nationalization and localism in electoral systems and ... · nupps - mostly caio, lucas, nina,...
TRANSCRIPT
Universidade de Sao Paulo
Faculdade de Filosofia, Letras e Ciencias Humanas
FABRICIO VASSELAI
Nationalization and localism in
electoral systems and party
systems
(corrected version / versao corrigida)
Sao Paulo
2015
Universidade de Sao Paulo
Faculdade de Filosofia, Letras e Ciencias Humanas
FABRICIO VASSELAI
Nationalization and localism
in electoral systems and party
systems
(corrected version / versao corrigida)
Tese de Doutorado apresentada a Faculdade de
Filosofia, Letras e Ciencias Humanas da Univer-
sidade de Sao Paulo, para a obtencao do tıtulo
de Doutor em Ciencia Polıtica.
Orientador: Prof. Dr. Fernando de Magalhaes
Papaterra Limongi
Sao Paulo
2015
i
“when representatives ceased to be (. . . ) mere proxies for indi-
viduals, the communities they represented were geographically
defined (. . . ) It was possible to stretch the fiction of one man
standing for another or for several others to the point where
he stood for a whole local community” (p.41).
“by the seventeenth century the local geographical definition
of representation had become an essential ingredient in it, just
as representation had itself become” (p.43)
Edmund Morgan, in Inventing the People - The Rise of Pop-
ular Sovereignty in England and America.
ii
Acknowledgments
From all cliches one can find in an academic dissertation, no one is more truthful
than a simple phrase present in almost all papers, books and theses. I am talking
about that sentence where authors thank others for the remaining qualities of their
work and assume the guilty for the flaws that still last. Indeed, our works would be
much worse without all course corrections, welcome additions and support brought
by people every now and then. Not to mention they would also be better weren’t
us too stubborn to listen even more to what the community has said...
No one else could deserve more such credits in helping me polishing my thesis
than my advisor, Fernando Limongi. Above his well known intellectual qualities,
I feel fortunate to have received Limongi’s confidence in what I was trying to do.
Which means thanking him first for what he never said. Don’t get me wrong, he
(very) often questioned me whether something I wrote was in the right direction or
simply pointed my (frequent) mistakes. But he never, not only once, questioned (at
least not to me!) whether I would be capable of delivering some of the ambitious
things I was trying to cook for this work. He let me go on, always allowing (and
encouraging) me to try to reach further, never holding me back. He was very
supportive of no matter the crazy thing I had in mind for testing, or of my many
(expensive) efforts in trying to make this thesis dialogue with the international
debate. For those who think such an attitude of him is a triviality, I have to say:
you can only be one of the rare readers (hi mom!) who are not in our business.
Still, the many things Limongi has said were very, very important too. His ad-
vices are part of most of the rights one may find in the remainder of this dissertation.
iii
As well as they are responsible for me leaving out quite a few wrongs. Some of my
(few) ideas were indeed born and raised during the dozens of times I had to explain
them to him again and again until they magically got clear (to him and to myself).
Besides, Limongi has the better of the intellectual qualities: he knows how to think.
He reads or listens an idea and can dissemble it in its smaller logical pieces, which
is so helpful! Thank you for that, for the long-term bet on me, for the thousand
signatures and for not kicking me out whenever I found yet another parallel task to
do. I will miss the easygoingness and the short (and wrong) emails. You’re forgiven
for the missed appointments due to the Champions League matches.
But science is not the project neither of an individual, nor of a duet. During
this thesis’ journey, I have received ideas, suggestions, criticism, funding, logistics,
data and so much more and from so many, that it’d be impossible to nominally
thank all of them. I am particularly in debt with professors Ken Kollman and Allen
Hicken, from the University of Michigan, for the life-changing opportunities they
gave me at the CLEA project, for receiving me as a visiting researcher, for the
joint projects and, above all, for the ideas to some of the chapters in this thesis. I
also have to acknowledge contributions made by professors Anckar and Karvonen,
from the Abo Akademi university, in Finland, where they also kindly received me as
visitor. Thanks Coimbra Group of European Universities for the award that made
such a visit possible. I also have to acknowledge comments to an earlier version of
my third chapter, made by professor Roger Bivand for a course on Spatial Analysis
that I took at the Norwegian School of Economics. Lastly, I also thank Juan March
Foundation for receiving me as a visitor, which was when this research was originally
(happily) conceived. Thanks Fundacion Carolina for the funding to go there. Above
all research institutions, I also thank FAPESP for the doctorate scholarship and for
iv
the extra scholarship to visit the U. of Michigan, for the funds to go to ICPSR and
to many congresses. The role FAPESP plays in our science is commendable.
Additional contributions to the three chapters of this dissertation also came from
scholars along the last years, during conferences, classes or friendly talks. I would like
to acknowledge, specifically, some ideas given by Gary Cox, Scott Morgenstern, Ed-
uardo Alleman, Octavio Amorim Neto, Paolo Ricci, Sandra Marquart-Pyat, Rogerio
Arantes, Jose Antonio Cheibub, Daniele Caramani, David Armstrong and George
Avelino. I also profited from the debates I had on parts of my work with professors
Sergio Praca, Glauco Peres, Lorena Barberia, Fernando Guarnieri and Jose Alvaro
Moises. By the way, a special thanks go to professor Moises, for always listening my
ideas and for opening so many opportunities to me. Special thanks also to professor
Rogerio Arantes for all the good he has done to our department, for giving the best
class I took and, above all, for being who leads us tricolores to the SPFC matches.
Many colleagues also contributed in reading earlier versions of this work or in
debating its questions. I feel particularly thankful to the substantive or method-
ological debates related to this dissertation I had with Umberto Mignozzetti, Diogo
Ferrari, Jill Wittrock, Inga Anna-Liisa Saikkonen, Mattias Karlsson, among many
others. Above all, I am thankful to Marcos Paulo de Lucca-Silveira and Sergio Si-
moni Jr., who carefully read and commented my papers. These two are among my
dearest interlocutors and they are some of the brightest bets of this department,
that’s for sure. I particularly thank Mr. Simoni for his vast knowledge of the liter-
ature (this guy likes to read!), smart insights, and for his willingness to jump in all
crazy quijotesc tasks we found in the department. Marcos “little friend” Paulo is
one of the most intelligent guys I know and is my best buddy in academia. I have
to thank him for too many things, but in the top 2, I thank him for our real team
v
play, and for how his vivid willingness to talk about good research often saves my
mood about academia when I am almost giving up on it. In the top 1, I thank him
for our acid academic facebook chats. These are cool.
It is also fair to acknowledge Amanda Campanini, Andre Kaysel, Andreza Da-
vidian, Danilo Medeiros, Camila Rocha, Carol Requena, Fabio Lacerda, Guilherme
Duarte, Lincoln Noronha, Lucas Petroni, Maira Rodrigues, Marcos again, Miguel
Barrientos, Paulo Flores, Raissa Wihby, Rafael Magalhaes, Rafael Moreira, Roberta
Soromenho, Samuel Godoy, San Assumpcao, Sergio again, Telma Hoyler, Thiago
Silva, Tiago Borges, Vıtor Oliveira, the collaborators of Leviathan, people from
NUPPS - mostly Caio, Lucas, Nina, Kayli, Vanessa and Vera - people from NECI,
among others, for giving me ideas, inspiring me, listening me talk or simply for part-
nering in many efforts to contribute to our department and to our Political Science.
These people are so generous! I will, of course, miss specially the bar crawling.
Talking about best buddies, I have to tell San I am grateful for all the friendship
and kindness she always gave me. I am happy she’s pursuing new challenges, but I
have to say she is sorely missed here.
At the same time, nothing would have been possible without the crucial support
given by my department’s administrative staff. Marcia, Rai and Vasne saved my
neck so many times that a book telling these stories would be bigger (also funnier)
than this small thesis. I thank you all so very much for always being there for me,
for the invariable good mood, for letting me watch soccer in any TV available. I will
miss a lot being there with you, not letting you work just to babble about something
that certainly should not be talked there if I had any good judgment left.
Outside the academic world (a.k.a. in real life), I was just as lucky. I had so
many people rooting for me. I have to thank my friends Leandro, Murilo and Isa for
vi
not getting too mad with my absences and for always being there when I needed.
I thank Rodrigo Abu and Cida for always being so nice to me when receiving me
back to Sao Paulo. Also, a wholeheartedly thanks go to Paula Rondinelli, for all she
has done and taught me over the years. I own her so much, and I will never forget
it. To my mom Analice and my bro Rennan, I have to thank for too many things
for one single thesis. She is responsible for what I am; and he, for what I want to
be. In a deeper sense, all I do is for them. Love you, buddies.
But the person I have to thank the most for being there for me and with me,
is Vanessa Vieira. How lucky I am to have met her! That’s by far the smartest
thing I’ve done during this Phd. This thesis was written with her support, love and
patience. I lost count of how many times she just stayed by my side, doing anything
else while I was hitting hard the keys of my computer’s keyboard. Always happy just
for being together! How is it even possible, don’t ask me. Vanessa often reminded
me that my realism should be balanced by always dreaming with impossible things.
She ruined my always rational barrier and in fact suddenly gave me much to dream
about. Since Vanessa is always proud of me anyways, what I hope with this thesis
done is that she is happy I can have back a bit of life for the two of us to share.
vii
Abstract
This research offers 3 independent studies on the questions of what is party na-
tionalization, how nationalization, regionalization or localism are affected by and
affect the electoral systems and the party systems. More specifically, in the 1st
chapter a new theoretical definition of party and party system nationalization is
presented, dividing such concept into four dimensions - the nationalization of party
organization, of the electoral supply, of the electoral demand and of the electoral
outcome. After that, such a theoretical framework is applied to the Brazilian case
to demonstrate how, in fact, more conceptual precision can alter empirical readings
about a given party system. The 2nd chapter explores one of the consequences of
party system nationalization, which literature has theorized but never tested di-
rectly. Namely, the idea that party nationalization would be what puts the electoral
circumscriptions together and what makes Duvergerian propositions move from the
local to the national level. To test that, party system nationalization is included for
the first time in a model of effective number of parties, after handling endogeneity
problems that have prevented scholars from doing the same. With such inclusion, it
will be proven and demonstrated that omitting party nationalization from models
of number of parties, which is a common practice, incurs in omitted variable bias.
In fact, such correct inclusion of party nationalization trough a system of simulta-
neous equations corrects that bias, altering some of the canonical interpretations
about party system fragmentation. Lastly, in the 3rd chapter I reevaluate the com-
mon idea that electoral systems with personal voting would lead to geographical
concentration (i.e. localization) of candidates’ electoral support. I offer a theoretical
discussion and then empirical evidence that such territorial pattern is not the rule of
what happens for instance in open-list PR. Besides, both concentrating and spread-
ing votes are electorally profitable results and very few candidates achieve levels of
concentration that predicts effective increases in the odds of being elected.
Keywords: party system, electoral system, nationalization, localism, territorial dis-
tribution
viii
Resumo
Esta pesquisa traz 3 estudos independentes, sobre temas ligados as questoes do
que e nacionalizacao partidaria e como nacionalizacao, regionalizacao e localismo
sao afetados por e afetam os sistemas eleitorais e partidarios. Mais especificamente,
no capıtulo 1 proponho uma nova definicao teorica de nacionalizacao dos partidos e
sistemas partidarios. Argumento que tal conceito pode ser dividido em 4 dimensoes,
que sao a nacionalizacao da organizacao partidaria, da oferta eleitoral, da demanda
eleitoral e dos resultados eleitorais. Em seguida, aplico esse quadro teorico ao caso
brasileiro para mostrar como, de fato, maior precisao conceitual altera a leitura
empırica que se faz de um sistema. No capıtulo 2, exploro uma das consequencias
da nacionalizacao partidaria, que vem sendo teorizada pela literatura mas nunca
testada de modo direto. Trata-se da ideia de que nacionalizacao seria o que conecta
as circunscricoes eleitorais e faz as proposicoes de Duverger passarem do nıvel local
ao nacional. Para testar isso, incluirei nacionalizacao dos sistemas partidarios pela
primeira vez num modelo de numero de partidos - apos lidar com problemas de endo-
geneidade que vem impedindo autores de fazerem isso. Assim, sera possıvel provar
e demonstrar que a nao inclusao de nacionalizacao vem causando vies de variavel
omitida nos modelos da literatura. Quando esse e corrigo, atraves da inclusao de
nacionalizacao por um sistema de equacoes simultaneas, altera-se algumas das inter-
pretacoes canonicas sobre a fragmentacao partidaria. Por fim, no capıtulo 3 reavalio
a ideia comum de que sistemas eleitorais com voto pessoal levariam candidatos a
ter apoio eleitoral geograficamente concentrado, portanto localista. Ofereco uma
discussao teorica e evidencias de que tal padrao territorial nao e a regra do que
vem ocorrendo, por exemplo, em sistemas de lista aberta. Alem disso, tanto con-
centrar votos como espalha-los vem dando dividendos eleitorais e poucos candidatos
conseguem atingir patamares altos de concentracao, a um nıvel que prediga real
aumento nas chances de eleicao.
Palavras-chave: sistemas partidarios, sistemas eleitorais, nacionalizacao, localismo,
distribuicao territorial
ix
Contents
List of Figures xii
List of Tables xv
Introduction 1
I.1 The subject of nationalization: general theoretical framework . . . . . 2
I.2 Road-map of this work . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1 The multiple dimensions of party nationalization - an application
to the Brazilian case 13
1.1 Multiple nationalizations . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 The debate about nationalization in Brazil . . . . . . . . . . . . . . . 27
1.2.1 Nationalization of organization . . . . . . . . . . . . . . . . . 34
1.2.2 Nationalization of the supply . . . . . . . . . . . . . . . . . . 39
1.2.3 Nationalization of the demand . . . . . . . . . . . . . . . . . . 42
1.2.4 Nationalization of the outcomes . . . . . . . . . . . . . . . . . 47
1.2.5 Overall scenario . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.3 Nationalization and fragmentation of party systems in Brazil . . . . . 50
1.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.5 Annexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Annex 1.A – Formula of Bochsler’s index of Party and Party system
nationalization . . . . . . . . . . . . . . . . . . . . . . . . . . 60
x
Annex 1.B – Parties’ organization: % of municipalities where par-
ties have presented candidates for the municipal Executive or
Legislative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2 Party system nationalization and the national effective number of
electoral parties 64
2.1 The omitted party system nationalization . . . . . . . . . . . . . . . . 66
2.2 The endogeneity obstacles for including party system nationalization 72
2.3 Description of dataset and of variables . . . . . . . . . . . . . . . . . 77
2.3.1 Endogenous variables . . . . . . . . . . . . . . . . . . . . . . . 79
2.3.2 Exogenous shared variable of interest . . . . . . . . . . . . . . 80
2.3.3 Exogenous exclusive (instrumental) variables . . . . . . . . . . 81
2.3.4 Other exogenous control variables . . . . . . . . . . . . . . . . 83
2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.6 Annexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Annex 2.A - Proof of endogeneity due to omitted variable bias in
models of ENEPnat that do not include PtyNat . . . . . . . . 98
Annex 2.B - Proof of endogeneity in models of ENEPnat that include
PtyNat, if they do not explicitly model the possible reciprocal
causation between these variables . . . . . . . . . . . . . . . . 101
Annex 2.C – List of countries’ elections per tier that are in the data . 104
Annex 2.D – Distribution of variables used in the main models of this
chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Annex 2.E - Measure sensitivity of the main results for the effects
between ENEPnat and PtyNat . . . . . . . . . . . . . . . . . 107
xi
3 Personal voting and localism: geographical distribution of electoral
support under open list PR systems 111
3.1 Does voting for a candidate mean voting for a local constituency service?113
3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.3 Spatial assessment: one needs parishes to be parochial . . . . . . . . 125
3.3.1 Geographical clustering: Moran‘s I through an Empirical Bayes
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.3.2 The impact of cities with big population: Local G index . . . 132
3.3.3 Territorial homogeneity of electoral support: Spatial GINI index135
3.4 Statistical models: does electoral localism pay off? . . . . . . . . . . . 140
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
3.6 Annexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Annex 3.A – Distribution of EBMoran’s I according to additional
specifications of the neighbor matrix WI×J . . . . . . . . . . . 151
Annex 3.B – Distribution of variables used in the main models of this
chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
References 153
xii
List of Figures
I.1 Frequency of expressions “nationalization” and “party nationaliza-
tion” over frequency of “politics”, in aprox.4.5 million books pub-
lished in English . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1 Venn diagram of the 4 dimensions of party and party system nation-
alization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.2 Nationaliz. of organization: homogeneity of % membership across
elec. circumscriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.3 Presence of parties’ organization across the Brazilian municipalities . 37
1.4 Nationalization of the supply: number of elec. circumscriptions where
parties offered candidates . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.5 Nationalization of the demand: elec. support homogeneity across
elec. circumscriptions (measured with Bochsler’s (2010) PNSsw . . . 44
1.6 Party system nationalization of demand among countries . . . . . . . 46
1.7 Nationalization of the outcomes: number of elec.circumscriptions where
parties won seats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.8 Effective num. of parties at the Lower Chamber - Brazil vs. World
and within Brazil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.9 Effec.num.of legislative parties at the Lower Chamber, according to
simulated perfect nationalization of the electoral demand . . . . . . . 56
Annex 1.B – Parties’ organization: % of municipalities where parties have
presented candidates for the municipal Executive or Legislative . . . . 63
xiii
2.1 Party system nationalization (PtyNat) with bootstrapped measure
uncertainty and Effective number of electoral parties (ENEPnat), per
country-tiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.2 Path diagrams of the models 1 to 4 . . . . . . . . . . . . . . . . . . . 85
2.3 Coefficient of the direct effects of SocialDiversity in 19 replicated ver-
sions of previous Model 3, each with a different measure for SocialDiversity 93
2.4 Conditional direct effects of Social Cleavages on the effective number
of parties (ENEPnat), in 19 replication versions of Model 5, each with
a different measure for SocialDiversity . . . . . . . . . . . . . . . . . 94
Annex 2.D – Distribution of variables used in the main models of this chapter106
Annex 2.E.1 - Changes in the direct effects from ENEPnat to PtyNat and
from PtyNat to ENEPnat according to the reliability assumed for
PtyNat as a measure of latent Nationalization . . . . . . . . . . . . . 108
3.1 Distribution of the geographical concentration of electoral support
(measured by EBMoran’s I) of all candidates . . . . . . . . . . . . . . 129
3.2 Concentration of population in each city and the average concentra-
tion of candidates’ electoral support in that city (using local-G) . . . 133
3.3 Logical outcomes of territorial distribution and how EB-Moran is ex-
pected to score* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.4 Distribution of the geographical homogeneity of electoral support
(measured by spGINI of all candidates . . . . . . . . . . . . . . . . . 139
3.5 Median cross-level predicted probabilities of being elected, according
to the level of geographical concentration of votes . . . . . . . . . . . 146
3.6 Random effects at the electoral circumscriptions - contribution of the
varying slopes of EBMoran’s I to its overall slope . . . . . . . . . . . 148
xiv
Annex 3.A – Distribution of EBMoran’s I according to additional specifi-
cations of the neighbor matrix WI×J . . . . . . . . . . . . . . . . . . . 151
Annex 3.B – Distribution of variables used in the main models of this chapter152
xv
List of Tables
2.1 Simultaneous Eq. Models of the reciprocal relationship between party
system nationalization (PtyNat) and national effective number of
electoral parties (ENEPnat) . . . . . . . . . . . . . . . . . . . . . . . 87
2.2 Indirect effects to effective number of electoral parties (ENEPnat)
and to party ystem nationalization (PtyNat) in previous Model 3 . . 92
Annex 2.C – List of countries’ elections per tier that are in the data . . . . 104
3.1 Bayesian multilevel logistic models of the odds of being elected . . . . 143
1
Introduction
This thesis presents three different studies on the relationship between elections,
party systems and the territorial distribution of politics. In specific, it is particularly
interested in how parties and candidates organize and compete across the territory
of countries. As well as in how the subsequent geographical distribution of voters’
electoral demand affects the electoral outcomes and the party systems that emerge.
It begins by offering, in chapter 1, a new theoretical definition of party and party
system nationalization. The concept is divided into four dimensions - nationaliza-
tion of party organization, of the electoral supply, of the electoral (realized) demand
and of the electoral outcome - with an application of such conceptual detailing to
the Brazilian case. Chapter 2 explores one of the consequences of party system
nationalization, that has never been tested directly. Namely, the idea that it would
be what makes Duvergerian propositions move from the local to the national level.
To test that, party system nationalization is included for the first time in a com-
parative model (with data on 62 countries) of national effective number of parties,
altering some of the canonical interpretations about party system fragmentation.
Lastly, chapter 3, with data on 5 Open List countries, questions the common idea
that personal voting would lead to geographical concentration (i.e. localization) of
candidates’ electoral support. More details will be given shortly, in the last section
of this introduction.
2
I.1 The subject of nationalization: general theo-
retical framework
In a more general tone, this thesis has, as its framework, what Pitkin (1967)
has called the local-national ‘classical dilemma’, with which, according to her, Po-
litical Science has actually been always concerned. The fact that the question of
local-national politics indeed permeated the development of modern politics has
to do, in the first place, with a historical dimension. The birth and expansion of
representative government were somewhat contemporary to the very formation of
national States as such. In 1896, John Commons observed that, on the one hand,
in England and in the United States “the original object which produced represen-
tative assemblies was nationalisation” (1896:11-17). While, reciprocally, when the
principle of representation was discovered “it permitted the unity of a nation, while
preserving the freedom of the localities” (idem). In the meantime, while national
units emerged, representation transformed.
Edmund Morgan (1988) points out that one of the first steps in the quest of
representative government was precisely the passage from representation of persons
to representation of places. For him, “in the Anglo-American world, when repre-
sentatives ceased to be, where they had ever been, mere proxies for individuals, the
communities they represented were geographically defined (. . . ) It was possible to
stretch the fiction of one man standing for another or for several others to the point
where he stood for a whole local community” (p:41). And, more specifically, “in
England and America at least, the community was geographically defined” (idem).
Historically, it was not, of course, only the case of England and the US. Besides,
many currently democratic countries, not to say most of them, began their pre-
3
democratic electoral history with electoral systems based on local districts just like
those two (see Boix 1999; Grofman and Lijphart 2003; Colomer, 2005; Calvo, 2009).
But if geography is in the very basis of representation, Agnew (1987) famously
argued that, for a long period, the empirical political literature would have neglected
attention to what we could call the location of politics. According to him, scholars
had treated the interpretation of political phenomena as if their locale and location
were irrelevant for the broader political happenings. Or, as Taylor and Johnston
(1979) put, at least as if they were irrelevant for the electoral universe of politics. In
fact, beyond some sparse intuitions on the possibility of political and social cleavages
being geographically demarcated (e.g. Rose and Urwin, 1975; Riker, 1982; Sartori,
1976; Kim and Ohn, 1992), most earlier comparative electoral studies disregarded
the importance of the geographical realization of elections. In reality, over time the
effects of electoral institutions became increasingly understood in a functionalist
manner, disregarding two important things. First, crucial aspects of the functioning
of these institutions have been theorized for the local levels (e.g. Duverger, 1954;
Cox, 1997). Second, specific territorial patterns can affect electoral and legislative
strategies and outcomes - therefore, altering the very outcomes of institutions.
Place matters. Parties organize themselves and compete with each other in the
territory, as well as elections are held in specific places. Parties, candidates, the
electorate, and the conformation of the electoral support - they can all have multi-
ple territorial faces and shapes, from local to regional, and then even national. That
is true both regarding the electoral and the legislative arenas. Both can be signifi-
cantly altered by the territorial formation of politics. Probably the most well-known
theory that explored such a possibility, was the wide-spread debate on distributivist
politics. The postulate of Edmund Burke that the locally elected politicians in the
4
Westminster should be free to not have to behave in a localist fashion, was brought
to the American scene. But the usual interpretations were not optimistic, and the
US representatives became classically seen as prone to localism, pork barrel and
constituency services (see Mayhew, 1974; Fenno, 1978; Cain et. al., 1987). It is
no surprise that, after it, plenty of studies have appeared exporting the idea even
further, and trying to diagnose, criticize and maybe redress localisms in different
countries and different electoral systems (e.g. Mainwaring, 1991, 1999; Ames, 1995,
2001; Samuels, 2001; Crisp and Ingall, 2002; Shugart et. al, 2005; Golden and
Picci, 2008; Allen, 2010; Tavits, 2010). After all, “clientelistic politics is seen as
the ‘natural’ outcome of place-based, communal and traditional values” (Agnew,
1987:65).
In a word, in the long run it seems Burke has more than won the normative
dispute and representatives are often seem as they should be national-oriented. More
than that, almost everything in politics which is (or is supposed to be) local, became
increasingly seen as linked to the idea of parochialism (e.g. Mainwaring, 1991, 1999;
Samuels, 2001). One of the reasons for it is that nationalization of politics has been
often seen as either the natural course of democratic and structured systems (e.g.
Lipset and Rokkan, 1967; Sartori, 1968), or at least as the empirical achievement
that allowed the emergence of modern party competition (Caramani, 1996, 2003,
2004). Likewise, for Claggett et. al. (1987:77), the nationalization of American
politics was “consistent with the decline of parochialism and localism under the
onrush of modernization”. After all, that process of nationalization frequently meant
“political alignments have crystallized around national social cleavages to produce
national patterns of political mobilization and partisanship. In this model of political
modernization, geographical political cleavages are given (. . . ) primordial status”
5
(1987:80). Which means to say they are seen as an unwelcome heritage from the
underdeveloped past.
This might help in understanding why the broader comparative study of the
territorial distribution of parties, candidates and electoral support remained under-
developed for quite a while. Probably, one of the main reasons is that nationalization
as a research object became a victim of the perceived success of nationalization, the
real process, in the developed democracies. Agnew (1987) noticed it when he pro-
posed in the 80s that, paradoxically, the reason for why these issues became less
prominent in the political sociology debate was precisely because the “declining
significance of place or locality” (p.3) has been seen as a given of the process of
modernization and of democratic consolidation.
In fact, until more recently, the question of whether nationalization indeed was
achieved, and then overcame the local or regional incentives, was not often explored.
The great exception was certainly the American case. Probably because, as men-
tioned, it was frequently cause of concern due to the local nature of the functioning
of its electoral system. After Schattschneider (1960) and Stokes (1967) both claimed
that the American politics had become quite nationalized after a long period of re-
gionalism, an intense debate was enacted. While many followed that interpretation
(Converse, 1972; Sundquist, 1973; Sorauf, 1980; Polsby, 1981), others offered evi-
dence on the contrary, it is, that nationalization was not exactly increasing (Mann,
1978; Pomper, 1980; Claggett et. al., 1984).
Still, even in what regards the US, the controversy did not survive the first half of
the 80s. Also, as mentioned, such modern debate on nationalization of the elections
and of parties, rarely reached other countries. Certainly never with intensity similar
6
to that of the American earlier debate1. However, since the beginning of the 2000s,
but specially in the last decade, that scenario seems to have been changing quickly.
Some of the main works on the subject have been published in that period (e.g.
Caramani, 1996, 2003, 2004; Chhibber and Kollman, 1998, 2004; Morgenstern et.
al., 2009). There clearly is a renewed interest in the theme of nationalization of
electoral politics. Even more so, for the party and party system nationalization -
which is the most wide spread contemporary approach on dealing with the local-
national electoral question.
Figure I.1: Frequency of expressions “nationalization” and “party
nationalization” over “politics”, in aprox. 4.5 million books published in English
1900
1902
1904
1906
1908
1910
1912
1914
1916
1918
1920
1922
1924
1926
1928
1930
1932
1934
1936
1938
1940
1942
1944
1946
1948
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Rel
ativ
e F
requ
ency
'nationalization'
1900
1902
1904
1906
1908
1910
1912
1914
1916
1918
1920
1922
1924
1926
1928
1930
1932
1934
1936
1938
1940
1942
1944
1946
1948
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Rel
ativ
e F
requ
ency
'party nationalization'
Source: raw data extracted from Google Books Ngram.Notes: omitting data from 1500 to 1900, which are more sparse.
The reasons for such renewed interest, currently in more specific terms and linked
to the development of party systems, are many. First, scholars like Caramani and
1One of the very few exceptions, which was also the single broad comparative effort for decades,was the work of Rose and Urwin (1975), where they proposed to analyze the impact of regionalcleavages on the electoral dispute in a selection of Western democracies.
7
also Chhibber and Kollman noticed that much could still be learned from more thor-
ough studies of the historical development of the territorial distribution of politics
in the Western democracies. Second, scholars seem to have noticed that although
in the developed democracies nationalization might have looked like a solved issue,
there was still an open market to assess the degree of nationalization of politics
in the case of recent democracies of Latin America (e.g. Jones and Mainwaring,
2003; Aleman and Kellam, 2008; Harbers, 2010), East Europe (e.g. Ishiyama, 2002;
Bochsler, 2006; Meleshevich, 2006), Africa (e.g. Wahman, 2015) and Asia (e.g.
Hicken, 2009). In the third place, regional politics not only did not disappear but
seem to be actually on the rise in many advanced democracies (e.g. Hooghe et. al.,
2010). And lastly, scholars finally started to test whether or not previously imagined
consequences of low nationalization are in fact a fact (e.g. Hicken, Kollman, and
Simmons, 2008; Lagos-Penas and Lagos-Penas, 2009).
I.2 Road-map of this work
Notice, however, that so far I have not defined precisely what the concepts of
nationalization or localism exactly mean, or what they are referring to. That was on
purpose. On the one hand, because, actually, the literature itself seldom develops
what one should understand as nationalization of politics or, even more specifically,
as a nationalized party or election. On the other hand, because that is precisely
the exercise I intend to unfold in the first chapter of this thesis. Indeed, the lack
of a more detailed definition for these concepts raises two orders of problems. On
the epistemological side, scholars resort indistinguishably to the same or correlated
terms and to the same literature, despite of talking about different things. On
8
the ontological side, the too broad idea of party nationalization in fact confounds
multiple phenomena. It means, the same party or party systems can be more or less
nationalized, depending on the dimension being considered.
I will propose a theoretical schema that splits party nationalization into four
dimensions, which are that of partisan organization, that of electoral supply, that
of the electoral demand, and that of the electoral outcome. Nationalization of
partisan organization deals with the degree to which parties are able to spread
their formal presence and their apparatus across the territory. Nationalization of
the supply means the homogenization of the presence of candidatures, that is, the
degree to which parties decide to withdraw or to face competition across districts.
Nationalization of the electoral demand refers to the degree of homogenization of
the electoral support that parties or candidates end up having. This has to do not
necessarily with the sociological homogenization of the electorate itself, but rather
to the homogenization of the appeal that the candidatures or their partisan labels
are able to exert across the eventual cleavages present in the electorate. Finally,
the nationalization of the electoral outcomes is the territorial homogenization of the
elected seats - which, by the distortion effects of electoral laws, can be different from
the nationalization of the electoral demand.
With such a detailing, my hope is that readers will be better prepared to ana-
lyze the local-national issues more accurately. By way of example, I will apply this
framework to the Brazilian case, using data on its elections for the Lower Cham-
ber (1945-2014). Brazil is a specially fitting case to testify the importance of the
proposed conceptual detailing. First, because, as I will show, its parties and party
systems can be differently characterized depending on which dimension one choses.
Secondly, because it allows me to demonstrate that also the investigation of the con-
9
sequences of nationalization depends on better definitions of the term. To illustrate
it, I will use simulation to show that while the country has very high nationalization
of the supply, which according to Cox’s theory (1997, 1999) should not inflate the
party system fragmentation, it is actually Brazil’s low nationalization of the demand
that causes such inflation.
In chapter 2, I take this question one step further. I will investigate, in a com-
parative setting and trough proper estimation, the relationships that exist between
party nationalization and the fragmentation of party systems in 62 democracies
(1945-2012). This is a particularly important question, which deals with one of
the most pivotal electoral consequences that have been ever theorized regarding the
lack of nationalization. But only theorized, not directly tested. In fact, despite the
concepts employed, in one way or another most scholars have recognized party sys-
tem nationalization as being theoretically crucial to make the effects of Duverger’s
propositions move from the local to the national level. Surprisingly, however, nation-
alization has never been included as one of the explanatory variables in an empirical
model of party system fragmentation – what creates the risk of omitted variable
bias.
Consider that an important explanatory covariate of party system fragmentation,
like as social diversity measured at the national level (but not only: think also of
federalist status, upper tier sizes, among others), might conceivably be correlated at
the same time with the degree of party system nationalization - the covariate that
has been omitted. In this case, how to be certain that social diversity (or others)
really has any (or at least as much) effect on party system fragmentation as we are
used to think? In other words, the inclusion of nationalization in the model equation
of number of parties not only will show us the importance of nationalization upon
10
the formation of party systems, but will also potentially alter the canonical and
conventional understanding of the causes of party system fragmentation.
However, scholars probably had their reasons for never including nationalization
in a model of number of parties. I argue that two of these reasons are likely to
be the fair concerns with the measurement and reciprocal causation types of endo-
geneity that may exist between these phenomena. The first has to do with the fact
that literature on electoral systems usually measures nationalization trough indices
that use number of parties in their formulae (inflation indices; see Chhibber and
Kollman, 1998; Cox, 1999; Moenius and Kasuya, 2004). Hence, it is impossible to
later use such measures as predictors in a model of number of parties. I propose to
solve this problem by using Bochsler’s Gini based measure of party system nation-
alization instead of the party inflation indices usually employed. The second type of
endogeneity is more problematic. It means the possibility that nationalization and
number of parties either cause each other or have a common omitted predictor.
To deal with such a problem, there is no better way than assuming the possible
reciprocity and modeling it explicitly. I will do that, specifying a model of reciprocal
causation through a nonrecursive system of equations. By doing so, I will show that
the effect of party nationalization on the number of parties is in fact clear and
strong, while the other way around is doubtful and weak. Then, I also show how
this inclusion of party system nationalization in the model of number of parties,
through a system of equations, indeed changes the role played by variables present
in the literature about this subject, like the canonical social diversity.
Lastly, in chapter 3, I will go back to the often assumed idea that personal voting
systems would breed localism, the cradle of parochial politics. With the burkerian
paradigm and the derivative debate on the UK and US in mind, authors tend to
11
see personal voting as not only having a probabilistic propensity towards parochial-
ization, but as an almost sufficient condition for it (see Ames, 2001; Mayhew, 1974;
Cain, Ferejohn and Fiorina, 1987; Carey and Shugart, 1995; Bowler and Farrell
1993; Samuels, 2001; Shugart et. al., 2005; among others). However, since those
countries use a single member district, where the electoral circumscriptions are local
in scope, we still know very little about the extent to which pork barrel loci are
a consequence of the personal voting or of the districtalization of the electoral sys-
tems. That means, we do not know whether what weights the local-national balance
towards local politics is the fact that voters vote for specific candidates to the Lower
Chamber or the fact that the whole election is designed by law to operate at local
levels.
Broadly speaking, the idea is to verify whether specific electoral institutions
are expected to affect the degree of localism or nationalization of the system. My
argument is that it is not the personal voting rule that brings parochial behavior.
The problem is that to properly test it, one should not look to systems where
the elections are local de jure, but to systems that also have personal voting, but
in at-large multi-member electoral circumscriptions. Which means, systems where
personal voting is a given, but districtalization varies. That is exactly the case of
the open list proportional representation.
Under those systems, voters always have the choice of choosing a specific can-
didate. But candidates not necessarily have local electoral bases. On the contrary,
local electoral support is not a given, but rather will have to be demonstrated.
Hence, I will analyze elections to the Lower Chamber in five countries that use such
a system, with results disaggregated at the local administrative level (municipali-
ties) which lies within the countries’ multimember electoral circumscriptions (e.g.
12
states, provinces, regions). The first aim is to verify the distribution of candidates’
electoral support across the territory trough spatial indices, in order to assess how
often candidates do have concentrated electoral support.
After that, I will use those measures as predictors of the odds of being elected,
to test whether electoral localism is electorally profitable. Results suggest that we
should not follow common assumptions without caution, since the majority of can-
didates do not have electoral support highly concentrated. Besides, concentrating
votes is related to the overall geographic concentration of population. The elec-
toral performance of parties has great effect for candidates even under Open List.
And spreading votes homogeneously across the territory is also profitable, just like
concentrating.
Thereby, in the end I hope I will have been able, first, to qualify the concept
of party and party system nationalization, showing the importance of a more thor-
ough conceptualization. Second, to directly investigate one of the most important
electoral consequences of nationalization, which is its effect on the party system
fragmentation. And third, to sustain the claim that localism and parochial politics
are not a direct function, or a necessary consequence, of specific electoral systems.
