and Barbara Vis ( 2012 ) Why Do Social Democrats Retrench the Welfare State ? A Simulation

Why do social democrats retrench the welfare state? Next to violating core social democratic principles, recent studies show that social democratic parties and not so much other parties are systematically punished after retrenchment by (severe) vote losses. We propose and show that the strategy a party employs policy-seeking, office-seeking or vote-seeking is crucial for explaining why and when social democrats retrench, because it influences how they respond to environmental incentives (economy, public opinion, activist opinion and opposition strategy). The intra-party balance of power determines which strategy a party employs. We simulate an agent-based model to examine the implications of the different party strategies for the behavior of social democrats regarding welfare state retrenchment. This model includes 1) agents (here: parties) with specific decision heuristics that operationalize the different strategies, 2) voters who use a proximity-rule and 3) a coalition formation model. To mimic reality as much as possible, we calibrate the model with real economic data, whereby these data affect voters' policy preferences. We find that social democrats with a vote-seeking strategy retrench when inflation pushes the mean voter to the right. Office-seeking social democrats behave similarly, but respond with a strong time lag to the economy and to public opinion. Policy-seeking social democrats do not retrench. Moreover, we find that when faced with a radical left-wing competitor, social democrats shift towards the centre irrespective of their strategy -, increasing the likelihood that social democrats retrench. An empirical illustration of the behavior of five social democratic parties corroborates the simulations' results. To the welfare state literature, we offer a dynamic model that predicts under which conditions social democrats engage in welfare state retrenchment. To the literature on simulations, we offer an application of a model that makes predictions for a specific policy field, using real economic data and party decision rules.

Why do social democrats (such as the Dutch PvdA in cabinet Kok, British Labour under Blair and the German SPD in the Schröder cabinet) implement welfare state retrenchment?Retrenchment is a reduction of welfare state generosity, i.e. a lowering of the level and/or the conditionality of benefits for the unemployed, the sick, the disabled and the elderly.For long, a main assumption in the welfare state literature was that retrenchment is electorally risky for all political parties alike.Since retrenchment of for instance public pensions hurts all voters, all parties risk severe electoral punishment (vote loss) when they pursue it.While perhaps intuitively plausible, recent studies show that this assumption is incorrect.Specifically, Schumacher et al.
(forthcoming) find that social democrats systematically lose votes after welfare state retrenchment, whereas other parties (liberal and conservatives ones) do not.Related, Giger and Nelson (2011) find that liberal and religious parties gain votes after retrenching the welfare state.Against the backdrop of this new empirical evidence, a key puzzle in welfare state research thus is why social democrats engage in retrenchment.In addition to the probability of electoral punishment, retrenchment also goes against the core social democratic value of egalitarianism.
1.2 A vast body of literature explains welfare state retrenchment by stressing different environmental incentives such as the state of the economy or public opinion (for a review see Starke 2006).We propose, conversely, that although these environmental incentives do matter, social democrats differ in whether and how they respond to changes in the state of the economy or to shifts in public opinion (but see Green-Pedersen 2002;Kitschelt 2001;Zohlnhöfer 2003).Due to the variation in the balance-of-power between activists and leaders, parties differ in their motivations (policy or office) (Müller andStrøm 1999, Schumacher et al. 2010).These different motivations explain why some parties respond to changes in public opinion, whereas other parties respond to changes in what their own voters want.To understand the puzzle of why social democrats enact welfare state retrenchment, we model under which combination of party motivations and environmental incentives social democrats' expected utility from welfare state retrenchment outweighs the utility of refraining from it.

