Public Policy, Price Shocks, and Conﬂict: Price Shocks and Civil War in Developing Countries∗
Brett L. Carter† Robert H. Bates‡ November 7, 2011
Abstract Those who study the role of agriculture in the political economy of development focus on government policy choices on the one hand and the impact of price shocks on the other. We argue that the two should be studied together. We ﬁnd that civil unrest (Granger) causes government policies, pushing governments in poor and medium income countries to shift relative prices in favor of urban consumers. We also ﬁnd that while civil wars are related to food price shocks, when government policy choices are taken into account, the relationship disappears. We thus learn two things: Policies that placate urban consumers may inﬂict economic costs on governments, but they confer political beneﬁts. And when estimating the relationship between price shocks and political stability, equations that omit the policy response of governments are misspeciﬁed.
∗Research on this project was launched with support from the Political Instability Task Force. Our thanks go to other members of the initial research team: Colin Brown, Didi Kuo, Richard Nielsen, and Mircea Popa. We also thank Adam Glynn, Stephen Block, and participants of the Civil Conﬂict Workshop at Harvard University for their timely suggestions.
†Ph.D. Candidate, Department of Government, Harvard University. Email: blcarter@fas.harvard.edu. ‡Eaton Professor of the Science of Government, Department of Government, Harvard University
1

Table 1: Food Riots, 2008

Bangladesh Brazil Burkina Faso Cameroon Cote d’Ivoire Egypt Ethiopia Haiti India Indonesia Mexico

Mozambique Pakistan Myanmar Panama Philippines Russia Senegal Somalia Tajikistan Yemen

1 Introduction
Particularly in the developing areas, governments treat food supplies as a security matter. This was certainly the case in the West, when countries that are now developed were themselves developing.1 It remains true today. Thus Table 1, which lists countries that experienced food riots during the price rise of 2007-2008. Recent events in North Africa and the Middle East provide additional evidence of the political importance of food prices.
In this article, we seek to move beyond qualitative accounts and salient examples to provide statistical evidence of the link between food shortages and political security.
In section 2, we brieﬂy review the literature. In section 3, we describe our sample and our measures and in section 4 our methods of estimation. A novel feature of the latter is the incorporation of the expected policy response of governments into equations that capture the political response to food price rises. In section 5 we present our ﬁndings, the most striking of which is that the relationship between price shocks and civil wars becomes insigniﬁcant when government policies distort markets in ways that favor the consumers of agricultural products.
2 Literature
Two literatures inform this article. The ﬁrst addresses agricultural policy and explores the manner in which governments intervene in markets so as alter prices. The second explores the relationship between food price rises and political stability. In the sections that follow,
1See, for example, Tilly (1971), Tilly (1990), and Kaplan (1976).
2

we demonstrate that, while commonly pursued separately, for reasons both of methodology and substance, the two themes should be addressed together.
2.1 Agricultural Policy
Writing in 1977, Michael Lipton highlighted the degree to which governments in the developing areas adopted policies biased in favor of the urban and industrial sector. While Lipton (1977) addressed a wide range of policies, others have focused more narrowly, concentrating on how governments intervene in agricultural markets. Some (Johnson (1973)) focus on international markets and trade policy; others, on domestic markets, both for agricultural products and farm inputs (Schultz (1976)). Still others probe the political foundations of agricultural policy. The literature on interest groups (Bates (1981); Olson (1985); see also Hayami (1957) and Hayami (1986)) stresses the ironic diﬀerence in the power of the farm lobby in developed and underdeveloped countries. In the former, producer interests dominate, despite their small numbers and the small size of agriculture in the national economy; in the latter, consumer interests dominate, even though the urban sector constitutes a small proportion of the total economy and industrial and manufacturing ﬁrms are few in number. Varshney (1993) notes, however, that the political advantage enjoyed by urban interests in the developing world erodes in democracies, where the size the rural population confers a political advantages (see also Bates and Block (2009)).
In the sections that follow, when we address public policy, we make use of the Relative Rate of Assistance (Anderson (2009)), which provides a measure of the degree to which governments intervene in favor of rural or urban producers. Using this measure, we explore the impact of diﬀerences in the economic structure and political institutions upon the policy choices of governments.
2.2 Commodity Price Shocks
Deaton and Miller (1995) number among the ﬁrst systematically to explore the political impact of food price shocks. For each country in Africa, they constructed an index in which they weighted each of 21 commodities by its relative importance in the total value of that country’s exports. To deal with the possible impact of endogeneity, they made use of instrumental variables. Applying vector auto-regression to data from 32 African countries, 1960-1990, Deaton and Miller (1995) ﬁnd the impact of commodity price shocks largely favorable to Africa’s economies. Total GDP increases. So too do each of its components:
3

consumption, investment, and government expenditure. The researchers then introduce a measure of “irregular exit”2 as their index of political instability. Commodity price shocks aﬀect this measure, they note, through their impact on national income. It is when prices turn negative and national incomes decline, they ﬁnd, that the likelihood of irregular exits rises.
In contrast to Deaton and Miller (1995), Dell et al. (2008) report that food price rises lead to declines in income, reductions in the growth rate of the economy and increases in political instability and that these relationships prevail only among poor countries. The diﬀerence between two sets of ﬁndings can be attributed to the instrument which Dell et al. (2008) employ – increases in local temperatures – that would correlate with local crop loss rather than with movements in global markets. In Deaton and Miller’s world, were a country to produce rice, it would beneﬁt economically from an increase in rice prices; the value of its the crop would rise. In the world of Dell et al. (2008), a weather shock signals the possibility of a crop failure, with losses of output and a rise in food prices, but a fall in the national product.
As did Deaton and Miller (1995), Dell et al. (2008) employ the probability of irregular exit as a measure of political instability. The probability of irregular exit rises with temperatures, they report, and thus with food prices. They also ﬁnd temperature increases to be associated with political “interregnums” or periods in which no discernable government holds power.
Writing in 2008, Besley and Persson (2008) construct country speciﬁc Laspeyres indices for a global sample of roughly 125 countries for 45 commodities and for the period 1960 to 2000. They ﬁnd a positive and signiﬁcant relationship between their index and the occurrence of civil wars. Decomposing the bundle of commodities into agricultural goods, petroleum products, and minerals, they attribute this result largely to the impact of price rises for farm products. The relationship is stronger, they ﬁnd, in poor countries and in nondemocracies.
The last contribution is that of Arezki and Bruckner (2011), who study the relationship between a food price index based upon international prices and export shares on the one hand and a series of political variables on the other: Polity’s measures of democracy and political restraint; Banks’ data on riots and anti-government demonstrations, and PRIO’s measure of civil unrest.
As do Dell et al. (2008), Arezki and Bruckner (2011) ﬁnd that food price shocks impact measurably on poor but not on high income countries. Based on estimates drawn from a
2Irregular exits include assassinations, coups, and the forced resignation of the head of state. See also Bienen and van de Walle (1991)
4

