Geographically-Targeted Spending inMixed-Member Majoritarian Electoral Systems

Amy Catalinac∗

Lucia Motolinia †

Abstract

Can governments elected under mixed-member majoritarian (MMM) electoral systems usegeographically-targeted spending to increase their chances of staying in office and if so, how? De-spite being used in thirty countries, little attention has been paid to this question. We posit thatunder MMM, majority-seeking parties will have compelling reasons to engage in ‘preelectoralcoordination’ with small parties, whereby supporters are instructed to cast their district votesfor one party and their proportional votes for the other, and use the geographically-targetableresources under their control to encourage supporters to comply with their instructions. Wetest these propositions using original data on municipality-level electoral returns and centralgovernment transfers from Japan (2003-2013) and Mexico (2012-2016). In both cases, munici-palities where supporters complied received more money after elections. Our findings have broadimplications for research on MMM, distributive politics, and the politics of Japan and Mexico,respectively.1

∗Assistant Professor, New York University. Email: amy.catalinac@nyu.edu.†Ph.D. Candidate, New York University. Email: lucia.motolinia@nyu.edu.1We thank Makoto Fukumoto, William Godell, Sean Kates, Kyuwon Lee, Andrew Little, John Marshall, Gwyneth

McClendon, Pia Raffler, Massimo Pulejo, Alison Post, Steven Reed, Arturas Rozenas, Tara Slough, Daniel M. Smith,David Stasavage, Steve Vogel and participants at the NYU Comparative Politics Speaker Series, the University ofCalifornia Berkeley’s Comparative Politics Speaker Series, MPSA 2019, and the Harvard Symposium on JapanesePolitics, August 29, 2018 for comments on earlier drafts.

1

Thirty countries around the world today use an electoral system called ‘mixed-member ma-

joritarian’ (MMM) to choose members of their national legislative bodies. At first glance, MMM

resembles its more well known cousin, ‘mixed-member proportional’ (MMP), in that both select

some legislators in a ‘district tier’ (usually comprising single-seat districts or ‘SSDs’) and some in

a ‘proportional tier’ (usually comprising closed-list proportional representation or ‘PR’). There

is a critical distinction between the two, however. Under MMM, the number of seats a party

wins is the sum of those it wins in both tiers, whereas under MMP it is determined solely by

those won in the proportional tier. This means that under MMM, majority-seeking parties have

to be able to win seats in both tiers, as opposed to being able to concentrate on the proportional

tier under MMP. They must be capable, in other words, of placing first in as many districts

as possible (Herron, Nemoto and Nishikawa, 2018; Riera, 2013; Shugart and Wattenberg, 2003;

Reilly, 2007; Bawn and Thies, 2003).

We advance our understanding of MMM in two main ways. First, we extend existing work

on why majority-seeking parties have reasons to pursue ‘preelectoral coordination’ with small

parties under dual-ballot MMM to offer a general theory of why they do so and how such co-

ordination can be achieved under both single- and dual-ballot MMM. Our contribution here is

to explain that, even when voters have a single vote, parties interested in coordinating can find

ways to parse this vote into two components: one of which can be cast for one coordinating

partner and the other of which can be cast for the other. Like all forms of preelectoral coordi-

nation, this is done to increase the number of seats won by both parties. Second, we posit that

once a majority-seeking party dependent upon preelectoral coordination reaches government,

it can use geographically-targeted spending to encourage supporters to comply with its chosen

coordination strategy, where ‘comply’ means split one’s vote in accordance with instructions.

Our contribution here is to show that MMM, like other electoral systems, creates a strategic en-

vironment in which governing parties can use geographically-targeted spending to significantly

enhance their prospects of winning the next election.

To evaluate these propositions, we turn to Japan’s House of Representatives (HOR), which

has used dual-ballot MMM since 1996, and Mexico’s Chamber of Deputies (COD), which has

used its current version of single-ballot MMM since 1988. Together, these countries make up 20%

of voters worldwide who elect representatives under MMM. Both chambers select approximately

60% of their members in SSDs and 40% in regional constituencies according to PR. During

2

our period of study, both chambers were dominated by a coalition comprising a majority-

seeking party and a small party. Both coalitions used preelectoral coordination to capture

their majorities, had access to discretionary transfers awarded annually to municipalities for

the purpose of funding projects, and could discern vote shares at the level of the municipality.

Fixed-effect regressions of original data reveal that in both cases, municipalities complying with

the dominant coalition’s preelectoral coordination strategy received more money after elections.

Concretely, our results show that Japan’s Liberal Democratic Party (LDP)-Komeito gov-

erning coalition used discretionary transfers to municipalities to reward LDP supporters for

switching their PR votes to the Komeito in SSDs where LDP candidates ran and Komeito can-

didates stood down (the vast majority) and Komeito supporters for switching their PR votes

to the LDP in SSDs where Komeito candidates ran and LDP candidates stood down (a small

fraction) in the 2003, 2005, and 2012 HOR elections. In Mexico, on the other hand, the single

‘fused’ ballot afforded voters in COD elections adds one vote to the chosen candidate’s tally in

the SSD race and one vote to the candidate’s party in the PR race. The Institutional Revolu-

tionary Party (PRI)-Ecological Green Party of Mexico (PVEM) coalition found a way to convert

this vote into a vote in both tiers, at least for party supporters. Instructing supporters to vote

for the SSD candidate, an affiliate of one coordinating partner, under the other partner’s label

ensured that each vote counted as an SSD vote for one partner and a PR vote for the other

partner. Our results show that the PRI-PVEM coalition used COD-controlled discretionary

transfers to municipalities to reward PVEM supporters for switching their PR votes to the PRI

in SSDs where the coalition ran a PVEM candidate and the PRI stood down (about a quarter

of SSDs) in the 2012 and 2015 COD elections.

Our findings shed new light on the inner workings of the coalitions that dominated Mexican

and Japanese politics during the period of study. Beyond the fact that the electoral strategy

we identify herein likely contributed to their dominance, it may also help explain why the

coalition’s policies do not always reflect the preferences of the smaller coalition partner. In

Japan, for example, the LDP-Komeito coalition has enacted changes to Japan’s security policy

that have left pundits scratching their heads as to why the Komeito, known as a pacifist party,

acquiesced. While future research is needed, it is possible that geographically-targeted spending

also functions as a tool to buy the support of the smaller partner for policies it would ordinarily

find unpalatable. In support of this, we find that Komeito supporters who switched their

3

PR votes to the LDP received considerably more spending on their communities than LDP

supporters who switched their PR votes to the Komeito. This also helps make sense of the fact

that the Land, Infrastructure, Transport, and Tourism ministerial portfolio has been assigned

to the Komeito for the vast majority of years it has been in government (2004-8; 2012-present).

Our study also contributes to research on mixed-member systems and distributive politics,

respectively. By our calculations, 33% of voters worldwide elect representatives under mixed

electoral rules today (Bormann and Golder, 2013). Reflecting this, there is now a vast literature

on their effects (e.g. Herron, Nemoto and Nishikawa, 2018; Batto et al., 2016; Rich, 2015; Moser

and Scheiner, 2012; Krauss, Nemoto and Pekkanen, 2012; Naoi and Krauss, 2009; Reilly, 2007;

Bawn and Thies, 2003; Shugart and Wattenberg, 2003). Very few of these studies, however, fall

under the rubric of distributive politics, and those that do examine their impact on government

spending more broadly (Thames and Edwards, 2006), not on geographically-targeted spending.

This is surprising given that work on electoral systems per se is a core focus of the distributive

politics research agenda (Golden and Min, 2013). Numerous studies have elucidated the ways in

which governing parties elected under different electoral systems can use geographically-targeted

spending to maximize their chances of staying in power (e.g. Funk and Gathmann, 2013; Tavits,

2009; Dahlberg and Johansson, 2002; Ames, 1995). Mixed systems are not absent from this

work, but tend to be the focus of study only insofar as they provide a laboratory where scholars

interested in testing their propositions about a single system can do so in a controlled setting

(Kerevel, 2010; Pekkanen, Nyblade and Krauss, 2006; Stratmann and Baur, 2002; Rickard,

2012). Our study is the first, to our knowledge, to posit that the distinct strategic environment

created by the combination of systems influences geographically-targeted spending.

In drawing this conclusion, we contribute to two debates central to research on mixed sys-

tems. The first concerns the extent to which they produce outcomes that can be reduced to the

sum of their constituent systems versus outcomes that reflect the distinct incentives created by

their combination (e.g. Herron, Nemoto and Nishikawa, 2018; Pekkanen, Nyblade and Krauss,

2006; Cox and Schoppa, 2002). Our findings support the latter. It is MMM’s combination of

SSDs and PR, as well as its requirement that majority-seeking parties win seats in both tiers,

that gives majority-seeking parties the ability to coordinate SSD candidacies and PR votes with

smaller parties in ways that promise more seats for both. If the system was comprised solely

of SSDs, well-organized small party supporters would be less likely to exist. If the system was

4

PR-only, on the other hand, small parties would be less willing to step back from electoral com-

petition. Because rewarding supporters for complying with one’s coordination strategy means

steering spending toward people casting at least one of their votes for another party, we would

be similarly unlikely to observe this pattern of spending under a system comprised solely of

SSDs or PR. Future research should consider how the distinct strategic environment created by

MMM’s combination of tiers influences other aspects of government policy.

A second debate concerns the extent to which mixed systems give voters the ‘best of both

worlds’, usually construed to mean representatives attentive to local concerns and parties ca-

pable of aggregating the concerns of broad swaths of voters (e.g. Kerevel, 2010; Shugart and

Wattenberg, 2003; Bawn and Thies, 2003; McKean and Scheiner, 2000). At the heart of this

debate is a question about representation. While the incentives for majority-seeking parties to

embrace preelectoral coordination under dual-ballot MMM have been recognized by others, we

are the first to posit that once in government, they may be able to leverage government resources

to cement it. If government resources are being used to cement coordination, opposition par-

ties will be at a serious disadvantage. Future research should examine whether the difficulties

opposition parties tend to experience establishing preelectoral coordination can be explained,

at least partially, by their lack of resources to cement it. Future research should also examine

whether MMM makes it harder than previously acknowledged for opposition parties to unseat

coalition governments. If so, then this suggests MMM is closer to the ‘worst of both worlds’

(Bawn and Thies, 2003; McKean and Scheiner, 2000).

1 Theory

It is well-established that parties competing under disproportional electoral rules have incentives

to engage in ‘preelectoral coordination’, which entails agreements not to compete against each

other (Ferrara and Herron, 2005; Golder, 2006). To understand why, consider the archetypal

example of a disproportional system: one comprised solely of SSDs. Because votes cast for losing

candidates in a system comprised solely of SSDs are wasted, the number of votes cast for a party

rarely matches the number of seats it wins. Vote-seat distortions are particularly pronounced

for small parties. Small parties may have considerable support across the nation as a whole, but

if none of their SSD candidates are capable of placing first, they will emerge from an election

with no seats at all, despite having captured an enviable number of votes. From the perspective

5

of majority-seeking parties, small parties represent a resource that can be harnessed to improve

their chance of winning a majority. If a majority-seeking party can convince a small party not

to run in an SSD where its candidate’s prospects of victory are uncertain, it can dramatically

improve its candidate’s chance of winning, especially if the small party’s supporters can be

persuaded to support the majority-seeking party’s candidate. Forming such arrangements in

every SSD where its candidates’ victories are not foregone conclusions can dramatically improve

its chances of capturing a majority.

Because a system comprised solely of SSDs is so disproportional, however, it will be imme-

diately obvious that small parties are much less likely to exist, rendering it difficult (oftentimes

impossible) for majority-seeking parties to develop an electoral strategy centered around using

them as a crutch to win more SSDs. Nevertheless, contrasting a system comprised solely of

SSDs with MMM helps highlight the fact that under MMM, small parties have reason to exist

and majority-seeking parties have reason to want to use them as a crutch. Under MMM, the

number of seats a party wins is the sum of those it wins in a district (usually SSD) and pro-

portional (usually PR) tier. Small parties that cannot place first in SSDs can concentrate on

winning seats in the PR tier. Majority-seeking parties, however, are not afforded this luxury. If

the mixed system was MMP, where the total number of seats a party wins is determined by the

number it wins in the PR tier, it could concentrate on winning seats in that tier. Because its

seat share is determined by the seats won in both tiers under MMM, it must be able to place

first in as many SSDs as possible.

Early work on coordination in mixed-member systems focused on ‘coordinated entry’ into

SSDs, defined as more than one party agreeing to field a single SSD candidate (Ferrara and

Herron, 2005). The study found that coordinated entry was more common in mixed systems

where the SSD tier dominates, as it does under MMM; district magnitude in the proportional

tier is low; and voters are afforded two ballots. In systems with the first two characteristics,

small parties are punished more severely in the translation of votes to seats, which increases their

willingness to coordinate. When voters cast a single ‘fused’ ballot, on the other hand, which

counts as a vote for the candidate in the SSD and the candidate’s party in PR, parties seeking

PR seats must run candidates in SSDs, which reduces small parties’ willingness to coordinate.

