University of Birmingham Civil war and U.S. foreign influence

Westudyhowforeigninterventionsaffectcivilwararoundtheworld.Inanin ﬁ nitelyrepeatedgamewecombine agamblingforresurrectionmechanismforthein ﬂ uencingcountrywiththecanonicalbargainingmodelofwarin the in ﬂ uenced country to micro-found sudden shifts in power among the domestic bargaining partners, which are known to lead to war due to commitment problems. We test two of our model predictions that allow us to identifythein ﬂ uenceofforeigninterventiononcivilwarincidence:(i)civilwarsaroundtheworldaremorelike-ly under Republican governments and (ii) the probability of civil wars decreases with the U.S. presidential ap- proval rates. These results withstand several robustness checks and, overall, suggest that foreign in ﬂ uence is a sizable driver of domestic con ﬂ ict. © 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/3.0/). war in the rest of the world. According to our theoretical model, government's ideology and presidential approval determine the willingness to intervene abroad of a potential intervening country. As these motives for intervention are mainly domestic, they constitute an exogenous source of variation in the foreign in ﬂ uence received by a country in (potential or ongoing) con ﬂ ict. We ﬁ nd that the incidence and onset of civil war increase under Republican administrations and decreases in the level of presidential approval. These results show how the risk of civil war is affected by the political situation in the U.S. and suggest that the international dimension of domestic con ﬂ icts is very relevant to understand civil wars.


Introduction
There is a large and growing recent economic literature on the motives and consequences of civil wars, focusing almost exclusively on domestic determinants. While the theory is devoted to understanding why costly conflicts between two domestic parties are not deterred, the empirical research emphasizes diverse country-specific factors that affect the risk of civil conflict. Examples include slow income growth, proportion of natural resources, secondary school attainment (Collier and Hoeffler, 2004;Collier et al., 2005), income inequality (Sambanis, 2005), poverty (Djankov and Reynal-Querol, 2008), ethnic polarization (Montalvo and Reynal-Querol, 2005) or even the effect of diseases (Cervellati et al., 2010). This paper complements the theoretical and the empirical literature by developing a systematic investigation of the role of U.S. foreign influence as a determinant of civil (domestic) conflicts in other countries. 1 While there are many examples of civil wars characterized by the involvement of foreign governments supporting one of the sides in conflict, even before and after the end of the Cold War, 2 identifying the effect of foreign influence on civil conflicts is a challenging task. Interventions in foreign conflicts are often secretive and indirect and therefore unlikely to be fully reflected in available data. As an additional difficulty, many are the ways for foreign governments to intervene in domestic civil wars. They can provide covert encouragement, allow for (and promote) arms transactions, supply war intelligence and resources, and give sanctuary to rebels or support a third state that is also involved in the civil war. But even if the "right" measure Journal of Development Economics 110 (2014) [64][65][66][67][68][69][70][71][72][73][74][75][76][77][78] ☆ Previously circulated as "Civil war and foreign influence". We are thankful to the editor, Christopher Udry, and two anonymous referees who provided excellent comments and suggestions. We also thank Sami Berlinski, Francesco Caselli, Benjamin J. Cohen, Joan María Esteban, Hannes Mueller, Gerard Padró i Miquel, James Reade, Sevi Rodríguez Mora, Martín Rossi and Jaideep Roy for valuable comments and discussions. We are grateful to Tim Besley and Torsten Persson for sharing their data on political conflict, and Daniel Berger, William Easterly, Nathan Nunn and Shanker Satyanath for allowing us to use their data on CIA operations. Hauk thanks the LSE for its hospitality and acknowledges financial support from the Programa de Movilidad de Profesorado project PR2009-0375, from CICYT project number ECO2009-12695 and from the Severo Ochoa Programme for Centres of Excellence in R&D (SEV-2011-0075) 1 The possibility of foreign influence has typically been overlooked in economic studies.
of foreign influence was available, it would be difficult to identify causality.
We propose a theoretical model that shows how a (foreign) third party can trigger and prolong costly conflicts between two domestic sides. This model extends the (infinitely) repeated canonical bargaining model of war where two domestic parties bargain how to split the country's spoils and wara costly lotteryis the outside option of the game. In our model, a foreign government decides in each period whether to intervene by offering foreign support to one of the domestic players before the domestic bargaining takes place. Whether support is offered depends on the time varying internal political situation of the potentially intervening country. This makes the nature of intervention fleeting and always leaves one of the domestic parties in a transient advantageous situation. This party might prefer war to peace to lock in its momentary advantageous situation. The existence of a foreign potential intervening country induces a commitment problem among the domestic players by leading to a sudden shift in power between the incumbent government and the opposition which as Powell (2006) has shown can cause civil war. Our model provides a plausible example and a microfoundation of this sudden power shifts by explicitly modeling the costs and benefits of the potentially intervening country.