panel of data drawn from 61 low income countries, 1970-2007, they ﬁnd food price shocks systematically related to declines in Polity scores, increases in riots and demonstrations, and increases in the likelihood of civil unrest. Their ﬁnding is robust to the exclusion of large food producers, which they list as China, Guatemala, India, Ivory Coast, Pakistan, Thailand, Uganda, Ukraine, and Vietnam. They make no eﬀort, however, to control for movements in the prices of other commodities, such as petroleum products.
Portions of this literature also focus on particular countries, such as Colombia (as by Dube and Vargas (2006) and Angrist and Krueger (2008)), or on the role of climate (as by Miguel et al. (2004) and Hsiang et al. (2011)). Given the focus of this paper, we do not review their contributions here.
2.3 Discussion
We build on the literature in several ways. Given that the prices of other goods – and especially that of petroleum – can also aﬀect national incomes and political stability, we strive to isolate the impact of changes in food prices from that of changes in the prices of other commodities. Unlike many of the previous papers, we include measures of energy use and production in our estimates. It is particularly important to do so as the prices of diﬀerent commodities may vary in concert, particularly when responding to monetary imbalances.
More signiﬁcantly, perhaps, this paper departs from many of its predecessors in that it combines the two literatures. That is, while focusing on the impact of food price shocks, it takes into account the policy response of governments. Doing so, it ﬁnds and reports evidence that the impact of food price shocks is itself a function of the policy choices of governments.
We thus seek to contribute to two literatures, both by exploring the policy response of governments and the manner in which their responses shape the impact of price rises upon domestic stability.
3 Sample and Variables
Our measure of government policies, calculated by Anderson (2009), exists only for a subset of lower income nations; moreover, we found – as did Dell et al. (2008) and Arezki and Bruckner (2011) – that food price shocks fail to aﬀect political outcomes to a measurable degree in
5

higher income countries.3 Our dataset therefore consists of roughly 1, 100 observations, drawn from approximately 30 developing countries between 1961 and 2001.4
As our measure of price shocks, we created import and export Laspeyres indices, using FAO measures of food consumption and World Bank data on international reference prices5. These indices provide measures of food price imports and exports for each country and for each year. For country i and year t, the expression of the Laspeyres index reads

Laspeyresit =

(pct × qci1980) (pc1980 × qci1980)

where pct gives the global market price of commodity c in year t and qcit gives the quantity consumed (exported) of commodity c in country-year it’s consumption (export) basket.6
The data cover crops that account for nearly 70% of the calories consumed globally. We employ 1980 as the base year.7
Figure 1 employs data from 12 countries to illustrate changes in the value of our aggre-
gated food price index over time and country. Clearly, the boom of the 1970s accounts for
a major portion of the temporal variation. And since countries confront the same global
food prices and choose relatively similar consumption baskets, the major portion of the total
variation occurs cross-temporally. That this is the case facilitates causal inference. For since
all countries receive roughly the same treatment, we can attribute country level variation
in responses to price shocks to diﬀerences in their political institutions and socioeconomic
structures.
As can be seen in Figure 1, despite occasional sharp upward movements, food prices
have been steadily declining. When estimating the relationship between our price index
and measures of political instability, we therefore experimented with a variety of corrections
for trend. None improved the ﬁt, however; indeed, they reduced it. While we shall report
evidence of a relationship between the price index and measures of political instability, these measures failed to correlate with trend-ﬁltered versions of the index.8
3A ﬁnding that we, like they, attribute to Engel’s law, by which poorer people spend a larger proportion of their incomes on food.
4Of the 94 low income countries in our dataset, we observe the outcome variable in the ﬁrst stage of our model for only 32. Hence we are reluctant to employ imputational methods: This would entail imputing values of the ﬁrst stage outcome variable to some 60 countries for which no observations were recorded.
5Anderson (2009) 6For more, see Deaton and Muellbauer (1980). 7For each year, we dropped countries whose production accounted for 20% or more of the major food crops. An important exception occurred when we dropped price setters in rice markets, strongly suggesting that domestic insecurity signiﬁcantly aﬀects food prices in rice markets. 8It is as if consumers were responding politically to food price rises, even when these increases represented

6

We employ two measures of political instability. The ﬁrst is the Peace Research Institute’s (PRIO) measure of civil war, for which there are three variants: PRIO 25, which assumes a value of 1 if a civil war in a given year results in at least 25 battle deaths; PRIO 100, which assumes a value of 1 if a civil war results in at least 100 battle deaths; and PRIO 500, which assumes a value 1 if a civil war yields at least 500 battle deaths. Use of the three measures enables us to probe the impact of food price shocks not only upon the onset but also the intensity of civil wars. A second is Arthur Bank’s conﬂict index, which we draw from his Cross-National Time Series Data Base. For each country and each year, Banks records the occurrence of strikes, assassinations, revolutions, riots, and purges, which he then weights according to his assessment of their impact on political stability. We explore the relationship of food price shocks to the natural log of Banks’ variable, seeking thereby to reduce the impact of potentially inﬂuential outliers.
As a measure of government policy, we employ an index of the relative rate of assistance (RRA), as calculated by the World Bank.9 The index reﬂects the relative magnitude of the shifts of the nominal rates of assistance (NRA) between agriculture and other sectors of the economy. More formally, for purposes of illustration, if the sole intervention by government is an ad valorem tariﬀ, the NRA for a given commodity is then

