Representation vs. Responsiveness: How ideology and votes shape party policy change

Representation vs. Responsiveness: How ideology and votes shape party policy change October 2009 Abstract: Parties in modern democracies represent specific groups of voters. They offer distinct policy programs representing their (sub)constituencies. Ideologies deliver the framework structuring the programmatic platforms parties offer. Shaped by their ideology, socialist parties are geared towards representing workers' interests, Christian Democrats towards representing those of Christians and so on. Parties represent these interests in parliament. But parties should also respond to public opinion shifts and adapt their policy platforms. Following public opinion shifts, parties as vote-seekers compete for votes. The dual challenges - to represent and to respond to public opinion - may provide conflicting incentives for a party s choice of policy positions. This paper studies parties reactions to the challenges of representing specific constituencies and competing on the vote market at the same time. I argue that both considerations affect the choice of party positions: Parties indeed strive for votes but are restricted by voters ideological expectations. The findings suggest that parties act as Downsian vote-maximizers as long as voters' ideological expectations permit. However, vote-maximizing behaviour does not occur if the policy shifts involved contradict voters expectations about a party s ideological policy stands. The findings have important implications for political representation and models of party competition. Thomas Meyer Center for Doctoral Studies in Social and Behavioral Sciences (CDSS) Mannheim Centre for European Social Research (MZES) University of Mannheim A5, 6, Room 340 68131 Mannheim, Germany Phone: +49 (0) 621 181 2066 Fax: +49 (0) 621 181 2080 Email: tmeyer@mail.uni-mannheim.de Prepared for the Comparative Subconstituency Representation Workshop, October 21 st 2009 at the University of Essex 1

1. Introduction 1 Parties are necessary in modern democracies because they represent specific voter groups: Parties articulate and aggregate policy preferences, and translate those policy preferences into distinct policy options (Mair 2009: 5). Parliament reflects the politicized segments of society with parties representing the interests of their (sub)constituencies. Ideology shapes the party s image uniting the voters policy preferences with the party s programme. This idealtypical model of democracy is closely linked to mass parties (Duverger 1954) and a proportional vision of democracy (Powell 2000). Additionally, parties are expected to respond to changing demands from the electorate by adapting their policy programs to changing preferences of the electorate (Adams, Clark et al. 2004; Adams, Clark et al. 2006; Adams, Haupt et al. 2009). In other words, the main function of parties is to respond to public opinion by translating changing public opinion in corresponding public policy. If this process works, dynamic representation is at work (Stimson, Mackuen et al. 1995; Stimson 1999). 2 From this perspective, parties are Downsian vote-seekers competing with rival parties. This market model with parties offering policies to voters largely ignores ideology and fits to the majoritarian vision of democracy (Powell 2000). This paper studies parties reactions to the challenges of representing specific constituencies and competing for votes at the same time. Are modern parties indeed mainly vote-seekers increasingly neglecting their representative functions (as argued by Mair 2009)? Or are parties rather more constrained by their ideology tailored to representing the interests of specific constituencies? I argue that both ideology and votes shape party behaviour: Parties aim at maximizing their vote share but they are constrained by the ideological image voters have in their minds. If parties leave their ideological territories leapfrogging their competitors (Budge 1994), voters do not accept these new policy positions all. As Enelow and Hinich (1984: 116) put it: A candidate who attempts to run as a moderate one year and a conservative the next may lose credibility with the voters. His career may ultimately be destroyed, simply because the voters no longer believe he has any predictive label at all. Parties representing policy preferences which are in line with voters ideological expectations are free to shift policy positions to hunt for votes. If, however, a party s policy program 1 I thank Bernhard Miller for valuable comments and suggestions on earlier drafts of this paper. 2 Stimson uses the terms representation and responsiveness interchangeably (Stimson 1999). Hence, the term dynamic representation is somewhat misleading for the purposes of this paper. In fact, it rather means responsiveness defined as the dynamic representation of the median voter. 2

differs from voters expectations, incentives from the electoral market and the parties respective ideologies force parties to shift their policy positions. These incentives of representation and responsiveness may reinforce each other. If the electoral market and the party s ideology are in line and suggest a position shift in the same direction, parties will react by shifting their policies accordingly. If, however, representation and responsiveness demand different reactions, parties do not shift their policy positions: Vote-seeking policy shifts are prevented by ideological boundaries. Conforming to ideological expectations, however, costs votes. Hence, parties stay put. I test the hypotheses brought forward in the last paragraph by using data on party policy positions gathered by the Comparative Manifestos Project (CMP) (Budge, Klingemann et al. 2001; Klingemann, Volkens et al. 2006). Analyzing party position shifts in ten Western European countries 3 during the post-war period, I study whether parties react to ideological and vote-seeking incentives. The results show that the potentially contradictory incentives of representation and responsiveness do indeed shape the parties behaviour in the interactive fashion. I hypothesize that parties are most likely to shift their policy positions if the voteseeking and the ideological incentives point in the same direction. If the incentives contradict each other, parties hesitate to shift at all. Finally, parties follow vote incentives if their actual policy position conforms to the voters ideological expectations. In other words, parties are ideologically guided vote-seekers. The paper proceeds as follows: The next section reviews the literature predicting party policy change as a consequence of representation and responsiveness incentives. Next, I present three hypotheses how ideology, votes, and combinations of these two factors shape party behaviour. After a description of the data and the methods used, I present the results I the fifth section. After that I check the robustness of the results by performing three sensitivity analyses. The last section concludes. 2. Representation vs. responsiveness: The incentives for party policy change The assumption that parties aim at maximizing their vote share goes back to Duncan Black (1948) and Anthony Downs (1957). This model posits that parties choose policy positions so that the chosen policy stands are the best response to the positions of their competitors. Downs s prediction for two-party systems is the famous median voter theorem which predicts 3 The countries are Austria, Belgium, Denmark, Finland, Germany, Great Britain, Ireland, the Netherlands, Norway, and Sweden. 3