1.3
In our expected utility framework, parties receive utility from obtaining policy payoffs [1] and office pay-offs. [2]They differ, however, in their appreciation of these payoffs.Activist-dominated parties value policies, whereas leadership-dominated parties value office (Müller and Strøm 1999).If activists have the resources to topple the leadership when it fails to carry out the activists' preferences, the party leadership follows the activists.These activistdominated parties change position when the voters of the party change position (Schumacher et al. 2010).We label this a policy-seeking strategy.Alternatively, if the party leadership faces a weak party activist base, party leaders can pursue the prestige and salary associated with office.Schumacher et al. (2010) demonstrate that leadership-dominated, office-motivated parties respond in two ways.First, if the mean voter shifts, these parties shift their position in the direction of this voter.We label this strategy vote-seeking.Second, if parties are excluded from government, parties shift towards the position of the government.We label this strategy office-seeking. [3]We use a simulation to analyze which combination of strategies and environmental incentives motivate social democrats to retrench or not.To do so, we simulate elections in which 1) parties take a policy position, 2) voters vote for the most proximate party and which includes 3) a coalition formation procedure after which the newly formed government enacts policies.
http://jasss.soc.surrey.ac.uk/15/3/4.html 1 12/10/2015 1.4 Drawing inspiration from recently published agent-based models (ABMs) of multi-party competition in information-scarce environments (Laver 2005), we formulate decision rules that operationalize the three party strategies (also see Kollman et al. 1992;1998).Parties apply decision rules for deciding 1) their position and 2) whether or not to join a coalition government.We extend these simulations by letting real economic data affect voter preferences, by deriving party decisions rules from studies on party position shifts (Adams et al. 2005;Ezrow et al. 2011) and by letting social democrats compete in party systems that resemble most multi-party democracies.By doing so, our simulations resemble the actual political and economic environment of the 1980 to 2008 period.We present four simulations.First, we run a simulation of a three-party system comprising a Liberal Party, a Christian Democratic Party and a Social Democratic Party in which we assess the implications of policy-seeking and office-seeking strategies.This three-party system resembles Austria, Germany (before 1990) and the UK.In the second simulation, we introduce a party to the left of the Social Democrats-the Democratic Socialists-and again examine the implications of the three strategies.This type of party system resembles Germany (after 1990), the Netherlands and is also comparable to the Scandinavian party systems. [4]In simulations 3 and 4, we focus on the same party systems but now assess the implications of the vote-seeking strategy.

1.5
The paper is structured as follows.First, we discuss how the literature fails to make party strategies explicit.Next, we develop a model that operationalizes the three party strategies and perform simulations to study the implications of the model's assumptions for welfare state retrenchment.Subsequently, we corroborate the validity of the simulations' results by means of an empirical illustration of the relationship between party strategies and retrenchment, focusing on five European social democratic parties.The final section concludes.
Assumptions of Party Strategies in the Welfare State Literature 1.6 What are the assumptions about parties' strategies in the comparative welfare state literature?A vast body of work argues that socioeconomic changes fuel demand for specific social policies.For example, high levels of unemployment spur demand for social policies protecting the unemployed.Factors such as globalization (Adelantado and Calderón Cuevas 2006;Ha 2008;Swank 2005), de-industrialization (Iversen and Cusack 2000;Manow et al. forthcoming), "post-industrialization" (Bonoli 2007;Castles 2004) and budgetary pressures (Pierson 1994) have been identified as causes of welfare state retrenchment.Under the assumption that parties employ a vote-seeking strategy, aimed at winning office, parties respond to socioeconomic changes by providing social policies that the median voter supports.Pierson (2001), for example, argues that despite large-scale budgetary pressures, radical welfare state retrenchment has not taken place due to two factors.First, the welfare state's core programs (such as pensions and sick pay) receive broad support (Boeri et al. 2001;2002) and, second, institutional hurdles hinder cutbacks.Hence, it is rational for parties to converge towards a single policy position rather than offering distinct policy alternatives.According to this body of work, parties' dominant strategy is office-seeking while policy-seeking takes the backseat-if a seat at all.

1.7
By identifying the strength of social democratic power resources as the driving force of welfare state development, the power resources tradition provides a powerful alternative to the socioeconomic view on welfare state development (Castles 1982;Esping-Andersen 1990;Huber and Stephens 2001;Korpi and Palme 2003).According to Korpi (1989: 313), parties are 'organized expressions of collective action geared towards safeguarding the diverging interests of groups differently located in the social structure'.Parties provide policy alternatives to their constituents and it, therefore, matters which parties are in government.By connecting ideology directly to policy outputs, this tradition (implicitly) assumes that parties have fixed policy preferences (Kitschelt 2001).In other words, they are policy-seeking.

