The New Federalism of Mexico's Party System

1 Introduction

Decentralized political institutions, whose authority is divided between central and regional governments, have been widely lauded for their economic and political benefits. On the political front, regional autonomy helps manage deep-rooted social divisions and gives minority groups political space in subnational politics. On the economic front, decentralization is purported to create competition between subnational units for investment and human capital, leading to improved property rights, economic growth and human welfare (Chandler 1987; Davoodi and Zou 1998; Dyck 1997; Manor 1998; Riker 1964; Rodden 2002; Ross 2000; Stansel 2002; Stein 1999; Stepan 1999; Suberu 2001; Weingast 1995).

Critics of federalism point to its institutionalization of regional conflict, and its detrimental impact on national agenda formation. Federalism creates an additional layer of political competition and formalizes existing regional competitions and disagreements. Subnational jurisdictions compete for federal transfers, pork, and even natural resources. Certainly, local pressure for resources does not depend on federalism, but a federal form of government cements the lines of competition and naturally organizes dispersed and otherwise fleeting and unorganized shared interests (Chandler 1987; Scharpf 1995; Ross 2000; Mainwaring 1997; Suberu 2001). The result is that national policy agendas may be stalled or sidetracked by regional conflicts. ¹

One manifestation of powerful, competing subnational agents is the division of national parties along state or provincial lines. National legislators may find their allegiance and accountability divided between agents of national and subnational interests. When these interests are not aligned, legislators find themselves torn between “competing principals” (Carey 2007). In other words, sometimes legislators vote with a national party, but other times they are more responsive to subnational interests, to the detriment of national policy agendas, debates, and the consolidation of the party system.

However, while political scientists have recognized the importance of centralization and decentralization for many facets of politics, many of the differences across and within systems are unexplained. Most systems with decentralized parties are federal, but many formally federal systems are highly centralized. Indeed, there is great diversity in the influence of subnational political actors over national politics across various federal systems around the world. At one extreme are cases like Brazil and Argentina, where substantial policy and political authority resides in subnational units. At the other extreme are cases that we normally think of as de facto unitary systems, but which in fact have many of the same formal institutions as federal systems. ² Previous work suggests that part of the explanation for these differences is in the role of informal institutions in structuring federalism, but empirical evidence is rare and usually requires comparisons among very different cases.

In this paper, we seek to contribute to the literature on federalism and party system nationalization. We focus on the nationalization of legislative policymaking by examining party unity on roll-call votes. While a large literature has examined party system nationalization through electoral results, much less attention has been focused on the degree to which parties are divided by subnational conflict when confronting a policy agenda (see, for example, Schattschneider 1960; Jones and Mainwaring 2003; Kasuya and Moenius 2008; and Chibber and Kollman 2004). Though understudied, this aspect of party systems is extremely important as it directly affects the extent to which parties can form and maintain a clear and consistent ideological brand for voters, as well as the extent to which parties can unite and advance their agenda.

Our analysis focuses on the case of Mexico, an ideal case for exploring the role of subnational conflict in national legislative politics due to its unique institutions and recent transition from one-party to multi-party rule. Under the one-party rule of the Revolutionary Institutionalized Party (PRI, Partido Revolucionario Institucional), career opportunities were highly centralized, and national parties became extremely cohesive. When the PRI lost power, party cohesion dropped dramatically, and as we show, much of this decrease was due to state divisions within parties. Our analysis also reveals that this transformation reflects the interaction of resource (de)centralization, electoral institutions, and the degree to which parties are ideological or distributive.

We proceed in three additional steps. We first provide an overview of the mechanisms of federalism that affect nationalization of parties. Then we apply these frameworks to the Mexican federal system, examining institutions that shaped levels of decentralization during and after the decline of one-party system behavior, and test these hypotheses using an original dataset. Finally, we consider the implications and limitations of our work, along with future research that it may spark.

Although many aspects of federalism are much studied, its impact on party systems has received only limited attention. Most of this work comes from Europe, and most scholars agree that a combination of formal and informal institutions determines whether federalism leads to regionalization of parties. One central focus is on societal diversity, and another is on whether the formal institutions of federalism – especially the geographic construction of subnational units – correspond closely with distinct local shared interests and preferences. Mayer (1970) calls this correspondence a case of “congruent” federalism, cases without that correspondence being merely “legalistic” federalism. Both the informal and formal institutions are required, the argument goes, for parties to split on subnational lines. A second line of research focuses on the formal institutions of federalism: the degree to which governmental authority and action are decentralized. For example, Thorlakson (2003) argues that the key factor in the impact of federalism on parties is the decentralization of power to subnational units. For Thorlakson, this power can include authority over both policymaking and resource allocations, and decentralization leads to less vertical integration and more decentralized parties. Other scholars combine formal and informal features of political systems to explain variance in federalism's impact. Chandler finds that combining formal decentralization and regional social differences accentuates the “fragmenting or centrifugal effects” of federalism (Chandler 1987: 156).

More recent work argues that the key variable is the decentralization of resources, not of policy, and that this allocation has gradual, long-term effects on the nature of the party system. Chhibber and Kollman (1998, 2004) argue that where resources are centralized,

In essence, there is pressure on identities to nationalize when pork is centralized. Other scholars have responded by noting that this resource argument is most appropriate in pork-oriented systems, and is limited in its effects to the ruling party that controls national or subnational resources. Most tests of these hypotheses rely on cross-country case studies or comparative quantitative analysis.