that both parties choose policy positions at the median voter s position. Downs s theory is static predicting party policy position using the Nash equilibrium concept. Party policy shifts only occur if the environment changes. Parties are responsive only in following (median) voter position shifts. Assuming that parties also implement their proposed policies, dynamic representation (Stimson, Mackuen et al. 1995; Stimson 1999) is at work: Public opinion changes and public policy responds. Following the model of dynamic representation, several studies deal with the question whether parties respond to shifts in public opinion (Adams, Clark et al. 2004; Adams, Clark et al. 2006; Adams, Haupt et al. 2009). James Adams and colleagues (2004) find that parties indeed react to shifts in public opinion. However, parties only respond to public opinion shifts away from the party s policy position. If public opinion shifts towards the party s policy position, the party does not respond. Moreover, party type affects the parties responses to public opinion shifts (Adams, Clark et al. 2006): Whereas mainstream parties follow public opinion shifts, niche parties (defined as Green, Communist, or Nationalist parties) do not react to changes in the voters policy preferences. The findings of both studies suggest that vote-seeking is not the only factor shaping party behaviour. First, the direction of public opinion shifts matters. Parties only react to policy shifts which put them at risk of losing votes at the next election. However, they do not react to public opinion shifts in the direction of their current policy position although some shifts may increase their vote share. Hence, parties may also change policy positions for ideological reasons. Second, the electoral strategies of mainstream and niche parties differ. Although alternative explanations such as the party leader s dependence on the rank-and-file s manpower and membership fees may explain the different party strategies, the findings of Adams and his colleagues suggest that mainstream and niche parties differ in their behaviour and the success of their vote-seeking strategies. Niche parties are not vulnerable to public opinion shifts because niche parties represent extreme or noncentrist ideological clienteles (Adams, Clark et al. 2006: 513; emphasis added). In other words, niche parties value ideology higher than votes. Downs s assumption of pure vote-seeking parties was also questioned before. Wittmann (1983; 1990) formulates a model of party competition in which parties are policy seekers aiming at gaining office to implement these policies. As Wittmann states [i]t would be strange if the voters were interested in policies and not the members of the political party, especially so because government policy is a public good shared by all (Wittmann 1990: 66). Exploring the reasons for policy-seeking behaviour, Robertson (1976: 35) notes that it is 4

mistaken to assume that a particular position on an ideological spectrum is of the same consequence of all parties. In fact, the party s ideology constrains its policy choice and votemaximizing policy positions may not be taken up due to ideological constraints (Robertson 1976: 38). Even if party leaders are vote-seeking, parties depending on their members labour and financial resources do not shift their policies away from their members preferences (Robertson 1976: 32). It has been argued that the parties dependence on their rank-and-file diminishes because elections campaigns are getting less labour- and more capital-intensive (Farrell 1996), party membership declines (Mair and van Biezen 2001), and state subsidies to parties increase (Katz and Mair 1995; van Biezen 2004; van Biezen 2008). In sum then, Robertson s arguments for policy-seeking parties might be less valid for today s parties. Nevertheless, I argue that parties are still constrained by their ideology. Even if party members cannot prevent party leaders from choosing policy positions which deviate from their policy preferences, voters may not accept party policy shifts resulting in policy positions deviating from the party s ideology. Sánchez-Cuenca (2008) argues that party policy positions have to be consistent with the party s ideology. Otherwise, voters will not be attracted to this party even if it is the closer party as defined by the proximity model (Sánchez-Cuenca 2008: 51). His empirical findings for Spain suggest that the consistency of parties policy positions with their respective ideology indeed plays a role. Parties with policy positions deviating from their ideology suffer at the polls. In a similar vein, Budge states that the movement of parties is limited because of ideological constraints. In particular leapfrogging is largely ruled out (Budge 1994: 448). Although party shifts are feasible, they mainly occur within the parties ideological territories (see Enelow and Hinich 1984: 116). Voters use the parties ideologies as information shortcuts for party policies. The party s ideology entails information on what the party stands for and which groups it represents: Socialist parties represent the workers interests, Christian Democrats those of Christians, liberal parties represent the interests of those interested in less state control. If a party shifts its policies deviating from the expected policy area (i.e. if it leapfrogs its competitors), it loses credibility and suffers at the polls. Recently, theories of party competition studying party position shifts integrated the challenges of representation and responsiveness and their impact on party strategies. Studying the effects of past election results on party position shifts, Somer-Topcu (2009) argues that policy shifts away from a party s policy ideals increase uncertainty and risks because parties do not know how voters, activists, or donors will react to change, or whether the party would lose its credibility in the eyes of the voter (Somer-Topcu 2009: 238). In other words, parties ignoring the representation duties are at risk of losing their credibility. At the same time, parties have 5