Subsequent research continued the focus on how coordination could be achieved under dual-

ballot MMM (e.g. Liff and Maeda, 2019; Klein, 2013; Smith, 2014; Reed and Shimizu, 2009; Cox

6

and Schoppa, 2002). A central insight is that even if a majority-seeking party could convince

a small party to stand down in an SSD, the fact that voters have a second vote, for seats in

a second tier, risks impeding any coordination in the first tier (Nemoto and Tsai, 2016). For

this reason, we posit that it makes more sense for coordinating parties to incorporate both tiers

into their coordination. How might they do so? Consider two parties, one majority-seeking

and the other small. The majority-seeking party could ask the small party to stand down in

SSDs where the majority-seeking party’s candidates’ prospects are uncertain and implore its

supporters there to cast their SSD votes for this candidate. In return, the majority-seeking

party can implore its supporters in those SSDs to cast their PR votes for the small party. In

return for standing down in the SSD, the small party recoups PR votes.

Because parties that stand down in an SSD forgo the chance to win PR votes under single-

ballot MMM, it has always been assumed that coordination is less likely to occur here. We

posit that majority-seeking parties can make stand-down agreements attractive to small parties

under single-ballot MMM by incorporating the proportional tier into the exchange. How can

this be achieved? Consider the same two parties. In SSDs where the majority-seeking party’s

prospects are less certain, it could offer to field a candidate jointly supported by itself and the

small party, while also presenting voters with separate lists in the proportional tier. Concretely,

this would mean that voters in those SSDs are presented with a single ballot upon which the

joint candidate’s name appears twice, once next to one coordinating partner and once next to

the other. The majority-seeking party could then cede SSD candidacies to the small party in

exchange for having both parties’ supporters choose the jointly-supported candidate under the

majority-seeking party’s label. In exchange for standing down in these SSDs, the majority-

seeking party ensures that all votes cast for the small party’s candidate translate into PR votes

for itself. If the SSD is won, the small party wins a seat, but regardless, the majority-seeking

party keeps all the PR votes.

Ultimately, whether parties pursue coordination depends on demand and supply. If a

majority-seeking party can comfortably field enough SSD candidates capable of placing first,

it would be unlikely to go to the trouble of coordinating. Conditional upon preferring to co-

ordinate, the availability of a small party with supporters in exactly those SSDs where the

majortiy-seeking party’s candidates struggle is another factor influencing its ability to do so. If

a small party fitting the bill exists but is already coordinating with another majority-seeking

7

party, for example, a second majority-seeking party may be unsuccessful at forging coordina-

tion. Thus, our conjecture is not that every majority-seeking party in every MMM system will

coordinate; only that coordination will be an attractive tool to enlarge one’s seat share, when

demand and supply are met. We also emphasize that there is no one-size-fits-all coordination.

The exact nature of the coordination forged between two parties will depend upon their geo-

graphic distributions of electoral support. Whether a majority-seeking party finds it fruitful to

cede any SSD candidacies to its smaller partner, for example, will depend upon their candidates’

relative strengths.

While preelectoral coordination can pay off in terms of seats, it places a burden on those

who ordinarily support one of the coordinating parties. In SSDs where parties are forging such

exchanges, supporters have to be told that, to maximize their preferred party’s collective inter-

ests, they have to cast one of their ballots for a party that is not their first choice. Coordination

also carries risks for the coordinating parties: both fear being exploited by the other, where

‘exploited’ means that one partner instructs its supporters to switch one of their ballots to the

other, but the latter fails to reciprocate (Nemoto and Tsai, 2016). Not only would this yield a

sub-optimal outcome in terms of seat share, but it would also lead to a credibility loss in the

eyes of one’s supporters.

This is where geographically-targetable spending can be useful. It has been shown over and

over again that governing parties use targetable resources to increase their chances of staying

in power (e.g. Tavits, 2009; Golden and Picci, 2008; Stokes, 2005; Dahlberg and Johansson,

2002). We posit that once in government, parties dependent upon coordination can try to

compel supporters to cast their votes strategically by making the distribution of resources to

their communities contingent upon it. While the secret ballot makes it near-impossible to verify

whether individuals are doing so, in many democracies votes are counted at a geographically-

defined unit within each district, such as a municipality, precinct, or polling station. When this

is the case, coordinating parties will be able to use changes in vote shares to discern whether

supporters in a given unit are complying. Whether a majority-seeking party’s supporters have

followed instructions to cast their PR votes for the small party, for example, can be verified by

examining whether PR votes for the majority-seeking party decreased and those for the small

party increased in a given unit, relative to the previous election. Once verified, coordinating

parties with access to geographically-targetable resources will be tempted to use those resources

8

to reward units where supporters complied and penalize units where they did not. By making

it clear that resources will be withheld from units where compliance was not forthcoming,

coordinating parties can kill two birds with one stone: they can encourage their supporters to

comply and minimize the risk of being exploited by their partner.

Implicit in the notion that coordinating parties will rely on over-time changes in votes to

discern the level of compliance in a given geographic unit is the assumption that perfect com-

pliance will not yet have been reached. If it had been, there would be no more vote-switching

to observe. We acknowledge this, but suggest that asking one’s supporters to cast their second

ballot for another party, against whom it likely competed in previous elections, is likely to be

met with resistance, at least initially, before the regime of rewards and penalties has had time

to sink in. Qualitative accounts of how incumbents go about asking their supporters to switch

their second vote corroborate this (Reed and Shimizu, 2009).

2 Dual-Ballot MMM in Japan

To evaluate these propositions, we turn first to Japan. Japan is a bicameral parliamentary

system, of which the House of Representatives (HOR) is the more-powerful House. HOR mem-

bers serve four-year terms, but Prime Ministers can dissolve the HOR at any time. The HOR

has used dual-ballot MMM to elect its members since an electoral reform in 1994.2 Initially,

300 members were chosen in SSDs and 200 members from party lists in eleven regional blocs

according to PR. Over time, the number of members elected in both tiers has declined. As of

2017, 289 are chosen in SSDs and 176 via PR, for a total of 465. The number of seats in each

PR bloc has been adjusted over time for population changes and currently ranges from 6 to 28.

Votes cast in each PR bloc are used to apportion seats in that bloc, which is done via d’Hondt.

In HOR elections, voters receive two ballots. On one, they write the name of their preferred

SSD candidate. On the other, they write the name or abbreviation of their preferred party.

Formed in 1955, Japan’s Liberal Democratic Party (LDP) captured majorities in every HOR

election between 1958 and 1990, giving it uninterrupted control of government. In the 1993 elec-

tion, it captured a plurality but lost control of government when other parties won enough seats

to form a coalition government. In early 1994, this government introduced MMM. Midway

2Prior to 1994, it used SNTV-MMD (‘single non-transferable vote in multi-member districts’).

9

through 1994, the LDP reentered government via a coalition of its own. Toward the end of

1999, the LDP convinced a small party, the Komeito, to join its coalition. The Komeito’s reli-

gious underpinnings meant that the parties were unlikely bedfellows. Just three years earlier,

LDP candidates had made “The Enemy is Komei” the ‘centerpiece’ of their campaign (Reed and

Shimizu, 2009, 38). For years, LDP politicians had even argued that the Komeito’s existence

threatened the constitutional principle that religious organizations could not exercise political

authority (Liff and Maeda, 2019, 59). Helpfully, though, the Komeito had a history of preelec-

toral coordination (Christensen, 2000, chapter 4). The LDP-Komeito coalition proved durable:

the two parties have governed together ever since, with the exception of 2009-12.3

Preelectoral coordination between the LDP and Komeito began in the first election the

partners faced, in 2000. Initially, leaders concentrated on how to divvy up SSDs. The LDP

reportedly floated the idea of standing down in as many as 25 SSDs in favor of Komeito candi-

dates (Reed and Shimizu, 2009). It ended up running in 271 SSDs, while the Komeito ran in 18.

In 4 of the 18, both parties ran candidates. By the 2003 election, they had stopped competing

against each other in any SSD. In this and all subsequent HOR elections, LDP candidates have

run in approximately 90% of SSDs (and Komeito candidates have stood down) and Komeito

candidates have run in approximately 3% of SSDs (and LDP candidates have stood down).4

Moreover, standing down has been accompanied, in most SSDs, by explicit ‘recommendations’

that one’s supporters cast their SSD votes for the other party’s candidate. While in 2000, the

LDP recommended 14 of the Komeito’s 18 SSD candidates, since 2003 it has recommended all of

them. In 2000, the Komeito recommended 58% of the LDP’s SSD candidates. This percentage

has risen since then, reaching 96% in 2017 (Liff and Maeda, 2019, 61).

Votes in the PR tier appear to have become incorporated into the exchange via LDP candi-

dates seeking to guarantee they would receive Komeito votes in their SSD by promising that, in

exchange for those votes, they would instruct their supporters to vote Komeito in PR. In one

case, LDP-affiliated Mihara Asahiko established a campaign organization to help persuade his

supporters to vote Komeito in PR and even shared his supporters’ names and addresses with the

local Komeito organization, which visited their homes to solicit PR votes (Reed and Shimizu,

2009). While we lack systematic data on how many LDP candidates instruct their supporters

3The LDP-Komeito lost the 2009 election, but regained control of government with a landslide win in 2012.4Calculated from the number of SSDs where LDP candidates ran and Komeito candidates did not, and Komeito

candidates ran and LDP candidates did not, respectively, in Reed and Smith (2015).

10

to do this, the share of PR votes cast for the Komeito increased between 2000 and 2005, which

Reed and Shimizu (2009) attribute to LDP supporters following instructions. As a result, it is

generally believed that in ‘LDP SSDs’, the Komeito instructs its supporters to cast their SSD

votes for the LDP’s candidate (and their PR votes for the Komeito) in exchange for the LDP

candidate instructing its supporters to cast their PR votes for the Komeito (and their SSD

votes for the LDP’s candidate) (Klein, 2013; Smith, 2014). While less is known about ‘Komeito

SSDs’, it would make sense for the reverse to be true: LDP supporters are likely instructed to

cast their SSD votes for the Komeito’s candidate (and their PR votes for the LDP) in exchange

for the Komeito candidate instructing its supporters to cast their PR votes for the LDP (and

their SSD votes for their party’s candidate).

Our theory leads us to expect that the LDP-Komeito coalition will try to elicit compliance

by making the distribution of valued geographically-targeted funding conditional upon it. In

Japanese elections, votes are counted at the level of the municipality and almost all municipalities

are contained within a single SSD.5 Being able to observe how many PR votes were cast for

each coordinating partner in each municipality gives the coalition the tools to verify whether

their supporters complied. Concretely, in LDP SSDs, compliance can be verified by examining

whether PR votes for the LDP declined and those for the Komeito increased in municipality m

relative to the previous election. In Komeito SSDs, compliance can be verified by examining

whether PR votes for the Komeito declined and those for the LDP increased.

Our claim that HOR politicians make the distribution of government resources to munic-

ipalities contingent upon their voting behavior has support in the Japanese politics literature

(Catalinac, Bueno de Mesquita and Smith, 2019; Saito, 2010). These and other studies credit

a potent mix of conditions facilitating this: municipalities’ dependence on the central govern-

ment for resources; the large pool of ‘national treasury disbursements’ (NTD) made available

by the central government each year, which is allocated at the discretion of bureaucrats for the

purpose of funding projects in municipalities; and the ability of LDP incumbents to discern

how municipalities vote and lean on bureaucrats to allocate NTD in ways that benefit certain

municipalities over others (e.g. Saito, 2010; Scheiner, 2006; Hirano, 2006; Horiuchi and Saito,

2003; Reed, 1986). There is no question that NTD is a valued resource: approximately 16%

5Our empirical analysis focuses on the 2003, 2005, and 2012 HOR elections. In these, 1%, 3.6%, and 9% ofmunicipalities, respectively, spanned more than one SSD. The number changed because municipal mergers reducedthe number of municipalities.

11

of the average municipality’s revenue in 2015 came from this type of transfer (Yamada, 2016).

Concretely, we test the following:

• Hypothesis I: Municipalities in LDP SSDs (where LDP candidates run and Komeito

candidates stand down) are rewarded with more NTD after elections when they increase

PR votes for the Komeito and decrease them for the LDP.

• Hypothesis II: Municipalities in Komeito SSDs (where Komeito candidates run and LDP

candidates stand down) are rewarded with more NTD after elections when they increase

PR votes for the LDP and decrease them for the Komeito.

3 Single-Ballot MMM in Mexico

To illustrate our theory’s applicability to a different type of MMM system, we turn to Mexico.

