PR MM (Comb.) Success Less South Korea 2003 FPTP MM (Ind.) Success More United Kin. 1997 FPTP MM (Comb.) Failure More United Kin. 2011 FPTP AV Failure More proposed switches from one type of electoral system to another are considered 9. As they directly relate to the principle of representation at stake in the political system (Nohlen, 1984), these proposals are more likely to attract the attention of the general public. It is worth noting that this definition is slightly more restrictive than the one chosen by Lijphart (1994), which also includes changes in the ballot structure, introductions of an electoral threshold and modifications in district magnitudes. To identify the electoral reform proposals fitting these criteria, the relevant pieces of the existing literature, and in particular the numerous case-studies contained in edited books, have been surveyed. When missing, this information was completed by direct contacts with national experts 10. 23 proposals that cover 14 countries and more than 40 years have been found. The table 3.2 gives a brief summary in reporting the year when the proposals reached the political agenda of the government, the type of the status-quo and the proposed electoral system, and the wether the process succeeded or not. The table 2.2 also includes the binary approximation of the more or less inclusive character of the electoral reform proposals needed to fix the sign of the equation and then 9 To classify electoral systems, the typology of Golder (2004) has been used. 10 The full list of the pieces of literature and the national experts surveyed can obtained upon request to the author. Usual disclaimer applies. 10

estimate the models. While it is straightforward to label the switches from some sorts of PR (List PR, Single Transferable Vote (STV), or even combined or independent MM) to a pluralitarian or majoritarian electoral system (FPTP, AV, or TRS) as less inclusive and vice-versa, the picture is often more complex. In particular, many proposals consist in a change within these two overall families of electoral systems, and thus required more in-depth analysis. Foremost, since the proportional component usually fully compensate for the entire disproportionality brought by the pluralitarian or majoritarian component, combined MM are as inclusive as other PR electoral systems. However, in such systems, the so-called psychological coordination effect between parties has been said to increase the cost of entry in parliament (Ferrara and Erron, 2005). Switching from List PR or STV to any kind of MM is then labelled as a move towards less inclusiveness. Likely, since the PR component of independent MM does not compensate for anything, this electoral system should be considered as more inclusive than FPTP, AV, and TRS yet less inclusive than all sorts of PR including combined MM. Then, the effect of a switch from a pluralitarian to a majoritarian electoral system on inclusiveness is tricky. Examining the political consequences of the potential introduction of AV in the United Kingdom in 2011, Renwick (2011a) argued that it would not make life easier for small parties, meaning not the Labour, the Tories, and the LibDem, in perspective of entering parliament. However, the very possibility offered by AV to rank the parties by order of preference would have cancelled out the fear of wasting one s vote, and would have therefore given those parties much much bargaining power. In the sense, the proposal to replace FPTP by AV is interpreted as a change towards a more inclusive electoral system. The same applies to the bill introduced in 2011 by the Hungarian ruling party- ie. Fidesz -that replaced the majoritarian component of the combined MM by FPTP. Finally, among the proposals identified, two electoral systems are rather exceptional and are thus hard to assess. On one side, the Japanese Single Not Transferable Vote (SNTV) constitute a very special case. Although highly proportional in the translation of votes into seats, this electoral system was said to be very exclusive when one bigger party manages to coordinate in the various multi-member constituencies and in consequence to block the entry to parliament to all other parties. Hence, the Liberal Democratic Party had been able to secure a comfortable majority of seat in the lower house of the Japanese parliament for about 50 years following 1945 (Cox, 1996). Its replacement by an independent MM in 1993 is therefore labelled as more inclusive. On the other side, the List PR implemented in Italy in 2005 was also very particular. It gives extra seats to the coalition of parties with the greater number of votes until it benefits from a majority of support in parliament. This electoral system is therefore usually assimilated to the pluralitarian or majoritarian family as it put high entry cost in parliament to parties not in one the two main coalitions (Baldini, 2011). Party positions for these electoral reform proposals have been coded using a binary coding-frame, or in other words, wether they were in favor or against the proposal 11. The 11 As a principle, all parties represented in the lower house of the parliament when the proposal reached 11

