Capital Mobility, Veto Players, and Redistribution in Latin America During the Left Turn^*

Abstract

This article offers an explanation for the intensity of redistributive policies from Latin American leftist governments during the so-called 'left turn'. It challenges the idea that the recent radical leftist governments are the product of region-specific characteristics. Based on theoretical models that analyze the implications of inequality in democracies and the moderating effects of different economic and institutional structures, it argues instead that the left's redistributive policies were more intense in countries where capital mobility is low and there was no pro-elite legislative veto player. To evaluate this explanation, we offer original evidence from time-series -cross-section regression models of social spending, based on data from thirteen Latin American countries over the years 2003 to 2015. The results show partial support for the explanation: there is strong evidence that social spending decreased with the ideological distance between the president and the pro-elite veto player, but only weak evidence that capital mobility attenuated this effect.

Redistribution; social spending; capital mobility; veto player; Latin America; left turn

Most Latin American democracies experienced significant political and social transformations at the dawn of the twenty-first century. After a decade dominated by policies that prioritized pro-market economic reforms and rejected state-led redistribution, beginning at the end of the 1990s a left turn swept through the region, with the electoral victory and subsequent implementation of policy agendas emphasizing social spending and redistribution. Consequently, income inequality fell substantially in practically all the countries governed by the left (CORNIA, 2014; LÓPEZ-CALVA and LUSTIG, 2010).

Leftist policies produced social impacts via different paths, however. According to a significant part of the contemporary literature on Latin America, democracies recently governed by the left can be divided into two groups, according to those paths (CASTAÑEDA, 2006; WEYLAND, 2009). The first group is formed of countries such as Brazil, Chile, and Uruguay, in which presidents were elected from traditional left-wing parties, defending alternative policies without antagonizing the prevailing economic order. Such governments enacted moderate redistributive policies, mainly guaranteeing the provision of a minimum income to the poorest sectors of society, without advancing major changes to the structures of taxation and government transfers (HUNTER and SUGIYAMA, 2009; PRIBBLE and HUBER, 2013). In such cases, the left's success among voters and the subsequent implementation of redistributive policies came about without affecting political stability. This path is similar to the one taken by various social democratic parties in consolidated democracies, which, for many analysts, constitutes the expected pattern of evolution.

The second group includes countries such as Bolivia, Ecuador, and Venezuela, where a new left emerged in a context of crisis of the traditional party system and explosive growth of social movements, subsequently electing presidents who emphasized a break with the economic order. These leftist governments implemented much more intense redistributive policies, substantially increasing both taxation of the richest sectors of society and public transfers to the poorest (CORRALES, 2010; KENNEMORE and WEEKS, 2011). In such cases, the emergence of the left is associated with a radicalization of the political process and of government, accompanied by political instability in the form of violent protests, secession attempts, and even coups (RITTINGER and CLEARY, 2013). According to a substantial number of studies on the subject, such a path does not fit with either the theoretical models or empirical patterns associated with the experiences of consolidated democracies. Rather, it constitutes a specific characteristic of the region, related, for example, to its tradition of populism or to transformations undergone by its party systems.

In this article, we question the idea that Latin America's recent radical leftist governments¹ reflect characteristics specific to the region, arguing instead that they constitute empirical manifestations of more general processes of democratic consolidation in a context marked by, on one side, acute inequality and, on the other, an economic and institutional structure conducive to intense redistributive policies. Our claim is that although Latin American countries entered this century with high levels of inequality, and practically all of them experienced favorable macroeconomic conditions during most of its first fifteen years or so, leftist governments implemented radical redistributive policies only in countries where there was relatively low capital mobility and the legislature was not a veto player. The theoretical foundations for this claim come from theories about how inequality affects the intensity of redistributive policies in democracies (MELTZER and RICHARD, 1981), its implications for economic growth (ALESINA and RODRIK, 1994), and how this effect is moderated by different economic and institutional structures (BATES and LIEN, 1985; PERSSON and TABELLINI, 1994). However, we must emphasize that our aim is not to offer a new explanation for what are considered Latin American political idiosyncrasies, but rather to demonstrate that well-established theories about the interaction between the political and the economic realms in consolidated democracies remain useful, with only minor adaptations, for explaining policy differences among recent leftist governments in Latin America.

We assess our claim by testing its implications for levels of government social expenditure, using time-series-cross-section (TSCS) data from the thirteen countries which comprise the so-called left turn in Latin America, over the years 2003 to 2015. This period roughly coincides with the so-called commodities price boom, during which the region enjoyed above average economic growth rates, thus facilitating redistribution. Since our goal is to explain why the left adopted radical redistributive policies in some countries but not in others, our sample only includes the countries where at least one leftist president was elected. We use the level of social expenditure (data from the Economic Commission for Latin American and the Caribbean (ECLAC)) as proxy for the government's redistributive effort, as have several previous studies (e.g.: AVELINO, BROWN and HUNTER, 2005), according to McCarty and Pontusson (2009, p. 666). Evidence from TSCS linear models with country-specific random effects and Driscoll-Kraay standard errors lends partial support to our explanation: we find strong evidence that social spending decreased with the ideological distance between the president and the (pro-elite) legislative veto player, but only weak evidence that the degree of capital mobility attenuated this effect.

The remainder of this article is organized into four sections. The next section provides a brief review of the recent literature on the Latin America's left turn, outlining different explanations for the variation in policies among leftist governments. The third section selectively discusses the theoretical literature on inequality and redistribution, specifically those studies that analyze the redistributive implications of democracy and the moderating effects of economic and institutional structures. Based on the main contributions to this literature, we offer an explanation for the variation found in redistributive policies implemented by recent leftist governments in the region. The fourth section evaluates the empirical validity of the explanation. The fifth and final section briefly summarizes the main findings and offers some final considerations.

Brief literature review

Latin America's left turn has inspired many studies addressing different aspects of the phenomenon. One notable characteristic of these studies is the way in which they depart from the traditional view in the comparative literature, which supposes that accurate accounts of political phenomena in Latin America require the formulation of new, regional-specific theories². It is our purpose to advance and improve upon this scientific program, embodied in the institutionalist and the political economy paradigms, which analyzes the Latin American case using an integrated theoretical lens capable of accounting for as many cases as possible.

Our focus is on the policies implemented by the left. Here, studies vary in terms of their emphasis on similarities and differences. Those pointing to common elements of the Latin American left highlight a common agenda centered on equality, redistribution, and social inclusion (ARDITI, 2008). Studies emphasizing differences, on the other hand, tend to draw attention to the actual economic and social policies, usually classified as either moderate or radical (CASTAÑEDA, 2006; WEYLAND, 2009). The moderate left preserves the pillars of the market economy and does not interfere with prices and contracts, tackling poverty in a gradual fashion, with fiscal prudence and respect to private property. The radical left, on the other hand, intervenes in the economy, controlling prices and nationalizing companies, and seeks to reduce poverty by drastically raising public spending and intensely redistributing income.

The literature offers various explanations for such differences among leftist governments, some being of a political and others of a more economic nature. Among the former, studies initially focused on the evolution of the party system. According to this perspective, left radicalism in government results from the party system's low degree of institutionalization (FLORES-MACÍAS, 2010) or from the electoral context in which left-wing parties emerged and consolidated (MADRID, 2010). Regarding the first explanation, however, Weyland (2009, p, p. 150) argues convincingly that the causal relationship is in fact the reverse of that postulated, with the emergence of radical leftist parties contributing to the destabilization of the party system. Concerning parties' electoral trajectories, although this may eventually explain ideological differences, it is not sufficient to account for variation among leftist policies, as Campello (2015), Murillo, Oliveros and Vaishnav (2011) and Weyland (2009) have shown.

Another explanation in line with low institutionalization arguments is found in Biglaiser (2016), according to whom far-left policies were adopted where the electorate granted the president a clear mandate, in the form of a wide margin of votes and a legislative majority. The argument is that this type of electoral result is a strong sign that the electorate will reward legislative support for the presidential agenda, thus increasing/decreasing defections among the opposition/government's base. Despite a lack of solid theoretical foundation, this explanation is consistent with the idea that democracies in the region are 'delegative' in nature (O'DONNELL, 1994), in the sense that presidents enjoying the support of a large majority of voters can govern virtually without political or institutional constraints.