13
Chapter 1
The multiple dimensions of party
nationalization - an application to
the Brazilian case
In this chapter I propose to circumscribe and to unfold the broad concept of
nationalization of parties and party systems into four dimensions. Simply put, there
are many types of party nationalization one may talk about when using such general
label. The epistemological problem, as it is always the case, lies in the fact that
by analyzing a given party, or a given party system, one can develop very different
understandings and achieve very different conclusions depending on what is taken
as nationalization. An ontological issue which accompanies that one is, however,
even more decisive. Namely, parties and party systems can in fact be highly and
lowly nationalized at the same time, in different dimensions of nationalization.
The case of the Brazilian Lower Chamber illustrates such situation quite well,
with a diverse set of parties in what regards nationalization. It is not uncommon, for
instance, for a Brazilian party to offer candidates trough the whole country, but to
end up actually only having a regionalized electoral support. Likewise, parties can
be well nationalized in terms of organization, but not necessarily offer candidates
trough the whole country. Additionally, even parties that are able to gather similar
electoral support all over the country, often are not able to win seats in all electoral
14
circumscriptions.
My first aim is to define, in the next section, the main logical types of national-
ization we may think of when studying parties and party systems. Next, I exemplify
the application of such conceptual detailing by analyzing the Brazilian case. I will
show how important it is to define a priori what one understands by nationalization,
since the country’s party system can have divergent interpretations regarding each
dimension of nationalization. Lastly, I also show an example of how such conceptual
detailing might be relevant for important claims made in the literature, like Cox’s
(1997, 1999) on the cross-district linkage being a jointly sufficient condition for the
Duvergerian propositions to work at the national level. I will show, trough a simu-
lation, that in Brazil it is a very specific type of nationalization that matters when
it comes to affecting the country’s party system fragmentation. And the guilty one
is a type of nationalization which is actually different from the one Cox’s theory
would seem to expect.
1.1 Multiple nationalizations
The controversy over the meaning of party nationalization is not new. As well
acknowledged by the first time by Clagett, Flanigan and Zingale (1984), the concept
has broadly implied two different logics. On the one hand there is the idea of national
homogenization of the patterns of changes between two or more elections. Kawato,
for instance, described such interpretation as meaning that “if there is the same
direction and amount of electoral change in every district, then the electorate is
perfectly nationalized in movement” (1987:1237). Named dynamic nationalization
by Morgenstern et. al. (2009), it was the first broad concept of nationalization
15
to become popular in the international literature, after the works of Stokes (1965,
1967), and later adopted by many other scholars (e.g. Converse, 1969; McLean,
1973; Butler and Stokes, 1974; Brady, 1985; Taylor et al., 1986; Clagett aet al.,
1984; Kawato, 1987; Bawn et al., 1999; Aleman and Kellam, 2008).
On the other hand, there is another understanding of nationalization that became
the standard during the last decade. It is the logic of nationalization interpreted
as being the homogenization across the territory in a given election. Named static-
distributional nationalization by Morgenstern et. al. (2009), the first scholar to
convey such idea was Schattschneider (1960), when talking about the consequences
that the regionalization of the offer of candidates and of the electoral support in the
US had to the electoral competition in the beginning of the last century. Later, one
of the most common definitions became Kawato’s (1987:1237), to who “the elec-
torate with a nationalized configuration is one that shows few regional and district
differences in partisan support”. While for Clagett et al. (1984:80), although na-
tionalization “may not mean the homogenization of the electorate, it suggests that
distinctive regional political cultures and traditions are being replaced by a more
similar mixture of political sentiments across the nation”. The process of national-
ization of a party would mean, therefore, a territorial penetration of some sort, so
that the partisan distribution across a given country in a given election would become
increasingly homogeneous. After Schattschneider, such interpretation gained popu-
larity, being more recently adopted by a number of important works (e.g Sundquist,
1973; Rose and Urwin, 1975; Lee, 1988; Jones and Mainwaring, 2003; Caramani,
2004; and Chhiber and Kollman, 1998, 2004, Bochsler, 2010).
In fact, recently, literature seems to have been adopting that second logic as
the standard basic definition of nationalization. The reason for it follows Jones and
16
Mainwaring (2003:142) when they say that “both conceptions are meaningful, but
the term ‘nationalization’ of parties or the party system should be reserved”, specif-
ically, for the second, i.e. for the static-distributional nationalization. According
to them, those who adopted the other option actually “measured not the nation-
alization of the party system, but rather the nationalization of electoral trends (or
swings)” and logically “the concept of party system nationalization should refer to
the structure of the party system, not to whether electoral swings are similar across
districts”. I endorse such interpretation that “nationalized” is a construct which
relates more clearly to the synchronous distribution of a given characteristic across
the territory. One could even call it synchronous nationalization, in order to better
differentiate from the dynamic type.
Nonetheless, notice I vaguely mention “synchronous distribution of a given char-
acteristic” on purpose. As much as the generic static/distributional logic of na-
tionalization has been growingly becoming the literature’s standard, I claim that as
a concept it still lacks precise definition, since nationalization by these terms can
actually speak of many different things. In a word, what is more or less well-spread
across a given territory in a given election? Authors have been truly talking about
different things. The focus has varied between the territorial homogenization of
the electorate itself (e.g. Lipset and Rokkan, 1967; Sartori, 1968), of the electorate
response (e.g. Kawato, 1987; Clagett et al, 1984; Caramani, 2004), of the presence
of candidatures (e.g. Schattschneider, 1960; Cox, 1999), or of partisan organization
and partisan electoral strength (e.g. Schattschneider, 1960; Chhibber and Kollman,
2004; Mair, 1987). While for some like Caramani (2004), nationalization is proba-
bly a process that entails a variety of these aspects. From one author to the other,
the idea of static-distributional nationalization (be it trough the specific term or
17
trough a correlate) sometimes have full differences in meaning, other times there are
smaller differences in emphasis. Slightly different terms are often employed either
to talk about the same subjects as if they were the same, or to establish dialogs
between authors as if they could do so without further clarification. In fact, it is
not rare, for instance, that Cox’s cross-district linkage (1997, 1999), cross-district
aggregation (Chhibber and Kollman, 1998, 2004) and nationalization are treated
as synonyms - what they not necessarily are. Still, it is not a matter of right or
wrong; both the term and the idea of nationalization breed multiple meanings and
it is only a matter of better theoretical definition. Here, I attempt to contribute
to such detailing by unfolding into four dimensions the usually broader concept of
static-distributional nationalization present in the recent developments of the field
of electoral and partisan studies1.
The first dimension is the (a)nationalization of partisan organization. Basically,
it is concerned with the extent to which the parties exist, are present and have
organizational apparatus across the territory of a given country. The general idea
that the inner organization of parties would be relevant for - and consequently an
important part of - the nationalization of politics, is surely not new. Early on, Du-
verger (1954:228) talked about an “increased centralization of organization within
the parties and the consequent tendency to see political problems from the wider, na-
tional standpoint”. While Sartori (1968:281) hit the spot even more precisely, when
marking “the development of a stable and extensive (...) organization throughout
the country” as one of the main conditions for parties to become structured. Even
1My dimensions do not pretend to be exhaustive. One could always think of new dimensions.Some obvious ones that I leave out would be the degree of nationalization: of the political agenda,of the discourses, of the campaign themes, of the implemented policies, among many others. Onein particular that I leave out is the nationalization of the electorate in sociological terms, as willbe discussed. I chose to focus on the basic dimensions of the electoral process that are usuallypresent in the debate on party systems.
18
more specific, Rokkan linked the very process of nationalization of politics to “the
breakdown of the traditional systems of local rule through the entry of nationally
organized parties” (1970:227). As for what exactly means a nationally organized
party, Caramani (2003:436) highlighted that “to mobilize the most remote and pe-
ripheral electorates”, which was a quest successful parties progressively pursued in
order to face their adversaries, “parties needed a capillary network of local orga-
nizations”. Such network, with its “cross-local communication channels” (Lipset
and Rokkan, 1967:4), would be precisely what, in the hands of modern centralized
parties, allowed the overcoming of notabilat local (Caramani, 2003). It means a
party with national organization is a party which expands its activities and struc-
ture throughout a country, so it can eventually dispute the elections effectively not
only in specific regional spots.
A logical next step for such disseminated effective competition to become possible
is, of course, that a party decides to enter the electoral dispute throughout the
country. Which leads to our second dimension, the (b)nationalization of the electoral
supply. It aims on how well spread across the territory is the offer of candidates
presented by parties in a given election. Schattschneider (1960) was probably the
first to somehow call attention to the issue of spreading candidatures. As previously
mentioned, he pointed that the past decision of Democrats and Republicans to
withdraw from dispute in some areas due to regional issues shaped a regionalized
pattern to the American politics, with great consequences to the country’s electoral
competitiveness. Caramani (2003), by his turn, focused on the offer of candidatures
to claim that the expansion of parties entering the electoral disputes in different
electoral circumscriptions was related to the big historical change that he called the
end of silent elections, i.e. the emergence of really competitive dispute.
19
However, it was certainly in the works of Cox (1997, 1999) that this dimension
of the supply got the most centrality in the political literature. In fact, the modern
idea of electoral coordination focused much more on the Duvergerian (1954) problem
of the prudent withdraw of candidatures - to use Cox’s (1997) words - than on the
complementary Duvergerian question of the avoidance of wasted votes by the voters2.
The nationalization of the supply of candidates has to do precisely with the extent
to which parties decide to back away from dispute in some areas of the countries.
Consequently, nationalization of supply is, for all effective purposes, a synonym of
Cox’s famous concept of cross-district linkage.
It may be true that at some points Cox uses his concept more loosely, maybe in
broader sense. But he is always referring to candidates, not to electoral support. To
the coordination among political elites, which is precisely what the nationalization
of electoral supply is about, rather than to the electorate. That is why, for him,
“the question becomes, why do politicians seeking election to the national legislature
from different districts find it useful to run under a common party label” (1997:186).
Or, even more clearly, “will a national market in which candidacies and withdrawals
are traded emerge?” (1997:181). More than that, though, the nationalization of the
electoral supply entails not only a scenario in which that market becomes national,
but actually one in which a party or parties increasingly offer candidates throughout
2As it is well known, the propositions of Duverger are that single-member district electoralsystems with only one turn would lead to bipartism, while other systems, namely proportionalsystems, would lead to multipartism. The two main explanations for that are roughly the degreeof incentives given by the systems a) for a party to withdraw candidates where they see to not havechances and b) for the voters to change their choices to avoid wasting votes with candidates thatare unlikely to win. In systems like the ones with proportional rules, parties may offer candidatesjust in case, with hopes of getting one of the many seats distributed per district, just like votersmay cast sincere-preference votes because even if the chosen party is not the winner, it may winsome seats. There is the increase in the number of participants and of winners. While in systemslike the single-member district with one turn, where there is only one winner per district, therewould be no point in offering candidates just to compete or in voting for loose-all candidates.
20
the given country. And that is why, later, Cox himself properly defined his idea of
linkage, exemplifying that “at one extreme, each party in a country might field
candidates in just one district (..) At the opposite extreme, every party might run
candidates in every district” (1999:155).
To declare such dimension explicitly is specially important in order to separate
what is supply from what is the corresponding demand. Indeed, our third dimension
is what we could call the (c)nationalization of the electoral (realized) demand. Con-
trary to what Cox’s interpretation often leads us to assume, deploying candidates is
not enough to determine the behavior of the voters - as if supply and demand would
achieve a natural equilibrium. Or, in other words, as if there would rarely be can-
didatures without corresponding electoral support. Routinely, there are. Specially
in systems other than the single-member district with one turn (SMD) - which are
the far majority in the democratic World (see Golder, 2005) - but also in the SMD.
The undeniable greater tendency for supply-demand equilibrium under SMD does
not mean it is always, or even often, achieved. In general, under any system, while
it is true the possible responses from the electorate are previously bounded by the
political options offered by the elites (Schumpeter, 1950), these two dimensions of
electoral supply and demand cannot and should not be confounded as being the
same phenomena. Neither be seen as an always good enough proxy of each other.
Still, such mixing of concepts is very common. Even Cox somewhat mixed his un-
derstanding about the supply with the usual definitions of nationalization present in
the literature (see Cox and Knoll, 2003), which were mostly referring to something
that one could actually put at the demand side: the observed electoral support (e.g.
Claggett, et al, 1984, Rose and Urwin, 1970, 1975; Caramani, 2004; Morgenstern et.
al., 2009, Boschler, 2010).
21
I will subscribe to that same interpretation, considering nationalization of the
demand as based on the territorial distribution of electoral results. Even though,
an important digression is due as a cautionary note. Of course that is not the
only possible meaning of the term “electoral demand”. On the contrary, this can
be quite an ambiguous concept. For some audience, it may evoke the idea we
could call “sincere demand”, hence referring to the nationalization of the preferences
present in the electorate. Often, this is connected to the set of needs or cleavages
of the electorate, that change sociologically and structurally over time. In this
case, the nationalization of the electoral demand is seen as a over-reaching historical
phenomena, as a process regarding the homogenization of the electorate itself trough
the substitution of the local territorial cleavages for the functional ones (see Taylor
and Johnston, 1979 and the review made by Caramani, 2003, 2004). Two of the
problems with this interpretation, nonetheless, are that it is very difficult to be
empirically assessed and, moreover, it deals with diachronic phenomena that usually
changes over centuries (Caramani, 2003, 2004). However, the most important issue is
the following. Certainly, even in a simplistic rational choice perspective, there would
surely exist a priori preferences within one’s utility function, given by whatever
cleavages one might see in the electorate. Still, any effective, realized demand that
is observable would be, in any case, intrinsically the result of the evaluation of the
real options at stake.
It is why the other, more promising (for empirical purposes) way to interpret
electoral demand may be as simply the electoral purchase, in terms of votes, that
is synchronously realized in a given election. Or, in other terms, it is what we
could call “realized demand”. It means the effective final electoral support, which
is due to the mere strategic choices made by voters in a given election, given the
22
political options offered by parties. In this case, as aforementioned demand may
partially refer to Schumpeter’s “manufactured will”, since it is bounded by the
supply presented by the parties. But it also refers partially to the appeal parties
are able to exert over the electorate across the country where they offer candidates.
Therefore, the nationalization of the electoral realized demand talks about how much
the parties are able to appeal to all the electorate, i.e. to either appeal to voters
above regional differences that might still exist or to deal with electorate diversities
so to become appealing despite of them. That is why its usual assessment aims on
how homogeneous across a given country are the vote shares effectively gathered by
parties or candidates. This the way I will treat the nationalization of the demand,
as realized demand. In a word, as the nationalization of the effective final electoral
support.
Lastly, the fourth dimension of party nationalization is the (d)nationalization of
the electoral outcomes. It means, how nationalized are the allocated seats. Differen-
tiating voters’ electoral support and the electoral outcome is important, of course,
due to the dis-proportionality between votes and seats, caused in different degrees
by all electoral formulas. Let’s imagine, for instance, a plurality election in a given
country with N number of n electoral circumscriptions. Suppose all parties running
offered candidates in every n, making the system to be perfectly nationalized in
terms of supply. Also, suppose all parties ended up with very similar - although not
identical - percentage of total votes in each n. Hence, the 3 most voted parties, A,
B and C had everywhere close to 33% of the votes each. This would be also a highly
nationalized system in terms of the electoral supply. Now, let’s suppose party A
always won in the northern electoral circumscriptions, with around 34% of the votes
in each. By its turn, party B always won in the southern circumscriptions with
23
around 34% of the votes, while party C always won in the circumscriptions at the
central pat of the country. Clearly, the electoral outcomes in such a country would
be very regionalized, despite the electoral supply and the demand being perfectly
or very highly nationalized. Such reasoning can, of course, also affect proportional
electoral systems - specially those that have devices to favor bigger parties. As well
as proportional systems where the magnitudes of the electoral circumscriptions vary
widely3.
The possible differences between electoral demand and outcome are also good
examples of why distinguishing types of party nationalization is also interesting from
an ontological point of view, besides the clear epistemological gains. It means, it
is not only a matter of conceptual detailing for the sake of clarity. The reality
itself presents a scenario in which parties can be in different positions regarding
each dimension of nationalization. Logically, a party can be nationalized in terms
of organization, but not offer candidates throughout the territory. Also, a party
can have nationalized offer of candidates, but end an election having a regionalized
electoral support. And as discussed, a winner party can have a nationalized elec-
toral support, but end up having a regionalized distribution of allocated seats. It
means these dimensions have some degree of independence. Specifically, a party
being highly nationalized in one of them doest not mean it will be deterministically
nationalized in the next. Although, of course, one might claim it increases such
3After all, even if all parties have exactly the same share of electoral support in all circumscrip-tions of a country, the distribution of seats could end up being quite uneven if many circumscriptionshave few seats to be allocated (say, for instance, 1 to 3), while others have plenty (e.g. dozens).The reason is that no matter how similar is the electoral support across circumscriptions, in theones with low magnitude very few parties would be able to get seats, while plenty of others couldlogically be awarded a seat in the circumscriptions with grater magnitudes. Now suppose thesebigger circumscriptions are localized only in the south part of a country. Consequently, the partieswhich are only able to get seats in the bigger circumscriptions will be southern patterns in termsof electoral outcome.
24
likelihood. After all, for any given party, having more nationalized organization may
increase the odds of presenting candidates in more areas of the country. Presenting
candidates everywhere may reasonably increase the odds of achieving a less uneven
electoral support, and so on.
On the other hand, being regionalized in one dimension may in fact be determin-
istic to how the party will end up being in the next dimension. Let’s see. Considering
the same territorial units, a party with regionalized organization simply cannot offer
candidates throughout the whole territory. Also, a party with regionalized offer of
candidates simply cannot end up having a nationalized electoral support. Lastly, a
party with regionalized electoral support shall not have a nationalized presence in
the national Legislative4. It means that being regionalized in one dimension in fact
bounds the possibility of being nationalized in the next.
In summary, one could say that being regionalized in one dimension precludes
from being nationalized in the next dimension, while being nationalized in one di-
mension does not precludes from being regionalized in the next dimension (although
decreases likelihood). Such complex relationship between the dimensions gets clearer
if we think that in a sense each dimension is a subset of each other, as can be seen
in Figure 1.1. In that Venn diagram, when a circle expands, it does not mean its
inner circles will do the same, but when it shrinks it certainly bounds the space that
its inner circles can occupy - or even press then to shrink too.
The nationalization of the electoral supply is a subset of the nationalization of
the party organization, what makes intuitive sense. A party can only offer candi-
dates where it exists. Next, notice the less intuitive relationship that shows the
4Of course, this last step of the reasoning is not fully deterministic. Depending on how theelectoral rules translate votes into chairs, one could eventually find a scenario where a regionalizedelectoral support end up achieving a more nationalized seat distribution. This is probably a rareexception, however
25
Figure 1.1: Venn diagram of the 4 dimensions of party and party system national-ization
Natorganiz.
Natsup.
Natdem.
Natout.
nationalization of the electoral demand as a subset of the nationalization of the
electoral supply. As previously debated, indeed voters can only opt to vote for can-
didatures that exist. It means not only that electoral demands can be created by
the political supply to which voters are exposed, instead of the other way around.
But more importantly here, it means even if there is such a thing like an a priori
set of preferences that is homogeneous across the territory, it will be of no effect
if the corresponding political parties that could fulfill such demands do not offer
candidates, by any reason, in all the territory. Similarly, the electoral demand itself
mostly bounds the electoral outcome5.
However, once we move from one broad and fuzzy concept of nationalization to
a more refined understanding of its dimensions, we pay a price in easiness regard-
ing how to design the empirical data gathering. The reason is when we include
dimensions that are not necessarily from the electoral arena, the geographical units
in consideration do not have to be the electoral circumscriptions anymore. More
than that, each dimension has different degrees of flexibility in what regards the
5Again, a nationalized electoral support might frequently be distorted by the electoral rulesinto a more regionalized electoral outcome. A regionalized electoral support logically could betransformed in nationalized electoral outcome, but it is certainly quite an unlikely scenario.
26
geographical unit it will refer to. For instance, consider a country like Brazil, where
the electoral circumscriptions for the national Lower Chamber are the 26 states plus
the capital, adopting proportional representation (PR). It would not make sense to
ask whether or not the parties are nationalized across the municipalities of a state
in what regards the electoral supply, since offering a candidate in a state means
necessarily offering such candidature in all of that state’s municipalities. However,
depending on the research question it would be quite reasonable to ask whether
the parties are nationalized across municipalities in terms of electoral demand or
electoral organization, since parties can achieve dissimilar electoral support in each
city of a state, despite the electoral district being the whole state.
Evidently, the chosen geographical units always matters. Actually, in case one
considers different territorial units for each dimension, the previously appointed
relationship between the four dimensions ceases to be what we could roughly call
“one-way deterministic, one-way probabilistic” and moves towards being only proba-
bilistic, at most. Again in the example of Brazil, a given party may easily be present
and have organizational apparatus in every one of the 27 electoral circumscriptions,
since they are so few circumscriptions. But, if we look at a more disaggregated
geographical levels, like the municipalities in which the states are divided, the same
party can be actually very badly nationalized in terms of organization. It means,
the party can be present just in very few municipalities (e.g. the urban or the rural
ones, the big or the small ones, etc). Now, suppose that party eventually offers
candidates in every one of the 27 electoral circumscriptions. The absence of party
organization from many municipalities does not preclude that the party will have
homogeneous electoral support across states. However, it is reasonable to think that
such a result is less likely than it would be had the party a strong organization in
27
all of the country’s municipalities.
Before moving to the empirical examples with the Brazilian data, however, sec-
tion 2 starts by reviewing the literature on the nationalization of the Brazilian par-
ties and party systems, in order to prepare us to get to the analysis of the country’s
system.
1.2 The debate about nationalization in Brazil
The broad issue of the nationalization of politics is an old acquaintance of the
Brazilian debates, both in academia and in the political history of the country. It
is well known that in 1930 the first coup d’etat of the Brazilian Republic aimed
precisely at, among other things, counterbalancing the political force of the regional
oligarchies that had ruled the country until then (e.g. Skidmore, 1967; Fausto, 1987;
Hentschke, 2007). Therefore, when the transition to democracy was at sight in the
1940s, naturally the project of the ending dictatorship involved avoiding the return
of the only-regional parties that existed since 1889.
Such objective was pursued trough both formal and concrete measures. The
new Electoral Code that finally brought democratic elections to Brazil in 1945 was
the first to explicitly request the creation of any new party should need signatures
from supporters located in more than one state of the country (Souza, 1976). It
was meant as a way to enforce, at the formal level, that all parties would have to
somehow seek to spread their organizations across the country. At the same time,
the falling government used its public apparatus to collect signatures and to create
two State-born parties all over Brazil (idem), while a third big party emerged from
joining the opposition of the regime across the country (Oliveira, 1973). The result
28
was that with two to three big national parties, i.e. serious national contenders from
the beginning, the following creation of regional parties by regional forces would
become less tempting6.
Lamounier (1982) has claimed that even the very adoption of the proportional
representation in Brazil in that period, after many decades of majoritarian electoral
systems, had to do in part with the concern with the nationalization of politics.
After all, Assis Brasil (1893), who was the country’s main scholar on and main
advocate of the PR system, had been claiming since the 19th century that such
system would be the way to create truly national representation in the country. For
him, if the electoral circumscriptions were big enough, with a big enough magnitude
and proportional representation, “the representative will be, by the virtue of the
law, protected from the pressure of the particular interests that are incompatible
with the common good (...) The members of the majority will be, in fact and by
right, attorneys of their whole party, for they would have received from the whole
party the votes to get elected” (Assis Brasil, 1893:198-200)7. As can be easily seen,
he was worried about overcoming localized forces and localized interests, in favor
of the Lower Chamber being pushed to handle nationalized issues. A concern that
was shared by plenty of other authors and politicians. Amado (1931), for instance,
was even clearer when he claimed the vote under proportional rules is “cast to the
ideas, to the party, to the group”, while under majority rule it is given to the local
candidate, to the “individual, to the godfather, to the friend, to the boss, to the
6A divergent understanding is offered by Lima Jr. (1983). For him, while the country in practicestarted with these few big national parties, the legal requirements for new parties didn’t block thebirth of parties that, while were not fully regional like the ones before 1930, were still much moreregionalized.
7In the original: “o representante estara, pela unica virtude da lei, obrigado contra a pressaodos interesses particulares que forem incompatıveis com o bem geral (...) Os membros da maioriaserao, de fato e de direito, procuradores de todo o seu partido, que de todo ele receberam os votosque os elegeram”
29
local chief” (p:71)8.
This reasoning was the very same that lead Duverger (1951) to later say that
while “majority vote accentuates geographical divisions of opinion” (p.331), the
proportional representation “compels the elector to vote for a party rather than for
personalities, that is to say, for a system of ideas and an organization of national
scale rather than for the champions of local interests” (p.333). Besides, for Rose and
Urwin, multi-member electoral circumscriptions that are typical in PR systems give
“much more incentive for parties to offer a full slate of candidates in all regions”
(1975:19). While Caramani (2003:436) puts it even more directly, that PR systems
“had a major effect on reducing territorial opposition [in Western Europe]. After
its introduction, competition mainly opposes groups and parties on an ideological
non-territorial basis”. In fact, most Brazilian analysts and scholars were seeking
to stimulate a Brazilian version of the nationalization process Caramani (2003) de-
scribed when studying the Western European countries. Which, of course, they felt
Brazil was still lacking. This is the process by which regionalized, geographically
bounded cleavages historically gave place to others non-territorial, like ideology.
In his words, Caramani (2003:434) states that “with the gradual disappearance of
territorial strongholds, competition transformed from territorial into functional”.
And, drawing from Sartori, he clarifies that the functional–ideological dimensions of
conflict “are mainly the left–right and religious dimensions” (idem).
Such quest for political nationalization, and more specifically for the nation-
alization of the partisan competition, however, has to be split into the different
dimensions I delineated in the last section. It is clear for the Brazilian literature
that, during the democracy of 1945-64, the previously mentioned incentives for the
8In the original: “dado as ideias, ao partido, ao grupo”, while under majority rule it is given tothe local candidate, to the “indivıduo, ao compadre, ao amigo, ao boss, ao chefe local”.
30
nationalization of party organization were effectively present. Maybe even incen-
tives for a higher nationalization of the electoral supply as well. After all, under
the new PR system with big district magnitudes the parties would tend to offer
candidates in electoral circumscriptions where they had less chance, once PR gives
more hope for minor parties. At the same time, literature on Brazil has always
been quite pessimistic about anything related to the nationalization of the demand.
But it is important to stress that, informed by a sociological reading, earlier stud-
ies on that period focused on the structural characteristics of the electorate itself,
rather than on the electoral choices. Therefore, differences between urban and rural
areas, or between rich and poor regions, have long been identified as hedges that
created multiple actual countries within one formal Brazil (e.g. Lambert, 1971),
and which posed considerable challenges to the proper development of a national,
institutionalized political system (see Britto, 1965; Soares, 1973). The analysis of
the nationalization of the electoral support itself got mostly neglected until Lima
Jr. (1983) and Santos (1987).
Indeed, the different levels of nationalization of electoral supply and of electoral
demand constitute a particularly central question. As Kawato (1987:1237) puts it,
“if reaction to the national party plays the most important role in an election and
the composition of the electorate in most districts is similar, then it is probable that
every district will show a similar level of partisan support” (emphasis mine). To be-
come fully nationalized, parties would have to be well spread across a country in as
least two of the dimensions defined previously: supply and demand. Regarding the
democratic period of 1945-1964, the general assessment of the Brazilian situation
is, until today, controversial. Lima Jr. (1983) claimed that over time the political
system was indeed becoming more nationalized, since during those years regional
31
parties would expand their presence around the country, while national parties pene-
trated in the regions and states. Later, Santos (1987), Lavareda (1991) and Nicolau
(2004) argued that, in reality, small and medium parties ended the period being
either more or at least equally regionalized as they were at the beginning. While
the larger, more important parties, were in fact growing more national. Still, it is
common to find case studies of the larger parties that claim the opposite, i.e. that
they never became actually national (e.g. Benevides, 1981, 1989; Hippolito, 1985)
be it in their presence, supply or demand.
By one hand, it has never been clear neither how nationalized the Brazilian par-
ties became, nor whether their level of nationalization was increasing with time.
The reason for the confusion is that scholars usually lacked an a priori specific defi-
nition of what they have been understanding for “nationalized”. By the other hand,
hardly an observer or a scholar would deny that the political issues under debate
became increasingly national (see Souza, 1976; Abrucio, 1998), and that cleavages
in fact became ideological, rather than as territorial as they were until the 1930s or
1940s. As a matter of fact, never again the regionalized politics were to become such
a prominent issue in Brazil, as it had been before. After that democratic period,
of course the issue even ceased making sense during the dictatorship that followed
the 1964 coup d’etat - a coup some attribute to the lack of governability caused by
the fragmentation and polarization of the political forces in the Legislative branch
(see Santos, 1979). The military junta soon forced a two-party only system that
lasted for almost twenty years. Such authoritarian interlude had a clear nationalist
discourse and centralized the country in what regards many of the state powers,
helping to further rub out past regional quests.
All summed, during the transition to the current democratic period in the 1980s,
32
the issue of nationalization deserved much fewer attention. But still, with quite di-
vergent conclusions. Lima Jr. (1983:301) claimed that the Brazilian states currently
are “politically different in their electoral and partisan manifestations”9. While
Mainwaring (1999:318-319) goes further to say that, in Brazil, “a rigor, os partidos
nacionais sao federacoes de partidos estaduais” and that “uma ultima consequencia
da importancia da dinamica estadual no sistema partidario e seu baixo grau de na-
cionalizacao - no sentido de que os padroes de voto variam significativamente entre
os estados”. By his turn, Samuels (1998) points that the recent politics in Brazil are
strongly state-based, while Ames (2001) concludes that Brazil does not even have
what could be called national parties. Recently, diagnostics following the opposite
direction started to show up, arguing that the Brazilian party system is becoming
more nationalized. Roma and Braga (2001) have claimed that since the presiden-
tial and legislative elections became simultaneous in 1994, the large parties have
been pursuing a national strategy when forming alliances10, in a reasoning about
the presidential elections that resembles Cox’s (1997). Braga has also sustained
that the partisan organization (2006) and the electoral presence of parties (2010)
had become more nationalized in Brazil since the 1990s. While Speck and Campos
(2014) argued that the TV airtime given to parties for propaganda would be one of
the reasons for such nationalization.
In any event, as it is easy to notice, the subject of party system nationalization
in Brazil remains quite open. Scholars have been debating about the general sub-
ject using very different concepts, very different understandings, summoning very
different evidence. Below, I hope to show how better defining nationalization helps
solving part of these questions. I will present data for the Brazilian case, on the
9In the original: “politicamente diferentes em suas manifestacoes eleitoral e partidaria”.10For a relativization of those findings, see Krause (2005).
33
four dimensions of party and party system nationalization, focusing on the country’s
Lower Chamber. In order to proceed, a brief introduction on the electoral system
used for these elections might be useful. In all democratic elections for the Lower
Chamber since 1945, Brazil has adopted an Open-List PR system, with the elec-
toral circumscriptions being the country’s states plus the capital city area (there
have been from 22 to 27 in the period). Magnitudes of electoral circumscription
ranged from 1 to 70, currently the minimum being 8 and the maximum being 70.
The Brazilian version of open list system has been working as follows.
In each election, voters have only one vote for the Lower Chamber, which can be
either cast for a specific candidate or for a party label. Electoral alliances between
parties, per electoral circumscription, have been allowed. The great majority of
voters cast votes for specific candidates and, in this case, the vote is firstly counted
for the party/alliance to calculate how many seats are going to be won, if any, by
that party/alliance. Next, the vote is used to determine which candidates of the
party/alliance are going to fill the seats won. As an example, suppose the case of
candidate C, from party P, which is a member of alliance A. A vote given for C is
first used to determine how many seats A will win in C’s electoral circumscription.
Let’s say A won S seats. Then, the S most voted candidates in the circumscription
are going to fill the S seats.
I will present data covering different periods of time depending on the type of
nationalization being analyzed, according to availability. All raw data comes from
Brazil’s official Electoral Justice branch, the Tribunal Superior Eleitoral (TSE).
34
1.2.1 Nationalization of organization
Since in both Brazilian democratic periods there have been very few electoral
circumscriptions and the country never had many registered parties (at most around
thirty), it is no surprise that pretty much in every national election all registered
parties have been somewhat present (i.e. had directories, offices, members) in every
electoral district (i.e. the states plus capital). Being so, if the geographical units
of interest are these circumscriptions, the Brazilian system has always been almost
perfectly nationalized in what regards nationalization of the partisan presence across
the territory. It is, therefore, of no use to present specific data on party presence
per circumscription since the results are the same in every state. But what I will
double-check is how nationalized has been the organizational strength of each party
in Brazil recently. As a loose proxy, I will use the share of official party members
that were held by each party, during the last four elections. Geographical data on
electorate affiliated to the registered parties were available only from 1994 onwards.
More precisely I will assess, for each party, the territorial homogeneity, across the
country’s electoral circumscriptions, of the percentage of partisan members that are
official members of the given party in each circumscription.
To measure it, I employ a weighted regional Gini index of territorial inequality.
Specifically, I will employ the complement-Gini index version proposed by Bochsler
(2010) for measuring party nationalization, which takes the usual weighted spatial
Gini and standardizes it by the number of electoral circumscriptions (in this case,
the states plus capital). It is usually called party nationalization score, or PNSsw,
with the subscripts standing for standardized and weighted. This measure has many
advantages. First, the Gini index and its interpretation are well known, unlike most
other common options for assessing party nationalization (for a comprehensive list
35
Figure 1.2: Nationaliz. of organization: homogeneity of % membership across elec.circumscriptions
PR
TB
PT
NP
CB
PT
DO
B PV
PR
PP
ST
UP
SC
PR
NP
MN
PF
LP
DT
PC
DO
BP
PS
PS
BP
MD
BP
PP
TB
PS
DB
PT
PL+
PR
ON
A
00.10.20.30.40.50.60.70.80.9
11994
PC
BP
CO
PR
TB
PS
TU
PT
NP
SN
PT
DO
BP
SL
PS
DC
PV
PR
PP
PS
PS
CP
RN
PM
NP
DT
PF
LP
SB
PC
DO
BP
MD
BP
PB
PL+
PR
ON
AP
TB
PT
PS
DB
00.10.20.30.40.50.60.70.80.9
11998
PC
BP
CO
PR
ON
AP
ST
UP
RT
BP
T D
O B
PG
TP
TN
PT
CP
SD
PD
TP
SL
PR
PP
HS
PAN
PS
TP
SC
PS
DC
PV
PM
NP
FL
PC
DO
BP
PS
PP
BP
MD
BP
SB
PT
BP
TP
LP
SD
B
00.10.20.30.40.50.60.70.80.9
12002
PR
BP
CB
PS
OL
PS
TU
PC
OP
RO
NA
PR
TB
PT
DO
BPA
NP
DT
PT
CP
RP
PS
CP
TN
PF
LP
PP
SL
PM
NP
SD
CP
PS
PC
DO
B PV
PM
DB
PH
SP
SB
PT
PT
BP
SD
B PL
00.10.20.30.40.50.60.70.80.9
12006
PC
BP
CO
PS
TU
PT
DO
BP
RB
PR
TB
PR
PP
SO
LP
TC
PC
DO
BP
MN
PS
LP
SC
PT
NP
DT
PS
DC
PH
SD
EM PV
PS
BP
PS
PP
PT
BP
RP
MD
BP
TP
SD
B
00.10.20.30.40.50.60.70.80.9
12010
PC
BP
CO
PE
NP
ST
UP
RO
SP
SO
LP
RT
BS
DP
T D
O B
PR
PP
SD
PT
CP
C D
O B
PR
BP
TN
PD
TP
MN
PS
LP
PL
PS
DC
PS
CD
EM
PH
SP
VP
PP
PS
PS
BP
TB
PM
DB
PR PT
PS
DB
00.10.20.30.40.50.60.70.80.9
12014
0% 2 5 8 10 12 15 18 20 22 25%Registered parties / Legend: countrywide elec.support:
Wei
ghte
d te
rrito
rial h
omog
enei
ty o
f mem
bers
hip
rate
s
of these options, see Caramani, 2004 and Bochsler, 2010). It assess precisely what
we want, which is territorial homogeneity versus heterogeneity, of the quantities of
interest. Second, the weighted Gini accounts for the sometimes huge differences
that may exist between electoral circumscriptions in terms of size of the electorate.