NRAit

=

Eit × Pt (1 + τmit) + Eit × Pt Eit × Pt

where τmit is the tariﬀ rate applied by country i in year t on competing imports m, Eit is the country-year speciﬁc domestic currency price of US dollars (exchange rate), and Pt is the dollar denominated global price of the commodity in year t. The measure thus captures the diﬀerence between global and domestic prices produced by the government’s tariﬀ.
The measure provides a means for calculating the relative rate of assistance (RRA), which is the index we employ to characterize government policy.10 For country i in year t, the RRA is

RRAit

=

1 + NRA/agricultureit − 1. 1 + NRA/nonagricultureit

The RRA is thus a function of the relative extent to which the government separates domestic from international prices for agricultural and non-agricultural commodities. When

but a return to a long-term equilibrium level. 9Anderson (2009) 10The measure can, and has been, adopted to incorporate additional forms of government intervention.

7

government policies favor agricultural producers to a greater degree than others, then this measure is positive; when they instead favor consumers, it is negative. Put another way, positive values suggest that the government favors farmers; negative, that it is biased in favor of urban interests.
As our measure of political institutions, we employ Political Competition, a variable taken from Polity IV. We do so because the ﬁrst-best measure of political competition, that of Beck et al. (2002), is not available in suﬃcient historical depth. Fortunately, Political Competition correlates closely (0.80) with the latter measure, enabling us to employ it instead.11
Lastly, we introduce measures of the importance of petroleum in the national economy and of monetary instability, which serve as control variables. Price shocks in food markets can be the result of monetary forces as well as physical shortages; and when that is the case, food prices move in concert with those of other commodities. Of these commodities, it is perhaps only petroleum that is able to induce political shocks comparable in magnitude to those resulting from food. In our eﬀorts to identify the impact of food prices, we therefore control for these confounding variables.
Table 3 provides a full list of the variables, describes them, and indicates the sources from which they were taken.
4 Speciﬁcation
To estimate the impact of food price shocks, we employ a two-stage, simultaneous equations model. The ﬁrst stage captures the policy response of the government; the second, the impact of price movements on political instability, conditional upon the government’s response.
4.1 First Stage Equation
By imposing tariﬀs, levying export taxes, and issuing licenses and permits, governments can limit trade and so separate the level of prices in domestic and foreign markets. Using these measures, governments can also alter relative prices for rural producers and urban consumers. The purpose of the ﬁrst stage equation is to account for the manner in which governments intervene, as captured by the RRA.
11The measure, collected by Beck et al. (2002), was developed by the Africa Program at Harvard University, who demonstrated that it Guttman scaled. It is also invertible, i.e. given a number on the scale, an observer can know the structure of political competition precisely.
8

We view the government’s policy response as being shaped by the economic and political characteristics of the nations in our sample, which is drawn from the ranks of the low and low-medium income countries. We learn from the literature that in such economies rural dwellers are numerous; if the government must compete for votes in order to stay in oﬃce, it will therefore have to consider letting domestic food prices rise, thereby generating an increase in real income for rural producers. On the other hand, urban consumers are spatially concentrated; by comparison with their rural counterparts, they therefore face low costs when organizing. Because they staﬀ key services, moreover, their power is disproportionate to their numbers. And not only do they spend a large portion of their incomes on food; but also, so do others. As food is a necessity, when its price rises, people spend less on other commodities, resulting in decreased purchases of non-essential items, recession in the manufacturing and retail sector, and rising unemployment.12 Given these eﬀects, we can expect urban consumers of food vigilantly to monitor food prices and vigorously to lobby for policies designed to keep them low. Indeed, in the absence of electoral competition, they can be expected to dominate the political process. Our choice of sample and our speciﬁcation of the policy equation is therefore designed to take into account the structure of the economy and the nature of political institutions: factors that aﬀect the policy response of governments.

Policyit|Income Level = αC + αT + α1Export Laspeyresit + α2Import Laspeyresit

+α3Political Competitionit

+α4Export Laspeyresit × Political Competitionit

+α5Import Laspeyresit × Political Competitionit

+α6Policyit−1 + αX

(1)

We estimate four variants of this equation. The results appear in Table 4. Models 1 and 2 are estimated with straightforward OLS, with the lagged RRA value and country and year ﬁxed eﬀects omitted. Model 2 includes robust and clustered standard errors, which correct for dependence within country observations. Models 3 and 4 add, respectively, country and year ﬁxed eﬀects and the lagged policy value.
12See, for example, Sen (1983).

9

4.2 Second Stage Equation
The estimates of Equation 1 yield the expected policy responses of governments to food price shocks, conditional upon the political and economic environment within which they inhere and (in the case of model 4) previous policy commitments. Equation 2 introduces these responses into models that capture the political impact of food price shocks. The introduction of an instrumented variables in the ﬁrst stage should purge the coeﬃcient relating government policy and political outcomes in the second of endogeneity bias.
As Wooldridge (2002) demonstrates, 2SLS is equivalent mathematically to regression using instrumental variables; the second stage estimates will be consistent, therefore, if the instrument is highly correlated with the instrumented explanatory variable. Because model 4 in Table 4 incorporates a lagged value of the dependent variable, estimates derived from it should generate consistent estimates in the second stage equation. The problem of inconsistency arises once again, however, because of the need to cluster standard errors by country; clustering reduces the precision of the ﬁrst stage estimates. When estimating the second stage, we therefore use the IV-Generalized Method of Moments (GMM) estimator developed by Hansen (1982) and extended in Baum et al. (2003) and Baum et al. (2007). The combination of country level clustering and IV-GMM ensures that our estimates correct for dependence among intra-country observations without sacriﬁcing the precision aﬀorded by a strong instrument.13
The system of equations is overidentiﬁed. In overidentiﬁed settings, standard IV techniques reduce the available instruments to the k necessary for exact identiﬁcation, sacriﬁcing eﬃciency in the process. An additional advantage of the IV-GMM estimator is that it employs information from all available instruments, thereby enhancing the eﬃciency of the second stage estimates.
The general form of the second stage equation can be written as:

yit|Income Levelit = βC + βT + β1Policyit + β2Export Laspeyresit + β3Import Laspeyresit

+β4Political Competitionit + βZ

(2)

13In the process, it also creates a weighting matrix such that covariance between the instrument and the second stage error term approaches 0 by deﬁnition, removing the remaining source of inconsistency in standard IV estimates.

