Determinants of invalid voting in Latin America

Abstract

Although elections in Latin America are often characterized by very high levels of blank and spoiled ballots, there has been little if any systematic research into the phenomenon. Previous research from Europe, the U.S., and Australia has usually seen invalid voting as deriving from socio-demographic factors (literacy, education, wealth), institutional factors (electoral system and ballot structure), and political factors (alienation and protest). We operationalize these models for Latin America, using a cross-sectional time series data set including 80 legislative elections held in 18 democracies between 1980 and 2000. Socioeconomic variables such as urbanization and income inequality are associated with levels of invalid voting, while institutional variables such as compulsory voting, electoral disproportionality, and the combination of high district magnitude and a personalized voting system tend to increase blank and spoiled ballots. Moreover, regime-level factors such as political violence and the level and direction of democratic change also shape the rates of invalid voting.

Full text

Determinants of invalid voting in Latin America
Timothy J. Power
a,
, James C. Garand
b,1
a
Centre for Brazilian Studies, University of Oxford, 92 Woodstock Road, Oxford OX2 7ND, UK
b
Department of Political Science, Louisiana State University, Baton Rouge, LA 70803-5433, USA
Abstract
Although elections in Latin America are often characterized by very high levels of blank and spoiled ballots, there has been little
if any systematic research into the phenomenon. Previous research from Europe, the U.S., and Australia has usually seen invalid
voting as deriving from socio-demographic factors (literacy, education, wealth), institutional factors (electoral system and ballot
structure), and political factors (alienation and protest). We operationalize these models for Latin America, using a cross-sectional
time series data set including 80 legislative elections held in 18 democracies between 1980 and 2000. Socioeconomic variables such
as urbanization and income inequality are associated with levels of invalid voting, while institutional variables such as compulsory
voting, electoral disproportionality, and the combination of high district magnitude and a personalized voting system tend to in-
crease blank and spoiled ballots. Moreover, regime-level factors such as political violence and the level and direction of democratic
change also shape the rates of invalid voting.
Ó2006 Elsevier Ltd. All rights reserved.
Keywords: Invalid voting; Informal voting; Blank ballots; Spoiled ballots; Latin America
1. Introduction
Although voting is one of the most frequently studied
aspects of politics, scholars still have an uneven under-
standing of electoral participation. Numerous and com-
plex models have been advanced regarding the decision
to participatedi.e., voter turnoutdwhile even more
attention has been devoted to the dynamics of voter
choicedi.e., which parties and candidates citizens will
choose when they enter the voting booth. Curiously,
however, very little attention has been given to a third
electoral phenomenon, which occurs when voters turn
out to the polls but do not record a vote for a party or can-
didate. Invalid votingdthe casting of blank or spoiled
ballotsdremains a gaping lacuna in the literature on
comparative political behavior. Why is this the case?
One reason for inattention to invalid voting may
be that research on electoral behavior has dispropor-
tionately focused on countries where this practice is
negligible. One of the few reliable comparative gener-
alizations that has been made about the phenomenon
concerns ‘‘the well-known relationship between man-
datory voting and invalid ballots’’ (Hirczy, 1994: 65).
Since most research on voting has focused on volun-
tary systems such as the U.S. or U.K., and also on
Western Europe where compulsory voting remains
rare, the established literature contains few references
Corresponding author. Tel.: þ44 1865 284460.
E-mail addresses: timothy.power@brazil.ox.ac.uk (T.J. Power),
pogara@lsu.edu (J.C. Garand).
1
Tel.: þ1 225 578 2548.
0261-3794/$ - see front matter Ó2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.electstud.2006.11.001
Electoral Studies 26 (2007) 432e444
www.elsevier.com/locate/electstud

to invalid voting.
2
A second reason may be that politi-
cal scientists prefer to study voting behaviors that shed
light on party identification and/or the allocation of po-
litical power. Data on valid voting are clearly more use-
ful for these purposes, although invalid balloting may
potentially offer clues about political alienation or
may even affect electoral outcomes under certain rules.
A third reason for the reluctance to study invalid voting
is that from a macro-political and comparative perspec-
tive, blank and spoiled voting is notoriously difficult to
interpret. Do high rates of invalid voting portend citizen
dissatisfaction with the political system? Do blank and
null votes signify deficiencies of the electoral system
or ballot structure? What about voter error, perhaps
due to socioeconomic characteristics of the electorate?
These persistent questions are especially familiar to
readers living in mandatory voting systems. After each
election in which substantial invalid balloting occurs,
journalists, politicians, and scholars attempt to advance
their preferred interpretations. From a methodological
perspective, the only way to sort out these competing
explanations of invalid voting is to test them empirically
via a multi-country research design in which there is
considerable variation on the dependent variable as
well as significant variance in political, institutional,
and socio-demographic contexts. But before rising to
this challenge, we note that we view this exercise as
much more than a statistical puzzle. The casting of
blank and spoiled ballots has serious implications for
citizen support of the electoral process and, by exten-
sion, democratic legitimacy writ large. Presumably, ar-
chitects of nascent democracies will prefer that the act
of completing a ballotda right that was denied to mil-
lions of Latin American citizens in the 1960s and
1970sdshould signify popular preferences for candi-
dates and parties rather than alienation, withdrawal, or
voter error. Only by examining rival explanations of in-
valid voting, i.e., by supporting some interpretations
and possibly ruling out others, can democratic practi-
tioners begin to conceive of reforms that might diminish
the casting of blank and spoiled ballots.
In order to test competing interpretations, we under-
take a pooled cross-sectional time-series analysis of in-
valid voting in Latin America, examining 80 legislative
elections held in 18 different countries between 1980
and 2000. Latin America is an ideal setting for a compar-
ative study of invalid voting. In Latin America we obtain
remarkable variation on the dependent variable: while
the average election during this time period saw about
11% blank and spoiled ballots, some countries have rates
of invalid voting in the range of 2e3%, while others have
averages that are ten times higher. Moreover, although
compulsory voting is the norm in Latin America, several
countries do not practice it and others do not enforce it to
a significant degree. We are fortunate to have a cross-
national measure of enforcement of mandatory voting,
which frees us from the usual practice of modeling com-
pulsory voting as a binary variable and affords us valu-
able variation within the one political institution that is
unambiguously associated with invalid votes.
3
Finally,
we also benefit from considerable variance on other
institutional variables across the region, as well as
from broad variation in regime-level and SES variables.
Following the excellent Australian study by
McAllister and Makkai (1993), we ask whether ‘‘insti-
tutions, society, or protest’’dor some combination of
the three interpretationsdbest explains cross-national
variance in invalid voting. Our cross-sectional time-
series method does not allow us to characterize any
one political system or explain any single election;
that is, we cannot explain, nor do we want to explain,
why blank and spoiled ballots in country Xspiked to
level Nin year Y. Nor do we intend to contribute to
the debate on the normative desirability of compulsory
voting (e.g., Lijphart, 1997; Hill, 2002). Rather, we aim
to advance some probabilistic generalizations about
whether ‘‘institutions, society, or protest’’ best explain
cross-national variation in invalid voting in Latin Amer-
ica, the world region where political democracy and
compulsory voting are most closely intertwined.
2. Invalid voting: theoretical approaches
As noted above, invalid voting has received very
little attention in the literature. The first comprehensive
review of contending theoretical perspectives was
2
Belgium retains compulsory voting, but the Netherlands and Italy
have abolished it. Regarding the English-language literature on the
topic, it is not surprising that Australia, as the only English-speaking
country with mandatory voting, has generated a number of the more
visible contributions on invalid voting (McAllister and Makkai,
1993; Hill, 2002). Australians use the term ‘‘informal voting’’ to de-
scribe blank or spoiled ballots.
3
The cross-national evidence is quite strong that compulsory voting
‘‘works’’ (Blais, 2006: 113). The careful studies by Franklin (1996,
1999, 2004) show that the presence of mandatory voting boosts turn-
out on average by 6e7%. In the broader sample examined by Blais
and Dobrzynska (1998), the effect is closer to 11%. All of these stud-
ies model compulsory voting as a dichotomous variable. Preliminary
research by International IDEA suggests that within the family of
mandatory voting systems, the level of enforcement matters consider-
ably (www.idea.int/vt). Our polychotomous measure of enforcement
in this paper reflects this assumption (see Appendix A).
433T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