Adopting a more institutionalist approach, Nunes (2014) argues that the radical leftist governments emerged due to the absence of at least one of two moderating political mechanisms: the occurrence of a runoff in the presidential election and the existence of an effective veto player in the legislature. While the first generates policy moderation by requiring that candidates win the support of the median voter, the second does so by requiring minority presidents to obtain support from opposition parties. Nevertheless, the first mechanism does not seem to operate as the author suggests. The theoretical results presented by Bordignon, Tommaso and Tabellini (2016) suggest that it is not the occurrence of a runoff that generates moderation, but rather the two-round rule. Considering that there are only two cases of leftist presidents elected by a simple majority under one round of balloting (Paraguay's Lugo and Nicaragua's Ortega, in 2006), the empirical relevance of Nunes' (2014) first mechanism appears highly doubtful.

Among the economic explanations, there is a virtual consensus that the macroeconomic climate in Latin America during the first decade of the twenty-first century favored the expansion of social policies, mainly due to the sharp income growth spurred by a significant and persistent increase in commodity prices. Weyland (2009) identifies this phenomenon as the main cause of the different intensities of leftist governments' redistributive policies, arguing that radical policies were adopted only in countries in which the commodities sector carried greater weight (as in Bolivia, Ecuador, and Venezuela). Additionally, Murillo, Oliveros and Vaishnav (2011) hypothesize that the higher capital inflow due to the increasing value of exports granted leftist governments the latitude to redistribute. On the other hand, Campello (2015, pp. 19-20) argues that integration into international financial market mitigated the left's redistributive policies.

What does the accumulated empirical evidence say? Among the political explanatory variables, the most consistent result is a negative correlation between the level of institutionalization of party systems and the intensity of leftist policies (FLORES-MACÍAS, 2010, p. 423; NUNES, 2014, p. 51). As previously mentioned, however, it is highly probable that the causal direction underlying this relationship is inverse to that postulated. With respect to the effect of the legislature, the findings to date cannot be considered conclusive, due to their contamination by measurement problems. Nunes (2014) measures the existence of a legislative veto player by means of an indicator that the parties are well structured, without considering their ideological location. Murillo, Oliveros and Vaishnav (2011) takes into account the ideology of the opposition in the legislature, but not of the legislative median, which is the crucial actor in the veto player theory. Biglaiser (2016), in turn, combines the president's legislative status with her margin of victory at the polls to gauge the nature of her mandate, which impedes an evaluation of the contribution of each of those variables to the strong positive correlation that the author detects between the electoral mandate and the intensity of leftist policies. Among the economic explanations, the only strong and consistent evidence is the positive correlation between the importance of rents from commodities and how far to the left the government's policies are (CAMPELLO, 2015; MURILLO, OLIVEROS and VAISHNAV, 2011; WEYLAND, 2009).

We consider the institutional explanation based on the veto player argument theoretically sound but empirically fragile, given the current limitations of the evidence. On the other hand, notwithstanding the strong (positive) association between the size of the commodities sector and the intensity of leftist policies, we find the economic explanation unsatisfactory since it cannot account for the political conflicts that accompanied the implementation of intense redistributive policies. For these reasons, in the next two sections we offer a modest contribution to the debate. First, we identify the causal connections proposed in the literature between the intensity of redistributive policies and the institutional and economic structures, and, based on these, build our explanation for the policy differences between Latin American leftist governments during the recent commodities boom. Then, we evaluate these relationships empirically.

Inequality and redistribution

In this section, we briefly present the micro foundations of the relationship between inequality and redistribution in democracies, to understand why leftist governments seeking to reduce inequality adopt redistributive policies with very different levels of intensity. Despite the large literature on the subject, for our purposes it will suffice to make use of the main insights from the seminal works of Meltzer and Richard (1981), Alesina and Rodrik (1994), and Persson and Tabellini (1994).

Imagine a perfectly competitive closed economy with no uncertainty, formed of individuals that only differ in terms of their relative endowments of capital (either physical or human) and (unqualified) labor, and whose income is an increasing function of the former. All labor income is consumed, but not necessarily that derived from capital, which may be invested. Any increase in production occurs via an increase in labor productivity, which in turn requires investment. Suppose also that there is a government elected by majority rule and according to universal suffrage, which directly provides basic infrastructure and services that benefit all citizens and are financed by a flat tax levied on capital income only. The government has the option of redistributing income, through a mechanism that consists of charging a surplus tax on capital income and transferring the corresponding revenue to all individuals, equally. The level of the surplus tax – i.e., the intensity of the redistributive policy – is the government's only choice variable and the only issue guiding voters' decision. The economy and politics operate in two periods: in the first (t), capitalists make decisions on investment and individuals elect a government; and in the second (t+1) individuals receive their earnings, the government defines and collects the surplus tax on capital income, and then distributes the corresponding revenue.

First, consider the decisions made by individual voters when electing a government. As the tax is only levied on capital gains and income is increasing on such yields, voters with lower income prefer more intense redistribution. Considering that elections are held under majority rule, the candidate elected is the one who promises to implement the redistributive policy preferred by the voter with median income. How intense is this policy? Imagine the usual scenario in which income is concentrated among the richest, meaning that the median income is lower than the average. In such a case, the median voter pays surplus tax in the amount of Ƭy^M and receives transfers of ý, Ƭ,y^M where , y^M and ý, represent the surplus tax, the median income, and the average income, respectively. Since ý y^M, (by assumption), the median voter will favor redistribution, though only up to the level in which her personal income equals the average income - beyond that, she would receive less in transfers than she pays in taxes (MELTZER and RICHARD, 1981).

The expectation of redistribution negatively affects investment decisions, as the surplus tax reduces the expected net income from capital. A reduction in investment implies a lower increase in labor productivity, negatively affecting the median voter's final income, because both labor and capital incomes (and therefore transfers) increase less. Taking these effects into account, the median voter prefers a more moderate redistributive policy, below that necessary for equalizing the median and average incomes post transfers (ALESINA and RODRIK, 1994). Notice, however, that the median voter (and the government elected) is subject to a time consistency problem. Suppose that capitalists make their investment decisions (in t) based on the winning policy platform, which consists of moderate redistribution. Once the investments have been made, however, the median voter benefits from the opportunistic implementation of the more intense redistributive policy (in t+1), i.e. the one that equalizes the median and average incomes. Given that the government seeks to please the median voter (for reelection purposes), it implements the more intense redistributive policy. Anticipating the government's opportunistic behavior, capitalists make investment decisions not according to the redistributive policy the median voter prefers in t, but rather the one which maximizes her utility in t+1, which is more intense. Despite the lower investment in t producing a less favorable result for the median voter, there is no way of avoiding it because the government cannot credibly commit to the moderate redistributive policy. The equilibrium policy is therefore more intense and less efficient.

In sum, higher (initial) income inequality generates (the expectation of) more intense redistributive policy, which leads to less investment, and finally, less economic growth. However, as we discuss next, the first nexus of this causal chain is conditioned by economic and institutional structures.

Capital mobility

The capitalists' strategic reduction in investment in anticipation of an excessive surplus tax is exclusively based on the premise that capital may be consumed instead of invested. Actually, the theoretical literature that analyzes the relationship between taxation and investment tends to work with models including multiple politically independent economies and the premise that capitalists can shift their assets between them. In this scenario, the lower the cost paid by capitalists to transfer their assets, the greater the government incentives for increasing the net rate of return on investment – in other words, for minimizing the surplus tax on capital (BATES and LIEN, 1985; BRETSCHER and HETTICH, 2002).

The cost of shifting or reallocating capital defines its mobility. Economies with low capital mobility are those mostly based on fixed assets or real estate, particularly investments in activities linked to the exploitation of natural resources, such as agriculture and mining. Economies based on manufacturing industries with intensive human capital are said to have intermediate capital mobility. Capital is more mobile when it assumes the form of a financial asset, allowing it to easily migrate to another country, especially in the current context of strong international integration.

Given that capital flight implies subsequent lower economic growth, negatively impacting the income of the median voter, governments from countries whose economies have a high degree of capital mobility are expected to adopt more moderate redistributive policies (BOIX, 2003, Chap. 02). However, this effect may not obtain when redistributive policies aim at improving health or education, given "the importance of human capital for employers needing to be competitive in the world economy" (HUBER, MUSTILLO and STEPHENS, 2008, p. 425).

Veto players

The idea that the government cannot credibly commit to more moderate redistribution strongly depends on the assumption that the decision-making process operates according to the majoritarian model of democracy, according to which the institutional structure favors the election of a single-party majority and the implementation of its policy agenda. In this scenario, the median preference in the electorate is converted into public policy almost automatically. This is not the case in the model of democracy known as 'consensual', which seeks to 'share, disperse and limit power' (LIJPHART, 1999). Consensual democracies promote the formation of multi-party government coalitions and afford a more relevant role to the legislature in policy-making. This leads to decision-making processes that contemplate minority preferences, meaning there is no guarantee that the electorate's median preference will be converted into public policy.