10

5 Results
5.1 First Stage
Turning ﬁrst to the coeﬃcients on import prices (Table 4): they are positive in all models, thus suggesting that when the costs of imports rise, governments seek to stimulate local production. Initially beneﬁcial to farmers, the intent is of course to restore prices by increasing food supplies. By contrast, the coeﬃcient on the export price index is negative in all models. It suggests that when global prices shift such that local farmers can gain higher prices in foreign markets than they can at home, governments prevent them from doing so. By banning exports or imposing export taxes, they seek to prevent local prices from rising to the level of prices abroad.
Note the coeﬃcient on political competition: it is positive and signiﬁcant in most models. When the value of the coeﬃcient is low, there is little or no electoral competition; political competition is suppressed or takes the form of rivalries between organized interests. As the value of the index rises, it points to the rising signiﬁcance of party competition. As the sample is drawn from low and middle income countries, the change in political institutions thus signals an increase in the power of the rural majority and therefore – as the coeﬃcients suggest – a shift toward pro-farmer policies.
Importantly for second stage estimation, model 4 attains an R2 of 0.76, implying that equation (1) explains nearly 80% of observed variation in government policy response. Our instruments for RRA are thus strong, and our estimates will therefore be consistent when we employ this equation in the second stage.
5.2 Second Stage
The ﬁrst four panels of Table 5 contain two equations. Both yield coeﬃcients measuring the relationship between the right hand side (RHS) variables – discussed above – and measures of political instability. The ﬁrst equation (OLS) in each panel resembles those reported elsewhere in the literature: it relates price shocks and the properties of institutions to the likelihood of civil wars and political unrest. The second equation (GMM) provides estimates of the second stage equation of the 2SLS model, which contains an additional variable: the expected policy response of the government, given the structural characteristic of the economy and the nature of political institutions.
Panels 1-4 in Table 5 address the relationship between price shocks and civil war (see the
11

deﬁnition of PRIO in Table 2); panel 4 addresses the relationship between price shocks and political unrest (see the deﬁnition of Banks in Table 2). We begin with panel 4.
5.2.1 Political Unrest
In models 7 and 8, as the prices of imports rise, unrest increases.14 While the magnitude of the estimates is the same, the loss of observations renders the coeﬃcient in model 8 statistically insigniﬁcant. In keeping with Deaton and Miller’s (1995) ﬁndings, rising prices for agricultural exports associate with lower rates of disorder – something they attribute to increased prosperity. Supportive of this interpretation are the coeﬃcients on energy use and OPEC membership: Closely related to levels of income, they too bear negative signs and are signiﬁcant in both models.
Note the coeﬃcient on government policy: it is large, negative, and signiﬁcant at the 1% level. Policy shifts in favor of the urban sector are associated with increases in urban unrest. Yet the direction of the causal relationship between the two variables remains unclear: Do policy shifts in favor of urban dwellers cause urban unrest, or does urban unrest cause governments to subsidize urban food costs? To address this question, we therefore conduct a Granger causality test (6), which establishes causality by asking whether one time series successfully forecasts another. As depicted in the right hand column, we ﬁnd strong evidence that urban unrest is a strong predictor of government policy.
On the basis of this ﬁnding, we respecify equations (1) and (2) so that food price shocks condition urban unrest, which in turn induces governments to lower food prices. The ﬁrst stage results appear in model 5 of Table 4; the second stage results in model 9 of Table 5. We ﬁnd that urban unrest exerts a large eﬀect on government policy: A standard deviation increase in ln Banks generates a standard deviation shift in relative prices, rendering them more favorable to urban consumers.
5.2.2 Civil Wars
Turning to models 1-6, focus ﬁrst on the OLS estimates, which, as noted above, resemble many of the equations previously reported in the literature. Negative and signiﬁcant, the coeﬃcients on the export price index suggest that when export prices fall, the likelihood of civil wars increases. A standard deviation reduction in the export Laspeyres index generates a 2% increase in the probability of civil war, or roughly two-thirds of a standard deviation.
14The decline in the number of observations results from the dropping of those in which the raw scores were negative, rendering the log undeﬁned.
12