provided by McAllister and Makkai (1993) in their case
study of Australia and was later amplified by Power and
Roberts (1995) in their case study of Brazil. These stud-
ies develop three broad theoretical approaches to the
study of invalid voting: the institutional approach, the
socioeconomic approach, and the political-protest or re-
gime-level approach. Each of these approaches draws
attention to a rival cluster of variables that are assumed
to explain variation in the casting of blank and spoiled
ballots. Many of these variables also figure prominently
in the related (but vastly larger) literature on voter turn-
out. This is not surprising given that models of turnout
and of invalid voting are both concerned with the factors
that influence a critical intervening variable in virtually
all of these models: voter efficacy.
The institutional approach claims that invalid voting
is unrelated to the political or socioeconomic character-
istics of the polity, but is rather a function of institu-
tional design. Building on influential institutionalist
accounts of electoral participation (Powell, 1986; Jack-
man, 1987; Jackman and Miller, 1995; Pe
´rez-Lin
˜a
´n,
2001), this approach draws attention to the legal and
formal structures that make voting desirable versus un-
desirable, or simple versus difficult. For example, where
mandatory voting is strictly enforced, citizens who
would normally abstain under voluntary voting condi-
tions may turn out to the polls only to cast a blank or
spoiled ballot. Electoral disproportionality, which tends
to punish small parties and/or minority viewpoints, may
have the same effect. Bicameralism may also reduce the
perceived efficacy of voting (Lijphart, 1984), as would
an excessively high number of political parties (Jack-
man, 1987; Kostadinova, 2003). Blais and Dobrzynska
(1998: 250) find strong evidence for Jackman’s hypoth-
esis about multipartyism, and claim that where the ef-
fective number of parties is lower, ‘‘people are more
likely to feel that they are electing the party that will
form the government.’’
Voter efficacy may also be depressed in systems
where electoral rules combine high district magnitude
with a personalized voting system, thus placing a heavy
burden on voters to learn about dozens if not hundreds
of candidates; this was a key finding of Power and Rob-
erts’ (1995) study of invalid voting under open-list pro-
portional representation (OLPR) in Brazil. The
Brazilian Senate uses the simple single-member district
plurality method (SMDP) and has fewer invalid ballots,
whereas the Chamber of Deputies uses the less user-
friendly OLPR and sees more invalid voting. In sum,
certain electoral rules, ballot structures, and party sys-
tem configurations may favor the casting of blank and
null votes.
The socioeconomic approach to invalid voting claims
that blank and spoiled ballots are a product of the social
structure. Socioeconomic development and urbaniza-
tion favor wider circulation of and more democratic ac-
cess to the political information necessary to complete
a ballot in national elections. Education and literacy con-
tribute to the political skill levels of individual voters. In
their Australian study, McAllister and Makkai (1993)
found that invalid voting occurred most frequently in
precincts with large numbers of immigrants with low
levels of English proficiency. In Latin America, large
numbers of voters remain illiterate or semi-literate, mak-
ing it difficult for them to complete a ballot paper; this is
the functional equivalent of the weak English skills
among foreign-born Australians. It is worth noting that
in both Australia and Latin America, citizens with
weak literacy skills are forced to the polls by compulsory
voting legislation, presumably inflating the numbers of
blank and spoiled ballots. Thus the socioeconomic ap-
proach to invalid voting gains greater plausibility in
mandatory voting settings, just as the socioeconomic ap-
proach to voter turnout has more plausibility in voluntary
voting settings (Powell, 1986).
Finally, the political-protest approach claims that in-
valid voting is an indicator of voter discontent. This dis-
content may reflect poor economic conditions, rejection
of incumbents, condemnation of the existing political
regime (i.e., an anti-system orientation), or some com-
bination of all of these factors. Power and Roberts
(1995) find that invalid voting in Brazil rose when the
1964e1985 military regime was most discredited, and
especially so when the generals tried blatantly to manip-
ulate electoral rules to their own advantage. Transitions
to democracy may coincide with a reduction in invalid
voting, as citizens embrace new electoral procedures
and subjective competence rises. Conversely, a per-
ceived authoritarian retrogressiondi.e., an erosion of
political rights or civil libertiesdmay cause an upturn
in blank and spoiled ballots as a form of protest. If pro-
test is really what is driving blank and spoiled ballots,
then invalid voting should logically be found alongside
other manifestations of anti-system sentiment, e.g., rev-
olutionary activity or political violence.
These three rival approaches to invalid voting are rel-
atively straightforward. However, rather than juxtapose
them as mutually exclusive explanations, we suspect
that invalid voting may in fact be the result of a number
of diverse variables drawn from each of the three frame-
works. Below, we merge key variables from these three
approaches into a multivariate model that assesses
the effects of a large number of explanatory factors
simultaneously.
434 T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

