Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/231803720

GubernatorialEffectsontheVotingBehaviorofNationalLegislators

ARTICLEinTHEJOURNALOFPOLITICS·MARCH2011

ImpactFactor:1.48·DOI:10.1017/S0022381611000326

CITATIONS

4

DOWNLOADS

172

VIEWS

138

2AUTHORS:

GuillermoRosas

WashingtonUniversityinSt.Louis

30PUBLICATIONS169CITATIONS

SEEPROFILE

JoyLangston

CentrodeInvestigacionyDocenciaEconomi…

13PUBLICATIONS63CITATIONS

SEEPROFILE

Availablefrom:JoyLangston

Retrievedon:16August2015

Gubernatorial Effects on the Voting Behaviorof National Legislators

Guillermo RosasAU1 Washington University in St. Louis

Joy Langston Centro de Investigacion y Docencia Economicas

Are subnational political elites, such as governors, capable of affecting the voting behavior of nationalrepresentatives even in the face of high legislative discipline? We address this question by estimating the exogenouscausal effect of gubernatorial influence on the voting behavior of national legislators in Mexico, whereconstitutional provisions guarantee that the political survival times of governors and legislators are fixed andknown ex ante and where we can thus know precisely which legislators will leave congress before their stategovernor’s term has expired. We posit that such legislators will be more willing to represent gubernatorialpreferences because they expect the governor to find them employment upon finishing their stay in congress. We findthat governors whose terms end after the term of the national assembly are able to systematically increase thevoting cohesion of legislators from their own party and state and that these effects are substantively important.

There is growing interest among comparativeanalysts of legislatures in understanding whetherpowerful subnational actors are able to influ-

ence the behavior of national legislators (Benton 2009;Cantu and Desposato 2009; Carey 2007; Carey andReinhardt 2004; Cheibub, Figueiredo, and Limongi2009; Desposato 2003, 2004; Jones and Hwang 2005;Pereira and Mueller 2004). The issue of subnationalinfluence over national legislators is especially impor-tant in new democracies and in those weakened bycontinuous economic and social upheaval, such asmany of those found in Latin America. Dıaz-Cayeros(2006) and Garman, Haggard, and Willis (2001) haveargued forcefully that the relationship between thestates and the federation can have a profound impacton policy making, especially in the fiscal arena.Coherent policymaking can indeed depend on theability of national politicians to discipline state actors,such as governors, so that subnational conflicts donot preclude national Pareto-improving solutions(Rodden and Wibbels 2005). Conversely, governorscan be instrumental in ensuring that copartisanlegislators from their states support bills that areimportant to national party leaders. In any case, thepossibility that governors or other state elites may affectthe voting behavior of their state’s delegation in thenational congress is an important topic in the agendaof federalism.

Persuasive theoretical arguments suggest thatsubnational political elites (governors) can in principleaffect the voting behavior of national representatives(deputies) in federal regimes. Empirical verification ofthese arguments, however, has remained elusive. Thesearch for ‘‘subnational effects’’ has led scholars tolook for instances in which national legislatorscoming from the same state vote in a similar wayto the detriment of national party unity. It indeedmakes sense to look for state effects wherever legis-lators face cross-pressures from national and stateactors but, as we explain below, gubernatorial effectscan exist even when the interests of state and nationalelites are aligned rather than at odds. Our firstcontribution is to confirm that state effects exist evenin federal legislatures with relatively disciplinedparties. We study whether gubernatorial influencesystematically determines the voting cohesion oflegislators that belong to the governor’s party andthat hail from the governor’s state, i.e., those thatbelong to the same ‘‘state-party delegation.’’ Weexamine whether a governor can push her state’sfederal deputies to vote as a cohesive unit, regardlessof whether they vote with or against their party’smajority. While this might seem a less direct way ofmeasuring state effects in the national legislature, it isin fact a different empirically verifiable implication ofthe idea that state political elites may intervene in

The Journal of Politics, Vol. 73, No. 2, April 2011, Pp. 1–17 doi:10.1017/S0022381611000326

� Southern Political Science Association, 2011 ISSN 0022-3816

1

national politics through federal representatives.1

With this approach, we have the leverage to discernwhether gubernatorial effects exist even when gover-nors might be in agreement with national leaders.

More importantly, previous studies of state effectsrely on observational data where the mechanism thatassigns gubernatorial influence is unknown. Underthese circumstances, distinguishing the causal effect ofgubernatorial pressure from the dictates of nationalparty leaders, voter demands, or ideological similaritiesamong legislators belonging to the same state is a talltask. To put it differently, demonstrating empiricallythat state effects exist does not necessarily mean thatthese effects can be attributed to gubernatorial influ-ence. Assessing the potential influence of governors onthe behavior of deputies in statistical analyses requires,at the very least, that we understand the process thatassigns varying degrees of gubernatorial influence.

In this work we use differences in the expectedlength of political careers of deputies and governorsas a proxy for gubernatorial influence, as we explainbelow. Given the possibility of reelection, theproblem researchers face is that we do not generallyknow at the time of measurement whether thepolitical careers of governors and deputies are aboutto finish (a strong incumbent surprisingly loses anelection), drastically change course (anticipating atough election, an incumbent chooses to switchparties), or be extended for the foreseeable future(an incumbent facing a tough challenge actually winsreelection). Lack of knowledge about the mechanismsthat ‘‘assign’’ length of political careers to governorsand deputies opens the possibility of measurementerror in determining the power of governors as wellas the ability of deputies to withstand their pressures.Omitted variable bias is also a concern, as the policypreferences of deputies and governors may both beshaped by subnational level factors that are notexplicitly controlled by the researcher. Finally, reversecausation may be at play, for example, if the votingbehavior of a particularly cohesive group of deputieshelps extend the political career of their state’sgovernor. Under these circumstances, measures ofstatistical association between governor characteristicsand the voting behavior of representatives cannotidentify a causal effect (Holland 1986; Rubin 1974).

Our second contribution is to solve this ‘‘un-known survival time’’ problem, providing convincingevidence that governors have a causal impact on the

behavior of national legislators. We estimate guber-natorial effects on the voting cohesion of party-statedelegations by taking advantage of variation in theterms of governors and congressional representativesthat is exogenous to the hypothesized relationof cause and effect. By considering the case ofMexico, our identification strategy capitalizes ontwo constitutional features of that country’s federalpolitical system: (1) the staggered calendar used toelect governors2 and (2) the contrasting politicaltime-horizons of governors, who are elected fornonrenewable six-year periods, and their deputies,who are elected for three-year periods without thepossibility of immediate reelection.

Because of these features, we know at the begin-ning of any given legislature that some governors willoutlast the terms of their state representatives to thenational congress while others will leave their postsbefore the end of the legislative term. Some governorsare thus in a position to reward loyalty by securingjobs for their deputies who must cycle out ofCongress at the end of their single three-year term.One would expect that deputies from states wherethe governor remains in office past the end of thelegislative term will be more willing to representthe governor’s preferences because they depend onthis political leader for a future job. Simply put, wehypothesize that the governors who outlast theirstate’s copartisan legislators are more influential.Because gubernatorial and congressional electoralcalendars are fixed, and because neither governorsnor legislators are capable of unilaterally extendingtheir terms of office, our design provides a plausibleclaim of exogeneity for ‘‘expected political survivaltimes.’’ While this constitutional combination mightnot exist in other nations, our findings shed lighton how legislative politics work in federal regimesbecause we provide abundant evidence that the causallink between gubernatorial influence and legislators’voting behavior is not spurious.

We present the case for the existence of guber-natorial effects in four sections. In the first section,we expand upon the reasons that lead us to expectgubernatorial effects on voting behavior; we focusparticularly on previous evidence of state effects inMexico and elsewhere and on the instruments thatallow Mexican governors to shape the voting be-havior of copartisan national deputies from theirstates. In the second and third sections we describe

1We focus on explaining differences in the voting cohesion ofsubsets of legislators, an oft-used strategy in the literature on stateeffects (Cantu and Desposato 2009; Carey 2009; Cheibub,Figueiredo, and Limongi 2009; Jones and Hwang 2005).

2Between 3 and 10 states (out of 32) elect new governors eachyear except the second of a president’s six-year administration.State constitutions determine the timing of elections.