Third, by applying the standardization proposed by Bochsler, we make the index
also comparable among countries with very different number of electoral circum-
scriptions11. Here, the quantity of interest that I will apply the index on is the
11Although here that is not particularly useful since there is no international comparison, lateron I will use the index again for measuring nationalization of the electoral demand and will put
36
partisan membership force rate. It is, for each party p of the set of parties P , the
percentage given by the number of voters who are members of party p in circumscrip-
tion c, i.e. pc, divided by the total number of voters that are members of any party
P in c, i.e. Pc. In the end, the index ranges from 0, which means extremely unequal
distribution of membership force across the electoral circumscriptions (or extremely
poor nationalization of organization), to 1, which means perfectly equal distribution
of membership force across electoral circumscriptions (or perfect nationalization of
organization)12. Figure 1.2 has the results.
In fact, it seems in the years of the last four elections to the Lower Chamber,
the Brazilian parties had not only been present in all electoral circumscriptions, but
they also have been able to penetrate their organization somewhat evenly too. It
means that, taking into account the demographic differences of the electoral cir-
cumscriptions, most parties have proven to have a quite nationalized distribution of
their membership force. Only very few parties show lower levels of nationalization
of the membership and these are precisely the very smallest parties in the country13.
It suggests Brazilian parties are in fact at least in a good position to compete evenly
across the circumscriptions. If they in fact do that, it is something I will show in a
moment. Before, however, it may be interesting to make a very quick detour and
show how results can be dissimilar when we break the concept of nationalization
and start playing with the geographical units being considered.
It means, there is something we can do to better understand the penetration of
the parties’ infrastructure across Brazil. We can focus on a different geographical
the Brazilian data in comparative perspective. Anyway, it is always useful to deliver figures thatcan be further compared by the readers themselves to data they might have on other countries.
12The original formula for Bochsler’s index can be found in Annex 1.A13From now on, any reference to the size of parties will be made in terms of final national results
for the Lower Chamber.
37
Figure 1.3: Presence of parties’ organization across the Brazilian municipalitiesP
RT
BP
TN
PS
TU
PC
BP
VP
T D
O B
PP
SP
RP
PM
NP
SC
PC
DO
BP
RN
PS
BP
L+P
RO
NA
PT
BP
TP
DT
PS
DB
PP
PF
LP
MD
B
0102030405060708090
1001994
PC
OP
TN
PC
BP
RT
BP
ST
UP
SN
PS
DC
PT
DO
BP
SL
PV
PR
PP
MN
PR
NP
PS
PC
DO
BP
SC
PS
BP
L+P
RO
NA
PT
PD
TP
TB
PF
LP
PB
PS
DB
PM
DB
0102030405060708090
1001998
PC
OP
CB
PS
TU
PG
TPA
NP
TN
PR
ON
AP
HS
PR
TB
PS
DC
PT
DO
BP
ST
PV
PS
LP
TC
PR
PP
MN
PC
DO
BP
SC
PS
DP
PS
PS
B PL
PD
TP
TP
TB
PS
DB
PF
LP
PB
PM
DB
0102030405060708090
1002002
PR
BP
CO
PS
OL
PC
BP
ST
UP
RO
NA
PAN
PT
NP
RT
BP
T D
O B
PH
SP
SD
CP
SL
PT
CP
VP
RP
PM
NP
C D
O B
PS
CP
SB
PP
S PL
PD
TP
SD
BP
TB
PT
PF
LP
PP
MD
B
0102030405060708090
1002006
PC
OP
ST
UP
CB
PS
OL
PT
NP
RT
BP
T D
O B
PS
DC
PH
SP
RB
PS
LP
RP
PT
CP
MN
PC
DO
B PV
PS
CP
PS
PS
BP
RP
DT
PT
BD
EM PP
PS
DB
PT
PM
DB
0102030405060708090
1002010
PC
OP
CB
PS
TU
PE
NP
PL
SD
PR
OS
PS
OL
PR
TB
PS
DC
PT
NP
T D
O B
PH
SP
TC
PR
PP
SL
PM
NP
RB
PC
DO
BP
SC
PV
PS
DP
PS
PS
BP
RP
DT
DE
MP
TB
PS
DB
PP
PT
PM
DB
0102030405060708090
1002014
0% 2 5 8 10 12 15 18 20 22 25%Registered parties / Legend: countrywide elec.support:
% o
f mun
icip
aliti
es w
here
par
ties
are
offic
ially
pre
sent
unit, since TSE has been also collecting data on party membership at municipal
level since 1994. Municipalities are the cities into which the Brazilian electoral cir-
cumscriptions are divided. Therefore, I will go down in geographical disaggregation.
Figure 1.3 shows in how many cities each Brazilian party has been officially present,
which means in how many cities they have had a directory, an office. To measure
it, I consider a given party was present in a given municipality if it had at least
one member in there, since to establish an office, a directory or any sort of official
presence a party would have to have at least one member.
As it can be seen, even in recent elections few Brazilian parties were present in
38
close to all municipalities of the country. Most parties have been present, actually,
in less than 50% to 60% of the municipalities. However, not surprisingly, the parties
that have a more well spread presence across municipalities are exactly the biggest
ones, who have had the greatest electoral support for the national Lower Chamber
elections. It even looks like the system can be divided in two general groups. One
with the parties that get the greatest share of electoral support and are present in
almost all municipalities of Brazil. The other with the parties that gather medium
to low levels of country-wide electoral support and are present in less than 60%
of the Brazilian municipalities. Certainly, using the diagonal reference line, it is
possible to see that the organizational-nationalization of the party system has been
increasing, with these low and medium nationalized parties spreading a bit more
over the years. For an additional but similar assessment, see the figure at Annex
1.B, showing similar results but with data for the percentage of cities where parties
have presented at least one candidate in the municipal elections - be it for mayor or
for local Legislative.
Still, that movement in Figure 1.3 has been a small one. It is unfortunate
that similar data are not available for previous years for us to compare the recent
situation to that of the beginning of the current system in the 80s to that of the
previous democratic period in 1945-1964. But if the visual pattern was consistent
backwards, it looks like at least in the 80s and 90s the system was probably steeply
divided between a small group of parties with national presence and a majority
group of parties with only regional presence. Of course, these results use a different
geographical unit than the one I employed before and will use from now on. But
it helps to understand that the territorial presence and the idea of nationalization
can have multiple faces. It is even reasonable to expect interconnections, like for
39
instance that the badly spread presence of parties in the cities of Brazil may affect,
for instance, the capability of the parties to win votes in certain areas of the country
despite of offering candidates.
1.2.2 Nationalization of the supply
Talking about offering candidates, I have just mentioned that in terms of the
electoral circumscriptions (states+capital), the Brazilian parties are in a good posi-
tion to offer candidates for the Lower Chamber all over the country, since they are
present in all circumscriptions and with well spread membership force. It means
they have a first condition for achieving a high cross-district linkage, to use Cox’s
(1997, 1999) expression. To asses it, Figure 1.4 shows in how many Brazilian elec-
toral circumscriptions each party has presented candidates for the Lower Chamber
in each democratic election since 1945.
The situation is quite different for each of the democratic periods of Brazil. From
1945 to 1962, there was a clear pattern of low nationalization of the partisan supply.
Only three parties offered candidates in close to all electoral circumscriptions. And
as it is shown by taking into consideration the cumulative line of electoral support
in the right Y-axis, these few nationalized parties in terms of electoral supply were
exactly the three biggest parties that dominated such electoral period. All other
parties had a less spread supply of candidates, with most being fairly regionalized
in such terms. There was a division, therefore, between national big parties and
regional small and medium parties, just like we saw it was claimed by the Brazilian
literature. And the system didn’t look like it was presenting any major change
across time, specifically it does not look like the smaller parties were getting more
nationalized in terms of the electoral supply - like Santos (1986) and Nicolau (2004)
40
Fig
ure
1.4:
Nat
ional
izat
ion
ofth
esu
pply
:num
ber
ofel
ec.
circ
um
scri
pti
ons
wher
epar
ties
offer
edca
ndid
ates
PR
PL
PAN
PDC
PRD
PRProg.
PPS
PRP
PTB
PCB
PSD
UDN246810121416182022
0102030405060708090100
1945
POT
PRB
PDC
PL
PTN
PSB
PRT
PRP
PST
PR
PSP
PTB
UDN
PSD
24681012141618202224
0102030405060708090100
1950
PST
PRP
PRT
PTN
PL
PSB
PDC
PR
PSP
UDN
PSD
PTB
24681012141618202224
0102030405060708090100
1954
PRT
PTN
PL
PRP
PST
PSB
PDC
PR
PSP
PTB
UDN
PSD
24681012141618202224
0102030405060708090100
1958
PRT
PL
MTR
PR
PRP
PTN
PST
PDC
PSB
PSP
UDN
PTB
PSD
24681012141618202224
0102030405060708090100
1962
PTB
PDT
PT
PDS
PMDB
24681012141618202224
0102030405060708090100
1982
PASARTPDI
PNRPRTPTNPMNPRPPTRPCN
PNPSPJ
PMCPPBPND
PHPMBPSCPDCPCB
PLPCdoB
PSBPTBPDSPDTPFL
PMDBPT
2468101214161820222426
0102030405060708090100
1986
PAPPBMPEBPRSPSU
PDPLHPNT
PSPAS
PRONAPSLPCNPRP
PTdoBPSDPSTPMNPSCPCBPTBPTRPSBPDC
PCdoBPDS
PLPDTPFL
PMDBPRN
PSDBPT
2468101214161820222426
0102030405060708090100
1990
PTdoBPCB
PTRBPRONA
PRPPRN
PSTUPSDPSC
PVPMN
PLPCdoB
PPPPSPDTPFL
PMDBPPRPSB
PSDBPT
PTB
2468101214161820222426
0102030405060708090100
1994
PCOPGTPTNPCB
PSDCPSLPSNPRN
PRTBPSTPANPSDPRP
PSTUPTDOB
PSCPMN
PVPPSPSB
PLPRONA
PTBPCDOB
PDTPFL
PMDBPPB
PSDBPT
2468101214161820222426
0102030405060708090100
1998
PCOPRONA
PCBPSTU
PSLPTCPRPPSD
PSDCPAN
PRTBPSTPHSPTNPGT
PTdoBPVPL
PSCPCdoB
PDTPFL
PMDBPMNPPBPPSPSB
PSDBPT
PTB
2468101214161820222426
0102030405060708090100
2002
PRBPCOPCBPRPPSLPAN
PSTUPHSPMN
PRONAPSC
PSOLPTC
PSDCPTdoB
PTNPRTB
PCdoBPDTPFL
PLPMDB
PPPPSPSB
PSDBPT
PTBPV
2468101214161820222426
0102030405060708090100
2006
PCOPCB
PSDCPSTU
PTNPTDOB
PSLPRTB
PTCPRPPHSPPSPRBPSB
PSOLDEMPMNPSC
PCDOBPDT
PMDBPPPR
PSDBPT
PTBPV
2468101214161820222426
0102030405060708090100
2010
PCOPCBPENPSLPPL
PSTUPTNPTC
PTdoBPRTBPSDC
PCdoBPPS
PROSPSDB
PDTPHSPMNPRP
PVDEM
PMDBPPPR
PRBPSBPSCPSD
PSOLPT
PTBSD
2468101214161820222426
0102030405060708090100
2014
Reg
iste
red
part
ies
N. of elec. circumscriptions where candidates to the Lower Chamber were presented
Cummulative % of votes for the Lower Chamber in the whole country
N. o
f ele
c. c
ircum
scrip
tions
whe
re p
artie
s pr
esen
ted
cand
idat
es fo
r th
e Lo
wer
Cha
mbe
r an
d fin
al e
lect
oral
sup
port
41
argued.
By the other hand, in the current democratic period, there is a trend of temporal
change, with the first election (i.e. 1986) after the transitional election (i.e. 1982)
presenting a pattern that was similar to the one of 1945-1964. It means, with a lowly
nationalized party system in terms of the electoral supply. Few parties in 1986
presented candidates in numerous electoral circumscriptions of Brazil, and these
were exactly the parties with the greatest electoral support. However, election after
election such scenario gradually changed, in a way that all parties gradually started
presenting candidates in more electoral circumscriptions. To the point that in 2002
the far majority of the parties offered candidates in close to all circumscriptions. In
the last elections, in 2014, the system was already strongly nationalized in terms
of the electoral supply. Or, in Cox (1997, 1999) wording, Brazil currently presents
a system with very high cross-district linkage, in which politicians do link to each
other under a same party label all over the country, in order to face elections. As
the cumulative line support shows, even parties with small and medium electoral
support are in such a situation.
In summary, although in 1945-1964 the electoral supply of Brazilian party sys-
tem was actually fairly non-nationalized, since 1982 it displayed a changing pattern
from non-nationalized to very strongly nationalized. This tendency could not be
explained solely by institutional factors like the presence of the PR system that
stimulates small parties to present candidates in different electoral circumscriptions,
since the two democratic periods had the same macro electoral institutions. One
possibility to explain why the system became rapidly supply-nationalized in the cur-
rent period, and didn’t in 1945-1964, may be the recurrence to electoral alliances
between small and big parties. Although such strategy was available and becoming
42
increasingly common in the earlier democratic period too (Soares, 1964; Schmitt,
1999), it was no way as prevalent as it is in the current regime. In the current period,
in every electoral circumscription, pretty much every party always enters an alliance
with another, in order to offer candidates for the Lower Chamber. Certainly, this
almost universal use of the electoral alliances raises the likelihood of electing at least
one deputy, what makes it interesting for smaller parties to try their luck an offer
candidates in as many electoral circumscriptions as they can14.
1.2.3 Nationalization of the demand
We have seen so far that in the supply side, the Brazilian party system has grown
over time to become highly nationalized. Nevertheless, I now analyze how it has
performed in the demand side. It is to say, how national or regional has been the
appeal of parties to the voters in Brazil. It can be the case that although presenting
candidates all over the country - which, again, is tempting, considering the electoral
system - parties may have specific regional clienteles.
In order to address this dimension, I will assess the territorial inequality of the
shares of votes obtained by each Brazilian party across the country’s electoral cir-
cumscriptions. I will measure that once again resorting to Bochsler’s (2010) PNSsw.
However, now I will apply it to the electoral support (share of total votes) of par-
14Of course, that explanation would ask for another: why parties enact electoral alliances moreoften in the current system than in 1945-1964? Although that is not my central question here, onereasonable guess would be the changes in the role and powers of state governors from one periodto the other. Lavareda (1991), Souza (2005), Schmidt (1999) and others claimed that, for the bigparties, these electoral alliances for the Lower Chamber were of interest in order to bring the smallparties to their alliances in the elections for state governors. If that is true, the more rewarding thegovernor position is, the more frequently the big parties would accept enacting electoral allianceswith the smaller parties for the Lower Chamber, provided that these accept to be on board inthe given state govern elections. Although there are few studies about the institutional powers ofgovernors in Brazil in the first half of the last century, many claim that after the 1988 Constitutiongovernors got either important powers (e.g. Abrucio, 1998; Samuels, 1998) or policy attributions(Arretche, 2007).
43
ties across electoral circumscriptions. Which means in the end, the index ranges
from 0, which means extremely unequal distribution of electoral support across the
electoral circumscriptions (or extremely poor nationalization of the demand), to 1,
which means perfectly equal distribution of electoral support across electoral cir-
cumscriptions (or perfect nationalization of the demand).
Figure 1.5 has the results for all Brazilian parties in all of the country’s demo-
cratic elections. Indeed, as expected, in 1945-1964 the overall degree of nationaliza-
tion of the electoral demand was quite low, following the also low nationalization of
the supply we saw previously. Very few parties had more than medium-level scores,
which were the biggest parties. The overall picture was of low nationalization of
the demand, and in fact it seems Lima Jr. was incorrect in his interpretation that
the regional parties were nationalizing over time. On the other hand, since the
80s the picture is more nuanced, with more mixed results. The country has had
some parties with better nationalization of their demand, but most have what we
could temporarily call medium or low levels of nationalization of their electoral de-
mand. More interestingly, this is a scenario that happens in spite of the very high
nationalization of the electoral supply.
Actually, the electoral demand in Brazil has been much less nationalized when
compared to the previous dimensions. Surely, it is possible to notice the parties with
medium and low nationalization have seen an increase over time in the homogeneity
of their electoral support, especially from 1990 to 1994. Still, it is worth noting that
few parties scored higher degrees of nationalization even in the most recent elections.
Similar to previous dimensions, it can be said that in general lines the parties with
higher nationalization of the electoral demand are the bigger parties, the ones that
end up having the greatest electoral support. However, this pattern is less clear
44
Fig
ure
1.5:
Nat
ional
izat
ion
ofth
edem
and:
elec
.su
pp
ort
hom
ogen
eity
acro
ssel
ec.
circ
um
scri
pti
ons
(mea
sure
dw
ith
Boch
sler
’s(2
010)
PN
Ssw
PL
PR
PRD
PPS
PRProg.
PDC
PAN
PRP
PTB
PCB
UDN
PSD0.1.2.3.4.5.6.7.8.91
1945
PRB
POT
PL
PST
PRP
PRT
PDC
PSB
PTN
PSP
PR
PTB
UDN
PSD
0.1.2.3.4.5.6.7.8.9119
50
PRT
PRP
PTN
PST
PL
PSB
PDC
PR
PSP
UDN
PTB
PSD
0.1.2.3.4.5.6.7.8.9119
54
PL
PTN
PST
PRT
PRP
PSB
PR
PDC
PSP
PTB
UDN
PSD
0.1.2.3.4.5.6.7.8.9119
58
PL
PRT
PTN
MTR
PSB
PRP
PST
PSP
PR
PDC
PTB
UDN
PSD
0.1.2.3.4.5.6.7.8.9119
62PDT
PTB
PT
PDS
PMDB
0.1.2.3.4.5.6.7.8.9119
82
PJPMNPRPPDI
PASARTPNRPRTPTNPTR
PSPN
PCNPNDPDCPPBPMC
PLPH
PSBPTB
PCdoBPSCPMBPDTPCBPDSPFL
PTPMDB
0.1.2.3.4.5.6.7.8.9119
86
PBMPEB
PSPASPRSPAPPSUPNT
PRONAPSCPLHPSDPSLPSBPSTPCN
PDPMNPTRPRPPCBPDC
PTdoBPDTPFL
PCdoBPDS
PSDBPTB
PLPRN
PTPMDB
0.1.2.3.4.5.6.7.8.9119
90
PTdoBPCB
PTRBPMNPSCPSB
PVPRPPPSPRN
PRONAPSD
PPPFL
PCdoBPL
PDTPPR
PSTUPTB
PSDBPT
PMDB
0.1.2.3.4.5.6.7.8.9119
94
PSLPTdoB
PCOPGTPSDPSTPMNPTNPSC
PSDCPRPPANPSBPSN
PSTUPLPV
PRTBPPS
PRONAPCdoB
PRNPCBPDTPTB
PMDBPFLPPB
PTPSDB
0.1.2.3.4.5.6.7.8.9119
98
PRONAPTdoBPSDC
PSLPSDPSCPST
PRTBPAN
PMNPCBPGT
PCdoBPTC
PVPSBPRPPCOPHSPTNPTBPFLPDTPPS
PLPPB
PSTUPMDBPSDB
PT
0.1.2.3.4.5.6.7.8.9120
02
PRBPMNPAN
PTdoBPRTB
PTNPTC
PRONAPCB
PSOLPSDC
PHSPSCPPSPCOPRP
PVPCdoB
PSLPSB
PSTUPFLPTB
PMDBPLPP
PSDBPDT
PT
0.1.2.3.4.5.6.7.8.9120
06
PCOPCB
PTdoBPRTB
PSLPSDCPMN
PSOLPTCPTNPHSPRP
PVPSCPSB
PCdoBPPS
PSTUPMDB
PRBPTB
PRDEMPDT
PSDBPPPT
0.1.2.3.4.5.6.7.8.9120
10
PRTBPTdoB
PCOPROS
PMNPSDC
PENPSLPTNPRP
PSOLPCBPTC
PVPSCPHSPPLPPSPRB
PCdoBPSB
PSTUDEMPDT
PPPSDB
PTBPSD
PMDBPRSDPT
0.1.2.3.4.5.6.7.8.9120
14
0%2
58
1012
1518
2022
25%
Lege
nd:
coun
tryw
ide
elec
.sup
port
Reg
iste
red
part
ies
Standardized/Weighted cross−circumscriptions homogeneity of the electoral support
45
than it was in the previous dimensions. Actually, in some years parties with not so
high nationalization won some of the greatest shares of electoral support (notice for
instance the case of PFL, the Liberal Front Party). Even more importantly, in recent
elections, especially the last one in 2014, the electoral preponderance of parties with
more nationalized demand diminished. Still, most of the Brazilian parties seem to
have at least what we could call medium level nationalization, not getting closer to
zero.
Notwithstanding, it might seem too abstract to judge the nationalization of
electoral demand in Brazil as being high or low without a comparative benchmark.
This is especially important when we talk about results achieved trough a secondary
index. For instance, thinking of the nationalization of the electoral supply, imagine
a party which offers candidates in 50% of the electoral circumscriptions of a given
country. Probably, it would not be too off the target to interpret, just with the
given information, that such a party had a not high nationalization of the electoral
supply. However, when we move to the measures of the nationalization the demand,
the meaning of each level of the index is not so straight-forward. In Brazil, most
parties score above 0.5 in the recent elections. But how low or how high is 0.5? I
claim it is actually quite low. It is very unlikely that a party would score extremely
low levels of nationalization of the demand, no matter what. In order to show the
Brazilian situation in perspective, I will also present a summary measure per party
system and compare it to the ones from other countries. The procedure is intuitive.
Recapitulating: for the last figure, I calculated for each party the Bochsler’s (2010)
standardized version of the weighted regional Gini, applied to the share of votes of
parties across electoral circumscriptions. Now, for each election, I get the results
of all parties and calculate a weighted average, i.e. weighting for the final size of
46
national votes share of parties. Bochsler calls it the party system nationalization
score, or PSNSsw.
To permit international comparison, I calculate the same for many other country-
elections since 1945. I use my own data with electoral results of countries at the
electoral circumscription level, as well as similar data from the Constituency Level
Electoral Archive - CLEA project (Kollman et al, 2014). In the end, it totals more
than 700 elections, in 82 countries. Figure 1.6 brings the results, where each dot
represents the PSNSsw of a party-system of a country-tier in a given election. The
Brazilian party systems are colored in red to ease visualization.
Figure 1.6: Party system nationalization of demand among countries
●● ●
● ●
●●
● ●● ● ● ● ●
1945
1947
1949
1951
1953
1955
1957
1959
1961
1963
1965
1967
1969
1971
1973
1975
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011
2013
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
● ●Brazil Others
Year
Nat
. of e
lec.
dem
and
at p
ty s
yste
m le
vel
In fact, when compared to other countries, the Brazilian party systems have
always been amidst the less nationalized in terms of electoral demand. Although the
dots corresponding to Brazil may be in the middle of the dot cloud, it is important
to keep in mind that it is a scatter-plot with color-density layer, which means the
darker the gray-to-black dots are, the more cases are concentrated in the region
of the plot. So, in reality, the great mass of democratic elections for which data
at the electoral circumscription level were available, stay quite above the Brazilian
47
elections. That means, in comparative perspective the Brazilian party systems have
been quite lowly nationalized in terms of electoral support of the parties. And in
both democratic periods
1.2.4 Nationalization of the outcomes
Of course, as mentioned at the beginning, even with the possible dis-proportionality
between votes and seats, it is unlikely that an election with medium-to-low level of
nationalization of the electoral demand would end up presenting nationalized dis-
tribution of seats, i.e. nationalized outcomes. On the contrary, it is more likely that
the seats will end up distributed similarly or less evenly than the electoral demand
is. In order to see what has been happening with the Lower Chamber seats in Brazil,
Figure 1.7 shows in how many electoral circumscriptions (states plus capital) the
elected parties were able to win seats, since 1945. Notice that the graphic also
brings a cumulative line to see how many of the total seats were won by parties
with given levels of nationalization of the outcomes. Also, the colors in the figure
show the results of a simulation for which data were available only for the 1994-2014
period. The red shades show the parties that would loose seats in case the official
distribution of seats disregarded the electoral alliances (the darker the red, the more
seats lost) while blue means the party would actually win more seats (the darker,
the more).
The result here is fairly clear. The electoral outcomes in Brazil are not nation-
alized. Almost no party was able to win seats in all the electoral circumscriptions.
On the contrary, the norm has been that only one or two parties, among the dozens
that win seats, are able to get at least close to winning in the whole country. In
2014, for instance, the cumulative line shows that around 70% of the seats were won
48
Fig
ure
1.7:
Nat
ional
izat
ion
ofth
eou
tcom
es:
num
ber
ofel
ec.c
ircu
msc
ripti
ons
wher
epar
ties
won
seat
s
PL
PDC
PRProg.
PPS
PR
PCB
PTB
UDN
PSD13579111315171921
1945
0.1.2.3.4.5.6.7.8.91
PRT
PSB
PTN
PDC
PRP*
PST
PL
PR
PSP
PTB
UDN
PSD
13579111315171921232519
50
0.1.2.3.4.5.6.7.8.91
PRT
PDC
PRP
PTN
PSB
PL
PR
PSP
PTB
UDN
PSD
13579111315171921232519
54
0.1.2.3.4.5.6.7.8.91
PTN
PRT
PST
PL
PRP*
PDC
PSB
PR
PSP
PTB
UDN
PSD
13579111315171921232519
58
0.1.2.3.4.5.6.7.8.91
PRT
MTR
PR
PL
PRP
PSB
PST
PTN
PDC
PSP
UDN
PTB
PSD
135791113151719212325
1962
0.1.2.3.4.5.6.7.8.91PTB
PDT
PT
PMDB
PDS
13579111315171921232519
82
0.1.2.3.4.5.6.7.8.91PSB
PSC
PCdoB
PL
PCB
PDC
PT
PTB
PDT
PDS
PMDB
PFL
135791113151719212325
1986
0.1.2.3.4.5.6.7.8.91
PMNPSDPSTPRSPTRPCBPSC
PCdoBPSB
PLPDC
PTPTB
PSDBPDTPRNPDS
PMDBPFL
1357911131517192123252719
90
0.1.2.3.4.5.6.7.8.91
PRNPRP
PVPPSPSDPSCPMN
PLPSB
PCdoBPTBPDT
PTPP
PSDBPPRPFL
PMDB
1357911131517192123252719
94
0.1.2.3.4.5.6.7.8.91
PRONAPSLPST
PVPMNPSCPPSPSD
PCdoBPL
PSBPDTPTB
PTPPB
PSDBPFL
PMDB
1357911131517192123252719
98
0.1.2.3.4.5.6.7.8.91
PMNPSC
PSDCPSL
PRONAPSTPSD
PVPSB
PCdoBPPSPDTPTB
PLPSDB
PPBPT
PMDBPFL
1357911131517192123252720
02
0.1.2.3.4.5.6.7.8.91
PANPRB
PTdoBPHS
PRONAPMN
PSOLPTC
PVPSC
PCdoBPTB
PLPDTPPSPSB
PPPSDB
PFLPT
PMDB
1357911131517192123252720
06
0.1.2.3.4.5.6.7.8.91
PSLPTCPHSPRP
PRTBPSOL
PTdoBPMNPRBPPS
PVPCdoB
PSCPTBPDTPSB
PRPP
DEMPSDB
PTPMDB
1357911131517192123252720
10
0.1.2.3.4.5.6.7.8.91
PRTBPSL
PTdoBPEN
PSDCPTCPMNPRP
PSOLPTNPHS
PVPPS
PROSPSC
PCdoBPRB
SDDEMPDTPTBPSBPSD
PRPPPT
PSDBPMDB
1357911131517192123252720
14
0.1.2.3.4.5.6.7.8.91
−100%
−80
−60
−40
−20
0
+20
+40
+60
+80
+100%
Lege
nd:
seat
s lo
st/g
aine
d if
disr
egar
ding
alli
ance
s
− N
o in
fo.
Par
ties
that
won
at l
east
one
sea
t
N. of elec. circumscriptions where deputies were elected (min=1)
Line of cumulative share of seats in the Lower Chamber
49
by parties which were able to get candidates elected in less than two-thirds of the
country’s electoral circumscription. Almost half of the seats were elected by parties
that only won seats in less than half of the territory. It is possible to see, as well,
that the party system has seen little to no change over time, since the beginning
of the democratic periods. Also, the situation seems to have been pretty similar
between both periods.
Lastly, from 1994 to 2014, it is also possible to see that, generally, the parties
with worst nationalization of the outcomes were precisely the ones who have been
profiting with the possibility of establishing electoral alliances. They are the ones
who would loose seats if the distribution of seats disregarded these alliances. While
the parties with higher levels of nationalization of the outcomes have been generally
in the opposite position, it is, they would profit the most if electoral alliances were
disregard when TSE calculates the distribution of seats. This result is in line with
claims made by the Brazilian literature, that electoral alliances might have been
helping the small parties at expense of the big ones, in what regards the elections for
the Lower Chamber (Nicolau, 1996; Schmitt, 1999). In fact, it seems these alliances
have been helping also to de-nationalize the electoral outcomes, it is, to strengthen
the potential of regionalized parties to win seats, while harming the potential of
better nationalized countries to end up with a nationalized outcome.
1.2.5 Overall scenario
The Brazilian case shows vividly how important it is to split the concept of
nationalization into its different dimensions, as well as how important are the geo-
graphical units one wants to take into consideration. On the one hand, the system
is both almost perfectly nationalized in terms of party presence/organization if the
50
electoral circumscriptions are the parameter, and also has become over the years
almost perfectly nationalized in terms of electoral supply/cross-district linkage. On
the other hand, the electoral demand is much less nationalized, being actually among
the lowest nationalized in the democratic World, with the electoral outcome present-
ing even lesser degrees of nationalization. In summary, the Brazilian system has been
presenting an interesting discrepancy between the high nationalization of what po-
litical elites have to offer and do offer, and the low capacity of these elites to appeal
homogeneously to the electorates over the country.
1.3 Nationalization and fragmentation of party
systems in Brazil
The importance of studying the nationalization of parties and party systems can
be of multiple orders. One of the most relevant suggestions at the theoretical level,
and which has stayed mostly untested empirically, is Cox’s (1997) claim that the
cross-district linkage (or roughly what I’ve called nationalization of the supply) is
what makes the Duvergerian propositions move from the local to the national level.
For him, “if all candidates find it necessary to join a party that runs candidates in all
districts, then local bipartism will indeed turn into national bipartism” (1997:201).
While he only mentioned the first of the two Duverger’s propositions, i.e. the one
about the plurality vote, that reasoning should logically fits the other as well. In
fact, although less formally, he extends the idea to other system in a later work (Cox,
1999). It means that, generalizing it for other electoral systems, in Cox’s terms the
lack of cross-district linkage would tend to increase the national-level party system
fragmentation. The reason is that even in systems like proportional representation,
51
if parties that offer candidates were not the same in every electoral circumscription,
the final national party system would become even more fragmented than the average
(or median) fragmentation found within electoral circumscriptions.
That’s a particularly important issue for the Brazilian case. Beyond the debate
on nationalization, the fragmentation of party systems in the Legislative branch
has been one of the up-most concerns both for the Brazilian political literature
(e.g. Goes, 1992; Kinzo, 1989, 1993, 1997; Lamounier, 1992, 1994; Lima Jr.,1999;
Nicolau, 1996; Nicolau and Schmitt, 1995) and also for the international literature
on Brazil (e.g. Ames, 1995; Geddes and Ribeiro Neto, 1992; Mainwaring, 1991,
1993a, 1993b, 1999; Mainwaring and Scully, 1995; Sartori, 1994). Already in 1872,
Belisario de Souza, one of the earliest Brazilian scholars on electoral systems, feared
the adoption of proportional representation by the Brazilian Monarchy with an eye
in the consequences to the legislative party system fragmentation: “In a country
carved up by several parties or factions, that system would raise grave difficulties
(...) How to form legislative majority? No party would have a majority capable of
governing. Even if not coming to that extreme point, the legislative majorities could
be so minute that any government would be always in crisis, at the mercy of the
whims of its members and of the coalitions of these with the oppositionist groups”
(1979:133-134)15.
Such attention is not undue. Figure 1.8(a) brings the effective number of leg-
islative parties in every democratic election of 153 countries since 1945. The data
mixes my own data with Golder’s (2005) data, covering now more than one thou-
15In the original: “Num paıs retalhado por varios partidos ou faccoes este sistema traria gravesdificuldades (...) Qual o meio de obter maioria parlamentar? Nenhum partido reuniria maioriacapaz de governar. Nao chegando mesmo a este ponto extremo, as maiorias parlamentares poderiamser tao diminutas, que qualquer governo estaria sempre em crise, a merce dos despeitos dos seuscorreligionarios e das coligacoes destes com os grupos oposicionistas”.
52
Figure 1.8: Effective num. of parties at the Lower Chamber - Brazil vs. World andwithin Brazil
●
●●● ●●
●
●● ●●
●● ●
1945
1948
1951
1954
1957
1960
1963
1966
1969
1972
1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011
2014
1
3
5
7
9
11
13● ●Brazil Others
(a) International comparison:num. parties in Brazil vs. other countries
1945
1950
1954
1958
1962
1982
1986
1990
1994
1998
2002
2006
2010
2014
1
3
5
7
9
11
13
(b) Within Brazil: num. of parties in eachelec. circumsc. (boxplots) vs. whole country (line)
Year
Effe
ctiv
e nu
m. o
f leg
isla
tive
part
ies
at th
e Lo
wer
Cha
mbe
r
sand and three hundred elections. The plot shows Brazil has not only been having
the most fragmented Lower Chamber in the World, but such fragmentation is also
on the raise. Of course, the usual reasons called to explain such scenario have
to do with electoral system characteristics, like the PR system, the high median
electoral magnitudes of the electoral circumscriptions, the possibility of electoral al-
liances per circumscription, among others. However, Samuels (1998) has suggested
that the regional characteristic of the Brazilian parties would have been contribut-
ing too, which was endorsed by Cox (1999) and also pointed somehow by Nicolau
(1996), when commenting on the effect caused by the “federative characteristic” of
the Brazilian party systems.
Figure 1.8(b) brings the effective number of legislative parties for the Lower
Chamber within Brazil, i.e. in each of the electoral circumscriptions (represented by
boxplots) as well as a line with the overall number of parties in the country. There
are important differences between the two democratic periods. Firstly, during 1945-
1964 there was a scenario of much less fragmentation (contrary to what literature
diagnosed; see Britto, 1965; Santos, 1979; Lamounier, 1982; Lavareda, 1991) than
53
in the current period. Secondly, it is possible to see the party system fragmentation
has been raising steeply since the 80s. In the third place, there has been a great
difference between the overall number of parties in the country and the number of
parties for the Lower Chamber within electoral circumscriptions. Samuel’s argument
(1998) later followed by Cox (1999) was precisely based on that characteristic. The
idea is when the number of parties for the Lower Chamber in the whole country is
much greater than the average (or median) number of parties for the Lower Chamber
within electoral circumscriptions, it could be only a result of each circumscription
having dissimilar party systems. In other words, we could say that in my plot, it
means the distance between the black line and the median mark of the boxplots
would represent the final effect of some low nationalization on the party system
fragmentation. Hence, that figure alone is an evidence that in fact nationalization
has been of great importance to understand the party system fragmentation in
Brazil.
Still, for the sake of precision, the problem with these lines of reasoning is al-
though they make sense, they are often not theoretically specified and they have
never been tested in a more formal, rigorous fashion. I mention that they are of-
ten not specified in a theoretical level because it is usually not clear what authors
define by regionalized, nationalized of federative characteristic. Cox is an excep-
tion, but the question is: if it is a fact that the low nationalization affects party
system fragmentation in Brazil, it cannot be the dimension of nationalization that
Cox was more talking about. As we have seen previously, Cox was mostly talking
about politicians taking part in parties that would offer candidates all over a coun-
try. However, Brazil is a country with almost perfect nationalization of the electoral
supply. Almost all parties offer candidates for the Lower Chamber in all electoral
54
circumscriptions. Therefore, if there is a dimension of nationalization affecting frag-
mentation in Brazil, it has to be another one, namely the low nationalization of
electoral demand or of the electoral outcomes. Consequently, it would mean that
Cox’s proposition would need to be slightly reviewed. It is, if there is evidence that
some other dimension of nationalization affects party system fragmentation even
when cross-district linkage is perfect, we will then have a great example of major
theory in the literature that must be revised in light of a better detailing of the
concept of nationalization.