3. Data and methods
Given the foregoing discussion, we suggest that in-
valid voting is a function of a variety of institutional, so-
cioeconomic, and protest-democracy variables. From
the institutional approach, we operationalize personal
voting, district magnitude, electoral disproportionality,
cameral structure, and mandatory voting; from the so-
cioeconomic approach, we draw urbanization, literacy,
per capita income, and social inequality; and from the
political protest model, we employ economic perfor-
mance, anti-system violence, founding election status,
levels of democratization, and overall direction of re-
gime change. Our model can be summarized in the fol-
lowing equation:
If our characterization of the three rival models is correct,
we hypothesize that b
1
,b
2
,b
3
,b
4
,b
6
, and b
12
will be pos-
itive, while the remaining coefficients will be negative.
4
3.1. Variable measurement
In order to estimate the parameters of our model, we
collect data on the variables in the model for 18 Latin
American countries in each year from 1980 to 2000.
A brief summary of these variables can be found in
Appendix A.
As a starting point, we note that the dependent vari-
able in our study is measured as blank and spoiled ballots
as a percentage of all ballots cast in national legislative
elections for each country and each election year. In
Fig. 1 we present the histogram and the kernel density
plot for the distribution of invalid ballots for the 80 coun-
try-year cases for which elections were held. Moreover,
in Table 1 we report the mean, standard deviation, min-
imum, and maximum on this variable for each of the 18
countries in our data set. As one can readily see, the over-
all distribution is skewed heavily to the right. Most
countries have percentages of invalid ballots below
10%; in fact, a majority of country-year cases fall below
this 10% figure. Only 15% of observations have invalid
ballot percentages of 20% or greater, though there are
a few cases in which the percentage of invalid ballots ex-
ceeds 40%. Turning to Table 1, we observe considerable
variation across countries in the percentage of invalid
ballots. Overall, the average percentage of spoiled bal-
lots across these country-election years is 10.58, with
a standard deviation of 10.14. At the low end, countries
such as the Dominican Republic (1.70%), Mexico
(2.94%), Paraguay (3.10%), Chile (3.29%), Uruguay
(3.32%), Honduras (4.66%), and Argentina (4.94%)
have average invalid ballot rates of less than 5% of votes
cast. At the other extreme are Brazil (33.29%), Peru
(24.30%), Ecuador (22.51%), Venezuela (14.58%),
Guatemala (11.72%), Colombia (10.52%), and Bolivia
(10.17%). Clearly, in some countries invalid ballots are
a commonplace occurrence, while in other countries in-
valid ballots are relatively rare.
3.1.1. Institutional variables
The first institutional variable in our model is per-
sonal vote, measured using Carey and Shugart’s
0
.02
.04
.06
.08
Density
0 10 20 30 40 50
% invalid ballots cast
Fig. 1. Distribution of invalid ballots cast in legislative elections in
18 Latin American countries, 1980e2000. The solid line represents
the kernel density plot for the distribution of invalid ballots cast.
Invalid Voting ¼aþb1ðPersonal voteÞþb2ðDistrict magnitudeÞ
þb3ðPersonal vote District magnitudeÞþb4ðElectoral disproportionalityÞ
þb5ðUnicameralismÞþb6ðCompulsory votingÞþb7ðUrbanizationÞ
þb8ðLiteracyÞþb9ðPer capita GDPÞþb10ðIncome inequalityÞ
þb11ðChange in per capita GDPÞþb12ðRevolutionary violenceÞ
þb13ðFounding electionÞþb14 ðLevel of democracyÞþb15 ðChange in level of democracyÞ
4
We notethat electoral disproportionality, urbanization, literacy,per
capita GDP, and income inequality are lagged by one year to control
for the possible delayed effects of these variables on invalid ballots.
435T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

(1995) ranking of electoral systems based on their ‘‘in-
centives to cultivate a personal vote.’’ The variable
ranges from 1 (i.e., the electoral system favors party-
based voting) to 9 (i.e., the electoral system favors an
emphasis on individual candidates and the personaliza-
tion of campaigns). For instance, the closed, blocked
party list in Argentina forces voters to cast their ballots
strictly in terms of partisan choices and is hence coded
1. This system affords voters little direct influence (they
cannot alter the nominations and rankings proposed by
parties) yet is an extremely simple balloting system,
since voters choose only among parties and benefit
from a considerable mental shortcut in the act of voting.
On the other hand, the open-list system in Brazildin
which voters, not parties, determine the list position
of candidatesdis coded an 8. The Brazilian system al-
lows voters great latitude in casting their votes for indi-
vidual politicians, yet places a heavy cognitive burden
on the voter to choose one person from among hundreds
of candidates (Power and Roberts, 1995). As noted, we
expect that the coefficient for this variable will be pos-
itive, indicating that invalid ballots should be higher in
those systems characterized by personalized electoral
systems rather than party-based systems. Simply put,
it is more difficult to vote in a personalized, candi-
date-based system than in a party-based system.
Second, district magnitude is the average number of
seats allocated in electoral districts used in lower house
elections. (For mixed electoral systems such as Mexico
or Bolivia, we downweight this value by the percentage
of legislative seats elected in single-member districts.)
We suggest that high district magnitude should result
in higher levels of invalid ballots, since the result of
high district magnitude is likely to be more political
parties, more candidates, and the resulting increase in
the difficulty of making a choice on election day.
Here our expectations resemble those of Jackman
(1987), Blais and Dobrzynska (1998), and Kostadinova
(2003) in their discussions of the depressing effect of
multipartism on voter turnout.
Third, we also consider the possibility that the per-
sonal vote and district magnitude have joint effects
that promote invalid ballots. A polity characterized by
the complexity inherent in both a highly personalized
electoral system and high district magnitude should ex-
hibit exponentially higher levels of invalid ballots. The
interaction between these two variables is essentially
a measure of the complexity of the political market.
The higher the value of this interaction variable, the
greater the ‘‘mental tax’’ imposed on voters and the
more likely they are to spoil their ballots or to just give
up and cast blank ballots. To capture these effects, we
create an interaction term by multiplying the personal
vote and district magnitude variables; we hypothesize
that the coefficient for this variable will be positive.
Fourth, electoral disproportionality is the distortion
of representation caused by translation of votes into
seats. We follow Lijphart (1984) in measuring this vari-
able as the average vote-seat share deviation of the two
largest parties in each election. Electoral disproportion-
ality punishes small parties, whose adherents may re-
spond by casting a blank or null vote. We speculate
that systems with substantial vote-seat distortion will
exhibit higher levels of political alienation, lower levels
of efficacy, and hence higher levels of invalid ballots.
Fifth, unicameralism is the degree to which the na-
tional legislature approaches the unicameral model.
Following the work of Lijphart, we code this variable
on a five-point scale, ranging from 0 (strong bicameral-
ism) to 4 (unicameral legislature). We hypothesize that
unicameralism will be negatively related to invalid
ballots. Voters in unicameral systems should feel more
subjectively competent, since they know that the single
legislative chamber will be the central locus of the po-
litical system and cannot be blocked by another cham-
ber. Specifically, we contend that unicameralism creates
clearer lines of authority and accountability, and this
simplifies that task of voting and makes casting a com-
plete ballot easier.
Sixth, we include a measure of mandatory voting in
our model. Most Latin American countries have some
form of compulsory voting, but there is also some
Table 1
Descriptive statistics on invalid ballots, by country, 1980e2000
Country Mean Standard
deviation
Minimum Maximum No. of
elections
Argentina 4.94 2.56 2.54 9.31 9
Bolivia 10.17 3.19 4.83 12.98 5
Brazil 33.29 11.13 20.03 43.71 4
Chile 3.29 0.45 2.53 3.61 5
Colombia 10.52 6.53 5.05 17.75 3
Costa Rica 5.61 7.52 0.19 16.34 5
Dominican
Republic
1.70 1.22 0.59 3.00 3
Ecuador 22.51 5.65 15.70 30.40 7
El Salvador 8.54 4.11 3.66 14.12 6
Guatemala 11.72 0.85 10.92 12.69 5
Honduras 4.66 2.89 2.53 8.92 4
Mexico 2.94 0.68 2.32 3.66 3
Nicaragua 6.81 0.96 6.13 7.49 2
Panama 5.61 0.40 5.33 5.89 2
Paraguay 3.10 1.11 1.90 4.08 3
Peru 24.30 12.92 11.77 45.10 5
Uruguay 3.32 1.35 2.05 4.76 4
Venezuela 14.58 11.82 3.58 31.89 5
Allcountries 10.58 10.14 0.19 45.10 80
436 T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