2 guillermo rosas and joy langston

our data, discuss operationalization of relevantindicators, and develop and test our model oflegislators’ voting behavior. Our empirical strategyinvolves estimating an item-response theory modelbased on roll-call data for three successive nationallegislatures in Mexico’s post-1997 democratic era.This strategy yields individual-level ideal points,which in effect provide summary scores of the votingbehavior of legislators.3 We focus on the variance ofthe ideal points of legislators that belong to the samestate-party delegation to build ‘‘delegation dispersionscores,’’ our proxies for voting cohesion—a strategyused by Jones and Hwang (2005). We then modelstate-party delegation dispersion scores as a functionof governors’ remaining time in office. If stronggovernors are indeed capable of promoting votingunity among copartisan deputies from their states,we would expect the length of gubernatorial tenureto depress state-party delegation dispersion scores.Indeed, we find compelling evidence of gubernatorialeffects among Mexico’s national legislators, eventhough these effects do not necessarily lower nationalparty cohesion. We elaborate on the contribution thisdiscovery makes to current debates in the conclusion.

The Possibility of GubernatorialEffects

Students of legislative behavior often note thatlegislators are beholden to multiple principals whoseinterests may at times be at loggerheads (Carey 2009).Examples of such principals include voters in thelegislators’ home districts, the country’s president,national party leaders or subnational politicalelites. Because of the prominent position of subna-tional political elites—especially governors—in federalsystems, scholars studying federal legislatures oftenexpect to uncover state effects in national legislativevoting. Indeed, under pressure from competingprincipals (national party leaders and governors)one would expect to see legislative parties unableto muster voting discipline, presumably resulting inlow levels of party unity.

There are, indeed, empirical findings that suggestthat federal legislators often vote with their statecounterparts, even when this might put them atcross-purposes with their legislative party leaders. Inparticular, an interesting debate has developed in theliterature on Brazil’s legislature concerning the powerand influence of state governors. Samuels (2000)notes that despite the legal possibility of consecutivereelection, many deputies do not stay in the BrazilianChamber, preferring to cycle back to the stateand municipal governments that control importantresources, a disposition he refers to as ‘‘regressiveambition.’’ Because governors and other local polit-ical elites control resources, it stands to reason thatnational deputies may be keen to do the governors’bidding in national congress. Despite the relativelystrong standing of Brazilian governors, however,Desposato (2003, 2004) and Cheibub, Figueiredo,and Limongi (2009) detect only weak state effects inthe national congress, where national party leadersand the executive are powerful enough to disciplinebackbenchers and therefore mute the potentiallydisruptive consequences of Brazil’s federal system.4

In a different setting, Hix (2002) argues thatmembers of the European Parliament—a quasi-federalinstitution—are more likely to vote with their‘‘national delegation’’ than with their supranational‘‘party bloc’’ when faced with competing demandsfrom different principals. In this case, sub-Europeannational elites are seemingly more powerful thanEuropean party leaders in shaping the voting behaviorof European parliamentarians. In contrast, Jones andHwang (2005) find no evidence of state effects in thevoting behavior of deputies in Argentina’s province-centric federal arrangement, a phenomenon theyattribute to the willingness of governors to delegatecontrol of their deputies to the national party in returnfor fiscal gains.

Because the literature mostly considers stateeffects as emerging from competing principals’ dilem-mas, tests to detect these effects often look for instancesof clear conflict between state elites and nationalelites to investigate how deputies vote in thesecircumstances. As Cheibub, Figueiredo, and Limongiput it, ‘‘[i]n deciding how to behave in the assembly,legislators must consider not only their personalpreferences but also those of their constituents, their

3In educational testing, item response theory (IRT) is used toinfer students’ abilities based on responses to exam questions(items). In political science, IRT models are commonly employedto infer the ‘‘ideological stance’’ of legislators based on theirobserved votes on bills, yielding inferences about ‘‘ideal points’’that are similar to Nominate scores (Poole and Rosenthal 1997).See Clinton, Jackman, and Rivers (2004) and Martin and Quinn(2002) for applications of IRT in legislative studies.

4The debate on Brazil centers less on whether governors affect thebehavior of national deputies than on whether their influence canbe overcome by party discipline. For more on this debate, seeAmes (2001), Armijo, Faucher and Dembinska (2006), Hagopian,Gervasoni and Moraes (2009), Pereira and Mueller (2004),Samuels and Shugart (2000).

xxx 3

parties, and regional and national leaders. Thesepreferences rarely coincide, and the issue becomeshow to detect whether existing conditions favor onetype of interest over another’’(2009, 2). Indeed, opendisagreement between national and subnationalelites that ends with deputies supporting the latterconstitutes the most obvious telltale evidence of theexistence of state effects.

However, the possibility of state effects on legis-lators’ voting behavior exists even in the absence ofcompeting principals and even in the presence ofrelatively high party discipline in congressionalvoting. To see why, imagine a country in whichgovernors are the only actors with the means,opportunity, and motivation to control the votes oftheir states’ deputies. If governors from the sameparty were able to coordinate and support similarpolicy positions, all legislators would vote togetherand we would never see within-party variation invoting behavior. Examining the frequency of voteswith or against the party majority would lead us toconclude erroneously that national party leadersimpose discipline when in fact state actors are solelyin control of the behavior of national representatives.In a less drastic scenario, national party leaders couldenlist the support of governors to ensure thatcopartisan deputies from their state systematicallyuphold the national party line. This type of stateeffect could even increase the voting cohesion of thecongressional party bloc, resulting in higher partydiscipline than would obtain in the absence ofgubernatorial pressure. In short, the possibility ofgubernatorial influence can exist even in situations inwhich the interests of multiple principals with a stakein the voting behavior of national legislators are notcompletely divergent. In these circumstances, wemight not expect state effects to systematically erodea party’s capacity to vote cohesively, but we wouldcertainly expect state elites to be a force to bereckoned with. Gubernatorial influence over legisla-tors’ voting behavior can thus appear in situationsof multiple principals, whether they compete overlegislators’ loyalties or not, a possibility that has notbeen widely considered.

Alongside this issue, the empirical literature hasyet to produce convincing evidence about the causalmechanisms that presumably produce state effects.Even where patterns of correlation in the votingbehavior of deputies from the same subnationaldelegation can be documented, it is not entirelyobvious that state effects should be attributed to theinfluence of governors or other subnational politicalelites. Legislators belonging to the same subnational

delegation may vote in similar ways for a variety ofreasons. Pressure from their state governor to voteconsistently in the same direction is certainly apotential mechanism, but so is the existence of sharedideological preferences or links to constituencieswith similar socioeconomic characteristics. For thisreason, designing tests that can help us gauge thecausal impact of subnational elites on deputies is anobvious concern. Our research design allows us tounequivocally trace differences in subgroup votingbehavior to gubernatorial influence.

Our tests take advantage of constitutional featuresof Mexico’s federal arrangement, where governorshave recently taken on more conspicuous politicalroles. Before the transition to democracy inthe late 1990s, the Mexican Congress was anextremely hierarchical and centralized legislativebody where the hegemonic Partido de la RevolucionInstitucional (PRI) consistently won more than 95%of all single-member districts, small oppositionparties were largely figureheads without the abilityto influence policy, and legislators rarely, if ever,voted against their caucus leaders (Nacif 2002). Yet,the federal political structure embodied in theMexican constitution has resurfaced more recently,especially after the PRI’s loss of the presidency in2000 (Hernandez Rodrıguez 2003).5

Mexico is characterized by wide cross-state di-versity in terms of levels of development, socialstructures, and political preferences, which is acommon feature of federal systems. This diversityforces politicians to adopt different vote-maximizingstrategies at the state level even as they nurture acommon national programmatic identity. Greene(2008) shows that state delegates to the 1999 PRINational Convention held ideological preferencesthat were determined by the type of opposition theyfaced in different states. If state representatives toCongress hold policy positions similar to those ofstate delegates to the PRI National Convention—aplausible assumption—the very possibility of effectiveelectoral competition could lead to distinct stateeffects in congressional voting. These effects wouldobtain regardless of gubernatorial pressure. Indeed,based on an analysis of legislative voting behavior,Cantu and Desposato (2009) substantiate the exis-tence of ‘‘state effects’’ within some party delegationsin the Mexican Legislature. In other words, when theunit of analysis is the state-party delegation, as

5As Dıaz-Cayeros (2006) notes, governors in the early post-Revolutionary period were willing and able to stand up to thenational government to push for policies they preferred.