to care for votes to survive or to gain office in order to implement their policy programs. Hence, parties are more likely to shift their policy positions if they lost votes at the last election. Moreover, the effect of vote losses in the past diminishes the more time passed since the last election was held. Although interesting, the paper leaves some questions unanswered. First, although parties losing votes at one election are morel likely to shift their policy positions at the next election, the direction of these shifts is not considered. It is reasonable to assume that parties may choose a vote-seeking strategy shifting towards the centre of the policy space. Alternatively, however, parties could also decide to shift their policy position towards the preferences of their party members when they have lost votes in the previous election. In other words, it is not directly shown that parties shift policies for electoral purposes. Second, the paper assumes that the parties policy positions are appropriate in the sense that they conform to the party s ideology. If this is not the case, policy-motivated parties should have an incentive to react by shifting their policy position in the appropriate direction. Combining electoral strategies with ideological constraints, Budge, Ezrow, and McDonald (forthcoming) develop a dynamic theory of party competition. Parties respond by shifting their policy positions in response to electoral results: If the last policy shift resulted in vote gains at the last election, parties shift in the same direction at the next election. If, however, the party suffered at the polls, parties reverse their shifts. Yet, parties are also constrained by their ideology. Knowing that the party has to represent specific policies, they do not shift more than twice in the same direction. The authors show that their model outperforms policyalternation models in predicting party position shifts. Although simple and intuitive, the model suffers from two shortcomings. First, it is not clear why parties are only allowed to make two (in contrast to three, four, or five) shifts in the same direction. Although the supposed intra-party factionalism effect on the selection of policy positions is at work, there is no reason to assume that factional control of the party changes every second (and not third, or fourth ) election. Second, although parties are modelled as organizations with policyseeking intra-party factions, the party policy shifts are not influenced by the rival parties policy positions and position shifts. Parties represent absolute policies and only compare their current policy positions with stands they took in the past. They do not compare their policy positions relatively by comparing their policy positions with rival parties policy programs. Given the logic of multiparty competition these omissions are remarkable. I argue that parties indeed have ideological incentives to present voters a policy platform which is, for ideological reasons, left or right of specific parties. Communist parties, for example, may not only 6

compare their proposed policies with those at the last election(s). Perhaps more importantly, they have incentives to be the most leftist party in the system unless they want to lose credibility with their constituency. Hence, party strategies will pay attention to rival party positions (see also Adams and Somer-Topcu forthcoming). 3. Ideology and votes: How parties react to different incentives The literature therefore identifies two challenges for parties: First, parties have to be responsive. If public opinion changes, parties have to adopt new policy platforms which conform to the changing environment. Parties performing a vote-seeking strategy gain votes by listening to demands of the electorate. These responsive parties put pressure on other parties to respond to the public policy shift if vote losses wanted to be avoided. Hence, the pressure to adapt to public policy shifts stems from the demand (i.e. the risk of vote losses) and the supply side (i.e. the rival parties strategies). Hence, we can formulate the first hypothesis: Hypothesis 1: Parties adapt their policy positions in response to demands in the political market. Hence, parties shift their policy positions to the left (to the right) if public option shifts to the left (to the right). Second, parties represent specific constituencies. Party ideologies shape voters expectations on party policy positions. Communists are extreme left-wing parties, Social Democrats are (moderate) left-wing parties, whereas Conservatives are right-wing parties. Parties leaving their ideological loci are at risk of losing credibility and, subsequently, votes. Hence, parties should stick to those loci which are predefined by the parties respective ideologies. As mentioned above, parties can be seen as ideologically guided vote-seekers. Do parties always choose policy positions consistent with their ideological loci? Assuming full information, this should be the case. However, parties usually choose party policy positions under uncertainty. While writing a policy programme, the information available to parties about their rivals policy positions is confined to the rivals policy positions at the last election. Because parties adapt their policy positions to past information, parties may accidently leave their policy territories. Take, for example, a Communist party which 7