1.8
Both the assumptions of gaining office pay-offs by means of a vote-seeking strategy and the assumption of (pure) policy-seeking cannot explain why some social democratic cabinets have retrenched welfare state programs (e.g., Nyrup Rasmussen II and III in Denmark, Schröder II in Germany and Kok I in the Netherlands) (Vis 2010).As indicated earlier, recent empirical findings demonstrate that social democrats systematically lose votes after implementing welfare state retrenchment.This indicates that these parties do not pursue a vote-seeking strategy.Because retrenchment goes headson against social democratic values and policy preferences, these parties were also not pursuing a policy-seeking strategy.
1.9 Some studies address this criticism and argue that a combination of poor economic and electoral prospects motivates parties to retrench the welfare state (Green-Pedersen 2001;Kitschelt 2001;Vis 2010;Zohlnhöfer 2003).However, this latter body of research fails to explain the link between party strategies and retrenchment and also does not systematically assess when parties favor policy pay-offs over office pay-offs.This paper offers this systematic assessment.
The Utility of Welfare State Retrenchment 1.10 Before we define the utility function, we first define the policy space in which parties and voters take up position.Welfare state retrenchment consists of a reduction of welfare state generosity, i.e. a lowering of the level or conditionality of benefits for the unemployed, the sick, the disabled and the elderly.
Apart from welfare state retrenchment, parties can also expand the welfare state or maintain the status quo.To represent the entire policy space, as well as to create some variation between radical retrenchment and modest retrenchment, we define P as a set of policy preferences denoted by a continuum with -5 and 5 as limits.The negative values indicate retrenchment, zero indicates keeping the status quo, and positive values indicate welfare state expansion.The policy preference of party p on the policy space P is denoted by p p .Additionally, we define p c as the government's policy choice and v i as voter i's policy preference, each as points on the policy space P (v i , p c , p p ∈ P).
1.11 We define the expected utility formula as follows (see formula [i]).As indicated, parties have policy motivations and office motivations and differ in the extent to which they value both.As a consequence, party (p) calculates the utility of a potential government policy choice (U c,p ) by adding the product of the office pay-offs for the party (O c,p ) and the degree to which it values office pay-offs (β p ) to the product of the policy pay-offs for the party and the degree to which it values policy pay-offs (1-β p ).The policy pay-off is the absolute distance between the position of the proposal (p c ) and the position of the policy preference of the government (p p ).
The paper's objective is to analyze under which conditions social democrats accept a policy c that is below zero (retrenchment) on the policy space P over a policy that is zero or higher (status quo or expansion).Recall that our simulation builds upon dynamic models of party competition (Kollman et al. 1992;1998;Laver 2005) and policy choice (Kollman et al. 1997).Because of our model's inherent dynamism, we cannot analytically solve a set of functions with a closed-form solution, as is customary in game theoretic models (e.g.Alesina et al. 2006;Hollanders and Vis 2012). [5]Instead we derive and analyze comparative numerical statics (Smirnov and Fowler 2007).
http://jasss.soc.surrey.ac.uk/15/3/4.html 2 12/10/2015 Steps in the simulations and simulation set-up 1.12 Table 1 lists the important steps of our model, which will be explained in detail in the following sections.The model consists of (1) parties taking position by some decision rule, (2) voters updating their preference with regard to economic changes and (3) casting a vote using a proximity voting model.Parties subsequently gain seats (4) and choose a coalition and become part of the government or opposition.The government implements the policy c on policy space P (5).

