Choosing the Path of Austerity: How Policy Coalitions Shape Welfare-policy Choices in Periods of Fiscal Consolidation

Transcription

1 Choosing the Path of Austerity: How Policy Coalitions Shape Welfare-policy Choices in Periods of Fiscal Consolidation Klaus Armingeon, Kai Guthmann, David Weisstanner University of Bern Institute of Political Science May 2014 Paper presented at the ECPR General Conference in Glasgow, September 2014 Abstract: This paper focuses on the effect of fiscal adjustment programs on public social expenditures. We show that budget consolidations are generally associated with welfare state retrenchment. But do the partisan complexion and the type of government condition the extent to which austerity policies rely on social spending cuts? These are the questions guiding our paper, which compares 17 OECD countries between 1982 and Our findings partly support a functionalist argument: If governments embark on austerity, their partisan complexion does not matter. Since welfare state retrenchment is electorally and politically risky, however, a broad pro-reform coalition is a crucial precondition for large fiscal consolidation programs to rely on substantial cuts to social security. We use a novel operationalization of fiscal consolidation based on budgetary decisions, rather than simply tracing deficits without consideration of the underlying policy changes. Our empirical results are based on long-run multipliers from autoregressive distributed lag models.

2 Introduction In the summer of 1982, the socialist French president Francois Mitterrand learned the hard way that under severe constraints of liberalized markets for goods and capital (particularly under conditions of a de facto fixed exchange rate regime in Europe) traditional Keynesian policies no longer work. Reluctantly, the French government had to take up on austerity policies (Hall, 1986, Chapter 8), marking the beginning of a period of permanent austerity in the mature democracies of the Western world (Pierson, 2001). Permanent austerity means that austerity is the dominant script in the discourse of governing parties. Despite that basic conviction that public deficits must be reduced, public households may still expand due, for example, to demographic or economic reasons. In fact, between 1983 and 2007 many countries frequently deviated from austerity courses, with Greece being the most dramatic example. After 2007 a brief straw fire of Keynesian demand management flared up in some countries in response to the sharp drop in output during the early phases of the Great Recession (Armingeon, 2012; Cameron, 2012). By 2010 however, in the aftermath of the recession, full-blown austerity measures were not only back on governments agendas (except in Japan), but they had returned with a force not witnessed in earlier decades. It is against this background that we study the impact of austerity policies on the welfare state. Austerity may come in many forms. Some governments try to consolidate public finances with higher taxes, others focus on cuts in spending, and, probably most frequently, governments opt for a mixture of spending cuts and tax increases. Spending cuts can be executed in many policy fields, from subsidies to agriculture, to education and military spending, to reducing the costs of public administration. One of the most important candidates for cuts, however, is the welfare state. We know that public social expenditures are closely correlated with indicators of income inequality (Huber & Stephens, 2001; Pontusson 2005). The more austerity is focused 1

3 on this spending category, the more likely it is that those at the bottom of the income hierarchy will bear the brunt of consolidation. The welfare state and inequality are major issues of dispute in the politics of modern democracies. We would commonly expect pro-welfare state parties Social Democrats and, with some qualifications, Christian Democrats to be most reluctant to consolidate the budget by means of welfare state retrenchment. In other words, there should be a strong partisan effect: if a left party holds governmental power, an austerity program may be less reliant on social expenditure cuts than under a centrist or liberal-conservative administration. At second glance, however, this is not very convincing. Austerity policies are often enacted when governments are facing dire economic situations. Once they run out of money, left parties may therefore based on a functionalist logic feel compelled to do exactly the same as their more rightist competitors would. If there are any partisan differences, we may even expect a Nixon goes to China scenario (Kitschelt, 2001): since left parties are more trustworthy as true defenders of the welfare state, they may also be the most successful in passing and implementing reforms. There may, however, be an alternative mechanism at work, one that is not dependent on any sort of partisan logic. It is based on the type of government, and the requirement of stable majorities necessary for big reforms, as well as the accompanying electoral risks (Pierson, 1994). The larger the pro-reform coalition, the more likely will the reform process be timeconsistent, and the less likely will a major opposition party be able to exploit the electoral vulnerability of the governing parties implementing the reform. Two types of government may be particularly well suited to reducing such electoral risks and thus may be best able to implement cuts to the welfare state: minority governments and surplus coalitions. The former have to negotiate policy-field specific measures with the major opposition parties, as they otherwise have no chance of realizing substantial reforms. The latter include more parties than necessary for a 2