Mexico is a presidential federal republic, with a bicameral legislature. Parties seek a majority of

seats in the 500-member Chamber of Deputies (COD) to shape the legislative process and lead

negotiations over the Federal Expenses Budget, which is proposed by the President but requires

approval by the Chamber. The Senate is smaller; comprising only 128 Members. Senators do

not have the same influence over the Federal Expenses Budget as Deputies. A mixed-member

system has been used to select members of the COD since 1964. The current version is MMM

and has been in place since 1988 (Weldon, 2003). In it, 300 Deputies are elected in SSDs and

200 via PR in five regional constituencies. Each constituency elects 40 Deputies. Deputies serve

three-year terms. Following an electoral reform in 2014, Deputies elected in and after 2018

are, for the first time in Mexico’s history, permitted to seek consecutive reelection (Motolinia,

2019). In elections, voters receive a single ballot, upon which appears a series of party-candidate

combinations. They mark the combination for whom they wish to vote. This translates into a

vote for the candidate in the SSD race and her party in the PR race.6

Formed in 1929, Mexico’s Institutional Revolutionary Party (PRI) controlled an absolute

majority in the COD until the 1997 election, when economic crisis and social unrest prompted

enough voters to cast their ballots for other parties (Magaloni, 2006). This proved the harbinger

for the election of the first non-PRI-affiliated President in more than seventy years in 2000,

6Parties must meet a minimum threshold to win PR seats; a cap against over-representation exists; and votes areconverted into PR seats by pooling them at the national level first, before allocating them to parties. These detailsare explained in Section A and B of the Online Appendix.

12

credited as evidence of Mexico’s transition to democracy. The PRI held onto a plurality in

the COD during this President’s administration, only to lose it in 2006, when a presidential

election held concurrently with legislative elections gave momentum to two opposition parties.

This relegated the PRI to third-largest party, a position it held until 2009, when it regained a

plurality. It controlled in excess of 40% of COD seats until 2018, when another concurrently-

held election saw a coalition of parties excluding the PRI regain the presidency and majorities

in the COD and Senate.

At least since Mexico’s transition to democracy, preelectoral coordination has been a staple

among parties contesting elections at all levels (Kellam, 2017). In 2003, the PRI convinced a

small party, the Ecological Green Party of Mexico (PVEM), to ally with it in the COD and

coordinate with it in elections. Founded in 1993, the PVEM allied with the National Action

Party (PAN) in the 2000 election, helping to facilitate the alternation of power. While its

ideological leanings are to the right of the PRI, its choice of coordinating partner appears to be

motivated mainly by the desire to survive (Spoon and Gomez, 2017). Since 2003, the PRI-PVEM

coalition has coordinated in all elections (Casar, 2012).

How does their coordination work? Under single-ballot MMM, coordinating partners are

permitted to field joint candidates in select SSDs. Since 2007, parties fielding joint candidates

in an SSD are also permitted to present voters with separate lists. This means that in ‘alliance

SSDs’, voters are presented with a ballot upon which the same candidate appears under the

names of all coordinating parties. Voters casting their ballots for the joint candidate have the

option of choosing which of the candidate-party combinations they prefer. They can choose all

or a subset of these combinations. When a voter chooses the joint candidate under more than

one coordinating party, one vote is recorded for the joint candidate in the SSD race and one vote

is divided up among the chosen coordinating parties for the purpose of allocating PR votes.7

We suggest that the PRI will propose fielding joint candidates with the PVEM in SSDs

where it anticipates a tough race. Then, it will offer candidacies in some of these alliance SSDs

to the PVEM in exchange for the PVEM instructing its supporters in those SSDs to vote for

the joint candidate under the PRI label. With this strategy, the PVEM benefits from having all

votes cast for both parties count as votes for its SSD candidate. The PRI, on the other hand,

7How votes cast for more than one coordinating party translate into PR votes for those parties is explained inSection C of the Online Appendix.

13

insures itself against the possibility that the PVEM-affiliated joint candidate loses. When all

votes for this candidate are cast under the PRI label, then regardless of whether or not she wins,

the PRI keeps these votes, which translate into PR seats.8

Studies of PRI-PVEM coordination have focused on coordinated entry into SSDs (Spoon

and Gomez, 2018, 2017; Montero, 2015). Evidence that PRI politicians incorporated PR votes

into coordination forged with other parties in previous periods comes from Weldon (2003, 4).

He cites interviewees in 1985, when the COD used a version of MMM with two ballots, who

told him that voters were instructed to cast one of their ballots for the PRI and the other for its

coordinating partner. Further evidence, albeit indirect, comes from the affiliations of the two

parties’ joint candidates. In the first two elections in which joint candidates were fielded (2003

and 2006), the PRI kept more than 92% of the joint candidacies. In 2007, an electoral reform

permitted coordinating parties to present separate lists.9 This is when we expect the PRI begins

to exchange SSD candidacies for PR votes. In every election since 2007, the number of joint

candidacies the PRI has kept for itself declined. Whereas it kept 90% in 2009, this dropped to

78% in 2012 and then 77% in 2015.10

We posit that when the PRI-PVEM coalition controlled enough seats to dominate the COD,

they tried to elicit compliance by making the distribution of geographically-targetable funds

conditional upon it. Votes in Mexican elections are counted and reported at the polling station

and easily aggregated to municipality. The great majority of municipalities are nested within

SSDs.11 Concretely, we posit that compliance can be verified by examining whether, in SSDs

with PVEM-affiliated joint candidates, votes cast for the joint candidate under the PRI label

increased and votes cast for the joint candidate under the PVEM or both parties’ labels decreased

in municipality m relative to a previous election.

For many years, government resources in the form of ‘patronage, pork and spoils’ were the

‘glue’ that bound voters to the PRI’s ‘party-led hegemonic regime’ (Magaloni, 2006). Almost

8Mexico’s ‘proportionality restriction’ and rules governing the allocation of publicly-provided campaign fundsgenerate additional incentives for majority-seeking parties to focus on PR votes. These are explained in the OnlineAppendix’s Section B.

9Until 2007, parties fielding joint candidates had to present joint lists. Because these lists were closed, the partieshad to decide ahead of time which position would be ‘filled’ by a candidate of which party. We anticipate thatcoordination would have involved exchanges of SSD candidacies for choice positions on the joint list.

10Even if we treat ‘watermelon’ candidates (those who are PVEM affiliates but former PRI politicians) as PRIcandidates, the number of SSD candidacies the PRI kept for itself was similar in 2003 and 2006 (96%) and beginsdeclining after the reform (95% in 2009, 91% in 2012, and 87% in 2016 (Spoon and Gomez, 2018).

11Our empirical analysis focuses on the 2012 and 2015 COD elections. In these, only 2.5% of municipalities spannedmore than one SSD.

14

every study to date has focused on how Presidents wield geographically-targetable resources

such as allocations to municipalities under the poverty relief program ‘PRONASOL’ (Collier,

1992; Molinar and Weldon, 1994; Hiskey, 1999), conditional cash transfers (De La O, 2015) and

other federal transfers (Costa-I-Font, Rodriguez-Oreggia and Lunapla, 2003; Campos, Garcia

and Ruiz, 2018; Rodriguez-Oreggia and Rodriguez-Pose, 2004). Hiskey (1999) provides evidence

that these resources are distributed after elections, once vote shares have been verified. The fact

that a President’s budget requires COD approval, however, creates room for Deputies to add

amendments furthering their interests. Kerevel (2015) finds that Deputies alter the President’s

budget to “a substantial degree” and documents a relationship between the number of SSD-

targeted amendments added by Deputies and the tier from which they are elected.

We posit that the PRI-PVEM coalition used the as-yet unstudied ‘Municipal and State

Infrastructure Strengthening Fund’ (‘FORTALECE’) to elicit compliance. Dubbed the “hand-

outs fund” (‘fondo para moches’) by the news media (e.g. Jimenez and Alcantara, 2016), it fell

under the ‘Economic and Salary Provisions’ section of the federal budget and consisted of an-

nual allocations to municipalities that were added as amendments by Deputies.12 Like NTD in

Japan, these allocations were awarded for specific projects, which municipalities have to apply

for (Garcia, 2017).13 In 2015, a total of 10 billion pesos (approximately $500 million USD) were

allocated to municipalities under this fund (Cervantes, 2017). These allocations are valued: in

2016, 70% of the average Mexican municipality’s income came from federal transfers (IMCO,

2018). Unlike NTD in Japan, which have been available for decades, FORTALECE funds appear

in the budget only in the period 2013-17, which limits our analysis to this period.14 Concretely,

we test the following:

• Hypothesis III: Municipalities in SSDs with PVEM-affiliated joint candidates are re-

warded with more FORTALECE after elections when they increase votes for this candidate

under the PRI label and decrease them under the PVEM or both parties’ label.

12A telephone interview with an official from Mexico’s Budget Transparency Agency confirmed that this fund isentirely under the control of Deputies (August 22, 2018, New York-Mexico City).

13This fund consisted of four separate funds until 2016, when they were combined into one. We combine allocationsunder each of the four for the years before 2016.

14In late 2017, the head of the PRI’s parliamentary COD group, Cesar Camacho, announced that the PRI haddecided to ‘remove any present or future temptation’ for Deputies to ‘steer funds toward municipalities of their choosing’by eliminating the fund (Expansion, 2017).

15

4 Data

We assembled comprehensive data on all municipalities in Japan (1996-2013) and Mexico (2011-

16). For Japan, data on election returns come from JED-M (Mizusaki, 2014). For annual

amounts of NTD, we relied on data from Nikkei NEEDs, supplemented where necessary with

data from official government sources.15 We also used Nikkei NEEDs for control variables

used in research on transfers: per capita income, population, ‘fiscal strength’, proportion of

residents employed in primary industries, proportion of residents aged 15 and under, proportion

of residents aged 65 and over, and population density (Horiuchi and Saito, 2003).16 For Mexico,

elections data is from the National Electoral Institute (INE, 2018). For FORTALECE, we relied

on the relevant section of the Federal Expenses Budget for allocations in 2013 (Camara de

Diputados, 2012) and on data collected by the Ministry of Finance and Public Credit’s Budget

Transparency Project for allocations from 2014 until 2017 (Observatorio del gasto, 2018). We

collected data on the following control variables: state of emergency declarations, population

and poverty, municipality size, and their designation as urban or rural.17

5 Rewarding Coordination in Japan

To examine Hypothesis I, we restricted our analysis to Japanese municipalities located exclu-

sively within ‘LDP SSDs’ (those with LDP candidates and without Komeito candidates) in

the 2003, 2005, and 2012 HOR elections.18 Table 1 presents the results of fixed effect regres-

sions using this sample of 4,497 municipalities. The dependent variable is Log(Transfersm,t+1):

the logarithm of per capita NTD received by municipalities in the fiscal years following the

15The NEEDs data is described here.16The first three are measured annually. The second three are measured in censuses conducted every five years.

For values in off-years, we used the value in the census year closest to the off-year. Population density was created bydividing a municipality’s population by its size in kilometers squared. ‘Fiscal strength’ is a government calculation ofthe proportion of the cost of delivering required services that a municipality can finance with its own revenue, derivedfrom taxation.

17For the first variable, we used data from the Governance Ministry’s Civil Protection Office (Secretaria de Gob-ernacion, 2018). For the second and third, we relied on data compiled by the National Population Council (ConsejoNacional de Poblacion, 2017). For the latter two, we relied on the National Institute of Statistics and Geography(INEGI, 2017). Municipalities are defined as rural if they contain less than 2,500 people, and in a state of emergencyif a natural disaster has placed a municipality at imminent risk.

18We begin our analysis in 2003 because this was the first election after the two parties began coordinating whereour three independent variables are observed (the Komeito did not run as a separate party in 1996, meaning that∆Komeito PR VSm,t is not observed in 2000). We exclude 2009 because the LDP-Komeito coalition lost the 2009election, meaning it was not in control of transfers in 2010.

16

2003, 2005, and 2012 HOR elections. We have three independent variables of interest. One,

∆Komeito PR VSm,t is the change in share of municipality m’s voting population casting their

PR votes for the Komeito at time t (the current election) relative to the previous election (with

higher scores indicating greater increases in share of voting population who voted Komeito in

PR). Two, Negative ∆LDP PR VSm,t is the change in share of municipality m’s voting popula-

tion casting their PR votes for the LDP at time t (the current election) relative to the previous

election, multiplied by -1 (with higher scores indicating greater decreases in share of the voting

population who voted LDP in PR). Three, the interaction of ∆Komeito PR VSm,t and Nega-

tive ∆LDP PR VSm,t captures the effect of a simultaneous increase in share of a municipality’s

voting population who voted Komeito in PR and decrease in share of its voting population who

voted LDP in PR.19

In all models, SSD-year fixed effects control for features of a municipality’s SSD in a given

election that could influence the amount of transfers received by all municipalities therein after

the election. These could include whether the coalition’s candidate was victorious or whether the

SSD ended up with two HOR representatives (due to a candidate from one party winning and a

candidate from another party losing but being resurrected via PR).20 Our inclusion of SSD-year

fixed effects means that we are comparing the amount of transfers received by municipalities

in the same SSD-year. In Models 2 and 4, we add municipality fixed effects. This controls for

time-invariant features of a municipality that could influence the amount of transfers it receives,

such as its ability to put together proposals for projects for which to seek funding and build a

consensus around those projects. Municipality fixed effects enable us to leverage changes in the

same municipality’s level of compliance over time.

All models in Table 1 also include the following time-varying municipality-level controls:

population (logged), per capita income (logged), proportion of residents employed in agriculture,

proportion of residents who are dependent (aged 15 and under or 65 and over), population

density, and fiscal strength. All models also include the logarithm of per capita transfers received

by the municipality the year of the election. The lagged dependent variable helps us guard

19We use shares of eligible voters rather than raw numbers of PR votes to account for the fact that municipalitiesvary greatly in size, even within the same SSD. Descriptive statistics are in Section D of the Online Appendix. Wepresent summarized results here. The full specification, which includes coefficients on the control variables, is presentedin Section E of the Online Appendix.

