t.1 Przewoski-Democracy

1.
Democracy
Democracy
In his opening speech to the Constituent Assembly, Adolfo Suarez, the
prime minister of the Spanish transition to democracy, announced that
henceforth "the future is not written, because only the people can write it"
(Verou 1976). Heralding this plunge into the unknown, he caught two
quintessential features of democracy: Outcomes of the democratic process
are uncertain, indeterminate ex ante; and it is "the people," political forces
competing to promote their interests and values, who determine what these
outcomes will be.
Democracy is a system in which parties lose elections.1 There are par-
ties: divisions of interests, values, and opinions. There is competition,
organized by rules. And there are periodic winners and losers. Obviously
not all democracies are the same; one can list innumerable variations and
distinguish several types of democratic institutions. Yet beneath all the
institutional diversity, one elementary feature - contestation open to par-
ticipation (Dahl 1971) - is sufficient to identify a political system as
democratic.2
Democracy is, as Linz (1984) put it, government pro tempore. Conflicts
1
Note that the presence of a party that wins elections does not define a system as
democratic: The Albanian People's party has regularly produced overwhelming victories. It is
only when there are parties that lose and when losing is neither a social disgrace (Kishlansky
1986) nor a crime that democracy flourishes.
2
Most definitions of democracy, including Dahl's own, treat participation on a par with
contestation. Indeed, there are participationist and contestationist views of democracy. The
emphasis on participation is essential if one wants to understand the development of democ-
racy in Western Europe, where battles over suffrage evoked more conflicts than the issue of
governmental responsibility. Moreover, such an emphasis is attractive from the normative
point of view. Yet from the analytical point of view, the possibility of contestation by
conflicting interests is sufficient to explain the dynamic of democracy. Once political rights
are sufficiently extensive to admit of conflicting interests, everything else follows, even if
effective participation is far from universal. And since, except in South Africa, broad re-
strictions of political rights are inconceivable under present conditions, a focus on contesta-
tion is sufficient to study current transitions to democracy.
Downloaded from https://www.cambridge.org/core. Universidad Complutense de Madrid, on 26 Apr 2021 at 14:39:06, subject to
the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/CBO9781139172493.003
DEMOCRACY 11
are regularly terminated under established rules. They are "terminated"

(Coser 1959), temporarily suspended, rather than resolved definitively.
Elections fill offices, legislatures establish rules, bureaucracies issue deci-
sions, associations arrive at agreements, courts adjudicate conflicts, and
these outcomes are binding until and unless they are altered according to
rules. At the same time, all such outcomes are temporary, since losers do
not forfeit the right to compete in elections, negotiate again, influence
legislation, pressure the bureaucracy, or seek recourse to courts. Even
constitutional provisions are not immutable; rules, too, can be changed
according to rules.
In a democracy, multiple political forces compete inside an institutional
framework. Participants in the democratic competition dispose unequal
economic, organizational, and ideological resources. Some groups have
more money than others to use in politics. Some may have more extensive
organizational skills and assets. Some may have greater ideological means,
by which I mean arguments that persuade. If democratic institutions are
universalistic - blind to the identity of the participants - those with greater
resources are more likely to win conflicts processed in a democratic way.3
Outcomes, I am arguing, are determined jointly by resources and institu-
tions, which means that the probability that any group, identified by its
location in the civil society, will realize its interests to a specific degree and
in a particular manner is in general different from any other group's.
The protagonists in the democratic interplay are collectively organized;
that is, they have the capacity to formulate collective interests and to act
strategically to further them (Pizzorno 1978). Furthermore, they are orga-
nized in a particular way entailed in the institutional framework within
which they act. To represent, political parties must be stratified into leaders
and followers; by definition, representative institutions seat individuals,
not masses. A relation of representation is thus imposed on the society by
the very nature of democratic institutions (Luxemburg 1970: 202). Indi-
viduals do not act directly in defense of their interests; they delegate this
defense. Masses are represented by leaders; this is the mode of collective
organization in democratic institutions.4 Moreover, as Schmitter (1974),
3
This is not to say that institutions are not biased. Institutions have distributional conse-
quences. Much more on this topic will follow.
4
Note that social movements are an ambiguous actor under democracy, and always short-
lived. Unions have a place to go: industrial relations institutions and the state; parties have
parliaments; and lobbies have bureaus; but movements have no institutions to direct them-
selves to.
https://doi.org/10.1017/CBO9781139172493.003
12 DEMOCRACY
Stepan (1978), Offe (1985), and others have insisted, most interests are
organized in a coercive and monopolistic fashion. Interest associations
acquire the capacity to act on behalf of their members because they can
coerce these members, specifically because they can sanction any indi-
viduals or subgroups who attempt to advance their particular goals at the
cost of the collective interest. To have market power, unions must be able
to punish workers who are eager to replace their striking colleagues; to
have a strategic capacity, employers' associations must be able to control
the competition among firms in the particular industry or sector. Democrat-
ic societies are populated not by freely acting individuals but by collective
organizations that are capable of coercing those whose interests they repre-
sent.
Democracy is a system of processing conflicts in which outcomes de-
pend on what participants do but no single force controls what occurs.
Outcomes of particular conflicts are not known ex ante by any of the
competing political forces, because the consequences of their actions de-
pend on actions of others, and these cannot be anticipated uniquely. Hence,
from the point of view of each participant, outcomes are uncertain: Democ-
racy appears to be a system in which everyone does what he or she expects
is for the best and then dice are thrown to see what the outcomes are.
Democracy generates the appearance of uncertainty because it is a system
of decentralized strategic action in which knowledge is inescapably local.
The fact that uncertainty is inherent in democracy does not mean every-
thing is possible or nothing is predictable. Contrary to the favorite words of
conservatives of all kinds, democracy is neither chaos nor anarchy. Note
that "uncertainty" can mean that actors do not know what can happen, that
they know what is possible but not what is likely, or that they know what is
possible and likely but not what will happen.5 Democracy is uncertain only
in the last sense. Actors know what is possible, since the possible out-
comes are entailed by the institutional framework;6 they know what is
likely to happen, because the probability of particular outcomes is deter-
5
These distinctions are based on Littlechild 1986.
6
I mean "know" in the logical sense: They have the information from which they can
deduce each consequence. They can deduce it because the possible outcomes are entailed by
rules, and rules can change only according to rules. The "institutional framework," under-
stood as the entire system of rules, is not fixed; it is repeatedly modified as a result of
conflicts. But these conflicts always occur within a system of rules that delimit the feasible
set. Obviously, none of the above implies that political actors always know what is possible in
the psychological sense: They err and they are surprised, particularly because the logical
relations involved are often "fuzzy."
https://doi.org/10.1017/CBO9781139172493.003
DEMOCRACY 13
mined jointly by the institutional framework and the resources that the
different political forces bring to the competition. What they do not know
is which particular outcome will occur. They know what winning or losing
can mean to them, and they know how likely they are to win or lose, but
they do not know if they will lose or win. Hence, democracy is a system of
ruled open-endedness, or organized uncertainty.
The uncertainty inherent in democracy does permit instrumental action.
Since actors can attach probabilities to the consequences of their actions,
they form expectations and calculate what is best for them to do. They can
participate, that is, act to advance their interests, projects, or values within
the democratic institutions. Conversely, since under the shared constraints
outcomes are determined only by actions of competing political forces,
democracy constitutes for all an opportunity to pursue their respective
interests. If outcomes were either predetermined or completely indetermi-
nate, there would be no reason for groups to organize as participants. It is
the uncertainty that draws them into the democratic interplay.
Results of democratic processes are read by applying the particular rules
that make up the institutional framework to the joint consequences of
decentralized actions. Yet in spite of its majoritarian foundations, modern
representative democracy generates outcomes that are predominantly a
product of negotiations among leaders of political forces rather than of a
universal deliberative process. The role of voting is intermittently to ratify
these outcomes or to confirm in office those who brought them about.7 In
all modern democracies, the deliberative process and day-to-day supervi-
sion over the government are well protected from the influence of the
masses. Indeed, a direct recourse to voters about specific policy issues is
often referred to as plebiscitarianism, a term with negative connotations.
Hence, voting - majority rule - is only the ultimate arbiter in a democracy.
Outcomes consist of indications to each political force to follow specific
courses of action, different for winners and losers. If these indications are
followed, losers get less of what they want than winners. To follow these
indications is to comply.
Because outcomes cannot be predicted exactly under democracy, com-
7
As Bobbio (1989: 116) put it, "collective decisions are a fruit of negotiation and agree-
ments between groups which represent social forces (unions) and political forces (parties)
rather than an assembly where voting operates. These votes take place, in fact, so as to adhere
to the constitutional principle of the modern representative state, which says that individuals
and not groups are politically relevant . . .; but they end up possessing the purely formal
value of ratifying decisions reached in other places by the process of negotiation."
https://doi.org/10.1017/CBO9781139172493.003
14 DEMOCRACY
mitment to rules need not be sufficient for compliance once the results are
known. If outcomes were certain, that is, if participants could predict them
uniquely, they would have known that in committing themselves to par-
ticular rules they were accepting specific outcomes; commitment to
rules would have been sufficient for compliance with results. Yet under
democracy commitment to rules constitutes at most a "willingness to
accept outcomes of an as yet undetermined content" (Lamounier 1979:
13). This is why procedural evaluations of democracy diverge from conse-
quentialist judgments. As Coleman (1989: 197) put it, "consenting to a
process is not the same thing as consenting to the outcomes of the pro-
cess." Since outcomes are uncertain for the participants, their ex ante and
ex post evaluations must diverge. And, as Lipset and Habermas agree, ex
post evaluations modify the ex ante commitments.8 Hence, compliance is
problematic.
In sum, in a democracy all forces must struggle repeatedly for the
realization of their interests. None are protected by virtue of their political
positions.9 No one can wait to modify outcomes ex post; everyone must
subject interests to competition and uncertainty. The crucial moment in any
passage from authoritarian to democratic rule is the crossing of the thresh-
old beyond which no one can intervene to reverse the outcomes of the
formal political process. Democratization is an act of subjecting all in-
terests to competition, of institutionalizing uncertainty. The decisive step
toward democracy is the devolution of power from a group of people to a
set of rules.
8
Lipset (1960) makes the distinction between "legitimacy" - ex ante commitment - and
"effectiveness" - ex post evaluation of outcomes. Habermas (1975) distinguishes "legality"
- ex ante acceptance of rules - and "legitimacy" - for him, the ex post evaluation. Both
maintain that ex post evaluations modify ex ante commitments, but neither notices that the
very problem of compliance arises only because the outcomes generated by rules are uncertain
ex ante.
9
Some interests, notably of those who own productive resources, may be protected by
their structural position in the economy: If everyone's material welfare depends on the
decisions of capitalists to employ and to invest, all governments may be constrained from
adopting policies that lower employment and investment. This is the theory of the structural
dependence of the state on capital. The controversial question is whether this dependence is so
binding on all democratically elected governments that the democratic process can have no
effect on the policies followed by governments. My view is that all governments are to some
degree dependent on capital but that this dependence is not so binding as to make democracy a
sham. There is room for the democratic process to affect the outcomes. See Przeworski and
Wallerstein 1988 for a formal analysis of this theory.
https://doi.org/10.1017/CBO9781139172493.003
HOW ARE OUTCOMES ENFORCED? 15
How are outcomes enforced under democracy?
The question: democracy, rationality, and compliance