Table 1: Steps in simulation
Step Action 1.
Parties take position p p using party decision rule 2.
Voters update their preferences.Increases in inflation push the electorate rightward; increases in unemployment push the electorate leftward  Korpi 1989).At least until then, we can therefore safely assume that the policy preference (p p ) of social democrats is between 0 and 5 on policy space P. To operationalize the initial preference of social democrats, we assign them value 2 on P.Although this paper's puzzle relates to the behavior of social democrats (SD), these parties compete with rivals for power and policy.Therefore, we introduce a centrist Christian Democratic party (CD), a right-wing Liberal Party (LP), and the so-called Democratic Socialist party (DS) as a more radical left-wing competitor of SD.We model the behavior of the latter parties below, but we do not analyze it.The reason is that retrenchment by religious and liberal parties is less puzzling because they do not systematically lose votes as a result of it, and because their policy preference is close to retrenchment.We therefore set the initial preference for Christian-Democratic parties at zero and for Liberal parties at -2.The Democratic Socialists are more radical than the social democrats, so we set these at 4.
1.14 We adopt Laver's (2005) AGGREGATOR rule as our policy-seeking decision rule, which is applied by parties that are activist-dominated as explained above.This means that the party looks at the last election, observes the location of the mean party voter and moves towards that position.We use this decision heuristic because an empirical study of 324 party shifts in 8 parliamentary democracies between 1977 and 2002 shows that policy-motivated parties are indeed responsive to the position of the mean party voter (Schumacher et al. 2010).With respect to parties' office-motivated behavior, Schumacher et al. (2010) provide empirical evidence that such parties shift position in line with the mean voter (vote-seeking strategy) or change their position if they are excluded from government (office-seeking strategy).We use the symbol α to denote whether a party chooses a vote-seeking strategy (α = 1) or a not vote-seeking strategy, i.e. an office-seeking or policy-seeking strategy (α = 0).The office-seeking decision rule is as follows: when in office in the period prior to the election, the party does not move; if the party is not in office it moves towards the position of the coalition position. [6]The vote-seeking rule directly applies the empirical finding: the party moves towards the mean voter at the previous election.Table 2 summarizes the operationalization of the three party decision rules.Vote-seeking α = 1, β=0 Shift towards the mean voter (t-1).
Voting rule-steps II & III in the simulation 1.15After parties have taken their position (step I in the simulation), voters update their preferences (step II).Specifically, voters adapt to changing circumstances.For example, Erikson and others find that the electorate shifts rightwards when inflation increases and leftwards when unemployment increases (Erikson et al. 2002).To capture this, we multiply the difference in the unemployment and inflation rate in the current and previous run by 0.05 (see appendix I).This produces a mean voter position shift that is comparable to real mean voter position shifts in European countries (Schmitt and Scholz 2005). [7]  1.16Moreover, we use inflation rates and unemployment rates similar to European experiences in the period 1980-2008 to make our simulations as realistic as possible.Most European countries experienced four economic phases in the 1980 to 2008 period.In phase 1, there is low inflation and low unemployment.In phase 2, inflation increases but unemployment remains low.This means that the mean voter shift rightwards.In phase 3, unemployment increases and the increase in inflation levels off.This means that the mean voter shifts back to the centre.Finally, in phase 4 inflation decreases and unemployment remains high (see appendix II).This means that the mean voter shifts further to the left.To enforce comparability between the phases, each economic phase in the simulation lasts 10 runs.Ten years also approximates the length in years for each economic phase in reality.After they have updated their preferences, voters vote for the party that is most proximate to their policy preferences (step III). [8]We opt for the proximity rule because it is the dominant paradigm in the voting literature (Adams et al. 2005). [9]  Coalition formation and policy implementation-step IV in the simulation http://jasss.soc.surrey.ac.uk/15/3/4.html 3 12/10/2015 1.17 Next, we calculate parties' utility for every possible coalition by applying utility function (i) (step IV in the simulation).To calculate a coalition's vote payoffs, we take the number of seats a party gains or loses if it had used the coalition position as party position in the last election.Office pay-offs represent the number of portfolios a party captures (see appendix I).We calculate policy pay-offs by the distance between the party position (p) and the coalition position (c).Parties choose the coalition with the highest utility. [10]The government coalition that is formed, is the one of which all parties involved believe that it has the highest utility.Finally, the coalition enacts its policy (step V in the simulation).The coalition policy p c is the average position of the policy preferences p i of all parties involved in the coalition, weighted by their number of seats.At the end of one run the simulation goes back to step I.
Simulation setup 1.18 We run four simulations.Simulation 1 tests the implications of office-seeking versus policy-seeking strategies in a three-party system.Simulation 2 does the same but in a four-party system.In this latter system, we add the Democratic Socialists (DS) and fix this party at an ideological position of 4. In simulation 1 and 2, we start with the minimum value of β (pure policy-seeking) and end with the maximum value of β (pure office-seeking).To test the implications of mixed strategies, we also examine six β values starting at 0 and increasing with increments of 0.2 to 1.In the simulations, the parties have an equal chance of choosing the policy-seeking, office-seeking or vote-seeking rule.In simulation 3, we test the implications of vote-seeking versus not-vote-seeking (parties use policy-seeking or office-seeking rules) in a three-party system.We pursue the same strategy but instead we vary the social democrats' α values.In other words, we vary the probability that social democrats choose the vote-seeking rule.In simulation 4, finally, we add the DS to the left of SD.The simulations are written in and conducted in R. [11] For each α and β value, we run 10 simulations in order to grasp the model's mean tendency.Each simulation has 40 runs, [12] hence in total we have 2,400 observations per simulation.We report the percentages of SD cabinets that retrench in each of the four economic phases and the α and β values in table 3.These percentages are thus based on 100 observations. [13]  Results