4 guillermo rosas and joy langston

opposed to the state delegation, Cantu and Desposatoshow that the incidence of congruent voting behavioris higher than one would expect by chance. Using adifferent metric and research design, Betanzo (2009)is able to substantiate cross-state variation in thenumber of legislator-sponsored initiatives to reformthe national fiscal code, a finding that suggests greatvariation in levels of state assertiveness concerningfiscal matters.

We still need to show, however, that voting unitywithin state-party delegations is at least partiallydriven by governors. To explore the possibilityof gubernatorial effects, we need to ask whethergovernors have the means, motivation, and oppor-tunity to press federal representatives from theirstates to vote together in a systematic fashion.First, governors spend resources to help federalrepresentatives get elected. The reason behindthis decision lies in Mexico’s peculiar fiscal system(Dıaz-Cayeros 2006; Flamand 2006). A governor thatis able to send a large contingent from his state to thenational congress will usually have more success atwinning federal funds in the annual budget rounds.6

State executives raise extremely low proportions oftheir state budgets and depend on the federal govern-ment for up to 95% of their annual spending.Therefore, the budget rounds that take place everyyear in the Chamber are crucial for the state’s abilityto spend in the next fiscal year.

Mexican governors are also extremely active inpromoting local candidates for national positionsand in nurturing their political careers. Since thelate 1990s, governors have increased their abilityto nominate congressional candidates and haveaugmented their control of public resources withoutsuffering a concomitant change in making theirexpenditures accountable to voters. Admittedly, formalprocedures to select candidates to federal congressvary widely across the three major parties (Wuhs2006, 43). The former opposition parties—PartidoAccion Nacional (PAN) and Partido de la RevolucionDemocratica (PRD)—have always had more decentra-lized candidate selection practices than the PRI and didnot change their selection procedures radicallyduring the transition to democracy (Bruhn 1997;Mizrahi 2003). They rely mostly on district-levelnominating conventions (PAN) and primaries opento all voters (PRD) to select congressional candidates.Due to the demands of rising electoral competition, the

PRI informally devolved much of the party’s selectionof single-member district (SMD) congressional candi-dates to its state governors even before their loss ofthe presidency. However, there is ample evidencethat despite variation in formal nomination procedures‘‘governors from all major parties have heldincreasing power over SMD candidacies for theChamber since [ . . . ] the mid to late 1990s’’ (Langston2010).

All else equal, local level control over nomi-nations undermines national party elites, but thesemay retain a measure of control if they have powerover campaign financing. Parties receive public fundsfor campaigning, and national party leaders arecharged with disbursing money to their state chaptersas they see fit. For example, more than a third ofall public funding transferred from the country’selectoral management body to political parties in2003 went to state affiliates, and almost one halfof those transfers were used to fund individualcampaigns (Poire 2005, 11–12). The rest of themonies was retained by the national party organiza-tions to spend on national media appeals to sell theparty label and its electoral platform as a whole—thatis, most of these funds were not spent on individualdeputy candidates. Governors of all three majorparties expend money and manpower to supportthe campaigns of copartisan deputy candidates,making up the difference between national partydisbursements and the amount individual deputiesneed to run competitive campaigns in single-memberdistricts.

Finally and most importantly, governors are notheld accountable for the vast amount of federalresources that they receive and spend each year(Flamand 2006; Pardinas 2008; Rıos 2009). Thefederal transfers that are earmarked for differentspending categories come legally under the purviewof the country’s national accounting agency, but inpractice this bureau finds it difficult to monitor orsanction misuse of funds. Another large portion offederal monies goes to revenue-sharing funds that arelegally considered part of the states’ budgets andtherefore are audited by the local assembly’s budgetoffice—which is almost always controlled politicallyby the governors. A large quantity of resources and alack of accountability in how they are spent allowgovernors to distribute a wide array of bureaucraticand political posts in their respective states. Accord-ing to Langston and Aparicio (2008), 62% of alldeputies return to their states for a political positionafter their three-year congressional term ends, whileover 72% of SMD deputies return to their states to

6Interviews with Francisco Suarez Davila, former PRI deputy andex-president of the Treasury Committee in the Chamber (July 10,2009), and with Ricardo Monreal, former PRD governor(February17, 2009).

xxx 5

continue their careers. Samuels’s ‘‘regressive ambition’’is certainly present in Mexico, where the politicalopportunity structure often drives ambitious politi-cians back to their home states. The ability to helpambitious deputies continue their political careers afterthe end of their non-renewable three-year term in theChamber allows governors at least shared control overtheir state’s deputies while they are in Congress.

We do not mean to imply from this discussionthat national party leaders lack influence on votingunity in Congress. To the contrary, national partyleaders and their agents in the legislature are crucialin determining committee assignments and haveagenda control over bills.7 As many specialists inMexican legislative politics have pointed out, caucusleaders hold an impressive array of intrachamberresources with which they can manage the votes oftheir legislators. These resources include money to hireassistants and buy office equipment, access to foreignjunkets, and, most importantly, committee leadershipassignments (Rivera Sanchez 2004; Weldon 2005).Legislative leaders control the flow of initiatives bothinto and out of committee and can remove their party’scommittee leaders with few regulatory constraints.

Nor do we mean to imply that governors willalways choose to direct state legislators to vote againstthe party line. However, disputes over the distri-bution of tax income and other fiscal matters couldpit governors against national party leaders; suchdisputes have been common over the past decade andhave led caucus leaders to take great care in negoti-ating with their party’s governors. Furthermore,many governors have political aspirations beyondthe state executive office and could conceivably usetheir influence over national representatives to builda recognizable personal reputation. Contrary to therigid hierarchy of the old political system, governorscan now confront the president without fear of beingremoved from their posts or allocated fewer federaltransfers (Figueras Zanabria 2009) and can thuswithhold support for prominent bills in a calculatedeffort to build notoriety. Yet, the reality of relativelyhigh party discipline in the Mexican Congress leadsus to discount the possibility of finding within-statesimilarities in voting patterns that are powerfulenough to overcome within-party similarities. Toparaphrase the French aphorism, we still expect two

PRI representatives, one of whom hails from Oaxaca,to have a more similar voting behavior than twoOaxacan representatives, one of whom belongs to thePRI. We do expect however that variations in thevoting behavior of national legislators will allow us todetect differences in the cohesion of the state-partydelegations to which they belong. Since state execu-tives have the means, motivation, and opportunity toinfluence the behavior of their deputies, we expectthese differences to be systematically driven byvariation in gubernatorial influence.

Research Design and DataDescription

We have argued that governors in Mexico, likegovernors in other federal political systems, havethe capacity and the motivation to affect the votingbehavior of copartisan national deputies that hailfrom the states they govern. Gubernatorial influenceis ultimately possible, especially in a system with noconsecutive reelection, because governors have somemeasure of control over the political careers oflegislators from their states. This control occurs boththrough legislative nominations that increase theprobability of selecting congenial agents and throughpromises of job offers to term-limited legislators.A governor will be better able to affect the politicalcareer of a legislator—and therefore his votingrecord—to the extent that her political life in thegovernor’s office exceeds the single three-year termof the legislator. Otherwise, legislators will feelless pressured to represent the preferences of theirgovernor. The constitutional prohibition of consec-utive reelection for legislators and state executivesrenders the assignment of differences in the expectedpolitical survival of governors and legislators trulyexogenous to observed voting behavior. Because theexpected congressional career of legislators is alwaysfixed at three years, governor terms provide sufficientinformation about differences in expected politicalsurvival. Furthermore, the staggered calendar ofgubernatorial elections guarantees variation in thelength of gubernatorial terms at the beginning of anygiven congressional term. These features make theMexican political system an ideal place to studygubernatorial influence over legislative voting.87The president, congressional representatives, and state legisla-

tures have the ability to initiate legislation. Every bill is turned toa committee, which is required to report bills back to the floorwithin a certain period. In practice, committees exercise greatdiscretion in the amount of time they keep bills off the floor(Nacif 2002).