With these preliminaries, we are ready to pose the central question con-
cerning the durability of democracy: How does it happen that political
forces that lose in contestation comply with the outcomes and continue to
participate rather than subvert democratic institutions? Suppose a govern-
ment seeks to establish control over the military. Why would the military
obey? Imagine that a legislature passes a law granting workers extensive
rights within enterprises. Why would the bourgeoisie not defend property
by antidemocratic means? Envisage a government policy that causes mas-
sive unemployment and widespread impoverishment. Why would the poor
not take to the streets to overturn it? Why would they all continue to
channel their actions via the democratic institutions that hurt their in-
terests? Why would they comply?
To understand why these questions matter, we need first to clear away
some underbrush. If democracy were rational in the sense of eighteenth-
century democratic theory, the problem of compliance would not emerge at
all, or at least it would assume a different form. If societal interests were
harmonious - the central assumption of the democratic theory of the eigh-
teenth century - conflicts would be but disagreements about identifying the
common good. They could be overcome by rational discussion: The role of
the political process would be only epistemic, a search for the true general
will. Politics, Wood (1969: 57-8) noted concerning American political
thought between 1776 and 1787, "was conceived to be not the reconciling
but transcending of the different interests of the society in the search for the
single common good." If representatives could free themselves from the
passion of particular interests, if institutions were properly designed, and if
the process of deliberation were sufficiently unhurried, unanimity would
prevail - the process would have converged to the true general will. Even
today some theorists see recourse to voting as only a time-saving device:
Voting merely economizes on the transaction costs inherent in delibera-
tion. 10 In this view, as Coleman (1989: 205) characterized it, "the minority
10
Summarizing with approval the views of Buchanan and Tullock (1962), Brennan and
Lomasky (1989; 3) present the argument as follows: "If the rule of unanimity were also
employed at the postconstitutional level, such that each individual possessed an effective veto
https://doi.org/10.1017/CBO9781139172493.003
16 DEMOCRACY
does not consist of losers, and the majority winners. Instead, minority
members have false beliefs about the general will; members of the majority
have true beliefs."
Is democracy in any sense rational?11 Democracy would be collectively
rational in the eighteenth-century sense if (1) there exists some unique
welfare maximum over a political community: common good, general
interest, public interest, and the like (Existence); (2) the democratic pro-
cess converges to this maximum (Convergence). Moreover, democracy
would be superior to all its alternatives if (3) the democratic process is the
unique mechanism that converges to this maximum - no benevolent dic-
tator could know what is in the general interest (Uniqueness).
The question whether democracy is rational in this sense evokes five
distinct responses, depending on whether (1) (a) such a welfare maximum
is thought to exist prior to and independent of individual preferences, (b) it
is thought to exist only as a function of individual preferences, whatever
these might happen to be, or (c) it is thought not to exist at all, because of
class or some other irreconcilable division of society; and whether (2) the
democratic process is thought to converge to this maximum. Rousseau
believed that general interest is given a priori and that the democratic
process converges to it. Conservatives in France and England at the time of
the French Revolution, as well as contemporary ideologists of various
authoritarianisms, maintain that such a welfare maximum does exist but
that the democratic process does not lead to it. Economic theorists of
democracy, notably Buchanan and Tullock (1962), have maintained that
the public interest is tantamount to the verdict of the democratic process,
which does identify it. Arrow (1951) demonstrated, under some assump-
tions, that even if such a maximum does exist, no process of aggregating
individual preferences will reveal it. Finally, Marx and his socialist fol-
lowers argued that no such general interest can be found in societies
divided into classes. Note that Schmitt (1988: 13, 6) simultaneously sided
over every collective determination, exorbitant bargaining costs would ensue. . . . Balloting
thus emerges as an efficiency-enhancing device itself resting on a foundation that eschews
majoritarianism."
11
To follow distinctions made by economists, we might first distinguish technical from
collective rationality. Democracy would be said to be technically rational if it effectively
served some otherwise desirable objectives, such as promoting economic development, or (a
view to which I adhere) minimized arbitrary violence. But in the present discussion our
interest is in the notion of collective, rather than technical, rationality.
https://doi.org/10.1017/CBO9781139172493.003
with Marx when he rejected Rousseau's assumption that "a true state . . .
only exists where the people are so homogeneous that there is essentially
unanimity" and attacked convergence when he observed that "the develop-
ment of modern mass democracy has made argumentative public discus-
sion an empty formality."
Recent discussions focus on the issue of convergence. In the light of
social choice theory, as argued particularly by Riker (1982), the democratic
process would not converge to a unique welfare maximum even if one
existed. The reasons are those offered by Arrow (1951): There is no pro-
cedure for aggregating preferences that would guarantee a unique outcome.
Hence, one cannot read voting results as identifying any unique social
preference. Moreover, McKelvey (1976) demonstrated that voting results
may be collectively suboptimal. Yet this view of the democratic process
relies on a tacit assumption that individual preferences are fixed and ex-
ogenous to the democratic process. Economists take preferences as fixed
and adjustment to equilibrium as instantaneous; this is why many of them
consider the democratic process as "rent seeking," that is, a waste of
resources (see, for example, Tollison 1982).
Yet the assumption that preferences are exogenous to the democratic
process is patently unreasonable. As Schumpeter (1950: 263) observed,
"the will of the people is the product, not the motive power of the political
process." Democracy may still discover or define the social welfare max-
imum if preferences change as a result of communication. Deliberation is
the endogenous change of preferences resulting from communication.12
The question, then, is whether deliberation leads to convergence.
Habermas and Joshua Cohen (1989) think it does. Their assumptions
are, however, too strong to be realistic. They have to claim that (1) the
messages are true or false, (2) people will accept the truth when confronted
with it, and (3) messages are issued in a disinterested way. The last as-
sumption is most dubious: If people behave strategically in pursuit in their
interests, they also emit messages in this way. But even if these assump-
12
To make this discussion less abstract, imagine that three young ladies venture to buy ice
cream, with enough money to buy only one flavor. Their initial preferences are respectively C
> V > S > N , V > S > C > N , S > C > V > N , where C stands for chocolate, V for
vanilla, S for strawberry, and N for none, and > should be read as "prefers over." Now,
suppose that the chocolate fan is told that this flavor leaves indelible spots on her dress.
Having received this information, she alters her preference, relegating chocolate to second
place, from C > V > S > N t o V > C > S > N . This is deliberation.
https://doi.org/10.1017/CBO9781139172493.003
18 DEMOCRACY
tions are granted, it does not follow that there is only one truth. The first
two assumptions may not suffice to lead the process to a unique welfare
maximum.13
In turn, Manin (1987), who offered a more realistic description of the
way deliberation works, concluded that deliberation stops short of con-
vergence to a unique maximum. In his view, deliberation educates prefer-
ences and makes them more general: It leads to the broadest agreement
possible at a particular time. But it stops there, leaving conflicts unre-
solved. Indeed, it is not apparent whether or not the intensity of conflicts is
reduced by Manin's process of deliberation. Perhaps conflicts between two
groups that are educated to believe that their interests are opposed are more
difficult to resolve than conflicts among fragmented "wanton" desires, to
use a term of Hirschman's (1985). After all, this was precisely socialists'
understanding of the deliberative process. In their view, this process leads
to a recognition of class identity and results in class conflict that cannot be
resolved by deliberation (see Przeworski and Sprague 1986).
Indeed, the coup de grace against theory of democracy as rational delib-
eration was administered in 1923 by Schmitt (1988), who argued that not
all political conflicts can be reconciled by discussion.14 At some point,
reasons and facts are exhausted, yet conflicts remain. At this point,
Schmitt observed, issues are decided by voting, which is an imposition of
one will upon a resisting will. From this observation, he concluded that
conflicts can be resolved only by recourse to physical force: Politics is an
antagonistic relation between "us" and "them" in which the ultimate
arbiter is violence.
The puzzle is thus the following. If one accepts, as I do, that not all
conflicts can be resolved by deliberation and that therefore democracy
generates winners and losers, can one ever expect the losers to comply with
the verdict of democratically processed conflicts? Why would those who
13
Go back to ice cream. Suppose that in response to the message about chocolate, the
strawberry devotee informs others that vanilla makes one fat. In turn, the vanilla lover notes
that strawberry contains red dye number 5, which causes cancer. Suppose further that all the
rational arguments are exhausted by these messages. Then the preferences that result from
rational deliberation may still cycle. Democracy will have educated the participants but will
not have led to a unique solution.
14 "Parliament," Schmitt (1988: 4-5) argued, "is in any case only 'true' as long as public
discussion is taken seriously and implemented. 'Discussion' here has a particular meaning
and does not simply mean negotiation. . . . Discussion means an exchange of opinion that is
governed by the purpose of persuading one's opponent through the argument of the truth or
justice of something, or allowing oneself to be persuaded of something as true and just."
https://doi.org/10.1017/CBO9781139172493.003
suffer as the result of the democratic interplay not seek to subvert the
system that generates such results?
Interests are often in conflict. Hence, there are winners and losers, and
compliance is always problematic. Yet Schmitt drew a conclusion that is
too strong because he failed to understand the role of institutions.15 Demo-
cratic institutions render an intertemporal character to political conflicts.
They offer a long time horizon to political actors; they allow them to think
about the future rather than being concerned exclusively with present out-
comes. The argument I develop below is the following: Some institutions
under certain conditions offer to the relevant political forces a prospect of
eventually advancing their interests that is sufficient to incite them to
comply with immediately unfavorable outcomes. Political forces comply
with present defeats because they believe that the institutional framework
that organizes the democratic competition will permit them to advance
their interests in the future.
Competing views of compliance

Before this argument is developed, it may be helpful to consider alternative
views of compliance.16
Think of democracy in the following way; To advance their interests, all
have to get past a particular intersection by any means of locomotion they
can put their hands on. Some people always arrive from the east; others
always from the south. Once they do arrive, a random device chooses the
lights: green is a signal to advance, red to wait.17 The probability of getting
the signal to pass or the signal to stop depends on the direction from which
one comes and the way the lights are set. If the lights are green in the east-
15
Indeed, his contemporary polemicist had already pointed out that Schmitt "has by no
means proven that Europe is confronted by the dilemma: parliamentarism or dictatorship.
Democracy has many other organizational possibilities than parliamentarism" (Thoma 1988:
81).
16
The question I pose is an empirical one: What are the conditions concerning the institu-
tions and the circumstances under which they operate that make political forces comply with
the outcomes of the democratic process and hence cause democracy to endure? There is an
enormous philosophical literature concerning moral justifications of democracy, in particular
of the coercion applied to force compliance. Since philosophers tend to confuse their nor-
mative opinions with reality, one often reads that democracy "is" this or that, rather than that
it would be this or that if people were guided by the morality of the particular author. While
some distinctions introduced in this literature clarify the issues, I find it largely irrelevant to
the empirical question at hand.
17
This allegory is derived from Moulin (1986: ch. 8).
https://doi.org/10.1017/CBO9781139172493.003
20 DEMOCRACY
west direction 80 percent of the time, those coming from the east have a
good chance to advance. If they are coming from the south, they are likely
to be told to wait. But if the lights are green 80 percent of the time in the
south-north direction, the situation is reversed. Hence, the likely outcome
depends on where one is coming from and on how the lights are set: the
resources that participants bring to the democratic competition and the
institutional framework within which they compete.
What will happen at any particular moment is uncertain in the sense
specified above: Actors know that the possible outcomes are the four
combinations of advance and wait, and they know the probability that the
light will be green or red (depending on where they are coming from) and
hence the probabilities of the two equilibrium outcomes, but they do not
know whether they will pass unobstructed or wait while others pass.
Suppose that participants obey the light. They pass alternatively, avoid-
ing collisions.18 Why do they do it? Why does a big car not force its way
through the intersection despite the signal?
Three alternative answers to this question are plausible. One is that
compliance is spontaneous - decentralized and voluntary. The second is
that there is a policeman at the intersection ready to send back to the end of
the queue anyone who tries to barge through out of turn. The last answer is
that people observe their turn because they are motivated by a moral
commitment to this social order even when it is not in their interest and
even when there is no one to punish them.
Elementary game theoretic terminology helps to flesh out these pos-
sibilities. Let us distinguish three classes of outcomes of strategic situa-
tions.
(1) Spontaneous self-enforcing outcomes, or equilibria. Each actor does

