legitimacy. In turn, at least according to democratic theory, a le-
Votes and Violence
19
gitimate government thereby acquires certain rights. A legitimate
government has a mandate to do what it said it would do, and this
entitles it to face down opposition to the implementation of its pro-
gram, at least within limits. In a democracy citizens agree to these
rules, and so opposition to a government’s elected program cannot
legitimately extend to the use of violence. This provides a further
reason for the reduction in political violence. Even if the most ex-
treme opponents of the government do not accept that the govern-
ment is entitled to enact its program, they will find it more difficult
to enlist mass support for violent opposition. They can no longer
reasonably claim that their struggle is just.
Democracy should thus deliver a double whammy against po-
litical violence: there is less objective basis for grievance, and for any
given grievance it should be harder to persuade people to resort to
violence against the government.
So confident have we been in asserting that democracy is the
answer to political violence that it seems almost churlish to look at
the evidence to test whether it is right. The peace-promoting ben-
efits of democracy have become one of the fundamental certainties
of the policy world, indeed perhaps one of the few unifying beliefs
across the political spectrum. George Soros and George Bush have
not agreed on much, but I suspect that they would be on the same
side on this one, along with millions of other people.
When the countries of the bottom billion started to democra-
tize I was as enthused as anyone. However, the ensuing years have
been more difficult than I had expected. I have little time for outside
commentators who turn into tut-tutting judges. Change is difficult
and there are strong forces resisting it. It is not that the societies of
the bottom billion have failed to live up to my expectations. Rather,
I was coming to suspect that I had missed things that in retrospect
were becoming evident. Indeed, there had surely been people with
doubts all along, but their voices had been drowned out in the ca-
cophony of enthusiasm for democracy. Essentially I came to suspect
20
WARS, GUNS, AND VOTES
that theories that were entirely appropriate for countries that were
more developed might have been overextended. The societies of the
bottom billion may simply be lacking the preconditions whereby the
accountability and legitimacy effects were going to work very well.
I have to say that I came to these doubts with deep reluctance. But it
was time to turn to the evidence.
You might expect that the relationship between democracy and
political violence would be settled academic territory. But somewhat
to my surprise I found that it was not. It was, in fact, about as close
to terra incognita as modern social science gets: I could not find a
single published paper. I teamed up with Dominic Rohner, a young
Swiss researcher, and got to work.
We got data on virtually all the countries in the world for the
period since 1960. Controlling for the other characteristics that were
likely to matter, how did democracy affect the incidence of political
violence? At first we could find no relationship. To me this nonre-
sult seemed intrinsically unlikely: surely something as salient as the
political regime simply had to matter. Then it occurred to us that the
relationship might well not be the same across the entire range of
economic development. After all, the societies of the bottom billion
were highly distinctive in being far poorer than the other democra-
cies. Maybe in poor countries the effect of democracy on violence
was not the same as in rich countries. Once we introduced this pos-
sibility we found that the political regime always mattered. In fact,
democracy had the opposite effect in poor countries to that in rich
countries. It was because the two effects were opposing that there
had appeared to be no effect at all. So what were the two opposing
effects?
We found that in countries that were at least at middle-in-
come levels, democracy systematically reduced the risk of political
violence. The prediction of the accountability-and-legitimacy view
of how democracy should make a society more tranquil was borne
out. But in low-income countries, democracy made the society more
Votes and Violence
21
dangerous. As if poverty was not miserable enough in itself, the ef-
fect of democracy adds insult to this injury. Whereas in societies that
are not poor it enhances their already safer conditions, in poor soci-
eties democracy amplifies the already severe dangers.
If democracy makes poor societies more dangerous, but societ-
ies that are not poor safer, there must be some threshold level of in-
come at which there is no net effect. The threshold is around $2,700
per capita per year, or around $7 per person per day. The societies
of the bottom billion are all below this threshold: most of them are
a long way below it.
To my mind the key implication of these results was that the
accountability-and-legitimacy theory of how democracy would help
the societies of the bottom billion must be missing something. In-
deed, it must be missing an elephant. Much of this book is devoted
to flushing out that elephant. But I have not quite finished with the
results of our investigation.