4 parliamentary majority and thus integrate many players that would otherwise be in opposition rallying against the reform (see Lijphart (2012: 79-93)). The underlying logic is the same: austerity programs implemented by broad pro-reform coalitions inherent to minority governments and surplus coalitions should be more likely to rely on the electorally risky business of welfare state retrenchment. Our research interest thus boils down to the following questions: How do different governments design austerity policies with respect to the welfare state? Is there any effect of (i) the partisan complexion or (ii) the type of government on the extent to which austerity policies rely on social spending cuts? We compare the experiences with austerity policies in 17 OECD countries between 1982 and Our theoretical argument is discussed in the next section. We then present our data sources and operationalizations, in particular the major independent variable of fiscal adjustment, based on path-breaking work on public budgets (Devries et al., 2011). This new data set provides information about the fiscal effects of austerity policies that have been adopted by governments and parliaments. We then discuss our research design and the statistical technique; finally, we present our empirical findings. The argument Austerity policies denote cuts in public expenditures and/or increases of public revenues which are intended to reduce a budget deficit. 1 Austerity can thus in principle be designed in three ways, each of which implying very different distributional consequences. First, a government may raise taxes. For example, in 2013 the new socialist French president François Hollande tried to consolidate public finances by introducing a new tax on the wealthy. Second, a govern- 1 We use the terms fiscal consolidation, fiscal adjustment, and austerity interchangeably. 3

5 ment may implement cuts to non-social-spending-related items of the budget, such as agricultural subsidies, or spending on infrastructure investment, defense, or education. Finally, fiscal adjustment may comprise of cuts to social expenditures (both benefits in kind and cash benefits, such as unemployment assistance or pensions) and thus largely imply welfare state retrenchment. Empirically, while most austerity programs tend to combine tax and expenditure measures, spending cuts are more extensively used as compared to tax increases (Alesina & Ardagna, 2009; Wagschal & Wenzelburger, 2008b). One of the largest items on public budgets in modern democracies is social spending. In 2009 public outlays accounted for 49% of GDP on average in the 23 mature OECD democracies, and about half of these (25% of GDP) were devoted to the welfare state (calculated from Armingeon et al. (2013)). Any program of fiscal consolidation is therefore very likely to include at least some cuts to welfare state expenditures. As a result of the central role that the welfare state plays in mitigating the inequality of wealth and income, these cuts should significantly alter the income distribution at the expense of the poor. This is also what empirical research finds: austerity hits those at the bottom of the income distribution more forcefully than other groups (Agnello & Sousa, 2012; Ball et al., 2013; IMF, 2012, p. 50 ff.; Mulas-Granados, 2005). From partisan theory we may derive hypotheses about the degree to which governments of varying partisan complexion should opt for social spending cuts in an effort to consolidate the budget. Overall, we may expect left governments to be the least willing to adjust through the welfare state, when compared to their centrist and right (i.e. secular-conservative or liberal) competitors (Esping-Andersen, 1990; Huber et al., 1993; Schmidt, 2010; Swank, 2013). But is this strict ordering of parties with regard to their welfare-preferences realistic in times of austerity? On the one hand, the welfare state is extremely popular in western societies. Even right parties have incentives to preserve it, mainly in terms of the political costs associated with doing so, i.e., electoral punishment. This is a major argument by Paul Pierson: since the welfare state 4

6 is so strongly embedded in society, and voters on the left and right are generally supportive of its many benefits, politicians of any orientation will try to avoid the electoral costs of welfare state retrenchment (Pierson, 1994, 1996). Rather, most policymakers on the left and right will try to design austerity measures in a way that achieves fairness in the distribution of adjustment burdens, thereby sparing the social spending categories from cuts whenever possible. Such balanced austerity was, for example, the claim of the Greek government in the Memorandums of Understanding with the IMF and the EU in 2010 (IMF, 2010a). On the other hand, however, the empirical evidence that balanced austerity gets implemented at all is weak. Governments have failed to honor such claims, when we consider, for example, cases of dramatic consolidation efforts such as in Greece, Spain, Portugal, or Ireland in the current crisis. In these countries, governments were forced to save money wherever they could, without any serious consideration of the social implications (Armingeon & Baccaro, 2012). Upon closer look, this comes as no surprise: First, influential economic research tends to find that spending cuts are more effective as compared to tax increases in reducing debts and deficits (Alesina & Ardagna, 2009, 2013; Guajardo et al., 2011). Second, since one of the largest spending items on public budgets in modern democracies is the welfare state, it is hardly conceivable even for left parties to spare it altogether. Finally, cuts to social spending may be the only feasible option available to policymakers when the government faces a situation like that many Southern European countries were experiencing until fairly recently: a sovereign debt crisis that severely limits governments ability to roll over their debt on sovereign bond markets. In such situations, consolidation policies will be effective only with the consent of financial markets, since it is here where interest rates on government bonds are determined and financial markets do not tend to reward austerity programs that shy away from resolute cuts. In essence, the above discussion can be boiled down to a functionalist argument that stresses the importance of policy constraints emanating from opportunities and functional requirements and resources. Times of austerity are, virtually by definition, hard times, and during 5