Additionally, when I mention that these blur relationships between nationaliza-
tion and fragmentation still lack a more formal test in the literature, I mean the
following. Looking at the difference between country-wide number of parties and
average/median number of parties within circumscription is a nice way of starting,
but cannot be the finish point. For one thing, a country can have the local party
systems distributed across the territory in infinite ways. Take my previous plot as
a reference. In case of perfect nationalization, who knows if the country-wide line
would approximate the median of the boxplot or some other point depending on
the magnitudes of the circumscriptions? In order to properly test the impact of
nationalization on party system fragmentation in Brazil, first of all I specify my
concept. Since nationalization in Brazil is very high in terms of party organization
and of electoral demand, I will focus on the other two dimensions of nationalization
that we have seen are not high in Brazil. Of course, to formally test whether such
lack of nationalization has an impact on the national party system fragmentation,
I cannot resort to historical data since Brazil had only 14 democratic elections for
the Lower Chamber during the country’s history. The solution I propose here is to
simulate how fragmented the Brazilian party system would be in terms of electoral
55
outcome in the hypothetical case of perfect nationalization of the electoral demand.
Fortunately, for the five elections from 1998 to 2014, the TSE provides the electoral
results for the Lower Chamber disaggregated at the level of the more than 450 thou-
sand polling sites16. I took these and programmed a script that does the following.
First, it picks one of these polling sites at random. Then, it extrapolates the rela-
tive electoral results of the picked polling site to all the others in the country. Next,
it calculates the electoral outcomes of each electoral circumscriptions, taking into
consideration the real observed parties and electoral alliances per circumscription.
Finally, it calculates the final overall effective number of legislative parties at the
Lower Chamber for that iteration. I repeat the procedure for around 20% of the
voting sites in each of the five elections.
The result, with the probability distribution of the simulated effective number of
legislative parties, can be found in Figure 1.9. Basically, what the density plots show
is how likely it would be for the effective number of legislative parties to decrease
in case of perfect nationalization of the electoral demand. The fairly biggest parts
of the distributions are to the left of the lines that mark the observed number of
parties. That makes it possible to conclude that in fact there would be a very high
probability the fragmentation would decrease in case of perfect nationalization of the
demand. In all of the five elections, actually, there is more than 83% probability that
the effective number of parties would diminish. The simulation also offers additional
information. It is possible to search for where is the peak of each distribution and
therefore to assess more or less what would be the most likely magnitude of change
in the number of parties. In general, it seems that it would be possible to expect
16I also tried the same approach, but just simulating across the 26 Brazilian states plus capitaldistrict, which are the electoral constituencies of the country. The results found support my claimeven more strongly. But due to the small N and to my attempt to be conservative, which meansto play against my hypothesis as much as possible, I sticked with the approach per polling sites.
56
Figure 1.9: Effec. num. of legislative parties at the Lower Chamber, according tosimulated perfect nationalization of the electoral demand
x=7.1
Pr=.83
2 4 6 8 10 12 14
(−72
%)
(−44
%)
(−16
%)
(+13
%)
(+41
%)
(+69
%)
(+97
%)
0.00
0.05
0.10
0.15
0.20
0.251998
x=8.5
Pr=.92
2 4 6 8 10 12 14 16
(−76
%)
(−53
%)
(−29
%)
(−6%
)
(+18
%)
(+41
%)
(+65
%)
(+88
%)
0.00
0.05
0.10
0.15
0.20
2002
x=9.3
Pr=.87
2 4 6 8 10 12 14
(−78
%)
(−57
%)
(−35
%)
(−14
%)
(+8%
)
(+29
%)
(+51
%)
0.00
0.05
0.10
0.15
0.20
2006
x=10.3
Pr=.83
2 4 6 8 10 12 14 16
(−81
%)
(−61
%)
(−42
%)
(−22
%)
(−3%
)
(+17
%)
(+36
%)
(+56
%)
0.00
0.05
0.10
0.15
0.20
2010
x=13.2
Pr=.85
2 4 6 8 12 16 20
(−85
%)
(−70
%)
(−55
%)
(−39
%)
(−24
%)
(−9%
)
(+6%
)
(+21
%)
(+36
%)
(+51
%)
0.00
0.05
0.10
0.15
2014
●simulatedobserved
Effective num. of legislative parties at the Lower Chamber(simulated % lost/gained)
dens
ity
a decrease in the number of parties in the Lower Chamber of the following order:
around -46% to -15% in 1998, around -30% in 2002, around -25% in 2006, around
-20% in 2010 and -15% in 2014.
These are, in most cases, fairly great decreases if one considers that effective
number of legislative parties are usually small quantities. Most countries in the
World, as we could see, have less than 5, in general around 3 effective parties at
their Lower Chambers. Therefore, a decrease of, say, 2 effective parties (which is
the number of parties in the US, for instance) in case of perfect nationalization
of the demand, is a result of substantive importance. As expected, these findings
regarding the Brazilian case exemplify the usefulness of better defining what one
57
understands nationalization and, especially, of splitting the concept into its further
dimensions. The application of Cox’s (1997, 1999) proposition would not directly
work in Brazil, which is a country with almost perfect nationalization of the electoral
supply, i.e. where parties are national in the sense that they offer candidates in
pretty much every electoral circumscription. Contrary to how Cox restates the
Duvergerian proposition, the cross-district linkage, if understood as the offer of
candidates throughout a territory, is not a sufficient condition for the party system
fragmentation within circumscriptions to become mirrored at the national level. It
is necessary to also look at the nationalization of the electoral demand and/or of the
electoral outcomes (depending on whether one is interested in the effective number
of electoral or legislative parties). Once we look at these dimensions, then Brazil
becomes a good example of low nationalization affecting party system fragmentation.
But at the same time, the result suggests that the impact of low nationalization
of demand or outcomes on the party system fragmentation in Brazil is diminishing
over time in relative terms. It is, the levels of nationalization are roughly the same
across these five elections and have an important impact upon the fragmentation,
but at a first glance it seems that such impact is decreasing rapidly. How can the
medium to low level of nationalization of the demand stay the same but its impact
diminish across elections? The reason is that the other factors affecting fragmenta-
tion have grown in importance. Mostly, this is a result that matches with what we
had just seen about the party system fragmentation within Brazil. The country has
been facing a considerable increase in the party system fragmentation for the Lower
Chamber within electoral circumscriptions. Therefore, despite nationalization re-
maining roughly constant, it is only natural that 1) the overall fragmentation would
increase if the within-circumscription fragmentation increases and 2) the relative
58
importance of nationalization for that fragmentation would decrease in face of the
increased relative importance of the within circumscription fragmentation.
1.4 Conclusion
In this chapter, it was my intent to achieve three aims. Firstly, to present a better
theoretical framework from which we could understand party system nationalization,
by further detailing what one means when talking about nationalization. Second, I
wanted to offer an example application for such conceptual detailing, by using the
case of the Brazilian party system. I hope I was able to show, trough the Brazilian
example, the importance of splitting the concept of nationalization. When we do
that, the political party systems of Brazil across the years have shown considerable
differences both over time and across different dimensions of party nationalization.
In 1945-64, the party system of the country was low nationalized in terms of electoral
supply, of electoral demand and of electoral outcomes (we do not have data on party
organization for any year of that period). But since the re-democratization in the
80s, the new party system has faced important changes.
Firstly, in the recent years for which data are available, the political parties
became almost perfectly nationalized in terms of organization if we consider the
electoral circumscriptions for the Lower Chamber (states plus capital), but are still
very badly nationalized if we go down in geographical unit disaggregation and take
a look at the number of cities where parties are present. Still, over time almost
all parties became perfectly nationalized in terms of electoral offer for the Lower
Chamber. Which means, they grew to offer candidates in every electoral circum-
scriptions, achieving the strongest cross-district linkage possible between politicians
59
across the country. However, although offering candidates wherever possible, the
parties have been unable to appeal to the electorate in a homogeneous fashion. For
the majority of the parties, the electoral demand in Brazil is not highly national-
ized, although for some of the biggest parties it is more nationalized. In the end,
the winner parties for the Lower Chamber are generally unable to win seats in all
electoral circumscriptions. On the contrary, the far majority of the seats belong to
parties that won seats in few parts of Brazil.
Therefore, the current Brazilian party system presents different scenarios de-
pending on which type of nationalization one takes into account. More than that,
depending on the concept of nationalization on choses, even the possible conse-
quences of party nationalization may change. As an example, I have shown that if
one extends to Brazil Cox’s (1997, 1999) proposition about the relationship between
cross-district linkage and party fragmentation, it would not work properly. While
Cox claimed that the offer of candidates was crucial to make Duvergerian proposi-
tions move from the local to the national level of countries, I show that in fact the
offer of candidates is not enough in the Brazilian case. It is necessary to take into
account the correct dimension of nationalization, which is the one of the electoral
demand, for Cox’s proposition to work. Hopefully, my propositions of better de-
tailing the concept of party nationalization will prove useful to help scholars in the
future. Particularly, in finding other applications for which results can be sensible
to the definition of what means to be nationalized.
60
1.5 Annexes
Annex 1.A – Formula of Bochsler’s index of Party and Party system
nationalization
Jones and Mainwaring (2003) were the first to proposed the use of the (inverse of)
Gini index of inequality to assess the degree to which a given party has equal/similar
electoral support across geographical units within a country. They claimed that “In
addition to being widely known and used, the Gini coefficient is technically superior
to most existing alternatives” (2003:142). However, the unweighted Gini coefficient
as they have proposed is expected to present more technical issues than previous
measures present in the literature, for at least two reasons. The plain Gini index is
at the same time sensitive to the number of geographical units being considered and
also does not account for discrepant population sizes among those units. Morgerstern
and Swindle (2009) warn that the first issue is especially problematic to the analyses
across countries, while the second affects specially the analyses within countries.
But not weighting each geographical unit by their population size also affects
cross-country comparison as it can greatly change the final nationalization score of
parties and of party systems. It is easy to understand the impact of not weighting
for different sizes of geographical areas. Geographical units with very few voters
should not penalize the final nationalization score in the case a given party had
abnormally few votes in there, as much as units with heavy proportion of voters
should penalize the score in case the same thing happened there.
Fortunatelly, Bochsler (2010) proposed solutions for both issues pointed above.
The insensitivity to different sizes of geographical units is solved by a weighted ver-
61
sion of the spatial Gini. While the sensitivity to the number of units is dealt with by
a post-calculation reweighting function. The weighted Gini calculated by Bochsler,
which he calls weighted party nationalization score (PNSw), has the following for-
mula:
PNSw = 1−Giweighted = 2 ∗∑D
i (popi ∗ (∑i
j ptyj − ptyi/2))∑Di popi ∗
∑Di ptyi
(1.1)
Where D is the total number of territorial units in the given country, ordered
according to the increasing vote share of the party pty being evaluated. Then, popi
is the voters population in district i and ptyi and ptyj are the votes of pty in districts
i and j, respectively. The resulting PNSw is, therefore, an inversed weighted Gini
index calculation to ensure that if a party has similar vote shares in all districts, the
index approaches 1. While in case it has very dissimilar vote shares across districts,
each district will lower the final score according to its share of the national voters.
After that, Bochsler’s proceeds to a final standardization for the countries’ num-
ber of districts. It s is what he calls weighted and standardized party nationaliza-
tion score (PNSsw), achieved by raising the former PNSw to the power of a given
weighting function (let’s call it S), which is proposed to be as follows:
PNSsw = (PNSw)S (1.2)
S =1
log10(E)(1.3)
E =1∑D
i (popi/popnat)2(1.4)
Where E is the effective number of territorial units in the given country, analo-
gous to the usual Laakso and Taagepera (1979) effective number of parties. Accord-
62
ing to Bochsler, this is used instead of the plain number of districts in the country
“since the units were weighted by their size (number of voters) for the calculation of
the Gini coefficient, they should be weighted as well for the standardisation in this
step” (2010:164). Therefore, the standardization follows an inversely proportional
base-10 log function of the effective number of districts. The standardization via
log function has the desirable effect of assuming “an increasing heterogeneity of the
vote as the number of districts rises, but with a decreasing marginal effect of the
number of districts. For example, splitting a single district into two should almost
always have a greater effect on decreasing static nationalization than moving from
fifty to fifty-one districts (or even fifty to sixty)” (Morgenstern et. al., 2014:4).
Lastly, the party system version of PNSsw, which is called PSNSsw, is just the
weighted mean of the PNSsw of each party in the given party system, with the
weights being the share of nation votes received by each party.
63
Annex 1.B – Parties’ organization: % of municipalities where parties
have presented candidates for the municipal Executive or Legislative
64
Chapter 2
Party system nationalization and the
national effective number of elec-
toral parties
It is surprising that party system nationalization has never been directly included
in the empirical models of party system fragmentation, beyond sparse side-related
efforts (e.g. Chhibber and Kollman, 1998, 2004; Cox, 1997, 1999). Certainly, schol-
ars have been offering increasingly elaborate treatments of the other covariates of
number parties. From the original sociological factors (Duverger, 1954; Grumm,
1958; Lipson, 1959; Lipset e Rokkan, 1969; Rose e Urwin, 1970; Campbell, 1989)
and from the dominant electoral institutions (Duverger, 1954; Rae, 1971; Sartori,
1976; Riker, 1982; Taagepera e Shugart, 1989, 1993; Lijphart, 1984, 1990), to the re-
cently modeled interaction between both (Amorim Neto and Cox, 1997; Coppedge,
1997; Jones, 1994; Ordeshook and Shvetsova, 1994; Taagepera, 1999; Penas, 2004;
Clark and Golder, 2006; Stoll, 2008). However, since we operationalize these co-
variates at the national aggregate level, missing what puts the electoral districts
together - to use Cox’s (1997) words – might be a problem.
The importance of what connects the results of each national electoral district
comes from the recognition that the national party system of a given country in a
given election is, in fact, a junction of the many (possibly dissimilar) party systems
65
that arise from each electoral district. Hence, the number of parties that are effec-
tively important (Laakso and Taagepera, 1979) in a country is, ceteris paribus, a
function of how homogeneous from one district to the other, i.e. nationalized, are
the electoral supplies offered by the parties, like Cox (1997) has claimed, but more
importantly the electoral demands represented by voters’ choices. Still, although
some role of party system nationalization in the causation of the effective number
of parties is sometimes acknowledged in a theoretical perspective (e.g. Cox’s cross-
district linkage), it has not been dealt with empirically. As I intend to demonstrate,
the risk of not including more specifically a measure for the degree of the nationaliza-
tion of the electoral demand in the empirical models of party system fragmentation
is to incur in omitted variable bias. A bias that makes many of the established find-
ings in the literature to become dubious. Nevertheless, I will propose that scholars
might have had their reasons for avoiding such proper specification of the models of
party system fragmentation. Mostly, they have been probably afraid of two types
of endogeneity that can reasonably exist between party system nationalization and
the number of national parties - the measurement endogeneity and the simultaneity
(reciprocal causality) endogeneity.
I will try to address both problems. Firstly, I will resort to the literature on
party system nationalization for an external Gini based measure (Bochsler, 2010)
that escapes the measurement endogeneity peril. As for the reciprocal causality
endogeneity, I follow the claim that the better way we can deal with such a menace in
the context of observational data is to model explicitly the reciprocity (c.f. Antonakis
et. al. 2010). By doing so, I will be able to include party system nationalization
as an explanatory variable in an empirical model of number of parties for the first
time. This will show that, first, in fact nationalization of the electoral support has
66
crucial role on the fragmentation of party systems. Second, by curing the traditional
omitted variable bias, the effects of some variables on the national number of parties
are changed. Namely, the famous effect of social diversity upon the party system
fragmentation changes from the usual direct effect pointed by literature to an indirect
effect only, which is mediated by the party system nationalization. In all the analysis
I will employ a new data on 62 countries with democratic electoral results since 1945
disaggregated at the electoral district level.
2.1 The omitted party system nationalization
At present, it seems firmly established that Duverger’s (1954) propositions, as
well as their consolidation by Cox (1997) in the M + 1 rule, operate at the electoral
constituency level only (cf. Leys, 1959; Wildavsky, 1959; Cox, 1997)1. It means that
it is within electoral districts2 that the number of seats at contest and the electoral
rules at play can restrict or permit the fragmentation of partisan choices made by
voters. These district-level electoral institutions would impose an upper limit to the
fragmentation of political choices that are demanded by - or at least related to - the
socio-political cleavages of these districts (Ordeshook and Shvetsova, 1994; Amorim
Neto and Cox, 1997; Coppedge 1997; Cox 1997; Penas, 2004; Jones 1994, 1997;
Taagepera, 1999; Clark and Golder, 2006; Geys, 2006).
However, because these effects all happen at the district level, each district of
a country might end up having different degrees of electoral choice fragmentation
1Although such acknowledgment took much longer to become common matter in the discipline,it is worth recalling that Duverger had established at the very beginning that “simple-majoritysingle-ballot system (...) tends to the creation of a two-party system inside the individual con-stituency; but the parties opposed may be different in different areas of the country” (1954:223).
2The terms electoral constituency, or even better electoral circumscription like I have used inthe previous chapter, are certainly more accurate than the widely used electoral districts. However,for consistency with the literature being reviewed and employed here, I will stick with the latter.
67
and, in fact, even different parties chosen by voters. By consequence, in reality the
final national party system and its degree of fragmentation would be the result of
the aggregation of these many district-level party systems that happen to be chosen
across the territory of a given country (Cox, 1997, 1999, Chhibber and Kollman,
1998, 2004, Hicken, 2009; Hicken and Stoll, 2008, 2011). For instance, even if
in a plurality system each district may have, in accordance to Duverger’s main
proposition, only two effective electoral parties, in case the two parties are not the
same across the districts, then the final national party system electoral fragmentation
would positively become greater than two. It is what happens, for instance, in
Canada and India (e.g. Riker, 1982; Gaines, 1999; Chhibber and Kollman, 1998,
2004).
Generalizing this idea to any type of electoral system, let D be the number of
districts d in a given country, each with magnitude Md where up to Md + 1 effective
parties tend to emerge (Cox, 1997). The final national effective number of parties
(ENEPnat) can range from the minimum value of Md + 1 if chosen parties were
the same in all districts; up to a theoretical maximum of∑D
d=1(Md + 1) if chosen
parties were different in each district. However, in practice, of course one can expect
ENEPnat to be very distant from this theoretical celling3, since approximating it
would require an unrealistically diverse party system across districts. Hence, while at
the district level the effective number of electoral parties (ENEPd) is approximated
by the Md + 1 rule, the ENEPnat will actually fall at some point of the interval:
min1≤d≤D
(Md + 1) ≤ ENEPnat ≪D∑
d=1
(Md + 1) (2.1)
3Empirically, it is certainly not usual for ENEPnat to approximate this ceiling∑D
d=1(Md + 1).In fact, it would be more precise to expect that (∃ENEPnat ∈ R+)(∀M,d ∈ N)|(ENEPnat �∑D
d=1(Md + 1)) ∝ (∑D
d=1(Md + 1) ≫ min1≤d≤D
(Md + 1))
68
At exactly which point of this interval, it will depend, therefore, on how similar
are the party systems that emerge in each of the national electoral districts. Which
means how nationalized it is, in the end, the country’s overall party system. The
most well-known theorization of that phenomenon comes from Cox’s commonly used
concept of cross-district linkage, that deals with the extent to which “(. . . ) would-be
legislators from different districts find it necessary or valuable to link together in
[common] national parties” (1997:201). For him, the decision of parties on entering
or withdrawing the competition in each district might have to do not only with
the conditions of the local (within district) competition, but might be also a sub-
product of candidates and parties bargaining across districts. Consequently, Cox
proposes that with regard to Duverger’s main proposition, “if all candidates find
it necessary to join a party that runs candidates in all districts [i.e. nationally],
then local bipartism will indeed turn into national bipartism” (1997:201). In other
words, Cox sees the nationalization of the electoral supply (offer of candidates) as a
necessary and a jointly sufficient condition for the Duvergerian propositions to work
at the national level. However, there is no reason to expect that a party that offers
candidates in all parts of a country would also necessarily gather a similar electoral
support everywhere. It is, while for parties who offer only a regionalized supply
of candidates it is logically impossible to end up having a nationalized electoral
support, for parties who offer a nationalized supply of candidates it is quite possible,
and actually common, to end up with a non-nationalized electoral support.
Consequently, it is of the nationalization of the electoral demand that party
system fragmentation is a direct function. Not of the nationalization of the electoral
supply/cross-district linkage. Probably one of the reasons why the focus has been
cast so much on the electoral supply, is the fact the appealing idea of a national
69
market for electoral coordination is much more difficult to be applied to voters.
It is dubious how much the voters would be able to coordinate strategically their
choices across the boundaries of their electoral districts as Leys (1959), for instance,
has imagined. More than that, it is also uncertain if such coordination would even
make practical sense (Cox, 1997, 1999). Still, even if it does not, there at least
two other ways of thinking about how the electoral demand can be more or less
nationalized. One has to do with the extent to which voters in different parts
of a given country deeply share their political identities and preferences. It has
to do, as such, with a sociological structural feature that has to do with what,
based on Schattschneider (1960) and using Kawato’s (1987) decription, we could
name as the over-time process of nationalization of the electorate. In Caramani’s
(1996:206) words, the “homogenisation of political characters (. . . ) [when] political
identities are moulded by wider environmental contexts, and parochial memberships
are replaced by cosmopolitan identities” (see also Sartori, 1976; Caramani, 1996,
2004; Chhibber and Kollman, 1998, 2004). The second way of thinking about de
degree of nationalization of the electoral support is how much parties are able to
evenly appeal to the electorate across the country in a given election. In fact, even if
there are no specific territorial cleavages in the electorate, parties may have diverse
electoral appeal in each area of the country, like it happens in Brazil. While even
if there are territorial cleavages, parties may be able to adapt accordingly in the
localized areas and end up having a nationalized appeal.
In fact, although usually forgotten, the exceptions Duverger (1954) himself made
to his propositions were a few countries with strong regional parties - such as pre-
war Belgium, Denmark and Sweden, as well as modern Canada. Noticing this
and reformulating the propositions to accommodate the exceptions, Rae probably
70
became the first to recognize explicitly the impact regional parties could have on the
functioning of Duvergerian reasoning: “Plurality formulae are always associated with
two-party competition except where strong local minority parties exist” (1971:95).
After him, others have suggested that, theoretically, regional parties (Riker, 1982;
Sartori, 1976; Geddes and Benton, 1997), regional social cleavages (Rose and Urwin,
1975; Kim and Ohn, 1992) or geographically heterogeneous party systems (Cox,
1997, 1999; Chhibber and Kollman, 1998, 2004) can raise the final national party
system fragmentation in comparison to what the electoral systems would allow to
expect otherwise.
However, while scholars may implicitly recognize such important role played by
party system nationalization, this is a dimension seldom accessed empirically in the
literature about party system fragmentation4. The following equation represents
the most general form of the model of number of parties given by the literature, i.e.
omitting the party system nationalization (PtyNat):
ENEPnat = β0 + βj
J∑j=1
Xj + ζ (2.2)
Yet, we know from theory that ENEPnat = f(PtyNat), so:
4This is especially problematic when we consider how scholars measure the quantities of interestin this field of study: it is rare that scholars do model such relationship empirically at the districtlevel (one exception is Geys, 2006, on the Swiss system). In reality, published works on thedeterminants of the party system fragmentation always resort (as I will do here, following them)to some sort of aggregation. They generally (and I will) model the final ENEPnat instead of eachdistricts’ ENEPd. They usually use (as I will) the national average/median of districts’ magnitudesas covariate, instead of each districts’ M . As well as they always employ (as I also will) socialdiversity measured at the national level of countries instead of at each electoral district of eachcountry. Consequently, a theory we know works at the local level can only be empirically testedby using national aggregations. One of the most important consequences of this problem, usuallyneither clear nor made clear, is precisely the fact that it leads us to end up encapsulating the issueof party system nationalization in our models and inadvertently hiding them from our analyses.It means, we empirically model ENEPnat, not ENEPd, but omit from the set of covariates theparty nationalization that has generated ENEPnat from the many ENEPd. To do that meansflirting with an especially problematic omitted-variable bias.
71
E[ENEPnat, P tyNat] 6= 0 (2.2a)
So if:
E[Xj, P tyNat] 6= 0 (2.2b)
Then:
E[ζ,Xj] 6= 0 (2.2c)
It means that by missing PtyNat at the right hand side, the usual models in
the literature are likely victims of endogeneity. The reason is that it only displaces
PtyNat (and its effect on ENEPnat) into the error term ζ. The consequence is that
βj found for covariate Xj will be biased in case such Xj is related to PtyNat. Even
worse, if as usual this is a model with more than one explanatory variable (J > 1),
we cannot even know the direction of the bias. Further mathematical proof is in the
Annex 2.A.
This can have important consequences for our previous knowledge about the
number of parties. Let’s assume the number of parties is indeed strongly related
to party system nationalization (what is quite likely according to theory) but we
model ENEPnat without using PtyNat as a covariate (as we know scholars have
been doing). At the same time, suppose that an important covariate such as social
diversity can be expected to affect both ENEPnat and PtyNat (what is reasonable,
as I will argue and demonstrate). How can we know whether the effect of social
diversity in the model of number of parties is not actually mostly or exclusively
related to the omitted variable party system nationalization?5
5Incidentally, the same applies the other way around, that is, for the literature on the predictorsof cross-district aggregation or of its related static party system nationalization. By not includingnumber of parties in equation 2.2, how can we know whether an explanatory variable in the model
72
It is hopefully clear at this point that to disentangle this, it is necessary to
include party system nationalization as an additional variable that explains number
of parties. Yet, there might be, of course, good reasons for scholars to have refrained,
for so long, from doing so. While to the best of my knowledge, surprisingly no one
has ever clearly stated a reason for such notorious absences, I would argue that
authors have been, quite likely, usually concerned with two other types of potential
endogeneity that could be present were they to include party system nationalization
as covariate of number of parties. One is the measurement endogeneity and the
other is the reciprocal-causation type of endogeneity. After handling both issues,
I will be able to test i) whether in fact nationalization of electoral support affects
party system fragmentation; ii) whether there is a reciprocal effect the other way
around; iii) if nationalization affects party system fragmentation, how its inclusion
in the model of number of parties might change the role of other covariates such as
established by literature (e.g. that of social diversity).
2.2 The endogeneity obstacles for including party
system nationalization
The concern about measurement endogeneity is, of course, easily justified if we
recall that literature has often assessed party system nationalization, cross district
aggregation or linkage by what has been called party inflation indices (Chhibber
and Kollman, 1998; Cox, 1999; Hicken, 2009; Hicken and Stoll, 2008, 2011). These
indices, from the original version proposed by Chhibber and Kollman (1998) to the
further enhancements made by Cox (1999) and by Moenius and Kasuya (2004),
of party system nationalization is not actually related to the omitted number of parties?
73
actually do not measure the cross-district homogeneity of partisan support in itself.
They measure the very impact that this has on the national number of parties,
since the core of all different versions of these indices rely on the subtraction of the
average number of parties at the districts from the total national number of parties
(ENEPnat −∑D
d=1ENEPd/D). The problem, here, is that since national number
of parties is a constituent part of the indices, party inflation obviously cannot be
used as an explanatory variable in models where the national number of parties is
the response variable6. In reality, this measurement endogeneity clearly precludes
any possibility of estimating a model where both national number of parties and
party inflation indices are present in different hand sizes of a same equation.
Let PtyInf represent party inflation measures. The following model cannot be
estimated as it is:
ENEPnat = β0 + β1PtyInf + βj
J∑j=2
Xj + ζ (2.3)
Because PtyInf ∝ ENEPnat.
Yet, we can opt for a different way to assess party system nationalization that
allows us to disentangle its effect on the national number of parties, instead of mixing
them together even more. There is quite a developed debate on the better ways to
measure the degree of homogeneity of the electoral support of parties and of party
systems (for a comprehensive summary of them, see Caramani, 2004; Bochsler, 2010;
6In fact, as a side note, I would also be cautious about the overall use of party inflation indicesas dependent variable as well. If our empirical knowledge on the determinants of the number ofparties was given by models at the district level as it should be, then there would be no problem,since in all versions of the inflation indices the
∑Dd=1ENEPd/D is subtracted and therefore goes
out. However, as our empirical knowledge actually comes usually from modelling ENEPnat, theconfounded effects when using party inflation indices as dependent variable can become unpre-dictable.
74
Morgenstern et. al., 2014). Most of the recent ideas are based on regional coefficients
of variations or regional measures of dispersion, what is good because none of these is
mathematically related to any usual way we measure number of parties (i.e. Laakso
and Taagepera, 1979). My option here will be for the increasingly used Bochsler’s
(2010) index of party nationalization. It is essentially the inverse of a weighted
regional Gini index of inequality calculated for the electoral support of parties across
electoral districts. There is no reason to suppose a priori that a Gini index formula is
endogenous to the national number of parties at the measurement or matemathical
levels.
Undoubtedly, however, it still leaves the problem of party nationalization and
national number of parties possibly being endogenous at the theoretical level. First,
because these two phenomena surely can be thought of as sharing political and
social determinants, as well as sharing omitted determinants. Secondly, and much
more interesting, because they can be reasonably thought of as reciprocally causing
each other. As I have mentioned, higher party system nationalization is usually
expected to decrease number of parties. But it is not absurd to think that more
effective parties in a country should difficult the nationalization of the party system
as well. For instance, more parties could mean increased difficulty in the cross-
district coordination, i.e. in deciding who enters and who quits competition across
electoral districts. As well as a greater probability that some party may try (and
eventually succeed) to conquer regionalized electorates of a country.
In the end, it still looks like a saddening choice between harmful omitted variable
bias and hurtful reciprocal endogeneity. It is not difficult to visualize such trade-off.
Consider that the conceivable reciprocal causation between number of parties and
party nationalization, specified in equation format, has the following general form:
75
ENEPnat = γ1,0 + γ1,1PtyNat+J∑
j=1
β1,jX′
j + ζ1 (2.4a)PtyNat = γ2,0 + γ2,1ENEPnat +
K∑k=1
β2,kX′′
k + ζ2 (2.4b)
Where γ represents the coefficients of the effect of one endogenous variable on
the other; β represents the coefficients of the effects of exogenous variables; X′j and
X′′
k are the sets of exogenous variables of each equation; and ζ1 and ζ2 represent the
error or disturbance terms of each equation.
One can easily notice why estimating the equation 2.4a alone would be prob-
lematic. The explanatory variable PtyNat would be correlated with ζ1, since an
increase in ζ1 would increase ENEPnat, who in turn would increase PtyNat in the
absent equation 2.4b, creating a loop. Hence, E(ζ1|PtyNat) 6= 0, meaning that
PtyNat introduces endogeneity in equation 2.4a. The same reasoning, of course,
is true for ENEPnat and ζ2 in equation 2.4b. This is a treacherous scenario for
researchers because it violates one of the most important assumptions for usual es-
timation techniques, since it means that a simple separate estimation of each of the
above will be biased and inconsistent. For further mathematical proof, see Annex
2.B. However, as we have seen, omitting PtyNat from equation 2.4a, as literature
has been doing, does not exactly solve the problem as well.
Fortunately, there is another option that improves on these two. One can stop
omitting PtyNat but, instead of estimating equation 4 only, one can estimate the
whole system of equations jointly. It means specifying explicitly the reciprocity,
hence testing and controlling for it, instead of vainly trying to avoid the problem.
76
Moreover, this also actually follows more closely the theoretical specification we
usually have in mind: if we think of the subject as a system of equations, why not
estimate it as such? In the econometric literature, these multi-equation systems
with reciprocal causation are known as nonrecursive simultaneous equation models
(c.f. Bollen, 1989; Kaplan, 2008; Greene, 2011, Wooldridge, 2010)7 and are con-
sidered one of the types of quasi-experimental techniques (Antonakis et. al., 2010).
This type of model can be estimated by the separate-equation instrumental variable
approach (2SLS, WSLS) or, as I will do here, truly simultaneously by iterating the
system to parcel out the endogeneity that the dependent variable of one equation
introduces in the other equation (3SLS, MLE)8.
For any of these options to become possible, each equation will need at least
one excluded covariate, i.e. at least one exclusive explanatory variable working as
an instrument, for the system to become mathematically over-identified9. If the
model includes a correlation between the error terms ζ1 and ζ2, then more than one
exclusive covariate will be necessary. There is also an additional advantage of using
a system of equations instead of estimating only equation 2.4a. More than just see
which covariates of ENEPnat still have significance when PtyNat is included, the
7A very good introduction for social scientists in general on the specification, estimation andassessment of nonrecursive models can be found in Paxton et. al. (2011).
8 Limited information techniques like 2SLS estimate one equation at a time, but using in-strumental variables to deal with the endogeneity. Their advantage is that they do not carrymisspecification error from one equation to the other. But they also disregard possible correlationsbetween the error terms of equations, i.e. between ζ1 and ζ2 in our system above. Full informationtechniques estimate the two (or more) equations at the same time, thus allowing for this possiblecorrelation between errors to be specified, but they also spread eventual misspecification error fromone equation to the other(s). In this research, I will opt to present results of ML estimations, as inone of the models presented for comparison I will specify correlation between the errors. However,all main models were also tested with 2SLS to check for misspecification robustness, and resultswere very similar.
9Although just-identification is enough for the estimations to be reliably performed, actuallywe want the models to be over identified. Otherwise, assessment tests become either unavailableor unreliable.
77
system of equations enables us to precisely find, calculate and test which are the
covariates that have direct effect on ENEPnat from the ones that have an indirect
effect through PtyNat.
2.3 Description of dataset and of variables
To make these analyses possible, I have built an original dataset10 that covers
nearly all democratic elections in 6211 countries, from 1945 to 2012. Starting from
1945, I have considered for inclusion any democratic period of all countries in the
world that had at least three consecutive democratic elections. The ones for which
data were found in time for this version of this chapter were included, covering all
continents and a wide variety of institutional, historical and social backgrounds. A
detailed list is in Annex 2.C. Notice that the cases I will work with are country-tier-
elections, not countries. For instance, each German election appears once for its
proportional tier, once for its single-member district tier12. Although this procedure
is not always used in the literature, it should be, once the effects of electoral rules
are supposed to be tier-specific. Of course, this choice can raise a concern about
the fact that different tiers in mixed systems quite possibly affect each other’s elec-
toral results (for a recent debate on this, see Crisp et. al., 2012), being therefore
inherently correlated. It is a justified concern, but one that can be dealt with if
authors resort to multilevel models or, in my case here, to clustered standard error
10The electoral data come from a broader original dataset that contains electoral results for eachparty, in each tier, disaggregated at the constituency level. Here, all partisan data will be usedaggregated for the party system at each election tier of each country.
11There will be 80 countries in the final version of the paper. Their data are already pre-processed, but there was not enough time to mount them into the dataset before this version ofthe paper was prepared.
12However, countries’ tiers that have only one nationwide constituency were, of course, droppedfrom the dataset, as it becomes pointless to calculate nationalization in such cases. The countrylist in the annex has a list of these dropped tiers.
78
Figure 2.1: Party system nationalization (PtyNat) with bootstrapped measure un-certainty and Effective number of electoral parties (ENEPnat), per country-tiers
estimation. It is always better to address this issue at the estimation stage, where
non-independence can be explicitly modeled, than to mix countries’ tiers by an ar-
bitrary procedure (average, summing, etc) at the data preparation level, or to throw
away data. Besides, models with many elections for each country should always
employ techniques that account for clustering anyway.
79
2.3.1 Endogenous variables
As previously explained, there will be two dependent variables, each also ap-
pearing as an explanatory variable of the other. The traditional effective number of
electoral parties as proposed by Laakso and Taagepera (1979) will be the measure
of party system electoral fragmentation at the national level (ENEPnat)13. Mean-
while, my measure of static party system nationalization (PtyNat) will be based on
the standardized Party Nationalization Score (PNSs) proposed by Bochsler (2010).
This score applies a weighted Gini index of regional inequality to the share of votes
each party receives in each electoral constituency, accounting for differences in pop-
ulation size across constituencies. Then, a log function is used to standardize it,
accounting for differences that exist in countries’ number of districts. I have used
bootstrapping to calculate the PNSs of each of the about 18300 parties-election-tier
covered by my data. Then, to get a party system version (PSNSs), I have used the
weighted average of all parties in a given election-tier-year. PNSs can range from
0 (total party system regionalization) to 1 (perfect party system nationalization).