One way to synthesize this literature is to observe the three ingredients essential for fragmenting federalism into national parties. First, legislators are career-oriented. This may mean they seek re-election to their current post, or that they seek other forms of political advancement. Second, one or more subnational actors, to some degree, have influence over their desired career progressions. State party organizations might control nominations, list positions, or media time. State governors might control pork and have extensive campaign resources. We will call these actors “subnational gatekeepers.” Their presence does not preclude the existence of national gatekeepers – national parties, national executives, or others. But their very presence could potentially influence national legislators. Third, these subnational gatekeepers have some policy disagreements – either with each other, or with the national gatekeepers. For example, subnational gatekeepers – even ones of the same party – could find themselves in disagreement over resource distribution. Such conflicts might include water rights, petroleum revenue sharing, the location and size of federal military establishments, and so on. Resource distribution is naturally a zero-sum game that can lead to conflict between subnational actors. At the same time, disagreement might be ideological – subnational gatekeepers from one state might simply be more liberal than those from another state – even when they are all in the same party. ³

When these three assumptions are met – when legislators are career-seeking, when there are subnational gatekeepers that control key resources that these legislators covet, and when subnational gatekeepers have some diverging interests on the national stage – subnational conflict can spill over into national politics. When that happens, national legislative parties will divide into competing state delegations. These divisions will lower party cohesion, but in a specific and systematic way. While national party cohesion will be low, state delegations within parties will be unified on either side of key issues. The cost will be weaker national parties, less consolidation of party labels and positions, and more difficulty advancing a policy agenda.

When looking at Mexico – or any other case of federalism – this discussion implies that understanding federalism's impact on national parties requires addressing three issues: (1) whether legislators are career-oriented; (2) the existence of subnational gatekeepers influencing legislators’ careers; and (3) the degree of disagreement among the subnational actors. We now examine recent Mexican political history, applying our framework and generating testable hypotheses.

In this section, we provide a brief overview of the Mexican system and its transformation, and offer a series of hypotheses on how recent changes have affected legislative politics. For many years, Mexico would not have been considered a suitable case for studying federalism's impact on parties. This might be surprising to a casual observer, because many of the prerequisites of party-splitting federalism seem to be present there. Mexico has subnational political units – states – with independent executives and legislatures. In addition, Mexican states are extremely diverse socially, economically and culturally. Although these factors should have created incentives for national parties to split along state lines, the informal institutions created by one-party rule led to a highly centralized system with strong national parties. More recently, the end of one-party rule has transformed the informal institutions that centralized politics, and led to a new federalism of parties in Mexico. We discuss this transformation in the following paragraphs.

Under one-party dominance, Mexico had a highly cohesive and centralized political system. The combination of unified government, party discipline, and the president's position as party leader guaranteed centralized control of the distribution of political and bureaucratic opportunities, which guaranteed loyalty to the party and president (Weldon 1997, 2002). The existence of Mexico's meta-constitutional powers meant that the future of representatives depended on their relationships with and loyalty to the president. Thus, deputies were accountable not to voters but to the president and the national party (Ugalde 2000: 131). ⁴ There were potentially powerful subnational actors – notably state governors – but their futures also were tied to central authority. Presidents typically chose their successors, so governors who wanted to advance to the presidency or to cabinet positions had to comply with presidential preferences. In this way, whatever limited subnational gatekeeping power existed was leveraged to reinforce centralization, rather than to counter it. ⁵

Thus, despite Mexico being formally a federal republic, one-party rule has created a highly centralized system, with power and resources flowing down from the national executive. ⁶ Legislators’ relationships with electorates were weakened, state parties were made reliant on their national committees, and governors were enfeebled, becoming mere administrators of central authority. State governors gradually lost their budgetary authority to the central government in exchange for career security, with post-gubernatorial posts in the federal legislature or the bureaucracy (Díaz-Cayeros 2003). In many ways, the Mexican system under one-party dominance strongly resembled that of Japan – a system with the formal institutions of federalism, but informal centralizing institutions of opportunity and pork created by one-party rule. In that case, the Liberal Democratic Party (LDP) distributed pork and career advancement downward from the national party through networks known as keiratsu (Scheiner 2006).

The Mexican system began to change in the 1990s in response to declining popular support and increasing competition from opposition parties (Greene 2007; Garrido 2012). Electoral reform led to more opposition victories and created career opportunities for disgruntled PRI politicians – if they did not want to follow orders from the national party, they could switch to an opposition party and have, for the first time, a reasonable chance of electoral success. Budget reform transferred more resources to state governors. The result was growth in opposition parties, culminating in the PRI's loss of majority control of Congress in 1997, and loss of the presidency in 2000.

At the national level, the result was that the informal meta-constitutional powers of the president were destroyed, transforming the Mexican executive from one of the strongest in the world to one of the weakest among presidential systems (Samuels 2003a). The Mexican president's authority within his own party was reduced by the presence of other career opportunities. Subnational actors, freed from the need to meekly obey the president and empowered with more budgetary resources, began to assert independent authority and influence. And in the national legislature, no party has had a majority since 1997, forcing the president to negotiate extensively with legislators to build majority coalitions. ⁷

These changes, we argue, have created a new federalism for Mexican parties. The end of unified one-party rule with the defeat of the PRI, the decentralization of resources, and the end of the PRI's monopoly on political opportunities have empowered subnational political interests and weakened national actors. Legislators seeking to advance their careers can no longer just toe the national party line: they have to work out a complex calculus when deciding which bills to support and which to oppose while con-tending with potential pressure from the president, national parties, state governors and state parties. Furthermore, we expect that the extent of these divisions will vary with the relative influence of national and subnational gatekeepers. In the following paragraphs, we consider several empirical implications of these changes for Mexican legislative parties. We begin with the broadest hypothesis:

Our other hypotheses suggest differences across and within parties. We expect that the dramatic changes in Mexico will have uneven impacts on different state delegations, as other institutions determine the relative influence of national and subnational interests in legislative politics. We focus on three specific factors: electoral rules, ideology and local resources.