8Dıaz-Cayeros (2006, 141–2) employed the period left in gover-nors’ terms—the same indicator we use here—to explain why thefederal government was able to institute an allocation agreementthat harmed the states’ fiscal interests in the late 1970s.

6 guillermo rosas and joy langston

Our analysis is based on roll-call votes from threecongressional terms—the LVIII (2000–2003), LIX(2003–2006), and LX (2006–2009) legislatures.9 Asis common in other legislatures, a large proportion ofvotes are extremely consensual. In our analyses, weexpunge all bills actively supported by a majority of99% of legislators in the three major parties. Thisdecision still leaves a high number of relativelyconsensual and therefore mildly informative votesin our data corresponding to bills that pass with thesupport of 95 to 99% of legislators in the three majorparties. The method that we use to infer legislators’ideal points appropriately discounts the amount ofinformation provided by such consensual votes, thusobviating the need to weight them down.10

Table 1 confirms that party discipline remainsvery high within the Mexican Congress. The top partof Table 1 displays the average proportion of legis-lators that voted with their party (% Aye), againsttheir party (% Nay), or registered an abstention (%Abs); the complement to 100 is the average percent-age of legislators absent from the floor on any givenvote.11 These statistics provide a good sense ofthe differential ability of parties to impose votingdiscipline. In our models, we limit our focus tolegislators from the three major parties. Even thoughlegislators from small parties became increasinglynumerous during the last decade, the size of statedelegations from these parties is still too smallto make meaningful statistical inference possible;furthermore, no state governorship has yet beencaptured by a small party.12 As expected, Mexicanlegislators vote with their caucus leadership more

often than not and, though not obvious frominspection of Table 1, contentious votes often breakdown along party lines.13

Consider now the distribution of governor term,the main independent variable in this study. Wesummarize this information graphically in Figure 1,where we identify the partisanship of Mexicangovernors with black (PRI), grey (PAN), and white(PRD) timelines. To explain our coding decisionsconsider the case of Baja California, the second statefrom the top in Figure 1. The graph indicates that thePAN governed this state at the beginning of theobservation period in 2000 (grey line). The sittinggovernor, who had taken office in 1995, had one yearleft in his term before a new governor was sworn in. Inconsequence, the terms of legislators from Baja Cal-ifornia outlast the governor’s term; this state’s dele-gations are all awarded a value of ‘‘1’’ and their votingrecord comprises the first two semesters of the LVIIIlegislature. Delegations starting in Congress with termslonger than the state’s governor are identified by acrossed pattern in Figure 1. A new PAN governor tookoffice late in 2001; because at this point the governorhad six years to go in his mandate, all party delegationsfrom Baja California are awarded a value of ‘‘6’’ andtheir voting records comprise the remaining foursemesters of the LVIII legislature. In other words, wecode our observations as if this state had beenrepresented by two separate sets of delegations duringthis legislature: the first set sat in Congress during thefirst two semesters of the LVIII legislature under alame duck governor (crossed-out segment), whereasthe second set was in Congress during the last foursemesters of the LVIII legislature under a brand newgovernor. Consequently, we estimate two ideal pointsfor each Baja Californian legislator: the first ideal pointsummarizes the legislator’s voting behavior under thefirst governor; the second one summarizes her votingbehavior under the second governor.14

9We downloaded all registered votes from the official websiteof the Camara de Diputados at gaceta.cddhcu.gob.mx and atwww.diputados.gob.mx/Votaciones.htm. We thank Jeff Weldonfor making us aware that procedural votes, votes on private bills,and votos en lo particular are not posted online (privatecommunication and Weldon (2002)).

10Estimates of ideal points are not driven by a few legislators thatvoting systematically against a majority in their party.

11Statistics in Table 1 are based on all votes, including extremelyconsensual ones. As is clear from the table, about 2% of allindividual-level vote choices were registered as active abstentions(i.e., the legislator declares herself neither in favor nor against thebill). We recode abstentions as votes against the party majority(on this issue, see Weldon 2002). In contrast, we assume thatabsences produce missing values at random. The main effect ofthis decision is to increase estimation uncertainty about legis-lators’ ideal points.

12Some legislators started their stints in Congress as members ofa major party but became independent at a later date. In thesecases, we ignore the voting record of these legislators as ofthe moment they leave the parties under whose banners they wereelected.

13A majority in one party votes against majorities in other partiesca. 20% of the time.

14We add two caveats regarding governor partisanship. First, theFederal District has a jefe de gobierno rather than a governor,which is irrelevant for the purpose of this study. The first electedjefe, Cuauhtemoc Cardenas, took office in 1997, but left two yearsinto his tenure to contest presidential elections. Andres ManuelLopez Obrador was elected in 2000, but also left early to contestpresidential elections. Interim jefes Rosario Robles (1999–2000)and Alejandro Encinas (2005–2006) are not counted as newgovernors. We also ignore interim governors in Baja California,Chiapas, Colima, Hidalgo, Jalisco, Morelos, and Tabasco (thesedo not appear in Figure 1). Second, two governors (SalazarMendiguchıa in Chiapas and Echeverrıa Domınguez in Nayarit)ran supported by a coalition of PAN and PRD. We code theformer as PRD and the latter as PAN.

xxx 7

To substantiate the claim that the assignment ofgovernor term is ‘‘as if’’ random, we broke down thecharacteristics of state-party delegations by length ofgovernor term. Our original dataset comprises a totalof 369 state-party delegations, 81 more than wouldobtain if the three major parties had representativesfrom all 32 states across all three legislatures (inpractice, the PRD does not have delegations from allstates).15 The large surplus follows from coding somestate-party delegations as two different delegationswhenever a new governor is elected midway throughthe legislature. We analyze a subset of 113 copartisanstate-party delegations, i.e., state-party delegations thatshare party with the state’s governor.16 We divide these113 state-party delegations according to whether theirgovernor is scheduled to remain in power for at leastthree years (new governor) or not (lame duck gover-nor) in order to see if the distribution of observablecharacteristics is similar across these two groups.Concerns about potential differences among thesetwo groups are reasonable, considering that state-partydelegations under lame duck governors comprise onlyabout one third of all observations.

We find that delegations under lame duck gov-ernors are on average slightly smaller (7.3 6 6.3

legislators) than delegations from states with newgovernors (9.4 6 6.6), which suggests inclusion ofdelegation size as a control in our models. Even then,the difference in mean size is not statistically signifi-cant.17 The distribution of party membership ofdelegations among these two groups also appears atfirst sight to be different; in both groups, the PRIaccounts for over half of the delegations, but PANdelegations are relatively more numerous than PRDdelegations among states with lame duck governors. Bethis as it may, we cannot reject the possibility that thedistributions of delegation party membership underlame duck and under new governors are similar.18

Though there are no statistically discernible differencesin the partisanship of delegations under new and lameduck governors, it is likely that relations betweengovernors and federal representatives are differentacross parties. For example, PRD deputies tend to bemembers of strong internal factions, and PAN gover-nors are somewhat constrained by a copartisan na-tional executive (the PAN has held the Presidency since2000). We prefer to be conservative and control for theparty membership of delegations in our models, butour results hold when we omit this variable.

TABLE 1 Descriptive voting statistics, distribution of seats across major parties, and number of votes thatprovide information

LVIII LIX LX

(2000–2003) (2003–2006) (2006–2009)

% Aye % Nay % Abs Seats % Aye % Nay % Abs Seats % Aye % Nay % Abs Seats

PRI 69.3 4.2 1.4 208 64.7 6.7 1.1 203 57.4 4.8 0.9 106PAN 82.0 6.6 1.3 207 64.3 6.9 1.0 148 71.6 5.8 0.5 206PRD 65.1 9.0 2.3 53 61.7 7.0 1.3 97 55.6 10.7 1.9 127

Other 32 52 61Uninformative votes 146 298 225Informative votes 171 386 241Total number of votes 317 684 466

15We code the state of origin of PR legislators as the state wherethey developed their political careers according to information inthe Sistema de Informacion Legislativa (sil.gobernacion.gob.mx/portal) and Trejo Delarbre (2003). We failed to obtain thisinformation for 14 legislators. In these cases, we coded thelegislator’s state of birth as an indicator of state origin.