moderate variation in how strongly these countries ad-
here to the principle of mandated voting. We measure
this variable usinga four-point scale, ranging from 0 (vol-
untary votingdonly Colombia and Nicaragua through-
out the time series, joined by Venezuela in 1999) to 3
(compulsory voting with legal sanctions that are enforced
in practice). Following Hirczy (1994), we speculate that
compulsory voting is positively related to invalid ballots.
When citizens are required to vote by law, discontented
voters who would normally register their discontent by
not turning out on election day may spoil their ballot as
a means of registering their protest, and voters with low
levels of literacy or political information may spoil their
ballot unintentionally or simply leave it blank.
3.1.2. Socioeconomic variables
We also include in our model a series of socioeco-
nomic variables that represent broad characteristics of
the electorate and serve as proxies for the relative capac-
ity of voters to complete their ballots accurately. First,
we suggest that urbanization is inversely related to in-
valid ballots. In urban areas information about upcoming
elections is more easily organized, distributed, and re-
ceived, and this should result in lower levels of invalid
ballots. We measure this variable as the percentage of
the total population living in urban areas. Second, liter-
acy is a direct measure of voter capacity. We speculate
that countries with high levels of literacy should exhibit
lower levels of invalid ballots, since the electorates in
these countries are more likely to possess the political
skills necessary to cast their votes properly. Our literacy
variable is measured as the percentage of the population
aged 15 years and older that can read and write. Finally,
we include a logged measure of per capita GDP, ex-
pressed in constant 1990 dollars. Here again, per capita
GDP should depress invalid ballots, since per capita
GDP is a proxy for the many dimensions of socioeco-
nomic development (e.g., industrialization, unioniza-
tion, feminization of the labor force, expansion of
tertiary education) that expand and enhance voter capac-
ity. Moreover, higher per capita GDP should, on average,
increase personal economic satisfaction, making voters
more content with the political system and less likely
to cast an invalid vote as a political protest measure.
We note that there is enormous variation on this variable
throughout Latin America: in 2000 per capita income in
Nicaragua stood at $532, whereas in Venezuela it was
$3036 and in Argentina it was $6254.
Latin America is also characterized by high levels of
income inequality. In contrast to our argument about pov-
erty, we suggest that income inequality will have a nega-
tive effect on invalid voting. Our reasoning is that in
highly unequal societies, overall voter turnout tends to
be lower. That means that the effective electorate will
be composed disproportionately of high-SES voters
who should be less likely to cast blank or spoiled ballots.
Mandatory voting laws should mitigate this effect to an
important degree, yet we hypothesize that social inequal-
ity will still have a marginal impact on the characteristics
of the effective electorate that turns out at the polls.
3.1.3. Protest-democracy variables
We also incorporate a cluster of variables that mea-
sure characteristics of the political regime such as per-
formance, pluralism, and legitimacy. We include
change in per capita GDP as a measure of economic per-
formance. Our hypothesis is that dissatisfied voters are
more likely to express their discontent with a blank or
spoiled ballot. To capture anti-system sentiment, we
employ an annualized measure of acts of revolutionary
violence drawn from the State Failure Task Force Data-
base. We expect that invalid voting should covary posi-
tively with political violence. Given that our 1980e
2000 time series coincides with the so-called Third
Wave of democratization in Latin America (Huntington,
1991), we also include a binary variable for the ‘‘found-
ing election’’ that inaugurates the democratic regime, in
the expectation that these elections should feature un-
usually high levels of salience and voter efficacy
(O’Donnell and Schmitter, 1986; Turner, 1993). The co-
efficient for the founding election variable should there-
fore be negative.
Our democracy measures are created by combining
Freedom House’s scales of civil and political rights, re-
coded so that higher values represent greater levels of
democracy. Elections held in more democratic coun-
try-years should generate higher levels of voter efficacy
and lower levels of invalid voting, net of all other vari-
ables. We also hypothesize that the direction of regime
change over time is equally important as the absolute
level of democracy. Voters in countries that are experi-
encing a deepening of democracy should be less likely
to cast invalid ballots, while voters in countries suffer-
ing an erosion of democracy should be more likely to
register their opposition to this retrogression by casting
an invalid ballot. Hence we include in our model a mea-
sure of year-on-year change in the Freedom House mea-
sure. We expect the coefficients for both Freedom
House variables to be negative.
3.2. Model estimation
Pooled cross-sectional time-series models often in-
volve violations of ordinary least squares (OLS)
437T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