One way to distinguish these two models analytically is by making use of the concept of 'veto players' (TSEBELIS, 1995). A veto player is any individual or collective agent whose agreement is necessary to the approval of a policy. Every party that is a member of a majority coalition is a veto player in the definition of the government's agenda, for example. Equally, when the median party in parliament is not the same which controls the executive the legislature becomes a veto player. Pure majoritarian democracies feature a single actor serving as the veto player: the majority party in the legislature, which also controls the government. Consensual democracies, on the other hand, can have multiple veto players. The main general result from this approach is that the production of new and relevant policies decreases when there are veto players with divergent preferences (TSEBELIS, 1995, p, p. 298). One implication of this result is that the institutional structure can serve as a mechanism for the government to credibly commit to a moderate redistributive policy (ACEMOGLU and ROBINSON, 2001; NORTH and WEINGAST, 1989; PERSSON and TABELLINI, 1994). The existence of a veto player representing voters whose income is higher than that of the median voter guarantees that the government will not be able to implement a more intense redistributive policy.

In summary, based on the theoretical literature reviewed in this section, we can conclude that the demand for redistribution increases with income inequality, and the extent to which the former is effectively translated into redistributive policy decreases with capital mobility or the existence of a pro-elite veto player. A variety of empirical studies have been conducted to test these relations. However, the results have proven controversial, mainly due to methodological problems. Examples of studies offering favorable evidence are: Borge and Rattsø (2004) and Milanovic (2000), on the positive effect of inequality on redistributive policy; and Carlsen, Langset and Rattsø (2005), Hallerberg and Basinger (1998) and Hays (2003), on the attenuating effects of capital mobility and veto players. For the specific case of Latin American countries, there is consistent evidence that capital mobility negatively affects the government's redistributive effort (KAUFMAN and SEGURA-UBIERGO, 2001; RUDRA, 2002; WIBBELS, 2006) and suggestive evidence that the existence of veto players restricts the government’s ability to change redistributive policies (HUBER and STEPHENS, 2012)³.

At first glance, these relations seem to offer a promising basis for explaining the intensity of the redistributive policies implemented by the recent leftist governments in Latin America. Figure 01 illustrates how the capital mobility and pro-elite veto player variables relate to the exemplary cases of radical (Bolivia, Ecuador, and Venezuela) and moderate left (Brazil, Chile, and Uruguay) governments. We observe radical redistributive policies only in countries with economies that heavily depend on the commodities sector and in which the left has won an absolute majority in the legislature – in other words, where capital mobility is low and there was no pro-elite veto player⁴. On the other hand, we observe moderate policies in countries with more complex economies (Brazil and Uruguay) or where the median party in the legislature was not left-leaning (Brazil and Chile)⁵.

Figure 01
Capital mobility, pro-elite veto players, and redistributive policy in selected countries

This explanation is partly in line with that provided by Campello (2015, Chap. 02) who uses the same theoretical approach to highlight the attenuating effect of international financial integration on redistributive policies. It also chimes with those proposed by Murillo, Oliveros and Vaishnav (2011), Nunes (2014) and Weyland (2009), in the sense that it emphasizes the attenuating effect of the legislative veto (as in the first two studies) and the dependence on commodities (like in the third). However, our explanation stands out for its emphasis on the effects of both economic 'and' institutional structures, according to which moderate redistributive policies are obtained due to sufficiently high capital mobility or the presence of a pro-elite legislative veto player.

It is also important to point out how our interpretation of the effect of commodity economies on redistributive policies differs from that of rentier-states' (WEYLAND, 2009). According to the former, radical redistributive policies were viable in countries in which the commodities sector carried more weight because capital mobility was lower and, therefore, majority governments (i.e., without a pro-elite veto player) could levy a higher tax on capital. According to the latter, those policies were viable not because the government levied a higher tax, but because it experienced a larger inflow of revenue stemming from the commodities sector, either by means of direct production or of participation in that sector (JENSEN and WANTCHEKON, 2004, p, p. 817; SMITH, 2004, p, pp. 233-234). According to the rentier-state perspective, therefore, governments from commodity-based countries could adopt more intense redistributive policies without soaring redistributive conflict (MORRISON, 2009; SMITH, 2004).

Studies on how leftist governments in commodity-based countries financed their social policies reveal mechanisms consistent with the redistributive approach. According to Hogenboom (2012), Bolivian president Evo Morales increased the public sector's share in the profits from two of the country's largest natural gas fields, as well as the taxes and royalties from private companies producing this commodity, from 18 to 82 percent; Ecuadorian president Rafael Correa practically doubled the tax on windfall profits from transnational oil companies; and Venezuela under Chavez increased royalties and taxes on foreign companies from 01 to 30 percent and 34 to 50 percent, respectively. On the other hand, Letelier and Dávila (2015) show how the Chilean legislature blocked the socialist's government aim to charge royalties on the copper industry right from the beginning of its term. It was only in 2005, five years after its inauguration, that the Chilean government could get legislative approval for the tax change, but only after substituting the royalties for the more moderate alternative of a specific tax on mining profits.

Empirical analysis

In this section, we perform statistical regressions to more systematically evaluate the empirical support for our explanation of the intensity of the redistributive policies implemented by the Latin American leftist governments of the 2000s. To do so, we first need to identify the countries that were part of the left turn.

The movement is widely regarded to have begun in 1998, with the first election of Hugo Chávez to the presidency of Venezuela, after which followed the electoral victories of leftist presidents in several other countries in the region, some of them more than once. However, political events in 2015 and 2016 indicate that the movement may be waning (CASTAÑEDA, 2016). Following Levitsky and Roberts (2011, p. 05), we consider only countries that elected at least one president who: 01. ran for office with an electoral platform centered on social policies, particularly the reduction of inequality; and 02. once elected, preserved a significant part of her platform in government. According to these criteria, we identified 19 leftist presidents in 13 countries, in office between 1999 and 2015. Table 01 lists the countries and their respective presidents, per year of election and period in office.

Table 01
Leftist presidents, Latin America, 1999-2016

Country	President	Party	Year of election	Years in government
Argentina	N. Kirchner	FPV-PJ	2003	5/2003-12/2007
	C. Kirchner	FPV-PJ	2007 2011	12/2007-12/2011 12/2011-12/2015
Bolivia	Morales	MAS	2005 2009 2014	1/2006-1/2010 1/2010-1/2015 1/2015-present
Brazil	Lula	PT	2002 2006	1/2003-12/2006 1/2007-12/2010
	Rousseff	PT	2010 2014	1/2011-12/2014 1/2015-5/2016a
Chile	Lagos	PPD	2000	3/2000-3/2006
	Bachelet	PS	2006 2013	3/2006-3/2010 3/2014-present
Dominican Rep.	Mejía	PRD	2000	8/2000-8/2004
Ecuador	Correa	MPAIS	2006 2009 2013	1/2007-8/2009 8/2009-5/2013 5/2013-5/2017
El Salvador	Funes Cerén	FMLN FMLN	2009 2014	6/2009-5/2014 6/2014-present
Guatemala	Colom	UNE	2007	1/2008-1/2012
Nicaragua	Ortega	FSLN	2006 2011 2016	1/2007-1/2012 1/2012-1/2017 1/2017-present
Paraguay	Lugo	APC	2008	8/2008-6/2012a
Peru	Humala	PNP	2011	7/2011-7/2016
Uruguay	Vázquez	FA	2004	3/2005-2/2010
	Mujica	FA	2009	3/2010-2/2015
	Vázquez	FA	2014	3/2015-present
Venezuela	Chávez	MVR	1998 2000 2006	2/1999-1/2001 1/2001-1/2007 1/2007-1/2013
	Chávez/Maduro	PSUV	2012	1/2013-presentb

• Notes: APC= Alianza Patriótica para el Cambio; FA= Frente Amplio; FMLN= Frente Farabundo Martí para la Liberación Nacional; FPV-PJ= Frente para la Victoria-Partido Justicialista; FSLN= Frente Sandinista de Liberación Nacional; MAS= Movimiento al Socialismo; MPAIS= Movimiento Patria Altiva i Soberana; MVR= Movimiento V [Fifth] República; PNP= Partido Nacionalista Peruano; PPD= Partido por la Democracia; PRD= Partido Revolucionario Dominicano; PS= Partido Socialista; PSUV= Partido Socialista Unido de Venezuela; PT= Partido dos Trabalhadores; UNE= Unidad Nacional de la Esperanza; ^a Impeached; ^b Maduro took office following Chávez’s death, on March 2013.
• Sources: Political Database of the Americas; Murillo, Oliveros and Vaishnav (2011).