Figure 2.1 shows the time-averaged PSNSs of each country-tier in the dataset, with
confidence intervals, as well as the time-averaged ENEPnat.Notice that the graphic
is ordered from the lowest to the highest estimate of PSNSs. It broadly confirms
the notion that to the extent PSNSs increases, the cloud of grey points formed by
ENEPnat tends to decrease. Additionally, this graphic shows that PNSs does not
vary according to the electoral type of tier being considered. This is good news.
Otherwise the very use of this measure would become problematic, because since we
13I have tried variants of this measure that claim to correct for the presence of the aggregated‘others’ category in the electoral results (e.g. Taagepera, 1997), even if few elections have suchcategory in my dataset. Using these alternatives did not yield any different results, so for the sakeof simplicity I opt for the usual index calculation.
80
already know electoral type of tier is strongly correlated with ENEPnat, it would
become unfeasible to disentangle the effects of PSNSs, ENEPnat and tier type
altogether.
2.3.2 Exogenous shared variable of interest
The canonical variable mobilized by scholars to explain countries’ effective num-
ber of parties, and which might also affect party nationalization, is social diversity.
A great variety of measures for it were proposed by the political and econometric
literatures, but roughly all are based on identifying the demographic size of the
linguistic, ethnic and, sometimes, religious groups present in each country. Then,
these indices calculate some sort of effective number of groups or its mathemati-
cally equivalent fractionalization index. From the many available, the measure I
will adopt as reference is the recent ethno-linguistic fractionalization measure based
on politically relevant groups, delivered by Cederman et. al. (2009). This is the
only recent dataset on ethnic groups in a sort of panel format instead of covering
a static point of countries in time14. However, Stoll (2008) has undeniably shown
that the role played by social diversity in the model of effective number of parties is
not robust to measurement variability, meaning that the measures of social diversity
we choose in fact alter the results we get. Therefore, I will also replicate my main
model other 19 times, each with a different measure of social diversity15. All of
14There were two older datasets that covered more than one time point. Krain (1997) hadspecific estimates of ethno-linguistic fractionalization for the 1940s, 1950s, 1960s and 1970s, whileRoeder (2001) used the 1960s and 1980s versions of the famous soviet data.
15For the sake of space, I will only replicate models using measures of ethnic, linguistic andethno-linguistic diversity, not religious. Examples of indices I will use from literature are: Selway’s(2011), Amorim Neto and Cox’s (1997) and Roeder’s (2001) ethno-linguistisc fractionalization;Alesina et. al.’s (2003) measures for Linguistic and Ethnic fractionalization; Fearon’s (2003) Ethnicdiversity index and Cultural diversity index ; Klaus et. al. (2012)’s Ethnologues’s ethnolinguisticfractionalization and Desmet et. al. (2012) linguistic diversity at language aggregation levels 1, 2,5 and 15. Regarding measures of religious fractionalization, literature on number of parties does
81
them in the format of fractionalization indices, i.e. the probability from 0 to 1 that
two individuals picked at random would not belong to the same social group. More
details on each social diversity measure will be found in the online supplementary
material.
2.3.3 Exogenous exclusive (instrumental) variables
The main covariate of effective number of parties in the literature is, of course,
some national aggregation of districts’ magnitudes. Likewise, here I employ the
usual average of magnitudes as a covariate in the ENEPnat equation, also adopting
the usual log transformation (ln(Mavg)). The justification of why it is also a good
exclusive variable of number of parties in the current setting is that there is no
reason - and no theory whatsoever that would suggest - to expect it to have any
direct effect on party nationalization. It means, there is no reason to expect the
average district magnitude of countries to impact on the party system nationalization
without first affecting the number of parties. It is true that many authors (e.g.
Caramani 2004; Hicken 2009; Bochsler, 2010; Hicken and Stoll, 2011) suggest that
number of districts may affect party nationalization; and since ln(Mavg) might be
related to number of districts, it could become important to include ln(Mavg) in
the models of party nationalization, therefore preventing it from being an exclusive
covariate of number of parties. However, as previously mentioned, because our
measure for PtyNat is standardized by the number of districts, this concern is not
necessary. Also, models that also include number of districts as control do not
present any difference in results. Moreover, additional tests with the inclusion of
number of districts or ln(Mavg) as control in the party nationalization equation
not often includes them and in any case, their inclusion does not change the results.
82
(therefore making the system just-identified) did not yield different overall results.
As for the party nationalization equation, the main instrumental variable will be
V trsGeoHomog, or the degree of how much homogeneous vs. concentrated is the
distribution of voters across the territory of a country in a given year. It is measured
using the same logic as PNSs, but for number of voters16. The main motive why this
can be used as an important instrumental variable for party nationalization is that,
first, it can be obviously expected that de degree of demographic regionalization
will affect the degree of nationalization of the political system. It is reasonable
to expect that regional parties will have extra incentives to emerge or to survive
where population is regionally concentrated or very unevenly distributed. Second,
V trsGeoHomog is a good instrument of party system nationalization in the setting
of this chapter because there is no reason to expect it to affect the number of parties
directly, it is, unless trough first affecting party system nationalization.
Two other variables will appear as shared variables in the first models and then
as exclusive covariates, one of ENEPnat and the other of PtyNat, in order to allow
for the comparison of different model specification.
The additional instrument for number of parties can be PersV oting, a bi-
nary identification of which country-tiers allow vote pooling, i.e. allow voters to
choose specific candidates instead of only parties (c.f. Karvonen, 2010; Norris 2002;
Colomer 2009). I have followed Renwick and Pilet (2011)’s typology17 and applied
it to information provided by Bormanna and Golder’s (2013). It is clear that the
16To anticipate concerns with possible measurement endogeneity between PtyNat and this mea-sure of V otersHomog, I have also tested the main models measuring this covariate differently.Instead of valid votes across electoral districts I have used demographic figures across the highestsubnational administrative divisions of countries. These data came from various editions of theThe World almanac and book of facts. As the general findings were the same, I have opted toreport here only the results using VotersHomog.
17They are: Open list PR, Block vote, Cumulative vote, Limited vote, SNTV, Single TransferableVote.
83
ability of choosing different candidates from a same party can alter the strategic
calculations voters do within their districts (e.g. Carey and Shugart, 1995). Du-
verger’s psicological effect can become much harder to happen, as strategic voting
may be harder to achieve (Cox, 1997). However, there is no established theory to
make us suspect a priori that it would affect party system nationalization.
While the additional instrument for party nationalization will be Federalism,
which is Gerring et. al.’s (2005) scale of unitarianism inverted and divided by two,
so in the end it ranges from 0 (descentralization) to 1 (strongly federative). Lijphart
(1994), Jones (1997), Geddes and Benton (1997) and Gaines (1999) offered some of
the theorizations about how federalism could be expected to affect the party system
fragmentation. The general idea is always that there is a “propensity of parties
in federal systems to split” due to “the viability of parties that play an important
role in provincial politics even though they have little weight nationally” (Geddes
and Benton, 1997:7). Which means, the expectation is that federalism would alter
the degree of nationalization of party systems and just then impact on the party
system fragmentation. There is no reason to expect that federalism would have a
direct impact on ENEPnat. On the contrary, it would be only indirectly, through
PtyNat. And that is precisely why Federalism should be safely included as an
exclusive covariate of PtyNat.
2.3.4 Other exogenous control variables
Additional exogenous controls will be also included in the model. UpperT ierSize
will be the percentage of seats distributed at compensatory upper tiers18, “(. . . )
18It is worth noticing that not necessarily compensatory upper tiers have to be neither nationwidetiers nor exclusive. Austria is a good example of country where after the primary proportionaltier where voters cast their votes, two other compensatory levels complement the seat distribution.One is an upper tier formed by macro regions and the next is a nationwide upper tier.
84
within which unused votes (and sometimes unallocated seats) from primary elec-
toral districts are aggregated and distributed” (Cox, 1999:157)Following Cox (1999),
Hicken and Stoll (2011) and others, I expect it to alter the incentives for cross-
district coordination of parties. First, because the more nationalized parties are the
ones who tend to better profit by the seats’ allocation at compensator upper tiers.
Second, because countries that have such tiers normally have nationwide electoral
thresholds for parties to be included in this last stage of seat distribution. Examples
of countries with upper tiers are Austria, Denmark, Norway, Sweden, Venezuela
before 2000, among others. President is a binary variable detecting presidential
systems, a control for the possible impacts of the existence of presidential powers
and presidential elections in a given country19.
Graphics with the statistical distribution of all variables can be found in Annex
2.D.
2.4 Results
Altogether, I will present results from 5 different models. Two of them will be
also replicated 19 times, as mentioned above, each with a different measure of social
diversity. The first model will serve for comparison purposes, as it is a naıve separate
OLS estimation of each equation in our system, one for ENEPnat omitting PtyNat
and another for PtyNat omitting ENEPnat. The other models are the proper
simultaneous equations, where the whole system of equations is estimated at the
same time by Maximum Likelihood estimation robust to non-normality (Asymptotic
19The literature about how these presidential system characteristics may impact on theENEPnat (Cox, 1997; Amorim Neto and Cox, 1997; Jones, 1994; Hicken, 2009; Hicken and Stoll,2011), on the party nationalization (Morgensterns et. al., 2009; Penas, 2004) or on both (Hockenand Stoll, 2011) is abundant. Therefore, President will be a variable included as shared covariatesin all models.
85
Figure 2.2: Path diagrams of the models 1 to 4
ENEPnatln(M)
𝜁1
Social Diversitynat
Upper TierSize
Models 1 (excludes bidirectional blue arrows) and 2 (includes bidirectional blue arrows) :
PSNSs
𝜁2
VotersHomog.
Federalism
Presidential
PersonalVoting
ENEPnatln(M)
𝜁1
Social Diversitynat
Upper TierSize
Models 3 (excludes dashed curved blue arrow) and 4 (includes dashed curved blue arrow):
PSNSs
𝜁2
VotersHomog.
Federalism
Presidential
PersonalVoting
-Note: endogenous paths are in blue and exclusive exogenous covariates are in red.
86
Distribution Free - ADF) and with clustered standard errors. Let’s recall that
simultaneous estimation of the two equations means that models 2 to 5 must have
excluded covariates in each equation working as instruments.
In more detail, model number 2 specifies only ln(Mavg) as exclusive covariate
of ENEPnat and V trsGeoHomog as exclusive covariate of PtyNat. Which means,
PersV oting and Federalism are specified as shared covariates of ENEPnat and of
PtyNat. This is important to show that besides not having theoretical justifica-
tions for V trsGeoHomog to affect directly ENEPnat or for PersV oting to affect
directly PtyNat, there are also no indication of such effects empirically. Models
3 and 4 specify PersV oting as an additional exclusive covariate of ENEPnat and
V trsGeoHomog as an additional exclusive covariate of PtyNat. Recalling, the rea-
son for that is we need more than one exclusive covariates per equation to estimate
and assess a model that allows covariance between the equations’ error terms (to
test the hypothesis of shared omitted variable). That model is the number 4, which
has the exact same specification as model 3 but allowing error covariance. Figure
2.2 presents the path diagrams of these fours models as form of visualizing each
specification.
Model 4 is a trickier model to estimate, because in case our additional instru-
ments are weak or theoretical grounds for considering the variables as exclusive
covariates are flawed, results of model 4 would become doubtful. Model 3 is more
parsimonious. However, testing the differences between a preferred model (the num-
ber 3) and its version with correlated disturbances, is a crucial step for the reliability
of results (see Antonakis et. al., 2010). The specialized software Mplus 7.11 was
used for estimation, with connection to R 3.0.2 through the package MplusAutoma-
tion.
87
Table 2.1: Simultaneous Eq. Models of the reciprocal relationship between partysystem nationalization (PtyNat) and national effective number of electoral parties(ENEPnat)
Model 1 Model 2 Model 3 Model 4OLS MLE MLE MLE
Coef. Rob.SE Coef. Rob.SE Coef. Rob.SE Coef. Rob.SE
ENEPnat ON:Intercept 2.73 (.31) 3.76 (.83) 3.87 (.78) 3.92 (.95)ln(PtyNat) -5.32 (2.46) -5.54 (1.93) -4.68 (2.70)ln(Mavg) .32 (.14) .23 (.11) .23 (.11) .26 (.13)PersVoting .78 (.36) .63 (.32) .61 (.30) .60 (.33)SocialDiversity 1.57 (.63) .51 (.66) .50 (.64) .70 (.76)President -.40 (.38) -.37 (.26) -.35 (.23) -.34 (.24)UpperTierSize .26 (1.60) 1.29 (1.60) 1.30 (1.56) 1.16 (1.61)Federalism .41 (.39) -.14 (.33)
R2: .15 .41 .41 .39
ln(PtyNat) ON:Intercept -.40 (.08) -.32 (.09) -.32 (.09) -.37 (.17)ENEPnat -.02 (.01) -.02 (.01) -.01 (.03)VtrsGeoHomog .33 (.09) .29 (.08) .29 (.08) .32 (.12)Federalism -.06 (.03) -.05 (.03) -.05 (.02) -.05 (.03)UpperTierSize .19 (.07) .19 (.06) .19 (.06) .18 (.07)SocialDiversity -.19 (.05) -.17 (.05) -.17 (.05) -.18 (.07)President .05 (.04) .04 (.03) .04 (.03) .04 (.04)PersVoting -.02 (.03) -.00 (.02)
R2: .29 .45 .45 .38
Disturbances covariance: fixed at zero fixed at zero fixed at zero -.03 (.07)
Over-all fit assessment:M.fit Chi2 p-value: .38 .79 .63CFI / TLI: 1.00 / 1.03 1.00 / 1.08 1.00 / 1.07Pr(RMSEA)≤.05 / SRMR: .73 / .01 .98 / .01 .93 / .01AIC / SampleSize adj.BIC: 1584 / 1612 1581 / 1606 1582 / 1608N 832 (both) 832 832 832
- Model 1 has separate equations estimated by OLS with clustered standard errors. Models 2 to 4 are systemof equations estimated jointly by Maximum Likelihood, with standard errors and chi-square test statisticsthat are robust to non-normality and non-independence of observations (clustered).- In Model 2, the exclusive covariate of ENEPnat is ln(Mavg) and the exclusive covariate of PtyNat isV trsGeoHomog. In Models 3 and 4, the exclusive covariates of ENEPnat are ln(Mavg) and PersV oting,while the exclusive covariates of PtyNat are V trsGeoHomog and Federalism.- Parameters with p-values greater than 0.10 are in grey, to ease visualization of inference.- In models 2 to 4, equation-level R2 are the Bentler-Raykov’s (2010) adjusted version for explained variancein nonrecursive models.- In these models, the measure for Social Cleavages is the ethno-linguistic fractionalization index based onpolitically relevant groups (Cederman et. al., 2009).
88
The Table 2.1 has the results of models 1 to 4. Model 1 gives the results with
the omitted variable bias, resembling the general approach that is common in the
literature. First, notice that even this naıve approach gives some support for my
theoretical expectations about the variables that should be exclusive covariates in
the next models. For instance, as expected ln(Mavg) and PersV oting do not have
statistically significant effects on PtyNat, only on ENEPnat even under this omitted
variable bias. The same can be said about Federalism, which seems to affect
only PtyNat20. A result confirmed by model 2 under the simultaneous equation
framework, where PersV oting and Federalism are still treated as shared variables,
being included in both equations. That is what is changed in Models 3 and 4, where
PersV oting is now also specified as an exclusive covariate of ENEPnat (in addition
to the exclusive covariate ln(Mavg)) and Federalism is now also specified as an
exclusive covariate of PtyNat (in addition to V otersHomog).
Model 3 has, in general, the same results as model 2, but it is the only model
where the direct effect of ENEPnat on PtyNat is also statistically significant. It
means, where there is ready evidence of the reciprocal causality between number of
parties and party nationalization. Model 4 shows, for instance, that this very same
model specification, but allowing the possibility of covariance between the error
terms of the ENEPnat and of the PtyNat equations, once again makes the reciprocal
causation not significant. However, there is evidence that the error covariance is
not statistically significant. What means that we cannot reject the null hypothesis
that ENEPnat and PtyNat have no additional shared covariate left out from our
model. Or phrasing it differently, we cannot reject the null that the model is not
20Of course, also V otersHomog is clearly a covariate with no expectation to affect directlyENEPnat, what is confirmed by alternative model specifications not presented in here due tosimilarity of result
89
omitting additional shared covariates between equations. Of course, this does not tell
anything about one-equation omitted variables, only about shared omitted variables.
Also, it can not be read as a positive proof that omitted variables do not exist -
which is something that can not be directly tested. It should be read as a negative
evidence: the model lacks evidence that shared omitted variables exist.
It also means that, as there is no evidence of error covariance, we should stay
with the more parsimonious models, like models 2 or 3. The chosen one is the
number 3, since it has the best fit, as it can be seen by the higher probability that
RMSEA is lesser than .05, by the greatest Chi2 p-value and by the lower CFI and
TLI measures21.
The greatest difference across models was the effect of ENEPnat on PtyNat,
i.e. the existence of reciprocal causation. This result proved to be quite sensible to
model specification. But in general, the models with better fit showed a statistically
significant reciprocal path22. This should be enough to claim that in general lines,
to include PtyNat in the models of ENEPnat without either modeling the recipro-
cal causality or at least using instrumental variable estimation, would mean to risk
having unaddressed endogeneity depending on the rest of the model specification.
21Talking about model selection, models 2 to 4 all have similarly good CFI, TLI and SRMRabsolute fit statistics for the system of equations, as well as reasonably good equation-level Bentler-Raykov’s (2010) R2 - which is an adjusted version for nonrecursive systems. In addition, allthe three models pass safely in the Chi2 tests of model fit, rejecting the null hypothesis thatmisspecification issues would have affected the fit to the data. Under ML, these Chi2 tests are alsothe tests for validity of instruments (Antonakis et. al., 2010), showing that the set of instrumentsin the two models are statistically valid in all our simultaneous models. Lastly, the AIC and BICmeasure of relative model fit are quite similar between models. Still, model 3 has a much betterprobability that RMSEA is lesser than 0.05, what is a sensible and therefore powerful indicationof absolute fit. As well as it has the greatest Chi2 p-value. Therefore, and because its version withdisturbance covariance proved to be not necessary, model 3 will be used hereafter as the preferredmodel specification. As robustness checks, additional specifications with different combinations ofexcluded variables were tried and yielded the same general results.
22In the Annex 2.E, I show that different expectations of measurement error for PtyNat alsobring different results for the reciprocal causation. To the extent we assume more measurementerror, it becomes more likely that the reciprocal path is statistically significant.
90
However, the results make even clearer that not including it at all is definitely a
problem. Looking back at model 1, it can be seen that in the usual OLS framework
with omitted variable bias, some shared covariates have statistically significant ef-
fect on both ENEPnat and PtyNat, like SocialDiversity and Presidential. Yet,
once we address the omitted variable bias by introducing ENEPnat and PtyNat as
covariates of each other, thus also explicitly specifying the possible reciprocal cau-
sation, the results for these shared covariates change quite radically. In all models
2 to 4 the direct effect from Presidential became only statistically significant to
ENEPnat, not anymore to PtyNat. While, even more important for the canonical
literature on this field, SocialDiversity loses the statistical significance of its direct
effect to ENEPnat, as well as the strength of such coefficient is halved. At the same
time, the direct effect of SocialDiversity to PtyNat stays significant, with similar
strength and expected sign.
This is a clear indication that party nationalization, or cross-district linkage, have
a mediator role for social diversity. It means, the effect of social diversity on number
of parties only comes through first altering the party nationalization. But instead of
just conjecturing such indirect effects, an additional gain we have for modelling the
relationship between ENEPnat and PtyNat through a system of equations is that we
can easily decompose the direct and indirect effects of all covariates, as well as their
statistical inference. Assessing that is important, because a significant relationship
between the endogenous variables not necessarily means that the covariates of one
of them have indirect effects on the other. Conversely, a non-significant relationship
between the endogenous variables not necessarily prevents indirect effects from being
relevant. Consequently, the coefficients of indirect effects must have their inference
tested separately. I present in Table 2.2 the indirect effects from all variables in
91
previous model 3.
Overall, we can see in this table that no covariate of ENEPnat has an indirect
effect on PtyNat , but all important covariates of PtyNat do have statistically
significant indirect effects on ENEPnat. A first interesting finding among these
covariates is the result of V trsGeoHomog. It shows that, actually, even the degree
of demographic regionalization of population in a country ends up strongly affecting
the national number of parties indirectly, what is quite possibly something we are
not used to think about. Even more importantly, as we can see social diversity and
the size of upper tiers, two commonly used variables to explain the number of parties,
are in fact only indirectly related with it. Compare the results to the model number
1 in the previous figure, with the results of omitted variable biased OLS estimation.
The indirect effects of SocialDiversity and of UpperT ierSize and Federalism on
ENEPnat are of similar strength to their direct effects on ENEPnat when PtyNat
is omitted. Therefore, their effects indeed exist, are significant and are quite strong.
However, they reach the national number of parties only by altering the party system
nationalization, not directly as we may be used to think, especially in the case of
SocialDiversity. The theoretical underpinning of this is important. It means that,
at least as far as the available measures can tell, we actually do not have evidence
of social diversity being relevant for the within district electoral coordination. The
evidence that scholars in general have been collecting seems to point, in reality, more
to the role that social diversity has in changing the cross-district coordination.
Recall, nonetheless, that we know since Stoll (2008) that different measures of
social diversity can bring different results about its impact on the number of parties.
Hence, we still have to check whether the results I have found are robust to different
measures of social diversity. I replicate model three 19 times, and in each I have
92
Table 2.2: Indirect effects to effective number of electoral parties (ENEPnat) andto party ystem nationalization (PtyNat) in previous Model 3
Indirect effects to: ENEPnat ln(PtyNat)Indirect effects from: Coeff. R.SE p-val Coeff. R.SE p-val
ln(Mavg) .02 (.02) .13 .00 (.00) .25PersVoting .06 (.05) .19 -.01 (.01) .33SocialDiversity 1.09 (.52) .04 -.03 (.02) .19President -.27 (.24) .27 .01 (.01) .24VtrsGeoHomog -1.77 (.88) .04 .03 (.01) .02Federalism .30 (.18) .10 .00 (.00) .16UpperTierSize -1.04 (.43) .02 -.00 (.02) .86ln(PtyNat) -.57 (.23) .01 .10 (.05) .03ENEPnat .10 (.05) .03 .00 (.00) .35
- Model was estimated by Maximum Likelihood, with standard errors robust to non-normality and non-independence of observations (clustered).- Parameters with p-values greater than 0.10 are in grey, to ease visualization of inference.- In this model, the measure for Social Cleavages is the ethno-linguistic fractionalizationindex based on politically relevant groups (Cederman et. al., 2009).
used a different variable to measure social diversity. Figure 2.3 shows, from these
replication models, the coefficients of the direct effects of SocialDiversity in each
equation. Other coefficients are omitted for the sake of space.
The difference is quite clear. While the impact of SocialDiversity on the number
of parties is almost never statistically significant, its impact on the party system
nationalization is almost always significant, with expected sign, and quite strong.
There are only two exceptions. One is Desmet et. al. (2012)’s measure of linguistic
fractionalization at their level 1 of linguistic tree aggregation. But even though,
this has clearly to do with their procedures to qualify language diversity according
to different levels of language in the Ethnologue’s language tree, since if we use
their measures at any other of their 14 levels, results are again consistent with other
social diversity proxies. The other only exception is Krain’s (1997). With this result,
it seems clear that, fortunately, the results presented here do not suffer from the
measurement sensitivity found by Stoll (2008) regarding the original debate.
93
Figure 2.3: Coefficient of the direct effects of SocialDiversity in 19 replicated ver-sions of previous Model 3, each with a different measure for SocialDiversity
−2 −1 0 1 2Coefficient
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●Classic Soviet EF (Fearon, 2003)ELF at the 1960s (Roeder, 2001)
Ef (Krain, 1997)ELF (Taylor and Hudson, 1972)
LF (Anckar et. al, 2003)EF (Anckar et. al, 2003)
EF (Annett, 2001)LF at Level 15 (Desmet et.at., 2012)LF at Level 10 (Desmet et.at., 2012)
LF at Level 2 (Desmet et.at., 2012)LF at Level 1 (Desmet et.at., 2012)
Ethnologue's ELF (Klaus et.al., 2012)Culture [Ethnoling] Frac.(Fearon, 2003)
EF (Fearon, 2003)LF (Alesina et.al., 2003)
EF (Alesina, 2003)ELF at the 1980s (Roeder, 2001)ELF (Amorim Neto & Cox, 1997)
LF (Selway, 2011)Pol.relevant EF (Cederman et.al., 2009)
ENEPnat
−0.3 −0.2 −0.1 0.0Coefficient
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
PSNSs
Direct effectfrom SocialDiversity to:
Models:
Lastly, I now introduce my model number 5, which is also exactly the same
as the preferred model 3, but it includes an interaction between ln(Mavg) and
SocialDiversity in the ENEPnat equation. It is important to try this specifica-
tion since it has recently become the standard way of translating the main theory
on the determinants of the number of parties. It means, the idea that the permis-
siveness of the electoral system works as a break or an incentive for latent social
and political cleavages to increase the number of parties (Duverger, 1954; Amorim
Neto and Cox, 1997; Ordeshook and Shvetsova, 1994; Clark and Golder, 2006; Stoll,
2008). Again, I replicate this model 19 times, each with a different measure of so-
cial diversity. Figure 2.4 shows the conditional direct effects of SocialDiversity on
94
ENEPnat for each value of ln(Mavg), in each of the replicated versions of model 5.
Figure 2.4: Conditional direct effects of Social Cleavages on the effective numberof parties (ENEPnat), in 19 replication versions of Model 5, each with a differentmeasure for SocialDiversity
−4−2
0246
1: Pol.relevant EF (Cederman et.al., 2009)
2: LF (Selway, 2011)
3: ELF (A. Neto & Cox, 1997)
4: ELF at the 1980s (Roeder, 2001)
−4−2
0246
5: EF (Alesina, 2003)
6: LF (Alesina et.al., 2003)
7: EF (Fearon, 2003)
8: Culture [Ethnoling] Frac.(Fearon, 2003)
−4−2
0246
9: Ethnologue's ELF (Klaus et.al., 2012)
10: LF at Level 1 (Desmet et.at., 2012)
11: LF at Level 2 (Desmet et.at., 2012)
12: LF at Level 10 (Desmet et.at., 2012)
−4−2
0246
13: LF at Level 15 (Desmet et.at., 2012)
14: EF (Annett, 2001)
15: EF (Anckar et. al, 2003)
16: LF (Anckar et. al, 2003)
−4−2
0246
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
17: ELF (Taylor and Hudson, 1972)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
18: Ef (Krain, 1997)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
19: ELF at the 1960s (Roeder, 2001)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
20: Classic Soviet EF (Fearon, 2003)
natural log of the average District Magnitude
Mod
erat
ed d
irect
effe
ct o
f Soc
ialD
iver
sity
on
EN
EP
nat
Figure 2.4 clearly shows that, once we include party nationalization in the equa-
tion of number of parties and also account for the possible reciprocity between them,
there is no unmediated (direct) effect of social diversity on ENEPnat, not even mod-
erated by districts’ magnitudes. Again, all effect of social diversity on the number of
parties actually is mediated by party system nationalization, coming to ENEPnat
only indirectly. The only exception is the measure of social diversity called Culture
95
Diversity, by Fearon (2003). It is an ethnic fractionalization index that takes into
account cultural distance of languages and is the only that has part of its condi-
tional effect not crossing the zero line in the graphics above. Even though, only
the very beginning of its plot is significant, meaning that only in cases with low
average magnitudes social diversity measured by such measure would have a real
impact on ENEPnat. Besides, it has the opposite direction than expected. Greater
average district magnitude appears to decrease the conditional effect of this measure
of social diversity on ENEPnat, what is inconsistent with the theory. Therefore, it
is clear that the results point in the general direction found in the previous figures,
i.e. social diversity only affects number of parties by first affecting party system
nationalization.
2.5 Conclusion
Primarily, I have shown a way through which we can start including party sys-
tem nationalization in the models of number of parties. I argue that, actually, no
empirical research modeling number of parties at the national level should ever omit
such a constitutive term. However, although I have found evidence suggesting that
there may be a reciprocal causation between party nationalization and number of
parties, thus leading to simultaneity endogeneity, I have also pointed that this result
is non-robust to different specification and is weak. Still, researchers that want to
stop omitting party nationalization from the models of number of parties should not
risk the possibility that their specification is suffering from endogeneity.
To deal with it, I have proposed that researchers can model this through a si-
multaneous system of equations. But of course, considering the lack of strong and
96
consistent evidence on the reciprocal causation, one could also model number of
parties with party nationalization as covariate and estimate the model using sim-
pler instrumental variable techniques. I would claim that this is probably the most
generalizable solution: to not omit cross-district linkage from the models of national
number of parties, but to estimate always in an instrumental variable framework.
I have even offered here some initial suggestions of good instruments for party na-
tionalization. Measures of federalism seem to be one. But above all, measures of
geographical distribution of population or of voters seem to be the best options. In
addition, if a researcher is operating in a simpler framework, out from the simul-
taneous equation world, another obvious option would be using a temporal lag as
instrument (getting closer to an Arellano and Bond approach). I hope these first
findings can help scholars to feel freer to explore directly the relationship between
number of parties and cross-district coordination.
Furthermore, I have shown that all this effort is not a detail. By including party
system nationalization in the model we actually alter consolidated results present
in the literature. Or rather, we displace them to their correct location. Given
the evidence usually mobilized by the literature, national-level measures of social
diversity and upper tier size are variables that do not affect strategic voting at the
electoral level. They change the degree of cross-district homogeneity of partisan
electoral support. They change party system nationalization. It is not my intention
to make the bolder theoretical claim that social diversity does not matter at all
within electoral districts. They may have a role there as well, it is just that we
really have no evidence as we usually think we do. The evidence that we usually
relied on is spurious. In the end, it is only a matter of defending once more the idea,
now empirically, that we should not make local conclusions using our aggregate data.
97
Interestingly, we still lack good theories about why and how social diversity would
alter the level of cross-district linkage of parties, candidates and voters. Maybe this
is a good start to incentive future efforts in this direction.
98
2.6 Annexes
Annex 2.A - Proof of endogeneity due to omitted variable bias in
models of ENEPnat that do not include PtyNat
Suppose that, as usual, a researcher does not include party system nationalization
(PtyNat) in a model of number of parties (ENEPnat), like the following:
ENEPnat = β0 + β1SocDiversity +J∑
j=2
βjXj + ζ (2.5)
Where SocDiversity is the social diversity (but could be any other variable of
interest), Xj is the vector of J additional explanatory variables and ζ is the distur-
bance term. The assumption of regular regression estimation is that all variables at
the right hand side of the equation are not correlated with ζ.
However, theory clearly tells us that E[ENEPnat|PtyNat] 6= 0. Therefore, be-
cause PtyNat was not explicitly included in the initial equation, it is actually being
absorbed by ζ:
ζ = γ1PtyNat+ ε (2.6)
Where ε is the true disturbance term of ENEPnat, were PtyNat included in the
model as a covariate.
Now suppose that cov(PtyNat, SocDiversity) 6= 0, i.e. that they are correlated.
Because PtyNat is within the error term ζ, the consequence is that SocDiversity
ends up being correlated with this error term as well.
To understand why, we substitute the formula of ζ into cov(SocDiversity, ζ):
99
cov(SocDiversity, ζ) = cov(SocDiversity, γ1PtyNat+ ε) (2.7)
= cov(SocDiversity, γ1PtyNat) + cov(SocDiversity, ε)
Here the last term can be canceled out since the expected covariance between
SocDiversity and ε is zero, otherwise it would mean that there are additional omit-
ted variables confounding the effect of SocDiversity on ENEPnat.
Then,
cov(SocDiversity, ζ) = γ1cov(SocDiversity, P tyNat) (2.8)
Therefore, if γ1 6= 0 (i.e. PtyNat indeed has an effect on ENEPnat) and if
cov(SocDiversity, P tyNat) 6= 0 (i.e. PtyNat is related to SocDiversity), then,
not including PtyNat in the original equation would make SocDiversity become
endogenous, because of being related to the error term, i.e. cov(SocDiversity, ζ) 6=
0. This scenario will yield biased and inconsistent estimation of β1, that is, of the
effect of SocDiversity on ENEPnat.
In order to see why, it is easier to represent the terms β1SocDiversity+∑J
j=2 βjXj
with matrices. Let X1×p be a matrix of all p explanatory variables in the model while
Bp×1 is the matrix with the coefficients of the effects of these variables on ENEPnat.
The parameter estimates will be usually given by the following formula:
B = (X′X)−1X′Y (2.9)
Then we substitute Y:
100
B = (X′X)−1X′(XB + γ1PtyNat+ ε)
= (X′X)−1X′XB + (X′X)−1X′PtyNatγ1 + (X′X)−1X′ε (2.10)
= B + (X′X)−1X′PtyNatγ1 + (X′X)−1X′ε
Since the expectation of ε is zero, once its mean is assumed to be zero, the last
term of this equation can be canceled out. So,
B = B + (X′X)−1X′γ1PtyNat (2.11)
Where (X′X)−1X′γ1PtyNat is the bias that makes B deviate from B. Notice that
such bias depends on both γ1, i.e. the effect of PtyNat on ENEPnat, and also on the
correlation between PtyNat and the other explanatory variables being considered,
due to the term (X′X)−1X′PtyNat.
Besides, as the correlation between SocDiversity and the omitted PtyNat has
the same sign as the correlation between PtyNat and ENEPnat, i.e. a negative
sign, we can also know in advance that the bias in the estimation of the effect of
SocDiversity on ENEPnat in the original model with omitted variable is positive.
That means, the omission of PtyNat artificially inflates the result of the effect of
SocDiversity.
101
Annex 2.B - Proof of endogeneity in models of ENEPnat that include
PtyNat, if they do not explicitly model the possible reciprocal causa-
tion between these variables
Suppose that a researcher naıvely includes party system nationalization (PtyNat)
in a model of number of parties (ENEPnat), like the following:
ENEPnat = γ1,0 + γ1,1PtyNat+J∑
j=1
β1,jX′j + ζ1 (2.12)
Where X ′j is the set of additional explanatory variables and ζ1 is disturbance
term. The assumption of regular regression estimation is that all variables at the
right hand side of the equation are not correlated with ζ1.
However, although not specifying it, the researcher is afraid that the following
relationship may also be true:
PtyNat = γ2,0 + γ2,1ENEPnat +K∑k=1
β2,kX′′k + ζ2 (2.13)
Where X ′′k is the set of exogenous variables of this equation and ζ2 is its distur-
bance term.
The proof that E[PtyNat|ζ1] 6= 0 in the first equation is as follows. First we
isolate PtyNat, what I will do here by substituting the first equation into the second.