7 hard times the set of feasible policy options available to policymakers tends to be severely limited. If a government embarks on a path of austerity, it is in all likelihood forced to cut back the welfare state. This leads to our first hypothesis: H1: Fiscal consolidation is associated with decreasing public social expenditures. Moreover, since under such circumstances little room of maneuver is left for the left or any other party, we generally do not expect the partisan complexion of government to make any difference for the design of austerity policies aimed at the welfare state there is no systematic partisan effect. This leads to a second hypotheses, which having formulated it as a null hypothesis we do not generally expect to be able to reject (more on that below): H2: The effect of fiscal consolidation on public social expenditures does not vary significantly with the partisan complexion of government. However, even if severe problem pressure constrains parties to such a degree that they begin behaving quite similarly in times of austerity, functional requirements and resources need to be cognitively identified by policymakers, translated into policies, and implemented. This leads to a discussion of the broader context in which democratic decision-making takes place. Due to their high salience and contentious nature in society, welfare state policies in particular are also a function of the ability of governments to mobilize support among societal actors, build broad pro-reform coalitions, and of the way policymakers are perceived by and enjoy the trust and support of the citizenry. In particular, relatively large and encompassing consolidation programs should depend on such favorable political conditions, which vary across national contexts and time. We have two partly competing arguments as to the conditions that should make governments more likely to consolidate the budget through substantial welfare state retrenchment the first of which tends to incorporate parts of the partisan logic, while the second does not. 6

8 First, the strong left parties found mainly in the Nordic countries, which were central to the foundation and expansion of the welfare state in their historical alliance with the trade union movement, may also be best able to cut it back. They are more trustworthy; they are better able to convince unions and other crucial welfare advocates to abstain from an unconditional defense of the existing safety net, and therefore they may even be able to mobilize support for a remodeling of the welfare state in order to make it sustainable in the long-run (see e.g. Ross (2000)). This can be referred to as an example of the Nixon goes to China effect (Kitschelt, 2001): a clear anti-communist politician such as Nixon is not vulnerable to allegations that he is too soft in negotiations simply because the ideological distance between him and his communist negotiation partners is so immense. In other words, the strong left parties ideological position and welfare state legacy signals to the electorate and societal actors that the welfare state will not be dismantled, but rather re-modeled, and thereby enables the left to build a pro-reform coalition that is more stable and encompassing than other parties could hope to achieve. From this perspective, we expect the strong social democratic parties found in the Nordic countries to be associated with more pronounced cuts to welfare state spending than their more rightist (or non-nordic) competitors (see also Obinger et al. (2010)). This leads to the following hypothesis modifying H2: H3a: When passed under the guardianship of the strong social democratic parties found in Nordic countries, fiscal consolidation is associated with a more pronounced decrease in public social expenditures than in cases where a strong left is absent. Our alternative argument starts from Paul Pierson s new politics perspective, which states that while partisan theory can explain the expansion of the welfare state, it is less powerful when retrenchment is concerned (Pierson, 1994). Once the welfare state has been established, it has built its own battalions. Since citizens have contributed to the social policy schemes, they have 7

9 acquired rights to social security and expect to get benefits in return. Any attempt to cut back on social spending is therefore electorally risky. Times of austerity are times of new politics : Politicians of any political orientation try to avoid punishment for welfare state retrenchment. While the punishment-hypothesis has been qualified in the academic debate (e.g. Armingeon and Giger (2008), Arndt (2011), or Giger and Nelson (2011)), there is little doubt that politicians take electoral risks seriously. For example, in negotiating the Grand Coalition in Germany in fall 2013, the chairman of the German Social Democrats stated that they would not repeat the mistake of liberalizing reforms, since these led voters to withdraw their support for the party. Governing politicians planning to cut back social security therefore have strong incentives to build a broad pro-reform coalition that includes as many parties in the party system as possible. If they do not, opposition parties outside of the coalition could exploit the electoral opportunities emerging from retrenching reforms in societies where citizens are strongly in favor of the welfare state. Two types of governments are particularly well suited in this regard. The first is an oversized coalition, since it includes more parties than necessary for a parliamentary majority. The second type of government is a minority government, because its members need to negotiate policy-specific majorities in parliament and therefore need to regularly get opposition parties on board (Lijphart, 2012: 79-93). The task of retrenchment should be much harder under single party governments or minimal winning coalitions, which are defined as including just as many parties as needed for a parliamentary majority. This argument goes back to ideas developed by Lijphart (2012) and Katzenstein (1985). More recently, Alexiadou (2013) condensed and reformulated the argument and put it to a rigorous test. Oversized and minority governments are conducive to social spending-based austerity for other reasons as well. First, for these reforms to fulfill their long-term promises, the proreform coalition needs to be time-consistent, i.e., reforms must not be repealed at the next possible occasion. Rational politicians interested in substantial policy change will therefore seek 8