16We exclude state-party delegations that do not have a coparti-san governor (say, the PAN delegation from PRI-governed Stateof Mexico), as our theory suggests that the governor’s influence islimited to copartisan deputies.

17The difference of means is not statistically significant at the 95%confidence level based on a two-tailed t -test (p-value50.104).Moreover, we fail to find a statistically significant associationbetween governor term (disaggregated) and delegation size.

18Among delegations with new governors, 42 are PRI, 17 arePAN, and 14 are PRD; among lame duck governors, the numbersare 22, 13, and 5, respectively. A test of the hypothesis that thefrequency of party membership among delegations with lameduck governors is similar to the frequency of party membershipamong delegations with new governors yields x2

2ð Þ � 1:54, withp-value 0.463.

8 guillermo rosas and joy langston

Modeling Gubernatorial Effects onCongressional Voting Cohesion

How can we distinguish the effects of gubernatorialpressure from the effects of ideological similarities orconstituent demands among legislators belonging tothe same party and state? Following on the insight ofPoole and Rosenthal (1997) we use ideal points basedon roll-call data as a summary of legislators’ votingbehavior. Our point of departure is the two-dimen-sional version of the Bayesian item-response theorymodel of Clinton, Jackman and Rivers (2004). Ourdecision to estimate two dimensions is based on theconsensus that Mexican politics have been stronglypatterned by a left-right distributive dimension(PRD, PRI, and PAN from left to right) and by anauthoritarianism/democracy regime divide that pit-ted the hegemonic PRI against the erstwhile opposi-tion parties (Domınguez and McCann 1996; Moreno1998; Robles 2009). In the two-dimensional model,

the probability that legislator i will vote in favor ofproposal j (i.e., yij 5 1) is a function of legislator-specific ideal points and item-specific parameters.19

Estimating ideal points based on no information otherthan the legislators’ voting record is the first step ofour analysis. In Figure 2 we show estimates of theseideal points. We mark with party-coded grey symbolsthe ideal points of all legislators. These plots confirmthat legislators are relatively cohesive within parties, as

FIGURE 1 Partisanship of Mexican governors, 2000–2009.

19Formally, the model is Pr yij 5 1� �

5 F b1jx1i þ b2jx2i � aj

� �,

where x contains ideal points and b and a are item-specificparameters. We identify this model by fixing the discriminationparameters b of four votes in each legislature; all other estimatesare scaled in relation to these four points. The anchoring votes pitPAN–PRI against PRD along the first dimension and PRI againstPAN–PRD along the second dimension. Two dimensions areclearly manifest in the LVIII and LIX legislatures, but the seconddimension fails to discriminate among parties in the LX Legis-lature (see Figure 2). Descriptions of the posterior distribution ofparameters are based on 500 draws from two chains thinnedevery 10th iteration after 50K burn-in scans (we use the Gelman-Rubin statistic to assess apparent convergence). Jags code for theMCMC sampler is available upon request.

xxx 9

in all legislatures we can more or less differentiateamong clusters corresponding to PAN, PRI, and PRD.After accounting for between-party variation, however,within-party variation is still large. For example, within-party variation along the first dimension in LegislatureLIX amounts to 30% of the total variance of idealpoints along the first dimension.20 Ample within-partyvariation in ideal points makes it possible to explorepotential state effects.

Indeed, the darker symbols overlaid on thelegislators’ ideal points show the centroids of state-party delegations, i.e., the average position of alllegislators that belong to the same contingent.21 Tofind these centroids, we use information on the partymembership and state of origin of legislators. In thebottom plots of Figure 2 we include uncertaintybounds about state-party delegation centroids; evenafter taking uncertainty into account, we notice thatseveral state delegations within the same party displayrecognizably different voting behaviors. This is acrucial finding: if the national party caucus werethe only actor with influence over legislators’ votes,one could still see differences among the ideal pointsof individual legislators belonging to the same party,but one would not anticipate these ideal points tocluster within recognizable state contingents. Differ-ences in the spatial location of state-party contingents,as captured by the darker symbols in Figure 2, arisebecause delegations from various states within thesame party do not always vote in the same direction.With evidence of variation in the spatial location ofstate-party contingents at hand, one can at the veryleast confirm that some state-level political force hasan effect on the propensity of copartisan legislatorsfrom the same state to vote together.22

One could presumably inspect the distinct spatialpositions of state-party delegations for traces of gu-bernatorial influence, but this is not the approach weemploy here. Instead, we focus on a second meaningattached to the notion of ‘‘state effects.’’ This secondmeaning concerns the dispersion of legislators’ ideal

points around their state-party delegation, i.e., theirdegree of voting cohesion as a group. This secondnotion of state effects is recognized in the literature,and is in fact exploited by Jones and Hwang (2005) intheir analysis of Argentina and by Cantu and Despo-sato (2009) in their study of Mexico. We also focus onthe second meaning of state effects and model thedispersion of legislators around the ideal position oftheir state-party delegation. Thus, the second step inour analysis starts with the construction of variancestatistics—which we refer to as delegation dispersionscores—for each state-party delegation. This step allowsus to gauge the potential effect of governor term on thevoting behavior of legislators.

If governors have the ability to sway the votes oftheir states’ representatives in Congress, we wouldexpect governor term to be a substantively significantpredictor of delegation dispersion scores because gover-nors with lengthy tenures are more likely to ‘‘take care’’of their states’ deputies than those who are leaving theirposts. We thus regress delegation dispersion scores ongovernor term. The exogenous nature of governor termmakes us confident that our estimates will not be biasedby reverse causality or measurement error. However, wecan still estimate the effect of gubernatorial influencemore precisely by controlling for other factors thatmight affect voting cohesion, even if these are notcorrelated with governor term. Legislative partisanshipand delegation size are two such factors. Political partiesvary in the degree to which they are factionalized orcentralized, and these characteristics can affect theability of national party leaders to rein in ambitiouscopartisan governors. The size of the state-party dele-gation is an obvious control, because smaller delega-tions tend to have, by virtue of sheer chance, lowerdispersion scores (Desposato 2003). None of thesecharacteristics are correlated with governor term (seefn. 18), but including them in our regression modelspermits a more precise estimate of gubernatorial effects.

We recognize one additional methodologicalquandary in our effort to estimate gubernatorial effects.Because our dependent variable is constructed fromuncertain estimates of ideal points, we need to accountfor this uncertainty in our models. We do so bydrawing 500 different sets of delegation dispersion scoresfrom the posterior distribution of ideal points. We thencarry out an equal number of regressions, one for eachset of delegation dispersion scores, on our independentvariables. We report regression coefficients averagedover the 500 runs; reported standard errors appropri-ately account for this added source of uncertainty.

We estimate models of delegation dispersion basedon all delegations, with at least two representatives,

20This statistic is the residual mean square of a regression of idealpoints on party dummies divided by the variance of all idealpoints, one per dimension. The corresponding statistics for alllegislatures are: 16% and 20% (LVIII), 30% and 31% (LIX), and13% and 85% (LX). As is obvious from inspection of the plots,party dummies have no predictive capability along the seconddimension in the LX Legislature.

21For example, if the positions of three legislators from the sameparty-state delegation are (1,1), (2,2), and (1,2), the state-partydelegation centroid is (4/3, 5/3).

22Cantu2009 provide evidence of differences in state-delegationRice scores; in their analysis, voting cohesion is indeed higher forstate-party delegations with copartisan governors.

10 guillermo rosas and joy langston

FIG

UR

E2

Su

mm

arie

so

fvo

tin

gb

ehav

ior

inth

ree

legi

slat

ure

s.T

he

top

plo

tssh

ow

idea

lp

oin

tso

fle

gisl

ato

rs(m

ean

so

fth

ep

ost

erio

rd

istr

ibu

tio

np

lus

50%

cred

ible

inte

rval

s).

Th

eb

ott

om

plo

tssh

ow

idea

lp

oin

tso

fp

arty

-sta

ted

eleg

atio

ns

(ave

rage

so

fth

ed

istr

ibu

tio

no

fid

eal

po

ints

of

legi

slat

ors

wit

hin

each

par

ty-s

tate

del

egat

ion

61

stan

dar

dd

evia

tio

n).

xxx 11

whose party membership matches the party membershipof their state’s governor (the average size of suchdelegations is 9.46 legislators).23 Thus, we pool togetherall PRI delegations from PRI-governed states, all PANdelegations from PAN-governed states, and all PRDdelegations from PRD-governed states.24 This is becausethe mechanisms we hypothesize as driving gubernato-rial effects—i.e., the governor’s ability to select con-genial types ex ante and to reward good behavior expost—should be most clearly seen amidst legislatorswho share a party with the governor. We expectgovernors with longer time horizons (governor term)to be able to impose voting homogeneity on statedelegations, therefore depressing levels of delegationdispersion.