what is best for her given what others (would) do. A car arrives at the
intersection from the south. The driver looks around and comes to the
conclusion that it is her turn to wait. She arrives at this conclusion because
she thinks that drivers coming from the east expect to pass. Her mental
18
These are the two outcomes that will occur if everyone complies with the signals. The
purpose of the institution of traffic lights is to eliminate the collectively suboptimal outcomes:
swear at the other {Advance, Advance} and swear at yourself {Wait, Wait}. In this sense,
democracy is a Pareto improvement over the state of nature in which everyone tries to force
the way. Yet this is a very weak argument for the rationality of democracy, since this state of
nature is merely an imaginary counterfact designed to justify the existing order. This is why
property rights arguments for efficiency are normatively unpersuasive.
https://doi.org/10.1017/CBO9781139172493.003
Nash equilibrium
Figure 1.1
signal is "red"; the best response to red is to wait (the alternative is

collision), and she waits. Drivers coming from the east interpret the signal
as green because they expect those arriving from the south to wait; their
best response is to advance (the alternative is to miss a turn and perhaps get
hit from behind), and they do. The outcome is {Wait, Advance}. This
outcome is equilibrium; no one wants to act differently given expectations
of others' actions, and the expectations are mutually fulfilled.
Suppose that leaders of political parties, Left and Right, decide how
dirty their campaigns should be. If Right plays clean, it is best for Left to
play dirty, and vice versa. If they select their strategies independently and
simultaneously, they will adopt some strategy combination {Dirty, Dirty}
that will be self-enforcing in the sense that neither party will want to do
anything else given what the opponent has done. Their expectations will
have been fulfilled: Left will have chosen some degree of dirty on the
assumption that Right chose a definite degree, and Right will have chosen
this same degree on the assumption that Left chose what it in fact did. This
equilibrium is portrayed in Figure 1.1.
Yet another example: Suppose the civilian government anticipates (cor-
https://doi.org/10.1017/CBO9781139172493.003
22 DEMOCRACY
rectly) that if it tinkers with the military, it will invite a coup, but if it leaves
them alone, the military will stay in the barracks. The government reads its
preferences as the discovery that it is better off with {Not Tinker, Stay in
the Barracks (Not Tinker)} than with {Tinker, Probable Coup (Tinker)}. It
decides not to tinker. This is also an equilibrium: The government does not
want to do anything else, anticipating the reaction of the military, and the
military do not want to do anything else given what the government did. 19
Expectations are again fulfilled: The government expects the military to
stay in the barracks, and they do.
What matters about such outcomes is that they constitute equilibria: No
one wants to act differently given what others (would) do in response. Such
outcomes are thus self-enforcing; they are enforced by independent spon-
taneous reactions.
(2) Bargains, or contracts. An outcome is such that at least one actor

would be better off doing something else, and it holds because it is ex-
ogenously enforced. There is some third party who punishes "defections"
from this outcome.
Suppose the two political parties agree not to engage in a dirty cam-
paign, even though it is most useful for each of them to do so if the other
does not. If parties want to win elections, this outcome will not hold
without external enforcement. Suppose the parties agreed not to exceed the
degree of dirt represented by point C (R, L) in Figure 1.1. Now, the leaders
of the Right party look at what the Left has promised to do and ask
themselves what it is best for them to do in response. They will smear the
Left party all the way to that point on their best response line, /?*(L). But
then the Left party will discover that if the Right has begun to talk about the
sexual mores of their leader, it is best for them to point out the sources of
wealth of their opponents. And so the agreement will unravel until it
arrives at the equilibrium outcome. For the initial agreement to stick, a Fair
Elections Commission must be able to punish dissuasively everyone who
transgresses. Bargains, or contracts, are agreements in which at least one
19
Note that this is a somewhat different equilibrium from the one we used to solve the
game between political parties. Political parties chose their strategies simultaneously, whereas
in the civilian-military game the government moved first, anticipating the best response of
the military. The first equilibrium concept is not very plausible, and the question of what
constitutes a reasonable notion of equilibrium is still wide open. But all these niceties need not
occupy us at the moment: Nash equilibrium is the simplest and the classic concept of game
theory.
https://doi.org/10.1017/CBO9781139172493.003
party has an incentive to renege but which hold because a third party
effectively sanctions defections.
But who is the third party who inflicts punishments under democracy?
In the end, there are two answers to this question. Either enforcement is
decentralized - there are enough actors who self-interestedly sanction
noncompliance to support the cooperative outcome - or it is centralized -
there is a specialized agency that has the power and the motivation to
sanction defections, even if this agency is not itself punished for failing to
sanction defections or for sanctioning behaviors that constitute com-
pliance.20 There are only two answers "in the end" because the issue is not
whether the state, in the Weberian sense, is necessary to sanction non-
compliance. In all democracies, state institutions specialize in doing pre-
cisely that. The question concerns the autonomy of the state with regard to
the politically organized civil society. If the sanctioning behavior of the
state is not itself subject to sanctions from the society, the state is autono-
mous; the cost of order to society is the Leviathan. But the Leviathan - an
externally enforced cooperative agreement - is not democracy.21 The cost
of peace is a state independent of the citizens. In turn, if the state is itself an
(albeit imperfect) agent of coalitions formed to assure compliance - a pact
of domination - then democracy is an equilibrium, not a social contract.
The state enforces compliance because it would itself be punished for not
doing so or for using its coercive power to prevent participation. And it
would be punished given the interests of the relevant political forces.
Hence, the notion that democracy is a social contract is logically in-
coherent. Contracts are observed only because they are exogenously en-
forced; democracy, by definition, is a system in which no one stands above
the will of the contracting parties. As Hardin (1987: 2) put it, "A constitu-
tion is not a contract, indeed it creates the institution of contracting. Hence,
again, its function is to resolve a problem that is prior to contracting."
(3) Norms. Equilibria and bargains are the only states of the world that are
feasible according to game theory. This theory asserts that all outcomes
20
Enforcement is decentralized if, when a car passes out of turn, someone is willing to
pass out of turn from the other direction, this time risking a collision because the present
sacrifice will increase his or her expected probability of passing in the future. The result is an
equilibrium, a "subgame perfect equilibrium" in game theoretic language.
21
As Kavka (1986: 181) observed, for Hobbes "the sovereign is not, qua sovereign, a
party to the social contract and is therefore not constrained by it." Kavka ended up arguing
(p. 229), in the same vein as I do, that this solution is not necessary to evoke compliance if the
government is "divided and limited."
https://doi.org/10.1017/CBO9781139172493.003
24 DEMOCRACY
hold only because they are mutually enforced in self-interest or are en-
forced externally by some third party. Specifically, this theory proscribes
outcomes that would be supported by something other than a strategic
pursuit of interests.
Yet the literature on democracy is full of the language of values and
moral commitments.22 In particular, those writing about transitions fre-
quently report precisely such normatively inspired commitments to democ-
racy. These tend to be called pacts. 23 Institutional pacts are agreements to
establish democracy even if a particular system of institutions is not best
for some political forces. Political pacts are collusive agreements to stay
away from dominant strategies that threaten democracy. Social - in fact,
economic - pacts are commitments by unions and firms to restrain present
consumption. Military pacts are deals, often secret, between civilian politi-
cians and the military that say, "We will not touch you if you do not touch
us." Such outcomes are said to be supported by values: They are collec-
tively optimal, individually irrational, and not externally enforced. Game
theory claims they do not exist.
I adopt the game theoretic perspective in what follows. I am not claim-
ing that normative commitments to democracy are infrequent or irrelevant,
only that they are not necessary to understand the way democracy works. 24
I am convinced that arguments about whether democracies are supported
by acting out of values or by strategic pursuit of interests are not resolvable
by direct reference to evidence. The two orientations have to and do
compete with each other in making sense of the world around us. The only
claim I am trying to substantiate is that a theory of democracy based on the
assumption of self-interested strategic compliance is plausible and suffi-
cient.
This claim is made possible by recent developments in game theory that,
22
A typical explanation of the feebleness of democracy in this perspective is well repre-
sented by the title of a recent Brazilian book: A cidadania que no temos (The citizenry we do
not have).
23
I am not claiming that all "pacts" to be found in the literature on transitions are pacts in
this sense. Some are bargains, and some are perhaps even equilibria. Despite its botanical
proclivities, this is not a literature distinguished by conceptual clarity.
24
This assertion does not imply that culture does not matter. Culture is what tells people
what to want; culture informs them what they must not do; culture indicates to them what they
must hide from others. I take it as an axiom that people function in a communicative and a
moral context. Buying votes, for example, is considered immoral in all democracies, though
it may be a collectively efficient behavior: If politicians trade promises of future benefits for
votes, why cannot they just pay up front?
https://doi.org/10.1017/CBO9781139172493.003
though still in the midst of a rapid flux, all add up to the message that
cooperation can be spontaneously enforced in systems with decentralized
self-interested punishments.25 The variety of circumstances in which this
assertion is true include repeated situations in which actors do not discount
the future and the probability of the game ending in any particular round is
low, repeated situations in which the game is expected to last indefinitely
and the actors discount the future at not too high a rate, and repeated
situations in which there is even a very low probability that one of the
actors is irrational. Many punishment strategies support compliance: tit for
two tats, two tits for a tat, three tits for two tats, and so on. 26
Thus, neither normative commitments nor "social contracts" are neces-
sary to generate compliance with democratic outcomes. Again, in all de-
mocracies the state is obviously a specialized agency for enforcing com-
pliance. Moreover, since the state monopolizes instruments of organized
coercion, there is a perpetual possibility that it will become independent,
that it will act in its own interest without effective supervision by political
forces. This is why the threat of the autonomization of the state is perpetual
and why institutional frameworks for controlling state autonomy are of
fundamental importance in any democracy.27 The central difficulty of po-
litical power in any form is that it gives rise to increasing returns to scale
(Lane 1979): On the one hand, incumbency can be used directly to prevent
others from contesting office; on the other hand, economic power translates
into political power, political power can be used to enhance economic
power, and so on. But compliance can be self-enforcing if the institutional
framework is designed in such a way that the state is not a third party but an
agent of coalitions of political forces. The answer to the question "Who
guards the guardian?" is: those forces in the civil society that find it in their
25
It appears that we were too precipitous in embracing Mancur Olson's (1965) vision of
the world as a macrocosm of prisoner's dilemmas generating ubiquitous collective action
problems. We now know that in a wide range of repeated situations, cooperative equilibria
can be spontaneously supported by self-interested actions. See Fudenberg and Maskin 1986
for several theorems to this effect. Note, in particular, their theorem 2, which shows that
under rather mild conditions (payoffs must be sufficiently varied), this result holds for n-
person games. Their explanation (p. 544) is the following: "If a player deviates [from
cooperation], he is minimaxed by the other players long enough to wipe out any gain from his
deviation. To induce the other players to go through with minimaxing him, they are ultimately
given a 'reward.' " Note furthermore that the punishment strategies that induce cooperation
need not depend on a history of past deviations; hence, players need not recognize one another
to inflict effective punishment for noncooperation (Abreu 1988).
26
A tit is a sanction in this language; a tat is an act of noncompliance.
27
See Przeworski 1990: ch. 2 for a review of literature on this topic.
https://doi.org/10.1017/CBO9781139172493.003
26 DEMOCRACY
self-interest. Democracy can be an equilibrium: a system of "self-govern-