10 broad long-lasting support and this is more likely to be secured through a formal oversized coalition or by negotiating policy-specific majorities on an ad hoc basis under minority governments. Second, even more important may be the signal to potential non-parliamentary opponents of welfare state retrenchment. Strong and potentially militant trade union movements, for instance, may seek to challenge a reform at all stages of the policymaking process. If these trade unions confront a very broad pro-reform coalition in the party system, however even including parties they usually see as allies they are likely to conclude that fierce opposition will not be successful. From this perspective, the ideology of governing parties and their historic relationships with the trade union movement are not the major determinant of the design of austerity (in contrast to hypothesis 3a). Rather, it is the breadth of the policy-coalition that reduces electoral risks and costs, silences interest group opposition, and ensures sufficient and time-consistent parliamentary majorities for reform (Alexiadou, 2013). While single party or minimal winning governments may also try fiscal adjustments, even substantial ones, they will have considerable difficulties in realizing these adjustments by means of welfare state retrenchment. Broad policy coalitions are more likely to successfully pursue austerity through far-reaching cuts to the welfare state. This leads us to our final hypothesis: H3b: When passed by oversized coalitions or minority governments, fiscal consolidations are associated with more pronounced decreases in public social expenditures than when a single party majority or minimal winning coalition government is in power. Data and Operationalization The focus of this paper is on the degree to which fiscal consolidation implies welfare state retrenchment. In other words, is the welfare state conditioned by the partisan complexion or 9

11 the type of government at the core of policies intended to reduce budget deficits and debts? Against this background, it becomes apparent why we conceptualize welfare state retrenchment exclusively in quantitative budgetary terms. We view the welfare state as simply one of the most important spending categories of government budgets and for its redistributive function the size of this category tends to be strongly associated with income inequality. The welfare state s more qualitative features, such as its institutional structure, its program coverage, or generosity, are not at the core of this paper. Our dependent variable is therefore operationalized as the annual change (first differences) in total public social expenditures as a percentage of GDP, as provided by the OECD. To measure social expenditure as a share of GDP is conceptually superior over measuring it as a share of total public spending. Imagine a country implementing a uniform 10% cut to all spending categories (that is, both welfare- and non-welfare-related budget items are equally hit). The net effect of fiscal adjustment on social expenditure as a percentage of total expenditure would then, misleadingly, be zero. The use of change rates (as opposed to levels) is, from a theoretical point of view, implied by our research question: we wish to know what changes during and after fiscal adjustment the size of the welfare state per se (as measured by levels of public social expenditures) is not of much relevance to us. A methodological justification for this operationalization is provided below. OECD public social expenditure data is available from 1980 onwards. Since we are using first differences and including a lagged dependent variable (LDV) in our statistical models, our observation period begins in 1982 (1991 for Austria). Our main independent variable is the occurrence of fiscal consolidation, but we also distinguish between different sizes of consolidation. The standard approach in comparative political economy to operationalize fiscal consolidation starts from changes in countries primary budget balance (Wagschal & Wenzelburger, 2008b, 2012) or its cyclically adjusted version (CAPB) (Alesina & Ardagna, 2009). A noticeable improvement in this balance (i.e., a sharp reduction in budget deficits) indicates fiscal adjustment (FA). For example, Alesina and 10

12 Ardagna (2009, p. 41) code a year as one of FA if the budget balance improves by 1.5 percentage points of GDP. More recently, however, the literature on the topic has become skeptical about the validity of this operationalization of FA based on the primary balance or CAPB (IMF, 2010b, p. 96; Wenzelburger, 2009). These indicators measure the outcome of consolidations, instead of looking at consolidation policies which are at the core of our theoretical interest here, and may or may not have caused measured outcomes. The CAPB is seen as a remedy to capture discretionary policy changes by adjusting primary balances for the effects of the business cycle. The methods to correct for cyclical changes, however, are usually inconsistent across different data sources and suffer from measurement errors that are likely to be correlated with economic developments (IMF, 2010b, p. 96). And even if an exact cyclical correction were possible, the political motivations behind the resulting figures would still be disregarded. To sum up, the overall validity of such outcome-based indicators of fiscal adjustment policies is questionable. In consequence, a different approach aims at explicitly identifying tax hikes and spending cuts passed by political actors at the general government level. This so-called historical or action-based approach underlies the work by Devries et al. (2011), who collected a dataset of policy measures (i.e., this is no data on budget balances/deficits) motivated explicitly by fiscal consolidation (rather than by restraining aggregate demand during periods of strong growth, for instance) in 17 OECD countries between 1978 and The authors identify a total of 173 country-years of FA based on policy documents like central bank reports, budget speeches, OECD, IMF, and EU sources, and provide estimates of the budgetary impact of these measures in percent of the GDP. Note that these measures were not only planned, but actually taken by governments; only their budgetary impact is based on estimates. We modified this data, based on the qualitative country information provided by Devries 2 These countries are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, Portugal, Spain, Sweden, United Kingdom, and the United States. 11

