See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/273214017

Acquiring the Habit of Changing Governments

Through Elections

Article in Comparative Political Studies · December 2014

DOI: 10.1177/0010414014543614

CITATIONS READS

50 1,566

1 author:

Adam Przeworski
New York University
222 PUBLICATIONS 28,532 CITATIONS

SEE PROFILE

All content following this page was uploaded by Adam Przeworski on 20 November 2015.

The user has requested enhancement of the downloaded file.

543614
research-article2014
CPSXXX10.1177/0010414014543614Comparative Political StudiesPrzeworski

Article
Comparative Political Studies
2015, Vol. 48(1) 101­–129
Acquiring the Habit of © The Author(s) 2014
Reprints and permissions:
Changing Governments sagepub.com/journalsPermissions.nav
DOI: 10.1177/0010414014543614
Through Elections cps.sagepub.com

Adam Przeworski1

Abstract
Changing governments through elections is a rare and a recent practice.
Yielding office the first time is foreboding because it entails the risk that
the gesture would not be reciprocated, but the habit develops rapidly once
the first step is taken. This article provides evidence for these assertions by
examining about 3,000 elections in the world since 1788.

Keywords
elections, democracy, regimes

Giving away votes to unworthy people, to the irrational passions of the parties
. . . is a suicide for a ruler and I will not commit suicide before a chimera.
—Domingo Santa María, the President of Chile between 1881 and 1886,
quoted in Collier and Sater (1996, p. 58).

We are not going to give up our country for a mere X on a ballot. How can a
ballpoint fight with a gun?
—Robert Mugabe, President of Zimbabwe.

Introduction
In a democracy, incumbent governments risk office in elections and yield it
when they lose. Why? The answer cannot be “because it is a democracy,” as

1New York University, New York City, USA

Corresponding Author:
Adam Przeworski, New York University, 19 West 4th Street, New York, NY 10012, USA.
Email: [email protected]

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
102 Comparative Political Studies 48(1)

the vast literature on elections in various regimes tautologically has it. True,
in the few lucky countries in which most political scientists happen to live,
partisan alternation in office through elections has become so routinized that
it appears as only natural. Parties offer platforms and personalities, people
vote, votes are counted, someone is declared the winner, the winner enters
unimpeded the White, Pink, or Blue House, and this sequence of events is
repeated at regular intervals. We can take it for granted, so we do. Yet exam-
ining history, one is struck how difficult it is for rulers to place their power at
risk in elections, to trust that if they do so their actions would be reciprocated,
to feel safe that they can lose office without losing much else, that political
conflicts can be processed with limited stakes. The historical facts are elo-
quent: While the first partisan alternation in history occurred, albeit painfully,
in the United States as early as in 1801, the phenomenon has been rare until
very recently. Still, as of 2008, 68 countries, including the 2 elephants, China
and Russia, never experienced a peaceful transfer of power between parties
as a result of an election.
In a democracy, incumbents expose themselves to the possibility of defeat
and yield office with full confidence that their opponents would do the same
if they become the rulers. This equilibrium may be sustained by the possibil-
ity that deviations—not holding competitive elections or not obeying their
outcomes—would provoke costly conflicts (as in Fearon, 2011, or Przeworski,
2005) but to relinquish the reigns of power all successive incumbents must
believe that the current opposition would reciprocate if it wins an election
and becomes the incumbent in turn. This belief is easy to sustain when the
past incumbent had already yielded office having lost an election. But how is
one to know what would happen if he loses an election when nothing of the
sort ever happened before? It may not have happened because no elections
were held, because the opposition was not allowed to compete effectively, or
only because voters freely kept the incumbent in office. But whatever the
reasons, the prospect of defeat is a plunge into unknown. Will the defeated
incumbents survive intact, as political contenders with a prospect of some
day returning to office, or do they risk death, imprisonment, or exile, personal
as well as political annihilation?
When I once commented to a Polish communist reformer that if the regime
would hold a competitive election it would lose, his response was “What mat-
ters is not whether we would win or lose but what we would lose.” This
“what” is quintessential. When incumbents fear that an electoral defeat may
mean a loss of life, freedom, or even just fortune, their risk is just too high
(see Makarenko, in press, on “tolerable uncertainty”). And such fears may be
grounded. The fact that most rulers who did expose themselves to defeat and
lost were not punished1 is not informative: Only those among them who do

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 103

not expect to be punished would take this risk. Neither safeguards nor assur-
ances are enough: Before holding the plebiscite of 1989, the Chilean military
protected itself by all kinds of institutional devices and amnesty laws, yet
many still ended in jail. In Central African Republic, General Kolinga, who
took office by a coup in 1981 and got himself elected in 1986, was pressured
to hold a competitive election in 1992 and lost to Ange-Félix Patassé, who
promptly stripped Kolinga of his military rank and persecuted several of his
ministers. In Malawi, Hastings Banda, who was in power for 27 years, lost an
election in 1994, only to be arrested and charged with murder allegedly com-
mitted 10 years earlier. Several post-military countries in Latin America
relentlessly persecute violators of human rights while post-communist gov-
ernments incessantly invent new “lustration” laws to harass anyone suspected
of having collaborated with the previous regime.2
Releasing the reigns of power is like letting go of the trapeze without
knowing whether one would be caught, a matter of belief about someone
else’s good faith, of trust. Put yourself in the shoes of General Pinochet, the
Polish Communists in 1989, South African Whites in 1994, the Algerian mili-
tary in 1991, or President Putin today. If you allow elections to be competi-
tive, lose, and yield office, what will the winners do to you? After all, the
opposition does not just want to defeat you but to destroy you: It accuses you
of breaking laws, of amassing a fortune; in the case of Putin even of having
bombed a Moscow building as a pretext to intensify the Chechen war. The
risk is foreboding.
One may think, however, that people learn from experience. The incum-
bent may fear that the opposition would exact a revenge if it is victorious but
the victorious opposition will have already seen that the incumbent did
release its hold on power. It may still well be that a party that yielded office
once would not do it the next time: Individuals involved may not be the same,
circumstances may change. Moreover, an entire sequence of elections may
occur peacefully, with or without alternations, and then some exogenous
event may lead to a coup, usurpation of power by the current incumbent, civil
war, or some other constitutional irregularity. Nevertheless, having gone
through alternation once should at least assuage the fears.
If this argument is correct, the first alternation should be difficult and rare
but the probability that governments would be selected through elections
should increase sharply once one had transpired. And when more alternations
occur, future conflicts should be more likely to be processed by elections,
rather than by force. This is indeed what the historical patterns indicate.
Periods in which no alternation had yet occurred are unlikely to witness one
in any subsequent election and not very likely to experience one at all, ending
in some form of a violent conflict. Incumbents are highly unlikely to lose an
election and are quite prone to disobey the verdict of the polls when they do

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
104 Comparative Political Studies 48(1)

lose. Once the first alternation does occur, however, electoral defeats become
much more frequent and the rate of compliance increases, so that the electoral
mechanism becomes entrenched.
The aim of this analysis is to identify the conditions under which incum-
bents are willing to lose elections and yield office, the events adding up to a
partisan alternation in office through elections, “alternation” for short. Note
that alternations need not coincide with transitions to democracy; indeed,
they rarely do coincide. While alternations are a sine qua non of democracy,
transitions to democracy need not entail alternations: Most transitions
occurred when authoritarian rulers were overthrown by force. Conversely,
most alternations took place when previous elections were already
competitive.
The remainder of this article first provides historical background on the
outcomes of national-level elections in the world after 1788, based on the
Political Institutions and Political Events (PIPE; Przeworski, 2013) data col-
lection and delves into the conceptual relation between alternations and tran-
sitions to democracy. A simple analytical formulation elucidates the choices
confronting successive incumbents. First alternations within particular spells
of continuously held elections are then analyzed. With these initial condi-
tions, analysis moves to the dynamic of subsequent alternations and their
impact on the survival of the electoral mechanism. A summary and conclu-
sion close the article.