Data and methods

Measuring the intensity of redistributive policy (henceforth IRP) is an extraordinary challenge because of the different tools and strategies for implementing such policy. The former includes taxation, social expenditures and regulations - for example, on minimum wages or land reform. The strategies are related to the definition of the pace (either gradual or precipitous), the target population, tax progressivity and compensations. The proper evaluation of different redistributive policies requires considering these various aspects. To the best of our knowledge, however, there is no valid measure of overall IRP.

To use actual income redistribution as proxy for IRP is not without problems when one considers the distinction between policies and outcomes. Since redistribution also depends on the operation of market forces, which are often difficult to identify and isolate, IRPs cannot be inferred directly from observed redistribution. In the case of the recent Latin American leftist governments, although there is reasonable agreement that their redistributive policies have contributed significantly to the reduction of inequality, the magnitude of this contribution is still under debate (CORNIA, 2014; LÓPEZ-CALVA et al., 2014). Alternatively, several studies use social spending as a proxy for IRP, as pointed out in McCarty and Pontusson (2011, p. 666). Although the former is only one aspect of the latter, there are compelling arguments and evidence that social spending strongly reflects the government's redistributive effort (AVELINO, BROWN and HUNTER, 2005; CASTLES and OBINGER, 2007; HUBER and STEPHENS, 2012, , Chap. 03). For these reasons, to test our explanation about the left's IRP we estimate time-series-cross-section (TSCS) regression models of overall social spending as percentage of the Gross Domestic Product (GDP).

Our sample only includes the thirteen countries listed in Table 01, and the thirteen years from the period 2003-2015. This is sufficient for the purpose of our empirical investigation, which is testing the effects of the economic (capital mobility) and institutional (veto player) structures on the intensity of the redistributive policies implemented by Latin American leftist governments during the recent commodities boom. It is important to bear in mind that our goal is simply to evaluate whether these two variables account for differences between the policies enacted by these specific governments, and not to test their relevance for redistributive policies in the region (much less in general). This limited objective derives directly from the scholarly debate about the differences between the radical and the moderate left, as discussed in the first and second sections of this article. Besides, there is no reason to expect any bias from the exclusion of countries where no leftist president was elected, since they cannot add relevant information for the evaluation of our explanation as to why the left adopted radical policies in some countries, but not in others⁶.

Graph 01 shows the evolution of social spending in each of the thirteen countries of the left turn, during the 2003-2015 period. The data is from the Economic Commission for Latin American and the Caribbean (ECLAC). Although there are some gaps in the series - Peru (2003-2004), Paraguay (2003-2006), and especially in Uruguay (2003-2010) and Venezuela (2010-2015) - we can see that social spending increased in almost all countries, most notably in Argentina, Chile, Ecuador and Paraguay. Graph 02 shows the average yearly change (in percentage points) in total social spending by country, during the leftist governments (as identified in Table 01). There are substantial differences: the left substantially increased social spending in most of the countries, though much more intensely in Paraguay (under president Lugo) and Argentina (under Néstor and Cristina Kirchner).

Graph 01
Annual social spending as % of GDP, per selected countries (Latin America, 2003-2015)

Graph 02
Average yearly change in social spending (as % of GDP) during leftist governments, by country (Latin America, 2003-2015)

As mentioned above, out of the total 169 country-year dyads, there are 20 missing values in the social spending series, most of them for Uruguay (eight observations) and Venezuela (six observations). We substituted these missing observations, as well as a few others in the independent variables, using Honaker and King's (2010) method of multiple imputation for TSCS data. Multiple imputation uses the information from the observed portions of the dataset, including many potentially relevant variables, to generate multiple values for the missing cells. Honaker and King's (2010) algorithm considers the usual characteristics of TSCS data, which are heterogeneity and correlation on the longitudinal or the cross-sectional dimensions (or both), thus generating more 'reasonable' estimates for the missing values. In our application, we generated five complete imputed datasets using all the variables included in the regression analyses as well as their first-order lags⁷.

To proper specify a regression model for TSCS data, it is first necessary to evaluate the presence of unit heterogeneity, heteroskedasticity and error dependency. Using the complete dataset resulting from replacing the missing values with the average of the (five) corresponding multiple-imputation values, we found strong evidence of country heterogeneity, heteroskedasticity, contemporaneous and serial correlation, but no evidence of time heterogeneity⁸.

The modelling of country heterogeneity, whether as fixed or random effects, is a controversial issue (CLARK and LINZER, 2015; PLÜMPER, TROEGER and MANOW, 2005). The main purpose of the fixed-effects specification is to avoid bias in the coefficient estimates due to correlation between independent variables and omitted (non-observed) factors that affect the dependent variable at the group-level. This strategy may come with great costs though. By accounting for group-level variation, it reduces the model's capacity to estimate the effects of time-invariant independent variables. Besides, the estimation of country-specific effects may substantially inflate the standard errors, thus reducing the efficiency of coefficient estimates from variables that explain within-group variation. It is the first of these costs that concerns us the most, since capital mobility, one of our explanatory variables, shows low longitudinal variation. To evaluate whether a model required fixed effects, we performed Mundlak's (1978) procedure, which is robust to heteroskedasticity and serial correlation, unlike the usual Hausman test (WOOLDRIDGE, 2010, p, p. 332)⁹. The results indicate that, for most specifications, there is no significant difference between the estimates from fixed and random effects models. In such cases, we use the latter strategy, for efficiency reasons.

How to treat serial correlation in TSCS data is another controversial issue. There are two basic strategies: adjusting the standard errors or including the lagged dependent variable as a correlate. The first approach treats time dependence as a nuisance, whereas the second, as substantive information. Beck and Katz's (1996) defend the second approach, pointing that it 'causes researchers to think about the dynamics of their model'. On the other hand, Achen (2000) calls attention to the fact that this strategy makes the coefficient estimates of theoretically relevant variables vanish, since the correlation between the dependent variable and its lag is usually very high, even when there is no substantive reason to expect the first to be a function of the latter. For the specific case of social spending, we agree with Achen (2000, p, p. 02) that there are no theoretically substantive grounds for including the lagged term, and thus estimate (static) models that simply adjust the standard errors for serial correlation¹⁰. In any case, we do check the robustness of the results by also including the (first) lag of the dependent variable.

Another time dependence issue is non-stationarity, a process through which the statistical properties of a time series are functions of (as opposed to fixed in) time. Regressing a nonstationary variable on another one may produce statistically significant relationship between them even if they are not causally related, for they are both functions of time. Since the dependent variable shows a clear longitudinal pattern (Graph 01), we tested for panel data unit roots (a feature of non-stationarity) and were able to confidently reject it¹¹.

Finally, to adjust for the error structure we use Driscoll and Kraay's (1998) standard errors (DKSE) with a correction for first-order serial correlation. DKSE is an adaptation of the (heteroskedasticity and autocorrelation consistent) Newey-West estimator that is also consistent in the presence of contemporaneous correlation. DKSE's method is very much like the usual panel corrected standard errors (PCSE) used by Beck and Katz (1995), in the sense that it adjusts the covariance matrix for the same data characteristics but preserves the least squares estimator of the coefficients. However, unlike PCSE, DKSE also allows for the estimation of random (country-specific) effects, which is the main reason why we decided to use it¹².

Variables

We include independent variables that account for our main explanatory factors - capital mobility and the ideological distance between the president and the pro-elite legislative veto player -, alternative explanations, and some controls emphasized by previous analyses of social expenditures.

'President's ideology' is the estimate of the president's location on the Wiesehomeier-Benoit's left-right scale (WIESEHOMEIER and BENOIT, 2009), which ranges from 01 (extreme left) to 20 (extreme right). The estimates are based on perception data from expert surveys about parties and presidents' positions on policy dimensions, including taxes and spending. For our purposes, it is important to note that respondents ranked this dimension about average in overall importance, and it ranks last in terms of party divergence, characteristics that led the authors to conclude that "political actors do not differentiate their positions much on this economic policy issue" (WIESEHOMEIER and BENOIT, 2009. p. 1438). Therefore, we can safely assume that our measure of ideology reflects neither the president's nor political parties' propensities to spend. The data are from Andy Baker's dataset of 'Latin American Election Results with Party Ideology Scores 2.0' (BAKER and GREENE, 2011)¹³. The sample mean and standard deviation are, respectively, 8.8 and 4.9, and the range goes from 2.0 to 18.5. We thus expect the president's ideology (in this case, her degree of conservatism) to have a negative effect on social spending. However, since the sample consists mostly of leftist presidents, there may not be enough variation to capture this effect.