In this step, I will omit the intercepts of both equations to simplify the mathematical
steps. Dropping the intercepts for this proof is usual practice and does not alter
anything. So,
102
PtyNat = γ2,1
(γ1,1PtyNat+
J∑j=1
β1,jX′j + ζ1
)+
K∑k=1
β2,kX′′k + ζ2 (2.14)
PtyNat = γ2,1γ1,1PtyNat+ γ2,1
J∑j=1
β1,jX′j + γ2,1ζ1 +
K∑k=1
β2,kX′′k + ζ2 (2.15)
PtyNat− γ2,1γ1,1PtyNat = γ2,1
J∑j=1
β1,jX′j + γ2,1ζ1 +
K∑k=1
β2,kX′′k + ζ2 (2.16)
PtyNat(1− γ2,1γ1,1) = γ2,1
J∑j=1
β1,jX′j + γ2,1ζ1 +
K∑k=1
β2,kX′′k + ζ2 (2.17)
PtyNat =1
1− γ2,1γ1,1
(γ2,1
J∑j=1
β1,jX′j + γ2,1ζ1 +
K∑k=1
β2,kX′′k + ζ2
)(2.18)
Assuming γ2,1γ1,1 6= 1, we can transform the above into the reduced form equa-
tion for PtyNat:
PtyNat = Π2,1X′ + Π2,2X
′′ + Z2 (2.19)
Where:
Z2 =γ2,1ζ1 + ζ21− γ2,1γ1,1
(2.20)
As a result, if the reciprocal path in fact exists, i.e. γ2,1 6= 0, then PtyNat is
a function of ζ1. It means that PtyNat and ζ1 are not independent in the first
equation. In more detail, let’s see why cov(PtyNat, ζ1) 6= 0. We substitute the last
structure form equation for the PtyNat inside this cov function:
103
cov(PtyNat, ζ1) = cov
(γ2,1
∑Jj=1 β1,jX
′j + γ2,1ζ1 +
∑Kk=1 β2,jX
′′k + ζ2
1− γ2,1γ1,1, ζ1
)(2.21)
= cov
(γ2,1
∑Jj=1 β1,jX
′j
1− γ2,1γ1,1, ζ1
)+ cov
(γ2,1ζ1
1− γ2,1γ1,1, ζ1
)(2.22)
+cov
(∑Kk=1 β2,jX
′′k
1− γ2,1γ1,1, ζ1
)+ cov
(ζ2
1− γ2,1γ1,1, ζ1
)
By definition, the first, third and fourth terms in the above equation are equal
to zero. The first and the third because it is assumed that cov(X ′j, ζ1 = 0) and that
cov(X ′′k , ζ1) = 0, since they are specified as exogenous variables. The fourth term
because here we are also assuming cov(ζ1, ζ2) = 0, which mean a model without
shared omitted variables for the sake of simplicity. One of the models in the paper
considers the possibility that this is not the case, but its results show that such null
specification between the error terms holds true. Going further and canceling these
three terms here, we end up with:
cov(PtyNat, ζ1) = cov
(γ2,1ζ1
1− γ2,1γ1,1, ζ1
)(2.23)
=γ2,1
1− γ2,1γ1,1V ar(ζ1) (2.24)
Therefore, provided that there is a reciprocal path, i.e. γ2,1 6= 0, cov(PtyNat, ζ1)
will certainly be different from zero as well. Thus, PtyNat will be endogenous.
Of course, all the reasoning in this proof also applies for ENEPnat and ζ2 in
equation two.
104
Annex 2.C – List of countries’ elections per tier in the data
Country Tier Elect.years N.Elec. Country Tier Elect.years N.Elec.
Albaniaab PR 2009-2013 2 Japan MMD 1947-1993 18
SMD 1996-2005 4 PR 1996-2012 6
Australia AV 1946-2010 26 SMD 1996-2012 6
Austria PR 1945-2008 20 Koreab SMD 1948-2008 19
Barbados MMD 1966 1 Latvia PR 1993-2011 7
SMD 1971-2008 9 Lithuaniaab SMD 1992-2008 5
Belgium PR 1946-2010 21 Luxembourg PR 1945-2013 15
Bolivia PR 1985-2009 7 Macedoniaa PR 2002-2011 4
SMD 1997-2009 4 SMD 1994 1
Brazil PR 1945-1962 5 Malta PR 1966-2013 11
1982-2010 8 Mexico PR 1991-2012 8
Bulgaria PR 1991-2009 6 SMD 1991-2012 8
SMD 2009 1 Netherlands PR 1946-2012 21
Canada SMD 1945-2011 22 New Zealandb SMD 1946-2011 23
Chile MMD 1989-2009 6 Norway PR 1945-2009 17
Colombia MMD 1958-1990 12 Peru PR 1963 1
PR 1991-2010 6 1980-2011 8
Costa Rica PR 1953-2010 15 Poland PR 1991-2011 7
Croatiaab PR 1995-2007 4 Portugal PR 1976-2011 13
SMD 1992 1 Romaniaab PR 1990-2004 5
Cyprus PR 1981-2011 7 Russiab SMD 1993-2003 4
Czech Rep. PR 1990-2013 8 Sloveniaa PR 1996-2011 5
Denmark PR 1945-2011 24 South Africab PR 1994-2009 4
Domin.Rep.a PR 1962-2010 12 Spain PR 1977-2011 11
Ecuador PR 1979-2009 12 Sri Lanka PR 1960-1977 6
Estonia PR 1995-2011 6 SMD 1989-2010 7
Finland PR 1945-2011 19 Sweden PR 1948-2010 20
Francea PR 1986 1 Switzerland PR 1947-2011 17
SMD 1973-1981 3 Taiwanb MMD 1992-2004 5
1988-2012 6 SMD 2008-2012 2
Germany PR 1949-2009 17 Trin.y Tobago SMD 1966-2010 12
SMD 1949-2009 17 Turkey MMD 1950-1957 3
Ghana SMD 1996-2008 4 PR 1961-2011 13
Greecea MMD 1952,1956 2 Ukraine SMD 1994-2012 4
PR 1946,1951 2 United King. SMD 1945-2010 17
1958-2007 14 United States SMD 1946-2012 34
Honduras PR 1981-2009 8 Uruguay PR 1954-2009 11
Hungary PR 1990-2010 6 Venezuelaa PR 1958-2000 11
SMD 1990-2010 6 SMD 2005-2010 2
Iceland PR 1959-2013 21 Zambia SMD 1968 1
Indiaa SMD 1977-2004 9 1996-2011 5
Ireland PR 1948-2010 18
Italy PR 1948-1992 11 Overall MMD 47
1994-2008 5 PR 506
SMD 1994-2001 3 SMD+AV 276
Jamaica SMD 1962-2011 11 Total 829
105
a Missing elections: Albania (1992/smd); Croatia (2011/pr); Dominican Republic(1990/pr); France (1945-1977/smd); Greece (1950/pr); India (1951-1977/smd); Lithua-nia (2012/smd); Macedonia (1998/smd); Romania (2008-2012/smd); Slovenia (1992/pr);Venezuela (1993/pr, 1993-2000/smd).b In some or all elections, these countries also had an additional PR tier with only onenation-wide constituency, which was dropped from the sample since they make it mean-ingless to talk about cross-district homogeneity: Albania (1992-2005); Croatia (1992 and1995); Korea (1963-2012); Lithuania (1992-2012); New Zealand (1996-2011); Romania(2008-2012); Russia (1993-2011); South Africa (1994-2009); Taiwan (1992-2012).
106
Annex 2.D – Distribution of variables used in the main models of this
chapter
ENEPnat
0 5 10 15
0.000.050.100.150.200.250.30
PtyNat
0.0
0.2
0.4
0.6
0.8
1.0
0
1
2
3
4
M
0 10 20 30 40
0.00
0.02
0.04
0.06
0.08
0.10
VtrsGeoHomog
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.5
1.0
1.5
2.0
2.5
UpperTierSize
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0
5
10
15
20
Federalism
0.0
0.5
1.0
0.0
0.5
1.0
1.5
SocDiv (Cederman et.al., 2009)
−0.
2
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.5
1.0
1.5
2.0
PersVoting
0.0
0.5
1.0
0.0
0.5
1.0
1.5
President
0.0
0.5
1.0
0.0
0.5
1.0
1.5
2.0
2.5
Variable
Den
sity
107
Annex 2.E - Measure sensitivity of the main results for the effects
between ENEPnat and PtyNat
One should not forget the warning made by Mustillo and Mustillo (2012), that
measures of party system nationalization are not fixed, but estimates. They have,
therefore (and as I have shown in Figure 2.1), an uncertainty associated with them.
Such uncertainty is hidden within the measure, meaning that the default assumption
of zero measurement is not only unlikely, but certainly violated. This leads results
to be biased to an unknown extent. Although this is a difficult issue to solve, it is
not difficult to be addressed. Once more, it is just better to explicitly model the
measurement error rather than assume it, as usual, to be zero. In order to do so, it is
only necessary to recall that any observed measurement is, in the end, an indicator of
a latent unobserved construct. In the context of this chapter, it means that PtyNat
measured by PSNSs is actually a single indicator of a latent unobserved variable
that we could call Nationalization:
PtyNat = λ1Nationalization+ δ (2.25)
When we assume zero measurement error, all we do is to fix (consciously or not)
λ1 to 1 and δ to zero (c.f. Bollen, 1989). Of course, it is not possible to know
how much error there is in our measure of PtyNat as we do not really observe
Nationalization. But if we usually assume zero measurement error, what in other
words means one hundred percent reliability on the used measure, why couldn’t we
simply make different, less optimistic assumptions? The reliability ρ of PtyNat in
the above equation is given by the following:
108
ρ =explainedvariance
totalvariance=λ21var(Nationalization)
var(PtyNat)(2.26)
Still fixing λ1 to 1, we can re-estimate previous models with different assumptions
about ρ, i.e. about var(Nationalization)/var(PtyNat). I have done so, by estimat-
ing one hundred replications of the chapter’s model 3, each time decreasing in one
percent point the assumed reliability of PtyNat as a measure of Nationalization.
Figure below shows two graphics with selected coefficients from these replications.
One has the coefficient and confidence interval of the direct effect from PtyNat to
ENEPnat, at each level of assumed reliability of PtyNat. The other is the coefficient
and confidence interval of the direct effect from ENEPnat to PtyNat at each level os
assumed reliability of PtyNat. The other coefficients are omitted for obvious space
limitation, but their results do not change in any important way when altering the
reliability of PtyNat.
The first graphic shows that the direct effect from PtyNat to ENEPnat in pre-
vious model 2 is fairly robust to measurement error in the PtyNat variable. The
results we get when assuming 100% reliability are the same we get until the level
of assumed reliability is of only 10%. As bad as our observed measure may be, it
is reasonable to suppose that it is not that bad. By the other hand, the second
graphic shows a different story about the direct effect from ENEPnat to PtyNat.
It is possible to notice that the result showed in Table 2.1 for this direct effect, i.e.
non-significance and coefficient around −0.02, is fairly dependent on PtyNat not
having measurement error. If our level of assumed reliability for this measure drops
from 100% to around 90%, it is enough to change the inference. The direct effect
from ENEPnat to PtyNat becomes significant and the less reliability we assume for
the measure of PtyNat, the greater becomes the magnitude of the coefficient of the
109
Annex 2.E.1 - Changes in the direct effects from ENEPnat to PtyNat and fromPtyNat to ENEPnat according to the reliability assumed for PtyNat as a measureof latent Nationalization
−10
−6
−2
Assumed reliability of PSNSs
Coe
ffici
ent o
f
PS
NS
s→
EN
EP
nat
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
−0.
05−
0.02
0.00
Assumed reliability of PSNSs
Coe
ffici
ent o
f
EN
EP
nat→
PS
NS
s
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
direct effect from ENEPnat to PtyNat.
The interpretation of this finding is that, actually, the result about the impact
of the number of parties on the party system nationalization that we have seen
in the previous table, is not totally reliable. It is not necessarily wrong, but it is
too much dependent on PtyNat having at least 90% of reliability as a measure of
the latent Nationalization. Therefore, I would claim that the correct conclusion
about the reciprocal causation is the following. We have clear evidence that party
system nationalization has a strong impact on the effective number of parties at
the national level. While the result for the other way around is less clear in what
regards its statistical significance (it depends on how well we are measuring party
110
system nationalization), but is in any case weak in what regards its coefficient. In
the end, we have good evidence that nationalization is an important variable to be
included in the models of number of parties, and few evidence about any important
reciprocal causation.
111
Chapter 3
Personal voting and localism: ge-
ographical distribution of electoral
support under open list PR systems
In this chapter, I examine the often-assumed relationship between voting for
candidates instead of for parties, and the emergence of political localism at the
electoral arena. More precisely, first I investigate to which extent candidates for the
national Legislatives under personal voting (PV) systems end up with a localized
electoral support instead of a territorially dispersed one. Next, I also verify how
much electorally profitable a localized support has been in terms of winning odds.
Literature has largely taken the connections between PV and localism as a given.
More than that, it is not uncommon for PV to be criticized for it. Just as institution-
alized parties are, often, normatively expected to be nationalized across a country
(e.g. Sartori, 1968; Jones and Mainwaring, 2003), candidates from PV systems are
usually expected to gather similar electoral support across the inner areas of their
electoral circumscriptions. Like most established parties in developed democracies
moved from a regionalized electoral support to a nationalized one (Caramani, 1996,
2003), candidates under PV are expected to cease being local, parochial, to become
candidates of their whole electoral circumscription. In a sense, to “nationalize”
within the circumscription.
112
Should they fail and therefore end up having geographically concentrated votes
in specific local areas, candidates are invariably associated to negative ideas such
as those of pork barrel, parochialism and localism (e.g. Mayhew, 1974; Fiorina,
1977; Ames, 1995, 2001; Allen, 2012). Regionalized and localized politics are in-
variably disparaged. When foundational scholars as Mayhew (1974) and Fiorina
(1977) claimed that, seduced by local politics, national representatives would tend
to overemphasize localized interests in the legislative arena, they were, as a matter of
fact, reviving regarding the American system the same concern that made Edmund
Burke advocate his normative paradigm of the free mandate when thinking of the
Westminster.
The problem with the American and Westminster cases, however, is that while
their single member district system (SMD) certainly is a PV system, since voters
directly chose candidates, they are also based on localized electoral circumscriptions.
It means under SMD localism is a given of the formal rules, making it difficult to
proper separate whether the Burkerian concerns should be put more on the shoul-
ders of the given localism or of the PV feature. Even so, such uncertainty has not
prevented scholars from frequently exporting the concern that PV would breed lo-
calism – and consequently parochial politics – to systems where localism is not given
de jure (see Katz, 1986, Shugart and Carey, 1995, Shugart et. al., 2005). It turns
out that the open list proportional representation (OLPR) type of electoral system
constitutes the ideal scenario for investigating it more fittingly. OLPR is also a PV
system since voters cast direct votes for candidates. But it happens in at-large elec-
toral circumscriptions with much greater magnitudes, where parties are allocated
seats in proportion to the votes received and PV determines which candidates take
the seats. It is an ideal scenario to check how much PV is in fact related to localism
113
since localism is not a given of the formal rules.
Here, I contribute towards testing it by mobilizing electoral results for the na-
tional (Lower) Chamber of five countries that use OLPR: Belgium (2007 election),
Brazil (2006), Ecuador (2009), Finland (2007) and Latvia (2011). These are electoral
data for each candidate, disaggregated at the local administrative level (municipali-
ties) that lies within countries’ electoral circumscriptions (cantons/states/provinces/regions).
The intention is two-folded. First, it is to verify how often candidates have or have
not been presenting geographically demarcated electoral support (to where they
could eventually deliver pork) analogous to what happens in single member district
(SMD) systems. Second, it is to verify whether this pattern is electorally profitable
in comparison to spreading votes territorially. In the next section, I further develop
the theoretical framework about the link between PV and parochialism. Then, in
the following, I measure the territorial distribution of votes of candidates in dif-
ferent ways. Mainly, trough an Empirical Bayes Index of spatial autocorrelation
and trough a weighted regional Gini index. In the last session, I finally present
a multilevel model to assess the electoral gain of concentrating electoral support
geographically.
3.1 Does voting for a candidate mean voting for
a local constituency service?
Amidst the controversy over whether worldwide electoral systems are moving
towards the personalization of the electoral choices at the expense of the partisan
vote (e.g. Cain, Frejohn and Fiorina, 1987; Manin, 1997; Colomer, 2009; Karvonen,
2010), literature has been paying increasing attention to such personal dimension
114
of the vote. Be it in systems that adopt the personal vote as a formal separate
choice from the partisan vote (i.e. preferential voting systems), be it as an informal
component of how voters would be more and more making choices in closed partisan
voting systems (Norris, 2002; Margetts, 2010; Karvonen, 2010). However, while we
are just beginning to focus our attention on and to understand the PV, it has already
been accused of plenty of unfortunate consequences.
The most common idea is that PV links candidates more directly to voters,
disregarding the mediation of parties. In different degrees and formulations, this
issue has been remarked by numerous authors (Ames, 2001; Mayhew, 1974; Cain,
Ferejohn and Fiorina, 1987; Carey and Shugart, 1995; Bowler and Farrell, 1993;
Samuels, 2001; Shugart et. al., 2005; among many others). Such a direct link
would have further undesirable impact in the legislative arena, triggering from the
fragmentation of party systems (Katz, 1986; Lijphart, 1994; Taagepera, 1994) to the
impairment of party strength and cohesion (Blais, 1991; Katz, 1986; Petersson et.
al., 1999). Other broader consequences to larger political outcomes could include
also affecting the focus and quality of implemented policies (Hicken and Simmonds,
2008), the increase of the particularism of transfers (Rickard, 2009) and increased
corruption (Chang and Goldgen, 2006; Kunicova and Rose-Ackerman, 2005; Persson
et., al 2003). Not to mention a series of additional penalties PV would bring to
different economic outcomes (Diaz-Cayeros et. al., 2009; Milesi-Ferreti, Millesi-
Feretti et. al., 2001).
Permeating most arguments about the unfortunate effects of PV, there is an
often present, but not always explicit assumption. It is that PV’s supposed direct
link between candidates and voters would cause the localization of politics, i.e. the
breeding of localism and parochialism. This intuition is, of course, a descendant
115
of the classic debate on how the PV present in the American SMD system would
have stimulated pork-barrel practices and constituency-services. The concept was
popularized by Cain, Ferejohn and Fiorina (1987), whose definition is cited from
Fenno (1978): “many activities can be incorporated under the rubric of ‘district
service’ or ‘constituency service’, but the core activity is providing help to individ-
uals, groups and localities in coping to the federal government (. . . ) Private groups
and local governments need assistance in pursuing federal funds”. This intuition
later matched the classic theorization of Mayhew (1974) about the district system
adopted in the United States, besides other works as those from Lancaster (1986)
and Cain et al (1987). After all, pork barrel would be a result of members of the
Lower chamber (MC) trying to build dominance over their original districts, since
there is only one representative per district. Delivering enough resources and ser-
vices to their localities could, in practice, close the future competition and assure,
within districts, the reelection of their always reelection-seeker representatives.
However, in the context of multi-member PV systems, candidates and MCs usu-
ally come from much wider electoral circumscriptions and have many more competi-
tors in the legislative arena who also come from the same circumscription. Nonethe-
less, literature has frequently transposed the idea of the link between PV and pork
barrel politics from the American context to multi-member systems1. The usual rea-
soning is that “to build and maintain a personal base that can set them apart from
co-partisan, candidates focus their activities on particularistic distribution” (Allen,
1A frequently forgotten caveat of such transposition would be the fact that most of the strongstatements of the distributive theory regarding the American case inexplicitly depend on the exis-tence of district magnitude = 1. Only in this scenario an elected candidate would be certain thatshe or he does not have other legislators competing for future votes in the same electoral area.Thus, the possibility of pork barrel depends on the magnitudes of districts (Lancaster, 1986). Thisis a basic logical need for making plausible the giant logroll idea criticized by Krehbiel (1992). Fol-lowing this logic, as PR systems usually have much greater magnitudes it gets much more difficultfor a legislator under this system to adopt parochial behavior at the legislative arena.
116
2010:4). It means that, in order to compete against a high number of opponents,
including co-partisans, candidates would behave in particularistic fashion, looking
for particularistic goods to deliver to a particular clientele.
More than that, for many authors, it seemed just logical and natural, likewise,
to assume that the particularistic goods per excellence would be the local goods.
The clientele per excellence would be the geographical ones and, thus, candidates
would build their personalism by delivering service and goods to political homeland
(general arguments can be found, for instance, in Katz, 1986, Shugart and Carey,
1995, Shugart et. al., 2005). Just like it happened in the single-member PV system
of the United States. This overall logic has been applied to many countries using
different PV systems, such as Italy (Golden and Picci, 2008), Colombia (Crisp and
Ingall, 2002), Estonia (Tavits, 2010), Indonesia (Allen, 2010) and, largely, about
the Brazilian case (Ames, 1995, 2001; Pereira and Muller, 2002, 2003; Samuels,
2001; Mainwaring, 1991, 1999)2. Few of these works have put this idea as clear
as Shugart et. al. (2005). Take for instance one of their statements which is a
good representative of this overall interpretation: “Where voters vote on the basis
of the personal distinctiveness of politicians, candidates for elective office often seek
to advertise the ways in which they will serve local interest” (p.437).
Yet, the not often asked question is: why? Why should we logically expect
such a link between PV and localism? Could not a candidate compete against co-
partisans and against other adversaries using personal but not local attributes or
actions? For instance, his/her appealing personal attributes, his/her linkage with
syndicates, associations, religions, in a way to seek clienteles that are not necessarily
geographically demarked? It is far from clear why we usually assume this mix of
2Although the same claim is usually not made regarding other countries with Open-List PRsystems such as Finland, Norway or Sweden.
117
personalism and localism, almost as if they were equivalent. PV is institutionally
present in varied electoral systems and can be operated quite differently (see Cox,
1997, Karvonen, 2004). It is unknown how this personal connection between electors
and candidates would happen in contexts different from the SMD that characterizes
the American system. How the personal connection links to geography in order to
open gates to parochialisms and pork barreling in other types of electoral system?
Should we rightfully expect localism and parochialization even in a framework that
is not based on local districts with extremely low magnitude?
Specifically, the usual logical path goes as follows. The presence of PV is assumed
to foster personalism, and personalism is assumed to lead to localism, therefore PV
is assumed to foster localism:
PV −→ Personalisme;Personalisme −→ Localisme ∴ PV −→ Localisme (3.1)
Where:PV : is the existence of formal/institutional personal voting systems.
Personalisme: is the incentive from the electoral system for politicians to engage
in personalistic behavior in the electoral arena.
Localisme: is the further incentive for locality-focused, i.e. parochial, behavior in
the electoral arena.
Evidently, this is certainly a chain of too many and too strong assumptions. To
begin with, it is even debatable whether PV would naturally breed particularisms
and personalisms of any type. In other words, it is not undisputed that PV −→
Personalisme. But it is not my intent to develop this issue and thus I will have
to accept this first part of the statement as a given in order to proceed. So, let’s
118
assume Personalisme ' PV for now on. I do so because my interest relies in better
understanding and assessing the second part of this chain, namely the assumption
that Personalisme −→ Localisme, which leads here to PV −→ Localisme
Theoretically, such an assumption that personalized voting in general should be
expected to breed local politics just as the particular PV in the American system
would do, is in fact a case of spurious association. As mentioned, the literature on
the American system deals with an electoral mechanism that is at the same time
small-district-based and candidate-centered. It is a PV because voters do not have
to choose only a party for the House of Representatives but, at the same time, a
specific candidate. And it is small-district-based since voters make such choice in a
previously given specific local circumscription called district. The problem is that
due to this framework, it is difficult to know whether it is the PV or the district-based
nature of the system what fosters localism.
It means that we need to make explicit the following distinction, as obvious as
it may sound, since it is often forgotten: districtalization cannot be always taken
as equal to personal voting. Put more accurately, they may sometimes be undistin-
guishable (“equal”), but cannot be confused as the same set of phenomena. Using
set theory, the assumption is that:
E(Dist ∩ PV ) 6= Dist ∪ PV (3.2)
Where:Dist is the occurrence districtalization of the vote.
While it seems quite straight forward why one expects local-district-based sys-
tems to foster localization of candidates’ interests, pork barrel and parochialism, it
119
is not that easy to logically assume that personalized voting itself should always
necessarily cause the same situation. Thus, for PV systems to start fostering local-
ism, it may also be required for them to be framed by geographical demarcation of
the votes, once it is only the localized vote that can lead to localized politics. This
argument, that reformulates the first half of proposition (1), can be put as follows:
Dist −→ (PV −→ Localisme) ∴ Dist¬ −→ Localism¬e (3.3a)
Or, to keep with the set theory:
Localisme ⊂ (Dist ∩ PV ) (3.3b)
Localisme
DistPV
While it is even questionable whether the presence of electoral local-districts
should be taken as sufficient condition for a parochial political behavior of candidates
and of congressmen, it is in fact a necessary one. Therefore, PV systems can just
lead to such behaviors if districtalization is also present. Certainly, nevertheless,
by existence of districtalization I do not mean only the existence of legally pre-
defined districts as the American. Using Taylor and Johnston (1979) terms, we
could say that a given PV system can present either de jure districts or de facto
districts. The former means having districts pre-defined by law, while the latter is
120
the case when the election is not organized in a small local district-based voting
system, but electoral results present patterns of effective geographical concentration
of candidates’ electoral support in local areas.
However, an important note of caution is necessary. Although both the pres-
ence of districts and PV can be seen as necessary conditions for the emergence of
localism in the electoral politics, it is not difficult to see why not even their joint
presence can be stated as a jointly sufficient condition for the existence of parochial
legislative behavior. Such a view of sufficiency would disremember the fact that we
can find, and general do find, considerable differences between what happens in the
electoral arena and what happens in the legislative arena (Bowler, 2000; Cox, 1987;
Cox e McCubbins, 1993). Both arenas can be strongly separated by institutional
frameworks within the legislative arena that shape MCs behavior despite of which
incentives have emerged from the electoral arena. It means that one thing is to
say that legislators elected in local districts under PV will probably prefer parochial
politics and pork barrel to deliver to those districts in order to maximize their fu-
ture electoral outcomes. Another different matter is to say they are really capable
of endorsing and of carrying this desire, given the rules of the legislative game they
will eventually face. That’s why even the presence of local electoral districtalization
is not sufficient, although necessary together with the PV, to strengthen the total
incentives for parochial behavior in the legislative arena.
With this, we complete the reformulation of proposition (3.1), by rewriting its
second half, given that Parocl means the final parochial behavior at the legislative
arena:
121
Parocl ∝ f(Dist→ (PV → Localisme), P ermissiveLeg)|
(Dist¬ → Localism¬e ;Localism¬e → Paroc¬l ;PermissiveLeg¬ → Paroc¬l ) (3.4a)
Or, to keep with the set theory:
Parocl ⊂ (Localisme ∩ PermissiveLeg)|Localisme ⊂ (Dist ∩ PV ) (3.4b)
Localisme
Parocl
DistPV
PermissiveLeg
Given this schema, one can adopt three different logical approaches to verify
the link between PV and localism. The first is to verify empirically if the final
outcome really does happen at the legislative arena (∃Parocl). Second and third
approaches mean to verify if the two conditions for it are present (∃Localisme and
∃PermissiveLeg). The approach represented in proposition 3.1 usually has the
afore mentioned issue of spurious association because it resembles the problem of
omitted variable bias, since when scholars assume direct relationship between PV
and Parocl they are disregarding Dist and PermissiveLeg. However, empirical
122
critiques to the deterministic relationship established by 3.1 are often insufficient as
well. Take for instance the counterarguments present in the extensive literature con-
cerning the Brazilian case. They rely on either sustaining that the final result just
does not happen, i.e. @Parocl (Ricci, 2003; Amorim Neto e Santos, 2003; Figueiredo
e Limongi, 1995, 1998; Mesquita, 2009), or that the condition PermissiveLeg is not
true (Figueiredo e Limongi, 2002, 2005; Santos, 2003). Evidently, eventual demon-
stration of both arguments (@Parocl or @PermissiveLeg) is enough to what they
propose: to demonstrate the insufficiency of the traditional deduction represented
by proposition 3.1. But they do not help identifying the size of E(Dist ∩ PV ), i.e.
the extent to each personal voting and districtalization are related.
Identifying this can be actually quite important. By one hand, to infer that the
electoral system does not create parochial incentives (@Localisme ) only because in
the end legislators are proved to not behave parochially (@Parocl ), would mean
incurring in the type of formal fallacy known as denying the antecedent3. By the
other hand, to affirm that parochial incentives from electoral systems do not exist
(@Localisme) just because in the legislative arena there is a restrictive framework for
these incentives to flourish (@PermissiveLeg), would be not only illogical due to the
sequence of the events. More than that, by relying only on this we would not know
what should happen in case any changes were made to the intra legislative framework
in order to make it more permissive (i.e. creating a case of ∃PermissiveLeg).
The OLPR systems offer us the opportunity to access this possible feature of PV
in a very unique way. By one hand, it would be obviously impossible to test it in
SMD or multimember districts (MMD) systems because in there PV and Dist always
perfectly coexist, then E(Dist∩PV ) = (Dist∪PV ). By the other side, the question
3It means: in the alleged situation where the occurrence of X is said to make Y happen, to saythat Y not happening is a consequence of the fact that X did not happen, is not necessarily true.
123
would also be meaningless or hard to evaluate in proportional closed-list systems, as
although we know PV can still informally exist in these electoral systems (Norris,
2002), it is formally inexistent, so PV = ∅ ∴ (Dist∩PV ) = ∅. By the other hand,
in the PR-Open list the PV is given: under this system, the votes go first to the
parties of chosen candidates to define each party’s seats, then go to the candidates
themselves to define who in each party will take partisan allocated chairs (pooling
vote). At the same time, Dist is not given, i.e. there is no de jure local districts,
but it may be possible to exist some de facto ones. It means that with OLPR we
can fix PV and see the variation of Dist, making this the ideal combination for us
to test E(Dist ∩ PV ). It is exactly what I will do shortly, by mobilizing data from
Belgium, Brazil, Ecuador, Finland and Latvia.
3.2 Data
It is important to begin by describing a few details about the data used in the
following analysis. My sample covers the electoral results of all candidates for the
national (Lower) Chamber in five different country/elections that used PR Open
List systems. Belgium elected 150 legislators in 2007; Brazil, 513 in 2006; Ecuador,
103 in 20094; Finland, 200 in 2007; Latvia, 100 in 2011. Among these countries,
only Finland does not allow voters to cast party-list votes (i.e. to skip choosing a
candidate trough PV). While in the Belgium election of 2007 about 27.9% of the
voters cast a list vote, they were 9.8% in Brazil-2006, 33.6% in Ecuador-2009 and
41.9% in Latvia-2011. Other important differences between countries’ institutions
are: Latvia and Ecuador allow multiple voting and Latvia allows negative voting
4Ecuador also elects 15 legislators using the whole country as electoral circumscription, whichI dropped.
124
(which I dropped, considering only the ordinal votes). There are a few differences
between how final votes are summed in the five countries, in order to calculate
number of chairs for each party5. In total, Brazil and Finland have the purest PR-
Open List system, being also almost identical in both countries. The other three
countries represent here some variations of the pure formula.
The electoral figures for each of the 10340 candidates in the original data are
disaggregated at the local administrative level (municipalities) that lies within the
countries’ electoral circumscriptions (cantons/states/provinces/regions). These cir-
cumscriptions are respectively 10 Belgian cantons, 26 Brazilian federated states, 23
Ecuadorian provinces, 14 Finnish electoral macro-regions and 4 Latvian planning-
regions6. What means that, in the five countries, the electoral circumscriptions are
indeed not formally local. So, as mentioned, from these systems one can only ex-
pect the personal voting to lead to parochial incentives and pork barrel politics if
elections end up drawing de facto districts across the localities that lay inside those
bigger and wider electoral circumscriptions. In my data, such localities into which
cantons/states/provinces/macro-regions are divided are officially called municipali-
ties in the cases of Belgium, Brazil and Latvia, the administrative parishes in the
case of Ecuador7 and cities in Finland. It is across these lowest administrative en-
5The greatest is in Ecuador, where votes nominally cast for candidates were counted after beingweighted by how many personal votes were cast.
6A few notes are needed here. I dropped a 11th Belgian canton, Wallon-Brabant (-5 legislators),and a 24th Ecuadorian province, Galapagos Islands (-2), as both have too few municipalities forthe computations being done. I also dropped a Brazilian 27th electoral district (city of Brasılia,-8), a 15th Finnish electoral district (city of Helsinki, -21), and a Latvian 5th electoral district(city of Riga, -30), as they are single cities-districts and the spatial statistics computed here needstudy areas divided in municipalities.
7More specifically, the official Ecuadorian administrative parishes are the cities and ruralparishes. Urban parishes into which cities are divided are not legally administrative units, noreven have the same representative status and political representation. I follow strictly the Ecuado-rian oficial INEC: Instituto Nacional de Estadıstica y Censos (National Institute for Statistics andCensus) in considering rural parishes and cities as the official local administrative units.
125
tities of each electoral circumscription of the countries that I will test how often
candidates do concentrate votes geographically.
3.3 Spatial assessment: one needs parishes to be
parochial
Unfortunately, the size of the literature about localism, parochialism or pork
barrel politics seems to be inversely proportional to the quantity of works on how
to empirically define and assess these concepts. Consequently, it is not only a chal-
lenge to empirically verify the “degree to which individual politicians can further
their careers by appealing to narrow geographic constituencies on the one hand,
or party constituencies on the other” (Seddon et. al., 2001:1). It is even hard to
conciliate what is a “narrow geographic constituency” in the many diverse systems.
Here, I will use two different measures to assess different aspects of the geographi-
cal electoral support of candidates under PR Open-List systems. One is based on
spatial autocorrelation (to detect geographical clustering), the other is based on the
regional GINI index (to detect territorial homogeneity of electoral support).
3.3.1 Geographical clustering: Moran‘s I through an Empirical Bayes
Index
To look for electoral support concentrated in specific localities of a territory
means to check for geographical clustering. It is, to check whether there are ter-
ritorial strongholds where electoral support is similar across neighbor areas. The
usual way of testing that is trough spatial autocorrelation. Put in a simple form,
spatial autocorrelation means that neighbor observations of the same variable are
126
correlated, thus configuring an autocorrelation in similar sense to its well-known
counterpart from panel analysis. But here, in spite of a variable being correlated
with itself over time periods, it is correlated with itself across areal units in a space
dimension. The usefulness of this in order to look for de facto electoral districtal-
ization and pork barrel loci is that strong positive autocorrelation in space usually
reveals spatial concentration of similar values across close neighbors (see Cliff and
Ord, 1981; Goodchild, 1991).
Spatial autocorrelation has been frequently measured trough indices like Moran’s
I, Geary’s C Ratio and Global G statistic, which all describe the overall spatial rela-
tionship of a given variable across all areal units (see, for instance, Fotheringham et.
al., 2003). Even in the literature about electoral support, it is no original proposal.
Ames (2001), for instance, has already used spatial autocorrelation to claim that
Brazilian candidates would have geographically concentrated electoral support. I
will follow a similar path. Like Ames, I will also opt to use Moran’s I8, but instead
of working with the traditional formula for this measure I employ here the Bayesian
version proposed by Assuncao and Reis (1999) to “adjust Moran’s I for the variation
on population size” (p.2160) among geographical areas. Their version of the index,
called Empirical Bayes Index for spatial autocorrelation (EBI), is an improvement
over previous measures at least for the field of electoral studies, due to the fact that,
as we know, voters are usually unevenly distributed across countries’ territories. I
slightly adapted Assuncao and Reis EBI so it can be bounded within the range
8I prefer Moran’s I over its similar Geary’s C, due to both its more spread use in the PoliticalScience and to its commonly pointed desirable distributional characteristics (Cliff and Ord, 1981).And over Getis-Ordis’ Global G, because Moran’s I does not differentiate hotspots and cold spotsin the territory, i.e. concentration of high values from concentration of low values. At first glancethis differentiation could be seen as welcome for my purposes in this chapter, since it would avoidfalse positive detection of electoral parishes. However, at the same time we would incur in greater(and in my case less conservative) risk of having false negatives, because simultaneous hotspotsand cold spots cancel each other in the Global G calculation.
127
[−1, 1] like the traditional Moran’s I9, by subtracting the mean smoothed rate from
the numerator of their index. I will call it the Empirical Bayesian Moran’s I, for
differentiation.
Therefore, for each candidate in the sample, I calculated EBMoran’s I using as
input the candidate’s votes per city. Instead of using the proportion of votes, the
index uses the absolute number of votes of a given candidate and of total voters to
calculate the deviation of the estimated marginal z. Then, the formula is as follows:
EBMoran =n∑I
i
∑Jj wij(zi − z)(zj − z)
WI×J∑I
i (zi − z)2
(3.5)
zi = ci/ei − b√υi
Where:wij is the cell value in the spatial weight matrix WI×J , identifying whether mu-
nicipality i is a neighbor of municipality j; ci is the absolute number of votes a
candidate c had in municipality i; ei is the size of the electorate that voted in i;
the marginal expectation of ci/ei and υi is its variance.
For each municipality i in a given WI×J neighbor structure matrix, if municipality
j is its neighbor, the formula calculates how much the votes of a given candidate
deviates from the expected by the population distribution. Then, these deviations
zi and zj are multiplied and all products of deviations of all pairs ij are summed.
Here is where the final statistic comes from: for each pair of neighbor spatial units
i and j, if deviations zi and zj are both above or below the expected value given by
9Moran’s I and its derivations are bounded only when calculated with a raw standardized weighneighborhood matrix, which is a standard procedure that I also adopt here.