Background
Elections may or may not be contested. When they are contested, incumbents
may win or lose and the winners, whether the incumbent or the opposition,
may or may not assume office. The distribution of these outcomes is por-
trayed in Figure 1, which covers most national-level elections in the world
between 1788 and 2008 in which there was an incumbent which presented
itself in the election. The incumbent may be a person, or a party, or a succes-
sor designated by the government in place. Elections are contested if there is
more than one candidate for president in presidential systems or when voters
in at least some districts have a choice of candidates or lists in legislative
elections.3
Of the 2,886 elections in which there was some kind of incumbent, in
656 / 2,886 = 0.23 there was no opposition. Incumbents won (2,210 + 116) /
2,886 = 0.81 of all elections, obviously all in which there was no opposition
but also (1,497 + 73) / 2,230 = 0.70 of those that were contested. In (43 + 73
+ 57) / 2,886 = 0.06 elections, the winner, whether the incumbent or the
opposition, did not end up in office. Finally, in 603 / 2,886 = 0.21 elections,
incumbents lost and peacefully yielded office to the winners.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 105

Election Held

Incumbent Challenger
Wins Wins

Incumbent stayed Incumbent Challenger Challenger unable to

in power loses power assumes power assume power

1497 613 73 43 603 0 57 0

Contested Uncontested Contested Uncontested Contested Uncontested Contested Uncontested

Figure 1. Outcomes of elections.

Periods in which any kind of elections are held regularly, without being
interrupted by coups, autocoups, civil wars, or other major constitutional
irregularities, constitute “electoral spells.”4 An electoral spell begins when
someone is elected for the first time, either the first time in the history of a
country or after someone grabbed power by force.5 Among the 423 electoral
spells that ever began, 156 continued until an alternation. The rest either con-
tinued until the country was last observed or were interrupted by a constitu-
tional breakdown. An “alternation” is considered to have occurred if (a) the
incumbent, however he originally entered office but was already elected,
presents itself in an election and (b) the party or the person of the chief execu-
tive changes as a result of an election.6 Also included are instances in which
there are no formally constituted parties but a manifest government candidate.7
The winner of the election must enter the office without anyone else attempting
to squeeze in before and must hold office at least for 1 year or until the next
election, whichever comes first, but not necessarily the entire constitutional
term.8 The result on an election is “obeyed” when these conditions are satisfied;
it is not obeyed when the losing incumbent attempts to hold onto office or the
defeated opposition tries to force its way into office. Note that alternation is a
conjunction of two events: The incumbent lost and the result was obeyed.
While Przeworski (1991) defines democracy as a “system in which parties
lose elections” (p. 11), and while everyone agrees that the occurrence of alter-
nations is the litmus test of democracy, the conceptual relation between tran-
sitions to democracy and alternations is complex. Transitions to democracy
occur when the first competitive election is held after a period in which elec-
tions were not held or were uncontested or were so rigged that the opposition
had no chance to win. Most transitions occurred not when the authoritarian
rulers held competitive elections but when they were overthrown by the use or

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
106 Comparative Political Studies 48(1)

Table 1. Conceptual Relation Between Regime Transitions and Alternations,

Given That Current Election Is Contested.

Incumbent Incumbent
runs loses Transition Alternation Example
Previously elections were

   No — Yes No Argentina, 1983

Not held Yes No Yes No Ghana, 1992
   Yes Yes Yes No Chile, 1989

   No — Yes No Niger, 1993

Uncontested Yes No Yes No Mongolia, 1990
   Yes Yes Yes Yes Hungary, 1990

   No — No No Georgia, 2004
Contested Yes No No No Botswana, 1964-to date
Yes Yes No Yes Luxembourg, 1974

threat of force. Hence, many transitions do not entail alternation, while most
alternations occur when the regime is already democratic.
As shown in Table 1, transitions and alternations coincide only in the cases
in which previous elections were uncontested, the former incumbent pre-
sented itself in first contested elections, lost, and obeyed, as is in Brazil in
1985 as well as Hungary and Poland in 1989. Because an alternation can
occur only if the current ruler was elected, no alternation can occur if elec-
tions were not previously held, as in Argentina before 1983 or Ghana before
1992. Alternation also cannot occur unless the formerly elected incumbent
presents itself (which excludes Niger in 1993) and it does not occur when the
forces that have supported the dictatorship present themselves and win, as in
Spain in 1977 or Mongolia in 1990. In turn, transition does not occur if previ-
ous election(s) were contested, whether an alternation takes place, as in
Luxembourg in 1974, or the same party continues to win, as in Botswana
from the independence in 1964 until the present.

The Argument
I hesitate to call the argument that follows a “model” because it is just a math-
ematically adorned restatement of the verbal reasoning. Nevertheless, formu-
lating the argument in the mathematical language does elucidate the choices
available to incumbent rulers.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 107

Each incumbent faces at each time a choice between two lotteries. In one,
he can hold elections which he cannot lose or not hold them at all, facing the
possibility of being removed by force. In the second, he lets the voters decide.
Let q be the probability that an incumbent who assures himself of not being
defeated in an election survives in power for one period (measured in electoral
terms). Let the one-period utility of surviving in office be U(w) and the utility of
being deposed by force be U(d). The expected value of the first lottery is then
EV ( q ) = qV ( w ) + (1 − q )V ( d ) , (1)

where V (.) are the expected present values, discounted flows of future util-
ity, associated with each outcome.
In the second lottery, the incumbent holds an election which he may lose.
What the incumbent does not know is what would happen if he does lose:
With probability θ , he may be treated just as a future electoral opponent,
obtaining V (l ) , and with probability 1− θ , he may experience the same fate
as if he were deposed by force: death, imprisonment, or exile. Let p be the
probability that the incumbent wins an election. The expected value of the
electoral lottery is9

EV ( p ) = pV ( w ) + (1 − p ) θV ( l ) + (1 − θ )V ( d )  , (2)

with V ( w) > V (l ) > V (d ) .

The probability of winning is obviously 1 if no opposition is tolerated, as
under communism. But the incumbent may be also certain to win even if
opposition vigorously competes: p = 1 even when the incumbent is certain
to win (or at least report) only 50.1% of the vote (or whatever share of the
vote is necessary to remain in office). In general, if v j is the vote share party
j is certain to obtain and v¬j is the certain share of party ¬j,
p j = (0.5 − v¬j ) / (1 − v j − v¬j ) < 1 if v j < 0.5, v j + v¬j < 1. Hence, the incum-
bent can assure himself of victory forcing some undecided voters to vote for
him, preventing some opposition voters from voting, or just fabricating
votes.10 For simplicity, I assume that any incumbent j can only choose to
either make his victory certain or to hold an election in which his probability
of winning depends on the distribution of the electoral support, p = p j (v j , v¬j )
resulting from electoral competition that is “free and fair.”11 It is important to
keep in mind that p can be high but, as long as it is less than 1 an accident
may still occur.
Suppose now that the incumbent is certain that if he loses the election he
will experience hard landing, θ = 0 . He then prefers the electoral lottery only
if p > q . But if he attaches some positive probability to the event that if he

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
108 Comparative Political Studies 48(1)

loses he would be treated just as a future electoral opponent, he opts for the
electoral lottery even if p < q , specifically if

p−q V (l ) − V ( d )
> −θ ≡ −θα < 0, (3)
1− p V ( w) − V ( d )

or as long as
q − θα
p> ≡ p* ( q, θ, α ) , (4)
1 − θα

V (l ) − V ( d )
where α = is the value of being in opposition relative to being
in office. V ( w ) − V ( d )
This condition is portrayed in Figure 2. When θ = 0 , the ruler holds com-
petitive elections only if p > q . As θ increases, he is willing to hold them
even if he has little of a chance to win; indeed, when α is sufficiently high,
even if he is certain to lose. This is because losing an election is not that much
of a disaster, obviously worse than winning it but far better than physical and
political annihilation. It is not whether the incumbent would lose but what he
would lose that matters.
This condition does not guarantee, however, that the incumbent would
obey the verdict of the polls if he loses. I leave open the possibility than he

Figure 2. Conditions for competitive elections.

q = 0.6, α = .66.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 109

might not obey because, as shown below (Table 5), among the 168 instances
in which an incumbent lost an election for the first time within a particular
spell in 29 of them, the winner was not able to assume office. It is reasonable,
however, to assume that an electoral defeat results in a loss of the incum-
bent’s power to impose himself by force, so that when the incumbent loses an
election q falls to some level q’ . One might think, with Geddes (2005) or
Simpser (2013), that an electoral defeat signals that the incumbent is not pop-
ular or at least that it does not control the state apparatus necessary to manip-
ulate the outcomes of elections, so that he is less likely to impose himself by
force. Alternatively, one might follow Przeworski, Rivero, and Xi (2014) in
thinking that at least some of those who bear arms are constitutionalists,
ready to obey a duly elected government.