Using the same dataset, complemented with information from the 'Political Database of the Americas'¹⁴ and Adam Carr's election archive,¹⁵ we identified the median party in the lower chamber of the legislature. The variable 'Distance from the pro-elite veto player' equals the absolute difference between the median's ideology and the president's whenever the former is located to the right of the latter, and zero otherwise. The sample mean and standard deviation are, respectively, 2.0 and 3.7, and the range goes from 0 to 15.1. Based on the hypothesis that the existence of a pro-elite veto player attenuates the IRP, we expect social spending to decrease with the ideological distance between the president and the pro-elite legislative veto player.

We use two variables to test the effect of capital mobility: 'Financial openness' and 'Commodity economy'. The first is the Chinn-Into yearly index, which measures a country's degree of capital account openness, based on restrictions on cross-border financial transactions (CHINN and ITO, 2006)¹⁶. The sample mean, standard deviation and range are, respectively, 71, 31 and [0,100]. The second variable is the weight of the commodities sector in the economy, measured as the percentage share of the agricultural and mineral sectors in the GDP, based on data from ECLAC¹⁷. For this variable the respective summary statistics are: 16.7, 8.0, and [6.3, 48.8]. By assumption, capital mobility is increasing on financial openness and decreasing on commodity economy. For the sake of clarity of interpretation, we thus use the complement of the second variable (Non-commodity economy = 100–commodity economy), so that, from the hypothesis that the IRP is decreasing on capital mobility, we expect financial openness or non-commodity economy (or both) to negatively affect social spending. Our explanation also posits that either capital mobility or the existence of a pro-elite veto player is sufficient to reduce the left's IRP. In econometric terms, it means that the expected (negative) effect of each of those factors is decreasing on the values of the other. Therefore, we also test the interactions between the veto player distance and the capital mobility variables, for which we expect positive effects.

In what relates to the (alternative) explanations discussed in Section 02, the inclusion of non-commodity economy allows for an indirect test of Weyland's (2009) hypothesis that the left's redistributive policies were more intense in countries that benefited more from the commodities boom, due to the larger inflow of nontax revenues. However, as opposed to ours, this explanation implies a null effect for the interaction between that variable and the veto player distance. Following Murillo, Oliveros and Vaishnav (2011), to test the positive effect of capital inflow from exports we also include the variable 'Current account', which is the sum of net exports of goods and services as percentage of GDP, based on data from the World Bank. The variable 'Popular vote', which is the percentage of votes received by the president in the first round of the election, captures Biglaiser's (2016) hypothesis that IRP increased on the size of the popular support for (leftist) presidents. On the other hand, to account more specifically for the potential negative effect of electoral competition on the left's IRP (BARRILLEAUX, HOLBROOK and LANGER, 2002; GARAY, 2016), we also include 'Margin of victory', which is the difference between the vote shares received by the president and the runner-up in the first round of the election.

Finally, as controls, we include economic and socio-demographic variables that previous studies indicate as potentially relevant (AVELINO, BROWN and HUNTER, 2005; HUBER and STEPHENS, 2012; KAUFMAN and SEGURA-UBIERGO, 2001). The first and most important is 'trade openness' (measured as total imports and exports as percentage of GDP), which is common place in studies that evaluate the impact of globalization on governmental spending, even though the direction of the effect is the object of much controversy, with some authors arguing that it should decrease social spending due to market pressures, and others that social spending in fact increases to compensate for higher exposure to external shocks (GARRETT, 1998; RODRIK, 1998). 'GDP per capita' (in purchasing power parity, per $1,000) controls for the level of economic development and its expected positive impact. This variable enters the model lagged by one year to avoid endogeneity. The level of unemployment, the percentage of the population aged 65 or older, and the percentage of the population living in urban areas are usual controls for the demand side of social spending. However, we include only the first of these variables, since the other two are extremely correlated between them and with GDP per capita (all pairwise correlations are above 0.80).

Results

Table 02 presents the results from five TSCS models of social spending. Models 01 through 03 treat country heterogeneity as random effects, in accordance with the test results reported at the bottom of the table. The first and second models estimate the effects of the president's share of the popular vote and her margin of victory separately, due to the strong correlation (r= 0.74) between these variables. We keep popular vote in the subsequent models since its inclusion contributes slightly more to the adjusted R². Models 03 and 04 test separately the (partial and interactive) effects of financial openness and non-commodity economy, the variables that capture the concept of capital mobility. Again, the reason for not testing them simultaneously is to avoid collinearity issues. The fourth specification is the one that best fits the data, accounting for one third of the variance. Since this specification is the only one for which country fixed effects cannot be rejected (with 90% confidence), we report the results from both the random (4A) and the fixed effects (4B) versions.

Table 02
TSCS models of annual social spending (as % of GDP) in 13 Latin American countries (2003-2015)

	Random effects		Random effects		Random effects		Random effects		Fixed effects
	(1)		(2)		(3)		(4A)		(4B)
President's ideology	-0.113	[0.048]**	-0.096	[0.050]*	-0.087	[0.046]*	-0.073	[0.045]	-0.104	[0.049]**
Distance from pro-elite veto player	-0.170	[0.052]***	-0.141	[0.050]***	-0.117	[0.053]**	-0.141	[0.052]***	-0.171	[0.053]***
Financial openness a	-0.021	[0.012]*	-0.017	[0.011]	-0.013	[0.013]	—		—
Non-commodity economy a	-0.026	[0.070]	-0.016	[0.067]	—		-0.027	[0.052]	0.017	[0.092]
Distance x Financial openness	—		—		-0.001	[0.002]	—		—
Distance x Non-commodity economy	—		—		—		0.014	[0.008]*	0.013	[0.007]*
Current account	0.017	[0.035]	0.013	[0.035]	0.014	[0.037]	-0.004	[0.038]	0.007	[0.035]
Popular vote	—		0.020	[0.015]	0.023	[0.016]	0.028	[0.014]*	0.026	[0.013]*
Margin of victory	0.003	[0.021]	—		—		—		—
Trade openness	-0.041	[0.017]**	-0.040	[0.017]**	-0.036	[0.016]**	-0.028	[0.016]*	-0.043	[0.017]**
GDP per capita ($1000, t-1)	0.248	[0.074]**	0.258	[0.074]**	0.256	[0.067]***	0.301	[0.072]***	0.289	[0.071]***
Unemployment	-0.042	[0.084]	-0.015	[0.090]	-0.011	[0.086]	0.003	[0.092]	0.029	[0.067]
Constant	11.31		9.89		9.35		7.86
Fixed (vs. random) effects test b	0.642		0.765		0.223		0.076		0.076
Adjusted R2c	0.256		0.274		0.288		0.315		0.437
N	169		169		169		169		169

• Obs.: Least squares coefficient estimates and, in brackets, Driscoll-Kraay standard errors with AR1. Pooled estimates based on five multiple-imputed datasets.Notes: a. Centered on the sample mean; b. p-value from the F test of Mundlak’s (1978) procedure; c. Overall R² for Models 01 through 4A, and within R² for model 4B; *p < 10%, **p < 5%, ***p < 1%,in two-tailed tests.
• Source: Owned by the authors.

The coefficient estimates, standard errors and significance levels are based on the point estimates generated from the five multiple-imputed (MI) datasets, according to the rules specified in Rubin and Schenker (1986, p. 367). The (pooled) coefficient estimates are the average of the MI coefficient estimates, their standard errors are an increasing function of the MI variance estimates (within-imputation variance) and the variance of the MI coefficient estimates (between-imputation variance), and the significance tests are based on an approximation of the t statistic with degrees of freedom that depend on the relative sizes of the within and between MI variances.

The evidence provides weak to moderate support for the partisan hypothesis that social spending decreased as the president's ideological position moved to the right. Increasing the president's location on the Wiesehomeier-Benoit scale by one unit decreased annual social spending by about 0.10 percentage points (p.p.). More importantly, there is strong support for the hypothesis that social spending decreased as the ideological distance between the president and the pro-elite legislative veto player increased, and weak support for the hypothesis that this effect was attenuated by capital mobility, measured as the weight of the non-commodity sector in the economy. On the other hand, there is no support for the hypothesis that social spending was lower where capital mobility is higher.

Graph 03 illustrates the estimate of the total effect of the distance from the pro-elite veto player, evaluated at different levels of non-commodity economy. When the latter variable equals 50% (near its sample minimum), the estimated effect is -0.61, significant at the 05% level (standard error= 0.251). The magnitude of the estimate decreases as non-commodity economy increases, such that when the latter is at its sample average (83%) the former equals -0.17, significant at the 01% level (the partial coefficient estimate from model 4B). The estimated effect is not significant only for the sample values of non-commodity economy greater than 88%.