Recall that at higher levels of income societies are safer. It turns
out that all the benign effect of higher income depends upon the
society being democratic. Indeed, it is more striking than that: in the
absence of democracy, as a society starts to get rich it becomes more
prone to political violence. Democracies get safer as income rises,
whereas autocracies get more dangerous. If it helps, you can think of
this as two lines, an upward-sloping one showing how democracies
get safer as income rises, and a downward-sloping one showing how
autocracies get less safe. The level of income at which democracy
has no net effect on violence, $2,700, is simply the point at which
these two lines cross over. Applying this to the society with the most
astounding income change of our times, China has now passed the
income threshold—per capita income has soared past $3,000. So, if
China runs to form, year by year its spectacular economic growth is
now making it more prone to political violence unless it democra-
tizes.
Our initial work had been pretty heroic in the sense that we
22
WARS, GUNS, AND VOTES
had hastened over a host of statistical cans of worms. Much of our
work now turned to opening these cans and seeing if the results
survived. For example, income is likely to be affected by both con-
flict and the political regime. Causality might in fact be running
in the opposite direction to our interpretation. We checked on this
and satisf
ied ourselves that this was not the explanation: our results
were not spurious, at least not on this count. In the small world of
the statistical study of political violence, the foremost rival team has
been James Fearon and David Laitin at Stanford. Like us, they had
a model of the factors that tend to produce violence, but it differed
in detail from our own. We decided that a good test of the result
that democracy increased the risk of violence for the bottom billion
would be to see whether it survived if we introduced it into their
model. Unfortunately for these societies, it did survive. To my mind
the most remarkable result came when we investigated a range of
different forms of political violence. We looked at assassinations, ri-
ots, political strikes, and incidents of guerrilla activity as well as full-
blooded civil war. To my amazement, the same pattern was true for
them all: at low income, democracy increased political violence.
I do not believe that these results reveal unalterable relation-
ships: later I will argue that democracy can be made to work in the
societies of the bottom billion. But consider for a moment what
would be the implication if they were unalterable. They would im-
ply that judged by the objective of peace, there would be a preferred
sequence for economic and political change. The ideal stage at
which to democratize would be once a society had already reached a
moderate level of development.
As Dominic and I digested these results we started to puzzle
over the obvious question: why? The question actually decomposes
into three distinct puzzles. First, why was the benign effect of de-
mocracy that reduced the risk of political violence dependent upon
the level of income: what was it about income that made democracy
differentially peace-promoting in richer societies? The second was
Votes and Violence
23
the converse question of why autocracies become more dangerous at
higher levels of income. Finally, and most mysteriously, once these
income-related effects of democracy and autocracy were allowed for,
there remained a further pure effect of democracy that was making
societies more at risk of violence. Like some unobservable dark mat-
ter it was lurking as a constant across societies. What was it? These
were not easy questions.
The key insight came by the simple psychological technique of
imagining myself in the position of being a former dictator in one
of the countries of the bottom billion who had caved in to pressure
from donors to democratize. How had I kept the peace before and
how did democratization change my problem? I was evidently not
the first person to wonder about how a dictator might best stay in
power. Herodotus reports that when Periander became the young
dictator of Corinth, he sent a messenger to the old and experienced
dictator of Miletus, Thrasybulus, for advice. Thrasybulus had clung
to power very effectively; had he any tips for someone just start-
ing out on the same career? Thrasybulus took Periander’s messen-
ger into a field of corn and, as he talked, repeatedly and systemati-
cally snapped off the heads of all the tallest stems. The messenger
returned baffled, but Periander got it. Although social science has
advanced in the two and a half thousand years since Herodotus,
I think that this still gives a pretty fair take on the technology of
power retention. If we are to generalize from Thrasybulus, the key
is to be preemptive: purge potentially dangerous people before they
act. Does democracy affect my ability to undertake such purges?
Well, the awkward problem with preemptive purges is that they are
not compatible with the rule of law: the technique depends upon
punishing people even though they haven’t done anything. This sort
of conduct collides with even fairly modest levels of democracy.