Assumption: An electoral defeat reduces the incumbent’s capacity to

impose himself by force to q’ < q .

The condition to obey an electoral defeat is then that the expected value of
being in electoral opposition, θV (l ) + (1 − θ)V (d ) , is larger than the expected
value of imposing oneself by force with q’ < q , which is q’V ( w) + (1 − q’ )V (d ).
This condition is satisfied as long as q’ < θα and it still does allow the pos-
sibility that the incumbent would not obey an electoral defeat.
Suppose now that the first alternation had occurred: The incumbent ran an
election with p < 1 , lost, obeyed, and the opposition is in office. The winner
faces the same choice between the two lotteries as the previous incumbent.
The difference, however, is that the winner knows that the defeated incum-
bent revealed himself to be of the type who allows himself to lose. Thus, he
can now believe that θ = 1 . Assume that the distribution of partisan support
does not change in the first period after the alternation had occurred,12 with
p still characterizing the defeated incumbent. The expected value of the
electoral lottery for the winner is then

EV (1 − p ) = (1 − p )V ( w ) + pV ( l ) , (5)

while the expected value of not holding competitive election is

( ) ( )
EV 1 − q’ = 1 − q’ V ( w ) + q’V ( d ) , (6)

so that the winner holds a competitive elections as long as

V ( w) − V ( d )
p < q’ = q’ / (1 − α ) ,
V ( w) − V (l ) (7)

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
110 Comparative Political Studies 48(1)

that is, as long as the former incumbent is not highly likely to return to office
in the next election.
Now, the incumbent must anticipate that if q’ / (1 − α) ≤ p < 1 and by
some fluke he happens to lose, the winner will not reciprocate by holding a
competitive election. Yet, as long as q < 1 , the incumbent still holds elections
with p < 1 . This case can occur only when α < 1− q’ , so that being in oppo-
sition is costly, but the incumbent expects to win (1 − p ) −1 consecutive elec-
tions, and because p is high, the prospect of the defeat materializing is
distant.
∗
Proposition 1a: If p > p , the incumbent in a spell that never experi-
enced an alternation holds elections with p < 1 ; otherwise, he does not
hold elections or assures himself of electoral victory.
Proposition 1b: When elections are held with p < 1 and the incumbent
loses, he obeys if q’ < θα and does not yield office otherwise.
Proposition 1c: If the incumbent loses and obeys, the winner reciprocates
by holding competitive elections if p < q’ / (1 − α) and does not hold elec-
tions or assures himself of victory otherwise.
Proof: It follows from the above, except for the claim that the incumbent
holds elections with p∗ < p < 1 even if the winner would not hold elec-
tions or assured himself of victory, the proof of which is in the appendix.

The comparative statics are obvious. For any q , a lower p is required for
the first incumbent to hold a competitive election as either α or θ increases.
In turn, the second incumbent is more inclined to reciprocate as α increases.

First Alternations
The first alternation in history took place in the United States following a
deeply divisive election of 1800 (about which Dunn, 2004; Weisberger,
2000). In Great Britain, the first alternation occurred in 1835 when the king
accepted the electoral victory of the Liberals. The first alternation in Latin
America was in New Granada (now Colombia) in 1837: General Santander,
who believed that the country was not ready for a civilian chief executive,
supported General Obando to be his successor, yet a civilian, Dr. José Ignacio
de Márquez won the plurality of electoral votes and the Congress confirmed
his victory (Posada Carbó, 2000). According to Bushnell (1993), “Santander
then delivered his office to someone he had opposed—taking pains to point
out, in a proclamation, that he had thus respected the will of the people and
the law of the land” (p. 90).13
As shown in Table 2, 19 countries experienced at least one alternation
before 1900, but only in 8 or 9 of them (depending how one thinks of Vichy

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 111

Table 2. Partisan Alternations in Office as a Result of Elections During the 19th

Century.

Country System First Second Third

The United States Presidential 1800 1828 1840
The United Kingdom Monarchy 1835 1841 1852
Colombia Presidential 1837 1848 1930
Belgium Monarchy 1847 1856 1870
Dominican Republic Presidential 1849 1853 1978
Honduras Presidential 1852 1928 1932
Portugal Monarchy 1860 1864 1865
Liberia Presidential 1855 1869 1877
Spain Monarchy 1857 1865 1919
Italy Monarchy 1867 1876 1892
Argentina Presidential 1868 1916 1989
The Netherlands Monarchy 1874 1877 1879
France Parliamentary 1877 1881 1885
Canada Monarchy 1878 1896 1911
Sweden Monarchy 1884 1905 1911
New Zealand Monarchy 1884 1887 1928
Costa Rica Presidential 1889 1909 1923
Greece Monarchy 1890 1950 1951
Norway Monarchy 1891 1903 1912

Italic fonts indicate that coups, civil wars, or other constitutional breakdowns occurred
between the previous alternation and the date appearing in italics.

France) control over the office of the chief executive alternated peacefully
without interruption by coups or civil wars from then on.
Elections preceding the first alternations were contested everywhere
except for Argentina before 1868 and Costa Rica before 1889. In turn, in
Dominican Republic in 1849 and Spain in 1857, elections in which first alter-
nations occurred were preceded by coups. Hence, these are the cases in which
alternations constituted transitions to democracy. In all other cases, contested
elections were held regularly before the first alternation happened to
transpire.
In almost all instances, victory of the opposition met with considerable
resistance. Jefferson may have prevailed only as a consequence of Madison’s
threat to mobilize the Virginia militia (Dunn, 2004). In Colombia, the defeated
candidate, General Obando, rose against President Márquez 2 years later, but
was defeated. In Great Britain, the king appointed a Tory prime minister in
spite of the Tory electoral defeat in 1834 and only the second victory of the

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
112 Comparative Political Studies 48(1)

opposition forced him to accept the Melbourne government. In Belgium,

Liberals had to win twice before assuming office in 1847, in Denmark minor-
ity right-wing governments stayed in office in spite of repeated defeats
between 1872 and 1901, in the Netherlands the same was true between 1856
and 1871.
The resistance against alternation had different sources in different sys-
tems.14 One might expect that partisan alternations would be easier in monar-
chies because the monarchs can provide the guarantee to the losers that they
could return to office. In these systems, the opposition to alternation came
from monarchs, for whom allowing an alternation resulting from elections
signified the loss of power to appoint governments at their will (Przeworski,
Asadurian, & Thomas, 2012). King Wilhelm of the Netherlands abdicated
rather than accept a parliamentary government, as did King Prajadhikop of
Thailand, while Queen Victoria was insulted by having to accept it. Most
monarchies resisted too long and ended up being abolished.15 In turn, in the
systems in which losers could not rely for protection on monarchs, the deci-
sion to relinquish the reigns of power was more stark, because no third actor
could protect the losers.
While obviously an alternation can occur only if an election is contested,
previous elections within a particular spell may or may not have been.
Uncontested elections were frequent in Latin America during the 19th cen-
tury but became rare exactly when Lenin’s invention of one-party systems
spread around the world. In turn, following the Soviet elections of 1937,16 the
countries which fell under Soviet domination and several newly independent
countries in Asia and Africa held regular “elections” in which no one was
selected because all opposition was suppressed. Yet incumbents also won an
overwhelming proportion of elections in which there was some opposition.
The frequency and the longevity of electoral spells in which elections were
contested but incumbents never lost is striking. Table 3 lists spells of con-
tested elections which incumbents repeatedly won during at least 40 consecu-
tive years.
These cases are a source of a major difficulty in classifying regimes and
thus in identifying regime transitions. Are these spells democratic (after all
elections were contested) or are they authoritarian (after all the opposition did
not seem to have much of a chance to win)? Was the United States a democ-
racy before 1800? George Washington was elected in 1788 to 1789 and
reelected in 1792 without opposition and while the 1796 election was con-
tested, it was won by Washington’s vice-president, John Adams, who
declared, “There is nothing I dread so much as a division of the republic into
two great parties, each arranged under its leader, and concerting measures in
opposition to each other” (quoted in Dunn, 2004, p. 39). Luxembourg, Italy,