Table 2 displays OLS estimates of governor termeffects and other parameters. We present four ver-sions of the model, each based on dispersion scoresalong dimensions 1 and 2. Model 1 pools all availableobservations together, as explained before, whereasModels 2 to 4 are based exclusively on PRI, PAN, andPRD delegations, respectively.25 Based on the modelthat pools all observations together, we point out thatdelegation size is a positive and statistically significantpredictor of dispersion, confirming the need toaccount in our models for the ‘‘size effect’’ notedby Desposato (2003). PAN delegations are on averageless cohesive than PRI delegations along dimension 1,whereas PRD delegations are on average more cohe-sive than PRI delegations along dimension 2.

Net of the effects of delegation size and legislativeparty, we find that governor term is indeed a negativeand statistically significant predictor of delegation dis-persion.26 In other words, governors with longer timehorizons in their posts preside over state-party dele-gations that show more cohesive voting behavior, allelse constant. We consistently obtain negative coeffi-cient estimates when we break delegations down bypolitical parties in Models 2, 3, and 4—though thecoefficient of governor term is not significant at conven-

tional levels for PRD state-party delegations alongdimension 1 and is significant only at the 93% levelfor PRI state-party delegations along dimension 2.27

The models displayed in Table 2 are based on thevoting records of both SMD and proportional repre-sentation legislators. In Mexico, PR legislators areelected in five circumscriptions of magnitude M 5

40. Each circumscription contains several states, afeature that forced us to code the provenance of PRdeputies as the state where they developed professio-nally (see fn. 16). We pool PR and SMD legislatorstogether because governors are active both in placingSMD deputy candidates and negotiating many of thePR candidacies and because many of the PR represen-tatives also return to their states after the end of theirsingle terms in office (Langston and Aparicio 2008).Note also that because PR deputies may be dispropor-tionately selected from individuals with careers tied totheir national party organization, their inclusion in ouranalyses would presumably bias estimates against ourexpectation of finding sharp gubernatorial effects.However, we also reestimate our main model basedexclusively on information from ideal points of SMDdeputies (Model 5 in Table 2). We do this becauseSMD deputies are more likely to have been nominatedby state elites and because these are the individuals forwhich state provenance information is most accurate.Though the number of legislators on which our resultsare based is diminished drastically, in practice we onlylose ten state-party delegations—those made up ex-clusively of PR deputies or those where only one SMDdeputy remains after excluding PR deputies.28 As canbe seen in Model 5, this produces a mild efficiency lossin our estimates. Despite this loss, coefficient estimatesfor governor term remain of similar substantive magni-tude and are still statistically significant.29

23We transform dispersion scores logarithmically to eliminate theskew produced by the zero bound on this indicator.

24We standardized the posterior distributions of ideal points inorder to pool delegation dispersion scores across legislatures.Because no single legislator could have voted in all threelegislatures, we lacked a common anchor that would give usthe ability to estimate ideal points on a common scale.

25We offer four similar specifications in the online appendix,except we code governor term as a dichotomous indicator thattakes the value of 1 when there are at least four years remaining inthe governor’s term. Substantively, the results are identical to theones presented in Table 2.

26One-tailed t-tests for the relevant coefficients in the pooledmodel yield p-values of about 0.01 or smaller.

27The fit of Model 4 is poor, as seen from adjusted R2 and Fstatistics; unfortunately, the number of PRD delegations fromPRD-governed states is rather small.

28As in our previous test, we consider whether the distributionsof size and partisanship are balanced across delegations with newand with lame duck governors. In this case, delegation size issmaller among new (6.1 6 5) than among lame duck governors(7 6 6), but this difference is not statistically significant. Thereare no significant differences in the distributions of parties acrossdelegations with new and lame duck governors.

29As an additional robustness test, we included a dummy indicatorto distinguish party-state contingents elected during the secondhalf of the Fox administration (2003–2006). The rationale is thatdispersion scores could be much higher if legislators know thatpositions in the federal bureaucracy may open up as the non-renewable presidential term comes to an end. Including thisindicator did not produce appreciable gains in the precision ofour governor term coefficient estimates, and the coefficient on thedummy variable itself was not statistically significant.

12 guillermo rosas and joy langston

TA

BL

E2

Pre

dic

tors

of

del

ega

tion

dis

per

sion

(lo

gsc

ale)

alo

ng

two

dim

ensi

on

s.M

od

els

are

bas

edo

nd

eleg

atio

ns

wh

ose

par

tisa

nsh

ipm

atch

esth

ato

fth

est

ate

gove

rno

r.C

oef

fici

ents

are

OL

Ses

tim

ates

(sta

nd

ard

erro

rsap

pro

pri

atel

yac

cou

nt

for

un

cert

ain

tyin

the

esti

mat

ion

of

idea

lp

oin

ts).

Mo

del

1M

od

el2

Mo

del

3M

od

el4

Mo

del

5

Po

ole

dp

arti

esP

RI

PA

NP

RD

12

12

12

12

12

Inte

rcep

t2

1.98

3†

(0.2

17)

21.

174†

(0.1

61)

22.

155†

(0.3

22)

21.

469†

(0.2

32)

20.

703†

(0.1

66)

20.

633†

(0.1

86)

21.

615†

(0.6

88)

21.

295†

(0.4

74)

22.

027†

(0.2

51)

21.

280†

(0.1

84)

Del

egat

ion

size

0.03

4*(0

.011

)0.

023*

(0.0

08)

0.05

7*(0

.020

)0.

039*

(0.0

14)

0.01

7*(0

.009

)0.

018*

(0.0

10)

0.02

5(0

.023

)0.

012

(0.0

16)

0.03

1*(0

.017

)0.

032*

(0.0

13)

Gov

ern

orte

rm2

0.09

7*(0

.042

)2

0.10

9*(0

.031

)2

0.10

5*(0

.063

)2

0.06

9(0

.046

)2

0.06

7*(0

.032

)2

0.17

1*(0

.036

)2

0.13

7(0

.135

)2

0.12

8(0

.093

)2

0.09

2*(0

.050

)2

0.10

4*(0

.037

)P

AN

gove

rnor

1.22

5†

(0.1

80)

0.25

2(0

.133

)1.

355†

(0.2

14)

0.39

4†

(0.1

57)

PR

Dgo

vern

or0.

085

(0.2

14)

20.

330†

(0.1

58)

0.17

3(0

.256

)2

0.22

2(0

.186

)A

dj.

R2

0.36

00.

205

0.11

70.

105

0.17

00.

458

0.00

60.

024

0.30

10.

165

F-s

tati

stic

16.7

318.

207

5.19

54.

684

3.98

713

.288

1.05

91.

233

11.9

946.

033

N11

311

364

6430

3019

1910

310

3

*p-v

alu

e,

0.05

,o

ne-

tail

edte

st;

†p-

valu

e,

0.05

,tw

o-t

aile

dte

st.

xxx 13

To assess the relative magnitude of estimatedeffects, recall that we scaled ideal points to have unitvariance. The coefficient estimate for governor term inModel 1 suggests that a governor starting her termwill be able to reduce the dispersion scores of statedelegations by almost half the variance of ideal pointscompared to a lame duck governor with one year left.Based on estimates in Model 1, Figure 3 showsgraphically the effect of governor term on dispersionscores of delegations from the three major parties. Ineach plot, ellipses represent the expected dispersionscores of average PRI, PAN, and PRD delegationsfrom states governed by their own party; the dottedellipses across all plots have constant size andrepresent the cohesion of an average delegation ofthe corresponding party. The three plots correspondto expected dispersion scores under a lame duckgovernor (governor term51 in the left-most panel), agovernor whose term will barely outlast the legislativeterm of the state’s delegations (governor term53.5,center), and a new governor (governor term56, right).The numbers in the center of the color regions showthe ratio of the area of the color ellipse to the area ofthe dotted ellipse. This ratio clearly conveys thesubstantive impact of gubernatorial influence. Forexample, the dispersion score of a PAN delegationunder a lame duck governor is twice as large as thedispersion score of the average PAN delegation (2.01).We would expect the dispersion score of the samePAN delegation to drop to slightly more than two-thirds of the score of an average PAN delegation undera brand-new governor (0.72). As is clear from Figure3, a similar impact obtains across all parties even afteraccounting for differences across legislative parties.