128
the population distribution, than their product will be a positive number. But if zi
is above the expected and zj is below, or zi is below with zj being above, the product
of their mean will be a negative number. Similarly to Moran’s I, the EBMoran’s
I will range between -1 to 110. Following Cliff and Ord (1981), here this range
would also mean: -1 = negative autocorrelation (perfect concentration of dissimilar
values11); 0 = random spatial distribution; 1 = positive spatial autocorrelation
(perfect concentration of similar values). Notice, however, that big negative values
are, hence, unlikely for our purposes since it is quite uncommon to expect a candidate
to have concentration of dissimilar electoral support. I will come back to this point
in a moment.
The spatial matrix WI×J is where we input information about, within a given
electoral circumscription, which municipalities are neighbors of each other. I opt to
use the matrix type known as Queen contiguity, which basically identifies neighbor-
hood by contiguity of borders. But I present results both for Queen contiguity of
orders 1 and 2. The first order type (Queen 1) means that, for each municipality,
its neighbors are only the other municipalities who share direct boundaries. While
the second order type (Queen 2) means that, for each municipality, the neighbors
are the ones who share direct boundaries and also the first neighbors of the direct
neighbors. Of course, it seems evident that Queen 2 is more realistic than Queen
1 for our purposes. Suppose a candidate has strong electoral support in city A, a
weak support in the neighbor city B, but again a strong support in city C - which
is neighbor of B, but not of A. Even if A and C are not contiguous cities, they can
10That is true particularly in the case of row-standardized neighbor matrices, i.e. when theweight matrix has all values dived by the sum of their rows, so the sum of any row equals 1. Thisis the case of the matrices used in this chapter.
11Again, notice that negative values in Moran’s I do not mean concentration of low values neareach other, as it happens with Global G. It means concentration of dissimilar values near eachother.
129
Figure 3.1: Distribution of the geographical concentration of electoral support (mea-sured by EBMoran’s I) of all candidates
Countries
spG
INI
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
Belgium Brazil Ecuador Finland Latvia
(a) neighbors = contiguous municipalities (WIxJ = Queen1)
Not elected Elected
Countries
spG
INI
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
Belgium Brazil Ecuador Finland Latvia
(b) neighbors = contiguous municip. + 1st neighbors of neighbors (WIxJ = Queen2)
Countries
Geo
grap
hica
l clu
ster
ing
(EB
Mor
an's
I)
Note: solid lines represent quantiles .25, .5 and .75.
130
be part of a same electoral stronghold. I show both results as a robustness check of
sensitivity to the matrix specification12, and also for further discussion.
To implement such calculation of the spatial auto-correlation, I took the poly-
gon maps of countries divided into their lowest administrative level divisions and
adjusted these maps for their situation in the years of the elections, using official
information on the creation of new cities and merging of old cities. Then, for each
of the electoral circumscriptions (cantons/states/provinces/regions) I calculated the
spatial weights. Lastly, to calculate candidates’ EBMoran’s I statistic, I run one
thousand Monte Carlo simulations for each of the 2514 Belgian, 4840 Brazilian,
1272 Ecuadorian, 1754 Finnish and 761 Latvian candidates, so the statistics were
inferred based on random permutations13. Figure 3.1 brings the distribution of the
EBMoran’s I of each candidate per country, and per electoral outcome.
It is essential to keep in mind that each electoral circumscription is a different
study area for the calculation of EBMoran’s I, with a singular neighbor structure
of its municipalities. And as neighborhood structures are part of the formulae of
EBMoran’s I (the WI×J matrix), thus affecting the index, it is not advisable to di-
rectly compare magnitudes of results from one area against another (for a recent
example on this discussion, see Van Meter et. al., 2011). This figure aggregates the
study areas (electoral circumscriptions) of each country, therefore the most impor-
tant here is the trend. First of all, it should be noticed that the EBMoran’s I of
candidates reach greater levels of geographical concentration when computation is
done with WI×J of type Queen 1, which means the less realistic one. Still, in 3.1(a)
12I have also tried Queen matrices with order greater than 2, as well as other types of matrices.The results do not differ from the ones yielded with the matrices presented here, as can be seen inthe Annex 3.A.
13Both calculations of the spatial indices and also of the neighborhood matrices were done byusing the spdep and rgeos packages of the R software, version x64 3.0.1. Fine adjustments at theshape polygon maps of the countries were made with ArcGIS Desktop 10.
131
in general 75% of the candidates scores EBMoran’s I of less than 0.4, and the dis-
tributions clearly show that most cases are concentrated in even lower values. The
exception is, probably, the distribution of elected candidates in Finland, which have
their bimodal distribution with one peak going towards 0.6. Overall, the median
elected candidate had a EBMoran’s I of around 0.3, and the median non-elected
candidate scored around 0.15. If one considers WI×J of type Queen 2, which is ar-
guably a more realistic formulation of municipal neighbor structure for our cases,
than the values of EBMoran’s I decrease even more. In 3.1(b), distributions go even
closer to zero, with median elected candidates scoring 0.2 and median non-elected
scoring around zero.
With any of the neighbor matrices, and in spite of the differences of neighbor
structures between electoral circumscriptions, the general results is that few to very
few candidates really had electoral support highly concentrated geographically in
the elections included in my data. More importantly, no matter what one might
consider as a low or high results, it is not negligible that most candidates in all
countries have values of EBMoran’s I that are close enough to zero for the reader to
question him or herself whether these electoral supports are something more than
randomly distributed across the territories. This is certainly an indication that con-
centrating votes is, at least, not exactly an unanimous pattern. Another support
for such conclusion is in the reasonably big variation that exists between candi-
dates. It means, the EBMoran’s I of candidates in all countries, maybe with the
exception of Belgium, range from very low, even negative values, to somewhat high
positive values. If nothing else, this is an additional evidence that concentrating
votes geographically is not a necessary behavior for candidates under OLPR sys-
tems. At the same, it is easy to see that there was a general pattern of more elected
132
candidates having geographically concentrated electoral support than defeated can-
didates. In all countries, the median EBMoran’s I of the elected was at around the
third quartile of the non-elected, which might be an indication that while very few
candidates have high concentration of electoral support, concentrating it might still
be an advantageous scenario.
In any case, of course this is not a formal test yet. The formal assessment will
come in more detail in the last section of this chapter. But before getting there,
the reader should be made aware of two flaws of relying only on measures of geo-
graphical clustering like the EBMoran’s I to detect the importance of geographical
concentration of votes. One of the flaws is that victorious candidates may be just
concentrating votes in bigger cities, just following the overall geographical concen-
tration of voters. The other flaw is that while spatial autocorrelation measures are
ideal to detect geographical concentration of electoral support, they are not apt of
measuring the opposite: the well spread electoral support. This is both a substan-
tively important and, as we will see, a methodologically important matter.
3.3.2 The impact of cities with big population: Local G index
Let’s deal first with the issue of the impact of the geographical concentration
of population/voters on the concentration of electoral support of candidates. I
will use here the Anselin (1995)’s LISA: local indicators of spatial associations.
LISA are local versions of spatial autocorrelation indices (e.g. local Moran’s I,
local Geary’s C), capable of identifying the contribution of each municipality to
the spatial autocorrelation of the electoral support of each candidate in his/her
electoral circumscription (state/province/canton). In the same way, Getis and Ord
(1992) have proposed a local version of their global-G statistic, which is capable
133
of differentiating spatial clusters of high values (hotspots) and of low values (cold
spots). It means, in our case, to differentiate, for each candidate, the cities in which
he or she concentrates more (hotspots, when local-G > 0) or less of their electoral
support (cold spots, when local-G < 0).
Figure 3.2: Concentration of population in each city and the average concentrationof candidates’ electoral support in that city (using local-G)
This is the option I follow here14. Accordingly, for all municipalities in a given
14Recall that, before, I have discarded Global G as a good measure, preferring the ones basedon Moran’s I such as the EBMoran’s I, because in the Global G the cold and hotspots cancel
134
electoral circumscription (study area) I calculated how much each municipality con-
tributes (its local-G) for the overall spatial autocorrelation of each candidate. It
means, for instance, that we can assess how much, on average, the city of Turku in
Finland has contributed for the geographical concentration of votes of candidates
that run in the electoral area of Varsinais-Suomen, where Turku is located. We
end up having one score for each city regarding each candidate. Then, I took the
average of local G contributions given by each city for all candidates, so our cases
here become one value per municipality. We can call it average concentration of
candidates’ votes in each city. For comparison purposes, I also calculated the local-
G of the whole electorate itself, to get a score of the contribution of each city for
the geographical concentration of the electorate. In the former example, it would
mean to assess how much the Finish city of Turku contributes to the demographic
concentration in Varsinais-Suomen. In all cases, here I used again binary Queen
contiguity matrix of second order as the spatial weights matrix.
Now, we can see whether it is true and to which extent that, on average, the cities
with more concentration of population receive more concentration of candidates’
electoral support. The scatterplots in Figure 3.2 positively demonstrate that, on
average, the geographical concentration of electoral support of candidates follows
the very concentration of the demographic distribution of the electorate across the
territory. Notice how the lowess line of fit and, in some cases, even a linear line
of fit, closely approximate the diagonal line of reference that represents perfect
linearity. This result means that, besides the number of candidates who concentrate
votes geographically being a far minority, the concentration of votes in the territory
each other. However, this time we can chose to use local G instead of local EBI or local Moran’sI because we are moving the unit of analysis. From candidates to city, i.e. disaggregating, sohotspots and cold spots do not cancel each other anymore as in Global G.
135
mostly follows the electorate itself15. In that sense, it starts getting a bit harder to
sustain that the majority of those candidates with more geographically concentrated
electoral support would be pursuing the formation of electoral parishes for electoral
purposes. Candidates mostly seem to be following the votes, i.e. concentrating their
gathering of votes where there are concentrations of voters. Or should one expect
that candidates would try to get less votes in the biggest cities on purpose? Now
let’s move to the second issue I have mentioned above: how spatial autocorrelation
is incapable of measuring the opposite of geographical concentration.
3.3.3 Territorial homogeneity of electoral support: Spatial GINI in-
dex
While the evidences so far seem to be sufficient to question the usual assumptions
about the automatic link between PV and localism, more work is still needed. The
problem is that, if we proceed to econometric analysis using EBMoran’s I alone as
a variable to measure spatial patterns of electoral support of candidates, we would
incur in one of two errors. If not in both - measurement error and misspecification.
Autocorrelation measures are ideal to detect geographical concentration of votes
(taking proximity into account), but are less than ideal to measure the opposite
scenario, the spreading of votes. The reason for that lies in the trick about what
means the lower range of spatial autocorrelation statistics such as Moran’s I and
EBMoran’s I (i.e. below zero and close to zero).
15The most numerous exception are a few cases in the peak above the diagonal reference line inBrazil. Interestingly, however, all these cities are in two of the Brazilian electoral circumscriptions,which are Parana and Santa Catarina. These are two of the states with most even distribution ofthe electorate across the territory. Santa Catarina is, for instance, the only state where the capitolis not even the greatest city. Therefore, it is maybe natural that in these states where many citiesare of medium size and none is gigantic compared to the rest, candidates are able to profit fromgeographical concentration in different parts of the territory
136
Remember that, as afore mentioned, neither the negative nor the null autocor-
relation do mean homogenization of votes across units. As it is pointed by Lee and
Wong (2000) when talking about Moran’s I, “if the value of one areal unit is above
the mean and the value of the neighboring unit is below the mean, the product of
the two mean deviations will be negative, indicating the presence of negative spatial
autocorrelation”. So, negative autocorrelation shows the dissimilarity between what
happens in two neighbor areas, not the homogeneity. It is possible to visualize these
outcomes by comparing the schema in Figure 3. Naturally, it is probably impossible
to find electoral geographical patterns that resemble the chess board-like schema (a)
in this figure. Accordingly, when studying electoral distribution of votes in a given
country’s territory, it is not feasible to expect values of measures based on Moran’s
I to be either systematically below zero or even strongly below zero16. The more
realistic geographic pattern of electoral support that is opposite to the parish-like
concentration of votes exemplified in schema (c) would be, then, not (a) but in the
worst case (b) and in the ideal case, the (d).
In both cases, Moran’s I or EBMoran’s I approximate zero. Therefore, since
spatial autocorrelation statistics do not allow for a further differentiation between
the more homogeneous voting of schema (d) in Figure 3 and random voting in schema
(b), if we did not look for an additional measure and proceed to econometric models
using only EBMoran’s I as an independent variable, we would be leaving much
of the impact of spreading-votes-pattern out of our models. And so, biasing the
equation in favor of the parochial-pattern result. So, it seems necessary to try
16Actually, even in logical grid spaces as those of Figure 3, it seems that in higher orders ofQueen or Rook neighbor matrices, dissimilar results near each other are increasingly difficult tobe differentiated from random spatial pattern. Hence, schema (a) scores a negative Moran’s I farfrom zero only when using a Rook contiguity matrix of order 1. Queen contiguity matrices of anyorder or Rook contiguity with orders greater than 1 give, in practice, scores close to zero.
137
Figure 3.3: Logical outcomes of territorial distribution and how EB-Moran is ex-pected to score*
(a)
I ∼ −1
(b)
I ∼ 0
(c)
I ∼ 1
(d)
I ∼ 0, . . . ,@I
Inspired by the approach of Lee and Culhane (2009).*Using Rook contiguity W matrix in order to assess the extreme logical possibilities.
another tactic to access the spreading-pattern or, in other words, the homogeneity
versus heterogeneity of candidates’ electoral support across the municipalities that
lay within their electoral circumscriptions. Probably the most accessible and yet
reliable option is to apply the concept of the well-known regional Gini index. Here,
it would measure the spatial inequality of electoral support of each candidate of
our sample, across the municipalities of his/her electoral circumscription. In the
same way that it is routinely done for examination of regional inequalities, but here
using as input the percentages of votes of candidates in each municipality of a given
electoral circumscription.
In the political science this experience is not new, but it is usually applied to
the study of nationalization, as I myself have done in the previous chapters. Take
for instance the index created by Jones and Mainwaring (2004) to study the na-
tionalization of electoral votes, i.e. territorial homogeneity of votes. It is precisely a
Gini-based distribution of the votes of each party across the electoral districts, called
PNS (Party nationalization score). Of course, here I would apply the index for each
138
candidate, not for parties. And across local administrative units of Belgium, Brazil,
Ecuador, Finland and Latvia, that are inside electoral circumscriptions. However,
Bochsler (2010) has shown that this index, as several others, is biased by the number
of spatial units. Consequently, Bochsler proposed a standardized version of this PNS
from Jones and Mainwaring (the PNSsw), which also accounts for the differences of
number of voters in each spatial unit. This is especially important because of what
we have just found above - candidates frequently follow the concentration of popu-
lation in the territory. Hence, the index, that I will call here spatial Gini (spGINI)
to avoid confusion due to the fact we are not applying it here to a national scale,
will score 1 for a candidate with perfectly homogeneous territorial distribution of
his electoral support, i.e. with equal percentage of votes in every municipality (zero
spatial inequality=homogeneity). While a candidate with a totally concentrated vot-
ing across municipalities will score 0 (total spatial inequality=perfect heterogeneity.
The correlation between spGINI and EBMoran’s I is not big: -0.15 (statistically
significant). Figure 3.4 has the descriptive summary of this measure for my sample.
Differently from what we saw regarding the EBMoran’s I, the spGINI varies
a bit more across countries. Belgium and Ecuador generally have the higher fig-
ures, meaning that it is more common in these countries for candidates to achieve
very good territorial spread of their electoral support across the municipalities that
lay within their electoral circumscriptions. On the other side, Brazil and Finland
present more diverse patterns. Quite a few candidates seem to have moderate to
low territorial homogeneity of the electoral support, which again does not mean
geographical concentration but only a not even distribution of votes. At the same
time, this figure does not show a clear pattern of elected candidates having less well
spread electoral support than non elected candidates. But likewise the EBMoran’s
139
Figure 3.4: Distribution of the geographical homogeneity of electoral support (mea-sured by spGINI of all candidates
Countries
spG
INI
Not elected Elected
0.0
0.2
0.4
0.6
0.8
1.0
Belgium Brazil Ecuador Finland Latvia
Countries
Geo
grap
hica
l hom
ogen
eity
(sp
atia
l GIN
I)
Note: solid lines represent quantiles .25, .5 and .75.
I results, this figure shows a noticeable dispersion of scores. That means candidates
ended up their elections with diverse patterns of territorial distribution of electoral
support. Which, again, is by itself a great indication that the link between PV and
Localisme is not direct and certainly not deterministic. Multiple situations coexist
under the same OLPR system. But now we have all the instruments we need to
move on to the statistical test of whether having geographical electoral parishes were
or not an electorally profitable electoral pattern in the elections covered here.
140
3.4 Statistical models: does electoral localism pay
off?
In order to answer this question, I will model the impact of EBMoran’s I and
of spGINI on the odds of a candidate being elected. Both concentrating votes ge-
ographically and spreading the votes across the territory are expected to improve
the odds of winning17. This model will be specified as a bayesian multilevel logistic
regression equation. Therefore, the unit of analysis in the first level are, of course,
each candidate i, and the outcome variable is a binary identification of whether each
candidate was or not elected. At the second level, the clusters j will be the electoral
circumscriptions. Besides the general importance of using multilevel models to take
the shared electoral contexts of candidates into account, it will be also specifically
helpful in order to deal with the aforementioned fact that EBMoran’s I is dependent
on the geographical configuration of municipalities in each electoral circumscription.
In other words, besides modeling the non-independence between candidates, using
multilevel models here also isolates in the error term of the 2nd level the variance
introduced by each circumscription’s neighbor structure. All models will be specified
with intercepts and slopes for EBMoran’s I and spGINI varying per electoral circum-
scription, with correlation between the random effects. The general specification of
17Notice that I do not include here the local G, used for Figure 3.2, as an explanatory variable.Recall that in the analysis with the local G measure the unit of analysis were the municipalities,while the statistical model uses the candidates as cases.
141
the preferred model is represented in equation 3.6:
elected ∼ Bern(θi) logit(θi) = β0ij + β1ijx1 + β2ijx2 +K∑k=3
βkXk
β[0:2]j = γ[0:2]j γ[0:2]j ∼ N(0, ξ) β[3:K]j ∼ N(0, 10) (3.6)
ξ = diag(σ)P diag(σ) σ ∼ Cauchy(0, 2) P ∼ LKJcorr(2)
In the general model specification, x1 is the variable EBMoran’s I, x2 is the
spGINI and Xk is the vector of fixed effects predictors, including the controls18. As
a brief description of the control variables, Incumbent is a binary predictor that
identifies whether each candidate was elected for the national (Lower) Chamber in
the former election. PctP ty is the percentage of total votes earned by the party
of candidate i at the circumscription k. It is expected to increase the electoral
performance, showing that despite of geographical patterns of electoral support,
the partisan label is still crucial even in PV systems. PctClosedList is the total
percentage of votes in a circumscription k that were cast directly for parties, i.e. from
voters who chose not to do PV. It is expected to decrease the electoral performance of
individual candidates, because when controlled for PctP ty, PctClosedList measures
only pooling votes that were lost by all candidates. NCand and NMunicip are,
respectively, the total number of candidates and the total number of municipalities
in a given circumscription k. Both enter the model not for substantive reasons, but
to account for the possibility that they might have residual relationship with both
EBMoran’s I or sptGINI, and also marginally with the odds of winning (at least in
the case of NCand). Graphics with the statistical distribution of all variables can
18Also notice in the third line of the equation that the covariance matrix for the correlatedrandom effects applies the separation approach proposed by Barnard et.al.(2000) and uses theLKJ correlation distribution, defined by Lewandowski et.al.(2009), as hyperprior.
142
be found in Annex 3.B.
I will present results for 3 different models, each one in two versions. Model
number 1 will include EBMoran’s I, but exclude spGINI, while model 2 will invert
that. Model 3 will include both variables. Each of these models will appear once
with EBMoran’s I calculated using a neighbor matrix WI×J of type Queen 1, and
another time with Queen 2. All models were fitted using Stan software version 2.6.0,
in 4 chains with 50 thousand iterations each, half burned-in, and thinned at each
tenth iteration to ease computational manipulation of the results. Usual convergence
checks show no evidence of non-convergence for any of the models19. The results of
the estimates appear in Table 3.1.
Comparing the models 1a or 1b to models 2a or 2b, one can see why it was im-
portant to also calculate a variable like spGINI, which is, in a way, a complementary
measure to EBMoran’s I. Because spGINI is capable of detecting votes well spread
in the territory, while EBMoran’s I is not, models 1a and 1b without spGINI could
give a false impression that only concentrating votes in the territory gives electoral
payoff, since on average the effect of EBMoran’s I in both these models has 100%
probability of having the right sign and is of great magnitude. However, if we substi-
tute EBMoran’s I for spGINI, like in models 2a and 2b, we also find that spreading
votes across the territory pays off similarly well. On the contrary of that being a
paradox, it actually shows that both things should be modeled together so we avoid
having only a partial picture of the question being studied. It is exactly what it is
done in models 3a and 3b, where the two complementary measures appear together.
One can easily notice that, first of all, no matter which neighbor matrix we calcu-
late EBMoran’s I with, these models that include both EBMoran’s I and spGINI are
19Traceplots of the chains, as well as R-hat statistics and effective sample size of the parameterscan be found online at: http://vasselai.net/usp/final/chapter3
143T
able
3.1:
Bay
esia
nm
ult
ilev
ello
gist
icm
odel
sof
the
odds
ofb
eing
elec
ted
(a)W
I×J
=Q
uee
n1
(b)W
I×J
=Q
uee
n2
mod
els:
1a2a
3a
1b
2b
3b
vari
able
s:β
σP
r.β
σP
r.β
σP
r.β
σP
r.β
σP
r.β
σP
r.
fixe
deff
ects
:In
terc
ept
-4.5
8.5
61
-6.6
3.7
11
-8.4
8.7
71
-3.9
5.5
71
-6.6
3.7
21
-7.5
2.7
61
EB
Mor
an’s
I3.
37.3
41
3.8
5.3
71
4.9
2.4
71
5.2
8.5
11
spG
INI
3.46
.57
14.7
4.5
31
3.4
6.5
51
4.5
7.5
51
Incu
mb
ent
3.26
.10
13.
31.1
01
3.1
5.1
01
3.2
7.1
01
3.3
1.1
01
3.1
7.1
01
Pct
Pty
5.24
.45
16.
96.4
41
5.4
9.4
61
5.4
9.4
51
6.9
6.4
51
5.8
7.4
61
Pct
Clo
sed
Lis
t1.
62.6
1.9
9-1
.06
.70
.06
-1.1
1.7
4.0
71.5
5.6
11
-1.0
7.6
9.0
6-1
.32
.78
.04
ln(N
Can
d)
-.12
.08
.94
-.11
.09
.90
-.08
.09
.82
-.09
.09
.85
-.11
.09
.90
-.07
.10
.75
ln(N
Mu
nic
ip)
.18
.08
.99
.34
.08
1.3
6.0
91
.04
.09
.68
.34
.08
1.1
9.1
0.9
7
ran
dom
effec
ts:
σ2
σ2
σ2
σ2
σ2
σ2
Inte
rcep
t.4
41.
691.3
0.3
61.6
91.5
1E
BM
oran
’sI
3.46
4.3
75.5
26.7
6sp
GIN
I3.
242.6
63.2
42.7
9
WA
IC:
4315
.69
4522
.40
4151.4
04292.3
04521.8
04139.7
2D
IC:
4360
.64
4671
.72
4246.6
04338.0
84658.3
54234.5
4
cor(Y,Y
)2:
.35
.33
.37
.35
.33
.37
N.c
lust
er:
7777
77
77
77
77
N.o
bs:
1033
310
333
10333
10333
10333
10333
-β
andσ
are
,re
spec
tivel
y,th
em
ean
and
the
standard
dev
iati
on
of
the
para
met
er’s
post
erio
rdis
trib
uti
on.
Pr.
isth
epro
babilit
yth
at
the
para
met
erhas
the
exp
ecte
dsi
gn.
-P
ara
met
ers
that
hav
e0
wit
hin
thei
rH
PD
[5%
,95%]
inte
rval,
i.e.
that
hav
ele
ssth
an
90%
Pr.
,are
colo
red
ingre
y.-
Para
met
ers
are
inlo
g-o
dds.
144
clearly the ones with the best relative fit, as it is shown by WAIC and DIC statistics
being very much smaller in models 3a and 3b than in the other models20. Also the
absolute fit represented by the correlation between the outcome variable and the
logit of the posterior mean of the predicted values is slightly greater in models 3.
Considering that no evidence of relevant multicollinearity between EBMoran’s I and
spGINI was found, the fact the preferred models are those with number 3 is an initial
evidence of the importance of including both measures to get a better assessment of
the importance of geographical distribution of votes under OLPR systems. In any
event, the most interesting part, of course, are the results regarding the effects of
our predictors of interest. From now on, I will focus on models 3a and 3b to better
detail these results.
In these models, the posterior mean of the effect of EBMoran’s I actually shows
the average fixed-effect of the most extreme geographical concentration of electoral
support (EBMoran’s I=1). While the posterior mean of the effect of spGINI shows
the average fixed-effect of the perfectly homogeneous distribution of electoral sup-
port across municipalities (spGINI=1). The most important things to notice are
that, firstly, no matter which neighbor matrix was used to calculate EBMoran’s I,
both the effects of EBMoran’s I and spGINI have 100% probability of having the ex-
pected sign, i.e. of increasing by some amount the odds of a candidate being elected.
Secondly, in both cases such amount seems to be of great magnitude. Which means
that, ceteris paribus, both concentrating votes geographically or spreading votes ge-
ographically increased considerably the chances of winning for the candidates in the
elections covered by my sample. When EBMoran’s I was calcualted with a neighbor
20The difference in DIC even surpasses by far the standard threshold of a difference of 2, proposedby Raftery (1995) for the related BIC measure, as the rule of thumb of relevant differences of fitbetween models
145
matrix of type Queen 1, the average effect of perfectly homogeneous distribution
of the electoral support on the chances of being elected was even slightly greater
than that of the most extreme index of geographical concentration. While when
EBMoran’s I was calculated with a neighbor matrix of type Queen 2, it was the
opposite. As for the other results, the most relevant is to point out that, as ex-
pected, being an incumbent at the national (Lower) Chamber had a positive and
strong effect on the odds of being elected, as did the total percentage of partisan
votes. The latter means that even under a PV system where there is more than one
candidate per party running in the same electoral circumscription, party electoral
strength matters.
Now, let’s focus a bit more on the effect of concentrating votes geographically.
One way to have a better idea of its effect on the electoral outcome of candidates is to
specifically calculate predicted probabilities of winning for each value of EBMoran’s
I. In a single level model, it would be trivial. It would be just a matter of fixing
a given value for each other predictor, then let EBMoran’s I vary. However, in a
multilevel-model it can get more complicated, due to the need of taking the ran-
dom effects into account. A computationally reasonable solution is to calculate a
point-prediction for each case in the data, considering the posterior of intercepts
and slopes of EBMoran’s I and of spGINI in each 2nd level cluster (i.e. each elec-
toral circumscription of the five countries), and then summarize. Hence, the overall
prediction will be the median prediction of all the predictions run at each of the 77
electoral circumscriptions in the data.
Figure 3.5 shows that, in fact, overall, higher levels of EBMoran’s I (e.g.> .45)
are expected to bring, on average, a steep increase in the probability of a candidate
being elected. While lower levels of positive EBMoran’s I (e.g.<0.45) seems to bring
146
Figure 3.5: Median cross-level predicted probabilities of being elected, according tothe level of geographical concentration of votes
−.5
−.4
−.3
−.2
−.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
0.1.2.3.4.5.6.7.8.91
(a) neighbors = contiguous municipalities
WIxJ = Queen1)
distrib. of EBMoran's I
Overall
−.5
−.4
−.3
−.2
−.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
0.1.2.3.4.5.6.7.8.91
(b) neighbors = contiguous municip. + 1st neighbors
of neighbors (WIxJ = Queen2)
.25−.75 qtile .1−.9 qtile
Overall
0
.25
.5
.75
1 Belgium Brazil
0
.25
.5
.75
1 Belgium Brazil
0
.25
.5
.75
1 Ecuador
−.5
−.2
5 0
.25 .5 .75 1
Finland
0
.25
.5
.75
1 Ecuador
−.5
−.2
5 0
.25 .5 .75 1
Finland
−.5
−.2
5 0
.25 .5 .75 1
0
.25
.5
.75
1 Latvia
−.5
−.2
5 0
.25 .5 .75 1
0
.25
.5
.75
1 Latvia
Geographic concentration of electoral support (EBMoran's I)
Pre
dict
ed p
roba
bilit
y of
win
ning
147
almost no change. That pattern varies a little between countries and the strength
of the effect of EBMoran’s I also depends on the neighbor matrix used to calculate
it. However, it is possible to see that in all cases, the density distribution (dashed
lines) of real EBMoran’s I in the data does not cover most the effect of EBMoran’s
I. That density shows where the real data lay, so very few cases in the data achieve
the higher values of EBMoran’s I that are the ones that really increase the predicted
probability. In substantive words, it means that only very rarely candidates present
a level of geographical concentration of votes that is capable of predicting a more
profound impact on the probabilities of getting elected. Again, notice that this is
true for all countries and with both definitions of weight matrix.
Lastly, it is worth exploring the random effects of the electoral circumscriptions.
More precisely, we can look at how much the particular posterior mean of the effect
of EBMoran’s I at each electoral circumscription contributes to its overall posterior
mean. It is a good way of assessing if the results seen in table 3.1 are consistent
across each electoral area of the countries, or if them vary widely. In Figure 3.6,
I show the contribution of these within cluster random effects on the fixed effect
of EBMoran’s I. For the sake of space, I will only present the result for model 3b,
i.e. the one where EBMoran’s I was calculated with the neighbor matrix of type
Queen 2. But the same patterns are yielded by model 3a as well. As it can can be
seen in the graphics, in fact the impact of concentrating votes geographically on the
odds of being elected is different across the electoral circumscriptions. It is much
greater in some parts of the countries, while much smaller in others, indicating that
not only the pattern of geographical distribution of electoral support varies among
candidates and that very few have strongly concentrated votes in the territory, but
also that the payoff of concentrating votes in the territory is not the same in every
148
Figure 3.6: Random effects at the electoral circumscriptions - contribution of thevarying slopes of EBMoran’s I to its overall slope
●
●●
● ● ●●
●● ● ●
●●
● ●●
●
●
● ●● ● ●
●●
●
Ser
gipe
Per
nam
buco
Rio
G.d
o N
orte
Ror
aim
a
Mat
o G
.do
Sul
Pia
ui
Esp
.San
to
Am
apa
Am
azon
as
Bah
ia
Cea
ra
Ala
goas
Goi
as
Mat
o G
ross
o
Sta
. Cat
arin
a
Acr
e
Mar
anha
o
Rio
G.d
o S
ul
Par
aiba
Toca
ntin
s
Par
ana
Par
a
Sao
Pau
lo
Ron
doni
a
Rio
de
Jane
iro
Min
as G
erai
s
−8−7−6−5−4−3−2−1
01234567 Brazil
●
●
● ●
● ● ● ●
● ● ● ●
●●
Ete
laS
avo
Ham
e
Poh
jois
Sav
o
Lapl
and
Ala
nd
Sat
akun
ta
Oul
u
Pirk
anm
aa
Cen
tral
Fin
land
Kym
i
Nor
thK
arel
ia
Vaa
sa
Uus
imaa
Var
sina
isS
uom
i
−7
−6
−5
−4
−3
−2
−1
0
1
2
3
4
5 Finland
●
●
●
● ●
●●
●
●
●
Leuv
en
Wes
t.Fla
nder
s
Bru
ssel
s.H
alle
.Vilv
oord
e
Lim
burg
Ant
wer
p
Luxe
mbo
urg
Eas
t.Fla
nder
s
Nam
ur
Hai
naut
Lieg
e
−6
−5
−4
−3
−2
−1
0
1
2
3
4 Belgium
●
● ●
● ●● ●
● ● ● ● ● ● ●● ● ● ●
● ● ●●
●
San
taE
lena
Sto
Dgo
Tsa
chila
s
Nap
o
Can
ar
ElO
ro
Mor
onaS
antia
go
Zam
oraC
hinc
hipe
Pas
taza
Tung
urah
ua
Imba
bura
Bol
ivar
Man
abi
Ore
llana
Azu
ay
Suc
umbi
os
Car
chi
Esm
eral
das
Loja
Pic
hinc
ha
Cot
opax
i
LosR
ios
Chi
mbo
razo
Gua
yas
−6−5−4−3−2−1
01234567 Ecuador
●●
●
●
Zem
gale
s
Kur
zem
es
Vid
zem
es
Latg
ales
−4
−3
−2
−1
0
1
2
3
4
5 Latvia
Electoral circumscriptions
Ran
dom
effe
cts
on th
e sl
ope
of E
BM
oran
's I
parts of countries. Different electoral patterns emerge, which suggests that if they
were to be pursued consciously as strategy, different strategies would be at hand,
depending on the candidates and depending on where they are competing. This
is, in fact, yet another clear evidence that, although there is definitively room for
electoral localism under the PV of the OLPR systems, such pattern is not close to
be as an hegemonic and decisive as the literature usually assumes. Although I do
not present similar graphics for other parameters for the sake of space, results are
the same, i.e. different levels of effect across electoral circumscriptions.
149
3.5 Conclusion
This chapter aimed at the common assumptions about what would be the incen-
tives of electoral systems that adopt the personal voting mechanism in what regards
the geographical distribution of candidates’ electoral support. In a word, does such
a system really leads candidates to end up having electoral bunkers to where they
might deliver pork in the future? Substantively, one first conclusion was that, in
general, most candidates do not have great levels of geographical concentration of
electoral support, when measured by a version of the widely used Moran’s I.
Notice that it is not that concentrating votes geographically isn’t the only pat-
tern presented by candidates, what by itself already demystifies the common idea of
that automatic link seen in proposition 3.1, between PV and localism. More than
that, I have found evidence that having high levels of geographical concentration of
votes is actually the exception, not the rule. The scenario is, indeed, far from the
assumptions that are usually made. It is true, on the other hand, that elected can-
didates generally concentrate their votes more than defeated candidates. However, I
have found that most candidates are only following the demographic distribution of
electors, what means that great part of those who concentrate votes might be only
following the geographical concentration of voters.
Regarding the pay off of each electoral geographical pattern, both concentrat-
ing votes and spreading votes proved to be profitable. Of course, it is difficult to
say whether a given distribution of the final result is the one intended by a former
deliberate strategy of a given candidate. Still, it seems that I have offered enough
evidence to sustain that in fact different territorial configurations pay off electorally,
be them designed configurations or not. Or, what is more important, there is more
150
than one way to look for votes under PV were candidates able to fully control that.
There is no unique incentive for an automatic tendency of candidates under PV to
seek the formation of electoral parishes where to deliver parochial goods. Also, the
effect of concentrating votes geographically varies across the electoral circumscrip-
tions of countries. It means that in some, concentrating votes pay off much more
than in others, which is yet another evidence of the multiplicity that might exist
under a PV system.
Additionally, the results have shown that while the potential contribution of con-
centrating votes geographically, to the odds of being elected, can be very strong, in
reality it is rare that candidates achieve scores of concentration that are predicted
to really increase those odds substantively. Or in other words, geographically con-
centrated votes can be really valuable, but few are the candidates that have been
able to mine such an asset in high scale. Lastly, a side result also showed that
the partisan performance is still of great importance for the odds of being elected,
even under PV systems. There is, in conclusion, no empirical motivation to keep
assuming a link between PV and localism - an assumption that means, after all, to
assume what remains to be proven. On the contrary, there is evidence that such
relationship is less frequent, weaker or at least more nuanced. In a broader sense, my
findings suggest that we should be more aware about the meaning and the strength
of the incentives we expect from electoral institutions. Both politicians and voters
can play the institutional game, and win or loose it, in varied ways.
151
3.6 Annexes
Annex 3.A – Distribution of EBMoran’s I according to additional spec-
ifications of the neighbor matrix WI×J
queen1
−0.5
0.0
0.5
1.0
Bel
gium
Bra
zil
Ecu
ador
Fin
land
Latv
ia
−0.5
0.0
0.5
1.0
queen2
−0.5
0.0
0.5
1.0B
elgi
um
Bra
zil
Ecu
ador
Fin
land
Latv
ia−0.5
0.0
0.5
1.0
queen3
−0.5
0.0
0.5
1.0
Bel
gium
Bra
zil
Ecu
ador
Fin
land
Latv
ia
−0.5
0.0
0.5
1.0
powerdist1
−0.5
0.0
0.5
1.0
Bel
gium
Bra
zil
Ecu
ador
Fin
land
Latv
ia
−0.5
0.0
0.5
1.0
doublepower
−0.5
0.0
0.5
1.0
Bel
gium
Bra
zil
Ecu
ador
Fin
land
Latv
ia
−0.5
0.0
0.5
1.0
Country
EB
Mor
an's
I
152
Annex 3.B – Distribution of variables used in the main models of this
chapter
EBMoran's I
−0.