2.1
Table 3 shows the percentage of SD cabinets that retrench in simulations 1 and 2. The findings show that a mostly policy-seeking SD (β ≤ 0.4) retains a left-wing policy position because its mean party voter never dips below 1 (see appendix II).A policy-seeking SD (almost) never accepts retrenchment, regardless of the economic phase.This is because the SD keeps more-or-less the same group of voters and stays relatively close to its starting position of 2 and retrenchment thus remains too far away from its policy preference.Because it retains its non-centrist position and is unwilling to join a coalition that takes up a centrist or right-wing position, the Christian Democrats (CD) are more likely to form coalition governments with the Liberal Party (LP).Consequently, SD joins the cabinet infrequently (< 20%).An unintended consequence of this radicalism is that LP and CD retrench more than would have been the case if SD had been a viable coalition partner.

2.2
The results presented in table 3 demonstrate that a purely office-seeking SD (β = 1) accepts retrenchment even when voters are swinging to the left.This is because this SD cares about office, not about its policy distance to voters.A not-purely office-seeking SD (0.6 ≥ β ≤ 0.8), conversely, is less likely to accept retrenchment.This is because policy considerations keep the party to a position on the policy space P above zero.In this case, the SD also cares more about policy distances and therefore it does not join a coalition government that is too far from its policy preference.Interestingly, in all but one case, the office-seeking SD is more likely to retrench in phases where voters swing to the left (phases 3 and 4) than when they swing to the right (phase 2).This conclusion has important consequences for welfare state research.An office-seeking SD responds to economic changes and concomitant voter preferences shifts but only with a significant time lag.This is because the office-seeking rule adapts a party's position to the coalition position if it is out of office.There are quite a number of centre-left governments in phases 1 and 2, and therefore an office-seeking SD does not move in these phases.However, once CD and LP form coalitions in phases 2 and 3, an office-seeking SD shifts to a centre-right position.This SD subsequently gains office again and stays for some time at the centre-right, even when voters are shifting to the left in phases 3 and 4. Hence, the position of an office-seeking SD always lags the voter's position for a number of runs.Note: Percentages of social democratic cabinets that retrench for office-seeking (β = 1) versus policy-seeking (β = 0) for three parties (simulation 1) and four parties (simulation 2).