The effect of governor term on the voting cohe-sion of state-party delegations should be predomi-nantly observed among state delegations from thegovernor’s own party and not on deputies from thesame state but from different parties. Still, wechecked whether governors systematically affectedthe voting cohesion of legislators from their statesthat did not belong to their own party, which createdsix potential combinations of delegation and gover-nor parties (i.e., PRD delegations from PRI-governedstates, PRI delegations from PAN-governed states,etc.) along both dimensions, estimating the effect ofgovernor term while controlling for delegation size. Formost of these models, the estimated coefficients ongovernor term were negative but statistically insignif-icant at conventional levels with one exception: theeffect of governor term on voting cohesion is negativeand statistically significant when considering the in-teraction of PRI governors with PRD delegations

(along both dimensions) or with PAN delegations(along the second dimension). These effects were notsubstantively large for PAN delegations, but they wererelatively sizeable for PRD delegations. This is asurprising finding: we think it possible that politicallysavvy governors (especially from the PRI) couldinfluence the voting behavior of state delegations fromthe PRD, because the members of the PRD share sofew copartisan governors and as a result find itnecessary to cut deals with governors from otherparties to survive politically. Indeed, we find evidencethat returning legislators from other parties, especiallythose from the PRD, benefit as well from the gover-nors’ largesse.30 An alternative possibility is that PRDdeputies vote cohesively against new governors fromother parties. Without careful examination of theideological location of state-party delegations—a taskwe leave for future work—we cannot conclusivelyarbitrate between these hypotheses.

Conclusion

We have proposed a strategy to uncover potentialgubernatorial effects in the voting behavior of legis-lators in a federal democracy, one with relatively highdegrees of national party discipline. Following Jonesand Hwang (2005), we use information from roll-callvotes to estimate legislator ideal points which we thenuse to build state-party delegation dispersion scoresthat are modeled as a function of gubernatorial timehorizons. Our research design takes advantage ofconstitutional features of the Mexican political systemto gauge gubernatorial influence over the cohesion ofstate-party delegations: because the lengths of thepolitical careers of governors and legislators areexogenously set by electoral laws, we can more crediblyestimate gubernatorial effects on legislative voting. Weconclude that gubernatorial effects are indeed discern-ible in the voting behavior of state delegations, even inthe face of high national voting cohesion. In a politicalsystem where, barely two decades ago, presidents couldcredibly threaten to remove problematic governors, we

30We have information about the first job that all 1,168 deputiesserving between 1997 and 2006 obtained upon leaving Congress.Among these, 583 were ‘‘political orphans’’ in that their state’sgovernor belonged to a party different than their own. Most ofthese orphans return to their states seeking jobs in municipaloffices, state government offices or their party’s state organiza-tion. Orphans thus have an incentive to maintain a cordialrelation with their state’s governor when possible. We find that arelatively large proportion of PRD orphans obtained a position instate government (21%) while PAN and PRI orphans got statepositions at lower rates (13% and 4%, respectively).

14 guillermo rosas and joy langston

find ample evidence that governors with longer timehorizons presided over more cohesive state-partydelegations than governors at the end of their terms.They do this even as national party leaders are able tokeep legislators voting together more often than not.This finding suggests that governors are capable ofaffecting national legislation through the votes of theirstates’ delegates. It also shows that measuring stateeffects can be fruitfully undertaken in more than oneway: to capture the ability of governors to minimizedispersion in their state-party delegations, regardless ofwhether this affects national voting cohesion, is animportant step forward (see also Cantu and Desposato2009; Cheibub, Figueiredo, and Limongi 2009; Jonesand Hwang 2005).

The characteristics of the Mexican case that allowus to identify with high precision the extent ofgubernatorial influence on legislators’ voting behaviorare not easily found elsewhere, but the basic structureof interaction among governors, deputies, and nationalparty leaders that we see in Mexico is of course adefining feature of many federal regimes. Therefore, theMexican case is important because the particularconstitutional combination of no consecutive reelec-tion and staggered electoral calendar for governorsallows us to precisely measure what most likely exists inother federal nations, but is difficult to observe. We stilldo not know the conditions under which governorswill choose to reach out to sway the voting behavior oftheir states’ representatives in Congress. For starters, wedo not believe that it is always in the interest ofgovernors to break ranks with their parties’ legislative

leaders. Governors, in Mexico and elsewhere, mostlycare about claiming success for redistributing resourcesfrom the federal to the state level via the annual budgetnegotiations, which means that they may focus selec-tively on wielding state votes in budget matters, taxbills, or distributive concerns and submit to theirparty’s caucus leadership on all other matters. Wehasten to add that even strong governors with policypreferences inconsistent with those of their copartisansmay decide against swaying the votes of their statedelegation if this means breaking with the party line.Governors, like other politicians, derive benefits frombeing associated with a recognizable party label and sowill break ranks as infrequently as possible. Governorsmay thus choose to selectively delegate to the caucusleader the power to whip state delegations, which is inessence the dynamic of ‘‘conditional party govern-ment’’ that Jones and Hwang (2005) have recognizedin Argentina’s congress and that is likely to exist inother federal regimes. Without a doubt, future researchwould benefit from theoretically guided argumentsabout the conditions under which we could expectgovernors to work together with party leaders inpromoting the party’s brand name. Knowing with highcertainty that gubernatorial effects exist should provideresearchers with the impetus to carry out such work.

Acknowledgments

We gratefully acknowledge support from the MurrayWeidenbaum Center on the Economy, Government,

FIGURE 3 Color-coded ellipses represent the ideal point dispersion of an average-sized delegation fromthree different parties under a lame-duck governor with one year left, a mid-career governorwhose term will barely outlast the delegation’s term (3.5 years left), and a new governor withsix years left. The dotted ellipse represents the average dispersion score across all delegations(numbers correspond to the ratio of color-coded ellipse area to dotted ellipse area).

xxx 15

and Public Policy at Washington University andresearch assistance from Marcela Ballina, MarianaMedina, Noel P. Johnston, and Bernardo Perez. Ourgratitude extends to Brian Crisp, Matt Gabel, Jeff Gill,Nate Jensen, Andrew Martin, Detlef Nohlen, SusanStokes, Margit Tavits, and the journal’s anonymousreviewers for comments and suggestions, and to ScottDesposato and Jeff Weldon for invaluable input.Previous versions of this paper were presented atthe 2009 APSA annual meeting and at conferencesin Washington University and the Universidad deSalamanca. We alone bear responsibility for anyerrors. An online appendix with data and supportingmaterials necessary to reproduce the numerical re-sults for this article can be found at grosas.wustl.eduand at http://journals.cambridge.org/Jop.

References

Ames, Barry. 2001. The Deadlock of Democracy in Brazil. AnnArbor: University of Michigan Press.

Armijo, Leslie Elliott, Philippe Faucher, and Magdalena Dem-binska. 2006. ‘‘Compared to What? Assessing Brazil’s PoliticalInstitutions.’’ Comparative Political Studies 39 (6):759–86.

Benton, Allyson L. 2009. ‘‘What Makes Strong Federalism SeemWeak: Fiscal Resources and Presidential-Provincial Relationsin Argentina.’’ Publius 39 (4):651–76.

Betanzo, Alejandra. 2009. The Partisan Determinants of MexicanState Representative’s Preferences towards Fiscal Decentral-ization: The Case of Mexico. PhD thesis, Universitat PompeuFabra.

Bruhn, Kathleen. 1997. Taking on Goliath: The Emergence of aNew Left Party and the Struggle for Democracy in Mexico.University Park: Pennsylvania State University Press.

Cantu, Francisco, and Scott W. Desposato. 2009. ‘‘The NewFederalism of Mexico’s Party System.’’ Unpublished manu-script. University of California, San Diego.