20Japan permits dual candidacy, meaning a candidate can run in an SSD and also appear on a party’s list (enablingher to be resurrected should she lose her SSD) (Pekkanen, Nyblade and Krauss, 2006).

17

against the possibility that the election results have no independent effect on transfers once the

amount of transfers the municipality received the year of the election is accounted for. If this

was the case, it would suggest that another factor was causing municipalities to exhibit greater

compliance in elections and receive more transfers. Models 3 and 4 include an additional control:

∆LDP SSD VSm,t, which captures the change in share of municipality m’s voting population

casting their SSD votes for the LDP candidate at time t relative to the previous election.

This controls for the possibility that changes in electoral support for the LDP’s SSD candidate

could be driving any observed effect of ∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t on

transfers. Finally, because the nature of a municipality’s SSD determines how LDP and Komeito

supporters are supposed to cast their PR votes, we cluster standard errors by SSD.

The positive, significant coefficients on ∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t (all

models) show that in LDP SSDs, controlling for time-invariant and time-varying municipality-

level features, as well as features of the municipality’s SSD-year and changes in electoral strength

of the LDP’s SSD candidate, municipalities that increased PR votes for the Komeito while

decreasing them for the LDP in the 2003, 2005, and 2012 HOR elections received more money

the year after the election. Substantively, Model 1 shows that a one percentage point increase

in ∆Komeito PR VSm,t in a municipality where the share of eligible voters casting PR votes

for the LDP declined by 10% (Negative ∆LDP PR VSm,t=10) is predicted to increase the

amount of NTD received by 1,139 yen per person (approximately $10.50 USD). In contrast,

a one percentage point increase in ∆Komeito PR VSm,t in a municipality where the share of

eligible voters casting PR votes for the LDP increased by 10% (Negative ∆LDP PR VSm,t=-10)

is predicted to decrease the amount of NTD received by 1,695 yen per person (approximately

$15.60 USD).21

Figure 1 is a marginal effects plot of the expected change in Log(Transfersm,t+1) (y-axis) as

values of Negative ∆LDP PR VSm,t (x -axis) increase at different levels of ∆Komeito PR VSm,t.

Recall that increases in Negative ∆LDP PR VSm,t mean decreases in LDP PR vote share. De-

creases in LDP PR vote share have the effect of increasing transfers in municipalities where

the Komeito PR vote share increased by 10% (the black dotted line), decreasing transfers in

municipalities where the Komeito PR vote share stays the same (the gray dotted line), and

21Average marginal effects are calculated with all other continuous variables held constant at their sample means. InSection E of the Online Appendix, Figure E.2 presents the average marginal effect of ∆Komeito PR VSm,t at differentlevels of Negative ∆LDP PR VSm,t.

18

Tab

le1:

InL

DP

SS

Ds,

mu

nic

ipal

itie

sth

atin

crea

sed

PR

vote

sfo

rth

eK

omei

tow

hil

ed

ecre

asin

gth

emfo

rth

eL

DP

inth

e200

3,20

05,

an

d20

12

HO

Rel

ecti

on

sw

ere

rew

ard

edw

ith

mor

em

oney

afte

rel

ecti

ons.

Mu

nic

ipal

itie

sth

at

dec

rease

dP

Rvo

tes

for

the

LD

P(w

ith

out

chan

ges

toK

omei

toP

Rvo

tesh

are)

wer

ep

enal

ized

wit

hle

ssm

oney

,w

hil

em

un

icip

alit

ies

that

incr

ease

dP

Rvo

tes

for

the

Kom

eito

(wit

hou

tch

an

ges

toL

DP

PR

vote

shar

e)w

ere

nei

ther

pen

aliz

edn

orre

war

ded

.

Dep

enden

tV

aria

ble

:L

og(T

ransf

ers m

,t+1)

Model

1M

odel

2M

odel

3M

odel

4∆

Kom

eito

PR

VSm,t∗

Neg

ativ

e∆

LD

PP

RV

Sm,t

0.00

249*

**0.

0021

1**

0.00

249*

**0.

0021

0***

[0.0

0093

][0

.000

81]

[0.0

0093

][0

.000

81]

∆K

omei

toP

RV

Sm,t

-0.0

0518

-0.0

0018

-0.0

0518

-0.0

0111

[0.0

0778

][0

.008

51]

[0.0

0808

][0

.008

59]

Neg

ativ

e∆

LD

PP

RV

Sm,t

-0.0

0996

**-0

.009

13**

-0.0

0997

**-0

.007

17[0

.003

99]

[0.0

0434

][0

.004

20]

[0.0

0497

]∆

LD

PSSD

VSm,t

-0.0

0001

0.00

162

[0.0

0203

][0

.002

10]

Log

(Tra

nsf

ers m

,t)

0.72

027*

**0.

4978

7***

0.72

027*

**0.

4976

2***

[0.0

2646

][0

.034

22]

[0.0

2647

][0

.034

09]

Con

stan

t-1

.822

02**

*9.

1497

6***

-1.8

2202

***

9.36

568*

**[0

.296

20]

[3.2

7514

][0

.296

12]

[3.3

0196

]C

ontr

ols

Yes

Yes

Yes

Yes

SSD

-Yea

rF

EY

esY

esY

esY

esM

unic

ipal

ity

FE

Yes

Yes

Obse

rvat

ions

4,49

74,

497

4,49

74,

497

R-s

quar

ed0.

8079

20.

7588

10.

8079

20.

7588

8N

um

ber

ofm

unic

ipal

itie

s2,

293

2,29

3R

obust

stan

dar

der

rors

atth

edis

tric

tle

vel

inbra

cket

s**

*p<

0.01

,**

p<

0.05

,*

p<

0.1

19

decreasing transfers to an even larger degree in municipalities where the Komeito PR vote share

also decreased by 10%. Hainmueller, Mummolo and Xu (2019) explain why valid inferences as

to the effects of an interaction with at least one continuous variable can be drawn only after re-

searchers have verified that the interaction effect is linear and there is sufficient common support

for the moderator. Section E implements their diagnostics, which support both assumptions.

-3.6

-3.4

-3.2

-3-2

.8Li

near

Pre

dict

ion

-10 0 10Negative ΔLDP PR VS

ΔKomeito PR VS=-10 ΔKomeito PR VS=0ΔKomeito PR VS=10

Figure 1: We plot the predicted change in Log(Transfersm,t+1) in LDP SSDs as Negative ∆LDP PR VSm,t

increases at one-percentage-point increments (indicating decreases in LDP PR vote share), as∆Komeito PR VSm,t is held constant at different values. These marginal effects are from Table 1, Model 1.

We also examined whether the effect of ∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t

could be explained by other parties’ supporters switching their PR votes to the Komeito and the

LDP’s PR vote share declining for another reason. We created Negative ∆Non-LDP/Komeito PR VSm,t,

which is the change in share of municipality m’s voting population casting their PR votes for nei-

ther the LDP nor the Komeito at time t relative to the previous election, multiplied by -1 (with

higher scores indicating greater decreases in share of the voting population who voted for another

party in PR). We then redid Table 1 with Negative ∆Non-LDP/Komeito PR VSm,t instead of

Negative ∆LDP PR VSm,t, keeping everything else about the specifications identical. The re-

sults are in Section G of the Online Appendix. None of the coefficients (on ∆Komeito PR VSm,t,

Negative ∆LDP PR VSm,t, or on ∆Komeito PR VSm,t∗Negative ∆Non-LDP/Komeito PR VSm,t)

are significant in any specification. This suggests that our results cannot (at least, not easily)

be attributed to other factors bringing about changes in the two parties’ vote shares. In sum,

our results support Hypothesis I.

20

Next, we examine whether a similar reward regime, but in reverse, operates in the small

number of SSDs where the LDP stands down and a Komeito candidate runs (Hypothesis II).

Table 2 presents the results of a regression on the 63 municipalities in ‘Komeito SSDs’ in the

2003, 2005, and 2012 HOR elections. The specification is identical to that in Model 1 of Table

1.22 The negative, significant coefficient on ∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t

demonstrates that in Komeito SSDs, controlling for time-varying municipality-level features and

features of the municipality’s SSD-year, municipalities that increased PR votes for the Komeito

while decreasing them for the LDP in the 2003, 2005, and 2012 HOR elections received less

money the year after the election. In LDP SSDs, this pattern of behavior is rewarded, but in

Komeito SSDs, it is penalized. Substantively, in Komeito SSDs, a one percentage point increase

in ∆Komeito PR VSm,t in a municipality where the share of eligible voters casting their PR

votes for the LDP declined by 10% (Negative ∆LDP PR VSm,t=10) is predicted to decrease the

amount of NTD received by 10,525 yen per person (approximately $96.80 USD).23

Figure 2 is a marginal effects plot of the expected change in Log(Transfersm,t+1) (y-axis) as

values of Negative ∆LDP PR VSm,t (x -axis) increase at different levels of ∆Komeito PR VSm,t

in Komeito SSDs. Decreases in LDP PR vote share have the effect of decreasing transfers to

a large degree in municipalities where the Komeito PR vote share increased by 10% (the black

dotted line), decreasing transfers to a lesser degree in municipalities where the Komeito PR vote

share stays the same (the gray dotted line), and slightly increasing transfers in municipalities

where the Komeito PR vote share also decreased by 10%. The black dotted lines in Figures 1 and

2 show that the same behavior is rewarded in LDP SSDs and penalized in Komeito SSDs. Sim-

ilarly, Section F of the Online Appendix reveals that the assumptions of linearity and common

support for the moderator are supported. It is striking that we observe a statistically significant

coefficient on ∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t, in the directed aligned with our

hypothesis, and in a specification that controls for prior transfers and SSD-year fixed effects, on

a sample of only 63 observations. In sum, this supports Hypothesis II.

22We present the summarized results here and the full specification, which includes coefficients on the controlvariables, in Section F of the Online Appendix. We cannot conduct the other specifications in Table 1 because thenumber of observations is too low. Controlling for ∆Komeito SSD VSm,t (change in share of a municipality’s eligiblevoters who cast their SSD votes for the Komeito candidate), for example, reduces the number of observations to just23.

23Average marginal effects are calculated with all other continuous variables held constant at their sample means. InSection F of the Online Appendix, Figure F.2 presents the average marginal effect of ∆Komeito PR VSm,t at differentlevels of Negative ∆LDP PR VSm,t.

21

Table 2: In Komeito SSDs, municipalities that increased PR votes for the Komeito while decreas-ing them for the LDP in the 2003, 2005, and 2012 HOR elections were penalized with less moneyafter elections. Municipalities that decreased PR votes for the LDP (without changes to KomeitoPR vote share) were also penalized, while municipalities that increased PR votes for the Komeito(without changes to LDP PR vote share) were neither penalized nor rewarded.

Dependent Variable: Log(Transfersm,t+1)Model 1

∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t -0.008***[0.001]

∆Komeito PR VSm,t -0.007[0.010]

Negative ∆LDP PR VSm,t -0.057***[0.002]

Log(Transfersm,t) 0.530***[0.053]

Constant -2.663***[0.227]

Controls YesSSD-Year FE YesObservations 63R-squared 0.659

Robust standard errors at the district level in brackets*** p<0.01, ** p<0.05, * p<0.1

-4-3

-2-1

Line

ar P

redi

ctio

n

-10 0 10Negative ΔLDP PR VS

ΔKomeito PR VS=-10 ΔKomeito PR VS=0ΔKomeito PR VS=10

Figure 2: We plot the predicted change in Log(Transfersm,t+1) in Komeito SSDs as Negative∆LDP PR VSm,t increases at one-percentage-point increments (indicating decreases in LDP PR vote share),as ∆Komeito PR VSm,t is held constant at different values. These marginal effects are from Table 2, Model1.

22

5.1 What Else Can Our Results Tell Us About Transfers?

Our results show that in LDP SSDs, municipalities where PR votes increased for the Komeito

and decreased for the LDP received more money after these elections. In Komeito SSDs, the

reverse is true: municipalities where PR votes increased for the Komeito and decreased for

the LDP received less money after these elections. This supports our claim that in our period

of study, the LDP-Komeito governing coalition used transfers to motivate LDP supporters to

switch their PR votes to the Komeito in LDP SSDs and Komeito supporters to switch their PR

votes to the LDP in Komeito SSDs. The coefficients on the uninteracted variables can tell us

more about how the coalition uses transfers.

First, let us take a closer look at LDP SSDs. In Table 1, the coefficients on ∆Komeito PR VSm,t

are negative and insignificant. This means that municipalities where PR votes increased for the

Komeito and stayed the same for the LDP were neither penalized with less money, nor rewarded

with more. This suggests that the coalition rewarded LDP supporters for switching their PR

votes to the Komeito (the positive, significant coefficient on ∆Komeito PR VSm,t∗ Negative

∆LDP PR VSm,t), but did not appear to reward Komeito supporters for mobilizing more PR

votes for their party. An implication of this is that Komeito supporters were not able to gain

more money for their community by mobilizing extra PR votes for their party in LDP SSDs.