occupies the leftmost policy position in a given system. The actual policy position fits to the party s ideology so that vote-seeking motivations prevail. At the subsequent election, the Communist party shifts its policy platform to the right. After all policy platforms become public knowledge, the Social Democrats, could turn out to be on the Communist s left if the Social Democrats have decided to move to the left simultaneously. Due to uncertainty about the rivals strategy, those parties have therefore accidently switched positions thus leapfrogging each other. In such cases, the representation component of party strategies requires a leftward shift of the Communists, and a rightward shift of the Social Democrats at the subsequent election to restore the expected order. Hence, for ideological reasons, parties shift policy positions if the current party platforms do not conform to the ideological expectations. This leads to Hypothesis 2: Parties adapt their policy platforms to their ideological loci. Hence, parties shift to the left (to the right) if the current policy platform is too far to the right (to the left) relative to the rival parties policy positions. Hypotheses 1 and 2 suggest that parties hunt for votes while simultaneously correcting deficits in ideological expectations. Because both responsiveness and representation put pressure on parties to change policy positions and because parties have to deal with those pressures simultaneously, it is reasonable to suggest that that the effects interact. Table 1 lists the combinations of representation- and responsiveness-demands on political parties and the hypothesized effects mentioned in hypotheses 1 and 2. Using parties with appropriate (i.e. conforming to a party s ideology) policy positions which are left of the political market centre as a reference category, responsiveness demands parties to shift to the right (indicated by +) if the party s policy position is on the left of the political market centre. If the party s policy position is on the right, parties should shift their policy position to the left (indicated by -). Demands of representation demands require parties to shift to the right (to the left) if a party s policy position is too far to the left (the right) compared to the expectations shaped by a party s ideology. If the party s current policy position corresponds to its ideology, representation does not predict a policy shift at all (indicated by 0). 8

Table 1: Incentives of responsiveness and representation to shift party positions Representation Policy position too far on the left Policy position conforms to expectations Policy position too far on the right Policy position on the left of the political market A +/+ C reference E Responsiveness Policy position on the right of the political market B -/+ -/- - and + indicate expected shifts to the left and the right, respectively. 0 indicates that no policy shift is expected The first value stands for the effect proposed by representation, the second stands for the effect proposed by responsiveness. +/- D 0/- F Table 1 shows that the strategies suggested by representation and responsiveness create different situations for political parties: First, the effects may reinforce each other: If parties are on the left of the political market centre and if those policy positions are at the same time too far to the left compared with their respective ideologies (combination A in Table 1), both representation and responsiveness require a policy shift to the right (indicated by +/+). Similarly, both factors motivate parties to shift to the left (indicated by -/-) if policy positions are on the right of the political market centre and simultaneously too far to the right compared to the parties ideological expectations (combination F in Table 1). Second, the incentives provided by ideology and the political market may contradict each other: For representational reasons, parties want to shift to the left (the right) whereas responsiveness requires policy shifts to the right (to the left) (combinations B and E in Table 1). Because of these contradictions, responding to representation or responsiveness requirements is less likely (indicated by +/- and -/+, respectively). Third, in case a party s policy position conforms to its ideological locus, its representation incentive does not require a policy shift at all. Compared to parties on the left of the political market (combinations C, the reference category), voteseeking incentive motivates parties right of the market centre to shift their policy positions to the left (combination D in Table 1). I expect that policy shifts are most likely if representation and responsiveness incentives reinforce each other. If parties hold policy positions conforming to the expected ideological party locus, the size of the policy shifts is moderate. Finally, parties are least likely to shift their policy positions if the motivations of representation and responsiveness conflict. This leads to 9

Hypothesis 3: Parties are most likely to respond to representation and responsiveness requirements if both factors suggest policy shifts in the same direction. In case parties hold policy positions corresponding to their ideological expectations, parties just respond to incentives of the political market. If representation and responsiveness require policy shifts in different directions, parties stay put. 4. Measurement and models The sample contains policy shifts of political parties in ten West European parties (Austria, Belgium, Denmark, Finland, Germany, Great Britain, Ireland, the Netherlands, Norway, and Sweden) between 1945 and 2005. To measure party policy positions and party position shifts, I rely on data collected by the Comparative Manifestos Project (CMP) (Budge, Klingemann et al. 2001; Klingemann, Volkens et al. 2006). Using hand-coding, the election manifestos of political parties are subdivided in so-called quasi-sentences. Each of these sentences in then assigned to one of 56 policy categories. Using information of these categories, it is possible to derive a variable measuring the parties policy positions on a left-right scale. I use these leftright positions at two subsequent elections to measure the dependent variable policy shift. Values larger than zero stand for policy shifts to the right. Position shifts to the left are indicated by values smaller than 0. 4 To measure ideological expectations, I rely on country-specific left-right scales which order parties in an appropriate way. Although there are some cases in which no clear-cut left-right expectations exist, it is often possible to sort parties from left to right according to their ideologies. It is, for example, unquestionable that Communist parties are expected to be the most leftist parties in a given party system. Moreover, we expect socialist and social democratic parties to have left policy preferences which are, however, more moderate than those of communist parties. Similarly, conservative parties are ideologically on the right. For 4 Although the CMP data is often criticized for its lacking validity (Pelizzo 2003) or methodological weaknesses (see e.g. Benoit and Laver 2007; but also Budge and Pennings 2007; Benoit, Laver et al. 2008), it is widely used in articles published in high-ranking journals (Budge 1994; Adams, Clark et al. 2004; Adams, Clark et al. 2006; Tavits 2007; Somer-Topcu 2008; Adams and Somer-Topcu forthcoming). Furthermore and despite the criticism, the CMP data is the only data source measuring party positions over time. Neither other methods using content analysis (Laver, Benoit et al. 2003; Slapin and Proksch 2008) nor expert surveys have been used to systematically measure party policy positions over time. 10