Regarding electoral rules, differences in personalism and in ballot access suggest different incentives for responding to pressure from subnational gatekeepers. Mexico has a mixed electoral system, with 300 deputies elected from single-member districts (SMD), and 200 from closed-list proportional representation lists (CLPR). ⁸ Most SMD legislators are nominated locally or selected in party primaries, and they tend to be political entrepreneurs that go on to seek other geographically based offices. The proportional representation deputies are elected from five mega-districts, each covering multiple states, and in most cases, these deputies are nominated by their national party organizations. They also tend to be tracking toward party careers – many have jobs in the national party organizations after their legislative tenure.

The result is that we should see differential impacts of the transformation of the political system for each of these types of deputies. SMD deputies have careers tied to local opportunities and will consequently be responsive to local interests (Carey 2003). Proportional Representation (PR) deputies have little reason to pay attention to local politics; their career advancement is much more likely to depend on good relations with the national party hierarchies. This suggests that any impact of Mexico's new federalism should be greatest among the SMD deputies and weakest among PR deputies, which is our second hypothesis:

Besides characteristics of legislators, the relative strength and resources of subnational actors will also affect their ability to successfully lobby national legislators and thus create state-based divisions within parties. In particular, we focus on whether legislators have a co-partisan governor in their home state. A same-party governor is a potentially powerful and important ally for a sitting legislator. ⁹ They control budget resources and can support legislators’ campaigns and projects. As the most important members of their state party, they also have influence over nominations for future candidacies. Legislators whose home-state governor is from another party are in an entirely different situation – they are extremely unlikely to receive any support from their governor, and their state party will be weaker and less influential without the presence of a dominant office-holder. This leads us to our next hypothesis:

Finally, an additional factor that may shape the impact of federalism is the nature of legislators’ electoral markets – in particular, the extent to which electoral success is ideological or distributive. There are two components to this: First, more ideological parties and elections create incentives for national platforms and unity in defense of a party brand name (Cox and McCubbins 1993). Second, distributive politics can be highly centralizing – or decentralizing – depending on the nature of budget authority. If pork is centralized, then legislative behavior should also be highly centralized and responsive to central budget authorities. If pork is decentralized, legislators should be highly responsive to the local actors that control resources (Desposato and Scheiner 2009). More ideological parties will be less susceptible to subnational divisions; more distributive parties may be more susceptible, depending on the extent of local resources and the alternatives offered at the national level.

For Mexico, this implies differentiating among the three main parties, the PRI, the National Action Party (PAN, Partido Acción Nacional), and the Party of the Democratic Revolution (PRD, Partido de la Revolución Democrática). Although none of these parties represents idealized extremes of distributive or programmatic parties, the literature clearly ranks them by strength of ideology. All evidence suggests that the PAN and PRD are both more ideological than the PRI, and that the PRI is ideologically diverse and relies more on clientelism and distributive politics. For example, scholars have found substantial variance in PRI positions across Mexico, with much less variance in PAN and PRD platforms (Klesner 2005: 132–134). In addition, survey data show evidence that ideology is more variable within the PRI among both mass supporters and elites than it is between the PRD and the PAN (Moreno 2009: 365–367). Although there are factions within both the PAN and the PRD (Shirk 2001; Bruhn 1997), intra-party conflicts for these parties are about personalities and electoral strategy rather than ideological positions.

The PAN is considered the most ideological of the three main parties. Among PAN members, there is consistent support for market forces and conservative social policies (Edmonds-Poli and Shirk 2009: 259). Similarly, in survey data, members’ ideological self-identification has been concentrated on the right of the spectrum (Moreno 2009: 366). In addition, since 2000, the PAN has held the presidency, giving it more national authority to enforce party cohesion. ¹⁰ On the other extreme of the ideological spectrum, the PRD is more diverse than the PAN, but also projects a consistent ideological signal. The PRD was formed in 1987 through the merger of a number of small leftist parties plus a splinter faction from the PRI that was discontented with the candidate-selection process of the presidential elections and the federal government's neoliberal economic policies during the eighties (Craig and Cornelius 1995: 259). Despite the heterogeneity of its members’ origins and weak institutionalization (Wuhs 2006: 47–48; Bruhn 2008: 217), disagreement within the party is typically about internal leadership and electoral strategy rather than policy platform (Craig and Cornelius 1995: 279). Similarly, PRD partisans’ ideological self-identification is consistently located on the left side of the spectrum (Moreno and Méndez 2007: 64–65).

In contrast, the PRI has a reputation for being an ideologically diverse party with a focus on redistribution and a great reliance on clientelism compared with other parties. Formed in 1929 as a coalition of different and heterogeneous groups, the party lacks a clear and defined ideology (Rodríguez and Ward 1994: 171). Though roughly identified as centrist, the “PRI jumps between the two poles, depending on the political circumstances of the moment” (Langston 2007: 359). Voters perceive the weakness of its ideological position, and the evidence shows that ideological self-placement has no effect on a vote choice for the PRI in federal elections (Guardado 2009). This lack of ideological consistency became even more salient when the party lost the presidency and conceded the power of the party to PRI governors, who represent the interests of core constituencies with very different preferences (Estévez, Díaz-Cayeros, and Magaloni 2008: 54). Subsequently, the PRI has been able to maintain its support through patronage among the elite and by redistributive policies for the voters (Lawson 2000: 270).