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 113

Table 3. Long-Lasting Spells of Regularly Contested Elections Which Incumbents

Always Won.

Country Year ended Years lasted Reason ended (if ended)

Luxembourg 1974 126 Alternation
Norway 1890 74 Alternation
Romania 1939 73 Defeat in war
The Netherlands 1873 59 Alternation
Brazil 1889 55 Monarchy abolished
Mexico 1999 54 Alternation
Denmark 1900 52 Alternation
Germany 1918 51 Monarchy abolished
Italy 1995 50 Alternation
Chilea 1890 50 Civil war
Tunisia 2008 49 Continued as of 2008
Japan 1992 47 Alternation
Spainb 1923 48 Coup
Taiwan 1999 46 Alternation
Austrian Empire 1910 44 Alternation
Botswana 2008 44 Continued as of 2008
Argentina 1915 42 Alternation
Nepal 2004 42 Autocoup
Portugal 1973 40 Coup

a. The legislature was temporarily closed in 1841, otherwise the continuity would be dated
back to 1831, with the duration of 60 years.
b. Control over government alternated regularly between two parties but alternations always
occurred before elections and governments newly chosen by the King never lost.

and Japan smell democratic and are invariably considered as such but Mexico,
Taiwan, and Botswana are generally seen as authoritarian. To call the cases in
which elections are pluralistic but the opposition never wins “hybrid” is just
an admission of the inability to decide. Przeworski, Alvarez, Cheibub, and
Limongi (2000) coped with this issue by using a retrospective rule to classify
as democratic periods of continued rule of the same party whenever an alter-
nation subsequently occurred under the same rules of competition. But they
also threw their hands up in the air when history was not kind enough to
generate this event and the electoral spell was right-hand censored, as in
Botswana. Geddes, Wright, and Frantz (2013, Section 1.3) claim that their
coding rules resolve this conundrum by classifying as autocracies regimes in
which rulers came into power by “any means besides direct, reasonably fair,
competitive elections.” But in the end, one still wonders on what grounds

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
114 Comparative Political Studies 48(1)

they consider Italy or Japan democratic and Mexico, Taiwan, or Botswana

autocratic. In fact, none of the criteria they list as identifying transitions from
autocracy to democracy apply to Mexico between 1934 and 1988 as well as
in 1994 and to Botswana at any time. Why was the defeat of Partido
Revolucionario Institucional (PRI) in Mexico in 2000 a transition to democ-
racy but the debacle of Christian Democracy in Italy in 1995 not? Alternations
occurred in both countries in these years after long periods of one-party dom-
ination, so one needs other criteria to decide whether or not these were also
regime transitions. The first alternation may thus occur either when previous
elections within a particular spell were uncontested17 or when they were con-
tested but the incumbent party always won, that is, when the regimes was
previously not democratic or its status was ambiguous.
To study statistically the determinants of the first alternation within each
spell, examine the implications of the argument outlined above. The incum-
bent is willing to risk losing office in a competitive election, that is, to run an
election in which his probability of winning depends on the actual distribu-
tion of support, if this probability is greater than the critical value
p∗ = (q − θα) / (1 − θα) . Assume that the incumbent knows or has beliefs
about the values of q, θ and α but he is uncertain about his support, so that
p is a random variable with support on [0,1] . The probability that the
incumbent decides to run a clean election, rather than assure himself of vic-
tory and face potential resistance by force, is then Pr( p ≥ p∗ ) = 1 − F ( p∗ ) .
Now, note that p∗ increases in q and declines both in θ and α . Hence, the
probability that the incumbent risks a defeat declines in q and increases in θ
and α .
To test this argument, we need variables that affect values of these param-
eters, not easy to find for the period under consideration here, which goes
back to the year of independence in each country. Here are the potential can-
didates. One may suppose that the incumbent is more likely to expect that he
would be treated as a future electoral opponent if (a) he has not repressed all
opposition in the past (the variable contested indicates whether opposition
was allowed in the previous election), (b) he had been in office for a shorter
time (lag head age), having fewer opportunities to commit punishable trans-
gressions, and (c) there were fewer instances of constitutional breakdown in
the past history of the country (the variable electoral spell shows the number
of past breakdowns). In turn, the incumbent expects that he would suffer less
being in opposition if (a) the country has a longer standing constitution (age
constitution), (b) if the constitution provides for checks and balances in the
form of bicameralism (bicameralism) and executive veto (outside blocker),
(c) the system is parliamentary, and (d) the country has a higher per capita
income (gdpcap, from Maddison, 2010).18 Finally, in the light of Geddes,
Wright, and Frantz (2014), as well as Escribà Folch (2014), the fate of the

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 115

Table 4. Determinants of the First Alternation Within a Spell (Partial Derivatives

or Effects of Dichotomous Variables).

Variable Effect 95% CI Effect 95% CI

Contested 0.100*** [0.057, 0.142] 0.078*** [0.045, 0.111]
Lag head age −0.001 [−0.005, 0.003] −0.000 [−0.003, 0.002]
Electoral spell 0.006 [−0.001, 0.013] 0.003 [−0.003, 0.010]
Age constitution 0.001** [0.000, 0.001] 0.001*** [0.000, 0.001]
Bicameralism −0.058** [−0.103, −0.012] −0.022 [−0.057, 0.019]
Outside blocker 0.015 [−0.023, 0.052] −0.012 [−0.039, 0.016]
Parliamentarism 0.032 [−0.022, 0.085] 0.028 [−0.015, 0.072]
Monarchy −0.051 [−0.106, 0.005] −0.035 [−0.085, 0.016]
Forced entry −0.088*** [−0.134, −0.043] −0.092*** [−0.123, −0.060]
Gdpcap 0.001** [0.000, 0.001]
Year 0.0001*** [0.000, 0.001]
Electoral age 0.006** [0.001, 0.011] 0.004** [0.000, 0.001]
Electoral age^2 −0.000** [−0.000, −0.000] −0.000** [−0.000, −0.000]
Electoral age^3 0.000** [0.000, 0.000] 0.000** [0.000, 0.000]
N 834 1,200
Spells 236 337
Lroc 0.73 0.75

CI = confidence interval; lroc = area under the receiver operating curve.