other party families, such a placement is more difficult: Liberal parties, for example, occupy the centre of the policy space in some countries (e.g. in Great Britain or Norway) whereas they are on the right in others (e.g. in the Netherlands or Belgium). The same holds for Christian Democrats. Hence, I use data from country experts to create country-specific ideological policy scales (presented in the appendix, see Huber and Inglehart 1995; see country chapters in Müller and Strøm 2000; Benoit and Laver 2006). Comparing the parties actual policy positions (measured with CMP scores) with these ideological loci, I measure whether these positions conform to the expected position or if they are too far to the left, or too far to the right. Using the expected position as a reference category, I code two dummy variables indicating whether a party is too far on the left or too far on the right compared to its ideological expectation. For the analysis which follows, I use the position at the last election (i.e. the lagged values) to predict a party s position shifts to the left or the right. The best way to measure a party s responsiveness is to compare party position shifts with shifts in public opinion. However, data on the median voter s policy position is missing for some countries and time periods. Eurobarometer questions on voter left-right placements, for example, are only available since the 1970s. Moreover, the Eurobarometer surveys are not available for long time periods in Austria, Finland, Norway and Sweden. Furthermore, although these left-right placements can be compared cross-nationally, this does not hold for Belgium, Germany, and Ireland (Huber 1989; see also Adams, Clark et al. 2004: 597). The alternative Kim-Fording measure for median voter positions (Kim and Fording 2001) is not a feasible option because the median voter positions are obtained from party positions. Hence, voter positions (and their changes over time) are not exogenous to party position shifts. In the absence of data on voter position shifts, I select another measure for responsiveness which concentrates on the supply side of responsiveness. For each election, I calculate the centre of the political market by estimating the mean party policy position. Parties responding to the market should shift their policy positions towards the centre of this market. Hence, responsive parties left (right) of the centre at one election should shift their policy positions to the right (left) at the next election. The lagged value of the dummy variable party right of political market centre (t-1) indicates whether the party was left (0) or right (1) of the political market centre at the last election. The data is times-series cross-sectional Hence, the observations are not independent and the several assumptions of the ordinary least square regression are violated (Beck and Katz 1995; Beck and Katz 1996). First, there may be unobserved heteroskedasticity between parties. In other words, parties differ due to (unobserved) factors thereby violating the homoskedasticity 11

assumption. Second, parties and their policy positions are not independent of each other. At elections, parties interact and are influenced by several election- and country-specific (probably unobserved factors). Third, observations of parties are not independent over time. Party positions and party position shifts at time t are influenced by party decisions made at t- 1. I correct for the third problem by including the lagged value of the dependent variable (i.e. the party position shift from t-2 to t-1) to capture the serial correlation. To correct for the second difficulty, I use a linear multilevel model clustering parties in elections. Finally, I include country dummies to correct for potential country effects. 5 In total, the model includes 781 party position shifts in 10 countries. 5. Results Table 2 reports the regression results for two linear multilevel models. The first (additive) model tests hypotheses 1 and 2 simultaneously. The second model tests hypothesis 3 which posits that the effect of ideology and votes depend on the strategy suggested by the respective other factor (see Table 1). Hypothesis 1 states that parties are responsive by reacting to changes in the market environment. In the first (additive) model, the dummy variable party right of political market centre (t-1) indicates whether the party is right of the electoral market centre (measured as the parties mean policy position) at the last election. Compared with the reference category (i.e. parties left of the political market centre at the last election), parties on the right have voteseeking incentives to shift their policy positions to the left. Hence, I expect a negative (and significant) coefficient. Table 2 shows that this is indeed the case. On average, party position shifts are around 12 points on the CMP left-right space. Therefore, the effect size shown in Table 2 (4 points) is also remarkable. In sum, then, empirical evidence supports hypothesis 1. 5 I do not directly correct for potential differences between parties (i.e. the heteroskedasticity between parties). To do so, one could include dummy variables for all parties in the sample. Such a model contains over 150 variables and is at risk of overspecification (with n=781). Note, however, that the results of such a model do not alter the direction nor the significance levels of the results reported below. 12