Moreover, this eﬀect grows slightly larger – and is more precisely estimated – as the PRIO battle death threshold increases. When food prices rise, these results suggest, the civil conﬂicts that ensue are likely to be particularly violent.
But now turn to the even numbered (GMM) equations. The coeﬃcient on RRA suggests that governments that tend to respond to price shocks with measures that favor urban consumers are signiﬁcantly less likely to face civil wars. Strikingly, the coeﬃcients on export prices are now insigniﬁcant, suggesting that governments that adopt pro-consumer policies neutralize the destabilizing impact of price declines in export markets.
In interpreting these ﬁndings, it is useful to turn to the qualitative literature on “provisioning,” a term that refers to the feeding of public servants – including the armed forces – and urban centers – especially political capitals. The literature notes that the management of food supplies was critical to the keeping of peace in ancient Rome (Garnsey (1988)) and 18th Century Paris (Kaplan (1976)) and remains critical to the maintenance of order in contemporary Nairobi (Bates (1989)). According to these accounts, the delivery of grain at aﬀordable prices enabled emperors, kings, and presidents to campaign for political support from urban populations. By providing heads of state the opportunity to consolidate their support, provisioning transformed food price shocks into political opportunities. In the hands of wiley politicians, food price shocks can render civil wars less likely. While we ﬁnd this interpretation appealing, we have yet to explore it systematically.
5.3 Robustness Checks
Were political instability to occur in a major producing country, it might aﬀect global prices and introduce endogeneity bias: In response to this possibility, we reestimate equations (1) and (2), omitting countries that produce 20% or more of the world’s major staple crops: maize, wheat, and soybeans.15 The results for our core IV-GMM model appear in Table 7, and closely resemble those in Table 5. They suggest that when governments respond to urban unrest by reducing RRA by a standard deviation, the probability of civil war declines by between 3% and 5%, depending on the number of battle deaths. And when governments pursue policies that favor urban consumers, food prices once again have no meaningful eﬀect.
We next consider whether these relationships have evolved over time. Given the impact
15Arezki and Bruckner (2011) represents the only other attempt to remove this form of endogeneity bias. They remove several large producers from their sample: China, Guatemala, India, Indonesia, Pakistan, Thailand, Uganda, Ukraine, and Vietnam. Since food production is relatively variable, however, we prefer our time variant approach.
13

of the Cold War on the levels and incidence of civil unrest,16 we re-estimate the model before and after the end of the Cold War. To conserve degrees of freedom, we do so by interacting our export and import food price indices with a dichotomous variable that assumes value 1 from 1990 on. The results appear in Table 8, and again closely resemble our baseline results in Table 5. When governments adopt policies that shift RRA in favor of urban consumers by a standard deviation, these results suggest, the probability of civil war once again declines by between 3% and 5%, depending on the number of battle deaths.
Lastly, we test whether the dynamics we ﬁnd in Table 5 are unique to low income countries. The results in Table 8 suggest that while policy shifts in favor of the urban sector appear to slightly reduce the probability of civil war, the eﬀect is much smaller than in poor countries; it is also imprecisely estimated. Surprisingly, we ﬁnd some evidence that increases in food export costs increase the probability of civil war, though the magnitude of this eﬀect is quite small.
6 Conclusion
As noted in Table 1, the rapid rise in food prices between 2007 and 2008 triggered widespread political unrest. More recently, discontent over food prices exacerbated urban protests in North Africa and the Middle East and so contributed to the events now known as the “Arab Spring.” These events have rekindled interest in the relationship between subsistence crises and political disorder. In this paper, we have sought to contribute to the literature on this subject.
Those who study the politics of agriculture in development focus not only on the determinants of political order; they also study government policy. Most often, they ﬁnd that governments in the developing world adopt policies that favor the interests of urban consumers and discriminate against those of farmers. In doing so, they emphasize the costs of these policies, both to farmers and to the economy. A major lesson of this paper is that the policy choice and political impact should be studied together. Doing so reveals that the economic costs of government policies must be weighed against their political beneﬁts. It also reveals that by omitting the responses of governments, the speciﬁcations employed in previous studies of the impact of food price shocks have been misspeciﬁed. When governments mitigate the impact of price rises to urban consumers, contrary to previous ﬁndings, there is then no relationship between food price shocks and civil war.
16See, for instance,Goldstone and Gurr (2003).
14

Our use of a two stage model suggests a mechanism that might account for this ﬁnding: political expectations. In countries where the structure of the economy, political institutions and the past behavior of the government are such that the government can be expected to placate urban consumers in the event of food price rises (thus the result of the ﬁrst stage), then price shocks will not result in civil wars (thus those of the second). When the organized, powerful, and vulnerable urban sector can rationally expect a favorable response from its government, their protests do not prepare the ground for civil war.
We realize that we have yet to pin down the precise path running between price shocks and conﬂict; our analysis is thus incomplete. What we have established is that urban bias lowers the likelihood that food price shocks will trigger civil war. The policy choices of governments in the developing areas become more intelligible when viewed in the light of this ﬁnding.
15

References
Kym Anderson. Distortions to Agricultural Incentives: A Global Perspective, 1955-2007. The World Bank, Washington, DC, 2009.
Joshua D. Angrist and Anne Krueger. Rural windfall or a new resource curse? cocoa, income and civil conﬂict in colombia. Review of Economics and Statistics, 90(2):191–215, 2008.
Rabah Arezki and Markus Bruckner. Food prices and political instability. IMF Working Paper, WP-11-62, 2011.
Robert H. Bates. Markets and States in Tropical Africa: The Political Basis of Agricultural Policies. University of California, Berkeley, 1981.
Robert H. Bates. Beyond the Miracle of the Market: The Political Economy of Agarian Development in Kenya. Cambridge University, New York, 1989.
Robert H. Bates and Steven A. Block. The politics of agricultural protection in africa. World Bank Agricultural Distortion Working Paper Series, 87, 2009.
Christopher F. Baum, Mark E. Schaﬀer, and Steven Stillman. Instrumental variables and gmm: Estimation and testing. Stata Journal, 3:1–31, 2003.
Christopher F. Baum, Mark E. Schaﬀer, and Steven Stillman. Enhanced routines for instrumental variables/gmm estimation and testing. Boston College Economics Working Paper, 667, 2007.
Nathaniel Beck, David Epstein, Sharyn O’Halloran, and Simon Jackman. Alternative models of dynamics in binary time-series–cross-section models: The example of state failure. Paper prepared for delivery at the 2001 Annual Meeting of the Society for Political Methodolgy, Emory University, 2002.
Timothy J. Besley and Torsten Persson. The incidence of civil war: Theory and evidence. NBER, 2008.
Henry Bienen and Nicolas van de Walle. Of Time and Power: Leadership Duration in the Modern World. Stanford University, Stanford, 1991.
Angus Deaton and Ron Miller. International Commodity Prices, Macroeconomic Performance, and Politics in Sub-Saharan Africa. Princeton University, Princeton, 1995.
16