Graph 03
Estimated effect of Distance from pro-elite veto player, at values of non-commodity economy

To assess the substantive relevance of these effects we calculated the expected change in social spending from meaningful changes in the independent variables. Thus, increasing the ideology of leftist presidents (i.e., with values below ten on the Wiesehomeier-Benoit scale) from its sample minimum, 2.0 (El Salvador since 2009), to its maximum, 8.4 (Argentina), we estimate that yearly social spending increases by an average of 0.64 p.p., which corresponds to a fraction of .22 of its sample standard deviation (s.d.). Using the same criteria, by increasing the distance from the pro-elite veto player from zero to 15 (El Salvador, 2009-2011) we estimate yearly social spending to decrease on average by 9.15 p.p. (3.14 s.d.) when non-commodity economy is near its sample minimum, and by 2.55 p.p. (0.88 s.d.) when it is at its sample average. Thus, the estimated impact of the president's ideology can be regarded as small to moderate, whereas the impact of her divergence with the pro-elite veto player, moderate to high.

Among the alternative explanations, there is only weak support for Biglaiser's (2016) electoral mandate hypothesis that the IRP increased with the (leftist) president's share of the popular vote. The evidence does not lend support to the hypotheses that the IRP increased with the capital inflow from (net) exports (MURILLO, OLIVEROS and VAISHNAV, 2011), with the nontax revenue from the commodities sector (WEYLAND, 2009) or that it was higher where the left faced weaker electoral competition (i.e., a higher margin of victory). Regarding the control variables, only trade openness and GDP per capita are statistically significant. The first had a negative effect on social spending, which contradicts the compensation hypothesis (GARRETT, 1998; RODRIK, 1998), whereas the second had an (expected) positive effect. Unemployment, meanwhile, had no significant effect.

In sum, the econometric evidence lends partial support to our explanation of the intensity of redistributive policies implemented by the Latin American left during the recent commodities boom. There is moderate evidence that social spending was lower where the president's ideological preference was more to the right, and strong evidence that it decreased with the ideological distance between the president and the (pro-elite) legislative veto player. There is weak evidence that the degree of capital mobility, measured as the share of the non-commodity sector in the economy, attenuated the veto-player effect. However, there is no evidence that social spending decreased with capital mobility.

To evaluate the robustness of these results, we performed additional analyses, considering each of the following: 01. including the first lag of the dependent variable; 02. substituting GDP per capita either for the percentage of the population living in urban areas or the percentage aged at least 65, two variables that other studies include but which are highly correlated with income per capita in our sample; and 03. excluding from the sample Venezuela and Uruguay, the two countries with the highest number of missing values in the dependent variable. The inclusion of the lag of the dependent variable causes the statistical significance of the coefficient estimates to vanish, with the only notable exception being that of GDP per capita. As we argued before, there is no theoretical reason to expect current social spending to be a function of past social spending. Therefore, we interpret the sensitivity of the results to this specification as a sample characteristic. Controlling for either urban or aged populations, instead of GDP per capita, substantially decreases the fitness of the models (for example, the adjusted R² for model 4A reduces from 0.32 to 0.14 and 0.19, respectively), but only the inclusion of the second variable affects our results, making the interaction in model 4A no longer statistically significant. Given the behavior of the R², this loss of significance is more likely to reflect collinearity issues. Finally, the only consequence of excluding the observations from Venezuela and Uruguay is to strengthen the significance of the coefficient estimate for that interaction, which is a strong indication that the results are not driven by the imputations.

Summary and concluding remarks

Based on theoretical models that analyze the implications of redistribution in highly unequal democracies, this article offered an explanation for the differences in the intensity of redistributive policies implemented by Latin American leftist governments during the 2000s and 2010s. In contrast with explanations that focus on region-specific factors, we claimed that more radical policies were adopted by the left only in countries where capital mobility was sufficiently low and there was no legislative pro-elite veto player.

To test our explanation, we estimated TSCS models of social spending, using a sample of thirteen Latin American countries over the thirteen years between 2003 and 2015. The countries are all those where at least one leftist president was elected (the only ones relevant for testing differences among leftist governments), and where the time period roughly coincides with the commodities boom, when conditions for redistribution were favorable throughout the region. The results provide partial support for our hypotheses: there is strong evidence that social spending decreased with the ideological distance between the president and the (pro-elite) legislative veto player, but only weak evidence that the degree of capital mobility attenuated this effect. Moreover, the redistributive theory of democracy seems to fit the facts better than the main alternative offered by the literature, namely the rentier state theory (WEYLAND, 2009), according to which the intensity of redistributive policy is conditioned not by the existence of economic and political constraints on the taxation of the wealthy elite, but by the growth of the government's non-tax revenue, due to higher commodity prices.

Our findings provide potential contributions to both the academic and the political debates. With respect to the former, we add new empirical evidence that strongly supports the veto players theory and partially supports the redistributive theory of democracy. Although testing these theories was not the purpose of this study, the results encourage their application to the recent context of Latin American politics. Besides, the evidence that legislative veto players contributed to attenuating redistributive policy contradicts arguments that cast doubt on the relevance of formal political institutions in Latin America, particularly those designed to constrain the chief executive (O'DONNELL, 1994).

Regarding the political debate, our findings shed light on the recent conflicts and transformations that Latin American democracies have experienced. Regarding the former, we offer an interpretation that emphasizes their redistributive nature, locating the conflicts among the different elements that comprise the political dynamics of very unequal democracies, instead of blaming particularities of the region. This more general perspective may be useful to reformers who look for remedies based on the international experience. The findings also allow us to reinterpret some of the political phenomena characteristic of the Latin American left turn, such as, for example, institutional transformations aimed at broadening and deepening direct popular participation in politics (AVRITZER, 2009; POGREBINSCHI, 2013). Mechanisms such as referenda and plebiscites, to which several radical leftist governments resorted once elected, can be understood as a resource to circumvent pro-elite veto players in the legislature. Participatory institutions, by their turn, can be thought of as a mechanism for reducing the costs to the poor of collective action, increasing their capacity to preserve leftist policies under future right-of-center governments.

Finally, our explanation needs to be applied with the utmost caution to the recent reshuffling of leftist governments in Latin America and the emergence of an aggressive right-wing agenda in countries like Brazil, Argentina and Chile. In this article, we identified the reasons why radical redistributive policies were adopted in countries like Ecuador and Venezuela, instead of in Brazil or Chile. The same reasons need not apply to the reverse movement - i.e., the extent to which the new right-wing governments will be able to roll back earlier welfare-based policies. Following Pierson's seminal work on policy retrenchment (PIERSON, 1996), we believe that the factors that explain variations in the intensity of redistributive policies are not the same as those that help us figure out why the reversal of these policies may occur in a more profound way in some Latin American countries than in others.