Table 2: Explaining party position shifts with ideology and vote-seeking incentives Additive model Interacting model (Hypotheses 1 and 2) (Hypothesis 3) Party right of political market centre (t-1) -4.029** -2.918* (-3.57) (-2.07) Party left of ideologically exp. position (t-1) 5.370** 6.960** (3.40) (3.73) Party right of ideologically exp. position (t-1) -3.444* -2.461 (-2.28) (-0.97) left of ideology (t-1) * right of pol. market (t-1) -5.150 (-1.58) left of ideology (t-1) * right of pol. market (t-1) -1.771 (-0.59) Party policy shift (t-1) -0.314** -0.310** (-9.70) (-9.58) Sweden -0.430-0.662 Norway -3.498-3.455 Denmark -0.665-0.808 Finland -3.908-3.934 Belgium -0.295-0.357 Netherlands -0.883-1.250 Austria -1.747-2.045 Great Britain 0.0702-0.166 Ireland -1.398-1.543 (-0.35) (-0.39) Constant 2.788 2.387 (0.91) (0.78) Observations 781 781 z statistics in parentheses + p < 0.1, * p < 0.05, ** p < 0.01 To test hypothesis 2, the first (additive) model in Table 2 contains two dummy variables measuring the party s policy position compared with its ideologically expected policy position at the last election. Compared to parties with appropriate policy positions, parties with policy position left of the expected position are expected to shift to the right. Hence, I expect a positive (and significant) coefficient. In contrast, parties with policy positions on the right of the ideologically expected policy locus at the last election are expected to respond by shifting their policy positions to the right. I therefore expect a negative coefficient. Table 2 reveals empirical support for hypothesis 2: Using the CMP left-right scale, parties left (right) of their 13

ideologically expected positions at the last election shift their policy positions 5 (3) points further to the right (to the left) than parties with ideologically expected policy positions. Both effects are significant at the 1% and 5% level, respectively. Using the average size of party position shifts (12 points) as a yardstick, the size of the effects is also remarkable. In sum, then, the additive model reveals empirical support for hypothesis 2. The second model in Table 2 presents results to evaluate hypothesis 3. Compared with the first model, this model interacts the two ideology variables with the dummy variable measuring the party s policy position vis-à-vis the political centre (see also Braumoeller 2004; Brambor, Clark et al. 2006). As a result, the effect of ideological expectations and market incentives now depends on the strategies suggested by the respective rival theory. Because a direct interpretation of the effects presented in Table 2 is difficult, Table 3 lists the effect sizes and their significance level for each combination A to F of ideological and market incentives in line with the presentation in Table 1. Table 3: Interacting effects of responsiveness and representation on party position shifts Responsiveness Policy position on the Policy position on the left of the political right of the political market market Representation + p < 0.1, * p < 0.05, ** p < 0.01 Policy position too far on the left Policy position conforms to expectations Policy position too far on the right A 6.960** C reference E -2.641 B -1.107 D -2.918* F -7.149** I expect that the effects of the two independent variables interact: Parties are most likely to shift their policy positions if the motivations provided by the parties ideologies and the political market point in the same direction (i.e. for the combinations A and F). Hence, parties with policy positions left of their ideologically expected policy position which are also left of the political market centre are expected to shift their policy positions to the right. I therefore expect a positive (and significant) coefficient for the policy shift variable. Similarly, parties with policy positions which are ideologically too far on the right and at the same time right of the political market are motivated to shift their policy positions to the left. Hence, I expect a negative (and significant) effect on the policy shift variables. If the parties policy positions 14

conform to their ideological expectations, parties have no incentive to shift their policies for ideological reasons. Hence, parties are free to follow the political market: Parties on the left shift to the right (combination C, the reference category), parties on the right shift to the left (combination D). I therefore expect a negative (and significant) coefficient. Finally, parties are not expected to shift their policy positions if the motivations for party policy shifts provided by their ideology and their vote-seeking strategy point in different directions (combinations B and E). Therefore, I expect no significant effect if ideology requires parties to shift to the left while vote-seeking incentives force parties to shift to the right and vice versa. Table 3 reports strong support for hypothesis 3. Parties respond to incentives of the political market but they do so only if these incentives do not contradict the ideological expectations. Parties left of the political market centre shift to the right but this effect depends on the party s policy position compared to its ideological expectation: Compared to parties with policy positions conforming to the party s ideology, parties left of their ideological locus (combination A) shift their policy positions around 7 points further to the right. If, however, the party s ideology suggests a policy shift to the right (combination E), parties do not shift their policy positions according to the market incentives. In a similar vein, parties on the right of the political centre do not shift their policy position further to the left if ideology requires a policy shift to the right (combination B). The effect is negative but does not reach a significant level. If the party s policy position corresponds to its ideological locus (combination D), however, it follows the market incentives shifting towards to the left. Compared to competitors with ideologically appropriate policy positions on the left of the political market centre, parties on the right are more likely to shift their policies to the left. The effect size is moderate (around 3 points on the CMP left-right space) and significant at the 5% level. The centripetal effect of the political market is strongest, however, if the incentive to shift the party s policy position to the left is supported by ideological expectations (combination F): Compared to parties on the left of the political centre with ideologically expected policy positions, these parties shift their policy positions further to the left. The effect (around 7 points on the CMP left-right scale) is large and significant at the 1% level. In sum, then, Table 3 reveals support that parties react to both representation and responsiveness incentives. 15