4
−0.
2
0.0
0.2
0.4
0.6
0.8
0
1
2
3
4
5
spGINI
0.2
0.4
0.6
0.8
1.0
0.0
0.5
1.0
1.5
2.0
PctPty
0.0
0.2
0.4
0.6
0.8
0
2
4
6
8
10
12
PctClosedList
0.0
0.1
0.2
0.3
0.4
0.5
01234567
NCand
2 3 4 5 6 7 8
0.0
0.1
0.2
0.3
0.4
0.5
NMunicip
2 3 4 5 6 7
0.0
0.1
0.2
0.3
0.4
Incumbent
0.0
0.2
0.4
0.6
0.8
1.0
0
2
4
6
8
Variable
Den
sity
153
References
Abranches, Sergio H. H. de. (1988), ”Presidencialismo de Coalizao: O Dilema Institucional
Brasileiro”. In: Dados, v. 31, n. 1, pp. 5-38.
Abrucio, Fernando Luiz (1998). Os baroes da federacao: os governadores e a redemocratizacao
brasileira. Sao Paulo: HUCITEC: Depto. de Ciencia Polıtica da USP.
Agnew, John A. (1987). Place and Politics: The Geographical Mediation of State and Society
Routledge Library Editions: Political Geography, Taylor and Francis. Kindle Edition.
Aleman, E., and Kellam, M. (2008). The nationalization of electoral change in the Americas,
27(2), 193–212.
Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R. (2003). Fractionaliza-
tion. Journal of Economic Growth vol. 8 (no. 2), 155–194, June.
Allen, Nathan (2012). “Clientelism and the Personal Vote in Indonesia”. Paper presented at
the Western Political Science Association Annual Conference. Portland, March 23rd.
Amado, Gilberto (1931). Eleicao e Representacao. Rio de Janeiro: Sa Cavalcanti Editores.
Ames, Barry (1995). Electoral Rules, Constituency Pressures, and Pork Barrel: Bases of Vot-
ing in the Brazilian Congress. In: The Journal of Politics, Vol. 57, No. 2. May, pp. 324-343.
Ames, Barry (2001). The deadlock of democracy in Brazil. Michigan: Michigan University Press.
Amorim Neto, Octavio and Cox, Gary. Electoral institutions, cleavage structures and the number
of parties. American Journal of Political Science, v. 41, p. 149-74, 1997.
Amorim Neto, Octavio and Santos, Fabiano (2003). The Inefficient Secret Revisited: The Legisla-
tive Input and Output of Brazilian Deputies. Legislative Studies Quarterly, Iowa City, Iowa, EUA,
v. 28, n. 3, p. 449-479.
Anselin, L. (1995). Local indicators of spatial association – LISA. Geographical Analysis, 27,
93-115.
Antonakis, John, Bendahan, Samuel, Jacquart, Philippe and Lalive, Rafael. 2010. On making
154
causal claims: a review and recommendations. In: The Leadership Quarterly, n. 21
Arretche, Marta (2007). A Centralizacao no Estado Federativo Brasileiro. Tese de Livre-Docencia
– Faculdade de Filosofia, Letras e Ciencias Humanas, USP, Sao Paulo.
Assis Brasil, Joaquim Francisco (1893). Democracia representativa: do voto e do modo de votar.
Rio de Janeiro: Typ. de G. Leuzinger, 175 pp.
Assuncao, Renato and Reis, Edna (1999). A new proposal to adjust Moran’s I For population
density. Statistics in Medicine, 18.
Avelino, George; Biderman, Ciro; Silva, Glauco Peres da (2011). A Concentracao eleitoral nas
eleicoes paulistas: medidas e aplicacoes. Dados, Rio de Janeiro, v.54, n.2.
Barnard, John; McCulloch, Robert and Xiao-Li Meng (2000). Modeling covariance matrices in
terms of standard deviations and correlation, with application to shrinkage”. In: Statistica Sinica
10(2000), 1281-1311.
Belisario de Souza, Francisco de (1979). O Sistema Eleitoral do Imperio. Brasılia: Senado Federal.
Benevides, Maria Victoria de M. (1981). A UDN e o udenismo – ambiguidades do liberalismo
brasileiro. Rio de Janeiro: Paz e Terra.
Bentler, P. M., and T. Raykov. 2000. On measures of explained variance in nonrecursive structural
equation models. Journal of Applied Psychology 85: 125-131. Benevides, Maria Victoria de M.
(1989). PTB e o trabalhismo: partidario e sindicato em Sao Paulo. Sao Paulo:Brasiliense.
Bochsler, Daniel (2006). ”Ethnic diversity, electoral system constraints and the nationalisation
of political parties. A triangle model, applied on the Central and Eastern European countries.”
CEU Political Science Journal 1 (4):6-37.
Bochsler, Daniel (2010). Measuring party nationalisation: a new Gini-based indicator that corrects
for the number of units. Electoral Studies, v. 29, n. 1, p. 155-68, March.
Boix, Carles (1999). “Setting the Rules of the Game: The Choice of electoral systems in ad-
vanced democracies.” American Political Science Review 93, no. 3: 609–24.
Bollen, Kenneth (1989). Structural Equations with Latent Variables. Wiley-Interscience.
Bormanna, Nils-Christian and Matt Golder (2013). Democratic Electoral Systems around the
155
world, 1946–2011. Electoral Studies, Volume 32, Issue 2, June 2013.
Bowler, S. (2000). “Parties in Legislatures: Two Competing Explanations”, in R. Dalton e M.
Wattenberg (eds.), Parties without Partisans: Political Change in Advanced Industrial Democra-
cies. Oxford, Oxford University Press, pp. 157-179.
Bowler, S. and Farrell, D.F. (1993). ”Legislator Shirking and Voter Monitoring: Impacts of Eu-
ropean Parliament Electoral Systems upon Legislator-Voter Relationships,” Journal of Common
Market Studies 31 (1): 45-69.
Braga, Maria do S.S. (2006). O processo partidario-eleitoral brasileiro: padroes de competicao
polıtica (1982-2002). Sao Paulo: Hu-manitas/Fapesp.
Brancatti, Dawn. Global Elections Database [computer file] (2013). New York: Global Elections
Database [distributor], Date Accessed 2013. Website: http://www.globalelectionsdatabase.com.
Britto, Luiz Navarro de (1965). ”A representacao proporcional”. In: Revista de Informacao
Legislativa, n. 6, pp.17-30.
Cain, Bruce E., Ferejohn, John and Fiorina, Morris (1987), The PersonalVote: Constituency
Service and Electoral Service. Cambridge, Harvard University Press.
Calvo, Ernesto (2009). ”The competitive road to proportional representation: partisan biases
and electoral regime change under increasing party competition”. In: World Politics, V.61, N.2,
April, pp 254-295.
Campbell, P. e Alistair, C. (1989). French Electoral Systems and elections since 1789. Alder-
shot: Gower.
Caramani, Daniele (1996). The nationalisation of electoral politics: conceptual reconstruction
and review of the literature. West European Politics, v. 19, n. 2, p. 205-24, April.
Caramani, Daniele (2003). ”The End of Silent Elections: The Birth of Electoral Competition,
1832–1915”. In: Party Politics. V.9, n.4, pp.411-443.
Caramani, Daniele (2004). The nationalization of politics - the formation of national electorates
and party systems in Western Europe. Cambridge: Cambridge University Press.
Carey, John M e Shugart, Matthew S. (1995). Incentives to Cultivate a Personal Vote: A Rank
Ordering of Electoral Formulas. Electoral Studies, vol. 14, no 4, pp. 417-439.
156
Cederman, Lars-Erik; Brian Min; Andreas Wimmer (2009). Ethnic Power Relations dataset,
http://hdl.handle.net/1902.1/11796 UNF:5:k4xxXC2ASI204QZ4jqvUrQ==V1
Chang, Eric C. C., and Miriam A. Golden (2006). “Electoral Systems, District Magnitude, and
Corruption.” British Journal of Political Science 37 (1): 115-137.
Cheibub, J.A. and Camargos, M. (2002). Electoral strategies and electoral success in an Open-PR
system. Yale: Yale University. mimeo.
Chhibber, P.and Kollman, K. (1998). Party Aggregation and the Number of Parties in India
and the United States. American Political Science Review v.92, n.2, p.329-342, June.
Claggett, W., Flanigan, W. and Zingale, N. (1984). Nationalization of the American Electorate.
American Political Science Review, v. 78, n. 1, p. 77-91, March.
Cliff, A. D. and Ord, J. K. (1981). Spatial Processes: Models and Applications . London: Pion.
Colomer, Josep M. (2005). ”It’s the Parties that Choose Electoral Systems (or Duverger’s Laws
Upside Down)” Political Studies, 53, 1, 2005.
Colomer, Josep M. (2009). Personal Representation: The Neglected Dimension of Electoral Sys-
tems.
Converse, P. (1972). Change in the American electorate. In The human meaning of social change,
A. Campbell and P. Converse (eds.). New York: Russell Sage.
Coopedge, M. (1997). District magnitude, economic performance and party-system fragmenta-
tion in five Latin-American countries. Comparativa Political Studies, v. 30, p. 156-85.
Cox, Gary. (1987), The Efficient Secret: The Cabinet and the Development of Political Par-
ties in Victorian England. Cambridge, Cambridge University Press.
Cox, Gary (1997). Making Votes Count: strategic coordination in the World’s electoral systems.
New York: Cambridge University Press
Cox, Gary (1999). Electoral rules and electoral coordination. Annual Review of Political Sci-
ence, v. 2, p. 145-61.
Cox, Gary, and Knoll, Jonathan S. (2003). “Ethnes, Fiscs and Electoral Rules: The Determi-
157
nants of Party System Inflation” Paper presented at the Annual Meeting of the American Political
Science Association, August 28-31.
Cox, Gary and McCubbins, M. (1993), Legislative Leviathan: Party Government in the House.
Berkeley,University of California Press. Fiscal decentralization and particularistic spending across
countries.
Crisp, B. F., and R. E. Ingall (2002). “Institutional Engineering and the Nature of Representation:
Mapping the Effects of Electoral Reformin Columbia.” American Journal of Political Science 46(4):
733–48.
Crisp, Brian F.; Joshua D. Potter, and John J. W. Lee (2012). Entry and Coordination in Mixed-
Member Systems: A Controlled Comparison Testing the Contamination Hypothesis. The Journal
of Politics, Volume 74, Issue 02, April.
Denemark, D. (2000). “Partisan Pork Barrel in Parliamentary Systems: Australian Constituency-
Level Grants.” Journal of Politics 62(3): 896–915.
Desmet, Klaus; Ignacio Ortuno-Ortın and Romain Wacziarg (2012). Journal of Development
Economics.
Diaz-Cayeros, Alberto. Kenneth M. McElwain, Vidal Romero, Konrad A. Siewierski (2009). Fiscal
decentralization and particularistic spending across countries, Manuscript.
Duverger, Maurice (1951) ”The Influence of the Electoral System on Political Life”, International
Social Science Bulletin 3: 314–52.
Duverger, Maurice (1954). Political Parties. London: Methuen and CO. LTD.
Fausto, Boris (1986). A Revolucao de 1930: historiografia e historia. Sao Paulo: Brasiliense.
Fearon, J. (2003). Ethnic and cultural diversity by country. Journal of Economic Growth, vol. 8
(no. 2), 195–222 June.
Fenno, R. (1978). Home style: house members and their districts. Boston: Little Brown.
Figueiredo, Argelina C. and Limongi, Fernando (1998). ”Bases Institucionais do Presidencial-
ismo de Coalizao”. Lua Nova, v. 44, n. 81-106.
Figueiredo, Argelina C. and Limongi, Fernando (1999). Executivo e legislativo na nova ordem
158
constitucional. Rio de Janeiro, Sao Paulo, Editora FGV.
Figueiredo, Argelina C. and Limongi, Fernando (2002). ”Incentivos eleitorais, partidos e polıtica
orcamentaria”. Dados - Revista de Ciencias Sociais, v. 45, n. 2, pp. 303-44.
Figueiredo, Argelina C. and Limongi, Fernando (2005). ”Processo Orcamentario e Comporta-
mento Legislativo: Emendas Individuais, Apoio ao Executivo e Programas de Governo”. Dados -
Revista de Ciencias Sociais, v. 48, n. 4, pp. 737-76.GERRING J. (2007), “Techniques for choosing
cases”, in John Gerring, Case Study Research,pp.86-150.
Fiorina, Morris, P. (1997). Congress: the keystone of Washington establishment. New Haven,
Yale University Press.
Fotheringham, A. Stewart; Chris Brunsdon and Martin Charlton (2002). Geographically Weighted
Regression - the analysis of spatially varying relationships. West Sussex: John Wiley and Sons,
LTD.
Gaines, Brian J. (1999). “Duverger’s Law and the Meaning of Canadian Exceptionalism”. Com-
parative Political Studies 32: 835-861.
Geddes, B. and Benton, A. (1997). The effect of Federalism on Latin America party systems.
Conference on The Transformation of Argentina. Argentina: Universidad de San Andres.
Geddes, Barbara and Ribeiro Neto, Artur (1992). Institutional sources of corruption in Brazil,
Third World Quarterly, 13:4, 641-661.
Gerring, J.; S Thacker and C. Moreno (2005). Centripetal Democratic Governance: A Theory
and Global Inquiry. American Political Science Review 99(4):567-81.
Getis, A. and Ord, J. K. (1992). The analysis of spatial association by use of distance statis-
tics. Geographical Analysis 24: 189–206.
Geys, B. (2006). District magnitude, social heterogeneity and local party system fragmentation.
Party Politics, v. 12, n. 2, p. 281-97.
Goes, Walder de (1992). Em busca de um novo sistema polıtico. In: Velloso, Joao Paulo dos
Reis (org.). O Brasil e as reformas polıticas. Rio de Janeiro: J. Olympio.
Golden, Miriam and Lucio Picci (2008). “Pork-Barrel Politics in Postwar Italy, 1953–94”. In:
American Journal of Political Science, Vol. 52, No. 2, April, Pp. 268–289.
159
Golder, Matt (2005). ”Democratic Electoral Systems Around the World, 1946-2000.” Electoral
Studies 24: 103-121.
Golder, Matt (2006). Presidential coattails and legislative fragmentation. American Journal of
Political Science, 50(1), 34-48.
Golder, Matt and Clark, William (2006). Rehibilitating Duverger’s theory: Testing the mechanical
and strategic modifying effect of electoral laws. Comparative Political Studies, 39(6), 679-708.
Greene, W. (2011). Econometric Analysis. Prentice Hall; 7 edition, February 13.
Grofman, Bernard and Arend Lijphart (2003). The Evolution of Electoral and Party Systems
in the Nordic Countries. new York: agathon press.
Grumm, J. G. (1958). Theories of electoral systems. Midwest Journal of Political Science, v.
2, p. 357-76.
Harbers, I. (2010). Decentralization and the development of nationalized party systems in new
democracies: evidence from Latin America. Comparative Political Studies, 43(5), 606–627.
Hentschke, Jens R. (ed) (2007). Vargas and Brazil: New Perspectives. New York: Palgrave
Macmillan.
Hicken, A. (2009). Building Party Systems in Developing Democracies. Cambridge : Cambridge
University Press.
Hicken, Allen and Heather Stoll (2008). Electoral Rules and the Size of the Prize: How Polit-
ical Institutions Shape Presidential Party Systems. Journal of Politics 70(4), 1-19.
Hicken, Allen and Heather Stoll (2011). Presidents and Parties: How Presidential Elections Shape
Coordination in Legislative Elections. Comparative Political Studies 44(8).
Hicken, Allen, Ken Kollman, and Joel Simmons (2008). “Public Goods Provision and Party Sys-
tem Nationalization.” Working paper. Department of Political Science. University of Michigan,
Ann Arbor.
Hicken, Allen, and Joel W. Simmons, 2008, ”The Personal Vote and the Efficacy of Education
Spending”, American Journal of Political Science, Vol. 52, No. 1, pp. 109–124.
160
Hippolito, Lucia (1985). De Raposas e Reformistas: o PSD e a experiencia democratica brasileira:
1945-64. Rio de Janeiro: Editora Paz e Terra.
Honaker, James and Drew Linzer (2006). Analysis of Voteshares Across Party System. Paper
presented at the annual meeting of the American Political Science Association, Marriott, Loews
Philadelphia, and the Pennsylvania Convention Center, Philadelphia, PA, Aug 31.
Hooghe, Liesbet; Gary Marks and Arjan H. Schakel (2010). The Rise of Regional Authority:
A Comparative Study of 42 Democracies, Routledge.
Ishiyama, J. T. (2002). Regionalism and the nationalization of the legislative vote in post-
communist Russian politics . Communist and Post-Communist Studies, 35(2), 155–168.
Jones, Mark (1993). The political consequences of electoral laws in Latin America and the
Caribbean. Electoral Studies, v. 12, n. 1, p. 59-75.
Jones, Mark (1994). Presidential election laws and multipartism in Latin America. Political
Science Quarterly, 47, 41-57.
Jones, Mark P. (1997). Federalism and the Number of Parties in Argentine Congressional Elec-
tions. The Journal of Politics, v. 59, p. 538-49.
Jones, Mark P. and Mainwaring, Scott (2003). The nationalization of parties and party systems:
an empirical measure and an application to the Americas, . Party Politics, v. 9, n. 2, p. 139-66.
Kaplan, D. (2008). Structural Equation Modeling: Foundations and Extensions. SAGE Publi-
cations, Inc; 2nd edition.
Karvonen, L. (2004). Preferential Voting: Incidence and Effects, International Political Science
Review / Revue internationale de science politique, Vol.25, No. 2 (Apr., 2004), pp. 203-226.
Karvonen, L. (2010). The Personalisation of Politics. A Study of Parliamentary Democracies.
Colchester 2010: ECPR Press, 124 pp.
Kawato, S. (1987). Nationalization and partisan realignment in congressional elections. Amer-
ican Political Science Review, v. 81, p. 1235-50.
Kinzo, Maria D. G. (1989), “O quadro partidario e a Constituinte”. Revista Brasileira de Ciencia
Polıtica, 1 (1): 91-124.
161
Kim, J.O. and Ohn, M.-G. (1992). A theory of minor-party persistence: election rule, social
cleavage and the number of political parties. Social Forces, v. 70, n. 3, p. 575-99.
Kinzo, Maria D. G. (1993), Radiografia do quadro partidario brasileiro. Sao Paulo, Fundacao
Konrad-Adenauer-Stiftung.
Kinzo, Maria D. G. (1997). Governabilidade, estrutura institucional e processo decisorio no Brasil.
Parcerias Estrategicas, vol.1, n.3, p.19-37.
Kollman, K., Hicken, A., Caramani, D., Backer, D., and Lublin, D. (2014). Constituency-level
elections archive [data file and codebook]. Ann Arbor, MI: Center for Political Studies, University
of Michigan [producer and distributor].
Krause, Silvana (2005). Uma analise comparativa das estrategias eleitorais nas eleicoes major-
titarias (1994-1998-2002): coligacoes eleitorais x nacionalizacao dos partidos e do sistema par-
tidario brasileiro. In: Krause, Silvana and Schmitt, Rogerio (org.) Partidos e coligacoes eleitorais
no Brasil. Rio de Janeiro: Fundacao Konrad Adenauer; Sao Paulo: Editora Unesp, pp.115-42.
Krehbiel,Keith. (1992). Information and Legislative Organization, Ann Arbor, Michigan Uni-
versity press.
Kunicova, Jana and Susan Rose-Ackerman (2005). ”Electoral Rules and Constitutional Struc-
tures as Constraints on Corruption”, British Journal of Political Science, Vol. 35, No. 4, 573-606.
Laakso, M. and Taagepera, R. (1979). Effective number of parties: a measure with application to
West Europe. Comparative Political Studies, v. 12, n. 1, p. 3-27, April.
Lago-Penas, Ignacio and Santiago Lago-Penas (2009). “Does the Nationalizationof the Party Sys-
tems Affect the Composition of Public Spending?”. Economics of Governance, 10:85-98.
Lambert, Jacques (1971). Os Dois Brasis. Sao Paulo: Companhia Editora Nacional.
Lamounier, Bolıvar (1982). ”A representacao proporcional no Brasil: mapeamento de um de-
bate”. In: Revista de Cultura e Polıtica, Sao Paulo, n.7, pp.5-42.
Lamounier, Bolıvar (1992). Estrutura institucional e governabilidade na decada de 1990. In:
Velloso, Joao Paulo dos Reis. O Brasil e as reformas polıticas. Rio de Janeiro: J. Olympio. p.
23-47.
Lamounier, Bolıvar. (1994), ”A Democracia Brasileira de 1985 a Decada de 90: A Sındrome
162
da Paralisia Hiperativa”, in J. P. dos Reis Velloso (coord.), Governabilidade, Sistema Polıtico e
Violencia Urbana. Rio de Janeiro, Jose Olympio Editora.
Lancaster, Thomas D. (1986). “Electoral Structures and Pork Barrel Politics”. In: International
Political Science Review / Revue internationale de science politique , Vol. 7, No. 1, Public Policy
and Government Structures, pp. 67-81.
Lavareda, Antonio (1991). ”A democracia nas urnas: o processo eleitoral brasileiro”. Rio de
Janeiro: IUPERJ/Rio Fundo Editora.
Lee ,Jay and David W. S. Wong (2000). Statistical analysis with ArcView GIS. John Wiley
and Sons, INC.
Lee, S. (2009). Bias from Misspecified Spatial Weight Matrices in SAR Models: Theory and
Simulation Studies.
Lewandowski, D., Kurowicka, D., and Joe, H. (2009). Generating random correlation matrices
based on vines and extended onion method. Journal of Multivariate Analysis, 100:1989–2001.
Leys, C. (1959). Models, theories, and the theory of political parties. Political Studies, v. 7,
p. 127-46.
Lijphart, A. (1984). Democracies: patterns of majoritarian and consensus government in twenty-
one countries. London: Yale University Press.
Lima Jr., Olavo Brasil de (1983). Partidos Polıticos Brasileiros: A Experiencia Federal e Re-
gional (1945-1964). Rio de Janeiro, Graal.
Lima Jr., Olavo Brasil de (1999). Reformas de sistemas eleitorais: mudancas, contextos e con-
sequencias. Revista Dados, Rio de Janeiro, v. 42, n. 1.
Linzer, Drew (2012). The Relationship between Seats and Votes in Multiparty Systems. Po-
litical Analysis, N.20.
Lipset, S. and S. Rokkan (1967). Party Systems and Voter Alignments: Cross-National Per-
spectives. New York: The Free Press, p.1-64.
Lipson, L. (1959). Party system in the United Kingdom and the older Commonwealth: causes,
resemblances and variations. Political Studies, v. 7, p. 12-31.
163
Lublin, David (2013). Election Passport. American University. http:///www.electionpassport.com.
Mahoney, J. et ali. (2009), The Logic of Historical Explanation in the Social Sciences”, Com-
parative Political Studies, vol. 42, n. 1, pp. 114-146
Mainwaring, Scott (1991). “Polıticos, partidos e sistemas eleitorais: o Brasil numa perspectiva
comparada”. Novos Estudos CEBRAP, 29:34-58.
Mainwaring, Scott (1993a). Democracia Presidencialista multipartidaria: o caso do Brasil. Lua
Nova, Sao Paulo , n. 28-29, Apr.
Mainwaring, Scott (1993b). ”Presidentialism, Multipartism, and Democracy: The Difficult Com-
bination”. Comparative Political Studies, vol. 26, n. 2.
Mainwaring, Scott and Scully, Timothy (1995). Building democratic institutions: party systems
in Latin America, Stanford, Stanford University Press.
Mainwaring, Scott (1999) Rethinking party systems in the third wave of democratization: the
case of Brazil. Stanford, Stanford University Press.
Mair, P. (1987). Party Organization, Vote Management, and Candidate Selection: Toward the
Nationalization of Electoral Strategy in Ireland. InPennimanand Farrell, p.104–30.
MAR (2009). Minorities at Risk Dataset. College Park, MD: Center for International Devel-
opment and Conflict Management. Retrieved from http://www.cidcm.umd.edu/mar/ on: Feb
2014.
Mayhew, David R. (1974). Congress. The Electoral Connection. New Haven/London, Yale Uni-
versity Press.
Meleshevich, A. (2006). Geographical Patterns of Party Support in the Baltic States, Russia
and Ukraine. European Urban and Regional Studies, 13(2), 113–129.
Manin, Bernard (1997). The Principles of Representative Government. London: Cambridge Uni-
versity Press.
Mann, T. E. (1978). Unsafe at any margin. Washington, D.C.: American Enterprise Institute
for Public Policy Research.
164
Mesquita, L. (2009). Emendas ao Orcamento e Conexao Eleitoral na Camara dos Deputados
Brasileira. (Mestrado) - Departamento de Ciencia Polıtica, Universidade de Sao Paulo, Sao Paulo.
Milesi-Feretti, G-M, R. Perotti, and M. Rostagno (2001). Electoral Systems and Public Spending.
In IMF Working Paper WP/01/22. New York
Moenius, J. and Kasuya, Y. (2004). Measuring Party Linkage across Districts: Some Party System
Inflation Indices and their Properties. Party Politics, V.10/No.5.
Moenius, J. and Kasuya, Y. (2008). The nationalization of party systems: Conceptual issues
and alternative district-focused measures. Electoral Studies, Volume 27, Issue 1, March.
Morgan, Edmund (1988). Inventing the People. New York, W. W. Norton and Company.
Morgenstern, Scott, Swindle, S. M. e Castagnola. (2009). A. Party nationalization and insti-
tutions. The Journal of Politics, v. 71, n. 4, p. 1322-41, October.
Morgenstern, Scott; John Polga-Hecimovich and Peter M. Siavelis (2014). Seven Imperatives for
Improving the Measurement of Party Nationalization with Evidence from Chile. Electoral Studies,
Volume 33, March.
Mustillo, Thomas, and Sarah Mustillo (2012). ”Party Nationalization in a Multilevel Context:
Where’s the Variance?” Electoral Studies.
Nicolau, Jairo (1996). Multipartidarismo e democracia. Rio de Janeiro, FGV Editora.
Nicolau, Jairo (2004). “Partidos Polıticos na Republica de 1946: velhas teses, novos dados”.
Dados, Rio de Janeiro, v. 47, n. 1.
Nicolau, Jairo M. and Schmitt, Rogerio A. (1995) Sistema eleitoral e sistema partidario. Lua
Nova, Sao Paulo, n.36, pp. 129-147.
Norris, P. (2002). Ballot Structures and Legislative Behavior. Paper for the conference “Ex-
porting Congress? The Influence of the U.S. Congress on World Legislatures” Jack D. Gordon
Institute for Public Policy and Citizenship Studies, Florida International University, University
Park Campus, Miami, Florida December 6-7.
Oliveira, Lucia Maria Lippi (1973). O Partido Social Democratico. Dissertacao de Mestrado –
Instituto Universitario de Pesquisas do Rio de Janeiro (IUPERJ).
165
ORdeshook, P. and Shvetsova, O. (1994). Ethnic heterogeneity, district magnitude and the number
of parties. American Journal of Political Science, v. 38, p. 100-23.
PAxton, Pamela M., John R. Hipp, Sandra Marquart-Pyatt. 2011. Nonrecursive Models: En-
dogeneity, Reciprocal Relationships, and Feedback Loops. SAGE Publications, Inc; 1 edition.
Pereira, C. and Mueller, B. (2002). ”Comportamento Estrategico em Presidencialismo de Coalizao:
As Relacoes entre Executivo e Legislativo na Elaboracao do Orcamento Brasileiro”. Dados - Re-
vista de Ciencias Sociais, v. 45, n. 2, pp. 265-301.
Pereira, C. and Mueller, B. (2003). ”Partidos Fracos na Arena Eleitoral e Partidos Fortes na
Arena Legislativa: A Conexao Eleitoral no Brasil”. Dados - Revista de Ciencias Sociais, v. 46, n.
4, pp. 735-71.
Persson, Torsten, Guido Tabellini, and Francesco Trebbi (2003). ”Electoral Rules and Corrup-
tion”, Journal of the European Economic Association, Vol. 1, No. 4: pp. 958-989
Polsby, N. W. (1981). The Washington community, 1960– 1980. In The new Congress, T. E.
Mann and N.J. Ornstein (eds.). Washington, D.C.: American Enterprise Institute for Public Pol-
icy Research.
Pomper, G. with S. Lederman (1980). Elections in America. New York: Longman.
Rae, D. W. (1971). The Political Consequences of Electoral Laws. Yale: Yale University Press.
Raftery, Adrian E. (1995). Bayesian Model Selection in Social Research. In: Sociological Method-
ology, V.25, pp.111-163.
Renwick, Alan and Jean-Benoit Pilet (2011). The Personalization of Electoral Systems: The-
ory and European Evidence. Paper prepared for presentation at the ECPR General Conference,
University of Reykjavik, 25-27 August.
Ricci, Paolo (2003). “O conteudo da producao legislativa brasileira: leis nacionais ou polıticas
paroquiais?”. Dados (Rio de Janeiro), Rio de Janeiro, v. 46, n. 4, p. 699-734.
Riker, W. H. (1982). The two-party system and Duverger’s Law: an essay on the history of
Poliltical Science. American Political Science Review, v. 76, p. 753-66.
Roeder, P. (2001). Ethnolinguistic fractionalization (ELF) indices, 1961 and 1985. Retrieved
Dez 2013 from http://weber.ucsd.edu/ proeder/elf.htm.
166
Roma, Celso and Braga, Maria do Socorro (2001). ”Sistema partidario, eleicoes e a questao
federativa no Brasil, 1986-2000”. XXV Encontro Anual da ANPOCS, Caxambu, outubro, mimeo.
Rose, R. and Urwin, D. W. (1970). Persistence and change in Western party systems since 1945.
Political Studies, v. 18, p. 287-319.
Rose, R. and Urwin, D. W. (1975). Regional differentiation and political unity in Western Nations.
v. 1. n. 06-007. London: SAGE Publication.
Samuels, David (1998). Careerism and its consequences: federalism, elections, and policy-making
in Brazil. PhD thesis. University of California, San Diego. 358 pp.
Samuels, David. (2001). “Political Ambition and Pork-Barrel Politics in Brazil.” In Legisla-
tive Politics in Latin America, ed. Scott Morgenstern and Benito Nacif. New York: Cambridge
University Press.
Santos, Wanderley Guilherme (1979). The Calculus of Conflict: Impasse un Brazilian Politics
and the Crisis of 1964. Stanford University, Ph.D. dissertation.
Santos, Wanderley Guilherme (1987). Crise e castigo: partidos e generais na polıtica brasileira.
Sao Paulo. Ed. Vertice.
Sartori, Giovanni. (1968). The Sociology of Parties: A Critical Review. In: Stammer, O., (ed)..
Party Systems, Party Organizations, and the Politics of New Masses. Berlin: Institut fur Politis-
cheWissenschaft an der Freien Universitat, Berlin, p.1-25.
Sartori, Giovanni (1976). Parties and party systems: a framework for analysis. Cambridge: Cam-
bridge University.
Sartori, Giovanni (1994). ”Neither Presidentialism nor Parliamentarism”, in: J.J. Linz e A. Valen-
zuela (ed.) The Failure of Presidential Democracy, Johns Hopkins University Press, Baltimore-
London, pp.106-118.
Schmitt, Rogerio (1999). Coligacoes eleitorais e sistema partidario no Brasil. Tese de doutorado.
Rio de Janeiro: IUPERJ.
Schattschneider, E. E. (1960). The Semisovereign people: a realist’s view of Democracy in Amer-
ica. New York: Holt, Rinehart and Winston.
167
Schumpeter, Joseph A. (1950). Capitalism, Socialism and Democracy. New York: Harper and
Row, 1942, 381 pp.; Third edition, 431 pp.
Seddon, Jessica; Alejandro Gaviria; Ugo Panizza; Ernesto Stein (2001). Political Particularism
Around the World
Selway, Joel (2011). The Measurement of Cross-cutting Cleavages and Other Multidimensional
Cleavage Structures. Political Analysis 19(1): 48-65.
Shugart, M., Melody Ellis Valdini,and Kati Suominen. (2005). “Looking for Locals: Voter In-
formation Demands and Personal Vote-Earning Attributes of Legislators under Proportional Rep-
resentation.” American Journal of Political Science 49 (2): 437–449.
Skidmore, Thomas E. (1967). Politics in Brazil 1930-1964: An Experiment in Democracy. Oxford
University Press.
Soares, Glaucio Ary (1964). “Aliancas e coligacoes eleitorais: notas para uma teoria”. Revista
Brasileira de Estudos Polıticos, n. 17, pp. 95-124.
Soares, Glaucio Ary (1973). ”Desigualdades eleitorais no Brasil”. In: Revista de Ciencia Polıtica,
v.7m n.1, p.25-48.
Sorauf, F. J. (1980). Party politics in America. Boston: Little, Brown.
Souza, Maria do Carmo Campello de (1976). Estado e partidos polıticos no Brasil: 1930-64.
Sao Paulo: Alfa-Omega.
Souza, Vivaldo de (2005). O comportamento das coligacoes eleitorais entre 1954-1962: pleitos
majoritarios coincidentes com proporcionais. In: Krause, Silvana and Schmitt, Rogerio (org.) Par-
tidos e coligacoes eleitorais no Brasil. Rio de Janeiro: Fundacao Konrad Adenauer; Sao Paulo:
Editora Unesp, pp.27-42.
Speck, Bruno and Campos, Mauro (2014). Incentivos para a fragmentacao e a nacionalizacao
do sistema partidario a partir do horario eleitoral gratuito no Brasil. Revista Teoria e Pesquisa
01, 2014; 23(2), pp.12-40.
Stokes, D. E. (1967). Parties and the nationalization of electoral forces. In The American party
systems: stages of political development, W. N. Chambers and W. D. Burnham (eds.). New York:
Oxford University Press.
168
Stoll, Heather (2008). Social Cleavages and the Number of Parties How the Measures You Choose
Affect the Answers You Get. Comparative Political Studies. Volume 41 Number 11. November.
Sundquist, J. L. (1973). Dynamics of the party system: alignment and realignment of politi-
cal parties in the United States. Washington, D.C.: The Brookings Institution.
Taagepera,, Rein (1993). Predicting the number of parties: a quantitative model of Duverger’s
mechanical effect. American Political Science Review, v. 87, n. 2, p. 455-64, June.
Taagepera, Rein (1994). ”Beating the Law of Minority Attrition,” in W. Rule and J.F. Zim-
mermann (eds), Electoral Systems in Comparative Perspective: Their Impact on Women and.
Minorities. Westport, CT: Greenwood Press.
Taagepera, Rein and Shugart, M. S. (1989). Seats and votes: the effects and determinants of
electoral systems. New Haven: Yale University Press.
Taagepera, Rein (1997). Effective Number of Parties for Incomplete Data. Electoral Studies,
Vol. 16, No. 2.
Taagepera, Rein (1999). The number of parties as a function of heterogeneity and electoral system.
Comparative Political Studies, v. 32, n. 5, p. 531-48, August.
Tavits, Margit (2010). ”Effect of Local Ties on Electoral Success and Parliamentary Behaviour:
The Case of Estonia”, Party Politics, 16: 215
Taylor, P. and R. Johnston (1979). Geography of Elections. New York: Holmes and Meier.
Van Meter, E; A. Lawson, N. Colabianchi, M. Nichols, J. Hibber, D. Porter, A. Liese. (2011).
Spatial accessibility and availability measures and statistical properties in the food environment.
Spatial and spatio-temporal epidemiology, V.2, N.1.
Vasselai, Fabricio; Mignozzetti, Umberto G. (2014). O Efeito das Emendas ao Orcamento no
Comportamento Parlamentar e a Dimensao Temporal: Velhas Teses, Novos Testes, In: Dados -
Revista de Ciencias Sociais, vol. 57, num. 3, pp. 817-853.
Wahman, Michael (2015). Nationalized Incumbents and Regional Challengers: Opposition- and
Incumbent-Party Nationalization in Africa. GIGA Working Papers, Hamburg, Germany.
Wildavsky, A.(1959). A methodological critique of Duverger’s political parties. Journal of Politics,
v. 21, p. 303-18.