The coefficients on Negative ∆LDP PR VSm,t, on the other hand, are negative and signifi-

cant, except in Model 4, where it is negative and insignificant. A negative, significant coefficient

on Negative ∆LDP PR VSm,t means that municipalities where PR votes decreased for the LDP

and stayed the same for the Komeito are penalized with less money after elections, which implies

that transfers may also have been used to encourage LDP supporters to cast their PR votes

for the LDP. In the specification that leverages over-time variation in the same municipality’s

voting behavior and controls for changes in the share of eligible voters casting their SSD votes

for the LDP candidate (Model 4), however, ∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t

remains positive and significant, while Negative ∆LDP PR VSm,t loses its significance. This

means that the coalition rewards compliance, not increases in PR votes cast for the LDP.

Turning to Komeito SSDs (Table 2), the coefficient on Negative ∆LDP PR VSm,t is negative

and significant, which means that municipalities where PR votes decrease for the LDP and stay

the same for the Komeito are penalized. In Komeito SSDs, then, transfers are used to motivate

23

LDP supporters to mobilize more PR votes for the LDP. The coefficient on ∆Komeito PR VSm,t,

on the other hand, is negative and insignificant. This means that municipalities where PR votes

increase for the Komeito and stay the same for the LDP are neither penalized nor rewarded. In

Komeito SSDs, then, the coalition rewards municipalities where Komeito supporters switch their

PR votes to the LDP (the negative, significant coefficient on ∆Komeito PR VSm,t∗ Negative

∆LDP PR VSm,t), but refrains from penalizing municipalities where PR votes for the Komeito

increase. In Komeito SSDs, then, Komeito supporters can gain more money for their community

only by switching their PR votes to the LDP, not by casting them for the Komeito.

6 Rewarding Compliance in Mexico

To examine Hypothesis III, Table 3 presents the results of fixed effect regressions. The de-

pendent variable is Log(Transfersm,t+1): the logarithm of per capita FORTALECE received

by municipalities in the years following the COD elections held in 2012 and 2015, respectively.

Model 1 focuses on municipalities located exclusively within alliance SSDs with PVEM-affiliated

joint candidates; Model 2 focuses on municipalities in alliance SSDs with PRI-affiliated joint

candidates; and Model 3 focuses on municipalities in non-alliance SSDs, where both parties

fielded their own candidate.24

In Models 1 and 2, we have three independent variables of interest: ∆PRI VSm,t, or change

in share of municipality m’s voting population who selected the joint candidate under the PRI

label only at time t (the current election) relative to the most recent similar election;25 Negative

(∆PVEM or PVEM-PRI VSm,t), or change in share of municipality m’s voting population who

selected the joint candidate under either the PVEM label or both parties’ labels at time t

relative to the most recent similar election, multiplied by -1 (with higher scores indicating

greater decreases in share of the voting population selecting the joint candidate under either the

24We present the summarized results here and the full specification, which includes coefficients on the controlvariables, in Section I of the Online Appendix.

25Some COD elections are concurrent with presidential elections, which has consequences for levels of support andturnout. We anticipate that the PRI-PVEM coalition would have compared municipality m’s vote shares with itsvote shares in the most recent similar election, meaning that 2012 will be compared to 2006 (both concurrent) and2015 will be compared to 2009 (neither were). Our independent variables were constructed to reflect this. Comparingmunicipality m’s vote shares in 2012 with those in 2006 would have presented an additional wrinkle: coordinatingparties had to present joint lists in 2006, meaning that only the number of votes cast for the PRI-PVEM coalition isobserved. Our construction of the independent variable for 2012 reflects what we can expect a coalition interested inverifying the extent of compliance would have done, and is detailed in Section H of the Online Appendix.

24

Tab

le3:

Inall

ian

ceS

SD

sw

ith

PV

EM

-affi

liat

edca

nd

idat

es,

mu

nic

ipal

itie

sth

atin

crea

sed

vote

sfo

rth

isca

nd

idate

un

der

the

PR

Ila

bel

on

lyw

hil

ed

ecre

asi

ng

them

un

der

the

PV

EM

orb

oth

par

ties

’la

bel

sw

ere

rew

ard

edw

ith

mor

em

oney

afte

rth

e20

12an

d20

15el

ecti

ons

(Mod

el1).

Inall

ian

ceS

SD

sw

ith

PR

I-affi

liat

edca

nd

idat

es(M

od

el2)

and

non

-all

ian

ceS

SD

s(M

od

el3)

,m

un

icip

ali

ties

that

incr

ease

dvo

tes

for

the

PR

I(o

nly

,in

the

case

ofM

od

el2)

are

rew

ard

ed.

Dep

enden

tV

aria

ble

:L

og(T

ransf

ers m

,t+1)

Model

1M

odel

2M

odel

3Joi

nt

Can

did

ate

(PV

EM

)Joi

nt

Can

did

ate

(PR

I)N

on-A

llia

nce

Dis

tric

t∆

PR

IV

Sm,t∗

Neg

ativ

e(∆

PV

EM

orP

VE

M-P

RI

VSm,t)

0.00

214*

0.00

000

[0.0

0117

][0

.000

51]

∆P

RI

VSm,t

0.01

261

0.01

169*

0.01

760*

*[0

.018

49]

[0.0

0618

][0

.007

55]

Neg

ativ

e(∆

PV

EM

orP

VE

M-P

RI

VSm,t)

0.02

680*

0.01

310

[0.0

1370

][0

.008

48]

∆P

RI

VSm,t∗

Neg

ativ

e(∆

PV

EM

VSm,t)

-0.0

0037

[0.0

0214

]N

egat

ive

(∆P

VE

MV

Sm,t)

0.01

483

[0.0

1942

]SSD

-Yea

rF

EY

esY

esY

esO

bse

rvat

ions

170

1,13

563

5R

-squar

ed0.

6710

10.

6254

20.

7278

0R

obust

stan

dar

der

rors

clust

ered

atth

edis

tric

tle

vel

inbra

cket

s**

*p<

0.01

,**

p<

0.05

,*

p<

0.1

25

PVEM or both parties’ labels at time t relative to the most recent similar election); and their

interaction, which captures the effect of a simultaneous increase in votes for the joint candidate

under the PRI label only and decrease in votes for the joint candidate under the PVEM or both

parties’ labels. In Model 3 (non-alliance districts), our three independent variables of interest

are: ∆PRI VSm,t, or change in share of municipality m’s voting population voting PRI at time

t relative to beforehand; Negative (∆PVEM VSm,t), or change in share of municipality m’s

voting population voting PVEM at time t relative to beforehand, multiplied by -1; and their

interaction.

In all models, SSD-year fixed effects control for features of a municipality’s SSD in a given

election that could influence the amount of transfers received by all municipalities therein after

the election. Our inclusion of SSD-year fixed effects means that we are comparing the amount of

transfers received by municipalities in the same SSD-year. All models include the following time-

varying, municipality-level controls: population (logged), a dummy variable indicating whether

the municipality is rural or urban, population density, surface area of the municipality (in

squared kilometers), poverty index, and a dummy variable indicating whether the municipality

is in a state of emergency. We cluster standard errors by SSD.

Hypothesis III concerns municipalities in alliance SSDs with PVEM-affiliated joint candi-

dates (Model 1). The positive, significant coefficient on the interaction shows that, controlling

for features of a municipality’s SSD-year and other time-varying differences, municipalities in al-

liance SSDs with PVEM-affiliated candidates that increased votes for the joint candidate under

the PRI label only while decreasing them for this candidate under the PVEM or both parties’

labels received more money the year after the 2012 and 2015 elections. Substantively, Model 1

shows that a one percentage point increase in Negative (∆PVEM or PVEM-PRI VSm,t), which

corresponds to a one percentage point decrease in share of voters choosing the joint candidate

under the PVEM or both parties’ labels, in a municipality where the share of voters choosing the

joint candidate under the PRI label only increased by 10% is predicted to increase the amount

of FORTALECE received by 9.9 pesos per person (approximately 50 US cents).26

The positive, significant coefficient on Negative (∆PVEM or PVEM-PRI VSm,t), on the

other hand, shows that increases in votes for the joint candidate under the PVEM or both parties’

26Average marginal effects are calculated with all other continuous variables held constant at their sam-ple means. In Section I of the Online Appendix, Figure I.2 presents the average marginal effect of Negative(∆PVEM or PVEM-PRI VSm,t) at different levels of ∆PRI VSm,t.

26

labels are, when ∆PRI VSm,t is zero, penalized with less money after elections. Substantively, a

1% increase in votes cast under the PVEM or both parties’ labels results in a per person loss of

5.44 pesos (approximately 30 US cents). The non-significant coefficient on ∆PRI VSm,t means

that increases in votes for the joint candidate under the PRI label only are neither penalized

nor rewarded. Taken together, these results show that in alliance SSDs with PVEM-affiliated

candidates, FORTALECE is used to encourage supporters choosing the joint candidate under

the PVEM or both parties’ labels to switch to casting them under the PRI’s label only.

Figure 3 is a marginal effects plot of the expected change in Log(Transfersm,t+1) (y-axis)

as values of Negative (∆PVEM or PVEM-PRI VSm,t) (x -axis) increase at different levels of

∆PRI VSm,t. Decreases in the share of voters choosing the joint candidate under the PVEM or

both parties’ labels have the effect of increasing transfers in municipalities where the PRI vote

share increased by 10% (the black dotted line), slightly increasing transfers in municipalities

where the PRI vote share stays the same (the gray dotted line), and no change in transfers in

municipalities where the PRI vote share also decreased by 10%. Section I of the Online Appendix

present the diagnostics recommended by Hainmueller, Mummolo and Xu (2019), which show

that the assumptions of linearity and common support for the moderator hold.

44.

55

5.5

Line

ar P

redi

ctio

n

-15 -10 -5 0 5 10Negative (∆PVEM or PVEM-PRI VS)

∆PRI VS=-10 ∆PRI VS=0∆PRI VS=10

Figure 3: We plot the predicted change in Log(Transfersm,t+1) in alliance SSDs with PVEM-affiliated candi-dates, as Negative (∆PVEM or PVEM-PRI VSm,t) increases at one-percentage-point increments (indicatingdecreases in PVEM-PRI vote share), as ∆PRI VSm,t is held constant at different values. These marginaleffects are from Table 3, Model 1.

27

Models 2 and 3 serve as placebo tests. In alliance SSDs with PRI-affiliated candidates (Model

2), neither the coefficient on the interaction nor on Negative (∆PVEM or PVEM-PRI VSm,t)

are significant. In non-alliance SSDs (Model 3), neither the coefficient on the interaction nor on

Negative (∆PVEM VSm,t) are significant. This means that it is only in alliance SSDs where PRI

candidates stand down (Model 1) that FORTALECE is used to encourage voters to switch their

votes from the joint candidate under the PVEM or both parties’ labels to the joint candidate

under the PRI label. This supports Hypothesis III.

6.1 What Else Can Our Results Tell Us About Transfers?

In SSDs where the two parties fielded a joint candidate who was PRI-affiliated (Model 2),

the coefficient on ∆PRI VSm,t is positive and significant, while the coefficient on Negative

(∆PVEM or PVEM-PRI VSm,t) is insignificant. This means that the coalition rewarded in-

creases in votes cast for the PRI-affiliated joint candidate under the PRI label only, but neither

rewarded nor penalized increases in votes cast for the PRI-affiliated joint candidate under the

PVEM or both parties’ labels. Substantively, a 1% increase in votes for the PRI-affiliated joint

candidate under the PRI label translates into an extra 2.65 pesos per person (approximately 15

US cents). In these SSDs, then, transfers are used to encourage supporters to cast both com-

ponents of their vote for the PRI. Because votes cast for the joint candidate under the PVEM

or both parties’ label in SSDs where the joint candidate is PRI increase the chance of the PRI

capturing the SSD, PRI does not dissuade voters from casting their votes this way. But nor

does it encourage this.

In Model 3, the coefficient on ∆PRI VSm,t is similarly positive and significant, revealing that

the coalition rewards increases in votes cast for the PRI. Substantively, a 1% increase in votes

cast for the PRI translates into an extra 4.41 pesos per person (approximately 25 US cents).

Rewards for voting PRI are largest in these non-alliance SSDs. The coefficient on Negative

(∆PVEM VSm,t) is similarly insignificant. Thus, in non-alliance SSDs, transfers are used to

encourage supporters to cast both components of their vote for the PRI.

28

7 Conclusion

We have offered a new theory for how governing parties can increase their chances of winning

elections under MMM by engaging in preelectoral coordination with small parties and using

government resources to encourage supporters to switch their votes accordingly. Turning to

Japan’s HOR and Mexico’s COD, we showed that the LDP-Komeito and PRI-PVEM coalitions

did exactly this during the period we study.