2.3
The right side of table 3 displays the results of simulation 2, i.e. the party system that includes the Democratic Socialists (DS)-a radical left-wing competitor.Since the more radical left-wing voters vote DS, SD's mean party voter is more moderate than in a party system without DS.In the simulation, Social Democrats therefore take a more moderate policy position for all β values, allowing a policy-seeking rule for at least one in five times (β < 1).When SD is a pure office-seeker (β = 1), it responds only to the behavior of the Liberals and Christian Democrats, since these are the parties with which SD forms coalitions.In fact, when SD follows LP and CD to the centre, the latter often jumps over SD-to the left-filling up the vacuum between the SD and DS.The result hereof is that SD and CD often join centre-left cabinets that do not retrench.However, for all other cases (β < 1), SD retrenches more than in simulation 1 because it is a more centrist party and therefore there is less disutility in bridging the policy distance to retrenchment.For the same reason, SD retrenches more in phases 2 and 3 than in simulation 1.This again implies the importance of the time lag of economic and public opinion factors.An unexpected, yet important finding is that the presence of a left-wing competitor does not make it less likely that the Social Democrats retrench (Kitschelt 2001).In fact, the presence of such a competitor moderates even the most policy-seeking SD.

2.4
Table 4 displays the percentages of SD cabinets that retrench in simulations 3 and 4, i.e. the simulations for vote-seeking strategies versus not-votehttp://jasss.soc.surrey.ac.uk/15/3/4.html 4 12/10/2015 seeking strategies.Starting with the three-party system (simulation 3), a pure vote-seeking SD (α = 1) typically accepts retrenchment mostly in phase 2 when the mean voter swings rightwards.Early in phase 3, when the mean voter creeps towards the left, the SD still accepts retrenchment, but at the end of that phase and also typically in phase 4, it is less likely to retrench than its pure office-seeking counterpart (β = 1) in the three-party system.For other vote-seeking values (0.6 ≥ α ≥ 0.8), SD is more responsive to economic change than in simulation 1, because it retrenches sooner and more in phase 2 when voters swing rightwards.Early in phase 3, SD is still considerably right-wing and accepts retrenchment, but with voters swinging leftwards it gradually becomes more left-wing and stops to accept retrenchment (almost) all together.Note: Percentages of social democratic cabinets that retrench for vote-seeking (α = 1) versus not-vote-seeking (α = 0) for three parties (simulation 3) and four parties (simulation 4).

2.5
In simulation 4, in which we assess the implications of vote-seeking versus not-vote-seeking in the four-party system, SD displays the most moderate policy positions of the four simulations.This is because, except for α = 1 (pure vote-seeking), a combination of all rules drives SD away from the already occupied left-wing.Because SD adopts such a centrist position, it often forms coalitions with the right-wing Liberal Party, often accepting retrenchment in such as coalition.In this simulation, SD also proves rather responsive to the economic tides since the percentage of SDs that retrench gradually declines over time.A vote-seeking SD in a system with a radical left-wing competitor (DS) is, in fact, the most likely of all types of SDs to retrench.This finding contradicts Kitschelt's (2001) claim that retrenchment is met with punishment by dissatisfied voters who turn to left-wing alternatives.Instead, we find that Social Democrats become less radical with a radical left party in the party system.Since there are more voters in the centre, a step towards the centre is more profitable than a step towards the left-especially when the centre is relatively empty.However, when the SD moves from the centre towards the left, it loses votes to the CD.The latter strategically moves with SD, absorbing larger parts of the electorate than SD is gaining.Voteseeking strategies respond more to direct economic events than do office-seeking strategies.This is because the vote-seeking decision rule dictates a direct response to the mean voter, instead of a response to a coalition position that is possibly further away from the mean voter.Yet, the two officemotivated strategies, office-seeking and vote-seeking, are similar since both are responsive to public opinion and the economy.The only exception is that the office-seeking strategy is responsive with a time lag.

Empirical illustration
3.1 A big question in comparative welfare state research is why social democrats retrench the welfare state given retrenchment's negative payoffs in terms of votes or policy.Our simulations show that the variation in party motivations (office and policy) and their concomitant strategies (office-seeking, voteseeking or policy-seeking) explain why and when social democrats retrench the welfare state.