procedure followed has been a mixed-strategy between qualitative hand-coding based on the existing literature just as for the identification of the proposal and a cross-validation of this pre-coding made by national experts. Proceeding this way allows addressing the critiques made by Budge (2000) concerning expert party positioning. In particular, it has been clearly stated to the experts that the information required was about a very specific behavior- ie. the support for an identified proposal -and not about abstract attitudes and at a precise moment in time. Also, it was made clear that the research is centered on party leaders and not on backbenchers, nor on activists. In turn, among the 119 positions coded, 67 are in favor of a more inclusive electoral system. Empirical analysis To confront values and self-interests in party support for electoral reform, the three hypotheses directly derived from the spatial proximity and the seat-maximization models are tested on an unique comparative dataset on electoral reform proposals of OECD countries presented just above. In doing so, the parameters of the two theoretical models will be estimated and the empirical relevance of each type of motivation assessed. To do so, a logit maximum-likelihood estimation is conducted. This specific technique is chosen since the predicted variable- ie. party support for more inclusive electoral system -is binary. Also, because the multi-level nature of the dataset presented above creates a significant risk of drawing incorrect inference, a specific correction term is included. Since the upper clusters (k) are not chosen on an ad-hoc basis but represents all the electoral reform proposals in OECD countries to date, this correction takes the form of a random error term (µ k ), which aims at absorbing potential effects that are not specified in the model at the upper level, just as ɛ ki does at the lower level 12. The estimated models thus follow the log of the following equation. Party support for more inclusive electoral system = β 0 + β 1 Standardized hypotheses variables + β 2 Direction + β 3 In office at T 1 + (6) µ k + ɛ ki The variables corresponds to the parameters of the spatial proximity and the seatmaximization model as presented above 13, at the exception of In at office T 1. This the government s agenda following the criteria presented above have been included in the analysis. For practical reasons mostly the difficulty of finding reliable information on their position some minor parties were excluded such as the Front National in Belgium, the Leefbaar Nederlands in the Netherlands. For similar reasons of data availability, the Mexican parties are not included in the analysis. 12 The Hausman test shows that there are no significant difference between fixed- and random-effect estimations. In this case, and as suggested by Beck and Katz (1995), the random-effect estimations, which is more efficient, should be preferred. The results of the Hausman are available upon request to the author. 13 As the central parameter of both theoretical models is timed by 2, this multiplier has been be simplified. 12

dummy variable reports wether the party was in office or in the opposition when the electoral reform proposal first reached the government s agenda and thus accounts for the distorting effect of party supports that were motivated by an agreement on a broader package deal, as it happened for example for the implementation of the direct election of the Prime Minister in Israel in 1993 (Rahat, 2008) 14. In order to compare the empirical relevance of the four hypotheses variables, these variables need to be standardized. However, since the party positions computed through the scaling method of Lowe et al. (2011) are not bounded, there is no other option but to subtract their mean to their actual value and to divide this difference by their standard deviation. These standardized variables share thus a mean of 0 and a standard deviation of 1. Although it complicates the final interpretation, it allows the estimated coefficients to be comparable measures of the empirical relevance of each type of motivation. Their distribution is given by the figure 2.1. The boxplots reports, respectively from the top to the bottom, the maximum, the 3rd quartile, the median, the 1st quartile and the minimum. Because the variables are standardized, their distribution is obviously similar. Hence the figure is useful to point at outliers. In line with conventions, the circles represents the cases with a value for this variable that is higher or equal to 1.5 of the interquartile range. One party engaged in the 1986 electoral reform process in New Zealand- ie. the National Party -and three parties facing the 1993 proposal to adopt a combined MM for the election of the Italian lower house- ie. the Lista Panella, the Social Movement-Right National, and the Green Federation- show an extremely negative attitude towards social groups. While the first is a conventional, although radical, party in the New Zealand two-party system at this time, the three others are very small and rather unstable parties. The Social Movement-Right National for example publicly demanded to come back to the fascist regime that governed Italy from 1922 to 1945. Not surprisingly then, this party, together with its objective ally at that time, the Republican Party, is also an outlier in the left-right scale. At the other pole of the spectrum, two Japanese left-wing radical parties - ie. the Clean Government Party and the Social Democratic Federation- can also be considered as extreme cases. The distribution of the standardized vote share is, as it, well-balanced as no case has a value of more than 1.5 of the interquartile range. However, due to the low number of cases, a specific attention deserve to be brought to these outliers. In particular, the same analyses should be conducted with and without them as they may artificially inflate or deflate the estimated coefficients. The table 2.3 reports the estimated coefficients of the various parameters of the spatial proximity models when all the cases are included. They give empirical evidence to the theoretical assumptions. First, all the coefficient of the standardized hypotheses variables have the expected sign, at the exception of belonging eft-wing party family that has a negative effect on party support for more inclusive electoral system. Second, all the diagnostics show rather satisfying goodness-of-fit, and in particular a degree of significancy 14 Many other specifications have been estimated. Including other potentially influent variables such as the party position towards decentralization, since small ethno-regionalist parties may sometimes have more seats under not-inclusive electoral systems, or previous although recent supports for more inclusive electoral system do not change the results. 13