ment" in which the distinction between the rulers and the ruled disappears
because, as Montesquieu put it, "le peuple . . . est a certains egards le
monarque; a certains autres, il est le sujet."28
Democracy as an equilibrium
Democracy is consolidated when under given political and economic con-
ditions a particular system of institutions becomes the only game in town,
when no one can imagine acting outside the democratic institutions, when
all the losers want to do is to try again within the same institutions under
which they have just lost. Democracy is consolidated when it becomes self-
enforcing, that is, when all the relevant political forces find it best to
continue to submit their interests and values to the uncertain interplay of
the institutions. Complying with the current outcome, even if it is a defeat,
and directing all actions within the institutional framework is better for the
relevant political forces than trying to subvert democracy. To put it some-
what more technically, democracy is consolidated when compliance -
acting within the institutional framework - constitutes the equilibrium of
the decentralized strategies of all the relevant political forces.29
This hypothesis is based on three assumptions. First, institutions matter.
They matter in two ways: as rules of competition and as codes of punish-
ment for noncompliance. That rules affect outcomes needs no discussion.
Just consider the following examples. The Spanish Union Centro Demo-
cratico, the party headed by Adolfo Suarez, and Roh Tae-Woo both re-
ceived 35 percent of the vote in the first democratic elections in their
respective countries. But Suarez won the election in a parliamentary sys-
tem: To form a government, he had to build a coalition, and he could
remain in office only as long as this coalition enjoyed sufficient support.
Roh was elected president for a five-year term and could rule during this
period, using decree powers, regardless of the short-term dynamic of polit-
ical support.30
28
I am quoting from a 1905 edition of L esprit des lois, edited and commented on by
Camille Julia, who footnotes this statement with a reference to Aristotle: "All should com-
mand each one and everyone all, alternatively."
29
By "political forces," I mean those groups that are already organized collectively and
those that can be organized under the particular institutional framework, as well as individuals
in their role as voters. I do not suppose that political forces are organized prior to and
independently of the particular institutional framework; institutions do shape political organi-
zation.
30
This example is due to Juan Linz.
https://doi.org/10.1017/CBO9781139172493.003
The point about institutions as codes of punishment is more complex.

Note that I argued earlier that actors may find it individually rational to
comply with some (cooperative) outcomes without invoking institutions:
When certain conditions are fulfilled, punishing deviations from coopera-
tion by others is the best strategy for each self-interested rational actor. Yet
the game theoretic account is based on the implicit assumption that some
actors have the capacity to punish. To administer sanctions, actors must be
able to undertake actions the effect of which is to lower the payoffs to
others. Institutions enable such punishments and make them predictable;
they have a priori rules according to which punishments are meted out, the
physical means of administering punishments, and incentives for spe-
cialized agents to administer them. Just think of taxes. To induce com-
pliance, there must be rules of punishment, a bureaucracy for the detection
of noncompliance, and a set of incentives for the bureaucracy to detect it
and to apply the rules. If the tax office lacks means of detection and if
bureaucrats can be easily bribed, punishment will not be effective. Institu-
tions replace actual coercion with a predictable threat.31
Second, there are different ways of organizing democracies. In some
democracies, directly elected presidents head governments independent of
support in legislative bodies. In other democratic systems, governments
must be supported by parliaments and last only as long as they can muster
support. Another important distinction concerns the manner in which in-
terests are organized and some aspects of economic policy are determined:
The preponderant role of political parties may be countered by the offi-
cially recognized role of union federations and employers' associations in
representing functional interests and in concerting with each other and with
governments about macroeconomic policies. Yet another important dif-
ference is between those systems that give almost unlimited powers to
current majorities and systems that tightly constrain majority rule, often by
providing special guarantees for religious, linguistic, or regional groups.
These are just illustrations. The list of important differences could be
continued to include electoral formulas, the presence or absence of judicial
review, the mode of civilian control over the military, the existence of a
professional civil service, and so on.
Finally, contrary to the current fashion, institutions make a difference
31
Game theorists take it for granted that punishment strategies are available to players. Yet
the issue is a complex one, as shown by Kavka (1986: ch. 4, sect. 3). In the state of nature,
punishments can be administered, but only by physical coercion. Institutions organize this
coercion, make it predictable, and rely on the threat.
https://doi.org/10.1017/CBO9781139172493.003
28 DEMOCRACY
not only in efficiency but, as Knight (1990) has forcefully reminded us,
through their profound distributional effects. It is well known, for exam-
ple, that first-past-the-post electoral formulas often generate "unearned
majorities": majorities of parliamentary seats out of minority electoral
support. Collective bargaining frameworks affect the results of wage nego-
tiations; property laws affect the assignment of responsibility for accidental
losses; rules governing university admissions determine the class composi-
tion of the student body.
Because they have distributional consequences - because they provide
different opportunities to particular groups - some institutional frame-
works are consolidated under particular economic and political conditions,
where others would not have been. The question, then, is what kinds of
democratic institutions will evoke the compliance of the relevant political
forces?
But what does it mean not to comply? This is not a place for hair
splitting; let me just distinguish what matters from what does not. In no
system do all individuals comply with all that is expected or required of
them. Since the marginal costs of enforcement are typically increasing, all
states tolerate some individual noncompliance, sometimes on a massive
scale. Noncompliance, in a somewhat counterintuitive sense, can also
mean individual withdrawal from participation: indifference to outcomes
resulting from democratic institutions. Nonparticipation at times assumes
mass proportions: At least 35 percent of the U.S. citizenry remains perma-
nently outside the democratic institutions.
These forms of individual noncompliance can threaten democracy when
they are on a mass scale, by creating a potential for sporadic street out-
bursts or ephemeral antidemocratic movements. But isolated individuals
do not shake social orders. This is why "legitimacy" understood in indi-
vidual terms, even with all the Eastonian distinctions, has little bearing on
the issue of regime stability. Only organized political forces have the
capacity to undermine the democratic system.
Thus, the only forms of noncompliance that matter for the self-enforce-
ment of democracy are strategies that (1) seek to alter ex post the outcomes
of the democratic process and (2) drastically reduce the confidence of other
actors in democratic institutions.32 Thus, not to comply is the same as to
subvert the democratic system in order to override its outcomes.
32
If any actor is able to reverse the outcome ex post, other actors must update downward
their expectations about winning the game according to the rules.
https://doi.org/10.1017/CBO9781139172493.003
Let me suggest schematically how spontaneous decentralized self-

interested compliance may work.
Examine the situation from the point of view of a particular actor, such
as the military or a coalition of the bourgeoisie and the military. At any
moment, the outcomes of the democratic process are such that these actors
either win or lose, where the value of having won is greater than of having
lost (W > L). The probability they attach to their chance of winning in any
future round is/?. 33 The courses of action available to these actors are either
to comply or to subvert. If they subvert, they get 5, where S includes the
risk that they will fail and will be punished;34 and if the compliance of
these actors is problematic, it must be true that W > S > L. 35 Suppose,
then, that they have just lost; let this be t = 0. If they comply, they will get
L(0); if they subvert, they will get 5(0). If they were guided only by
immediate interests, they would subvert. But institutions offer actors an
intertemporal perspective. Although they have just lost, the actors know
that if they comply in this round then they can expect to get C(l) = pW +
(1 — p)L in the next one, and although L < S, it may be true that L(0) +
C(l) > 5(0) + 5(1), which would lead them to comply at t = 0.
Let us generalize this argument. It is reasonable to assume that actors
discount the future, where the discount factor is 0 < r < 1, so that the
value they attach to compliance in the next round is rC, the round after that
r2C, and so on. The cumulative value of compliance is C*. If they subvert,
they can reverse the loss in this round and can expect to get 5 now and in
the future. The cumulative value of subversion is 5*. If C* > 5*, the losers
will comply at t = 0.
Note that the likelihood of successful subversion and the cost associated
with its failure depend on the willingness of other political forces to defend
the democratic institutions. One may thus be tempted to think in terms of a
"tipping equilibrium": a situation in which each actor's support of democ-
racy depends on the number of other actors who support it. Yet the actors in
the democratic game are not identical; democracy is not just a matter of
numbers. Obviously, the institutional framework of civilian control over
the military constitutes the neuralgic point of democratic consolidation.
33
This is the probability they attach at present; they may update this probability as they
learn whether they are losing or winning.
34
5 depends on the probability that an attempt to subvert the outcomes will be successful
and on the utilities of success and failure of subversion. If q is this probability, and D is the
value of successful subversion and F of its failure, then S = qD + (1 - q)F.
35
Some actors may be such that for them S > W > L: They will always try to subvert.
Others may be characterized by W > L > S: They never will.
https://doi.org/10.1017/CBO9781139172493.003
30 DEMOCRACY
One can complicate this story in several ways to make it more realistic,
allowing for more differentiated strategies, incomplete knowledge and
learning, and a more reasonable notion of victories and defeats.36 But one
fundamental conclusion has already emerged from this simplified model
and continues to hold when the model is made descriptively more realistic:
Compliance depends on the probability of winning within the democratic
institutions. A particular actor i will comply if the probability it attaches to
being victorious in democratic competition, p(i), is greater than some
minimum; call it p*(i). This minimum probability depends on the value the
particular collective actor attaches to outcomes of the democratic process
and to outcomes of subverting democracy and on the risk it perceives for
the future. The more confident the actor is that the relationship of political
forces will not take an adverse turn within the democratic institutions, the
more likely is this actor to comply; the less risky the subversion, the less
likely are the potential antidemocratic forces to comply.37
None of the above is intended as a description of historical events.
"Models" - I frequently feel forced to cite Theil (1976: 3) - "are to be
used, not believed." What the model suggests is that in analyzing any
concrete situation one should consider the values and the chances the
particular political forces attach to advancing their interests under democ-
racy and outside it. Democracy will evoke generalized compliance, it will
be self-enforcing, when all the relevant political forces have some specific
36
Note that the concepts of winning and losing are greatly simplified here. Ea^h group
defines its interests over a broad spectrum of outcomes and attaches values to particular
degrees and specific manners in which each of these interests is realized. Thus, winning and
losing are continually defined for multidimensional preference contours. But there is no
reason to get mired in mathematics if the logical implications remain the same as in a simple
model.
37
For those who are curious about the reasoning and not just the conclusions, here is the
model. If the actor has just lost, at time t, set as t = 0 for notational convenience, the payoffs
from complying are C* = L + %r<C(t) = L + [r/(l — r)]C. The payoffs from subverting are
S* and depend on the probability this actor attaches to the success of subversion and the rate at
which it discounts the nondemocratic future. Hence, the actor complies if C* > 5*, or if
Note that dp*/dr < 0: The more confidence a particular actor has in its future under democ-
racy, the lower the minimum probability required to evoke its compliance. In turn, let q be the
probability of the success of subversion, dS*/dq > 0. Then dp*/dq > 0: The less risky it is for
a particular group to subvert, the higher is the probability of winning required to make it obey
democratic outcomes.
Finally, observe that if p* is sufficient to evoke compliance when the actor has just lost, it
will be also sufficient if it has just won. Hence, p > p* is the minimal condition.
https://doi.org/10.1017/CBO9781139172493.003
minimum probability of doing well under the particular system of institu-
tions. 38
This probability is different for different groups. We learned earlier that
it depends on the specific institutional arrangements and on resources the
participants bring into the democratic competition. We now learn that it
also depends on the power a particular actor has to cause the downfall of
democracy. The military have weak prospects to pursue their interests
under democracy, but they can subvert democracy by force: Their W is low,
their S high. Hence, their p* may be quite high. The bourgeoisie can do
quite well under democracy and well outside it but need the military for
successful subversion. Unions and other organizations of wage earners can
do quite well in democratic competition, but they are often brutally re-
pressed if democracy falls; they may be the one group for which L > S and
which always prefers to comply.39 Moreover, the guarantees required by a
particular group may vary with historical conditions. In post-1976 Spain,
the military were almost indifferent as between S and L; they were so
starved by Franco that even a nonpolitical life under democracy seemed
satisfactory to them. In turn, the post-1983 Argentine military saw L as
much inferior to S; they knew that losing could mean long jail sentences for
many of them. These are just seat-of-the-pants speculations; what I want to
show is that even the simplified model has some power to distinguish
particular actors and different historical conditions.
Hence, the minimal chance required to stay within the democratic sys-
tem depends on the value of losing in the democratic interplay of interests.
Those political forces that have an outside option - the option of subverting
democracy or provoking others to subvert it - may stay with the democrat-
ic game if they believe that even losing repeatedly under democracy is
better for them than a future under an alternative system. After all, democ-
racy does offer one fundamental value that for many groups may be suffi-
cient to prefer it to all alternatives: security from arbitrary violence. As
Santiago Carillo, then secretary of the Spanish Communist party, put it in
1974, "One should have the courage to explain to the working class that it
is better to pay surplus value to this bourgeois sector than to create a
situation that may turn against them" (Carrillo 1974: 187).
Even from the purely economic point of view, faith in the efficacy of
38
The political forces that are relevant are those for which S > L. Those for which L > S
have no outside option and need no guarantee.
39
The Peronist unions in Argentina are the most likely exception.
https://doi.org/10.1017/CBO9781139172493.003
32 DEMOCRACY
democracy may be a source of commitment among those who see little