assumptions of homoskedasticity and uncorrelated er-
ror terms (Gujarati, 1995; Kmenta, 1986; Greene,
1993). While OLS estimates are unbiased in the pres-
ence of autocorrelation, these estimates are not effi-
cient, and the variability of OLS coefficients
contaminates tests of statistical significance. To account
for this, we estimate our pooled cross-sectional time-se-
ries models using feasible generalized least squares
(FGLS). The model assumes a heteroskedastic error
structure across panels with no cross-sectional correla-
tion and is estimated using panel-specific estimates of
first-order autocorrelation.
4. Empirical results
In Table we report the FGLS estimates for our model
of invalid voting in Latin American countries from 1980
to 2000. The model fits the data quite well (R
5
¼0.567,
Wald c
6
¼303.24, prob <0.000), with approximately
57% of the variance in invalid voting explained by the in-
dependent variables in the model. Moreover, at least two
variables from each of our three clusters of explanatory
variables have an effect on invalid voting that is both in
the expected direction and that surpasses conventional
levels of statistical significance. In fact, an examination
of the standardized regression coefficients reveals that
there are variables in each cluster that have strong effects
on invalid ballots, relative to other variables in the model.
Hence it is difficult to assert that one cluster of variables
is more important in determining levels of invalid ballots
than each of the other two clusters. Clearly, our compre-
hensive model does a very good job in explaining pat-
terns of invalid voting across countries and time.
4.1. Institutional variables
Our findings provide evidence that patterns of
invalid voting are explained in part by variation in the
institutional contexts found across Latin American de-
mocracies. First, we hypothesize that invalid voting
will be lower in those country-year cases in which vot-
ing is made easier for citizens by electoral structures
that facilitate party-based voting (as opposed to candi-
date-centered elections) and single-member districts
(as opposed to multi-seat districts). When voters face
complex electoral choices because the electoral system
emphasizes individual candidates and/or has multiple
candidates per seat, the complexity of decision-making
is increased and may result in an increase in invalid vot-
ing. Moreover, when both of these characteristics exist
simultaneously, there may be reinforcing effects that
promote invalid ballots. Our results suggest that there
is a strong joint effect of the personal vote and district
magnitude on invalid ballots, but that the individual ef-
fects of these two variables are not as expected. The co-
efficient for the personal vote variable represents the
effect of personalized elections on invalid ballots
when district magnitude equals 0, but here we find
that there is no independent effect of the personal vote
(b¼0.090, t¼0.26). Likewise, the coefficient for
district magnitude represents the effect of this variable
when the personal vote variable equals 0. In this in-
stance the coefficient for district magnitude is unexpect-
edly negative and statistically significant (b¼0.477,
t¼4.30). Given these two coefficients, we would be
hard pressed to suggest that these aspects of the com-
plexity of the electoral system have the expected effect
on invalid ballots. However, when we consider the joint
effects of these two variables, we find that the coeffi-
cient for the interaction of personal vote and district
magnitude is in the expected positive direction and
highly significant (b¼0.157, t¼4.97). This suggests
that electoral complexity, when compounded by both
personalized voting and multiparty districts, results in
a much greater rate of blank and spoiled votes. On the
other hand, having one but not the other does not result
in the expected effect on invalid ballots.
Second, we find that electoral disproportionality has
a strong positive effect on invalid ballots in legislative
elections, as expected (b¼0.405, t¼3.15). We specu-
late that the distortion of representation resulting from
the translation of votes into seats should result in higher
rates of invalid ballots, and our empirical results suggest
that this is the case. As the share of seats deviates from
the share of votes, voters cast invalid ballots in larger
numbers, presumably since this distortion results in
higher levels of political alienation and lower levels of
political efficacy.
Third, the effect of compulsory voting on the produc-
tion of invalid ballots is quite strong. We hypothesize
that invalid ballots should be much more likely in coun-
try-year cases with compulsory voting accompanied by
strong enforcement. When voters (1) are mandated to
go to the polls, (2) face strong sanctions for the failure
to vote, and (3) find that these sanctions are strongly en-
forced, they increase their propensity to cast an invalid
ballot. This pattern may be due to an alienation effect,
whereby discontented voters forced to the polls pur-
posely cast an invalid ballot. Alternatively, mandatory
voting will bring voters to the polls who do not have
the education or literacy to successfully complete their
ballots; in the absence of compulsory voting legislation
many of these low-competency voters would not be ex-
pected to turn out to vote in the first place.
438 T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

The bottom line is that at least some institutional
structures matter in determining levels of invalid bal-
lots. To be sure, unicameralism and the personal vote
have coefficients that fail to achieve conventional levels
of statistical significance, and the coefficient for district
magnitude is in the unexpected negative direction.
However, three of our institutional variables have strong
effects on invalid ballots that are consistent with expec-
tations. Electoral complexity (as represented by the
joint effect of the personal vote and district magnitude),
electoral disproportionality, and compulsory voting all
shape the propensity of voters to cast invalid ballots.
4.2. Socioeconomic variables
Our findings in Table 2 also point to the importance
of socioeconomic variables in determining levels of in-
valid voting. First, we speculate that urbanization will
depress invalid ballots, since citizens in urban areas
are more likely to be exposed to the kinds of political
organization and easily-accessible information that
will reduce invalid ballots. Our empirical results pro-
vide reasonably strong support for this hypothesis
(b¼0.172, t¼2.27). For every 10 percentage point
increase in urbanization, the percentage of invalid bal-
lots declines by a bit less than 2 percentage points of
ballots cast, controlling for the effects of other vari-
ables. Given the range on the urbanization variable of
about 55 percentage points, this suggests that the pre-
dicted difference in invalid ballots for the most and least
urbanized country-year cases is a bit less than 10 per-
centage points (i.e., 55% 0.172), which is not a trivial
amount.
Second, voter competence is often cited as a major
predictor of invalid ballots, so we suggest that greater
literacy will depress the likelihood of voters to cast an
invalid ballot. Our results are not highly consistent
with this prediction. The coefficient for literacy is neg-
ative but is significant only at the relaxed.10 level of sta-
tistical significance (b¼0.217, t¼1.47); for every
10 percentage point increase in literacy the percentage
of invalid ballots declines by about 2 points. In Latin
America it appears that countries with higher literacy
rates have lower rates of invalid ballots than other coun-
tries, though the relationship is not particularly strong.
Third, we consider the effects of logged per capita
GDP on invalid ballots. Income represents a wide range
of characteristics of the electorate that would be related
to voter competency, and hence we hypothesize that
voters in relatively wealthier country-year cases will ex-
hibit lower levels of invalid ballots. Our results in Table
2are not consistent with this hypothesis. The coefficient
for per capita GDP is not in the expected negative direc-
tion, and it fails to achieve statistical significance at
conventional levels (b¼1.234, t¼0.61).
Finally, the coefficient for income inequality is both
in the expected negative direction and statistically sig-
nificant (b¼0.527, t¼4.40). As noted, we contend
that income inequality will be negatively related to in-
valid voting, since in countries with high income in-
equality the electorate will be more compact and will
likely have a higher proportion of the high-SES, high
competency voters who are less prone to casting invalid
Table 2
FGLS regression results for model of invalid ballots in Latin Ameri-
can legislative elections, 1980e2000, with logged per capita income
and change in logged per capita income
Variable bt b
Institutional variables
Personal vote [þ]0.090 0.26 0.024
District magnitude [þ]0.477 4.30 1.101
Personal vote district
magnitude [þ]
0.157 4.97 1.383
Electoral
disproportionality [þ]
0.405 3.15 0.162
Unicameralism []0.218 0.51 0.039
Compulsory voting [þ] 2.843 3.60 0.250
Socioeconomic variables
Urbanization []0.172 2.27 0.250
Literacy []0.217 1.470.283
Logged per capita GDP [] 1.234 0.61 0.080
Income inequality []0.527 4.40 0.299
Protest-democracy variables
Change in logged per
capita GDP []
53.571 0.72 0.026
Revolutionary violence [þ] 1.209 1.89 0.094
Founding election []2.432 1.22 0.073
Freedom House []0.770 2.78 0.151
Change in Freedom
House []
3.020 6.67 0.239
Intercept 53.983 4.97 0
N75
Pseudo-R
2
0.567
Wald c
2
303.24
Prob (Wald c
2
) 0.0000
Log likelihood 200.32
Note: This model is estimated using feasible generalized least squares.
The model assumes a heteroskedastic error structure across panels
with no cross-sectional correlation and is estimated using panel spe-
cific estimates of the first-order autoregressive process. In addition,
the standardized regression coefficient (b) is calculated by multiplying
the unstandardized coefficient (b) by the ratio of the standard devia-
tion of each independent variable to the standard deviation of the de-
pendent variable (i.e., s
X
/s
Y
).
prob <0.01; prob <0.05; prob <0.10.
439T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