We show that a foreign intervention is more attractive the lower the personal costs of the foreign politician of going to war and the lower his popularity. A foreign intervention can serve as a gamble for resurrection for unpopular politicians. If the probability of reelections is associated with campaign contributions, a government will increase its re-election possibilities by relying more on the support of corporations which will make the government more likely to intervene abroad to enhance business opportunities for these cooperations. Moreover, a successful campaign can always be communicated by the intervening politician and is likely to boost approval. Since political support is bounded, this strategy is more attractive for unpopular politicians: the potential gain in approval rates due to a successful intervention is higher and the potential loss in approval rates is lower should a failed intervention be discovered than for more popular politicians.
Our empirical strategy consists of identifying and quantifying systematic domestic U.S. political factors as determinants of the incidence of civil war in the rest of the world. According to our theoretical model, government's ideology and presidential approval determine the willingness to intervene abroad of a potential intervening country. As these motives for intervention are mainly domestic, they constitute an exogenous source of variation in the foreign influence received by a country in (potential or ongoing) conflict. We find that the incidence and onset of civil war increase under Republican administrations and decreases in the level of presidential approval. These results show how the risk of civil war is affected by the political situation in the U.S. and suggest that the international dimension of domestic conflicts is very relevant to understand civil wars.
Why the U.S.? As we concentrate on civil wars that occurred from 1937 to 1997, the U.S. is the natural candidate for a potential intervening country. First, its superpower status and the size of its economy provide it with sufficient resources to intervene simultaneously in many countries during the period. Second, the data on observed foreign interventions tells us that the U.S. has extensively intervened in civil wars. 3 Third, the U.S. is characterized by a twoparty system and, importantly, the two parties, Republican and Democratic, have different views on the role of the U.S. in the international arenas. These differences are epitomized by diverse Republican approaches to foreign policy like the Roosevelt corollary of the Monroe's doctrine, and principles present in the Eisenhower or Bush doctrines. 4 This framework for foreign policy is rooted in the Republican ideology which differs from the general approach of the Democratic Party. As a consequence, the two parties systematically differ in their propensities to intervene in foreign affairs. Fourth, there is accurate data on presidential approval for the case of the U.S. Last but not least, given the secretive nature of interventions in civil wars abroad and the salience of domestic issues during election campaigns, the U.S. citizens are unlikely to decide their vote based on domestic conflicts in other (often distant and barely known) countries. 5 While there is evidence that, consistent with the spirit of our model, presidential approval is linked to foreign issues (e.g. Aldrich et al., 2006;Hurwitz and Peffley, 1987)), foreign policy issues only influence votes in so far as the public has coherent attitudes about foreign policy and the political parties uphold distinct foreign policy platforms and the foreign policy issue is made salient e.g. by the media (Aldrich et al., 2006). There is no indication that a civil war in another country becomes such a salient issue to affect the election of presidential candidates in the US.
Following recent empirical studies, we exploit panel data to identify a causal link between politics in the U.S. and the incidence of civil war relying on within-country variation. We adopt the empirical strategy developed in Besley and Persson (2011) and estimate the effect of a Republican government in office and the level of presidential approval. The results are striking and support our predictions. The incidence and onset of civil war increase under Republican governments and decrease with the U.S. presidential approval. Their impact is quantitatively important: the estimates imply that the incidence of civil war increases by 60% under Republican administrations. Also, a fall of 1 percentage point in the presidential approval rating rises the incidence of civil war by 2%. These results hold when we concentrate on the onset of civil war. However, these findings cannot be taken as conclusive evidence that U.S. interventions caused or prolonged civil wars. There may be omitted variables potentially correlated with both the U.S. political situation and the incidence of civil war. To mitigate this concern, we control for aggregated demand shocks (proxied by the growth of gross world product) and shocks in oil prices. We also address potential reverse causality problems. For example, it may be possible that the number of civil wars around the world can cause political changes in the U.S. by inducing citizens to vote for the Republican Party. Or that the Republican effect is driven by individual characteristics of the presidents, not their ideology. To understand the logic behind the relationship between civil wars and U.S. politics better we control for these potential sources of concern with no effect on our results.