References

• ACEMOGLU, Daron and ROBINSON, James A. (2001), A theory of political transitions. American Economic Review Vol. 91, Nº 04, pp. 938-963.
• ACHEN, Chirstopher H. (2000), Why lagged dependent variables can suppress the explanatory power of other independent variables. Paper presented at the Annual Meeting of the Political Methodology Section of the American Political Science Association, UCLA. Department of Political Science and Institute for Social Research. University of Michigan.
• ALESINA, Alberto and RODRIK, Dani (1994), Distributive politics and economic growth. Quarterly Journal of Economics Vol. 109, Nº 02, pp. 465-490.
• ARDITI, Benjamin (2008), Arguments about the left turns in Latin America: a post-liberal politics? Latin American Research Review Vol. 43, Nº 03, pp. 59-81.
• AVELINO, George; BROWN, David S., and HUNTER, Wendy (2005), The effects of capital mobility, trade openness, and democracy on social spending in Latin America, 1980–1999. American Journal of Political Science Vol. 49, Nº 03, pp. 625-641.
• AVRITZER, Leonardo (2009), La nueva izquierda, la crisis de representación y la participación social en América Latina. In: La 'nueva izquierda' en América Latina: derechos humanos, participación política, y sociedad civil. Edited by ARNSON, Cynthia J.; ARMONY, Ariel C.; SMULOVITZ, Catalina; CHILLIER, Gastón; PERUZZOTTI, Enrique, and COHEN, Giselle. Washington: The Woodrow Wilson Center Press. pp. 177-192.
• BAKER, Andy and GREENE, Kenneth F. (2011), The Latin American left's mandate: free-market policies and issue voting in new democracies. World Politics Vol. 63, Nº 01, pp. 43-77.
• BARRILLEAUX, Charles; HOLBROOK, Thomas, and LANGER, Laura (2002) Electoral competition, legislative balance, and American state welfare policy. American Journal of Political Science Vol. 46, Nº 02, pp. 415-427.
• BATES, Robert H. and LIEN, Da‐Hsiang Donald (1985), A note on taxation, development and representative government. Politics & Society Vol. 14, Nº 01, pp. 53‐70.
• BECK, Nathaniel and KATZ, Jonathan N. (1996), Nuisance vs. substance: specifying and estimating time-series-cross-section models. Political Analysis Vol. 06, pp. 01-36.
• BECK, Nathaniel and KATZ, Jonathan N. (1995), What to do (and not to do) with time-series cross-section data. American Political Science Review Vol. 89, Nº 03, pp. 634-647.
• BIGLAISER, Glen (2016), Mandate and the market: policy outcomes under the left in Latin America. Comparative Politics Vol. 48, Nº 02, pp. 185-204.
• BOIX, Carles (2003), Democracy and redistribution New York: Cambridge University Press. 284 pp..
• BORDIGNON, Massimo; TOMMASO, Nannicini, and TABELLINI, Guido (2016), Moderating political extremism: single round versus runoff elections under plurality rule. American Economic Review Vol. 106, Nº 08, pp. 2349-2370.
• BORGE, Lars-Erik and RATTSØ, Jørn (2004), Income distribution and tax structure: empirical test of the Meltzer–Richard hypothesis. European Economic Review Vol. 48, Nº 04, pp. 805-826.
• BRETSCHGER, Lucas and HETTICH, Frank (2002), Globalization, capital mobility and tax competition: theory and evidence for OECD countries. European Journal of Political Economy Vol. 18, Nº 04, pp. 695-716.
• BROOKS, Clem and MANZA, Jeff (2006), Why do welfare states persist? The Journal of Politics Vol. 68, Nº 04, pp. 816-827.
• CAMPELLO, D. (2015), The politics of market discipline in Latin America: globalization and democracy. New York: Cambridge University Press. 254 pp..
• CARLSEN, Fredrik; LANGSET, Bjørg, and RATTSØ, Jørn (2005), The relationship between firm mobility and tax level: empirical evidence of fiscal competition between local governments. Journal of Urban Economics Vol. 58, Nº 02, pp. 273-288.
• CASTAÑEDA, Jorge G. (2016), The death of the Latin American left. The New York Times The Opinion Pages. Available at ˂https://www.nytimes.com/2016/03/23/opinion/the-death-of-the-latin-american-left.html˃ Accessed on March 15, 2018.
  » https://www.nytimes.com/2016/03/23/opinion/the-death-of-the-latin-american-left.html˃
• CASTAÑEDA, Jorge G. (2006), Latin America's left turn: a tale of two lefts. Foreign Affairs Vol. 85, Nº 28, pp. 28-43.
• CASTLES, Francis G. and OBINGER, Herbert (2007), Social expenditure and the politics of redistribution. Journal of European Social Policy Vol. 17, Nº 03, pp. 206-222.
• CHINN, Menzie D. and ITO, Hiro (2006), What matters for financial development? Capital controls, institutions, and interactions. Journal of Development Economics Vol. 81, Nº 01, pp. 163-192.
• CLARK, Tom S. and LINZER, Drew A. (2015), Should I use fixed or random effects? Political Science Research and Methods Vol. 03, Nº 02, pp. 399-408.
• CONAGHAN, Catherine M. (2008), Ecuador: Correa's plebiscitary presidency. Journal of Democracy Vol. 19, Nº 02, pp. 46-60.
• CORNIA, Giovanni Andrea (ed)(2014), Inequality trends and their determinants: Latin America over the period 1990-2010. In: Falling inequality in Latin America: policy changes and lessons. Oxford: Oxford University Press. pp. 23 – 48.
• CORRALES, Javier (2010), The repeating revolution: Chávez's new politics and old economics. In: Leftist governments in Latin America: successes and shortcomings. Edited by WEYLAND, Kurt; MADRID, Raúl L., and HUNTER, Wendy. New York: Cambridge University Press. pp. 28-56.
• DRISCOLL, John C. and KRAAY, Aart C. (1998), Consistent covariance matrix estimation with spatially dependent panel data. Review of economics and statistics Vol. 80, Nº 04, pp. 549-560.
• EBERHARDT, Marcus (2011), XTCD: Stata module to investigate variable/residual cross-section dependence Statistical Software Components S457237, Boston College Department of Economics.
• ESAREY, Justin and MENGER, Andrew (2018), Practical and effective approaches to dealing with clustered data. Political Science Research and Methods pp. 01-19. doi: 10.1017/psrm.2017.42.
• FLORES-MACÍAS, Gustavo A. (2010), Statist vs. pro-market: explaining leftist governments' economic policies in Latin America. Comparative Politics Vol. 42, Nº 04, pp. 413-433.
• GARAY, Candelaria (2016), Social policy expansion in Latin America New York: Cambridge University Press. 406 pp..
• GARRETT, Geoffrey (1998), Global markets and national politics: collision course or virtuous circle? International organization Vol. 52, Nº 04, pp. 787-824.
• HALLERBERG, Mark and BASINGER, Scott (1998), Internationalization and changes in tax policy in OECD countries: the importance of domestic veto players. Comparative Political Studies Vol. 31, Nº 03, pp. 321-352.
• HAYS, Jude C. (2003), Globalization and capital taxation in consensus and majoritarian democracies. World Politics Vol. 56, Nº 01, pp. 79-113.
• HOGENBOOM, Barbara (2012), Depoliticized and repoliticized minerals in Latin America. Journal of Developing Societies Vol. 28, Nº 02, pp. 133-158.
• HONAKER, James and KING, Gary (2010), What to do about missing values in time‐series cross‐section data. American Journal of Political Science Vol. 54, Nº 02, pp. 561-581.
• HUBER, Evelyne and STEPHENS, John D. (2012), Democracy and the left: social policy and inequality in Latin America. Chicago: University of Chicago Press. 368 pp..
• HUBER, Evelyne; MUSTILLO, Thomas, and STEPHENS, John D. (2008), Politics and social spending in Latin America. The Journal of Politics Vol. 70, Nº 02, pp. 420-436.
• HUNTER, Wendy and SUGIYAMA, Natasha Borges (2009), Democracy and social policy in Brazil: advancing basic needs, preserving privileged interests. Latin American Politics and Society Vol. 51, Nº 02, pp. 29-58.
• JENSEN, Nathan and WANTCHEKON, Leonard (2004), Resource wealth and political regimes in Africa. Comparative Political Studies Vol. 37, Nº 07, pp. 816-841.
• KAUFMAN, Robert R. and SEGURA-UBIERGO, Alex (2001), Globalization, domestic politics, and social spending in Latin America: a time-series cross-section analysis, 1973–97. World Politics Vol. 53, Nº 04, pp. 553-587.
• KENNEMORE, Amy and WEEKS, Gregory (2011), Twenty‐first century socialism? The elusive search for a post‐neoliberal development model in Bolivia and Ecuador. Bulletin of Latin American Research Vol. 30, Nº 03, pp. 267-281.
• LETELIER S., Leonard. E. and DÁVILA A., Mireya (2015), The political economics of tax reform in Chile. New Political Economy Vol. 20, Nº 06, pp. 832-850.
• LEVIN, Andrew; LIN, Chien-Fu, and CHU, Chia-Shang James (2002), Unit root tests in panel data: asymptotic and finite-sample properties. Journal of econometrics Vol. 108, Nº 01, pp. 01-24.
• LEVITSKY, Steven and ROBERTS, Kenneth M. (eds)(2011), Latin America's 'left turn': a framework for analysis. In: The resurgence of the Latin American left Baltimore: The Johns Hopkins University Press. pp. 01-30.
• LIJPHART, Arend (1999), Patterns of democracy: government forms and performance in thirty-six democracies. New Haven: Yale University Press. 368 pp..
• LIPSET, Seymour Martin (1981), Political man: the social bases of politics. Baltimore: The John Hopkins University Press. 608 pp..
• LÓPEZ-CALVA, Luis F. and LUSTIG, Nora (eds)(2010), Explaining the decline in inequality in Latin America: technological change, educational upgrading and democracy. In: Declining inequality in Latin America: a decade of progress? New York and Washington: United Nations Development Program and Brookings Institution Press. pp. 03-24.
• LÓPEZ-CALVA, Luis F.; LUSTIG, Nora; PESSINO, Carola, and SCOTT, John (2014), The impact of taxes and social spending on inequality and poverty in Argentina, Bolivia, Brazil, Mexico, Peru, and Uruguay: introduction to the special issue. Public Finance Review Vol. 42, Nº 03, pp. 287-303.
• MADRID, Raúl L. (2010), The origins of the two lefts in Latin America. Political Science Quarterly Vol. 125, Nº 04, pp. 587-609.
• McCARTY, Nolan and PONTUSSON, H. Jonas (2011), The political economy of inequality and redistribution. In: The Oxford handbook of economic inequality Edited by NOLAN, Brian; SALVERDA, Wiemer, and SMEEDING, Timothy M.. London: Oxford University Press. pp. 665-692.
• MELTZER, Allan H. and RICHARD, Scott F. (1981), A rational theory of the size of the government. Journal of Political Economy Vol. 89, Nº 05, pp. 914-927.
• MILANOVIC, Branko (2000), The median-voter hypothesis, income inequality, and income redistribution: an empirical test with the required data. European Journal of Political Economy Vol. 16, Nº 03, pp. 367-410.
• MORRISON, Kevin M. (2009), Oil, nontax revenue, and the redistributional foundations of regime stability. International Organization Vol. 63, Nº 01, pp. 107-138.
• MUNDLAK, Yair (1978), On the pooling of time series and cross section data. Econometrica: Journal of the Econometric Society Vol. 46, Nº 01, pp. 69-85.
• MURILLO, Maria Victoria; OLIVEROS, Virginia, and VAISHNAV, Milan (2011), Economic constraints and presidential agency. In: The resurgence of the Latin American left Edited by LEVITSKY, Steven and ROBERTS, Kenneth M.. Baltimore: The Johns Hopkins University Press. pp. 52-70.
• NORTH, Douglass C. and WEINGAST, Barry R. (1989), Constitutions and commitment: the evolution of institutions governing public choice in seventeenth-century England. The Journal of Economic History Vol. 49, Nº 04, pp. 803-832.
• NUNES, Felipe (2014), The left variance: how electoral mechanisms shape leftist presidential behavior in Latin America. Revista Ibero-Americana de Estudos Legislativos Vol. 01, pp. 25-34.
• O'DONNELL, Guilhermo A. (1994), Delegative democracy. Journal of Democracy Vol. 05, Nº 01, pp. 55-69.
• PERALES, Francisco (2013), MUNDLAK: Stata module to estimate random-effects regressions adding group-means of independent variables to the model. Statistical Software Components S457601. Boston College Department of Economics.
• PERSSON, Torsten and TABELLINI, Guido (1994), Representative democracy and capital taxation. Journal of Public Economics Vol. 55, Nº 01, pp. 53-70.
• PIERSON, Paul (1996), The new politics of the welfare state. World Politics Vol. 48, Nº 02, pp. 143-179.
• PLÜMPER, Thomas; TROEGER, Vera E., and MANOW, Philip (2005), Panel data analysis in comparative politics: Linking method to theory. European Journal of Political Research Vol. 44, Nº 02, pp. 327-354.
• POGREBINSCHI, Thamy (2013), The pragmatic turn of democracy in Latin America. Paper presented in the American Political Science Association Annual Meeting. Chicago, IL.
• PRIBBLE, Jennifer and HUBER, Evelyne (2013), Social policy and redistribution: Chile and Uruguay. In: The resurgence of the Latin American left Edited by LEVITSKY, Steven and ROBERTS, Kenneth M.. Baltimore: The Johns Hopkins University Press. pp. 117-138.
• RITTINGER, Eric R. and CLEARY, Mattew (2013), Confronting coup risk in the Latin American left. Studies in Comparative International Development Vol. 48, Nº 04, pp. 403-431.
• RODRIK, Dani (1998), Why do more open economies have bigger governments? Journal of Political Economy Vol. 106, Nº 05, pp. 997-1032.
• RUBIN, Donald B. and SCHENKER, Nathaniel (1986), Multiple imputation for interval estimation from simple random samples with ignorable nonresponse. Journal of the American statistical Association Vol. 81, Nº 394, pp. 366-374.
• RUDRA, Nita (2002), Globalization and the decline of the welfare state in less-developed countries. International Organization Vol. 56, Nº 02, pp. 411-445.
• SANDI, R. (2006) Partido de Evo Morales se asegura control del Congreso boliviano. Nacion.com. Available at ˂www.nacion.com/ln_ee/2006/enero/18/ultima-la5.html˃. Accessed on January 18, 2006.
  » www.nacion.com/ln_ee/2006/enero/18/ultima-la5.html˃
• SMITH, Benjamin (2004), Oil wealth and regime survival in the developing world, 1960–1999. American Journal of Political Science Vol. 48, Nº 02, pp. 232-246.
• TSEBELIS, George (1995), Decision making in political systems: veto players in presidentialism, parliamentarism, multicameralism and multipartyism. British Journal of Political Science Vol. 25, Nº 03, pp. 289-325.
• WEYLAND, Kurt G. (2009), The rise of Latin America's two lefts: insights from rentier state theory. Comparative Politics Vol. 41, Nº 02, pp. 145-164.
• WIBBELS, Erik (2006), Dependency revisited: international markets, business cycles, and social spending in the developing world. International Organization Vol. 60, Nº 02, pp. 433-468.
• WIESEHOMEIER, Nina and BENOIT, Kenneth (2009), Presidents, parties, and policy competition. The Journal of Politics Vol. 71, Nº 04, pp. 1435-1447.
• WOOLDRIDGE, Jeffrey M. (2010), Econometric analysis of cross section and panel data Cambridge: The MIT Press. 776 pp..