3.2
As indicated, our model resembles best the political environment in the Scandinavian countries, the continental European ones and the UK.But does the model perform well when exposed to empirical data?We summarize the results of an empirical test of the model published elsewhere (Schumacher 2012), which shows that it does.The empirical test is a qualitative comparison of government terms of seven social democratic parties (British Labour, the Dutch PvdA, the Danish SD, the French PS, the Spanish PSOE, the German SPD and the Swedish SAP) in the period 1977 to 2005.Based on the intra-party balance of power, the PS and PSOE are mostly office-motivated; the SD, PvdA and Labour are mostly policy-motivated in the 1980s; and the SPD and SAP are in-between cases.For reasons of space, we only discuss the parties with the clearest motivations; the substantive conclusions do not change when the in-between cases would be included too.All these social democratic parties faced similar economic conditions in the 1980s and 1990s, but they responded markedly different to these similar conditions in terms of choosing welfare state policies.Table 5 indicates whether retrenchment takes place, whether parties are leader-dominated (i.e.office-motivated) or activist-dominated (policy-motivated), and whether the economy shifts voters to the right (-) (when inflation rises) or to the left (+) (when unemployment increases).

3.3
Considering each motivation separately, Table 5 shows that-as our simulation predicts-policy-seeking social democrats do not retrench.Specifically, the PvdA (1981-82) and SD (1979-84) did not retrench when in government.Also in line with the simulation results, the policy-seeking parties are mostly excluded from government, with the British Labour party even excluded for almost 20 years (from 1979 to 1997).These policy-seeking social democrats remain on the left, because they adapt to the position of their mean party voter (Marx and Schumacher forthcoming;Schumacher 2012).The mean voter, however, had shifted to the right due to the poor economic circumstances of the 1980s.Due to poor electoral performance, these three parties gradually diminished the power of activists in the early 1990s, giving more autonomy to the party leadership allowing for more office-motivated strategies (Hillebrand and Irwin 1999;Pedersen 2010;Seyd 1999). [14]Indeed after these reorganizations, the SD, PvdA and Labour shifted towards the political centre, abandoning their focus on the mean party voter.The PvdA (1991,1992,1994), SD (1993SD ( , 1995SD ( , 1998) ) and Labour (1997) implemented welfare state retrenchment measures on several occasions.Interestingly, these parties did retrench in a period in which the economy was driving voters leftwards.If they would be aiming for votes only, this should have motivated these parties to stay on the left and refrain from retrenchment.By moving towards the centre, these social democrats were strategically adapting their position to the centre-right cabinets at the time.Hence, they adopted the office-seeking rule, responding to public opinion and the state of the economy with a time lag. 1 Intra-party balance of power changed during government term and so did party behavior.

3.4
The leadership of the French and Spanish socialist parties is autonomous and therefore our model predicts that these parties use office-motivated rules (i.e.vote-seeking or office-seeking).Under the leadership of Mitterrand, the French PS centralized its organization, placing crucial decision-making power in the hands of the party leadership (Kitschelt 1994).At first, the PS presented a rather radical agenda.However, at the end of the 1970s, under high inflation, the mean voter shifted rightwards.In response, the PS made a strong volte-face in 1983 accepting retrenchment and austerity measures (Kitschelt 1994, Palier 2010) at a time when similar policies were considered heretical for the policy-seeking social democratic parties.Similarly, once the economy pushed the mean voter to the right in the early 1980s, the PSOE reneged on promises to expand the welfare state and shrunk it instead.However, when the economy pushed the mean voter to the left, the PSOE (1986PSOE ( -1993) ) and the PS (1997)(1998)(1999)(2000)(2001)(2002) expanded the welfare state.These parties are thus directly responsive to the economy, by responding to these shifts in the mean voter position.Specifically, and in line with our simulation results, these vote-seeking social democrats retrench when the economy pushes the mean voter to the right but they stop retrenching when unemployment sweeps the mean voter back to the left.In sum, the empirical results corroborate our simulation's finding that the variation in strategies (vote-seeking, office-seeking and policy-seeking) explains why social democrats respond differently to similar economic and electoral conditions.