Of course, the U.S. is not the only country intervening abroad. The Soviet Union during the Cold War and France influencing its former colonies are other important examples. We control for the Cold War, the 3 We mentioned examples in footnote 2. 4 These doctrines basically justify interventions abroad by emphasizing the defense of American values and the moral mandate of preserving (and installing) freedom around the world. The doctrine elaborated by Monroe, and amended under Roosevelt's presidency, was more oriented to preserve American interests in the western hemisphere (Sexton, 2011); While both the democrat Truman and the republican Eisenhower justified the right to intervene abroad as a measure to halt communism, Eisenhower was more precise on the goals of U.S. foreign policy. In Truman's words "..it must be the policy of the United States to support free peoples who are resisting attempted subjugation by armed minorities or outside pressures. " Truman (1947). In contrast, Eisenhower said that the United States would give economic and military aid to Middle Eastern Nation as it was essential to preserve this region from communism. As he put it U.S. intervention would "include the employment of the armed forces of the United States to protect and secure the territorial integrity and political independence of such nations requesting such aid, against overt armed aggression from any nation controlled by International Communism." Eisenhower (1957). 5 The voting behavior of Americans has been intensively studied (see Bartels (2010) for an overview). While early studies claimed that votes were only determined by domestic issuesstressing the importance of economic factors (see e.g. Lewis-Beck et al. (2008) and references therein)a more recent literature also emphasizes the importance of international issues (see Aldrich et al. (1989)). Kessel (2004) analyzes the presidential elections from 1952 to 2000 using the American National Election Studies whose open-ended questions provide a measure of valence toward candidate, party and issue objects in the elections. He shows that in all 13 elections economic and general issues were extremely important, but international issues also mattered in 11 of these 13 elections. effect of KGB interventions and former French Colonies in our empirical analysis. We show that even under the Cold War, the U.S. political situation has a significant effect on civil war in the rest of the world, even after controlling for the influence of the USSR though observed KGB operations. We also show that this is not the case when we restrict our attention to only former French colonies which provides additional support to our identification strategy. We then explore a precise channel of U.S. direct interventions in conflicts abroad. Given the secrecy associated with U.S. interventions it is difficult to claim they actually took place. However, the recent declassification of successful CIA operations before 1990 allows us to test whether these direct interventions operated by the CIA were an important channel inducing domestic conflict around the world. We show that (successful) CIA operations are more frequent under Republican administrations and when presidential approval is low.
Overall, our results suggest that U.S. foreign influence is a sizable driver of conflict around the world. Similar results are obtained in a time-series analysis where we estimate the number of ongoing and new civil wars per-year.
The remainder of the paper is organized as follows. In Section 2, we discuss the related literature. The theoretical model is developed in Section 3. Section 4 describes the data and reports the empirical exercises conducted to test the predictions of the model and discusses the role of the CIA as a channel of influence. Section 5 concludes.

Related literature
Unlike the economic literature, the political science literature on foreign interventions and transnational aspects of civil wars has been growing considerably in recent years. The earlier literature used the term foreign interventions mainly as referring to peace interventions in ongoing wars (Gartzke and Gleditsch, 2006;Regan, 2000;Walter, 1997). This clearly is complementary to our approach where the foreign interventions trigger or prolong an already existing war. 6 This possibility was already mentioned by Gleditsch (2007), who argues that motives for interventions in ongoing wars should be related to interventions causing war onset. He provides empirical evidence of the importance of ethnic, political and economic transnational linkages among neighboring countries: the probability of conflict in a given state is increasing in transnational ethnic links with the neighboring states, decreasing in the democratic degree of political institutions of neighboring countries and decreasing in trade integration with surrounding states. Gleditsch (2007), hypothesizes that the link is via external support of insurgencies whereas our model also works if support is given to the incumbent government. Moreover, we move away from neighboring countries in the strict spatial sense and consider the possibility of politically/economically motivated foreign interventions in general both theoretically and empirically. This is complementary to empirical studies on the spread of civil war which point to conflict in neighboring states, (Hegre and Sambanis, 2006) and the presence of refugees (Salehyan and Gleditsch, 2006) as a potential cause for civil war.
Foreign interventions in civil wars somehow blur the boundary between civil and inter-state wars. The question when a state prefers to support insurgencies instead of going to war and which type of rebel organizations receive and accept foreign support has been analyzed by Salehyan (2010) and by Salehyan et al. (2010). While this literature analyzes the trade-off foreign intervention versus direct war, it fails to explain why the foreign state is interested in either type of aggression. Our paper derives conditions for the endogenous occurrence of foreign interventions.