Angus Deaton and John Muellbauer. Economics and Consumer Behavior. Cambrdige University, New York, 1980.
Melissa Dell, Benjamin F. Jones, and Benjamin A. Olken. Climate change and economic growth: Evidence from the last half century. NBER Working Paper Series, 14312, 2008.
Oeindrila Dube and Juan F. Vargas. Are all resources cursed? coﬀee, oil, and armed conﬂict in colombia. Harvard University, 2006.
Peter Garnsey. Famine and Food Supply in in the Graeco-Roman World: Responses to Risk and Crisis. Cambridge University, New York, 1988.
Jack Goldstone and Ted Robert Gurr. State failure task force report, phase iii. 2003.
Lars Peter Hansen. Large sample properties of generalized method of moments estimators. Eonometrica, 50:1029–1054, 1982.
Kym Anderson Yujiro Hayami. The Political Economy of Agricultural Protection. Allen and Unwin, London, 1986.
Yujiro Hayami. Rice policy in japan’s economic development. American Journal of Agricultural Economics, 54(1):19–31, 1957.
Simon M. Hsiang, Kyle C. Meng, and Mark A. Cane. Civil conﬂicts are associated with the global climate. Nature, 476:438–441, 2011.
David Gale Johnson. World Agriculture in Disarray. St. Martin’s Press, New York, 1973.
Steven L. Kaplan. Bread, Politics, and Political Economy in the Reign of Louis XV. Martinus Nijhoﬀ, The Hague, 1976.
Michael Lipton. Why Poor People Stay Poor: Urban Bias and World Development. Urban Bias and World Development, Cambridge, 1977.
Monty G. Marshall and Keith Jaggers. Polity IV Project: Political Regime Characteristics and Regime Transitions, 1800-2004. George Mason University, Arlington, 2005.
Edward Miguel, Shanker Satyanath, and Ernest Sergenti. Economic shocks and civil conﬂict: An instrumental variables approach. Journal of Political Economy, 112:725–753, 2004.
17

Mancur Olson. Space, agriculture, and organization. Journal of Agricultural Economics, 67: 928–37, 1985.
Carmen M. Reinhart and Kenneth S. Rogoﬀ. This Time is Diﬀerent: Eight Centuries of Financial Folly. Princeton University, Princeton, 2009.
Theodore W. Schultz. Transforming Traditional Agriculture. Arno Press, New York, 1976. Amartya Sen. Poverty and Famines: An Essay on Entitlement and Deprivation. Oxford
University, New York, 1983. Charles Tilly. Coercion, Capital, and European States, AD 990-1992. Blackwell, New York,
1990. Louise Tilly. The food riot as a form of political conﬂict. Journal of Interdisciplinary History,
2:23–57, 1971. Ashutosh Varshney. Beyond Urban Bias. Frank Cass and Company, London, 1993. Jeﬀrey M. Wooldridge. Econometric Analysis of Cross Section and Panel Data. Mas-
sachusetts Institute of Technology, Cambridge, 2002.
18

Argentina

Austria

Bangladesh

Laspeyres Index 0.5 2.0

Laspeyres Index 1.0 2.0

Laspeyres Index 1.0 2.0

1960

1980 Year

2000

Congo

1960

1980 Year

2000

France

1960

1980 Year

2000

Indonesia

Laspeyres Index 1.0 2.5

Laspeyres Index 1.0 2.0

Laspeyres Index 0.8 1.6

1960

1980 Year

2000

Kuwait

1960

1980 Year

2000

Mongolia

1960

1980 Year

2000

Nicaragua

Laspeyres Index 1.0 2.0

Laspeyres Index 0.5 2.0

Laspeyres Index 1.0 2.0

1960

1980 Year

2000

Syria

1960

1980 Year

2000

Turkey

1960

1980 Year

2000

United Kingdom

Laspeyres Index 1.0 2.5

Laspeyres Index 1.0 2.0

Laspeyres Index 1.0 2.0

1960

1980 Year

2000

1960

1980 Year

2000

1960

1980 Year

2000

Figure 1: Laspeyres Time Series for a Random Country Sample

19

20

Variable RRA
PRIO 25 PRIO 100 PRIO 500 Banks

Table 2: Outcome Variable Deﬁnitions and Descriptive Statistics

Variable Type Mean (Min/Max) [−0.95, 1.30] -0.17
Dichotomous 0.14
Dichotomous 0.13
Dichotomous 0.11
[−0.62, 27.39] -0.06

Median -0.08
0.00 0.00 0.00 -0.62

Standard Deviation 0.26
0.35
0.34
0.31
1.80

Description/Source
A measure of the extent to which governments manipulate domestic prices in favor of the agricultural sector, developed by Kym Anderson and colleagues for the World Bank; see Anderson (2009). This variable assumes value 1 if a country experienced a civil war with greater than 25 battle deaths in a given year; see www.prio.no/CSCW. This variable assumes value 1 if a country experienced a civil war with greater than 100 battle deaths in a given year; see www.prio.no/CSCW. This variable assumes value 1 if a country experienced a civil war with greater than 500 battle deaths in a given year; see www.prio.no/CSCW. This index aggregates all instances of political violence within a given country-year, with more serious episodes – revolutions, riots, purges – weighted more heavily than strikes and assassinations. It is taken from Arthur Banks’ Cross-National Time Series database; see www.databanksinternational.com.