The different degrees of ideological heterogeneity and reliance on distributive politics suggest different patterns of responsiveness to subnational actors. In particular, the PAN and the PRD should be less responsive to subnational actors, and the PRI's behavior will vary before and after resource decentralization. Under PRI hegemony, when all career resources were concentrated in the hands of the president, the PRI should be highly centralized. After the PAN's takeover of the presidency and the decentralization of resources to governors, PRI legislators should naturally reorient themselves around the directives of their PRI governors. This suggests an additional interaction – namely, that these decentralizing effects for the PRI should be present only in states with PRI governors.

We are the first to examine the broad impact of federalism on Mexico's parties and to test for these specific effects. However, other important work has looked at related questions in Mexico. For example, Langston (2011), after conducting interviews with governors, noted that they are most likely to mobilize on fiscal issues (consistent with our argument that federal conflict will be primarily about resource distribution, above). More recently, Rosas and Langston (2011) study the relationship between gubernatorial influence over national legislators and the timing of elections. They find that that the closer the election, the lower party cohesion. Their analysis is limited in that they examine only legislators with co-partisan governors (just over half of all legislators). In addition, recent work suggests that their measurement is severely biased toward Type I error.

We test our hypotheses by seeking evidence of state divisions within Mexican parties on roll-call votes in the Chamber of Deputies. We utilize an original dataset that we collected of all published roll-call votes covering from the 57^th to the 60^th Legislatures of the Chamber of Deputies (1997–2006). ¹¹ We find significant evidence that Mexican parties have subnational divisions, and that the magnitude of these divisions varies with resource centralization, electoral systems, co-partisan governors, and ideological cohesion of the party.

Our tests take two forms. First, we adapt a spatial model to federal politics, where legislators face pressure from national and subnational actors. This approach uses a familiar model and methods, but trades breadth for narrow statistical power. More specifically, the spatial tests detect the presence of influential subnational actors when their influence is narrowly identified as a constant unidimensional pressure. However, spatial tests to detect the influence of subnational actors fail under alternative models of subnational actors’ influence. Second, we use a nonparametric test to examine the data with weaker assumptions. This approach has less statistical power, but can detect subnational defections whether spatial or non-spatial. For example, suppose that governors are only occasionally mobilized to pressure deputies, perhaps only on resource distribution bills. A spatial model will often fail to detect that pressure but a nonparametric permutation analysis will pick it up.

We first discuss several details of the data. Our core unit of analysis is the legislator-party-gubernatorial administration. During the period studied, many legislators switched parties, becoming independents or joining other parties. Switchers are treated as different observations in the data analysis -legislators’ votes are always counted with the party of membership at the time a roll-call vote was cast. If a deputy was a member of the PAN one day and the PRI the next, we count votes before the switch with the PAN, and votes after the switch with the PRI. Similarly, in some of our analysis we distinguish between legislators in their home governor's coalition and legislators opposing the governor. However, gubernatorial terms do not coincide perfectly with deputy terms; most governors are not elected concurrently with deputies. Again, our unit of analysis is the deputy-party-governor. If a PRI legislator enjoys a PRI governor until 1 July, and a PAN governor after that date, then votes are coded into government and opposition coalitions accordingly.

4.1 Spatial Analysis

Our first test is based on a simple, low-dimensional spatial model. In this test, we are assuming that legislators are arrayed in a low-dimensional space, each with a clear ideological position. Similarly, their national party leaders and subnational gatekeepers also have distinct positions in this ideological space, and compete to influence legislative behavior. One way to represent this framework is with the following simple model:

ψ i = α σ j + β γ k + ( 1 − α − β ) θ i

(1)

where ψ_i is the observed ideal point of legislator i, θ_i is the true underlying bliss point for legislator i, σ_j is the ideal point of party j, and γ_k is the ideal point of some state actor k – for example, a state party delegation k. ¹² The key parameters here are α and β, and they measure the influence of national and subnational gatekeepers over legislators. With the restriction that 0 ≤ α, β ≤ 1 and α + β ≤ 1, these parameters are the relative influence of the national and subnational actors, respectively. Each legislator's observed ideal point ψ_i is a weighted average of the preferences of the legislator, the national actor, and the state actor. For example, when α = β = 0, legislators ignore national parties and state parties when making decisions, and their observed ideal points reflect only their own preferences. When α=β=1/3, legislators are equally influenced by their national party, state party, and their own conscience. Again, this model is appropriate if one believes that subnational interests are part of a low-dimensional space and are attested to through constant influence over legislators.

12

Equation 1 could be generalized to more than one state actor, of course, but we assume just one to illustrate the model and challenges.

Unfortunately, straightforward estimation of (1) is impossible, because we do not have estimates for any of the unobserved underlying ideal points of parties (σ_j), state parties (γ_k), or legislators (θ_i). ¹³ However, with modest assumptions, we can still test for patterns of dispersion that reflect national and state actors’ influence using a simple analysis of variance framework. ¹⁴

13

Even if we did have reliable measures of some of these underlying preferences, identification might still be a problem without party-switching or other changes.

14

Assuming iid distributions of legislators (θ) within each party, and iid distributions of subnational actors (γ) also within each party, each state party's delegation will have different mean ideal points.