**significant at 5%, *** significant at 1%.

outgoing incumbents may depend on the type of the regime. Monarchy is

easy to identify, and it is included, but note that the chief executives under
this system are prime ministers. Finally, because the concept of the “military”
does not travel well across time (Rouquie, 1994),19 I use instead an indicator
variable of whether the current incumbent entered into office by force (forced
entry) or by constitutional means.20
These hypotheses are tested statistically with pooled probit regressions
(with standard errors clustered in each spell, country spell).21 Following
Carter and Signorino (2010), to control for the unobserved factors that gener-
ate duration dependence, the regressions also include the number of years
each spell had lasted up to the current date (electoral age), its square and
cube. Finally, because there are many missing observations for per capita
income, to check if the results are biased by the availability of these observa-
tions, these regressions are replicated with year in place of gdpcap.22 The
results are shown in Table 4.
These are paltry results, indicating that the uncertainty surrounding the
first alternation is not well captured by these observable conditions.23 Only
two variables clearly indicate path dependence: Electoral spells in which

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
116 Comparative Political Studies 48(1)

opposition was previously repressed are less likely to experience an alterna-

tion and incumbents who entered office by force are less likely to be subject
to this unpleasant event. Neither the longevity in office of the chief executive
nor the number of past institutional breakdowns matter much. First alterna-
tions are more likely in spells with older constitutions but the formal provi-
sions of the constitutions do not seem to reassure the incumbents: The sign of
bicameralism is opposite to the stipulated one, while executive veto plays no
role. Parliamentarism does not differ from presidentialism, while the effect of
monarchies is weak. Higher per capita income does increase the probability
of alternation. Finally, duration dependence is pronounced: The probability
of alternation first increases in electoral age, then it declines, and increases
again in the few instances where elections continue without alternation for a
very long time.
How much one should believe these results cannot be ascertained. Several
factors that likely influence decisions of incumbents cannot be systematically
observed. Perhaps most important among them is popular pressure on the
rulers to hold clean elections: “diretas já” in the Brazilian slogan of 1985. Yet
data for “unrest”—mass demonstrations, national strikes, and riots (Banks,
2005)—are available only after 1918 and when regressions are replicated
with this variable, it plays no role while the results for other variables remain
qualitatively the same. One would also expect that rulers are less inclined to
risk office when holding office is a source of enrichment, when assets held by
the ruling elite are less mobile, and when rulers originate from particular
ethnic or religious groups. Moreover, one may also think that because mem-
bers of the ruling group may be differentially affected by losing power, they
often divide whether to take the risk. In particular, those members of the rul-
ing apparatus who had committed acts of repression may fear the fate of
several agents of the Portuguese secret police in 1985, while those with clean
hands need not fear retribution. Hence, the omitted variables are many and
the resulting bias is unknown.
I suspect that first alternations are difficult to predict because they entail
idiosyncratic factors, accidents of history, sometimes very minor. Sparks
(1994) recounts that personal trust between a prominent White politician and
a Black businessman was crucial in the South African negotiations, and this
trust was built when a daughter of the latter broke a leg and the former rushed
to extend generous help. In some instances, personal understanding between
jailers and their victims developed during interrogations, for example,
between the Polish Minister of Interior, Władysław Kiszczak, and a leader of
the opposition, Adam Michnik. In Spain, mutual respect between Adolfo
Suarez, Francoist prime minister, and Santiago Carrero, the leader of the
Communist Party, was gradually built in secret meetings (Cercas, 2011). In

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 117

Kenya, the democratic opposition talked to the military, who then organized
a national day of honoring the retirement of President Moi, against his will.
Such random events inflate the errors of regressions but they are crucial in
overcoming the incumbents’ fear of the consequences of leaving power.
When a person, a party, or a clique had been in power for decades, their priors
are so uninformative that anything, even a minor event, can alter beliefs. And
beliefs are all that the incumbents can rely on to decide whether to put their
future at stake in elections.

Dynamic of Subsequent Elections

Once an alternation occurs, much of the uncertainty is unveiled. Having seen
that the incumbents were ready to lose and yield office, the winners should be
willing to do the same. If this is true, we should observe subsequent alterna-
tions to become more frequent once the first one occurs. Yet, as seen in Table
1, incumbents always enjoy some advantage, so that they may keep winning
elections even if they are “free and fair.” As a consequence, the time between
electoral defeats of incumbents may be long: In the United States, for exam-
ple, the second alternation occurred 28 years, seven electoral periods, after
the first one, and by that time none of the protagonists of 1800 was alive, the
party system had changed, as did other conditions. Hence, new doubts may
arise.
Examine first the effect of past alternations on the probability that another
one would occur in the next election, remembering that an alternation is a
conjunction of two events: Incumbents lose and obey the verdict of the polls.
The outcome of an election is the result of decisions of voters, given all the
actions by which the incumbents attempt to protect themselves and the oppo-
sition attempts to unseat them. The decision to obey belongs to the opposition
when the incumbents won and to the incumbents when they lost.
The probabilities in Table 5 concern results of particular elections. The
first one is the probability that an incumbent would lose the next election, the
second that both the winner and the loser would obey its result, the third is the
conditional probability that the incumbent would obey the outcome if he or
she lost an election, and the fourth is the conjunction of the two events: lose
and obey.
It is apparent that once an incumbent was defeated and yielded office
peacefully, the defeat of the next incumbent and the compliance with the
outcome become much more likely. First alternations are rare not only
because incumbents are unlikely to lose but also because they often do not
leave office when they do lose. Subsequent alternations are more likely
mainly because incumbents are more disposed to expose themselves to

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
118 Comparative Political Studies 48(1)

Table 5. Outcomes of Subsequent Elections Given the Number of Past

Alternations Within an Electoral Spell.

Past Number of
alternations electionsa Pr(lost) Pr(obey) Pr(obey|lost) Pr(alternation)
0 1,221 0.14 0.95 0.83 0.12
1 353 0.33 0.94 0.88 0.30
2 162 0.47 0.96 0.95 0.45
3 102 0.44 0.94 0.93 0.42
>3 413 0.41 1.00 1.00 0.41
Total 2,251 0.24 0.96 0.91 0.22

Spells which experience an alternation during the first election are considered here as having
one past alternation.
a. This is the number of elections that took place but the base of each probability is slightly
lower given that the information about the final outcome is not always available.

competition. Alternations become routine quite quickly: On average, it takes

9.0 elections for the first alternation to occur but it takes only 3.4 elections
for the second and only 2.3 for the third.
If the uncertainty about the treatment of the defeated incumbent is lifted
when the previous incumbent exposed himself to losing and left office upon
defeat, we should observe that none of the factors influencing the probability
of the first alternation should matter for the occurrence of subsequent ones.
This prediction is strongly supported when the analysis of the first alternation
is replicated given that at least one had occurred.24 In turn, the effect of the
number of past alternations is powerful: On average, each past alternation
increases the probability that another one would occur in the next election by
about .05. Hence, once one alternation had occurred, the next ones depend
only on the past number of alternations and random shocks.25 The habit of
changing rulers through elections is self-institutionalizing across institutional
and economic environments.
Having experienced alternations does not guarantee, however, that the
electoral mechanism would continue to function. It is a different question
whether an alternation would occur in subsequent elections and whether the
electoral mechanism would survive or be interrupted by a constitutional
breakdown.26 Table 6 shows the number of spells with a given number of past
alternations, the number that was censored, the number of spells that ended in
a constitutional breakdown, the probability that an electoral spell would be
broken at any time, and the probability that the electoral mechanism would
continue to be used, all as a function of the number of past alternations.
The probability that elections would cease to function does decline rap-
idly, dropping to zero when a spell witnessed more than six past alternations.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 119

Table 6. The Probability That at Least One More Alternation Will Occur Within
a Spell Given the Number of Past Alternations Within a Spell.

Past Pr(continue|
alternations Spells with Censored Broke Pr(broke)a uncensored)b
0 423 53 214 0.51 0.42
1 156 27 26 0.18 0.79
2 102 17 13 0.13 0.85
3 72 25 5 0.07 0.89
4 42 10 4 0.10 0.88
5 28 4 1 0.04 0.96
6 23 4 1 0.04 0.95
>6 18 18 0 0.00 1.00

All spells, including those in which alternation occurs in the first election within the spell,
are considered to have begun with zero past alternations. Hence, 423 is the total number of
spells.
a. Pr(broke) = Number of spells that ended unconstitutionally/Number of spells with a given
number of past alternations.
b. Pr(continue|uncensored) = Number of spells with one more alternation/Uncensored
number of spells with a given number of past alternations. This is the probability that at least
one more alternation would occur within a particular spell.