Conclusion
4.1 This paper systematically analyzes how social democrats respond differently to similar environmental incentives and shows that this is due to differences in party motivations and related strategies.Summarizing our results, we find that policy-seeking social democrats do not retrench, that voteseeking social democrats retrench when the economy pushes the mean voter to the right, and that office-seeking social democrats behave similarly as vote-seeking social democrats do but respond much later.In addition to unraveling these implications, our simulation contributes to the welfare state literature by revealing three interesting dynamics.First, we find that the radicalism of policy-seeking social democrats in times of high inflation enables Christian democrats and liberals to pursue a right-wing agenda without fear of losing the political centre.This is analogous to the British and Dutch situation in the 1980s, where left-wing radicalism gave centrist and right-wing parties the chance to retrench without being punished for it.

4.2
Second, Kitschelt (2001) has argued that the presence of left-wing alternatives makes it electorally unattractive for social democrats to retrench the welfare state.Consequently, social democrats (should) stick to the left.Our simulation suggests that the opposite happens.Social democrats are more moderate when faced with radical left-wing competition and are therefore more likely to accept retrenchment.While conflicting with Kitschelt's expectation, our results are in line with an empirical study of Pontusson and Rueda (2010), who show that social democrats have more moderate leftwing platforms when they face radical left-wing competitors.

4.3
Third, the simulations of office-seeking social democrats suggest that welfare state researchers are advised to model the effects of economic conditions on the welfare state differently.Currently, (quantitative) welfare state researchers typically introduce a one-year lagged effect of economic conditions on changes in the welfare state.Theoretically, this means that governing parties directly update their policy preferences in response to economic conditions.Although we find that this is indeed what vote-seeking parties do, our analysis indicates that office-seeking parties only adapt their preferences when in office and then move towards the coalition position.As such, office-seeking parties always trail the latest economic developments.In many instances, governing parties therefore do not respond within a year to economic changes.In fact, this may (partly) explain the mixed evidence typically found for the effect of for instance globalization.Some vote-seeking governments are quick to offer compensation for the victims of globalization (like the unemployed); their behavior supports a compensation argument.However, policy-seeking or office-seeking governments will not respond that quickly, which weakens the support for the compensation argument.Hence, whether globalization has an effect depends on the specific strategy a party follows.

4.4
To conclude, our study contributes to the welfare state literature by showing that the variation in party strategies explains the variation in policy outcomes.To the literature on simulations, we offer an application of a model that makes predictions for a specific policy field, using real economic data and party decision rules.Inspired by more abstract models (Kollman et al. 1992;Kollman et al. 1997;Laver 2005), our concrete application of welfare state politics illustrates the usefulness of simulation techniques for a field in which this technique is underutilized._ Appendix I. Model Components

Mean voter position
Mean of all voter preferences (see appendix II).

Party voter position
Mean of all party voters (at t-1) preferences.
3 Whether parties employ vote-seeking or office-seeking strategies may depend on the electoral system.In two-party majoritarian systems, keeping track with public opinion is a good strategy to win the median voter (i.e.vote-seeking), while in multiparty PR-systems staying close to the government (when in opposition) is a good strategy to obtain office pay-offs (Schumacher 2011).
but one competitor seems to be the standard.In the Netherlands, there are traditionally several parties to the left of the social democrats.Moreover, in all these countries there are more than two parties to the right of the social democrats.However, for the model it matters that the strategic positions to the left and right of social democrats are occupied; it matters less how many parties occupy these spots.Therefore, our four-party setup can reproduce a similar strategic situation as social democrats experienced in these countries.
the coalition's position, which leads to retrenchment or expansion Party positions and decision rules-step I in the simulation 1.13 Let us first discuss with position p p parties take (step I in the simulation).Social democrats have an ideological commitment to the welfare state and are considered to be the protagonists of welfare state expansion, at least up to the 1970s (Huber and Stephens 2001;

Table 2 :
Party decision rules

Table 5 :
Social democrats in government, retrenchment and types of party behavior