Our analysis uses a two-step process. Following Lewis’ (2000) recommendations, we estimate ideal points for several dimensions using W-NOMINATE (Poole and Rosenthal 1997), then run an analysis of variance on ideal points, testing for a reduction in unexplained variance with controls for states, state parties, electoral rules, and gubernatorial indicators. If subnational actors are influencing national legislators – that is, if β is positive -then on average, the ideal points for legislators influenced by subnational actors will all move toward the ideal point of the influential actors from their home state, and an ANOVA will detect differences within parties.

We tested for intra-party subnational divisions using simple ANOVA models with estimated ideal points as the dependent variable, and dummy variables for party, state, governor, and state party as the independent variables. We analyzed the first two dimensions, which explain most of the variance in the roll-call vote data. ¹⁵ The results are presented in Tables 1 and 2.

OPEN IN VIEWER

Table 1

ANOVA Analysis for the First W-NOMINATE Dimension

	(1)	(2)	(3)	(4)
57th Legislature
Model	258.518 ***	259.328 ***	260.209 ***	262.359 ***
Party	258.518 ***	197.852 ***	30.270 ***	2.972 ***
State		0.810 ***	1.047 ***	0.991 ***
State-Governor			0.881 ***	0.073 ***
State-Party				2.150 ***
Residual	7.629	6.818	6.016	4.607
R2	0.971	0.974	0.977	0.986
58th Legislature
Model	132.650 ***	134.106 ***	135.181 **	135.779 ***
Party	132.650 ***	107.424 ***	63.224 ***	1.720 ***
State		1.456 ***	0.697 **	0.425
State-Governor			1.076 ***	0.787 **
State-Party				0.598
Residual	8.601	7.144	6.070	5.471
R2	0.939	0.949	0.957	0.9631
59th Legislature
Model	182.804 ***	183.793 ***	184.340 ***	90.910 ***
Party	182.804 ***	141.353 ***	67.798 ***	1.746 ***
State		0.561 ***	0.601 ***	0.315 **
State-Governor			0.428 ***	0.251
State-Party				0.548 ***
Residual	4.017	3.455	3.028	2.481
R2	0.979	0.982	0.983	0.987
60th Legislature
Model	128.011 ***	129.224 ***	129.654 ***	131.631 ***
Party	128.011 ***	75.251 ***	43.743 ***	4.095 ***
State		1.213 ***	0.9366 *	0.571
State-Governor			0.430	0.356
State-Party				1.977 ***
Residual	10.262	9.049	8.619	6.642
R2	0.926	0.935	0.938	0.952

Note:

***

F <.001

**

F <.010

*

F <.050.

Source: Authors’ own calculation.

OPEN IN VIEWER

Table 2

ANOVA Analysis for the Second W-NOMINATE Dimension

	(1)	(2)	(3)	(4)
57th Legislature
Model	118.187 ***	119.567 ***	121.244 ***	123.764 ***
Party	118.187 ***	89.139 ***	31.072 ***	2.304 **
State		1.380 ***	1.733 **	2.191 ***
State-Governor			1.677 **	2.271 ***
State-Party				2.520 ***
Residual	16.132	14.752	13.076	10.555
R2	0.880	0.890	0.902	0.921
58th Legislature
Model	75.129 ***	77.658 ***	79.393 ***	80.685 ***
Party	75.129 ***	62.779 ***	19.768 ***	1.601 ***
State		2.529 **	1.650	1.078
State-Governor			1.734 *	1.306
State-Party				1.293
Residual	21.821	19.291	17.557	16.264
R2	0.775	0.801	0.819	0.832
59th Legislature
Model	90.910 ***	92.609 ***	94.034 ***	96.600 ***
Party	90.910 ***	71.315 ***	18.761 ***	1.471 **
State		1.699 ***	1.255 **	0.759
State-Governor			1.425 **	1.017 *
State-Party				2.565 ***
Residual	14.148	12.449	11.023	8.458
R2	0.865	0.881	0.895	0.919
60th Legislature
Model	43.227 ***	46.581 ***	48.357 ***	50.538 ***
Party	43.227 ***	39.365 ***	31.400 ***	3.890 ***
State		3.354 ***	2.647 **	1.313
State-Governor			1.776	1.352
State-Party				2.180
Residual	26.510	23.156	21.379	19.199
R2	0.620	0.668	0.693	0.725

Note:

***

F <.001

**

F <.010

*

F <.050.

Source: Authors’ own calculation.

15

The first two dimensions follow what Domínguez and McCann (1996) find to be the dominant dimensions in the Mexican political system: the first dimension is a classic left–right economic space, and the second dimension is a combination of a government/opposition split and continuing pressure to liberalize the political system (Robles Peiro 2009).

The analysis provides consistent evidence of subnational divisions in Mexican parties. Note first that the party factor explains most of the variation in ideal points for both dimensions, for all periods. For all the periods, an R² from a model only including party dummy variables was always above 0.926 for first-dimension ideal points, and 0.62 for second-dimension scores. The obvious conclusion is that most legislative behavior can be explained by party membership.

However, in every period studied, at least one state factor (and sometimes all of them – state, state governor, state party) was significant. For first-dimension results, all three state factors were significant for the 57^th Legislature (1997–2000), though the magnitude of their improvement in fit is less impressive. A model with just the party factor yields an ² of 0.971; after adding the state factors, the R² is 0.974. The marginal change in fit remains small – but significant – for first-dimension results for other periods.