ballots. Our empirical results are very consistent with
this assertion.
Again, it is clear that socioeconomic variables have
important effects on the propensity of voters to cast in-
valid ballots. Consistent with expectations, invalid bal-
lots are more numerous as a share of all ballots cast in
countries with low urbanization, low levels of income
inequality, and possibly low literacy. Clearly, while in-
stitutional variables structure patterns of invalid ballots
across countries and time, the aggregate competence of
the electorate has a strong effect as well.
4.3. Protest-democracy variables
One other possible explanation for variation in in-
valid ballots in Latin America is the performance and
legitimacy of political regimes. When voters view the
political system as illegitimate or otherwise disapprove
of the performance of the government or regime, on
possible response is for them to spoil their ballot as
a sign of political protest. This has nothing to do with
competence, but rather involves the purposive decision
by voters to register their discontent by casting a spoiled
ballot. We include several variables to capture the ef-
fects of anti-system sentiment on invalid ballots.
First, we find that political violence has a positive ef-
fect on invalid ballots (b¼1.209, t¼1.89). For every
additional act of political violence, there is, on average,
a bit more than a 1% increase in invalid ballots, control-
ling for the effects of other independent variables in the
model. Revolutionary violence is not extremely com-
monplace; the maximum number of acts of revolution-
ary violence in the data set is only 4 acts in a given
country-year. Our findings suggest that such acts do
have a discernible impact on invalid ballots, implying
that violence represents some level of discontent with
the political system that is translated into more invalid
ballots when voters go to the polls.
Perhaps our most interesting findings relate to the
level of and change in Freedom House democracy
scores. As noted, we have recoded the Freedom House
scoring so that high values represent higher levels of de-
mocracy and freedom, and so an increase (decrease) in
Freedom House scores indicates that a given country is
moving in a more (less) democratic direction. We sug-
gest that countries with high levels of democracy will
have lower levels of invalid ballots, since voters should
be more favorably oriented toward the political system;
conversely, less democratic systems should engender
less support, and this should results in higher levels of
invalid ballots as voters express their discontent at the
ballot box by spoiling their ballots. But we are
interested not only in the level of democracy but also
the direction of regime change. Using year-on-year
change in the Freedom House score, we suggest that
as a country deepens its democracy, voters will decrease
their propensity to cast invalid ballots as their sense of
efficacy and support for the regime increases. Con-
versely, countries that undergo a democratic erosion
from one year to the next should generate increased
levels of protest and discontent and, subsequently,
higher rates of invalid ballots.
Our empirical results are strongly consistent with
these expectations. The coefficient for our Freedom
House democracy measure is negative and significant
(b¼0.770, t¼2.78), suggesting that the percent-
age of invalid ballots cast is systematically lower in
highly democratic countries than in countries with
less robust democracy. In our Latin American sample,
the recoded Freedom House variable ranges from
0 (low democracy) to 10 (high democracy); given
this, the predicted differences in the percentage of in-
valid ballots between the most democratic and least
democratic election context is almost 8 points (i.e.,
10 0.77). Moreover, we find that the variable for
changes in levels of democracy has a strong negative ef-
fect on invalid ballots (b¼3.020, t¼6.67). As
countries strengthen their commitment to civil and po-
litical liberties from one year to the next, their voters be-
come less likely to cast invalid ballots.
Finally, we observe that the effect of founding
elections on invalid ballots does not achieve standard
levels of statistical significance (b¼2.432,
t¼1.22). Voters should feel extraordinarily effica-
cious in founding elections (i.e., the first free contests
since the end of the previous authoritarian regime)
and hence we expect the percentage of invalid ballots
to be lower in founding elections than in other elec-
tions, all else being equal. Our findings provide little
support for this expectation.
Clearly, the protest-democracy variables join the in-
stitutional and socioeconomic variables in having
a strong effect on invalid ballots. Variables associated
with dissatisfaction (or satisfaction) with the regime
and political systemdi.e., political violence, democ-
racy, and changes in democracydhave systematic ef-
fects on invalid voting in legislative elections.
5. Conclusion
In this paper we develop and test a model of in-
valid ballots in 18 Latin American countries from
1980 to 2000. The model considers the effects of
three clusters of independent variablesdinstitutional,
440 T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