For comparative politics scholars, we have three main contributions. First, we have identified

a new pathway through which governing parties that might have policies unpopular with the

median voter can use pork-barrelling to stay in power. Whether governing coalitions in the other

28 countries using MMM are doing this should be examined, as well as whether opposition parties

have a systematically tougher time winning elections under MMM for this reason. Second, we

show that parties interested in coordinating can find a way to do so, even when voters have a

single vote. Future research should investigate whether coordination happens in other systems

where it is not expected, and also whether changes in ballot structure, such as South Korea’s 2003

switch from single to dual-ballot and Italy’s 1993 switch from dual to single-ballot, produced

the changes in coordination we expect. Third, we show that when the parties in government

got there by coordinating, one of the ‘goods’ supporters are likely to be receiving is spending

on their community. Future research should investigate the relationship between the policies

passed by a coalition dependent upon coordination and the size of the payoffs in terms of local

spending accorded to each partner. It is possible that geographically-targeted spending is also

being used to buy a coalition partner’s support for policies it finds unpalatable.

For scholars of Japan and Mexico, we have shown that both countries’ longstanding govern-

ing coalitions won elections by coordinating and using government resources to overcome any

resistance to this coordination. While this strategy likely contributed to their dominance, the

LDP lost the 2009 HOR election and the PRI lost the 2018 COD election. Why this strategy

appears to have failed in these elections and why, in the Japanese case, the Democratic Party

of Japan, which won the 2009 election, was unable to replicate it to win again in 2012, are

important questions. Another important question is whether Japan’s LDP is neutralizing the

Komeito’s opposition to its desired defense policies by increasing the size of the payoff in local

spending to Komeito supporters.

29

References

Ames, Barry. 1995. “Electoral Strategy under Open-List Proportional Representation.” Amer-ican Journal of Political Science 39(2):406–433.

Batto, Nathan F., Chi Huang, Alexander C. Tan and Gary W. Cox, eds. 2016. Mixed-MemberElectoral Systems in Constitutional Context: Taiwan, Japan, and Beyond. University ofMichigan Press.

Bawn, Kathleen and Michael Thies. 2003. “A Comparative Theory of Electoral Incentives:Representing the Unorganized under PR, Plurality and Mixed-member Electoral Systems.”Journal of Theoretical Politics 5(1):5–32.

Bormann, Nils-Christian and Matt Golder. 2013. “Democratic Electoral Systems around theworld, 19462011.” Electoral Studies 32(2):360 – 369.

Camara de Diputados, Honorable Congreso de la Union. 2012. “Presupuesto de Egresos de laFederacion para el Ejercicio Fiscal 201 3.” Diario Oficial de la Federacion.

Campos, Mariana, Alejandro Garcia and Liliana Ruiz. 2018. “Arquitectura del Ramo 23.”Mexico Evalua, Centro de Analisis de Politicas Publicas .

Casar, Maria Amparo. 2012. “El reparto del poder: Otro gobierno sin mayoria.” Nexos .

Catalinac, Amy, Bruce Bueno de Mesquita and Alastair Smith. 2019. “A Tournament Theoryof Pork Barrel Politics: The Case of Japan.” Comparative Political Studies .

Cervantes, Jesusa. 2017. “El PRI impulsa la eliminacin del fondo de los moches en San Lazaro.”Proceso .

Christensen, Ray. 2000. Ending the LDP Hegemony: Party Cooperation in Japan. Universityof Hawaii Press.

Collier, Ruth Berins. 1992. The contradictory alliance : state-labor relations and regime changein Mexico. Berkeley: International and Area Studies, University of California at Berkeley.

Consejo Nacional de Poblacion, CONAPO. 2017. “Indice de Marginacion.”.

Costa-I-Font, Joan, Eduardo Rodriguez-Oreggia and Dario Lunapla. 2003. “Political Competi-tion and Pork-Barrel Politics in the Allocation of Public Investment in Mexico.” Public Choice116(1/2):185.

Cox, Karen E. and Leonard Schoppa. 2002. “Interaction Effects in Mixed Member ElectoralSystems: Theory and Evidence from Germany, Japan, and Italy.” Comparative Politics35(9):1027–1053.

Dahlberg, Matz and Eva Johansson. 2002. “On the vote-purchasing behavior of incumbentgovernments.” American Political Science Review 96(1):27–40.

De La O, Ana Lorena. 2015. Crafting Policies to End Poverty in Latin America: The QuietTransformation. Cambridge University Press.

Expansion. 2017. “El PRI ”reflexiona” tras los sismos y ofrece eliminar el fondo de moches.”Expansion .

30

Ferrara, Federico and Erik S. Herron. 2005. “Going It Alone? Strategic Entry under MixedElectoral Rules.” American Journal of Political Science 49(1):16–31.

Funk, Patricia and Christina Gathmann. 2013. “How do electoral systems affect fiscal policy?Evidence from cantonal parliaments, 18902000.” Journal of the European Economic Associ-ation 11(5):11781203.

Garcia, Imelda. 2017. “Fondo de Abundacia y Opacidad.” Reporte Indigo .

Golden, Miriam and Brian Min. 2013. “Distributive Politics Around the World.” Annual Reviewof Political Science 16(1):73–99.

Golden, Miriam and Lucio Picci. 2008. “Pork-Barrel Politics in Postwar Italy, 1953-94.” Amer-ican Journal of Political Science 52(2):268–289.

Golder, Sona N. 2006. “Pre-Electoral Coalition Formation in Parliamentary Democracies.”British Journal of Political Science 36(2):193–212.

Hainmueller, Jens, Jonathan Mummolo and Yiqing Xu. 2019. “How Much Should We TrustEstimates from Multiplicative Interaction Models? Simple Tools to Improve Empirical Prac-tice.” Political Analysis 27:163–192.

Herron, Erik S., Kuniaki Nemoto and Misa Nishikawa. 2018. Reconciling Approaches in theStudy of Mixed-Member Electoral Systems. In Oxford Handbook for Electoral Systems, ed.Erik S. Herron, Robert Pekkanen and Matthew Shugart. New York, NY: pp. 445–471.

Hirano, Shigeo. 2006. “Electoral Institutions, Hometowns, and Favored Minorities: Evidencefrom Japanese Electoral Reforms.” World Politics 58:51–82.

Hiskey, Jonathan. 1999. “Does Democracy Matter? Electoral Competition and Local Develop-ment in Mexico.” Ph.D. Dissertation .

Horiuchi, Yusaku and Jun Saito. 2003. “Reapportionment and Redistribution: Consequences ofElectoral Reform in Japan.” American Journal of Political Science 47(4):669–682.

IMCO, Instituto Mexicano para la Competitividad. 2018. “Indice de Informa-cion Presupuestal Municipal. Novena edicion.” https://imco.org.mx/temas/

indice-informacion-presupuestal-municipal/.

INE, Instituto Nacional Electoral. 2018. “Bases de Datos de los Procesos Electorales Federalesen Mexico (1991-2015.” http://siceef.ine.mx/downloadDB.html.

INEGI, Instituto Nacional de Estadistica y Geografia. 2017. “Sistema Estatal y Municipal deBases de Datos.”.

Jimenez, Horacio and Suzzete Alcantara. 2016. “Fondos para moches no son una sorpresa.” ElUniversal .

Kellam, Marisa. 2017. “Why Pre-Electoral Coalitions in Presidential Systems?” British Journalof Political Science 47(2):391411.

Kerevel, Yann. 2010. “The legislative consequences of Mexicos mixed-member electoral system,20002009.” Electoral Studies 29(4):691 – 703.

Kerevel, Yann P. 2015. “Pork-Barreling without Reelection? Evidence from the MexicanCongress.” Legislative Studies Quarterly 40(1):137–166.

31

Klein, Axel. 2013. Komeito - The First Third Force. In Japan Decides 2012, ed. Robert J.Pekkanen, Steven R. Reed and Ethan Scheiner. London, U.K.: Palgrave MacMillan.

Krauss, Ellis, Kuniaki Nemoto and Robert Pekkanen. 2012. “Reverse Contamination: Burningand Building Bridges in Mixed-Member Systems.” Comparative Political Studies 45(6):747–773.

Liff, Adam P. and Ko Maeda. 2019. “Electoral incentives, policy compromise, and coalitiondurability: Japan’s LDPKomeito Government in a mixed electoral system.” Japanese Journalof Political Science 20(1):5373.

Magaloni, Beatriz. 2006. Voting for autocracy : hegemonic party survival and its demise inMexico. New York, NY: Cambridge University Press.

McKean, Margaret and Ethan Scheiner. 2000. “Japan’s new electoral system: la plus ca change.”Electoral Studies 19:447–77.

Mizusaki, Tokifumi. 2014. Sosenkyo Deta Besu. JED-M Deta. Tokyo, Japan: LDB.

Molinar, Juan and Jeffrey Weldon. 1994. Transforming State-Society Relations in Mexico:The National Solidarity Strategy. Center for US-Mexican Studies, UCSD chapter Electoraldeterminants and consequences of national solidarity.

Montero, Sergio. 2015. “Coalition Formation, Campaign Spending, and Election Outcomes:Evidence from Mexico.” Working Paper .

Moser, Robert G. and Ethan Scheiner. 2012. Mixed-Member Electoral Systems. CambridgeUniversity Press p. 4269.

Motolinia, Lucia. 2019. “Electoral Accountability and Particularistic Legislation: Evidence froma Natural Experiment in Mexico.” Working Paper .

Naoi, Megumi and Ellis Krauss. 2009. “Who Lobbies Whom? Special Interest Politics underAlternative Electoral Systems.” American Journal of Political Science 53(4):874–892.

Nemoto, Kuniaki and Chiahung Tsai. 2016. Post Allocation, List Nominations, and PreelectoralCoalitions under MMM. University of Michigan Press pp. 165–193.

Observatorio del gasto, Transparencia Presupuestaria. 2018. “Recursos del Ramo 23 para Enti-dades Federativas y Municipios.”.

Pekkanen, Robert, Benjamin Nyblade and Ellis S. Krauss. 2006. “Electoral Incentives in Mixed-Member Systems: Party, Posts, and Zombie Politicians in Japan.” The American PoliticalScience Review 100(2):183–193.

Reed, Stephen R. and Daniel M. Smith. 2015. “The Reed-Smith Japanese House of Represen-tatives Elections Data Set.” https://sites.google.com/site/danielmarkhamsmith/data.

Reed, Steven R. 1986. Japanese Prefectures and Policymaking. Pittsburgh, PA: University ofPittsburgh Press.

Reed, Steven R. and Kay Shimizu. 2009. Avoiding a Two-Party System: the Liberal DemocraticParty versus Duverger’s Law. In Political Change in Japan, ed. Kay Shimizu, Steven R. Reedand Kenneth McElwain. Stanford University: Walter H. Shorenstein Asia-Pacific ResearchCenter pp. 29–46.

32

Reilly, Benjamin. 2007. “Democratization and Electoral Reform in the Asia-Pacific Region.”Comparative Political Studies 40(11):1350–1371.

Rich, Timothy S. 2015. “Duverger’s Law in mixed legislative systems: The impact of nationalelectoral rules on district competition.” European Journal of Political Research 54(1):182–196.

Rickard, Stephanie J. 2012. “Electoral Systems, Voters Interests and Geographic Dispersion.”British Journal of Political Science 42(4):855877.

Riera, Pedro. 2013. “Electoral coordination in mixed-member systems: Does the level of demo-cratic experience matter?” European Journal of Political Research 52(1):119–141.

Rodriguez-Oreggia, Eduardo and Andres Rodriguez-Pose. 2004. “The Regional Returns of Pub-lic Investment Policies in Mexico.” World Development 32(9):1545 – 1562.

Saito, Jun. 2010. Jiminto Choki Seiken no Seiji Keizaigaku. Tokyo, Japan: Keiso Shobo.

Scheiner, Ethan. 2006. Democracy Without Competition in Japan. Opposition Failure in aOne-party Dominant State. New York, NY: Cambridge University Press.

Secretaria de Gobernacion, SEGOB. 2018. “Declaratorias de Emergencia 2000-2015.”.

Shugart, Matthew Soberg and Martin P. Wattenberg, eds. 2003. Mixed-Member Electoral Sys-tems: The Best of Both Worlds? New York, NY: Oxford University Press.

Smith, Daniel M. 2014. Party Ideals and Practical Constraints in Komeito Candidate Nomi-nations. In Komeito - Politics and Religion in Japan, ed. George Ehrhardt, Axel Klein, LeviMcLaughlin and Steven R. Reed. Berkeley: Institute of East Asian Studies: University ofCalifornia pp. 139–162.

Spoon, Jae-Jae and Amalia Pulido Gomez. 2017. “Unusual Bedfellows? PRIPVEM ElectoralAlliances in Mexican Legislative Elections.” Journal of Politics in Latin America 9(2):63–92.

Spoon, Jae-Jae and Amalia Pulido Gomez. 2018. “Strategic alliances: Red, green, or watermeloncandidates in Mexican legislative elections.” Party Politics pp. 1–10.

Stokes, Susan C. 2005. “Perverse Accountability: A Formal Model of Machine Politics withEvidence from Argentina.” American Political Science Review 99:315–25.

Stratmann, Thomas and Martin Baur. 2002. “Plurality Rule, Proportional Representation,and the German Bundestag: How Incentives to Pork-Barrel Differ across Electoral Systems.”American Journal of Political Science 46(3):506–514.