In order to do so, we explicitly take the motives of politicians into account. We do not only look at purely economical motives but also at political and personal costs and benefits. One of the personal motives we put forward is related to the "diversionary theory of war" literature. A "diversionary war" is a war instigated by a country's leader in order to distract its population from their own domestic strife. This option is especially attractive to leaders facing a near inevitable removal from office since exercising the war option might enable them to signal a high military or foreign policy ability. 7 This incentive to gamble for resurrection is also present in our model, however, the risk of the gamble is considerably reduced due to the secretive nature of a foreign intervention. Since the public is unlikely to observe a failed foreign intervention but can be made aware of (or perceive the effects of) successful ones, one might expect that domestic problems have a stronger effect on interventions in civil wars than on open aggressions toward other countries. Indeed, we provide very robust empirical evidence of a positive link between low presidential approval rates in the U.S. and incidences of civil wars around the world while the enormous body of empirical studies on the diversionary theory of war provides rather mixed evidence. 8 Another personal motive we put forward is the personal cost of going to war which we identify with being Republican or Democrat when taking the model to the data. Our paper thereby adds to the open controversy on whether the U.S. foreign policy is based on a bipartisan foreign policy consensus or is partisan (that is, conditional on whether the government is Republican or Democrat) 9 by providing support for the latter.
We heavily draw on the existing literature of the canonical bargaining model of war (as e.g. in Dal Bo and Powell (2009)) to explain why a foreign intervention can trigger or prolong an already existing civil war into which we introduce a third party. Our model relies on a new commitment problem (Fearon, 1995;Powell, 2004Powell, , 2006. The foreign induced commitment problem we identify is another version of Powell's argument that rapid shifts in the distribution of power lie at the heart of war. Salehyan (2007) provides an additional argument: external sanctuaries in neighboring countries can complicate the underlying bargain between states and rebels.
Our paper is also related to the recent literature on foreign influence on domestic policy choices (Aidt and Hwang, 2008;Antràs and Padró i Miquel, 2011) and the influence of foreign countries on the dynamics of domestic political institutions. Aidt and Albornoz (2011) argue that foreign countries may have an economic interest in sponsoring coups, stabilizing dictatorships and facilitating constrained democratization abroad in order to protect their foreign direct investment. Easterly et al. (2008) estimate that (declassified) US and Soviet interventions abroad have caused a decline in democracy across the world of about 33%. In Bonfatti (2010) a key trading partner may be interested to keep an incumbent in power because the incumbent can be controlled more easily from the exterior than the challenger using the threat of trade sanction. Aidt et al. (2010) show the influence of IMF and World Bank programs on political regime transitions.
As explained by Blattman and Miguel (2009), most of the empirical civil war literature uses cross-sectional data and fails to exploit withincountry variation in panel data which leads to biased estimates by replacing time-varying explanatory variables by their cross-sectional mean. Consequently, cross-country variation in these explanatory observable variables is confounded with cross-country averages in unobserved parameters. To avoid this problem, our empirical strategy only exploits within-country variations. This way, we follow a new series of papers using panel data, mainly concerned by the effect of different 6 A foreign country could do both, an open peace intervention and providing covert support to one of the sides in conflict. Our empirical results suggest that war interventions dominate the peace interventions. 7 For theoretical models on the diversionary theory of war see e.g. Hess and Orphanides (1995); Smith (1996); Tarar (2006). 8 For example, Ostrom and Job (1986); Morgan and Bickers (1992); Hess and Orphanides (1995); Miller (1995Miller ( , 1999 find evidence for the diversionary theory while Meernik and Waterman (1996); Gowa (1998) (2007); and Gowa (1998). economic shocks on civil conflicts. This literature proposes different proxies to capture income growth or wage shocks in order to address potential endogeneity problems. Miguel et al. (2004) use rainfall variation to show a negative relationship between income and civil war in Africa. 10 Brückner and Ciccone (2010) and Dube and Vargas (2013) study the effect of changes in commodity prices in Sub-Saharan countries and Colombia, respectively. Besley and Persson (2011) use both proxies in a more general study on the determinants of political violence, which includes civil war and state repression. They also show how the effect of income shocks depends on political institutions. Our paper builds on Besley and Persson (2011). We focus on civil war only and include the novel dimension of foreign intervention.

Theoretical background
In this section, we provide a theoretical model based on the canonical bargaining model of war to illustrate how foreign influence could lengthen a civil war or generate its outbreak. The model predicts which political situations in a potentially intervening country are more likely to cause an intervention and can therefore serve to identify U.S. influence on civil war incidence around the world.