13 et al. (2011), to include only those measures that make permanent changes to the budget. Temporary and one-off measures such as a temporary rise in the VAT rate, or one-off proceeds from some privatization of a public enterprise, for instance are omitted. This reduces the number of FA cases from 173 to 164. Moreover, in order to ensure that FA measures are attributed to the correct government i.e., to the government that passed a measure as opposed to the (possibly different) one that presided over its implementation we modified the relevant FA and/or partisanship variables whenever the partisan complexion of the cabinet changed while a fiscal consolidation program was underway. Note that our empirical findings are substantively unaffected by these modifications, which is important since the main effect of FA on social spending is more precisely estimated based on the unaltered data. Take the French case for an illustration of our changes: Devries et al. (2011) report fiscal adjustment measures totaling 1.33% of GDP in 1996 and 0.2% of GDP in 1997 (excluding a temporary increase in corporation tax for business of 0.3% of GDP). Those measures were part of the social security reform adopted by the centerright government of Prime Minister Alain Juppé in 1996 and thus should not be attributed to the socialist government of Lionel Jospin that came to power in June We therefore change the relevant (permanent) FA measures from 1.33% to 1.53% of GDP in 1996 and from 0.2% of GDP to zero (= no FA ) in France is the only case where these modifications resulted in an FA measure crossing the 1.5% threshold (discussed below) or being eliminated altogether. An overview of all our changes (for 11 out of 164 country-years in total), which are based on qualitative country information provided by Devries et al. (2011) and several issues of the EJPR Political Data Yearbook, is available upon request. Since our analysis is restricted to the years after 1981 due to missing data on the dependent variable, the number of FA cases analyzed in this paper is reduced further to 149, leaving 317 no-fa-cases being coded as zero on the otherwise continuous FA variable. This continuous variable is thus censored in the sense that it does not or only 12

14 poorly distinguish between no-fa-cases (including fiscal expansions, which are also coded as 0) and FA of zero or very small size. Note that these could result from a spending cut that was offset by a tax cut, or vice versa (18 observations in our dataset). In order to distinguish more precisely between no FA cases on the one hand, and those of different sizes on the other, the data enters our statistical analyses in three dummy variables: a general FA dummy that equals 1 for all 149 FA cases in the sample and 0 otherwise, and two dummies to identify large and small fiscal adjustments ( FA small and FA large ). Using continuous versions of these dummies 3 does not, however, alter our findings. In line with the most commonly used threshold in economic research (see Alesina & Ardagna, 2009; IMF, 2012, p. 53; Wagschal & Wenzelburger, 2008a, p. 17 f.) consolidations exceeding 1.5% of GDP qualify as large. Table 1 provides an overview of all fiscal adjustment cases in our sample. [Table 1 about here] The other main independent variables are the political complexion of government on the one hand, and the breadth of the policy coalition on the other (data from Armingeon et al. (2013)). With respect to government partisanship, we use data for the percentage of cabinet posts occupied by left, centrist, and right parties. From these, we constructed three dummies, coded 1 if respective left, centrist, or right parties held more than 50% of cabinet posts in a given countryyear, which indicates whether a specific party ideology dominated the government. While cabinet shares do not always exactly reflect the real distribution of power within governments, our 3 In its original form, our continuous FA variable varies between and Derived from this are the continuous large FA variable, set to zero whenever FA was smaller or equal to 1.5% of GDP, and the continuous small FA variable, set to zero whenever FA was above that threshold (also see IMF (2012, p. 50 ff.) for this operationalization). In order to facilitate estimation of a linear relationship with this highly skewed regressor, we apply a logarithmic transformation (after adding 1). 13

15 dummies indicating an absolute majority of a specific ideology are more likely to do so. Moreover, around 50-70% of all governments in our sample are classified as either 100% left/centrist/right or 0% left/centrist/right even based on continuous cabinet shares. Unsurprisingly, then, using these percentage shares of parties in government (i.e., substituting the dummies for continuous variables) does not alter our findings. The breadth of the policy coalition is operationalized as a dummy based on the type of government. If a government is a surplus coalition in the sense that it has more parties than needed for a parliamentary majority, it is classified as a broad coalition (dummy = 1). Likewise, minority governments usually negotiate policy-specific majorities they would otherwise not be able to shape the legislative process and are thus also classified as broad coalitions. Single party or minimal winning coalition governments, in contrast, are classified as narrow coalitions (dummy = 0). In order to examine the conditional effect of party ideology and coalition breadth on the relationship between austerity and social expenditures, we construct interaction terms between each of our 3 FA variables and the partisanship dummy as well as with the coalition breadth dummy. Our control variables include standard economic factors such as economic growth (data from OECD (2012)), annual changes in unemployment rates (OECD, 2010b), and economic openness (Heston et al., 2012). In the short-run, as automatic stabilizers begin to work, poor economic performance leads to higher social expenditures. In the long-run, lower performance is expected to decrease the potential for social spending and can thus be perceived as an indicator for (higher) problem pressure. We control for additional driving forces of social expenditure as identified in the literature (see Kittel and Obinger (2003)). The first is the share of elderly people in the population (data from OECD (2010b)). The second is a measure for the funding structure of the welfare state, constructed as social security contributions divided by total tax revenues (data from OECD (2010a)). This contrasts tax-based versus insurance-based welfare 14