In Chile in 1924, Costa Rica in 1948, Ecuador in 2000, Germany in 1933, and
Thailand in 2000, electoral spells ended after three alternations; in Portugal
in 1870, Sri Lanka in 1982, Norway in 1940, and Philippines in 1972 after
four; in Chile in 1973 after five; and in Italy in 1922 after six. One might want
to eliminate Italy from this list because alternations were an outcome of an
agreement between parties,27 and one might want to drop Norway because
the constitutional breakdown was accompanied by the German invasion. But
if we were to include France in 1940 (the Vichy regime), the threshold would
have been nine.28 Hence, the number of past alternations that makes the elec-
toral mechanism immune to constitutional breakdowns is somewhere
between five and nine. Nevertheless, the effect of past alternations on the
survival of the electoral mechanism is powerful.
Probit regressions of breakdowns on the number of past alternations
within the particular spell (sum past alt), the electoral history of the country
(electoral spell), and per capita income (gdpcap) or year, presented in Table 7,
confirm that past alternations within the current spell increase the likelihood
that it would continue. In concordance with Przeworski et al. (2000), past
constitutional breakdowns increase the probability that the current electoral
spell would break as well, which suggests that analyses that ignore past his-
tory suffer form an omitted variable bias.29 Finally, as in analyses of regime
transitions, higher income makes constitutional breakdowns less likely.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
120 Comparative Political Studies 48(1)

Table 7. Explaining Constitutional Breakdowns.

Variable Effecta 95% CI Effecta 95% CI

Sum past alt −0.001** [−0.003, 0.0000] −0.004*** [−0.006, −0.002]
Electoral spell 0.002*** [0.001, 0.002] 0.002*** [0.001, 0.003]
Gdpcap −0.000*** [−0.000, −0.000]
Year −0.000*** [−0.000, −0.000]
N 7,301 10,016
Spells 292 415
Lroc 0.73 0.72

Country spell clustered errors. With cubic polynomial of electoral age, not shown. lroc =
area under the receiver operating curve.
a. “Effect” is the partial derivative or difference.
**significant at 5%; *** significant at 1%.

Thus, alternations protect from constitutional breakdowns. Yet is this pro-

tection due to changes of governments or the mere fact that elections are held
or the possibility for the opposition to channel its resistance through the elec-
toral mechanism? Cox (2008) and Geddes (2009) both argue that getting
elected protects rulers who enter office by force from the ambitions of those
who place them there. The generals who ride into the presidential palace,
whether on horse or in a tank, are followed by rungs of others, and those
immediately behind them would have wanted to be the first one as well.
Elections, the argument goes, protect rulers from their closest supporters
because they demonstrate that the ruler has some basis of support, whether
among the population or just within the state apparatus, independent of sheer
force. This argument is borne by the fact that breakdowns, as defined here,
occurred in 0.0506 of the 890 years in which rulers who entered by force
were not yet elected but only in 0.0255 of the 9,842 years when incumbents,
however they entered into office, had already been elected, t = 3.33, p(t) =
.0004.30 Moreover, rulers who enter by a coup seem to know it, rushing to
hold elections as soon as they can: Of the 537 years in which at least one
coup occurred, an election followed within roughly 30 months in 58% of
such instances. Hence, holding elections of any kind does prevent coups,
civil wars, and other major constitutional irregularities. The frequency of
breakdowns is reduced even more when the opposition has a chance to par-
ticipate in elections: Breakdowns occurred in 0.0364 of the 1,937 years
spanning uncontested elections but only in 0.0217 of the 8,489 years span-
ning election in which there was opposition, t = 2.91, p(t) = .0018. Yet the
experience of having changed a government through elections just once
immunizes even more: In the electoral spells in which no alternation had yet

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 121

occurred, breakdowns occurred in 0.0343 of the 3,814 years but after a sin-
gle alternation only in 0.0115 of the 4,094 years, t = 6.75, p(t) = .0000. To
summarize these numbers, observe that a breakdown occurs on the average
once in every 19.8 years when there are no elections, once in 24.7 when elec-
tions are held without opposition, once in 46.1 years when opposition is
allowed, and once in 87.0 years after a single alternation occurred. Hence,
while elections of any kind do prevent violent conflicts and contested elec-
tions increase this protection, a single experience of having witnessed a gov-
ernment changing as a result of people’s choice provides almost perfect
immunity from political violence.

Conclusion
The main conclusion of this analysis is that competitive elections are a self-
institutionalizing mechanism. Elections are a method of processing conflicts,
one among several possible. Contending political forces channel their efforts
into electoral competition rather than engage in violence when elections offer
them real prospects of prevailing at some time in the future, when govern-
ments can change as a consequence of elections. The first experience of elec-
tions in which the incumbent rulers subject themselves to the verdict of the
people, lose, and peacefully leave office has a powerful effect on shaping the
beliefs about these prospects: It almost doubles the probability that peaceful
electoral competition will continue indefinitely. Moreover, this effect holds
independently of political, economic, or social conditions, at least of those
that could be considered. Thus, a single experience of partisan alternation in
office as a result of elections goes a long way to consolidate democracy.
While once acquired the habit of changing governments through elections
is durable, it is not easy to acquire. Letting go of power for the first time is
extremely difficult. Wading through untreaded waters, rulers do not know
what else they would lose if they lose office, how deeply they would sink.
This uncertainty is reduced with each successive alternation, so that incum-
bents become willing to expose themselves to losing elections and to yield
office when they lose. Random events still intervene to generate constitu-
tional breakdowns even when political leaders had already learned to risk
their power in elections but after some number of alternations the habit
becomes safely entrenched.
This dynamic is reflected in the fact that, as shown in Figure 3, 81 out of
204 countries that existed at any time after 1918 and 68 out of 188 observed
as of 2008 never experienced an alternation, while countries that experienced
at least one went on to repeat the experience.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
122 Comparative Political Studies 48(1)

Figure 3. Distribution of the numbers of alternations experienced by particular

countries at any time before 2009.

Figure 4. Defeats of incumbents and partisan alternations over time.

The countries where the habit became entrenched are familiar: The United
States experienced 23 alternations, the United Kingdom and the Netherlands
22, followed by Denmark, Sweden, New Zealand, Finland, Iceland, Belgium,
Costa Rica, Australia, Ireland, and France. The newcomers are the 57

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 123

countries which experienced first alternation after 1988, with 49 of these

spells lasting as of 2008 (2 of these countries disappeared in the meantime),
so that the habit has obviously become widespread. But, as shown in Figure 4,
the world in which incumbents lose elections and peacefully leave office is a
new one.
Conflict, liberty, and peace do not coexist easily. When one looks at world
history in the perspective of centuries, the practice of peacefully changing
governments by letting the people freely choose them appears as no more
than a speck. The bad news emanating from this analysis is that the first step
requires crossing an abyss. The good news is that the habit is addictive.

Appendix

Proof
When q / (1 − α) < 1 , the winner does not hold competitive elections if
’

p > q’ / (1 − α) . Hence, the first incumbent holds elections with p < 1 if

( )
pV ( w ) + (1 − p ) E q’ > E ( q ) , (8)
q q
where E ( ) = V (d ) + ( )(V ( w) − V (d )) , where q applies to the ex ante
q’ q’
and q to the ex post lottery that occurs when the current ruler does not hold
’

competitive elections.
Substitutions show that the incumbent holds competitive elections when-
ever the winner would not if
q − q’
p> ≡ p’ . (9)
1 − q’
Comparing the two thresholds yields p’ > p∗ ⇔ θα > q’ . But if
∗
p > 0, θα > q and because q > q’ , it is always true that p’ > p∗ . In turn, it
q q − q’
is apparent that 1 − α > 1 − q’ (given that this inequality is true for α = 0 ).
∗
Hence, the incumbent holds competitive elections when p < p < q / (1 − α)
and the winner reciprocates when p’ < q / (1 − α) < p and the winner does
not hold elections or assures himself of victory.