Results looking at second-dimension ideal points are similar, with a few important differences. Most notably, the marginal increase in fit is much larger than on the first dimension, suggesting that more state conflicts are manifest in second-dimension bills than in first-dimension bills. For example, adding the state factors increases the R² from 0.880 to 0.921 in the 57^th period (1997–2000). As with the first dimension, the specific patterns of which factors matter most tell no consistent story. In the 58^th and 60^th Legislatures (2000–2003 and 2003–2006), the state factors have no significant impact. In the 57^th Legislature (1997–2000), all the variables are significant in the full model, while for the 59^th (2003–2006), the state variable lacks significance once the variables of state party and governor are included in the model.

The ANOVA analysis suggests four core conclusions. Parties explain the majority of variance on both the first and second dimensions. There are consistent significant state effects that divide parties. And the apparent magnitude of effects is higher on the second dimension than on the first. There is evidence that all subnational actors (state parties, governors, and electorates) are driving state divisions, but patterns of influence are inconsistent across both time and the dimensionality of the policy space.

4.2 Permutation Results

One possible conclusion from the spatial ANOVA analysis is that there seems to be some evidence that subnational actors affect national legislative parties in Mexico, but the effects are inconsistent and very small, especially when compared with the effects of large parties. However, one limitation of the spatial model is that it measures only a very particular type of federal effect. The ideal point model works best when state actors exert a constant influence over legislators in a low-dimensional ideal space. But if state conflicts are part of high-dimensional resource space, or mobilize actors only from time to time, a low-dimension spatial model may underestimate their impact or return inconsistent results. Given recent evidence that subnational actors are mobilized only infrequently and on distributional issues (Langston 2010), the results presented above might simply be confirming that federal-ism's impact in Mexico is not a simple, low-dimensional phenomenon.

For our second test, we use a nonparametric cohesion score analysis that corrects for the problems of the preceding ANOVA analysis. This test has less statistical power than the spatial model, but it can detect subnational divisions generated by many different behavioral models, including both spatial and non-spatial divisions.

The key test of the model is to compare state party cohesion on roll-call votes with overall party cohesion. For a party i, define the cohesion of state delegation j on a single vote as:

c i j = | Y i j − N i j | Y i j − N i j

Overall state party cohesion is thus the average of all state delegations, weighted by size:

∑ j = 1 n = 1 w j | Y i j − N i j | Y i j − N i j

If dissension within a party has nothing to do with state politics, we expect the levels of defection to be even across all states, and we likewise expect state party cohesion to be similar to overall national party cohesion. But if dissension reflects the politics of federalism, then defections will be concentrated in a subset of states (the subsets may vary over time, of course). A key feature of this method is the use of permutations to correct for bias in cohesion scores and to conduct inference. This method has been applied to study federalism in other countries and is well documented in the literature (Desposato 2003, 2004).

Figures 1 to 3 show the results of a nonparametric permutation analysis on our dataset. ¹⁶ Beginning with Figure 1, the dotted line shows national party cohesion, the dashed line shows mean state party cohesion, and the histogram shows permuted state delegation cohesion. Interpreting this data is easy: When the dashed line is covered by the histogram, we do not reject the null. When the dashed line is to the right of the histogram, these results are unlikely to happen randomly, and we reject the null hypothesis of no state effects. ¹⁷ More specifically, statistical significance is determined by quantiles of the permuted values; if actual state party cohesion exceeds 82 percent of the permutations, then one-sided significance is 0.18; if it exceeds 95 percent of the values in the histogram, then one-sided significance is 0.05. And if observed values exceed all of the permuted values, then the p-value is less than 0.0001. ¹⁸

OPEN IN VIEWER

Figure 1

Federal Influences on Roll Call Votes, 57^th-60^th Legislatures

OPEN IN VIEWER

Figure 2

Federal Influences on Roll Call Votes by Party, 57^th-60^th Legislatures

OPEN IN VIEWER

Figure 3

Federal Influences on Roll Call Votes by Party, Electoral Rules, and Gubernatorial Status, 57^th-60^th Legislatures

16

The detailed estimations are shown in Table A in the Appendix.

17

Note that the units vary across histograms, as the cross-period variation is large and would mask the quantities of interest.

18

There were 10,000 permuted values, so exceeding all of them implies a one-sided significance level of less than 1/10,000 = 0.0001.

We first examine overall party cohesion, which is represented by the dotted line and also shown in the first column of Table A in the Appendix. Overall party cohesion was very high in the 57^th Legislature (1997–2000), 0.963. Once the PRI lost the presidency and conflict was decentralized, cohesion fell substantially: during the 58^th Legislature (2000–2003) it was 0.835. Though it has risen over the course of the last two legislatures, it has not reached the high levels of party cohesion observed under the PRI. The change from the disciplined period of PRI hegemony is significant, and it suggests an important transformation of the party system. However, this weakening could reflect any number of things, not just an increase in subnational conflict and influence. Governors might have had an influence on legislative issues as the party lacked a clear ideological position, but its electoral debacle in 2006 pushed party officials to become more consistent to prevent another electoral defeat.

Do these changes reflect a simple weakening of party discipline and a rise in personal vote-seeking, or could they also reflect increasing divisions between state delegations? The permutation analysis answers this question, and provides support for a dramatic increase in state divisions in national parties. For all the legislative periods, state party cohesion exceeds the 10,000 null hypothesis permutations – providing evidence of state-based differences, significant at the .0001 level. Furthermore, there is evidence of a dramatic increase in the magnitude of the effects after the PRI's loss of hegemony. In the first period, the 57^th Legislature, party cohesion was very high, with very small – but still significant – state divisions within parties. Note how the dashed line is just barely outside the permuted range, and that all the values are very high (mean party cohesion = 0.963; state party cohesion = 0.975).