6. Sensitivity analyses Sceptical readers may argue that the findings presented in the previous section are at risk of omitted variable bias or artefacts of the assumptions made for the ideological expectations. In this section I tackle such objections. First, it may be argued that the number of parties competing at elections influence the size of party policy shifts. The more parties exist, the more crowded the policy space. Hence, the more parties exist, the smaller the policy shifts. Moreover, the higher the number of parties, the higher the likelihood that parties leapfrog each other and hence, the higher the likelihood that parties react to those deviations according to the theory presented above. Although the effect of the number of parties is largely controlled by the country dummies, I estimated additional models including the number of parties competing at each election (as documented in the CMP dataset). I both calculate models with the number of parties at the current and the last election as additional control variables. The direction of the effects, the effect sizes, and the significance levels of the variables are not affected. Second, previous research suggests that parties could also react to past election results. Adams and colleagues (2004) however do not find empirical support for this hypothesis. Somer- Topcu (2009) shows that parties indeed react to past election results but less so with increasing distance to the last election. Hence, because past election results could influence party incentives to shift to the left or the right, I estimated another model using the vote changes at the last election as a control variable. Again, the findings presented in Table 2 prevail. Neither the direction, nor the size of the effects or the significance levels change. Finally, the reader may argue that the findings presented in the previous section are due to misspecifications of ideologically expected loci (presented in the appendix). Although the ordering should mostly be uncontroversial, there may be specific objections. In Germany, for example, it is controversial whether the FDP (Liberals) is left or right of the CDU/CSU (Christian Democrats) (see e.g. Saalfeld 2000). The same holds for ordering of the Austrian FPÖ and the ÖVP (see e.g. footnote 1 in Müller 2000). Even worse, Mair (1986: 457) notes that the left-right dimension as such is less meaningful in Ireland than in most other West European countries. I therefore calculate the same model presented above with a restricted sample excluding Germany, Austria, and Ireland. Although the number of cases drops to n=638, the findings of the models presented above are not affected. In sum, additional models strongly support the findings presented above showing that they are robust against additional control variables and sample restrictions. 16

7. Conclusion This paper started out asking how parties change their policy positions. On the one hand, parties have to present distinct policy positions to the electorate representing specific (sub)constituencies and defending their interests against their competing parties. On the other hand, parties are expected to be responsive following shifts in public opinion. In this paper I show that both representation and responsiveness make parties to shift their policy positions away from their current position. I argue that the parties respective ideologies are proxies for their expected policy positions. Ideology constrains parties to shift their policy positions away from their ideological locus. These constraints first derive from policy-seeking party members. If the party leadership depends on the members resources (e.g. the membership fees or the members labour), parties are less likely to hunt for votes by responding to public opinion shifts. Second, voters also hinder parties from shifting away from expected positions because political actors changing their ideological images risk losing their credibility (see Enelow and Hinich 1984: 116). At the same time, parties react to changes in their environment by shifting their policy positions in line with public opinion. The two incentives of representation and responsiveness to shift party policy positions may complement each other. In this case, parties are likely to shift their policy positions in the suggested direction. In case the expectations of ideology and vote-seeking strategies conflict, parties do not shift their policy positions. I present empirical evidence that both factors ideology and votes shape parties behaviour. Using data from the Comparative Manifesto Project (Budge, Klingemann et al. 2001; Klingemann, Volkens et al. 2006) on party position shifts in ten West European countries, I first show that parties react to incentives of the political market: Parties with policy positions left of (right of) the political centre, shift their policy positions to the right (left) at the subsequent election. Hence, parties shift their policy positions towards the centre of the political market. Second, parties react to unexpected party policy positions which can occur because other parties change their positions simultaneously. If a party s policy position is further on the left (on the right) than suggested by its ideology, parties correct for this unexpected policy position by shifting to the right (to the left). In other words, parties react to prior policy shifts in which they leapfrogged their competitors. Third, I show that parties simultaneously consider ideological and vote-seeking strategies. If the demands to represent and to respond create similar motivations, parties are most likely to react accordingly. If the 17

demands contradict each other, proposing policy sifts in different directions, however, parties stay put. The findings have important implications for political representation and models of party competition. First, the empirical evidence suggests some fears raised on modern parties and their behaviour (see Mair 2008) are unfounded. Modern parties still value ideology and do not solely react to incentives from the electoral market. Although the divides between different voter groups may become blurry, parties still choose policy positions which conform to their respective ideologies. Second, the findings affect theories of voting and party competition. The findings show that parties do not intentionally leapfrog each other. I emphasized that that this may be due to policy-seeking party members and their intra-party influence on the selection of party policy positions. In addition, however, voters may simply not accept party policy positions deviating too far from the party s expected policy loci. In another paper (Meyer 2009), I find that voters indeed differ in receiving and accepting party position shifts. Moreover, the acceptance of party position shifts decreases if parties leapfrog their competitors. The evidence presented here suggests that parties react to the voters scepticism. They avoid policy shifts to policy territories which do not fit to their ideology. If leapfrogging nevertheless occurs, parties correct for it by shifting back to the expected position. These multiple, interacting and time-variant incentives are alien to the static theories of party competition predicting optimal party positions using Nash equilibria (see e.g. Adams, Merrill III et al. 2005; Schofield and Sened 2006). Future research should, however, take them into account to enlighten our understanding of party competition. 18