Tavits, Margit. 2009. “Geographically targeted spending: exploring the electoral strategies ofincumbent governments.” European Political Science Review 1(1):103123.

Thames, Frank C. and Martin S. Edwards. 2006. “Differentiating Mixed-Member ElectoralSystems: Mixed-Member Majoritarian and Mixed-Member Proportional Systems and Gov-ernment Expenditures.” Comparative Political Studies 39(7):905–927.

Weldon, Jeffrey. 2003. Mixed-Member electoral systems: A definition and typology. Oxford,United Kingdom: Oxford University Press chapter The Consequences of Mexico’s Mixed-Member Electoral System, 1988-1997.

Yamada, Kyohei. 2016. “From a Majority to a Minority: How Municipal Mergers in JapanChanged the Distribution of Political Powers and the Allocation of Public Services Within aMerged Municipality.” Urban Affairs Review pp. 1–33.

33

Online Appendix:Geographically-Targeted Spending inMixed-Member Majoritarian Electoral

Systems

A How Seats in the PR Tier are Allocated in Mex-

ico’s COD Elections

The number of valid votes cast for a party list in each of Mexico’s five PR constituencies is

pooled at the national level first, and then divided by 200 to ascertain the number of votes

needed for a party to obtain a single seat.1 For example, if 50 million valid votes were cast

in an election, a party would need 250,000 votes (50 million/200) to capture a seat. Then,

the party that won the most votes is identified. Dividing the total number of valid votes cast

for that party by this quotient yields the number of seats this party ought to receive, barring

caps on over-representation, explained below, are not violated. For example, if Party A receives

20 million votes, this means it is entitled to 80 seats (20 million/250,000 votes). Then, the

total number of votes Party A received in each of the five constituencies is divided by the same

quotient. Thus, if Party A received 3 million votes in Constituency 1, then Constituency 1

would receive 12 of Party A’s 80 seats (3 million/250,000 votes).

After allocating the total number of seats awarded to Party A and the distribution of those

seats across the five constituencies, the allocation for the remaining parties is done simultane-

ously in the following manner. First, the number of votes cast for Party A (in this example, 20

million) is subtracted from the total number of valid votes cast (in this example, 50 million) and

the result is divided by the number of seats remaining to be allocated (in this example, 80 were

awarded to Party A, leaving 120 of the 200 remaining). This quotient (30 million votes/120)

determines the number of votes each of the remaining parties needs to have obtained to win a

seat. Next, regional quotients are generated by dividing the total number of valid votes cast in

each constituency minus the total number of valid votes cast for Party A in each constituency by

the number of seats remaining to be allocated in each constituency. If 6 million valid votes were

1The total number of valid votes is the total number of votes minus votes cast for parties that did not reach thethreshold, independent candidates, and non-registered candidates, as well as null votes.

1

cast in Constituency 1 and 3 million of those were cast for Party A, enabling it to receive 12 of

Party A’s 80 seats, then the regional quotient for Constituency 1 is 6 million minus 3 million,

divided by the 28 seats (the total number of seats available minus 12). This regional quotient

(in this example, 107,143 votes) represents the number of votes each party needs to obtain a

seat in each constituency (in this example, Constituency 1). To obtain the final distribution

of seats, the total number of votes each party received in each constituency is divided by the

constituency’s regional quotient.2

B Mexico’s Cap on Over-Representation, Threshold,

and Rules About Publicly-Provided Campaign Funds

Here, we explain three pertinent features of Mexico’s MMM system. First, the Mexican Consti-

tution establishes two ‘caps’ to avoid over-representation: 1) no party can claim more than 300

Deputies, elected in either tier, even if it captures more than 52% of votes; and 2) no party’s

percentage of Deputies is allowed to exceed by more than 8% the percentage of votes the party

obtained in an election. There is one exception to this, which we explain below, but before doing

so, we note that in this latter calculation, the number of PR votes won by the party are used.

This may be another reason why a majority-seeking party coordinating with a small party in

Mexico may prefer to keep PR votes to itself: by doing so, it can avoid triggering this cap on

over-representation.

The exception to this is parties that, by winning SSDs, obtain a percentage of seats that is

larger than the percentage of votes received plus 10. Let us imagine that a party won 200 SSDs,

but all by very tight margins, meaning that it obtained 30 per cent of all votes cast. In addition,

the party receives 70 seats by PR. This means that the party has won 54% of seats in the COD

(270 seats out of 500), but has only won 30% of the votes. This party is thus in violation of

the second cap. To qualify for this exception, the party’s percentage of seats won in the SSD

tier (in this example, 40%) cannot be larger than the percentage of all votes obtained (in this

example, 30%) plus 10 (meaning: 40%). The party in our example qualifies for the exception.

2Throughout the process, leftover seats are assigned by largest remainder.

2

In the event any party does exceed its permitted number of seats, these seats are distributed

among the remaining parties.

Second, Article 54 of the Mexican Constitution stipulates that a party must have registered

candidates in at least two thirds of the 300 SSDs and obtained at least 3% of all valid votes to

receive a PR seat.

Third, a 1996 reform established that of the campaign funds provided by the government,

70% would be distributed among political parties with a national registry according to their vote

shares in the previous COD election, with the remaining 30% being equally distributed amongst

all parties. In this calculation too, the number of votes the party wins in PR are used. This is

another reason why a majority-seeking party in Mexico may prefer to maximize PR votes.

C How PR Votes Cast for More Than One Coordi-

nating Party are Divided Up

Here we explain how votes cast for a joint candidate under more than one coordinating party’s

label translate into PR votes for those parties. First, what does casting one’s vote for a joint

candidate under more than one coordinating party’s label look like? If two coordinating parties

present a joint candidate, this means the joint candidate appears twice on the ballot, under the

names of both coordinating parties. As Montero (2015) explains, a voter can select the joint

candidate under one party’s label, the joint candidate under the other party’s label, or the joint

candidate under both party’s labels.

When a voter casts her ballot in this manner, one vote is added to the joint candidate’s

tally in the SSD race. Then, one vote is divvied up equally among the number of chosen parties

for the purposes of PR. For instance, let us say ten votes were cast for the PRI-PVEM joint-

candidate in SSD A. Four were cast under the PRI label, three under the PVEM label, and

three under both party’s labels. First, the joint candidate receives 10 votes in the SSD race.

Second, each coordinating party receives 100% of the votes cast under their label only (the PRI

receives four and the PVEM, three). Third, votes cast under more than one of the coordinating

parties’ labels are divided equally among the selected parties (three votes are divided among two

parties, totalling 1.5 votes for each party). When dividing the votes does not generate an integer

3

number (as in this case), the residuals are summed (0.5 + 0.5) and assigned to the coordinating

party that obtained the largest number of votes under its label only (in our example, the PRI,

producing 2 votes for the PRI and 1 for the PVEM). Finally, the total number of PR votes for

each coordinating party is given by adding the total number of votes each party got under its

label only to the ‘divided up votes’ (the PRI receives 6 votes (4 + 2) and the PVEM receives 4

(3 + 1) votes).

D Descriptive Statistics

Table D.1: Descriptive Statistics for the Case of Japan

Mean Standard Deviation Min Max ObservationsKomeito SSDm,t 0.017 0.130 0.000 1.000 10174∆Komeito PR VSm,t 0.420 1.908 -20.155 16.019 6404Negative ∆LDP PR VSm,t 0.202 4.311 -40.627 40.670 9733Log(Transfersm, t) -3.396 0.744 -7.922 1.954 9818Fiscal Strengthm, t 0.441 0.286 0.000 2.850 9764Log(Populationm, t) 9.783 1.418 5.142 15.966 10374Log(Incomem, t) 0.072 0.266 -1.064 1.647 9822Dependent Populationm, t 0.381 0.054 0.000 0.629 9117Agriculturem, t 0.062 0.057 0.000 0.625 9116Population Densitym, t 1108.222 2498.595 1.304 19315.560 10374

Table D.2: Descriptive Statistics for the Case of Mexico

Mean Standard Deviation Min Max ObservationsLog(Transfersm, t) 4.612 1.217 -2.055 9.433 2064Negative ∆PVEM -1.991 5.321 -53.284 16.897 4864or PVEM-PRI VSm,t

∆PRI VSm,t 1.146 8.542 -69.193 34.696 4864Log(Populationm, t) 9.439 1.581 4.466 14.419 4913Poverty Indexm, t -0.000 0.997 -2.296 5.030 4913Surface Aream, t 796.408 2092.160 2.200 53237.801 4909Population Densitym, t 289.348 1187.954 0.144 17206.451 4908Ruralm, t 0.151 0.358 0.000 1.000 4918State of Emergencym, t 0.193 0.394 0.000 1.000 4915

E Supplementary Analyses for Hypothesis I

In Table E.1, we present the full specification of the main paper’s Table 1.

4

Tab

leE

.1:

InL

DP

SS

Ds,

mu

nic

ipali

ties

that

incr

ease

dP

Rvo

tes

for

the

Kom

eito

wh

ile

dec

reas

ing

them

for

the

LD

Pin

the

200

3,20

05,

an

d20

12

HO

Rel

ecti

on

sw

ere

rew

ard

edw

ith

mor

em

oney

afte

rel

ecti

ons.

Mu

nic

ipal

itie

sth

at

dec

rease

dP

Rvo

tes

for

the

LD

P(w

ith

out

chan

ges

toK

omei

toP

Rvo

tesh

are)

wer

ep

enal

ized

wit

hle

ssm

oney

,w

hil

em

un

icip

alit

ies

that

incr

ease

dP

Rvo

tes

for

the

Kom

eito

(wit

hou

tch

an

ges

toL

DP

PR

vote

shar

e)w

ere

nei

ther

pen

aliz

edn

orre

war

ded

(fu

llsp

ecifi

cati

on

).

Dep

enden

tV

aria

ble

:L

og(T

ransf

ers m

,t+1)

Model

1M

odel

2M

odel

3M

odel

4∆

Kom

eito

PR

VSm,t∗

Neg

ativ

e∆

LD

PP

RV

Sm,t

0.00

249*

**0.

0021

1**

0.00

249*

**0.

0021

0***

[0.0

0093

][0

.000

81]

[0.0

0093

][0

.000

81]

∆K

omei

toP

RV

Sm,t

-0.0

0518

-0.0

0018

-0.0

0518

-0.0

0111

[0.0

0778

][0

.008

51]

[0.0

0808

][0

.008

59]

Neg

ativ

e∆

LD

PP

RV

Sm,t

-0.0

0996

**-0

.009

13**

-0.0

0997

**-0

.007

17[0

.003

99]

[0.0

0434

][0

.004

20]

[0.0

0497

]∆

LD

PSSD

VSm,t

-0.0

0001

0.00

162

[0.0

0203

][0

.002

10]

Log

(Tra

nsf

ers m

,t)

0.72

027*

**0.

4978

7***

0.72

027*

**0.

4976

2***

[0.0

2646

][0

.034

22]

[0.0

2647

][0

.034

09]

Fis

cal

Str

engt

hm,t

0.01

344

-0.4

3850

**0.

0134

4-0

.442

86**

[0.0

5553

][0

.177

05]

[0.0

5565

][0

.177

00]

Log

(Pop

ula

tion

m,t

)-0

.011

81-1

.074

79**

*-0

.011

81-1

.095

82**

*[0

.012

29]

[0.3

1366

][0

.012

25]

[0.3

1636

]L

og(I

nco

me m

,t)

0.16

477*

*-0

.550

25*

0.16

478*

*-0

.551

30*

[0.0

7829

][0

.314

66]

[0.0

7802

][0

.315

17]

Dep

enden

tP

opula

tion

m,t

1.26

937*

**-0

.323

271.

2693

9***

-0.3

2766

[0.4

7491

][1

.398

13]

[0.4

7512

][1

.396

29]

Agr

icult

ure

m,t

0.09

327

-0.6

2286

0.09

328

-0.6

4946

[0.2

7014

][1

.802

27]

[0.2

7057

][1

.802

05]

Pop

ula

tion

Den

sity

m,t

0.00

001

0.00

023

0.00

001

0.00

023

[0.0

0001

][0

.000

15]

[0.0

0001

][0

.000

15]

Con

stan

t-1

.822

02**

*9.

1497

6***

-1.8

2202

***

9.36

568*

**[0

.296

20]

[3.2

7514

][0

.296

12]

[3.3

0196

]SSD

-Yea

rF

EY

esY

esY

esY

esM

unic

ipal

ity

FE

Yes

Yes

Obse

rvat

ions

4,49

74,

497

4,49

74,

497

R-s

quar

ed0.

8079

20.

7588

10.

8079

20.

7588

8N

um

ber

ofm

unic

ipal

itie

s2,

293

2,29

3R

obust

stan

dar

der

rors

atth

edis

tric

tle

vel

inbra

cket

s**

*p<

0.01

,**

p<

0.05

,*

p<

0.1

5

LMH

-.04

-.02

0

.02

.04

Mar

gina

l Effe

ct o

f Neg

ativ

e ∆L

DP

PR V

S on

Log

(Tra

nsfe

rs)

-20 -10 0 10 20Moderator: ∆Komeito PR VS

Figure E.1: Here, we present the Hainmueller, Mummolo and Xu (2019) diagnostic for Model 4 in Table1. Results show the estimated marginal effects using both the linear interaction model and the binningestimator. The conditional marginal-effect estimates of the binning estimator line up very closely with thelinear interaction effect from the original model. The histogram at the bottom of the figure corroboratesthat there is sufficient common support across values of ∆Komeito PR VSm,t.