chance of winning distributional conflicts within democratic institutions. If
democracy is believed to be conducive to economic development in the
long run, various groups may opt for this system even if they see little
chance of winning conflicts about distribution. The higher the anticipated
value of losing under democracy, the lower need be the chances of win-
ning. 40
This last hypothesis has implications for the perennial issue of the social
conditions of democracy. Put conversely, the model implies that if some
important political forces have no chance to win distributional conflicts and
if democracy does not improve the material conditions of losers, those who
expect to suffer continued deprivation under democratic institutions will
turn against them. To evoke compliance and participation, democracy must
generate substantive outcomes: It must offer all the relevant political forces
real opportunities to improve their material welfare. Indeed, a quick cal-
culation shows that in South America between 1946 and 1988 any regime,
democratic or authoritarian, that experienced positive rates of growth in a
given year had a 91.6 percent chance of surviving through the next twelve
months, a regime that experienced one year of a negative rate of growth
had an 81.8 percent chance, and a regime that experienced two consecutive
years of declining incomes had only a 67 percent chance.
Yet it is important to see what this hypothesis does not imply. First, it
does not mean that democracy must have a social content if the institutions
are to evoke compliance. If democracy is a system in which outcomes
always appear uncertain, "social content" cannot mean prior commitments
to equality, justice, welfare, or whatever.41 Such commitments are not
feasible; under democracy, outcomes are determined by the strategies of
competing political forces and are thus inevitably uncertain ex ante. Con-
stitutions that are an oath to promote the general welfare, enhance national
unity, advance the culture of the people, or provide decent conditions of
life for everyone42 may be necessary for catharsis, but they cannot be
40
This is true if the political forces that have a low chance of winning distributional
conflicts believe that democracy will result in increasing the total pie. Return to the expression
forp*. The derivative dp*ldL = - ( 1 / r j [W - (1 - r)S*]/(W - L)2. This derivative is
negative.
41
This has been a topic of my repeated debate with Francisco Weffort. For the most recent
salvo, see his "Incertezas da transic.ao na America latina," (1989).
42
Not to mention such clauses as those requiring every firm that employs more than ten
workers to hire at least 10 percent of new employees over forty-five years of age!
https://doi.org/10.1017/CBO9781139172493.003
complied with. They can be observed only to the extent that they express
laws, not oaths. 43 Democracy may end up having a social content if the
institutional framework favors social justice in spite of the unequal re-
sources with which different forces enter the democratic competition. But
this is a matter of institutions, not of substantive commitments.
Second, the assertion that democracy cannot last unless it generates a
satisfactory economic performance is not an inexorable objective law. A
phrase one repeatedly hears in newly democratic countries is "Democracy
must deliver, or else . . . " The ellipsis is never spelled out, since it is taken
as self-evident. When Argentine generals proclaim one after another that
"the economic situation is putting democracy at risk" {New York Times, 3
January 1990), they appear to be asserting an objective law of which they
are just unwitting agents: They expect the economic crisis to turn some
civilians against democracy, which will increase the probability of suc-
cessful subversion, to which they will respond, given their preferences, by
overthrowing democracy. Yet whether or not democracy survives adverse
economic conditions is a joint effect of conditions and institutions. As the
European experience of the Great Depression demonstrates, some institu-
tional frameworks are more resistant than others to economic crisis.
In conclusion, from the static point of view democratic institutions must
be "fair": They must give all the relevant political forces a chance to win
from time to time in the competition of interests and values. From the
dynamic point of view, they must be effective: They must make even losing
under democracy more attractive than a future under nondemocratic alter-
natives. These two aspects are to some extent interchangeable. They con-
stitute different ways of asserting that political forces comply with demo-
cratic outcomes when they expect that their future will be better if they
continue to follow the rules of the democratic game: Either they must have
a fair chance to win or they must believe that losing will not be that bad.
Thus, to evoke compliance, to be consolidated, democratic institutions
must to some extent be fair and to a complementary degree effective.
Yet under certain conditions these requirements may be contradictory,
particularly with regard to economic issues. Fairness requires that all major
interests must be protected at the margin; effectiveness may necessitate that
they be seriously harmed. To be effective economically, governments may
43
This juxtaposition is derived from the current Polish constitutional debate. See Trybuna
Ludu, 17 September 1989.
https://doi.org/10.1017/CBO9781139172493.003
34 DEMOCRACY
have to violate some property rights - for example, by adopting land

reform or by generating massive unemployment in a quest for allocative
efficiency. Institutions conducive to major economic transformation cannot
protect all interests; institutions that protect all interests are not an appro-
priate framework for major economic transformation.
Indeed, the traditional dilemma of the Left has been that even a pro-
cedurally perfect democracy may remain an oligarchy: the rule of the rich
over the poor. As historical experience demonstrates, democracy is com-
patible with misery and inequality in the social realm and with oppression
in factories, schools, prisons, and families. And the traditional dilemma of
the Right has been that democracy may turn out to be the rule of the many
who are poor over the few rich. Democratic procedures can threaten prop-
erty; political power in the form of universal suffrage and the right to
associate may be wielded to restrict property rights. Hence, the conditions
under which democracy becomes the equilibrium of decentralized strat-
egies of autonomous political forces are restrictive. This is why democracy
has been historically a fragile form for organizing political conflicts.
Institutional design
What does this abstract discussion imply about specific institutions? What
kinds of institutional arrangements are likely to last and to matter? Should
the constitution contain only rules about political competition and about
protecting minorities, or should it include substantive commitments? Is the
parliamentary system more likely than the presidential one to regulate
conflicts?44 Are some elements of a corporatist organization of interests
necessary to mobilize consent to economic policy at a time of crisis?
44
Linz (1984) has developed a number of arguments in favor of parliamentary, as opposed
to presidential, systems. I am particularly persuaded by his observation that presidential
systems generate a zero-sum game, whereas parliamentary systems increase total payoffs.
The reasons are the following. In presidential systems, the winner takes all: He or she can
form a government without including any losers in the coalition. In fact, the defeated candi-
date has no political status, as in parliamentary systems, where he or she becomes the leader
of the opposition. Hence, in terms of the model developed above, under ceteris paribus
conditions (under which W + L = T is the same in both systems), the value of victory, W, is
greater and the value of defeat, L, is smaller under presidential than under parliamentary
systems. Now, assume that political actors discount the future at the rate of r per annum.
Under the presidential system, the term is fixed for some period (/ = PRES), and the expected
value of the next round is rPRES [pW + (1 — p)L\. Under the parliamentary system, the winner
governs only as long as he or she can maintain sufficient support in the parliament, say for the
period / = PARL, SO that the expected value of the next round is rPARL [pW + (1 - p)L].
https://doi.org/10.1017/CBO9781139172493.003
The relationship between constitutions and political reality is not an

obvious one. Except for the United Kingdom and Israel, all countries have
written, formally adopted constitutions. Yet these constitutions have had
highly divergent roles in the actual political life of their countries. In the
United States the same constitution has survived for two hundred years,
during which time it has continually influenced political life, at least in the
sense that major political conflicts, with one major exception, have been
framed in terms of it. In Argentina, the constitution adopted in 1853
remained, on paper, in effect except for the brief period between 1949 and
1957. Yet in the past fifty years, political conflicts in Argentina have only
half the time been processed according to its provisions. In France, the
constitution has been changed several times since 1789; indeed, every
major political upheaval has produced a new one. Yet while it was in force,
each constitution did regulate the exercise of power and the pattern of
succession. Finally, to fill the last cell of this fourfold table, in South Korea
major constitutional reforms have occurred every three years and nine
months since 1948, and no succession has conformed to the rules. A
constitution that is long-lasting and observed, one that is long-lasting and
ignored, some that are changed often and respected serially, and others that
are modified frequently and remain irrelevant - historical experience is not
very informative.
Indeed, I discovered, much to my surprise, that we do not have suffi-
ciently reliable empirical knowledge to answer questions about institu-
tional design. We have intuitions about the impact of presidentialism ver-
sus parliamentarism, we know the effects of alternative electoral systems,
and we tend to believe that an independent judiciary is an important ar-
bitrating force in the face of conflicts, but our current empirical knowledge
leaves a broad margin for disagreements about institutional design. Is
democracy in Poland more likely to be consolidated under a strong or a
weak presidency? Under a plurality or under a system of proportional
representation? Under a constitution that affirms the commitment to com-
mon values or under one that leaves them open? We just do not know
enough to answer such questions when confronted with specific historical
conditions.
The reason we cannot answer such questions in a reliable way is that the
Elementary algebra will then show that unless the tenure expected under parliamentarism is
notably longer than under presidentialism, the loser has a greater incentive to stay in the
democratic game under parliamentarism.
https://doi.org/10.1017/CBO9781139172493.003
36 DEMOCRACY
consolidation of democracy may be a joint effect of conditions and institu-