16 regimes, where the former is expected to be affected more directly when governments consolidate (Kittel & Obinger, 2003, p. 31). Finally, we control for institutional arrangements by including the Rae index of legislative fractionalization of the party system (data from Armingeon et al. (2013)). Details on the operationalization and sources of all variables are found in the appendix. Method Taking all data availability restrictions into account, our empirical analysis is based on a timeseries cross-sectional (TSCS) dataset comprising 466 country-year observations between 1982 and 2009 (the data is balanced except for the missings in Austria). As discussed above, we have theoretical reasons to operationalize our dependent variable as first differences. There is, however, also a purely methodological justification for this approach. First, when specifying the dependent variable in levels, our data suffers from unit heterogeneity a problem that is commonly solved by adding unit dummies to the model, i.e., fixed-effects estimation. Second, the level-version of the dependent variable has a unit root (and therefore is non-stationary), as was indicated by a battery of augmented Dickey-Fuller tests for panel datasets. A solution that usually solves both problems simultaneously is to run the regression using first-differences, 4 a strategy we therefore apply for methodological reasons as well. Most of our control variables also enter the analysis in first differences, while we stick to levels whenever it made sense theoretically and when the level-variable turned out to have more explanatory power that the corresponding operationalization in changes. 4 An alternative specification commonly used in the context of non-stationary data in time-series- and TSCSanalysis estimation based on an error correction model (ECM) is not possible with our data since the crucial assumptions behind this method do not hold: neither are the time-series in question both stationary, nor are they co-integrated (and thus both have unit roots) (De Boef & Keele, 2008). Rather, our dependent variable (in levels) has a unit-root, while our main independent variables of interest are stationary. 15

17 A common downside of a model specification in first differences is the implicit assumption about the temporal effect of a change in x on y i.e., that a change in x causes a change in y only once (instantaneously or with a constant lag) and then fades immediately. This usually prevents the researcher to draw inferences about the long-term consequences of policy changes that are central to much of the research in comparative political economy. We mitigate this problem by estimating a distributed lag model where we include our FA variables not only at t, but also with a one-year as well as a two-year lag (t-1 and t-2). With respect to the lag structure of the control variables, we follow the recommendation by De Boef and Keele (2008) and start with a general model (i.e., include contemporaneous variables as well as their lags) and impose restrictions only when empirically justified i.e., when the respective variable did not have any substantial explanatory power. 5 We include a lagged dependent variable (LDV) to address serial correlation and apply simple OLS estimators to the pooled TSCS data. 6 In a first step, we thus estimate autoregressive distributed lag (ADL) models, formally specified as = (1) " + ( ) +! +# where represents the change in total public social expenditures in country i from time t-1 to t, is the constant, and is the coefficient of the LDV. stands for either the partisanship 5 We settled on a maximum lag length of two years for all explanatory variables, partly because we did not find any significant effect afterwards (for the FA variables), partly to keep the model as parsimonious as possible (for the controls). 6 Serial correlation was detected with a Lagrange multiplier test. Using simple OLS instead of panel estimators is justified based on a Breusch-Pagan Lagrange multiplier test indicating no significant differences across units (no panel effect). We do not include country- or time-dummies since doing so did not lead to substantively different results. Finally, we have no indications for severe problems with multicollinearity: the highest VIF in any of our models was When keeping in mind that distributed lag models by construction include variables measured at different time points, this does not seem extraordinarily high. The highest average VIF in any of our models is 3.2. All analyses were run in Stata

18 dummy or the dummy indicating the breadth of the policy coalition (that is, we estimate two separate series of models).,, and are three coefficients each (i.e. at t, t-1, and t-2) for the relevant FA dummy (FA, FA large, FA small), the relevant partisanship dummy (results reported for left governments only) or coalition dummy, respectively, and the relevant FA*partisanship or FA*coalition interaction term. Moreover, are coefficients for our six control variables at a maximum of three time points each. Finally, # is an idiosyncratic error. Our theoretical interest clearly is in the more long-run aggregate impact of fiscal adjustment on change rates in social expenditures. In a second step of our empirical analysis, we therefore proceed by calculating this long-run effect (known as the long-run multiplier (LRM) in the time-series literature) of fiscal adjustment from the coefficients of our contemporaneous and lagged FA variables as well as from the relevant partisanship or coalition interaction terms. Substantively, the resulting battery of LRMs (for each of our FA variables under governments of different partisan complexion or different coalition breadth) give the total effect of fiscal adjustment on change rates in social expenditures over a period of 3 years i.e., the immediate effect (impact multiplier) plus the effect that occurs with a one-year and a two-year lag. Slightly adjusted to the context of our analysis, the formula for calculating the LRM of the FA variable in the ADL model of equation (1) is given as $%&0 = ( = ( + + )/ (1 ) when the relevant government partisanship or coalition dummy equals zero, and as $%&1 = ( = ( )/(1 ) when it equals one (i.e. the LRM of the interaction term) (De Boef & Keele, 2008). In order to obtain the standard error for these LRMs, we follow the procedure described by Wooldridge (2013, pp ). That is, we solve the LRM formulas for and to obtain = ( (1 ) and = ( (1 ) and substitute that for and in equation (1), which leads to equation (2). Finally, equation (2) is estimated with the estimate for being obtained from equation (1) in advance. 17