Table 3: Explanatory Variable Deﬁnitions and Descriptive Statistics

21

Variable
Export Laspeyres Index

Variable Type (Min/Max)
[0.00, 18.65]

Mean 1.56

Import Laspeyres In- [0.01, 116.30] 1.55 dex

Political Competition [1, 10]

3.51

Energy Consumption [0, 1312845] 19430

OPEC Membership Banking Crisis Low Income Status

Dichotomous 0.04 Dichotomous 0.03 Dichotomous

Post-Cold War

Dichotomous 0.33

Median 1.32 1.36 2.00
1710 0.00 0.00
0.00

Standard Deviation 1.13
0.79
2.91
93150
0.20 0.16
0.47

Description/Source
A country-year measure of food prices, benchmarked at the representative consumers’ 1980 consumption basket. We draw global commodity prices from the World Bank and food consumption data from the FAO. A country-year measure of food prices, benchmarked at the representative consumers’ 1980 consumption basket. We draw global commodity prices from the World Bank and food consumption data from the FAO. This variable, drawn from the Polity IV dataset, measures “qualities of democratic and autocratic authority in governing institutions,” with −10 representing a hereditary monarchy and 10 a consolidated democracy; see Marshall and Jaggers (2005). A measure of primary energy consumption, given in thousands of metric coal-ton equivalents. The variable is taken from the Correlates of War 2 Project; see www.correlatesofwar.org. A time variant OPEC membership roster, as commissioned by the CIA’s Political Instability Task Force. Records episodes of banking crises, as presented in Reinhart and Rogoﬀ (2009). Following Arezki and Bruckner (2011), we restrict attention to low and low-middle income countries, as deﬁned by the World Development Indicators. This variable assumes value 1 from 1990 on.

Table 4: First Stage Results (Standard Errors in Parentheses)

Export Laspeyres
Import Laspeyres
Political Competition
Export Laspeyres × Political Competition
Import Laspeyres × Political Competition
Banking Crisis
RRA t − 1
Fixed Eﬀects Robust Clustered Standard Errors R2 N Signiﬁcance levels: †† : 15% † : 10%

RRA Model 1 -.050∗∗ (.016) .057∗∗ (.015) .035∗∗ (.007)
-.008† (.005) -.008 (.003) -.004 (.043)
No No 0.0703 1182 ∗ : 5%

RRA Model 2
-0.050 (0.037) 0.057 (0.037) 0.035∗ (0.013) -0.008 (0.009) -0.008 (0.008) -0.004 (0.035)
No Yes 0.0703 1182 ∗∗ : 1%

RRA Model 3
-0.014 (0.027) 0.010 (0.029) 0.035∗ (0.016) -0.010 (0.006) -0.005 (0.006) -0.016 (0.037)
Yes Yes 0.5703 1182

RRA Model 4
-0.020 (0.014) 0.007 (0.013) 0.010 (0.008) -0.002 (0.004) -0.002 (0.003) -0.015 (0.042) 0.657∗∗ (0.075)
Yes Yes 0.756 1182

ln Banks Model 5 -0.225†† (0.150)
0.312 (0.263) -0.119 (0.082) 0.103 (0.050) -0.008 (0.035)∗ -0.353 (0.329)
Yes Yes 0.3675 508

22

Table 5: Second Stage Results (Standard Errors in Parentheses)

23

PRIO 25

PRIO 100

OLS (1) GMM (2) OLS (3)

RRA

0.261∗ (0.156)

ln Banks

Export Laspeyres

-0.019††

(0.156) 0.011

-0.020††

(0.013)

(0.027)

(0.012)

Import Laspeyres

-0.005

-0.015

-0.011

(0.015)

(0.016)

(0.017)

Political Competition -0.012†

-0.026∗

-0.008

(0.006)

(0.011)

(0.006)

Energy Consumption -2.41e-07∗∗ -2.59e-07∗ -1.69e-07††

(8.21e-08) (1.17e-07) (1.11e-07)

OPEC Membership

-0.303 (0.211)

-0.337 (0.191)

-0.263†† (0.174)††

Fixed Eﬀects

Yes Yes

Yes

Robust Clustered SEs

Yes

Yes

Yes

N

2427

1061

2427

Signiﬁcance levels: †† : 15% † : 10% ∗ : 5% ∗∗ : 1%

GMM (4)
0.199†† (0.141)
(0.141) -0.004 (0.024) -0.037† (0.144) -0.024∗ (0.010) -1.72e-07 (1.22e-07) -0.217 (0.163)
Yes Yes 1061

PRIO 500 OLS (5)
-0.022∗ (0.009) -0.018 (0.017) -0.013 (0.006) -1.23e-07 (1.13e-07) -0.271 (0.167)
Yes Yes 2427

GMM (6)
0.142 (0.136)
(0.136) 0.009 (0.020) -0.035† (0.021) -0.012 (0.009) -1.31e-09 (1.16e-07) -0.198†† (0.137) Yes Yes 1061

Log Banks OLS (7)
-0.009 (0.129) 0.266† (0.134) -0.021 (0.035) -3.80e-06∗∗ (8.06e-07) -0.304†† (0.197)
Yes Yes 522

GMM (8)
-1.854∗∗ (0.549)
(0.549) -0.344∗∗ (0.119)
0.210 (0.160) -0.034 (0.044) -3.73e-06∗∗ (4.29e-07) -0.395† (0.235)
Yes Yes 289

RRA GMM (9)
-0.134∗∗ (0.049) -0.047†† (0.033) -0.018 (0.035) 0.006 (0.010) -5.43e-07∗∗ (1.68e-07) -0.088 (0.093)
Yes Yes 289

Table 6: Direct Granger Causality Test Results (p Values in Parentheses)

F Statistic

H0: No Granger causality from RRA to ln (Banks)
1.3467 (0.2526)

H0: No Granger causality from ln (Banks) to RRA
2.2985† (0.0775)

24

Table 7: Second Stage Results, with Price Setters Dropped (Standard Errors in Parentheses)

25

PRIO 25

PRIO 100

OLS (1) GMM (2) OLS (3)

RRA

0.214∗

Export Laspeyres

-0.022†

(0.108) -0.004

-0.020††

(0.012)

(0.025)

(0.012)

Import Laspeyres

-0.008

-0.021

-0.011

Political Competition

(0.015) -0.010††

(0.016) -0.021∗

(0.017) -0.008

Energy Consumption

(0.006)

(0.010)

(0.006)

-2.42e-07∗∗ -2.55e-07∗ -1.70e-07††

OPEC Membership

(8.23e-08) -0.299 (0.214)