tions. Institutions may have to fit conditions. Rousseau (1986: 1) may have
been correct when, in the course of designing a constitution for Poland, he
wrote, "One must know thoroughly the nation for which one is building;
otherwise the final product, however excellent it may be in itself, will
prove imperfect when it is acted upon - the more certainly if the nation be
already formed, with its tastes, customs, prejudices, and failings too deep-
ly rooted to be stifled by new plantings." And we have just not done
enough empirical research to acquire a reliable knowledge of such joint
effects.
Hence, I can venture only a rudimentary guess. Constitutions that are
observed and last for a long time are those that reduce the stakes of political
battles. Pretenders to office can expect to reach it; losers can expect to
come back. Such constitutions, Napoleon is alleged to have said, should be
"short and vague." They define the scope of government and establish
rules of competition, leaving substantive outcomes open to the political
interplay. Constitutions adopted to fortify transitory political advantage,
constitutions that are nothing but pacts of domination among the most
recent victors, are only as durable as the conditions that generated the last
political victory. In turn, constitutions that allow everyone to introduce
substantive demands, constitutions that ratify compromises by enshrining
substantive commitments (of which the social rights chapter of the Weimar
Constitution is the prototype) are often impossible to implement.45
To push this argument just one step farther, let me offer three - still
excessively abstract - observations. First, it is worth noting that electoral
majorities have been rare in the history of successful democracies; in the
postwar period only about one election in fifteen has resulted in a majority
of votes cast for one party. Hence, most democracies are ruled either by
explicit coalitions among parties none of which can rule on its own or by
minority governments based on implicit assurance of support. Second,
successful democracies are those in which the institutions make it difficult
to fortify a temporary advantage. Unless the increasing returns to power
are institutionally mitigated, losers must fight the first time they lose, for
waiting makes it less likely that they will ever succeed. Yet, third, govern-
ments must be able to govern, and this implies that they must be able to
45
As Lula put it in a preelection interview, "If we put in practice the social rights chapter
of the new constitution, we will make a revolution" (Luis Inacio Lula da Silva, interviewed in
Veja, 29 November 1990, p. 4).
https://doi.org/10.1017/CBO9781139172493.003
TRANSITIONS TO DEMOCRACY 37
prevent some demands from reaching the public sphere and certainly that
they cannot tolerate all important groups having veto power over public
policy.
These observations add up to two negative rules. To be stable and
effective, democratic institutions must not generate governments unrespon-
sive to the changing relations of political forces, governments free from the
obligation to consult and conceit when they formulate policy, governments
unconstrained to obey rules when they implement them. Yet they also must
not paralyze decisions and their implementation. All interests must be
represented in the making of policy, but none should be able unilaterally to
block its formulation and implementation. Another way to formulate this
conclusion is that a stable democracy requires that governments be strong
enough to govern effectively but weak enough not to be able to govern
against important interests.
If these observations are valid, democratic institutions must remain
within narrow limits to be successful. And under some historical condi-
tions there may be no space between the limits; consolidation of democracy
is not always possible.
Transitions to democracy
Self-enforcing democracy is not the only possible outcome of "transi-
tions": strategic situations that arise when a dictatorship collapses.46 A
breakdown of an authoritarian regime may be reversed, or it may lead to a
new dictatorship. And even if a democracy is established, it need not be
self-sustaining; the democratic institutions may systematically generate
outcomes that cause some politically important forces to subvert them.
Hence, consolidated democracy is only one among the possible outcomes
of the collapse of authoritarian regimes.
Given that under the current economic, political, and institutional condi-
tions autonomous social forces struggle to impose on others a system that
will fortify their political advantage, are there any institutions that will
voluntarily be adopted that, once in place, will elicit decentralized com-
pliance? When it is rational for the conflicting interests voluntarily to
constrain their future ability to exploit political advantage by devolving
46
The term "transitions" is not a very fortunate label for these situations, since it suggests
that the outcome is predetermined. Yet I decided to follow common usage in the immense
body of literature on transitions to democracy.
https://doi.org/10.1017/CBO9781139172493.003
38 DEMOCRACY
some of their power to institutions? When will they conclude a "democrat-

ic pact" that engenders compliance and thus makes democracy self-enforc-
ing?
We now confront nothing less than the classic problem of liberal political
theory. Ever since the seventeenth century, political philosophers have been
hunting for the secret of an alchemical transformation from the brutish
chaos of conflict to the serene life of cooperation. Beginning with the
Leviathan, proposals have been innumerable and, recently, increasingly
optimistic. We are told that the problem of social order can be solved by
conventions (Lewis 1969; Sugden 1986), by spontaneous evolution of co-
operation (Taylor 1976; Axelrod 1984), by norms (Ullman-Margalit 1977;
Axelrod 1986), by morals (Gauthier 1986), and by benevolent institutions
(Schotter 1981).
The generic problem can be posed as follows. Given some strategic
structure of interests classified by various mixes of conflict and coordina-
tion, the noncooperative solution to which has some normatively undesir-
able features, is there any device (the state, the plan, conventions, morals,
norms, institutions, lotteries) that will be voluntarily adopted and that,
once adopted, will evoke spontaneous (free and decentralized) compliance,
that is, behaviors that support normative desiderata, such as collective
(Pareto) rationality, some other welfare criteria, justice, fairness, equity, or
equality? Note that the philosophers' quest is for devices that evoke spon-
taneous compliance, not for institutions that force compliance, even if they
elicit behaviors that are normatively desirable.
This formulation is based on some assumptions that limit its
usefulness.47 The liberal point of departure - that hypothetical "indi-
viduals" confront the problem of cooperation in a state of nature - is not
helpful for analyzing problems confronting real actors in concrete histor-
ical conditions.48 The relevant actors are not abstract individuals but politi-
47
One reason why the Hobbesian formulation is not very useful in our context is that for
Hobbes the first reason for individuals to found a state is that it can defend them from invasion
by foreigners. Only a secondary reason is that it can protect them from injuring one another
(Leviathan, ch. 17). Although territorial conflicts flare up from time to time, the issue we are
analyzing is not founding a state but organizing a state on territory already given. Hence, the
Pareto superiority of having secure borders is not a major consideration in conflicts about
institutions in transitions to democracy.
48
The problem with game theory is that it combines a useful methodology with an
ideologically derived and patently unreasonable ontology of "individuals" who in addition
appear homogeneous in that they have available to them the same strategies and often the
same payoffs. My biases on this topic are treated at length in Przeworski 1985. Note that
https://doi.org/10.1017/CBO9781139172493.003
TRANSITIONS TO DEMOCRACY 39
cal forces: previously constituted collective organizations, some categories

of people who might become collectively organized if provoked, and indi-
viduals as voters. They enter conflicts in a context in which there are
always preexisting conventions, norms, and institutions.
Yet, with these caveats, the role of the democratic pact is to effect just
such an alchemical transformation. Pacts are (one-shot noncooperative49)
equilibria in strategies that consist of altering the current conditions in such
a way as to make decentralized voluntary compliance individually optimal.
They are agreements to disagree. And the only way to change these condi-
tions by agreement is to form new institutions.
Thus, solutions to the problem of democratization consist of institutions.
Resources of political forces are given; so are their preferences and the
conditions independent of everyone. The game is solved if a system of
institutions that engenders spontaneous compliance is an equilibrium of the
transition. The problem of establishing democracy is the following: Will
political actors agree to a framework of democratic institutions that will
evoke their compliance?
This question involves two separate issues.50 The first is whether under
given conditions there are any systems of democratic institutions that will
evoke spontaneous decentralized compliance once they are established.
Under some structures of interests, there may be no institutions that will
stop important political forces from trying to subvert them once they are in
place. The second is whether a self-enforcing system of democratic institu-
tions will be established as a result of conflicts concerning the choice of
institutions. For even if institutions could be found that would be self-
enforcing once installed, they need not constitute the equilibrium of the
Kavka (1986: 148) is careful to define the state of nature as "a model of societies of real
people dissolved by civil disorder or removal of the State."
49
By which I mean only not externally enforced.
50
These issues are collapsed in social contract theories. These theories pose the following
question: What kind of political order would hypothetical individuals in the state of nature see
as worth complying with? They differ with regard to the assumptions imposed on individuals.
If individuals are placed behind a veil that prevents them from knowing anything about their
welfare in the new social order, then the issue is why they would comply with this order once
they were in it and knew how well off they were (Braybrooke 1976). If, in turn, individuals
know their chances in the new order, then the question is why they will agree to one that they
know will cause them to comply with outcomes that make them badly off. Say the military
know that a democratic system will impose civilian control with which it would be best for
them to comply; they may prefer their own dictatorship. Hence, the questions whether
political forces will comply with a given institutional system once it is established and
whether they will agree to establish it are distinct.
https://doi.org/10.1017/CBO9781139172493.003
40 DEMOCRACY
transitional situation when the chances of the particular political forces are
very different under alternative institutional arrangements. Imagine that a
group of people enters a casino that contains a roulette wheel, a poker
table, a blackjack counter, and a crap stand. Is there a game that the
players, given the resources they have, will continue to play even if they
lose a few times in succession? And if there is, will the potential players
agree which one to play?
These are the generic issues inherent in any transition to democracy.
Appendix: Why do outcomes appear uncertain?