• 1
  We use the terms 'moderate' and 'radical' to characterize the intensity of redistributive policy, and by extension, of the government that implemented it.
• 2
  The origin of this view can be traced to Lipset's (1981) classical study, in which Latin American cases are treated as a cluster, separate from the other experiences of mass representative democracy.
• 3
  The literature on the redistributive effects of capital mobility and veto players is extensive. Since a comprehensive evaluation is beyond the purposes of this article, we simply refer to the main works that establish empirical support for these.
• 4
  In Bolivia, although the 2005 election did not produce a majority in the senate, the support declared by two opposing senators for the president and his party changed this situation in January 2006 (SANDI, 2006). In Ecuador, although the left won the majority vote in September 2007 exclusively to rewrite the constitution, it suspended the legislature and assumed its powers in November of that year (CONAGHAN, 2008, p, p. 57).
• 5
  Although they are part of the government coalition led by the left, the median parties in the Brazilian and Chilean legislatures – the Brazilian Democratic Movement Party (PMDB) and the Christian Democratic Party (PDC), respectively – are centrally-oriented. In Chile, non-directly elected Senators with conservative leanings were decisive from 1999 to 2005, due to the legislative stalemate between the government and the opposition.
• 6
  The sample does include center and right-of-center presidents, since only a few countries were governed by the left during the whole 2003-2015 period. Keeping those observations was necessary to preserve the time-series-cross-section structure of the data.
• 7
  We used the statistical package Amelia II, for the R software.
• 8
  According to the results from the following tests: the Breusch-Pagan test of cross-sectional heterogeneity, likelihood-ratio test of panel heteroskedasticity, Pesaran's test of contemporaneous correlation (EBERHARDT, 2011), Wooldridge's test of panel serial correlation, and joint F test of no year-specific-effect. We performed all tests and estimations in Stata 11.
• 9
  Using the Stata module 'mundlak', from Perales (2013).
• 10
  Brooks and Manza (2006) show that persistence in social spending is due to mass policy preferences, which have a high degree of temporal stability. According to the authors, "this characteristic does not disqualify [other temporally-stable variables] as causal factors behind social policymaking" (Brooks and Manza, 2006, p, p. 822).
• 11
  According to Levin, Lin and Chu’s (2002) test. By rejecting non-stationarity, we do not need to consider (dynamic) models based on first differences or error correction.
• 12
  It should be noted that when the number of clusters is small (as in our sample), standard errors may be biased downwards. Esarey and Menger (2018) provide a method for solving this potential bias, but it does not work with random effects or time series, which characterize our data. However, it is somewhat comforting that their simulations show that random effects perform better than any cluster adjustment method when the model is correctly specified (Esarey and Menger, 2018, p, p. 02).
• 13
  Available at http://spot.colorado.edu/~bakerab/data.html.
• 14
  Available at: http://pdba.georgetown.edu/.
• 15
  Available at: http://psephos.adam-carr.net/.
• 16
  Available at http://web.pdx.edu/~ito/Chinn-Ito_website.htm.
• 17
  Available at http://estadisticas.cepal.org/cepalstat/WEB_CEPALSTAT/Portada.asp.

Publication Dates

• Publication in this collection
  13 May 2019
• Date of issue
  2019

History

• Received
  06 Oct 2017
• Accepted
  19 Jan 2019