Acknowledgments
I appreciate comments by Abel Escribà, Joanne Fox-Przeworski, Bert Hoffmann,
Fernando Limongi, Frédéric Louault, Pasquale Pasquino, Gonzalo Rivero, Rubén

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
124 Comparative Political Studies 48(1)

Ruiz-Rufino, Ian Shapiro, Tianyang Xi, the editors of this journal, as well as anony-
mous reviewers.

Author’s Note
The original data and the Stata replication files are available at https://sites.google.
com/a/nyu.edu/adam-przeworski/

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research,
authorship, and/or publication of this article.

Funding
The author received no financial support for the research, authorship, and/or publica-
tion of this article.

Notes
1. Wright and Escribà Folch (2012) count that between 1946 and 2004 of the 199
rulers of autocratic leaders who lost office when regimes became democratic,
93 met with violent death, imprisonment, or exile. Geddes, Wright, and Frantz
(2014) report that of the 100 transitions to democracy between 1946 and 2010, in
40 cases the outgoing rulers suffered death, jail, or exile. These counts include,
however, the cases in which the incumbents were overthrown by force. Goemans
(2008), in turn, distinguishes leaders who lost office according to the rules of any
regime, among whom 8% were killed, jailed, or exiled within 1 year, and those
who left under the threat or the use of force, among whom 80% suffered this fate.
Counting only the 28 cases in which autocratic incumbents lost elections, only 3
were punished within 1 year (Abel Escribà, personal communication, April 24,
2014.), but some were jailed later.
2. Nalepa (2010) observes that actual punishments were rare after the fall of com-
munism. Nevertheless, many people were and still are prosecuted, temporarily
deprived of political rights, or found unfit to occupy public positions.
3. With minor exception of instances where political parties agreed to a unique can-
didate (as in Chile in 1891). Elections in which there was one organized party and
independent candidates (as in Portugal in 1954) are not considered contested.
4. Electoral spells must experience at least two consecutive elections in which the
winner of the first one assumed and completed the term in office, regardless
of what happened after the second election. Included in the analysis that fol-
lows are all spells that began after a country became independent and those ini-
tiated before independence if at least two consecutive elections occurred after
independence.
5. For example, in Malawi, Dr. Hastings Banda was elected in 1966 and upon being
reelected in 1970 proclaimed himself president for life, not holding elections
until 1994, when he presented himself and lost.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 125

6. Note that this definition biases upward the count of events considered as alterna-
tions because it includes instances in which the governing coalition remained
the same but the person and the party of the prime minister changed. In parlia-
mentary systems, partisan control often changes hands between elections. Such
events are not considered alternations here.
7. Changes between non-partisan, “caretaker” or “technical” governments, and par-
tisan ones are not considered alternations.
8. There were 18 instances in which victorious opposition assumed office but only
after either the defeated incumbents attempted to hold onto power or some third
party attempted to squeeze into office: they are not counted as alternations. Note
that alternations are dated in the year in which the relevant election occurred, not
the year new government assumed office.
9. For simplicity, I ignore the possibility that a ruler elected in competitive elections
may also be deposed by force. While such events do occur, holding an election
sharply reduces the probability of coups. See below.
10. The most recent anecdote is provided by the October 6, 2013, election in
Azerbaijan, where election results were inadvertently released before voting
took place ( www.washingtonpost.com/blogs/worldviews/wp/2013/10/09/). For
different “menus of manipulation,” see Posada-Carbó (2000), Lehoucq (2002),
Mozaffar and Schedler (2002), and Hyde (2011).
11. This assumption is buttressed by the fact that predicted probabilities of incum-
bents winning are bimodal, with one peak around p = .65 and the second around
p = .95, regardless of the specification of the regression equation.
12. One could assume that the incumbent and the potential opposition update their
beliefs about p and θ depending on the outcomes of elections, with θ adjust-
ing gradually rather than jumping to 1. I do not develop this model because the
results would be qualitatively the same.
13. Posada-Carbó (2008) points out the contrast with the events in Venezuela 2 years
earlier: “Although in Venezuela President Páez’s favourite, General Soublette,
also lost at the polls and Páez handed in power to the victor, Dr José M. Vargas
-a civilian-, the latter was ousted seven months later, when Páez returned to
the presidency and then practically ruled Venezuela for the next two decades”
(p. 31).
14. Geddes et al. (2014) as well as Escribà Folch (2014) show that the proportion
of autocratic rulers who were punished in various ways differs depending on
whether the autocracy was monarchic, personalist, military, or single-party.
15. Of the 32 constitutional monarchies, only 9 survived by allowing alternations:
Belgium, Denmark, Japan (where the Emperor lost even nominal executive
power in 1947), Lesotho (where the monarch lost executive power in 1993),
the Netherlands, Norway, Sweden (where the monarch lost executive power in
1974), Thailand, and the United Kingdom.
16. Interestingly, in 1935, Stalin wanted elections to be direct, based on universal
suffrage, secret, and open to competition, arguing that because class enemies had
been eradicated, the Communists have nothing to fear from contested elections.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
126 Comparative Political Studies 48(1)

The intra-party elections in 1936 took place under these rules but the communist
cadres persuaded Stalin that popular elections were too dangerous and the 1937
elections were uncontested (see Getty, 1991; Schlögel, 2014).
17. A direct passage from uncontested elections to competition in which they incum-
bent had presented itself, lost, and yielded office has been rare. Elections had
been previously contested in all the 36 monarchies that experienced the first
alternation, in 57 out of 62 parliamentary republics, and in 50 out of 61 presi-
dential systems, in total in 143 cases, 92% of 156 first transitions.Note again that
if the party that ruled unopposed does not present itself in the first competitive
election, there can be no alternation: by definition, an alternation can occur only
if the incumbent competes in an election.
18. The reasoning here, based on the ideas of Przeworski (2005) and Benhabib and
Przeworski (2006), is that if utility is concave in income, losing income matters
less to the losers when the country income is high. Specifically, if the utility of
1− σ
income is Constant Relative Risk Aversion (CRRA), U ( yi ) = yi , σ > 1,
1− σ
and winners obtain a larger multiple of average income than losers, the utility
difference between winning and losing declines in average income. The best
illustrations of the role of income are the elections in Costa Rica in 1948 and in
the United States in 2000: In both, the outcome was a technical draw, after which
the Congress in Costa Rica and the Supreme Court in the United States awarded
the victory to one of the candidates. The result in Costa Rica was a bloody civil
war, while in the United States supporters of the loser just drove their SUVs
back home to cultivate their gardens. The difference was that they had SUVs
and gardens.
19. In Latin America, during most of the 19th century, the “military” were landown-
ers calling themselves “generals,” who rode into presidential palaces accom-
panied by their peones, whom they called “soldiers,” but who never followed
institutionalized military careers. I do not know if that makes them “personalistic
militaries,” so I prefer to stay away from this term.
20. For example, Lt. Col. Mathieu Kérékou grabbed office by a coup in Benin in
1972 and had himself elected and reelected without opposition several times
until 1991, when he held a competitive election, lost, and yielded office, only to
be elected again in 1996 and reelected in 2001. He is classified as having entered
office by force in 1972 and constitutionally in 1996.
21. These analyses have been replicated with a random effects logit estimator. The
results are not reported because they are almost identical.
22. The average within-country correlation of year and gdpcap is .57.
23. I also examined the role of the extent of suffrage, measured alternatively on
an ordinal scale or by the proportion of the population eligible to vote, to find
that it has no statistically significant effect. Finally, expecting that the incum-
bent would have more to lose in unequal societies, I performed heroic efforts
to introduce inequality (measured in the earlier period by proportion of family
farms—from Vanhanen, 1997) and in the later period by interpolated values of