socioeconomic, and protest-democracy. We suggest
that the propensity of voters to cast invalid ballots
is affected by (1) institutional attributes of their polit-
ical systems that promote or depress invalid voting;
(2) the socioeconomic contexts in which they live,
which shape the competence of voters to navigate
the electoral process successfully; and (3) protest-de-
mocracy variables, which reflect the degree to which
voters are discontented with their political system and
hence are likely to register their frustration by spoil-
ing their ballots.
Taken as a whole, our model is very successful in
predicting levels of invalid ballots. Not only is the
overall goodness of fit for the model quite strong,
but we find support for hypotheses relating to each
cluster of independent variables. Why do voters cast
invalid ballots? Simply, they do so for most of the
reasons that have been anticipated in the disparate
streams of literature that we unify here. In fact, we
are struck by the level of support for all three of
the theoretical perspectives suggested by the institu-
tional, socioeconomic, and protest-democracy inter-
pretations of invalid voting. Among the institutional
variables, we find that the complexity of voting
brought on by the combination of personal voting
and high district magnitude generates significantly
higher levels of invalid ballots. Voters also respond
to the disconnect between votes cast and seats earned
by casting more invalid ballots, and invalid ballots are
also higher when voters are required to vote by strict
and enforced compulsory voting laws. The bottom
line is that institutions matter.
Regarding socioeconomic variables, we speculate
that electorates vary in their abilities to successfully
negotiate the sometimes complex electoral process,
and furthermore that these abilities vary with the so-
cioeconomic context. When political systems are
highly urbanized (i.e., where political organizations
are stronger and political information is easier to ob-
tain), when they possess higher aggregate levels of lit-
eracy, and when they feature high levels of income
inequality (and hence more compact electorates with
a relatively higher socioeconomic status), there is
a lower propensity of electorates to cast invalid
ballots.
Finally, political discontent matters for invalid bal-
lots. When voters confront high levels of revolutionary
violence (which, presumably, represents some level of
discontent), when they live in less democratic countries,
or when their political systems are undergoing change
in an anti-democratic direction, invalid voting occurs
with greater frequency.
Our research on invalid voting is important for both
scholars and practitioners of democracy in Latin Amer-
ica. The lesson for scholars is that there is no one-size-
fits-all interpretation of invalid voting, and that it is sim-
ply wrong to assert any one perspective (institutional, so-
cioeconomic, or political-protest) to the exclusion of the
others. Each theoretical perspective provided us with var-
iables that proved to be significant predictors of blank and
spoiled ballots (even in the presence of variables drawn
from rival perspectives) suggesting that invalid voting is
a complex phenomenon deserving of an eclectic, multidi-
mensional approach. The lesson for political practitio-
nersdand here we return to our initial speculation
about possible reforms of the electoral processdis that
it is indeed possible to influence the level of invalid voting
by intervening in the socioeconomic, regime-level, and
institutional contexts of elections. Surveying these con-
texts, we note that there is today a broad consensus in
Latin America about promoting human development
and deepening political democracy, but change in these
areas moves at a notoriously glacial pace. Therefore, it
is clearly the institutional context that is the most imme-
diately tractable for political elites. The evidence from
Latin America between 1980 and 2000 suggests rather
strongly that (1) reducing the complexity of the political
market, (2) improving the proportionality of the electoral
system, and (3) moving toward voluntary voting would
all work to reduce the level of invalid voting.
Of these three possible reforms, the first two are not
implausible in many countries. However, the easing of
compulsory voting would almost certainly drive down
aggregate voter turnout (Pe
´rez-Lin
˜a
´n, 2001; Fornos
et al., 2004; Blais, 2006), which is a cost that elites
may not be willing to pay. Declining turnout would
bring the debate back full circle to the broader norma-
tive concerns raised by invalid voting in the first place.
In the eyes of democratic reformers, one worrisome
trend might simply be replaced by another. The tradeoff
between electoral participation and invalid votingdtwo
phenomena frequently cited as indicators of democratic
health and legitimacydis likely to remain a vexing di-
lemma at the heart of Latin American democracy for
years to come.
Acknowledgments
The authors thank Carolina Fornos and Daniel J.
Nielson for provision of data. They are also grateful
to Jairo Nicolau, Geoffrey Evans, and two anonymous
reviewers for helpful comments on an earlier draft of
this article. The usual disclaimer applies.
441T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

Appendix A. Summary statistics
Appendix B. Description of variables and sources
B.1. Invalid voting in legislative elections
Blank and spoiled ballots cast for the lower house, or
for the unicameral legislature, as a percentage of overall
turnout. Sources: International IDEA (1997 and www.i-
dea.int), supplemented by data from the Political Data-
base of the Americas (www.georgetown.edu/pdba),
from the PARLINE database of the Interparliamentary
Union (www.ipu.org), and from the Elections Around
the World website (www.electionworld.org).
B.2. Personal vote
A rank ordering of electoral formulae based on their
incentives for politicians to cultivate a personal vote; in-
novated by Carey and Shugart (1995). A score of 9 rep-
resents the most candidate-centric voting system, and
a score of 1 represents the most party-centric system.
Source: dataset supplied by Nielson (2003).
B.3. District magnitude
Average district magnitude of the electoral districts
used in lower house elections. For mixed electoral sys-
tems, this value is downweighted by the percentage of
legislative seats elected in single-member districts.
Sources: dataset supplied by Nielson (2003) and supple-
mented by data in Payne et al. (2003).
B.4. Electoral disproportionality
Distortion of representation caused by translation
of votes into seats. Following Lijphart (1984), mea-
sured as the average vote-seat share deviation of the
two largest parties in each election. Sources: same
as for invalid voting, plus Political Handbook of the
World, and Keesing’s Record of World Events (various
years).
B.5. Unicameralism
Degree to which the national legislature ap-
proaches the unicameral model. Scoring follows Lij-
phart (1984). Countries with unicameral legislatures
receive a score of 4; countries whose chambers are
congruent and asymmetrical in a manner that favors
the lower house receive a score of 3; countries with
incongruent bicameralism receive a score of 2; coun-
tries with bicameral legislatures receive a score of 1,
and countries with strong bicameralism receive
a score of 0. Sources: coded by authors based on
Nohlen (1993), Jones (1995a,b), and legislative
websites.
B.6. Compulsory voting
Degree to which voter appearance at the polls is
mandated by national legislation. Countries with volun-
tary voting receive a score of 0 (Colombia and Nicara-
gua, plus Venezuela beginning in 1999); countries that
have a compulsory voting statute, but which have no
sanctions against nonvoters written into law, receive
a score of 1 (Costa Rica, Dominican Republic, El Salva-
dor, Guatemala, Panama, plus Venezuela only through
1998); compulsory systems possessing such legal sanc-
tions but leaving them generally unenforced receive
a score of 2 (Argentina, Bolivia, Brazil, Honduras,
Mexico, and Paraguay); and compulsory systems with
legal sanctions that are enforced in practice are given
the highest value of 3 (Chile, Ecuador, Peru, and Uru-
guay). Source: Payne et al. (2003, table 3.2).
B.7. Urbanization
Percentage of the total population living in the urban
areas. Source: United Nations (1980e2000).
Variable Mean Standard
deviation
Minimum Maximum
Invalid legislative
ballots
10.58 10.34 0.19 45.1
Personal vote 2.88 2.71 1 9
District magnitude 14.02 23.86 2 120
Founding election 0.11 0.31 0 1
Electoral
disproportionality
6.01 4.14 0.15 18.7
Unicameralism 1.96 1.86 0 4
Compulsory voting 1.76 0.91 0 3
Urbanization 64.21 17.00 37.16 92.24
Literacy 84.85 10.52 60.73 97.56
Logged per
capita GDP
7.52 0.67 6.15 8.81
Gini coefficient 49.74 5.87 39.19 60.64
Change in logged
per capita GDP
0.001 0.01 0.014 0.013
Revolutionary
violence
0.33 0.80 0 3
Freedom House 6.39 2.03 0 10
Change in
Freedom House
0.03 0.82 32
N¼75
442 T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