(1.18e-07) -0.308††
(0.193)

(1.11e-07) -0.263†† (0.174)††

Fixed Eﬀects

Yes Yes Yes

Robust Clustered Standard Errors

Yes

Yes

Yes

N

2403

1037

2427

Signiﬁcance levels: †† : 15% † : 10% ∗ : 5% ∗∗ : 1%

GMM (4)
0.211†† (0.145) -0.003 (0.025) -0.036† (0.020) -0.024∗ (0.010) -1.78e-07 (1.23e-07) -0.223 (0.164)
Yes Yes 1037

PRIO 500 OLS (5)
-0.022∗ (0.009) -0.018 (0.017) -0.013∗ (0.006) -1.24e-07 (1.13e-07) -0.271 (0.168)
Yes Yes 2403

GMM (6)
0.157 (0.140) 0.011 (0.021) -0.034† (0.020) -0.013 (0.010) -7.95e-09 (1.16e-07) -0.202†† (0.137)
Yes Yes 1037

Log Banks OLS (7)
0.022 (0.135) 0.266∗ (0.134) -0.026 (0.035) -3.79e-06∗∗ (7.81e-07) -0.300†† (0.202)
Yes Yes 509

GMM (8)
-1.808∗∗ (0.558) -0.364∗ (0.145) 0.201 (0.157) -0.047 (0.047) -3.77e-06∗∗ (4.52e-07) -0.358†† (0.222)
Yes Yes 276

Table 8: Second Stage Results, Post-Cold War (Standard Errors in Parentheses)

26

PRIO 25

PRIO 100

OLS (1) GMM (2) OLS (3)

RRA

0.261∗

Export Laspeyres

-0.018

(0.118) 0.008

-0.019††

(0.013)

(0.027)

(0.012)

Import Laspeyres

-0.003

-0.016

-0.008

(0.016)

(0.017)

(0.018)

Export Laspeyres

-0.022

-0.043

-0.014

× Post-Cold War

(0.048)

(0.085)

(0.045)

Import Laspeyres

-0.018

0.009

-0.025

× Post-Cold War Political Competition

(0.045) -0.011†

(0.082) -0.026∗

(0.044) -0.007

Energy Consumption

(0.006)

(0.011)

(0.006)

-2.36e-07∗∗ -2.58e-07∗ -1.63e-07††

OPEC Membership

(8.11e-08) -0.304 (0.211)

(1.18e-07) -0.347∗
(0.190)

(1.12e-07) -0.265†† (0.173)††

Fixed Eﬀects

Yes Yes Yes

Robust Clustered Standard Errors

Yes

Yes

Yes

N

2427

1061

2427

Signiﬁcance levels: †† : 15% † : 10% ∗ : 5% ∗∗ : 1%

GMM (4)
0.193 (0.137) -0.008 (0.025) -0.037† (0.021) 0.053 (0.077) -0.001 (0.080) -0.023∗ (0.010) -1.67e-07 (1.23e-07) -0.227 (0.164)
Yes Yes 1061

PRIO 500 OLS (5)
-0.021∗ (0.009) -0.014 (0.018) -0.028 (0.037) -0.025 (0.039) -0.012∗ (0.006) -1.16e-07 (1.13e-07) -0.273 (0.167)
Yes Yes 2427

GMM (6)
0.135 (0.135) 0.008 (0.022) -0.039† (0.021) -0.023 (0.067) 0.059 (0.050) -0.012 (0.009) 1.41e-09 (1.13e-07) -0.183 (0.139)
Yes Yes 1061

Log Banks OLS (7)
-0.005 (0.140) 0.276∗ (0.132) -0.014 (0.376) -0.422 (0.480) -0.022 (0.036) -3.71e-06∗∗ (6.83e-07) -0.318†† (0.200)
Yes Yes 522

GMM (8)
-1.930∗∗ (0.537) -0.398∗∗ (0.120)
0.204 (0.159) 0.648† (0.335) -0.640† (0.381)
-0.035 (0.042) -3.81e-06∗∗ (3.66e-07) -0.348††
(0.241) Yes Yes 289

Table 9: Second Stage Results, High Income Countries (Standard Errors in Parentheses)

27

PRIO 25

PRIO 100

OLS (1) GMM (2) OLS (3)

RRA

0.115

Export Laspeyres

0.035†

(0.110) 0.092∗∗

0.033††

(0.021) (0.032) (0.020)

Import Laspeyres

-0.010

0.014

-0.004

(0.018) (0.016) (0.015)

Political Competition

0.003

-0.000

-0.001

(0.005) (0.005) (0.005)

Energy Consumption

-1.55e-07 -1.17e-07 -1.72e-07

OPEC Membership

(2.12e-08) (1.98e-07) (2.17e-07)

-0.040

-0.037†

(0.021)

(0.021)

Fixed Eﬀects

Yes Yes Yes

Robust Clustered Standard Errors Yes Yes Yes

N

2222

1205

2222

Signiﬁcance levels: †† : 15% † : 10% ∗ : 5% ∗∗ : 1%

GMM (4) 0.049 (0.088) 0.060† (0.035) -0.005 (0.021) -0.002 (0.005)
1.05e-09 (1.81e-07)
Yes Yes 1061

PRIO 500 OLS (5)
0.022 (0.020) -0.008 (0.014) -0.005 (0.006) -4.15e-07 (2.57e-07) -0.023 (0.020)
Yes Yes 2222

GMM (6) 0.014 (0.073) 0.006 (0.035) -0.020 (0.025) -0.006 (0.005)
2.68e-07†† (1.82e-07)
Yes Yes 1061

Log Banks OLS (7)
0.242† (0.125) 0.084 (0.077) -0.024 (0.036) -4.21e-06∗ (1.81e-06) -3.737† (0.000)
Yes Yes 690

GMM (8) -0.557 (0.475) 0.112 (0.100) -0.109 (0.208) 0.028 (0.034)
-4.36e-06 (1.48e-06)
Yes Yes 289