Appendix: Ideological scales for expected party positions Austria: Communists Greens Social Democrats Christian Democrats FPÖ Liberal Forum 6 Belgium: 7 Communists Social Democrats Green parties Christian Democrats Liberals far right (Vlaams Bloc, FN) Denmark: Communists (including small left parties) Social Democrats centre (Radical Liberals, Center Party, Christian Democrats, Justice Party) right (Conservatives and Liberals) 8 Progress Party Finland: Communists (Socialists) Greens Social Democrats Center parties (KESK and SMP) Liberals Christian Democrats Swedish People s Party - Conservatives Germany: Communists (KPD and PDS) Greens Social Democrats Liberals Christian Democrats DSU Zentrum BG/BHE - DP 9 6 For the placement of the Liberal Forum see also Müller (2000: 87). 7 No distinction between Flemish and Walloon parties within party families; no left-right positions for VU, FDF and RW because clear ideological expectations are missing. 8 According to Damgaard (2000: 236), Liberals and Conservatives switched their policy positions over time. As in other north European countries, party families are not as decisive as in other countries. Rather, party systems are best understood in terms of five major groupings of parties (Damgaard 2000: 233; emphasis added). According to this, I distinguish party groupings rather than party families. 19

Great Britain: Labour Liberals (Liberals, Social Democratic Party, and Liberal Democrats) - Conservatives Ireland: 10 Socialists (WP/DL) Labour Fianna Fail Fine Gael Progressive Democrats Netherlands: Socialists (including the PPR) Greens Social Democrats D66 Christian Democrats DS70 List Pim Furtuyn 11 - Liberals Norway: Communists (Left Socialists) Labour Liberals Christian Democrats Center Party Conservatives Progress Party Sweden: Communists (Left Party) - Social Democrats - Greens - centre (including Center Party and Liberals) 12 - Christian Democrats Conservatives Notes: Data from country experts and experts (Huber and Inglehart 1995; see country chapters in Müller and Strøm 2000; Benoit and Laver 2006) Note that not all parties exist at all points in time (e.g. Green parties) 9 Coding for small parties according to Saalfeld (2000: 42) 10 No coding for Green parties because position vis-a-vis the Socialist parties is not clear. 11 List Pim Fortuyn placement due to Laver/Benoit expert study. Note that DS70 and List Pim Fortuyn do not compete against each other. 12 According to Berman (2000: 195), both parties occupy the centre of the policy space. Left-right distances between these two parties are marginal. 20

References: Adams, J., M. Clark, et al. (2004). "Understanding Change and Stability in Party Ideologies: Do Parties Respond to Public Opinion or to Past Election Results?" British Journal of Political Science 34(4): 589-610. Adams, J., M. Clark, et al. (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, J., A. B. Haupt, et al. (2009). "What Moves Parties? The Role of Public Opinion and Global Economic Conditions in Western Europe " Comparative Political Studies 42(5): 611-639. Adams, J., S. Merrill III, et al. (2005). A Unified Theory of Party Competition: A Cross- National Analysis Integrating Spatial and Behavioral Factors. Cambridge, Cambridge University Press. Adams, J. and Z. Somer-Topcu (forthcoming). "Do Parties Adjust Their Policies in Response to Rival Parties' Policy Shifts? Spatial Theory and the Dynamics of Party Competition in Twenty-Five Postwar Democracies." British Journal of Political Science. Adams, J. and Z. Somer-Topcu (forthcoming). "Moderate Now, Win Votes Later: The Electoral Consequences of Parties' Policy Shifts in Twenty-Five Postwar Democracies." Journal of Politics. Beck, N. and J. N. Katz (1995). "What to do (and not to do) with Time-Series Cross-Section Data." American Political Science Review 89(3): 634-647. Beck, N. and J. N. Katz (1996). "Nuisance vs. Substance: Specifying and Estimating Time- Series-Cross-Section Models." Political Analysis 6(1): 1-36. Benoit, K. and M. Laver (2006). Party policy in modern democracies. London ; New York, Routledge. Benoit, K. and M. Laver (2007). "Estimating party policy positions: Comparing expert surveys and hand-coded content analysis." Electoral Studies 26(1): 90-107. Benoit, K., M. Laver, et al. (2008). "Treating Words as Data with Error: Uncertainty in Text Statements of Policy Positions."??????(???):??? Bergman, T. (2000). Sweden: When Minority Cabinets are the Rule and Majority Coalitions the Exception. Coalition Governments in Western Europe. W. C. Müller and K. Strøm. Oxford, Oxford University Press: 192-230. Black, D. (1948). "On the Rationale of Group Decision-making." Journal of Political Economy 56(1): 23-34. Brambor, T., W. R. Clark, et al. (2006). "Understanding Interaction Models: Improving Empirical Analyses." Political Analysis 14(1): 63-82. 21