Figure 1: Distribution of the hypotheses variables (standardized) and outliers of at least 0.05. Yet an hypothesis is more supported by the empirical material than others, namely H2.1. More precisely, it appears that if a party goes to the right of the ideological spectrum for a unit equivalent to a standard deviation, its chances to support more inclusive electoral system decreases by 57%, all other things being equal 15. This effect is significant at 0.01 and so is the overall model. In contrast, even if an increase of an identical unit in the favorable attitude towards social groups increases party chances to support more inclusive electoral system by 73% all other things being (significant at 0.01), the overall fit of this model is not satisfying. In the same vein, the effect of ideology is rather poor when measuring the concept through the party family affiliation. While right-wing parties have 92% less chances to support more inclusive electoral system compared to the reference category mainly composed of centrist parties, belonging to the left-wing does not have any significant impact on this likelihood. This is without any the doubt the result of a not-flexible typology of party family that has not been able to stay in line with the actual evolution of the effective ideology of the OECD parties. For example, in the United Kingdom, the LibDems, which supported the proposed combined MM that would imply more inclusiveness in 1997, formally belong to the liberal family although being effectively on the left of the ideological spectrum. Also, it is important to note that since the party family variable is multinomial and cannot 15 The coefficients shown in the tables 2.3 to 2.5 are the logarithms of the odds of supporting more inclusive electoral system. The odds are then given by the exponential of the coefficients. 14

therefore be standardized, the associated odds cannot be directly compared to those for other standardized hypotheses variables. The coefficients estimated without outliers, as reported in the table 2.4, somehow confirms these findings. While the effect of the favorable attitude towards social groups is greater, its significancy decreases to a bare 0.05. In turn, this shows that the effect of this variables, although strong, is not robust and should be interpreted with cautious. On the opposite, the model estimating the effect of the left-right scale keeps very satisfying diagnostics. More specifically, an increase of a standard deviation towards the right pole of the ideological spectrum, decreases party chances to support more inclusive electoral system by 60% (significant at 0.01). From these analysis, it can be concluded that the spatial proximity model is only really empirically relevant when its central parameter- ie. party position over a social goal- is defined with the left-right scale. The more left-wing the party, the more it chances to support more inclusive electoral system. Now that the central parameter of the spatial proximity model has been estimated, the same needs be done for the seat-maximization model. The results of the table 3.5 give empirical evidence to the hypothesis that vote share at the preceding elections predicts party support for more inclusiveness. More precisely, the estimated model 4 shows that if a party electoral score increases by one unit equivalent to a standard deviation, its chances to support subsequent more inclusive electoral reform proposal decreases by 50%. This effect is significant at a level of 0.01 and so is the overall goodness-of-fit. Since potential associations between hypotheses variables of the two theoretical models may inflate or deflate the coefficients, they should be estimated together in order to properly assess their empirical relevance. Since party position on the left-right scale appeared to be the only robust specification of the spatial proximity model, this variable is the only one to be confronted to the party vote share at the preceding elections. The estimated coefficients both with and without outliers confirms the previous findings: when specified the way it has been done here, values appear to be as relevant empirically speaking, and even a bit more, than self-interests. While an increase of one standard unit increases towards the right-pole of the ideological spectrum decreases a party chance to support more inclusive electoral system by between 57% and 60%, a similar increase in its vote share at preceding elections decreases it by between 49% and 51%. Just as the overall goodness of fit of these estimated models, these effects are all significant at a level of 0.01. Putting those results into perspective, it appears that a favorable attitude towards social groups and the effective left-right ideology are better predictors of party support for more inclusive electoral system than expected seat payoffs. As a matter of fact, many of the case-studies in the existing literature confirm these findings. For example, the willingness of the Austrian People s party to implement a less inclusive electoral system than the List PR is use for the election of the lower house of the national parliament in 1989 has been said to be at least partially motivated by its willingness to get rid of a system that have been favoring the formation of unaccountable and unstable grand coalitions for years (Müller, 2005). Similarly, but inverted, some small populist right-wing parties such as or the Lega Nord in 1993 has supported the adoption of a less inclusive electoral system, which was in line with their ideology promoting strong and stable governments. Although 15