19 = + +( (1 ) + ( ) + +( (1 )( ) (2) " + (( ) ( ) ) +! +# Unfortunately, it was not possible to test our partisanship hypotheses (H2 and H3a) and the coalition breadth hypothesis (H3b) in a single comprehensive equation. This is because proper modeling would have required specifying a series of three-way-interaction terms (and computation of LRMs for these terms), which would have resulted in much more complex and much less interpretable models. Even more important, such a comprehensive specification would have significantly reduced the number of observations in the subgroups defined by the threeway-interactions. For example, there are only 4 cases in our data where a (i) broad coalition dominated by (ii) left parties passed a (iii) large fiscal adjustment program (see Table 1). In other words, data limitations would have prevented us from drawing robust inferences from these comprehensive models. We therefore stick with two separate series of models, to be discussed in the next section. Empirical Findings Tables 2 and 3 show the results of our main regression analyses. In Tables 4 and 5 we present the marginal effects of fiscal adjustments under left-/non-left governments and under broad/narrow policy coalitions, respectively. We do not report any results for governments dominated by centrist or right parties, since the respective models did not yield any additional consistent or substantive results. Moreover, we do not discuss the results for the controls as long as these are 18

20 in line with our theoretical expectations. Across both Tables 2 and 3, models (1) to (4) include all 466 country-year observations, while model (5) excludes the full time series of the three Nordic countries in the sample (Denmark, Finland, and Sweden). Similarly, Tables 4 and 5 present results for the restricted as well as for the full sample. [Table 2 about here] [Table 3 about here] As indicated by the coefficient for the LRM in model (1) in both tables, hypothesis 1 is strongly supported. Fiscal consolidation is associated with shrinking public social expenditures over a period of three years at least (i.e., an immediate effect plus effects with a one-year and a twoyear lag). In total, the change rate of public social expenditures is reduced by 0.28 over the period when a fiscal consolidation program was enacted (summary statistics for all variables are found in the appendix). The magnitude of this effect is comparable to the immediate positive effect that a 2 percentage point increase in the unemployment rate has on social expenditures (likely due to the associated increase in expenditures on unemployment benefits). As indicated in model (3) in both tables, large fiscal adjustment are not surprisingly associated with larger reductions in social expenditures than small adjustments are. More important, however, are the results of model (2) in Table 2 in combination with the results shown in the first row of Table 4. They lend support for hypothesis 2: we find no systematic impact of the partisan complexion of government on the relationship between austerity programs and the welfare state. The negative effect of FA on social spending is indicated to be even stronger under left than under non-left governments, but the difference between those two groups is insignificant (see first row, last column pair of Table 4). Functional requirements indeed seem to make all parties behave similarly in periods of fiscal consolidation. 19

21 [Table 4 about here] [Table 5 about here] What we also find based on Table 4 (see also the underlying models (4) and (5) in Table 2), is that in cases of large fiscal adjustments, left parties seem to be more successful in reducing welfare state spending than their non-left competitors. Looking closer, however, it turns out that this effect is mainly driven by the three Nordic countries in our sample. If we exclude these cases, the coefficient is no longer significant. 7 In other words, the Nixon-goes-to-China logic is confined to the strong left parties found in Nordic countries alone apparently lending some support for hypothesis 3a, at least with respect to large fiscal consolidation programs. But what about our alternative argument, that the breadth of the policy coalition in government determines the degree of retrenchment during fiscal consolidation, rather than the partisan complexion of government? The relevant results are reported in Tables 3 and 5 and first show that the effects of the policy coalition interactions are very similar to, but consistently stronger than those of the partisanship interactions discussed before. More importantly, however, our policy coalition results are not driven by the Nordic countries or by any other individual country in the sample. Moreover, the three crucial Nordic cases for which we did find a partisan effect turn out to be a subgroup of cases with broad policy coalitions. Since the 1960s, Sweden and Denmark have most frequently been governed by single- or multi-party minority governments; Finland by oversized party coalitions (data from Armingeon et al. (2013)). The qualitative evidence on retrenchment episodes in these countries (see e.g. Anderson (2001), Kuhnle (2007), Obinger et 7 In our empirical models, Sweden has the strongest influence on the results. The reported findings without Sweden, Finland, and Denmark are similar to excluding Sweden only. 20