One characteristic feature of democracy is that outcomes appear in a partic-
ular way uncertain to all participants. It is as if all do what they think is best
for them, and then some random device chooses the outcome; as if the
results were decided by a throw of dice. Are they in fact? And if they are
not, why do they appear as if they were? The purpose of this appendix is to
clarify the origins and nature of the uncertainty generated by democracy.
Let us first try a less frivolous description of the way democracy oper-
ates. A few examples may help intuition.
Electoral competition is an obvious one. Parties look at the electorate,
decide which issue positions will generate the most support, and choose
those that maximize the probability of winning under that platform. On
election day the result is read, and the parties receive the signal, more or
less uniquely defined in each democracy, to form the government or go into
opposition.
Proponents and opponents of public aid to private schools argue their
case in front of a constitutional court. They cite the constitution if the law
is on their side; the facts if it is not. The court deliberates and issues a
verdict, which is now the legal status quo.
Banks are pressuring the legislature to bail them out of their past mis-
deeds. Everyone knows that universalistic appeals beat particularistic ones:
Banks summon the specter of widows losing their lifesavings; politicians
claiming to represent taxpayers evoke the perils of deficits. The legislature
votes the bail-out, and the bureaucracy writes checks.
Note that in these illustrations there is no room for uncertainty. Given the
resources of the participants and the institutional framework, the outcome
is determined. Each actor can examine the distribution of resources, look
up the rules, and determine who will lose or win what if they all go through
https://doi.org/10.1017/CBO9781139172493.003
A P P E N D I X : WHY DO OUTCOMES APPEAR U N C E R T A I N ? 41
the motions; that is, if they follow their best strategies. And yet the actors
appear to behave as if they were not certain of the outcome.
The evidence that they do is twofold. If winning and losing are di-
chotomous, then those who expect to lose should simply do nothing, since
there is nothing they can do: The court will decide against them because the
other side has better arguments.51 Hence, if they do compete, it must be
because they are uncertain about the consequences of their actions. If
payoffs are continuous, the eventual losers are compelled to go through the
motions because otherwise they would do worse than they can do. Politi-
cians must complain about government largesse even if they know that they
will end up bailing out the banks, just in order not to lose votes. But I think
there is much prima facie evidence that political actors are often uncertain
about the outcome; everyone in a democracy has lived through at least one
election-night drama. My favorite admission of surprise was the editorial in
the right-wing Chilean daily El Mercurio the day after Salvador Allende
won a plurality in the presidential election of 1970: "No one expected that
an election via the secret, universal, bourgeois franchise could lead to the
victory of a marxist candidate."
What, then, is the source of uncertainty inherent in democracy?
Let us examine a few card games. The first one is called LEN. Players
come to the table and bid for the ace of spades. Whoever makes the highest
bid gets his money back and collects the money on the table and a dollar
from everyone who did not play. The rules are perfectly universalistic;
everyone can play. But one player is richer than the others, and wealth
uniquely determines the outcome.52 Hence, there is no uncertainty here.
This is why Lenin was correct to call his conception of democracy the
dictatorship of the bourgeoisie.53 Except for the certain winner, anyone
who pays more than a dollar to enter this game is a dupe.
51
For a dichotomous view of payoffs, see Riker 1962. This view was disputed by Stigler
(1972).
52
Think of (American) football. There are a field, a ball, and a set of rules. The rules are
blind to the identity of the teams. Referees and umpires adjudicate impartially whether actions
conform to the rules and administer specified penalties. But one team consists of 300-pound
players, the other of 150-pound weaklings. The outcome is certain.
53
"The bourgeois parliament, even the most democratic in the most democratic republic
in which the property and the rule of the bourgeoisie are preserved, is a machine for the
suppression of the toiling millions by small groups of exploiters. . . . As long as capitalist
property exists universal suffrage is an instrument of the bourgeois state" ("The Letter to the
Workers of Europe and America" [1919], Lenin 1959: 482). Lenin's most programmatic
statement on this topic is "Theses on Bourgeois Democracy and Proletarian Dictatorship
Presented to the First Congress of the Communist International," 4, March 1919.
https://doi.org/10.1017/CBO9781139172493.003
42 DEMOCRACY
Now let us play JON. Players bid for cards, face down. After all the cards
have been bought, they look at what they have. The player who has the ace
of spades wins, and payoffs are the same as before. In this game, if
everyone plays as well as possible, the wealthiest player will buy the most
cards and will have the best chance of getting the ace. If all the N players
are equally wealthy, their prior probabilities of winning are {l/N,
l/N, . . ., l/N}. In fact, the probabilities may be terribly unequal: The
prior probability distribution may be as skewed as {(N — l)/N, UN,
0, . . ., 0}. But all money can buy is a better chance, because pure chance
plays a role. Even a player who can afford only one card has one chance in
fifty-two of pocketing the prize. Is this what democracy is like?
One obvious argument against this analogy is that democracies - at least
modern ones - have no institutions that function as randomizing devices.54
Parliaments, bureaucracies, and courts are supposed to deliberate and
make decisions on justifiable grounds, not throw dice.
Note, however, that this is the explanation of uncertainty suggested by
social choice theory: Collective preferences cycle incessantly, the time of
reading them lacks particular justification, the outcome cannot be under-
stood in terms of individual preferences. But the uncertainty implied by
social choice theory is too radical; it permits no rational action. Social
choice theory portrays democracy as if it were LOTTO: Actors decide
whether to buy a ticket and wait for the winning numbers to appear on the
screen. The outcome is fair, but this is its only justification. This is not
enough to motivate participation in democracy; to participate, actors must
see some relationship between what they do and what happens to them. If
everyone believed the impossibility theorems, no one would participate.
True, Elster (1989) has shown that there are some circumstances when
collective rationality may call for a random decision: whenever the costs of
deciding are greater than the difference the decision makes - for example,
when a custody battle inflicts more damage on the child than landing with
the less-qualified parent. But in general, a democracy in which people
believed that outcomes were decided at random would be untenable.
Hence, I do not think that this is the way democracy is played. An
element of pure chance does enter the democratic game, but only ex-
ogenously: The accidental death of a leader may radically alter the situa-
tion. But this is where the role of chance ends.
54
There are instances in history of elections by chance and serious arguments in their
x
favor. See Elster 1989.
https://doi.org/10.1017/CBO9781139172493.003
APPENDIX: WHY DO OUTCOMES APPEAR UNCERTAIN? 43
Table 1.1
Column
King of hearts Any other card
R | Ace of spades Nothing, All All, Nothing

o w
1 Any other card Something, Something Something, Something
Another reason outcomes can be uncertain is that actors do not know

what to do. Some commentators on my earlier claim that democracy is
inherently uncertain concluded that this assertion implies that individuals
must be uncertain what to do. 55 Indeed, the Brazilians published one of my
articles under the title "Ama a incerteza e seras democratico": "If You
Like Uncertainty, You Will Be a Democrat."56 Now, it may be true, as
Manin (1987) argues, that democracy requires that citizens be willing to
change their preferences. But they need not like uncertainty and need not
be uncertain what to do.
Let NOR be a game in which actors do not know which strategies will
produce the best outcomes, because these outcomes depend on simul-
taneous actions by others: there are no dominant strategies. The game is
played as follows. Bids are made for cards, face down. Once all the cards
have been bought, the players (two for the sake of simplicity, named Row
and Column) play by each putting a card on the table face down and
simultaneously turning over the cards. The payoffs are given by {first
payoff to Row, second to Column}.57
Row does not know what to do. Playing the ace of spades is better than
pulling any other card if Column plays any card other than the king of
hearts; otherwise, it is worse. The same is true for Column. (Table 1.1.)
Some game theorists assert that the rational thing to do under the circum-
stances is to use a random device to choose one's actions. If a political
55
Notably Lechner (1986) and Hirschman (1986). There are in fact two distinct reasons
why actors may not know what to do. The one discussed in the text is that they do not know
which course of action is best for them. But I have an impression that Hirschman and Manin
(1987) have something else in mind, namely, that, educated by deliberation, actors are not
certain whether they should act on their own current preferences or yield to the preferences of
others. In the latter case, actors are uncertain about their own preferences rather than that
about courses of action.
56
Novos Estudos, 1985.
57
To limit the impact of resource inequality on the outcome, the Law of Fair Access
ensures that the same player cannot have both the ace of spades and the king of hearts.
https://doi.org/10.1017/CBO9781139172493.003
44 DEMOCRACY
party does not know whether it can gain more votes moving to the left or to
the right, because the outcome depends on where another party moves, it
should decide by throwing an appropriately weighted coin. If banks do not
know whether an argument about widows is more persuasive than one
about its employees threatened with losing their jobs, they may decide by
chance. In this case the outcome is uncertain because it emerges from
probabilistically chosen strategies: The combination of strategies that has
the property that no one would want to mix the strategies differently given
what others can do is unique, but the outcomes are only probabilistically
knowable.58
Lechner is right that NOR is not a plausible understanding of democracy,
because democratic actors value order, an order that will indicate to them
what to do. Disorder destabilizes democracies, argues Lechner, influenced
by the trauma of the chaotic years of the Unidad Popular government in
Chile. I agree, but I do not think that the uncertainty about outcomes
entails either chaos at the institutional level or uncertainty about one's own
actions.
The explanation of uncertainty that I find most persuasive has been
offered by Aumann (1987). He has shown that if actors do not know
something, if they are cognitively rational in the sense that they change
their beliefs about the world as a function of information they get, 59 and if
they act on these beliefs, then the strategies they choose independently will
be distributed probabilistically, as if they had been chosen jointly using a
random device.
What is it that actors do not know? One of the many powerful implica-
tions of Aumann's model is that they may not know all kinds of things, not
only those that traditional game theory allowed them to be ignorant about,
but also the strategies of other actors. Indeed, this is what actors do not
know in Aumann's account. Each actor may know the unique outcome
associated with each combination of strategies, and each may know what it
58
This idea seems to be going out of fashion. See Aumann 1987 and Rubinstein 1988: 9;
the latter says that "the naive interpretation of a mixed strategy, as an action which is
conditioned on the outcome of a lottery executed by the player before playing the game, is
intuitively ridiculous." In turn, a physically mixed strategy - mixing strategies in some
proportion - would not lead to uncertainty.
59
One important assumption underlying Aumann's model is the so-called Harsanyi doc-
trine, which asserts that the only source of knowledge is observation. Specifically, the as-
sumption is that all actors have the same priors, so that if they attach different probabilities to
crossing an intersection at any moment, it is only because what they have observed is
different.
https://doi.org/10.1017/CBO9781139172493.003
is best for others to do given what he or she does. Only the most minimalist
assumption is required to generate uncertainty: that I am not sure how
others see me. Leaders of a political party may know that if they keep the
opponent's skeletons in the closet, it will be best for others to reciprocate,
but if they are not sure whether opponents trust them not to cause scandals,
uncertainty will ensue. The minimal assumption is that I am not sure that
the opponents know my preferences or my character. If I allow that they
may see me as moralistic rather than victory-oriented or as reckless, I
cannot be sure what they will do.
Hence, the outcomes of the democratic process are not uncertain. They
only appear to be uncertain to every participant. But "appears" should not
be taken as an indication of remediable ignorance, as "false con-
sciousness."60 The appearance of uncertainty is necessarily generated by
the system of decentralized decision making in which there is no way to be
sure what others think about me. An omniscient observer could determine
the unique outcome of each situation, but no participant can be an ob-
server, because the observer's theory need not be universally shared by
other participants. And if it is not shared, then she cannot be certain how
others perceive her and hence what they will do. Note that the strategies are
chosen independently and deterministically. Each actor decides indepen-
dently what to do, and each actor knows what it is best to do at every
moment. Yet the outcomes associated with these combinations are dis-
tributed probabilistically.
To highlight the distinguishing features of uncertainty inherent in de-
mocracy, consider a stylized model of authoritarian regimes (which I treat
as synonymous with dictatorships, abandoning some important distinc-
tions).61
60
This lapse into marxist language is not accidental. Aumann's model provides microfoun-
dations for Marx's theory of fetishized knowledge. Fetishized knowledge is simply local
knowledge: the view of the system from the point of view of each agent. Individual agents
exchanging under capitalism do gain or lose from exchanges: If I sell for more than I bought, I
will gain and the buyer will lose labor values (but not necessarily utility). This is a valid local
theory of the capitalist system; everyone operating within this system must act on the basis of
this theory. Informed by marxist theoreticians, everyone may know that value is created only
by labor and that when all values entering exchange are summed up, their sum is zero:
Whatever I gained in exchange, someone else lost it. But this knowledge does not and cannot
alter individual behavior within the system. A critique of capitalism is not sufficient to alter
individual behavior.
61
And distinctions there are. Just think of the Soviet Union, which was variously dubbed a
totalitarian regime, an authoritarian one, a dictatorship of the proletariat, a dictatorship of a
party, an autocracy (samoderzhavie), a state capitalist system, a nomenklatura, a bureaucracy,
https://doi.org/10.1017/CBO9781139172493.003
46 DEMOCRACY
One essential feature of authoritarian regimes is that someone has an