-.06

-.04

-.02

0.0

2.0

4Ef

fect

s on

Lin

ear P

redi

ctio

n

-10 0 10Negative ΔLDP PR VS

Average Marginal Effects of ΔKomeito PR VS with 95% CIs

Figure E.2: Average Marginal Effect of ∆Komeito PR VSm,t with 95% Confidence Intervals from Model 1in Table 1 at different levels of Negative ∆LDP PR VSm,t.

F Supplementary Analyses for Hypothesis II

In Table F.1, we present the full specification of the main paper’s Table 2.

6

Table F.1: In Komeito SSDs, municipalities that increased PR votes for the Komeito whiledecreasing them for the LDP in the 2003, 2005, and 2012 HOR elections were penalized withless money after elections. Municipalities that decreased PR votes for the LDP (without changesto Komeito PR vote share) were also penalized, while municipalities that increased PR votes forthe Komeito (without changes to LDP PR vote share) were neither penalized nor rewarded (fullspecification).

Dependent Variable: Log(Transfersm,t+1)Model 1

∆Komeito PR VSm,t∗ Negative ∆LDP PR VSm,t -0.008***[0.001]

∆Komeito PR VSm,t -0.007[0.010]

Negative ∆LDP PR VSm,t -0.057***[0.002]

Log(Transfersm,t) 0.530***[0.053]

Fiscal Strengthm, t -3.048**[0.925]

Log(Populationm, t) 0.208*[0.087]

Log(Incomem, t) 0.536*[0.212]

Dependent Populationm, t 0.277[0.660]

Agriculturem, t 0.305[0.940]

Population Densitym, t 0.000***[0.000]

Constant -2.663***[0.227]

Controls YesSSD-Year FE YesObservations 63R-squared 0.659

Robust standard errors at the district level in brackets*** p<0.01, ** p<0.05, * p<0.1

7

L M H

-.4

-.2

0

.2

Mar

gina

l Effe

ct o

f Neg

ativ

e ΔL

DP

PR V

S on

Log

(Tra

nsfe

rs)

-20 -10 0 10 20Moderator: ΔKomeito PR VS

Figure F.1: Here, we present the Hainmueller, Mummolo and Xu (2019) diagnostic for Model 1 in Table2. Results show the estimated marginal effects using both the linear interaction model and the binningestimator. Because the confidence intervals of the binning estimators overlap with the linear interactioneffect from the original model, we cannot reject the possibility that the conditional marginal effect estimatesof the binning estimator are equal to those from the linear interaction effect in the original model. Thehistogram at the bottom of the figure corroborates that there is sufficient common support across values of∆Komeito PR VSm,t.

-.15

-.1-.0

50

.05

.1Ef

fect

s on

Lin

ear P

redi

ctio

n

-10 0 10Negative ΔLDP PR VS

Average Marginal Effects of ΔKomeito PR VS with 95% CIs

Figure F.2: Average Marginal effect of ∆Komeito PR VSm,t with 95% Confidence Intervals from Model 1in Table 2 at different levels of Negative ∆LDP PR VSm,t.

G Placebo Tests

In Table G.1 we present the placebo test explained in the main paper, where we redo Table 1

with Negative ∆Non-LDP/Komeito PR VSm,t instead of Negative ∆LDP PR VSm,t.

8

Tab

leG

.1:

InL

DP

SSD

s,m

un

icip

alit

ies

that

incr

ease

dP

Rvo

tes

for

the

Kom

eito

wh

ile

dec

reas

ing

them

for

all

non

LD

Por

Kom

eito

part

ies

inth

e200

3,

2005,

and

2012

HO

Rel

ecti

ons

wer

en

otre

war

ded

wit

hm

ore

mon

eyaf

ter

the

elec

tion

.

Dep

enden

tV

aria

ble

:L

og(T

ransf

ers m

,t+1)

Model

1M

odel

2M

odel

3M

odel

4∆

Kom

eito

PR

VSm,t∗

Neg

ativ

e∆

Non

-LD

P/K

omei

toP

RV

Sm,t

-0.0

01-0

.001

-0.0

01-0

.001

[0.0

01]

[0.0

01]

[0.0

01]

[0.0

01]

∆K

omei

toP

RV

Sm,t

-0.0

030.

002

-0.0

030.

002

[0.0

07]

[0.0

09]

[0.0

07]

[0.0

09]

Neg

ativ

e∆

Non

-LD

P/K

omei

toP

RV

Sm,t

-0.0

010.

002

-0.0

020.

001

[0.0

04]

[0.0

05]

[0.0

04]

[0.0

05]

∆L

DP

SSD

VSm,t

0.00

30.

003

[0.0

02]

[0.0

02]

Log

(Tra

nsf

ers m

,t)

0.72

2***

0.49

9***

0.72

2***

0.49

8***

[0.0

26]

[0.0

34]

[0.0

26]

[0.0

34]

Fis

cal

Str

engt

hm,t

0.02

0-0

.450

**0.

018

-0.4

54**

[0.0

55]

[0.1

78]

[0.0

56]

[0.1

78]

Log

(Pop

ula

tion

m,t

)-0

.013

-1.2

77**

*-0

.013

-1.2

26**

*[0

.012

][0

.306

][0

.012

][0

.305

]L

og(I

nco

me m

,t)

0.19

0**

-0.4

850.

182*

*-0

.506

[0.0

78]

[0.3

19]

[0.0

77]

[0.3

19]

Dep

enden

tP

opula

tion

m,t

1.27

9***

-0.5

941.

271*

**-0

.437

[0.4

69]

[1.4

10]

[0.4

70]

[1.4

02]

Agr

icult

ure

m,t

0.05

8-0

.696

0.06

3-0

.768

[0.2

76]

[1.8

02]

[0.2

74]

[1.8

07]

Pop

ula

tion

Den

sity

m,t

0.00

00.

000*

0.00

00.

000*

[0.0

00]

[0.0

00]

[0.0

00]

[0.0

00]

Con

stan

t-1

.815

***

11.1

69**

*-1

.806

***

10.6

06**

*[0

.294

][3

.226

][0

.296

][3

.214

]SSD

-Yea

rF

EY

esY

esY

esY

esM

unic

ipal

ity

FE

Yes

Yes

Obse

rvat

ions

4,49

74,

497

4,49

74,

497

R-s

quar

ed0.

807

0.75

80.

807

0.75

8N

um

ber

ofm

unic

ipal

itie

s2,

293

2,29

3R

obust

stan

dar

der

rors

atth

edis

tric

tle

vel

inbra

cket

s**

*p<

0.01

,**

p<

0.05

,*

p<

0.1

9

H Construction of Independent Variables for 2012

As the main paper explains, coordinating parties were required to present joint lists in the 2006

election, meaning that only the number of votes cast for the PRI-PVEM coalition is observed

in this election. The coalition would thus not have been able to distinguish how many votes, of

this total, were cast for either partner. In 2012, then, how might a coalition go about verifying

whether party supporters complied with instructions to split their votes? We posit that it is

likely to have used the results of the 2009 election to calculate the ratio of votes contributed

by each coordinating partner, and apply that ratio to the total number of votes received by the

coalition in 2006.

More concretely, let us say that in municipality m, the PRI-PVEM coalition received 10

votes in 2006 and 15 in 20093, out of which 9 were cast under the PRI label and 6 under the

PVEM label. Accordingly, the share of votes that the PRI contributed to the coalition’s total

in 2009 corresponds to 9 / (9 + 6) = 0.6 or 60%. Similarly, the share of votes that the PVEM

contributed to the coalition’s total in 2009 corresponds to 6 / (9 + 6) = 0.4 or 40%. To calculate

the number of votes that the coalition might expect corresponded to each partner in the 2006

election, we apply these 2009 ratios to the total number of votes received by the PRI-PVEM

coalition in 2006. By this calculation, of the 10 votes the coalition received in 2006, the PRI

was responsible for 6 of these (10 * 0.6) and the PVEM 4 votes (10 * 0.4).

I Supplementary Analyses for Hypothesis III

In Table I.1, we present the main paper’s full specification of Table 3.

3We exclude votes cast under both party’s labels to calculate these ratios.

10

Tab

leI.

1:

Inal

lian

ceS

SD

sw

ith

PV

EM

-affi

liat

edca

nd

idat

es,

mu

nic

ipal

itie

sth

atin

crea

sed

vote

sfo

rth

isca

nd

idat

eu

nd

erth

eP

RI

lab

elon

lyw

hil

ed

ecre

asi

ng

them

un

der

the

PV

EM

orb

oth

par

ties

’la

bel

sw

ere

rew

ard

edw

ith

mor

em

oney

afte

rth

e20

12an

d20

15el

ecti

ons

(Mod

el1).

Inall

ian

ceS

SD

sw

ith

PR

I-affi

liat

edca

nd

idat

es(M

od

el2)

and

non

-all

ian

ceS

SD

s(M

od

el3)

,m

un

icip

ali

ties

that

incr

ease

dvo

tes

for

the

PR

I(o

nly

,in

the

case

ofM

od

el2)

are

rew

ard

ed(f

ull

spec

ifica

tion

).

Dep

enden

tV

aria

ble

:L

og(T

ransf

ers m

,t+1)

Model

1M

odel

2M

odel

3Joi

nt

Can

did

ate

(PV

EM

)Joi

nt

Can

did

ate

(PR

I)N

on-A

llia

nce

Dis

tric

t∆

PR

IV

Sm,t∗

Neg

ativ

e∆

PV

EM

orP

VE

M-P

RI

VSm,t

0.00

214*

0.00

000

[0.0

0117

][0

.000

51]

∆P

RI

VSm,t

0.01

261

0.01

169*

0.01

760*

*[0

.018

49]

[0.0

0618

][0

.007

55]

Neg

ativ

e∆

PV

EM

orP

VE

M-P

RI

VSm,t

0.02

680*

0.01

310

[0.0

1370

][0

.008

48]

∆P

RI

VSm,t∗

Neg

ativ

e∆

PV

EM

VSm,t

-0.0

0037

[0.0

0214

]N

egat

ive

∆P

VE

MV

Sm,t

0.01

483

[0.0

1942

]L

og(P

opula

tion

m,t

)-0

.571

12**

*-0

.595

65**

*-0

.767

25**

*[0

.085

55]

[0.0

3845

][0

.040

72]

Pop

ula

tion

Den

sity

m,t

0.00

038

-0.0

0004

0.00

013

[0.0

0049

][0

.000

11]

[0.0

0014

]R

ura

l m,t

0.70

873*

**0.

2649

6-0

.023

32[0

.199

87]

[0.1

6927

][0

.124

53]

Pov

erty

Index

m,t

-0.1

9277

-0.0

7279

-0.0

6650

[0.1

3466

][0

.072

98]

[0.0

8012

]Surf

ace

Are

a m,t

-0.0

0000

0.00

001

0.00

001

[0.0

0002

][0

.000

01]

[0.0

0002

]Sta

teof

Em

erge

ncy

m,t

-0.0

5235

0.09

994

0.16

675

[0.1

5587

][0

.100

39]

[0.1

7202

]C

onst

ant

10.2

1101

***

11.6

2302

***

12.3

4533

***

[0.8

3591

][0

.419

86]

[0.4

1195

]SSD

-Yea

rF

EY

esY

esY

esO

bse

rvat

ions

170

1,13

563

5R

-squar

ed0.

6710

10.

6254

20.

7278

0R

obust

stan

dar

der

rors

clust

ered

atth

edis

tric

tle

vel

inbra

cket

s**

*p<

0.01

,**

p<

0.05

,*

p<

0.1

11

L M H

-.1

-.05

0

.05

.1

.15

Mar

gina

l Effe

ct o

f Neg

ativ

e (∆

PVEM

or P

VEM

-PR

I VS)

on

Log(

Tran

sfer

s)

-20 -10 0 10 20Moderator: ∆PRI VS

Figure I.1: Here, we present the Hainmueller, Mummolo and Xu (2019) diagnostic for Model 1 in Table3. Results show the estimated marginal effects using both the linear interaction model and the binningestimator. Because the confidence intervals of the binning estimators overlap with the linear interactioneffect from the original model, we cannot reject the possibility that the conditional marginal-effect estimatesof the binning estimator are equal to those from the linear interaction effect from the original model. Thehistogram at the bottom of the figure corroborates that there is sufficient common support across values of∆PRIm,t.

-.05

0.0

5.1

Effe

cts

on L

inea

r Pre

dict

ion

-10 0 10∆PRI VS

Average Marginal Effects of Negative (∆PVEM or PVEM-PRI VS) with 95% CIs

Figure I.2: Average Marginal effect of Negative (∆PVEM or PVEM-PRI VSm,t) with 95% Confidence In-tervals from Model 1 in Table 3 at different levels of ∆PRI VSm,t.

12