The idea that the ability to mount a purge would be reduced as
a result of democracy was a plausible explanation for the dark mat-
ter. If leaders could no longer mount preemptive purges they might
24
WARS, GUNS, AND VOTES
be less able to keep the lid on political violence. This might be why,
over and above those effects of democracy that depended upon the
level of income, there was the pure effect that increased political
violence. Herodotus had given us an idea; now it was time to test it.
We turned to a large political science data set on purges. Believe
it or not, these things are measured, country by country, and year by
year. We wanted to see whether democracy made purges more dif-
ficult, controlling for other possible influences. Sure enough, even
a modest degree of democracy radically reduces the frequency of
purges. From the perspective of keeping the peace through repres-
sion, democracy is a massive technological leap backward.
If you want a practical, real-world, up-to-the-minute example
of how democratization can make it harder to keep the peace, try
Iraq. Whatever the limitations of the present regime, it is clearly
massively more democratic than that of Saddam Hussein. Yet Hus-
sein presided over a relatively peaceful country. It was not an attrac-
tive peace, but it was a peace of sorts, and it most surely depended
upon preemptive repression rather than citizen consent.
So a weakening of technologies of repression is, I think, a likely
explanation for the dark matter: the higher risk of political vio-
lence that comes from democracy. Why, then, should the net effect
of democracy be increasingly favorable as income rises? I think the
answer lies in those effects that I started with: accountability and
legitimacy.
The stark and straightforward reason that in the bottom billion
the accountability and legitimacy effects of democracy do not reduce
the risk of political violence is that in these societies, democracy does
not deliver either accountability or legitimacy. So why does it fail to
do so?
O v e r t h e y e a r s I h av e had some very smart students, but undoubtedly the smartest was Tim Besley, now a highly distinguished
Votes and Violence
25
professor at the London School of Economics and a former editor of
the American Economic Review. Tim’s book Principled Agents? is the most serious theoretical attempt to answer the question of whether
having to face voters actually disciplines politicians. It is a compli-
cated book, but I think I can give you the gist of it. In our own
societies the answer to Tim’s question seems pretty obvious. If an
incumbent politician had not even tried to deliver what people want, electors would notice. The actions of political leaders are scrutinized
by the media, and if a politician were consistently to advance his
own interests at the expense of ordinary citizens he would not be
reelected. Politicians want to stay in power. Partly, let us hope, this
is because they feel a sense of vocation to do good, but also pretty
obviously because it is their choice of lifestyle: it is their profession,
and they do not want to
be unemployed. And so, between media
scrutiny and politicians’ appetite for power, political leaders are pin-
ioned to trying hard for the common good.
But in the societies of the bottom billion conditions are often
not like this at all. Suppose that voters have precious little knowl-
edge about the choices they face. Even the past performance of the
incumbent, which voters have just lived through, will typically be
open to multiple interpretations. Perhaps bad outcomes were due to
mitigating circumstances; perhaps the government was not to blame.
All too often, in the volatile economies of the bottom billion, this is
genuinely the case: the economy frequently gets derailed by shocks
beyond local control. A typical shock is that the price of the coun-
try’s export good crashes and the economy consequently collapses. I
can think of three African democracies in which this happened in
the run-up to an election. In each case the incumbent government
had done a pretty good job. One was in Benin during the run-up
to the 1996 election, removing a reforming president. It happened
again in Uganda in the run-up to the 1998 election: the world price
of coffee crashed. And it happened in Madagascar before the 2006
election with a combination of falling export prices and soaring costs
26
WARS, GUNS, AND VOTES
of imported oil. How was the electorate to tell whether the economy
crashing around their ears was crashing because of an unavoidable
external shock or because the government had been incompetent?
Of course, the government tried to explain, but governments had
always made excuses. How were they to know what to believe?
In addition to the problem of lousy information, perhaps some
voters are going to vote for or against the incumbent regardless of
performance because of their ethnic identity. Identity is the basis of
most voting in the bottom billion. Their societies are usually divided
into competing ethnic identities, and as a result ethnicity is by far the
easiest basis on which to organize political loyalty. The problem with
it is that because the loyalty isn’t issues-based, it isn’t performance-
based either. Votes are simply frozen in blocs of rival identities. A
consequence of having great blocs of votes frozen into support or
opposition is that the vote that an incumbent politician attracts is not
Paul Collier - Wars, Guns, and Votes Page 3