Table 3: Estimated coefficients of the spatial proximity model with all the cases Regressors Model 1 Model 2 Model 3 (attitude towards social groups) (ideology - scale) (ideology - family) Standardized hypotheses variables Attitude towards social groups 0.547** (0.212) Ideology (scale) -0.833** (0.251) Ideology (family) Left-wing family -0.273 (1.106) Right-wing family -2.528* (1.071) Constant and controls Intercept 0.697* (0.315) 0.727* (0.319) 1.098 (0.518) Direction -0.405 (0.404) -0.602 (0.422) -0.193 (0.409) In office at T 1-0.658* (0.398) -0.477 (0.410) -0.606 (0.408) Diagnostics Log-likelihood -75.953-72.801-75.345 χ 2 9.83* 14.01** 10.90* PseudoR 2 0.068 0.107 0.076 Degrees of freedom 3 3 4 N 119 119 119 Notes: Random-intercept estimations clustered by proposal (22 clusters); The predicted variable is the support for more inclusive electoral system; Standard errors are in parentheses; * <0.05, ** <0.01 (one-tailed). 16

Table 4: Estimated coefficients of the spatial proximity model without outliers Regressors Model 4 Model 5 (attitude towards social groups (ideology - scale - without outliers) - without outliers) Standardized hypotheses variables Attitude towards social groups 0.491* (0.247) Ideology (scale) -0.918** (0.283) Constant and controls Intercept 0.762 (0.324) 0.668* (0.322) Direction -0.415 (0.406) -0.596 (0.425) In office at T 1-0.796* (0.406) -0.400 (0.418) Diagnostics Log-likelihood -73.496-70.886 χ 2 8.90* 13.36** PseudoR 2 0.062 0.102 Degree of freedom 3 3 N 115 115 Notes: Random-intercept estimations clustered by proposal (22 clusters); The predicted variable is the support for more inclusive electoral system; Standard errors are in parentheses; * <0.05, ** <0.01 (one-tailed). 17

Table 5: Estimated coefficients of the seat-maximization model, and confrontation with the spatial proximity model Regressors Model 7 Model 8 Model 9 (expected seat gain) (confrontation, with all cases) (confrontation, without outliers) Standardized hypotheses variables Vote share at T 1-0.711** (0.230) -0.683** (0.238) -0.672** (0.239) Ideology (scale) -0.835** (0.263) -0.922** (0.297) Constant and controls Intercept 0.261 (0.327) 0.396 (0.343) 0.352 (0.346) Direction 0.216 (0.429) -0.169 (0.462) -0.170 (0.465) In office at T 1-0.216 (0.432) -0.002 (0.454) 0.061 (0.463) Diagnostics Log-likelihood -74.410-68.360-66.652 χ 2 12.58** 19.51** 18.62** Pseudo-R 2 0.087 0.162 0.156 Degree of freedom 3 4 4 N 119 119 115 Notes: Random-intercept estimations clustered by proposal (22 clusters); The predicted variable is the support for more inclusive electoral system; Standard errors are in parentheses; * <0.05, ** <0.01 (one-tailed). 18