22 al. (2010), or Alexiadou (2013, p. 709)) also shows that very broad policy coalitions among the major parties, supported by moderate trade union movements, were conducive to implementing unpopular welfare state reforms. We conclude that the conditional partisan hypothesis 3a is at best only weakly supported by empirical evidence. In contrast, all our analyses show a strong and significant effect of the breadth of the policy coalition on the degree of welfare state retrenchment in times of austerity. There is no difference between single party or minimal winning coalitions and broad policy coalitions in cases of small fiscal adjustments. If however, governments implement large austerity programs, hypothesis 3b finds strong support. 8 Of the control variables, all significant effects confirm our theoretical expectations, with one exception: higher shares of the elderly in the population are related to social expenditure cutbacks. This may reflect the long-run problem pressure of aging societies on the welfare state, in particular on health and pension expenditures, the largest items of social spending. In any case, all our main findings are substantively unchanged if we use changes of non-age-related public social expenditure as our dependent variable. We ran several robustness tests for our analyses (regression estimates available upon request). Among others, our findings are robust against including a number of additional controls, such as institutional variables for federalism and bicameralism, a dummy for Eurozone membership, or an interaction of the welfare regime (tax- vs. insurance-based) with our FA variables. None of these altered our results or were of any significance to our models. Moreover, our findings are robust to alternative operational definitions (i.e., continuous variables instead of dummies) for the partisanship or FA variables. But could there possibly be some endogeneity in our explanation? It may be the case that left parties attempt to avoid the adoption of an austerity program at all cost. Likewise, one 8 We find the same effect for fiscal adjustment programs in general, i.e., when we do not distinguish between large and small FA. This effect is not however robust against the exclusion of the Nordic countries. 21

23 could expect broad policy coalitions to exhibit a higher propensity for large fiscal adjustments, which is in turn correlated with a greater reliance on social spending cuts. To address this issue, we refer to the findings of our ongoing research (Authors (2014; but see also Hübscher and Sattler (2014) on the determinants of fiscal adjustments). These findings point to complex and conditional relationships between governing parties, the size of the governing coalition and the likelihood and size of fiscal adjustment programs, but none of these give rise to serious endogeneity concerns. 9 In particular, left governments are found to be no more or less likely to pursue austerity than non-left governments are, and neither do they implement consolidations that are any larger or smaller on average. Fiscal adjustments under broad coalitions are less common but larger (also see Table 1). However, as an additional test against endogeneity, the findings presented in this paper are substantively unchanged when we, in a stepwise procedure, exclude the largest adjustment cases adopted by broad coalitions in the large FA group, until consolidations under narrow and broad coalitions in that group are of the same size on average i.e., not statistically significantly different from one another. What both Hübscher and Sattler (2014) and our own ongoing research strongly suggest, is that functional requirements (as measured e.g. by fiscal pressure and unemployment) are crucial triggers of austerity. Conclusion When governments pursue a fiscal consolidation program, do left parties design austerity measures differently than their centrist and right competitors in terms of the degree to which 9 We distinguish between the decision to pursue austerity policies on the one hand, and the size of those consolidations on the other. While minimal coalitions are more likely to initiate adjustments than broad coalitions are, those programs are smaller on average. This effect is however different for governments dominated by right parties, which have a higher likelihood of FA under broad coalitions. In turn, the size of the adjustment under right governments is smaller as compared to non-right governments. 22

24 they target the welfare state? And does the type of government condition the extent to which austerity policies rely on social spending cuts? These were the questions that guided our paper. With respect to the first, the answer is no. Once the decision to consolidate public finances is made, there is generally little room left for the left. Irrespective of which party governs, the welfare state is not spared from spending cuts. This is in essence a functionalist logic: in the hard times of austerity, problem pressure and functional requirements seem to make all parties alike. We add to this, however, that even if all parties are inclined to behave similarly, functional requirements and resources need to be cognitively identified by policymakers, translated into policies, and implemented. This is most relevant in the context of very large consolidation episodes, when the room of maneuver for policymakers is severely constrained. Initially, we seem to observe a Nixon goes to China scenario and a conditional partisan effect in our data: when the fiscal adjustment program is large, the strong Social Democratic parties of Sweden, Finland, and Denmark the founders of the Nordic welfare model in their historical alliance with trade unions implement more pronounced cuts to social spending than their competitors to the right. Upon closer look, however, it seems that the crucial factor behind the parties ability to cut welfare spending is not their reputation as long-standing welfare state founders and defenders which would imply a partisan logic but rather their tendency to build broad multi-party pro-reform coalitions, be it in the form of a minority or a surplus government. And while broad coalitions are a common feature in the Nordic countries, their consequences for the design of austerity policy apply to other countries and contexts as well. The answer to our second research question is therefore yes: a broad pro-reform coalition is more likely to engage in welfare state retrenchment during a large fiscal consolidation episode than a narrow coalition such as a single party or minimal winning coalition government. 23