Carey, John M. 2007. ‘‘Competing Principals, Political Institu-tions, and Party Unity in Legislative Voting.’’ AmericanJournal of Political Science 51 (1): 92–107.

Carey, John M. 2009. Legislative Voting and Accountability. NewYork: Cambridge University Press.

Carey, John M., and Gina Y. Reinhardt. 2004. ‘‘State-levelInstitutional Effects on Legislative Coalition Unity in Brazil.’’Legislative Studies Quarterly 29 (1): 23–47.

Cheibub, Jose Antonio, Argelina Figueiredo, and FernandoLimongi. 2009. ‘‘Political Parties and Governors as Determi-nants of Legislative Behavior in Brazil’s Chamber of Deputies,1988-2006.’’ Latin American Politics and Society 51 (1): 1–30.

Clinton, Joshua, Simon Jackman, and Douglas Rivers. 2004. ‘‘TheStatistical Analysis of Roll Call Data.’’ American PoliticalScience Review 98 (2): 355–70.

Desposato, Scott W. 2003. ‘‘Comparing Group and SubgroupCohesion Scores: A Nonparametric Method with an Applica-tion to Brazil.’’ Political Analysis 11: 275–88.

Desposato, Scott W. 2004. ‘‘The Impact of Federalism onNational Party Cohesion in Brazil.’’ Legislative Studies Quar-terly 29 (2): 259–85.

Dıaz-Cayeros, Alberto. 2006. Federalism, Fiscal Authority, andCentralization in Latin America. New York: Cambridge Uni-versity Press.

Domınguez, Jorge I., and James A. McCann. 1996. DemocratizingMexico. Public Opinion and Electoral Choices. Baltimore: JohnsHopkins University Press.

Figueras, Zanabria, Vıctor Manuel. 2009. ‘‘Apertura polıtica ydescentralizacion fiscal en Mexico: El papel de los goberna-dores.’’ Nuevo Mundo Mundos Nuevos.

Flamand, Laura. 2006. ‘‘El juego de la distribucion de recursos enun sistema federal.’’ Polıtica y Gobierno 13 (2): 315–59.

Garman, Christopher, Stephan Haggard, and Eliza Willis. 2001.‘‘Fiscal Decentralization. A Political Theory with Latin Amer-ican Cases.’’ World Politics 53 (2): 205–36.

Greene, Kenneth F. 2008. ‘‘Dominant Party Strategy and Democ-ratization.’’ American Journal of Political Science 52 (1): 16–31.

Hagopian, Frances, Carlos Gervasoni, and Juan Andres Moraes.2009. ‘‘From Patronage to Program: The Emergence of Party-Oriented Legislators in Brazil.’’ Comparative Political Studies42 (3): 360–91.

Hernandez, Rodrıguez, Rogelio. 2003. ‘‘The Renovation of OldInstitutions: State Governors and the Political Transition inMexico.’’ Latin American Politics and Society 45 (4): 97–127.

Hix, Simon. 2002. ‘‘Parliamentary Behavior with Two Principals:Preferences, Parties, and Voting in the European Parliament.’’American Journal of Political Science 46 (3): 688–98.

Holland, Paul. 1986. ‘‘Statistics and Causal Inference.’’ Journal ofthe American Statistical Association 81 (396): 945–60.

Jones, Mark P., and Wonjae Hwang. 2005. ‘‘Party Government inPresidential Democracies: Extending Cartel Theory Beyondthe U.S. Congress.’’ American Journal of Political Science 49(2): 267–82.

Langston, Joy, and Javier Aparicio. 2008. ‘‘The Past as Future:Prior Political Experience and Career Choices in Mexico.’’Mexico: CIDE working paper.

Langston, Joy. 2010. ‘‘Governors and ‘‘Their’’ Deputies: NewLegislative Principals in Mexico.’’ Legislative Studies Quarterly35 (2): 235–58.

Martin, Andrew D., and Kevin M. Quinn. 2002. ‘‘Dynamic IdealPoint Estimation via Markov Chain Monte Carlo for the U.S.Supreme Court, 1953-1999.’’ Political Analysis 10 (2): 134–53.

Mizrahi, Yemile. 2003. From Martyrdom to Power: The PartidoAccion Nacional in Mexico. South Bend, IN: University ofNotre Dame Press.

Moreno, Alejandro. 1998. Party Competition and the Issue ofDemocracy: Ideological Space in Mexican Elections. In Gov-erning Mexico: Political Parties and Elections, ed. MonicaSerrano. Oxford: Institute of Latin American Studies, 38–57.

Nacif, Benito. 2002. Understanding Party Discipline in theMexican Chamber of Deputies: The Centralized Party Model.In Legislative Politics in Latin America, ed. Scott Morgensternand Benito Nacif. Cambridge: Cambridge University Press,254–84.

Pardinas, Juan. 2008. Decentralisation and Budget Accountabilityin the Twilight of Mexican Presidentialism. PhD thesis,London School of Economics.

Pereira, Carlos, and Bernardo Mueller. 2004. ‘‘The Cost ofGoverning: Strategic Behavior of the President and Legislatorsin Brazil’s Budgetary Process.’’ Comparative Political Studies37 (7): 781–815.

Poire, Alejandro. 2005. Follow the Money: Local Public Fundingand Internal Party Transfers in 2003. In What Kind ofDemocracy Has Mexico? The Evolution of Presidentialism and

16 guillermo rosas and joy langston

Federalism, ed. Matthew Shugart and Jeffrey Weldon. La Jolla:Center for US-Mexico Studies.

Poole, Keith T., and Howard Rosenthal. 1997. Congress: APolitical-Economic History of Roll-Call Voting. Oxford Uni-versity Press.

Rıos, Alejandra. 2009. ‘‘Accountability in Developing Countries:Oversight and Control of the Executive in Mexican States.’’Paper prepared for the 2009 annual meeting of the AmericanPolitical Science Association.

Rivera, Sanchez, and Jose Abel. 2004. ‘‘Cambio institucional ydemocratizacion: La evolucion de las comisiones en la Camarade Diputados de Mexico.’’ Polıtica y Gobierno 11 (2): 263–313.

Robles, Gustavo. 2009. Dimensiones espaciales de votacionlegislativa en la Camara de Diputados de la 60 Legislaturadel Congreso de la Union, 2006–2009. Master’s thesis,Instituto Tecnologico Autonomo de Mexico.

Rodden, Jonathan, and Erik Wibbels. 2005. ‘‘RetrospectiveVoting, Coattails, and Accountability in Regional Elections.’’Paper prepared for the 2005 annual meeting of the AmericanPolitical Science Association.

Rubin, Donald B. 1974. ‘‘Estimating Causal Effects of Treatmentsin Randomized and Nonrandomized Studies.’’ Journal ofEducational Psychology 66 (5): 688–701.

Samuels, David. 2000. ‘‘Ambition and Competition: ExplainingLegislative Turnover in Brazil.’’ Legislative Studies Quarterly25 (3): 481–97.

Samuels, David J. 2000. ‘‘The Gubernatorial Coattails Effect:Federalism and Congressional Elections in Brazil.’’ Journal ofPolitics 62 (1): 240–53.

Trejo, Delarbre, Bertha. 2003. Diputados y senadores, quien esquien en el Congreso: LVIII Legislatura. Mexico: Instituto deEstudios para la Transicion Democratica.

Weldon, Jeffrey A. 2002. ‘‘Factores institucionales y polıticos de ladisciplina partidaria en la Camara de Diputados de Mexico,1998-2002.’’ Mexico: ITAM working paper.

Weldon, Jeffrey A. 2005. ‘‘Institutional and Political Factors forParty Discipline in the Mexican Congress since the End of PRIHegemony.’’ Mexico: ITAM working paper.

Wuhs, Steven T. 2006. ‘‘Democratization and the Dynamics ofCandidate Selection Rule Change in Mexico, 1991-2003.’’Mexican Studies/Estudios Mexicanos 22: 33–55.

Guillermo Rosas is Assistant Professor of PoliticalScience at Washington University in St. Louis, St.Louis, MO 63130.

Joy Langston is Professor of Political Science atthe Division de Estudios Polıticos of the Centrode Investigacion y Docencia Economicas (CIDE),Mexico, DF 01210.

xxx 17