it was certainly an attempt to take benefits from the public anger that was hitting Italy during this period, this support was against their self-interests as the proposal was seriously threatened their parliamentarian representation but was (Baldini, 2011). The same is true for the small Dutch party Democrats 66 that have been, since the 1980s, continuously at the origin of many failed attempts to reducing the inclusiveness of the electoral system as way of renewing the country s political class (van der Kolk, 2007). Conclusions and contributions The existing literature on the subject often implicitly considers that self-interest is the sole motivation of party support for electoral reform. Even if some publications pointed out the importance of values in this respect, they never conducted a systematic comparison of the two types of determinants. Relying on an original theoretical model and an unique comparative dataset, this article showed that, when specified the way it is done here, they are as relevant empirically speaking as each other. Further, it appears that the left-right scale is even a better predictor than the expected seat payoff, in the sense that the more left-wing the party, the better its chances to support more inclusive electoral system. This article therefore contributes to the existing literature on electoral reform bringing the values back into the game. In over-emphasizing the impact of seat-maximization and other seat related motivations, everything appears just as the vast majority of researchers working on the subject has been obscured by the perspective offered by the very mathematical nature of electoral system and of its core operation of translation of votes into seats to advance the quantitative study politics (Taagepera and Shugart, 1989). It seems that since it is one of the rare field where concepts and theories can be easily measured and expressed by numbers and equations, they have all forgot the accumulated knowledge on party strategy and public policy. In particular, even if parties have the possibility increase mechanically their seat share in adopting a new electoral system, they will not forget that their behavior is constrained by the program they have been elected for, their electorates, and future elections. In turn, one could interpret this as a positive sign that our modern democracies has a healthier functioning that we think. References Aardal, Bernt. 2002. Electoral Systems in Norway. in Grofman and Lijphart (2002) pp. 167 224. Adams, James, Michael Clarck, Lawrence Ezrow and Garrett Glasgow. 2006. Are Niche Parties Fundamentally Different from Mainstream Parties? The Causes and the Electoral Consequences of Western European Parties Policy Shifts, 1976 1998. American Journal of Political Science 50(3):513 529. Adams, James, Samuel Merrill III and Bernard Grofman. 2005. A Unified Theory of Party Competition. Cambridge: Cambridge University Press. 19

Andrews, Josephine T. and Robert W. Jackman. 2005. Strategic Fools: Electoral Rule Choice under Extreme Uncertainty. Electoral Studies 24:65 84. Baldini, Gianfranco. 2011. The Different Trajectories of Italian Electoral Reforms. West European Politics 34(3):644 663. Beck, Nathaniel and Jonathan N. Katz. 1995. What To Do (and Not To Do) with Time- Series Cross-Section Data. American Political Science Review 3(89):634 647. Benoit, Kenneth. 2004. Models of Electoral System Change. Electoral Studies 23:363 389. Benoit, Kenneth. 2007. Electoral Laws as Political Consequences: Explaining the Origins and Change of Electoral Institutions. Annual Review of Political Science 10:363 390. Birch, Sarah, Frances Millard, Williams Kieran and Marina Popescu. 2002. Embodying Democracy: Electoral System Design in Post-communist Europe. Basingstoke: Palgrave Macmillan. Blais, André, Agnieska Dobrzynska and Indridi H. Indriadason. 2005. To Adopt or Not to Adopt PR: The Politics of Institutional Choice. Brithish Journal of Political Science 35(1):182 190. Blais, André, ed. 2008. To Keep or To Change First Past The Post? The Politics of Electoral Reform. Oxford: Oxford University Press. Bogdanor, Vernon and David Butler, eds. 1983. Democracy and Elections: Electoral Systems and Their Political Consequences. Cambridge: Cambridge University Press. Boix, Carles. 1999. Setting the Rules of the Game: The Choice of Electoral Systems in Advanced Democracies. American Political Science Review 93(3):604 624. Bowler, Shaun, David Farrell and Robin T. Pettitt. 2005. Expert Opinion on Electoral Systems: So Which Electoral System is Best? Journal of Elections, Public Opinion & Parties 15(1):3 19. Bowler, Shaun, Todd Donovan and Jeffrey A. Karp. 2006. Why Politicians Like Electoral Institutions: Self-Interest, Values or Ideaology? The Journal of Politics 68(2):434 446. Budge, Ian. 2000. Expert judgements of party policy positions: Uses and limitations in political research. European Journal of Political Research 37(1):103 113. Bueno De Mesquita, Ethan. 2000. Strategic and Nonpolicy Voting: A Coalitional Analysis of Israeli Electoral Reform. Comparative Politics 33(1):63 80. Calvo, Ernesto. 2009. The Competitive Road to Proportional Representation: Partisan Biases and Electoral Regime Change under Increasing Party Competition. World Politics 61(2):245 295. 20