After the PRI's loss of the presidency, institutional changes led to a massive decentralization of power. These changes are evident in the cohesion scores. Beginning in the 58^th Legislature (1997–2000), party cohesion falls precipitously and federal effects grow dramatically. For the 58^th Legislature (2000–2003), mean party cohesion is only about 0.83, while state party cohesion is 0.89. Since the null distribution – the range of permuted values – only covers 0.86–0.87, we reject the null hypothesis of no state party divisions in the national legislature, significant at the .0001 level. For the 59^th Legislature (2003–2006), overall cohesion rises modestly, but state party cohesion remains well above the range of permuted values, again indicating intra-party state divisions in roll-call votes. By the 60^th Legislature (2006–2009), overall cohesion has again risen, but significant subnational effects persist.

We can put these cohesion scores into perspective by standardizing them. To do so, we compare the relative size of the state effects with the divergence from perfect party cohesion, coming up with the percentage of defection attributable to state divisions:

f = c s − c p 50 1 − c s

where c_s is state party cohesion and cp50 is the median permuted state party cohesion.

For example, in the 57^th (1997–2000), state party cohesion was 0.9749, and median permuted cohesion was 0.9731. The maximum cohesion score is 1.0 for perfect group unanimity on all votes. So the overall divergence from perfect cohesion (1-0.9731) is 0.0269. And the magnitude of state party effects (divergence from the med) can be calculated as 0.9749-0.9731 = 0.0018. So 0.0018/0.0269 = 0.067, or 6.7 percent of the intra-party disagreements might be attributed to state divisions. By these calculations, state divisions grow substantially from the 57^th to 59^th Legislatures (1997–2006), then weaken in the 60^th (2006–2009), as shown in Table 3. From the 57^th (1997–2000), where state effects were just 7 percent, they rise to 18 percent in the 58^th (2000–2003), 27 percent in the 59^th (2003–2009), and then fall to 10 percent in the 60^th Legislature (2006–2009). The data show a massive shift in the nature of Mexican legislative politics in response to the institutional changes in the political arena, with a more than 400 percent increase in the effect of state divisions from the 57^th to the 59^th Legislatures (1997–2006). Discussing the results requires making several observations. First, there was a substantial drop in party cohesion, and state party divisions contributed significantly to that drop. Second, the drop in party cohesion also reflects non-federal effects – local vote- and career-seeking that do not correspond to the new federalism. Indeed, the percent attributable to state divisions never exceeds 30 percent, suggesting that many other factors also play a role in Mexican politics. Third, despite the fact that state effects explain only approximately 30 percent of the defections, the rise of state effects is impressive. For the 59^th Legislature, almost one quarter of defections is attributable to state divisions; this is a substantial figure in a country that previously had a highly centralized and unified party system.

OPEN IN VIEWER

Table 3

Measuring the Magnitude of State Effects

	Cohesion
Period	Overall	PRI	PAN	PRD
57th	6.7%	11.8%	3.8%	8.3%
58th	18%	27.3%	6.8%	9.3%
59th	27.2%	30.4%	8.8%	7.4%
60th	9.8%	14.8%	8.4%	8.5%

Note: Values are the percentage of deviations from perfect cohesion that can be attributed to state effects.

Source: Authors’ own calculation.

Figure 2 shows results for Mexico's three biggest parties, the PRI, PAN and PRD. In these graphs, each figure in the matrix shows results for a different party; the legislatures are numbered on the x-axis, and cohesion is measured on the y-axis. The vertical range of the red box shows the range of the 10,000 permuted subgroup cohesion scores. The solid line shows state party cohesion, and the dotted line shows national party cohesion. Note that the y-axis is not constant, but is scaled to each party. A glance at the dotted lines clearly shows the dramatic changes in overall party cohesion during this period. The PRI began with nearly perfect cohesion in the 57^th Legislature (1997–2000), which plummeted to under 0.80 for a national average for the 58^th (2000–2003) and 59^th (2003–2006), then rose above 0.90 again in the 60^th (2003–2006). The leftist PRD had high cohesion in the 57^th (0.96), which fell for the 58^th (2000–2003), rose for the 59^th (2003–2006), and again fell in the 60^th (2006–2009), this time to its lowest point.

The PAN began with high cohesion, but this fell to 0.86 during Fox's first term. In part, this reflected a transition for the party: with more than fifty years contesting government and playing the role of an opposition party, the PAN had some trouble adapting to the role of governing party. In addition, President Fox was a party outsider who lacked solid support in his own party. In contrast, President Calderon (2006) leveraged his status as a party insider to solidify relationships with legislative party leaders and improve coordination with the legislative branch on policy. Since then, PAN cohesion has been the highest of the three parties.

We can again interpret these results by calculating the percent of defections attributable to divisions, as above. Party-by-party standardized figures are reported in Table 3 and are consistent with our hypothesis: the biggest effects were observed for the PRI. For the 57^th Legislature (1997–2000), even with extremely high cohesion, 12 percent of defections were already explained by state conflict. ¹⁹ After 2000, once out of the presidency, state divisions within the PRI rose immediately to represent 18 percent of defections, increasing further to 27 percent during the following legislative period. The PAN, in contrast, has the lowest levels of state-based defections, never exceeding 10 percent. The PRD enjoyed a steady 7 to 9 percent over the entire period.

19

This is consistent with previous research, which has argued that while the PRI quelled open state conflict, much of the internal politics of the PRI reflected struggles between local organizations.