effective capacity to prevent any outcome from occurring. As was said of
Franco, "All the cards are in his hands, he does not make politics, he is
politics" (cited in Carr and Fusi 1979: 1). That "someone" can be the ruler;
an organization, such as the armed forces, the party, or the bureaucracy; or
even a less easily identifiable ring of groups and individuals. I speak below
of an authoritarian power apparatus and introduce distinctions only when
they enlighten the problem at hand.62 The power apparatus can act not only
ex ante, but even ex post; that is, not only can it establish rules that prohibit
actions that would lead to undesirable outcomes, but it can also overturn
such outcomes even if they result from following its own rules. Here is an
example drawn from Argentina. A minister of education appointed by the
military government charges a group of experts with preparing a mathe-
matics text for elementary schools. The textbook is prepared, and it is
approved by the minister, published, and distributed. It then falls into the
hands of the commandant of a local military zone, who orders that it be
removed from the schools. Note that the text in question is not an under-
ground pamphlet; it is a product of the authoritarian institutions them-
selves.63 In contrast to a democracy, the set of possible outcomes cannot be
deduced from the rules. 64 Under dictatorship, there is no distinction be-
and what not. My purpose is only to highlight what I see as the essential features of democ-
racy, not to provide a classification of forms of government. Most important, my discussion
collapses a distinction between what Montesquieu called despotism, where the will of the
despot is the order of the day, and dictatorships that rule through laws (monarchy: rule by laws
but not of law). For a discussion of various classifications of political regimes in history, see
Bobbio 1989: 100-25.
62
On the difficulties of identifying the centers of power under authoritarianism, see
Przeworski 1982. A more systematic analysis is offered in Cardoso 1972.
63
Note another aspect of this example: the absence of a clearly defined authority. There
are no rules that give the commandant of a military zone the authority to act on primary-school
textbooks. He has blanket power to act on anything. Another example: The Polish govern-
ment decided in the early 1960s to rebuild the center of Warsaw. An architectural competition
was announced, and the winning project was selected and approved by the government. But
one of the secretaries of the Communist party decided that the proposed buildings would
compete with a Stalinist monster that dominates the city and ordered their height reduced. He
could have done anything else he wanted.
64
This is not to say that retrospective action is not possible under democracy: The presi-
dent may appoint a surgeon-general, who may charge a group of experts with preparing a
report on AIDS; the report may be publicized; and the president may disclaim the report or even
fire his appointee. But we know ex ante that the president can do all this; he has the right to
repudiate, and he has the power to fire a member of his administration. He cannot repudiate,
however, a ruling of the Supreme Court or fire a Justice, and we know that, too. What I am
arguing is that under dictatorship we cannot know ex ante what the power apparatus can and
cannot do, because the feasible outcomes are not entailed by any set of rules.
https://doi.org/10.1017/CBO9781139172493.003
tween law and policy.65 In this sense, dictatorships are arbitrary. Under
democracy, an outcome of the democratic process can be overturned ex
post if and only if it violates previously established and knowable rules;
under dictatorship, the possible outcomes are not entailed by any set of
rules.
Does this argument imply that democracies generate less uncertainty
than dictatorships? I think this question cannot be answered, because the
response depends on the point of view.66 The difference is in the assump-
tions one must make to deduce the outcomes. In a dictatorship, they are
deduced only from the preferences of one actor; in a democracy, from
conflicting preferences and rules. Under a sufficiently capricious leader or
a sufficiently divided power apparatus, the authoritarian regime may keep
bewildering everyone with its twists and turns.67 Indeed, under dic-
tatorship the outcomes may be unpredictable: They can be predicted only
by knowing the will of the dictator or the balance of forces among the
conflicting factions. A democratic regime may, in contrast, yield highly
predictable outcomes even when parties alternate in office. Hence, ex post
an authoritarian regime may exhibit more variation of policies than a
democratic one. But examine the situation ex ante. Under dictatorship,
there is someone who is certain about the outcomes, and anyone who
knows what the power apparatus wants also knows what will happen.68
Under democracy, there is no such actor. Hence, the difference in uncer-
tainty is conditional in the following sense: In an authoritarian system it is
certain that political outcomes will not include those adverse to the will of
the power apparatus, whereas in a democracy there is no group whose
preferences and resources can predict outcomes with near certainty. Cap-
italists do not always win conflicts processed in a democratic manner,69
and even one's current position in the political system does not guarantee
65
This is the feature that Montesquieu saw as the fatal weakness of despotism.
66
For a spirited statement of a subjectivist approach to game theory, see Rubinstein 1988.
Rubinstein argues that if game theory is to make sense of the world around us, we should
interpret games not as physical descriptions but as assumptions about the perceptions and
reasoning procedures of the actors. Hence, what may be certain from the point of view of an
observer may appear uncertain from the vantage point of each actor.
67
Here is a Soviet view of the matter: Three men meet in a gulag. One asks another,
"What are you here for?" "I was against Radek," he says. "And you?" "I was for Radek."
They turn to the third man, thus far silent. "I am Radek," he says.
68
Assuming, obviously, that nature does not throw dice.
69
This is not an allusion to Marx, who argued in his writings on the 1848-51 period in
France that universal suffrage represents a perpetual threat to capital. It is instead an allusion
to Lenin, whose views were summarized above.
https://doi.org/10.1017/CBO9781139172493.003
48 DEMOCRACY
future victories. Incumbency may be an advantage, but incumbents do

lose.
Hence, instrumental action under authoritarianism is limited to cases in
which those actors who enjoy room for maneuver know that the power
apparatus is indifferent to some outcomes. Party secretaries from particular
localities may compete, for example, to get an investment provided for in
the plan; producers' associations from different sectors may defend them-
selves against the lowering of tariffs on competing imports. Acting instru-
mentally makes sense for them only if they know that the power apparatus
will not punish them for their actions and that it can tolerate the outcome
they want. It would be irrational for anyone to act as if the outcome were to
be determined by his or her actions under the existing institutional frame-
work. Everyone has to try to anticipate the reaction of the power appa-
ratus. 70
To test these distinctions, consider the following example. After 1954,
the Polish Communist regime regularly changed its agricultural policy.
Whenever peasants stopped producing food for the cities, the party would
tell them, "Enrich yourselves." And whenever peasants enriched them-
selves and their consumption began to appear ostentatious, the party would
confiscate all the riches. Hence, the policy followed predictable cycles:
Low productivity led to fiscal stimuli, visible inequality led to punitive
taxation, and so on. 71 Now, we could imagine a similar dynamic under
democracy: The Productivity party would campaign for fiscal stimuli; the
Equality party would advocate taxing the rich peasants. When food was
scarce, the Productivity party would win elections until peasants got too
70
Yet note that authoritarian regimes systematically hide information about their true
preferences. Their main concern is not to make it public that there are divisions within the
power apparatus or even that any counterarguments were considered legitimate in the discus-
sions inside the apparatus that led to a particular decision. What is communicated publicly is
only "the line": a decision portrayed as unanimous and undisputable. Yet for any educated
observer, the line is not credible information about the preferences of the rulers. I owe this
observation to conversations with Tang Tsou.
The secrecy of the power apparatus sometimes reaches the grotesque. When Chernenko
died, Soviet radio did not announce the fact for a day and a half; they let it be guessed by
playing only solemn music on the radio. In the meantime, he Monde announced the death of
another member of the Politburo and reported the rumor that yet another had been ousted. The
Soviet people did not know whether or not the dictator was still alive: Gabriel Garcia
Marquez's The Autumn of the Patriarch was performed in real life on the other side of the
globe.
71
Eventually, peasants did learn not to invest; workers were starved and threw party
bureaucrats out. But it took forty years.
https://doi.org/10.1017/CBO9781139172493.003
rich, whereupon the Equality party would be victorious. Ex post, there-

fore, the policy cycles, and the posterior probability that the tax rate is t
percent, may be identical in the two systems.
Ex ante, however, the uncertainty inherent in the two systems is different
in three ways. First, the party changed the rules ex post: The central instru-
ment of its policy was the domiar, a retroactive surtax. Income was earned
by peasants when the tax rate was 40 percent; long after this income was
earned and well after it was invested or consumed, it would be subjected to
an additional tax. This can happen under democracy, but only according to
established rules that permit retroactive taxation. Under dictatorship, it can
happen despite the rules. Second, the timing and the amount of the confis-
catory retribution was arbitrary in the sense defined above: It was not en-
tailed by any set of rules. Under democracy, peasants may expect that
when inequalities become conspicuous the tax rate will increase, but they
can also expect that the rules will change only according to rules. Finally,
under democracy, the new tax rate is determined jointly by the political ac-
tions of peasants and of other forces. Peasants can participate in determin-
ing the new tax rate; they can defend their interests. Given their reading of
public opinion and their knowledge of the rules, they can attach prior prob-
abilities to increases by any particular amount. Hence, they can calculate
expected values and act upon them when deciding how much to invest.
Under dictatorship, all they can do is to guess what the party will tolerate;
if they cannot guess, they do not know when they will get hit and by how
much.
None of the above implies that peasants will be better off under democ-
racy. If the power apparatus wants to develop agricultural production and if
it is willing to tolerate wealth, peasants will prosper. They will prosper
even if other people starve and even if everyone else would prefer lower
agricultural prices. Their interests are guaranteed by the will of the dic-
tatorship; but this is all their interests depend upon. There is little they
can do. 72
Democracy is thus a system that generates the appearance of uncertainty
because it is a system of decentralized strategic action in which knowledge
is inescapably local. Dictators are observers because they do not have to
consider what others think about them. If others guess incorrectly what the
72
The NEP is the obvious example. Told by Lenin to enrich themselves, the Soviet kulaks
waited for his death and the defeat of Bukharin to be massacred by Stalin.
https://doi.org/10.1017/CBO9781139172493.003
50 DEMOCRACY
dictator wants, he or she can correct the outcomes ex post. In turn, every-
one who knows what the dictator wants can predict what will happen.
Under democracy, no one is the dictator. Hence the appearance of uncer-
tainty.
https://doi.org/10.1017/CBO9781139172493.003

t.1 Przewoski-Democracy

Uploaded by

Copyright:

Available Formats

t.1 Przewoski-Democracy

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

t.1 Przewoski-Democracy

Uploaded by

Copyright:

Available Formats

1.

are regularly terminated under established rules. They are "terminated"

How are outcomes enforced under democracy?

The question: democracy, rationality, and compliance

Competing views of compliance

(1) Spontaneous self-enforcing outcomes, or equilibria. Each actor does

signal is "red"; the best response to red is to wait (the alternative is

(2) Bargains, or contracts. An outcome is such that at least one actor

self-interest. Democracy can be an equilibrium: a system of "self-govern-

The point about institutions as codes of punishment is more complex.

Let me suggest schematically how spontaneous decentralized self-

democracy may be a source of commitment among those who see little

have to violate some property rights - for example, by adopting land

The relationship between constitutions and political reality is not an

consolidation of democracy may be a joint effect of conditions and institu-

some of their power to institutions? When will they conclude a "democrat-

cal forces: previously constituted collective organizations, some categories

Appendix: Why do outcomes appear uncertain?

King of hearts Any other card

R | Ace of spades Nothing, All All, Nothing

Another reason outcomes can be uncertain is that actors do not know

One essential feature of authoritarian regimes is that someone has an

future victories. Incumbency may be an advantage, but incumbents do

rich, whereupon the Equality party would be victorious. Ex post, there-

You might also like