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 127

Gini coefficients (from World Development Indicators 2009), as well as a com-

bination of the two), but to no avail: Inequality never matters.
24. Almost all elections that followed the first alternation were contested (except
some in Liberia and the 1942 one in Colombia), so this variable had to be
dropped from the analysis. In turn, the number of past alternations, at least one,
is added to the specification.
25. The analysis was conducted without and with the variable that measures the
number of years since the past alternation and with and without controlling for
selection bias that may be due to the fact that the first alternation had occurred
in the particular spell. The detailed results are not worth reporting. They do not
differ between the samples including gdpcap (n = 754, with 99 country spells)
and year (n = 951, with 126 country spells). Note that even per capita income
does not matter.
26. A spell is considered to have ended in a breakdown if (a) there was a coup,
autocoup (including extensions of the current term in office), or a civil war that
resulted in a change of the chief executive, or (b) the winner of the election
did not directly assume office in violation of constitutional provisions (in some
monarchies the monarch was constitutionally empowered to appoint govern-
ments regardless of the result of an election), or (c) the legislature was closed
unconstitutionally.
27. The difference between tranformismo in Italy and rotativismo in Spain was
that in the former governments changed after elections, while in the latter they
changed before elections and were then ratified in elections. But the practice was
otherwise the same.
28. The difference between Norway and France is that Quisling made a coup the day
Germany invaded but not yet occupied the country while the Vichy regime was
installed only after France was defeated.
29. Almost all studies of regime transitions begin after World War II and do not con-
sider previous history. The bias that results is shown by the fact that the positive
effect of per capita income on the probability of transitions to democracy (as in
Boix & Stokes, 2003) disappears when previous experiences with democracy
are included in a specification. This effect also disappears with fixed effects (as
in Acemoglu, Johnson, Robinson, & Yared, 2008, 2009), and I suspect that these
effects just mask past history.
30. All the t tests assume unequal variances.

References
Acemoglu, D., Johnson, S., Robinson, J. A., & Yared, P. (2008). Income and democ-
racy. American Economic Review, 98, 808-842.
Acemoglu, D., Johnson, S., Robinson, J. A., & Yared, P. (2009). Reevaluating the
modernization hypothesis. Journal of Monetary Economics, 56, 1043-1058.
Banks, A. (2005). Cross-National Time Series Data Archive. Retrieved from www.
databanksinternational.com
Benhabib, J., & Przeworski, A. (2006). The political economy of redistribution under
democracy. Economic Theory, 29, 271-290.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
128 Comparative Political Studies 48(1)

Boix, C., & Stokes, S. C. (2003). Endogenous democratization. World Politics, 55,
517-549.
Bushnell, D. (1993). The making of modern Colombia. Berkeley: University of
California Press.
Carter, D. B., & Signorino, C. S. (2010). Back to the future: Modeling time depen-
dence in binary data. Political Analysis, 18, 271-292.
Cercas, J. (2011). The anatomy of a moment. New York, NY: Bloomsbury.
Collier, S., & Sater, W. F. (1996). A history of Chile, 1808-1994. Cambridge, UK:
Cambridge University Press.
Cox, G. W. (2008). Authoritarian elections and leadership succession, 1975–2000.
Unpublished paper, Department of Political Science, University of California,
San Diego.
Dunn, S. (2004). Jefferson’s second revolution: The election crisis of 1800 and the
triumph of republicanism. Boston, MA: Houghton Mifflin.
Escribà Folch, A. (2014). Accountable for what? Regime types, performance, and the
fate of outgoing dictators, 1946–2004. Democratization, 20, 160-185.
Fearon, J. (2011). Self-enforcing democracy. Quarterly Journal of Economics, 126,
1661-1708.
Geddes, B. (2005, September). Why parties and elections in authoritarian regimes?
Presented at the Annual Meeting of the American Political Science Association,
Washington, DC.
Geddes, B. (2009, September). How autocrats defend themselves against armed
rivals. Paper presented at the Annual Meeting of the American Political Science
Association, Toronto, Ontario, Canada.
Geddes, B., Wright, J., & Frantz, E. (2013). Autocratic regimes code book. Available
from dictators.la.psu.edu
Geddes, B., Wright, J., & Frantz, E. (2014). New data set: Autocratic breakdown and
regime transitions. Perspectives on Politics, 12, 305-309.
Getty, J. A. (1991). State and society under Stalin: Constitutions and elections in the
1930s. Slavic Review, 50, 18-35.
Goemans, H. E. (2008). Which way out? The manner and consequences of losing
office. Journal of Conflict Resolution, 52, 771-794.
Hyde, S. D. (2011). The pseudo-democrat’s dilemma: Why election monitoring
became an international norm? Ithaca, NY: Cornell University Press.
Lehoucq, F. (2002). Can parties police themselves? Electoral governance and democ-
ratization. International Political Science Review, 23, 29-46.
Maddison, A. (2010). Statistics on world population, GDP and per capita GDP,
1-2008 AD. Retrieved from www.ggdc.net/maddison/maddison-project/home.
htm
Makarenko, B. (in press). The role of elections in democracy. In A. Przeworski (Ed.),
Democracy in a Russian mirror. New York, NY: Cambridge University Press.
Mozaffar, S., & Schedler, A. (2002). The comparative study of electoral gover-
nance—Introduction. International Political Science Review, 23, 5-27.
Nalepa, M. (2010). Skeletons in the closet: Transitional justice in post-communist
Europe. New York, NY: Cambridge University Press.

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
Przeworski 129

Posada-Carbó, E. (2000). Electoral juggling: A comparative history of the corruption

of suffrage in Latin America, 1830-1930. Journal of Latin American Studies, 32,
611-644.
Posada-Carbó, E. (2008, September 18-21). Historical sequences and the origins of
democracy in Colombia, 1808–1886. Draft for presentation at the Conference
“Origins of Democracy in the Americas, 1770s-1870s,” Kellogg Institute,
University of Notre Dame, IN.
Przeworski, A. (1991). Democracy and the market. New York, NY: Cambridge
University Press.
Przeworski, A. (2005). Democracy as an equilibrium. Public Choice, 123, 253-273.
Przeworski, A., Alvarez, M. R., Cheibub, J. A., & Limongi, F. (2000). Democracy
and development. New York, NY: Cambridge University Press.
Przeworski, A., Asadurian, T., & Thomas, A. B. (2012). The origins of parliamentary
responsibility. In T. Ginsburg (Ed.), Comparative constitutional design (pp. 101-
137). New York, NY: Cambridge University Press.
Przeworski, A., Rivero, G., & Xi, T. (2014). Elections as a method of processing
conflicts (Working paper, June). Department of Politics, New York University,
New York.
Przeworski, A. (2013). Political Institutions and Political Events (PIPE) [Data set].
Retrieved from https://sites.google.com/a/nyu.edu/adam-przeworski/
Rouquie, A. (1994). The military in Latin American politics since 1930. In L. Bethell
(Ed.). The Cambridge history of Latin America, Volume VI, Part 2: 1930 to the
present (pp. 233-306). Cambridge, UK: Cambridge University Press.
Schlögel, K. (2014). Moscow 1937. Boston, MA: Polity Press.
Simpser, A. (2013). More than winning: Why governments and parties manipulate
elections. New York, NY: Cambridge University Press.
Sparks, A. (1994). Tomorrow is another country. Johannesburg, South Africa: Struik.
Vanhanen, T. (1997). Prospects of democracy: A study of 172 countries. London,
England: Routledge.
Weisberger, B. A. (2000). America afire: Jefferson, Adams, and the first contested
election. New York, NY: HarperCollins.
Wright, J., & Escribà Folch, A. (2012). Authoritarian institutions and regime survival:
Transitions to democracy and subsequent autocracy. British Journal of Political
Science, 42, 283-309.

Author Biography
Adam Przeworski is the Carroll and Milton Petrie Professor of Politics at New York
University. His most recent book is Democracy and the Limits of Self-Government
(Cambridge University Press, 2011).

Downloaded from cps.sagepub.com at Bobst Library, New York University on November 13, 2015
View publication stats