Finally, Figure 3 shows the permutation analysis with legislators categorized according to party, electoral rules (single-member district or closed-list proportional representation), and gubernatorial coalition status (government or opposition). Graphs labeled “CLPR” are for deputies elected under closed-list rules; graphs labeled “SMD” are for majoritarian districts. Similarly, graphs labeled “GOV” are for deputies from states where a co-partisan controls the governor's office and “OPP” for deputies whose home governor is from an opposing party.

The results are consistent with our core hypotheses. The largest and most consistent effects are for SMD legislators, especially those from the PRI and with co-partisan governors. Using the metric to measure impact presented above, for PRI SMD-GOV deputies, state party defections are always significant at the .05 level, and account for 20 to 34 percent of defections across the four periods. There is also evidence of state party defections for SMD deputies facing opposition governors, but their size is, on average, smaller. Evidence of state party defections among legislators elected under closed-list PR is weakest: only a few of the cases have significant state-based differences, and the magnitude of their impacts is almost always fairly small. Finally, the PRI is again the party most likely to suffer these divisions – it accounts for half of all significant subnational defections and has the largest number of substantive impacts.

The massive decentralization of Mexican politics did transform the political system, but with differential effects within and across parties. Politicians elected under personalistic rules had stronger ties to subnational gatekeepers than those elected under the multi-state closed-list system. The presence of a co-partisan governor increased opportunities and pressure for legislators to defect from their party line. And the PRI, with less ideological cohesion to unify as a national party, suffered the largest state divisions once resources were decentralized from the president to state actors.

In this paper, we have examined the new federalism of Mexico and its impact on the coherence of national political parties. Using an original dataset of roll-call votes from the last four legislative sessions, we have demonstrated the dramatic transformation that has occurred in the Mexican political system. Under PRI rule, Mexican politics was highly centralized with vertical networks of patronage and career advancement. There was no hint of subnational autonomy, and governors served only to reinforce legislators’ loyalty to the president and national party. Since the PRI's loss of the presidency, the system has been dramatically transformed. From one of the most centralized systems in the world, it has become one of the most decentralized. Evidence of conflict between state and national interests, which was either suppressed or dealt with quietly under PRI dominance, is now a frequent news item (Díaz-Cayeros 2005). Governors who once resigned on command now insult the president and challenge national authority over local elections.

An important part of this transformation is the federalization of national parties. Over the last four legislative sessions, subnational actors have had an increasing influence on national politicians. From virtually no impact in the first period, there are now significant subnational effects for all three major parties, which can be linked to state governors, parties, and electorates. Our research suggests that the interaction of several factors explains the extent of these new federal effects. In particular, the combination of personalistic electoral rules, resource decentralization, and a pork orientation maximize the frequency and magnitude of state party defections in national parties. All these seem to matter independently, but their combined effects were particularly dramatic. ²⁰

We highlight one of our important points for the study of comparative federalism. Our results emphasize the powerful role of distributional politics in regional party divisions. Parties relying more on ideology for electoral success were more likely to unite on national politics, even in the face of potential pressure from co-partisan governors. In Mexico, the more ideological parties show less evidence of state-based conflict and less responsiveness to system transformation. In contrast, the PRI – with its long history of distributive politics – is highly responsive to shifting resource control from the presidency to subnational executives.

For Mexico, these changes may be a double-edged sword. On one hand, they have reversed patterns of accountability, with subnational politicians now devoted to subnational interests and actors. For proponents of decentralization, these changes may be positive, as they may contribute to decreasing corruption, improving public goods provision, and uniformly improving representation. On the other hand, this decentralization has also created space for subnational authoritarianism and it threatens central government effectiveness. Regarding subnational authoritarianism, some politicians have created state fiefdoms. Combining clientelism and single-party dominance at the state level, these systems suffer from rampant corruption, electoral fraud, and ineffective government (Gibson 2005; Cornelius 1999, 2000; Snyder 1999). However, Mexico's central government has so far proven less effective at imposing good government than that of other federal systems – Brazil's, for example. In terms of the effectiveness of the central government being threatened, increasing state divisions might eventually weaken national parties’ ability to present a unified front to voters, to develop their own brand name – and thus consolidate the party system – and, ultimately, to advance a cohesive national agenda.

For students of federalism, the case of Mexico suggests a broader model of party system nationalization, and emphasizes the role of ideology and distributive politics in party system nationalization. In particular, the Mexican case reinforces the importance of distributive politics in nationalizing or regionalizing political parties, following Desposato and Scheiner (2009) and Chhibber and Kollman (2004). Where party identities are weaker, and distributive politics stronger, national legislators will be responsive to any actor that controls distributive goods. For PRI legislators, when resources were centralized, they were the most disciplined and cohesive of national parties. Once the PRI lost the presidency, PRI deputies shifted their attention to subnational gatekeepers – making that party the least disciplined and cohesive, partly due to subnational conflict. In contrast, the more ideological parties were less responsive to both the transformation of the political system and the presence of subnational governors.

At the same time, the process of change in Mexico illustrates the idea that centralization itself should be understood as endogenous to the political system. In the Mexican case, some of the most important centralizing factors were dependent on PRI dominance – and PRI dominance was dependent on those centralizing factors. One-party dominance concentrated power in the presidency and debilitated the independent influence of subnational gatekeepers. But the PRI's gradual loss of a congressional majority followed by its loss of the presidency ended the centralization of Mexican politics. The result was decentralization of political authority and resources, and with it, a new federalism for Mexico.