B.8. Literacy
Percentage of the population aged 15 years and older
that can read and write. Source: United Nations (1980e
2000).
B.9. Logged per capita GDP
The log of per capita gross domestic product mea-
sured in constant 1990 dollars. Source: data supplied
to authors by Statistical and Quantitative Analysis
Unit of the Inter-American Development Bank.
B.10. Gini coefficient of income inequality
Indicator of the size distribution of income, mea-
sured on a 0e100 scale where 0 denotes perfect equality
and 100 denotes perfect inequality. Source: World In-
come Inequality Database, available http://www.wider.-
unu.edu/wiid/.
B.11. Change in logged per capita GDP
Percentage change over previous year’s value for
logged per capita GDP.
B.12. Revolutionary violence
Number of acts of revolutionary violence occurring in
the election year. Source: State Failure Task Force Data-
base, available http://www.cidcm.umd.edu/inscr/stfail/.
B.13. Founding election
Founding elections are democratic contests that
mark a clear break from an authoritarian past, that
have unusual salience attached to them by major polit-
ical actors, and that represent the first major reallocation
of political power under a newly democratic regime (see
O’Donnell and Schmitter 1986). Legislative elections
embodying these characteristics were coded as 1 by
the authors, and 0 otherwise.
B.14. Freedom House score
Freedom House employs separate political rights
and civil liberties indexes, both of which are measured
from 1 (high freedom) to 7 (low freedom). We first
summed these into a composite index representing the
level of democracy. In Latin American countries over
the course of this study, the composite scale ranges
from 2 to 12. We then reverse the index and subtract
two from it, resulting in an 11-point variable ranging
from 0 to 10 and measured so that high scores represent
higher levels of democracy. Source: www.freedomhou-
se.org/ratings.
B.15. Change in Freedom House score
Absolute change over previous year’s value for Free-
dom House score.
References
Blais, A., 2006. What affects voter turnout? Annual Review of Polit-
ical Science 9, 111e125.
Blais, A., Dobrzynska, A., 1998. Turnout in electoral democracies.
European Journal of Political Research 33, 239e261.
Carey, J.M., Shugart, M.S., 1995. Incentives to cultivate a personal
vote: a rank ordering of electoral formulas. Electoral Studies
14 (December), 417e439.
Fornos, C.A., Power, T.J., Garand, J.C., October 2004. Explaining
voter turnout in Latin America, 1980 to 2000. Comparative Polit-
ical Studies 37 (8), 909e940.
Franklin, M.N., 1996. Electoral participation. In: LeDuc, L.,
Niemi, R.G., Norris, P. (Eds.), Comparing Democracies: Elec-
tions and Voting in Global Perspective. Sage Publications, Thou-
sand Oaks, CA.
Franklin, M.N., 1999. Electoral engineering and cross-national turn-
out differences: what role for compulsory voting? British Journal
of Political Science 29, 205e224.
Franklin, M.N., 2004. Voter Turnout and the Dynamics of Electoral
Competition in Established Democracies Since 1945. Cambridge
University Press, Cambridge.
Greene, W., 1993. Econometric Analysis, second ed. Macmillan,
New York.
Gujarati, D., 1995. Basic Econometrics, third ed. McGraw-Hill, New
York.
Hill, L., 2002. Compulsory voting: residual problems and potential so-
lutions. Australian Journal of Political Science 37 (3), 437e455.
Hirczy, W., 1994. The impact of mandatory voting laws on turnout:
a quasi-experimental approach. Electoral Studies 13, 64e76.
Huntington, S., 1991. The Third Wave: Democratization in the Late
Twentieth Century. University of Oklahoma Press, Norman.
International IDEA (Institute for Democracy and Electoral Assis-
tance), 1997. Voter Turnout from 1945 to 1997: A Global Report
on Political Participation. International IDEA, Stockholm.
Jackman, R.W., 1987. Political institutions and voter turnout in the
industrial democracies. American Political Science Review 81,
405e423.
Jackman, R.W., Miller, R.A., 1995. Voter turnout in the industrialized
democracies during the 1980s. Comparative Political Studies 27,
467e492.
Jones, M.P., 1995a. A guide to the electoral systems of the Americas.
Electoral Studies 14, 5e21.
Jones, M.P., 1995b. Electoral Laws and the Survival of Presidential
Democracies. University of Notre Dame Press, Notre Dame.
Kmenta, J., 1986. Elements of Econometrics, second ed. Macmillan,
New York.
443T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444

Kostadinova, T., October 2003. Voter turnout dynamics in post-com-
munist Europe. European Journal of Political Research 42 (6),
741e759.
Lijphart, A., 1984. Democracies: Patterns of Majoritarian and Con-
sensus Government in Twenty-One Countries. Yale University
Press, New Haven.
Lijphart, A., 1997. Unequal participation: democracy’s unresolved
dilemma. American Political Science Review 91, 1e14.
McAllister, I., Makkai, T., 1993. Institutions, society, or protest? Ex-
plaining invalid votes in Australian elections. Electoral Studies
12, 23e40.
Nielson, D.J., July 2003. Supplying trade reform: political institu-
tions and trade policy in middle-income democracies. American
Journal of Political Science 47 (3), 470e491.
Nohlen, D. (Ed.), July 1993. Enciclopedia electoral latinomericana y
del Caribe. Instituto Interamericano de Derechos Humanos, San
Jose
´, Costa Rica.
O’Donnell, G., Schmitter, P.C., 1986. Transitions from Authoritarian
Rule: Tentative Conclusions about Uncertain Democracies. Johns
Hopkins University Press, Baltimore.
Payne, M., Zovatto, D., Flo
´rez, F.C., Allamand, A., 2003. La Polı
´tica
Importa: Democracia y desarrollo en Ame
´rica Latina. Inter-Amer-
ican Development Bank and International IDEA, Washington,DC.
Pe
´rez-Lin
˜a
´n, A., 2001. Neoinstitutional accounts of voter turnout:
moving beyond industrial democracies. Electoral Studies 20,
281e297.
Powell, G.B., 1986. American voter turnout in comparative perspec-
tive. American Political Science Review 80, 17e43.
Power,T.J., Roberts, T.J., 1995. Compulsory voting, invalid ballots, and
abstention in Brazil. Political Research Quarterly 48, 795e826.
Turner, A., 1993. Postauthoritarian elections: testing expectations
about ‘first’ elections. Comparative Political Studies 26, 330e349.
United Nations, 1980e2000. Anuario estadı
´stico de Ame
´rica
Latina y el Caribe. United Nations Publications, New York.
444 T.J. Power, J.C. Garand / Electoral Studies 26 (2007) 432e444