The model ingredients
We study an infinitely repeated game with two long-run players and one short-run player. The long-run players consist of two domestic groups, the incumbent government I and the opposition O. The shortrun player is a foreign government with economic interests in the domestic country. 11 There are two types of foreign governments with a known distribution, which differ in the political costs associated with intervening abroad. At the beginning of every period of the infinitely repeated game the type of the foreign government is drawn by nature. Using a cost benefit analysis for the current period the foreign government then decides whether or not to strike a deal with the opposition. 12 This is observed by all players. The domestic parties then decide whether to peacefully split the country's spoils today or fight taking the entire future into account. If peace prevails, the game is repeated. A war ends the bargaining game forever.
In the absence of a war the domestic country's per period spoils are given by Π. The shares of the spoils going to the incumbent and the opposition result from bargaining. Bargaining failure leads to a destructive war reducing the per period spoils permanently to σΠ. These remaining spoils are fully kept by the winner of the war.
The probability to win the war depends on whether or not the foreign government has struck a deal with the opposition. Each domestic group has fixed resources for fighting given by r i with i = I,O. In the absence of a deal the opposition's probability to win a war is and increases to in case of a foreign intervention where r F are the resources contributed by the foreign government.
A foreign government will only contribute resources to the war if the total benefit of an intervention outweighs its costs. We denote the economic benefits of a successful foreign intervention by Π F . The economic costs are identical to the resources r F that the foreign government provides for fighting the war. Both types of foreign government face the same economic costs and benefits from intervention but differ in their political costs and benefits resulting in the low intervention cost type L drawn from a known distribution with probability λ and the high intervention cost type H, which occurs with probability 1 − λ. 13 The political costs and benefits are captured in two ways. On the one hand, there is an ideological component capturing the strictly personal costs c i with i = L,H of intervening in a domestic conflict abroad. 14 On the other hand, politicians enjoy different levels of popularity/approval u i with i = L,H which can be affected by a foreign intervention. 15 A secretive foreign intervention affects approval rates through two possible mechanisms. First, if the probability of re-election is associated with campaign contributions, then a government with low approval will increase its re-election probabilities by relying more on the support from corporations. This in turn makes the government more likely to intervene abroad to enhance business opportunities around the world. 16 Second, the secretive nature of foreign interventions makes them a safe bet. An unsuccessful involvement in a civil war is likely to go unnoticed by the public, while the head of government always has ways and means to get credit for new economic opportunities after a successful intervention even if the public does not know whether or not their country was involved. A successful ending of the civil war may spur government's popularity because of the possibility of signaling (e.g. by a state visit) global leadership and the new economic benefits associated with friendlier governments around the world. Since approval rates are bounded from above, the marginal gain from a successful intervention is higher for a head of state with lower initial approval.
Also, the potential downside of the intervention is low risk, because it only occurs if the intervention is unsuccessful and discovered by the public and is smaller for governments with low approval than for popular governments since approval rates are bounded from below. The higher upside potential and the lower downside risk make foreign intervention an attractive gamble for unpopular governments.
Our model captures this gambling for resurrection mechanism in a very stylized way. We assume that approval or reelection chances jump up to uNu i after a successful civil war abroad. 17 An unsuccessful foreign intervention will only affect the head of government's approval if discovered by the public which happens with probability β resulting in a drop in approval to a minimum level u b u i . 18 rainfall increases the incidence and onset of civil war. 11 These economic interests can take many different forms, such as foreign direct investment, trading opportunities, interest in natural resources, or interests grounded in geopolitical motives. 12 This assumption is made for the ease of exposition only. We would get qualitatively similar results if the deal was offered to the incumbent government. 13 Later on we will restrict parameters (Assumption 1) in such a way that in equilibrium the low cost type L always finds it optimal to intervene, while the high cost type H never does. 14 This personal component captures party ideology toward intervention, the party's sensitivity toward lobbying and concern about corporation business opportunities. A more pro-corporation party should be associated with a lower (or even a negative) c i . Indeed, there is evidence that this is the case for the U.S. where the Republican Party seems to be more influenceable by lobbies than the Democratic Party (see, for example, Jayachandran (2006)). 15 Approval matters because politicians derive ego-rents from being popular and because it determines future rents by affecting the possibility of re-election. 16 For example, Dube, Kaplan, and Naidu (2011) show that CIA operations to depose leaders abroad increase stock market values of corporations benefiting from the perspective of a new friendlier government in the foreign country. 17 Assuming that the marginal increase in approval is bigger the further the distance of initial approval with maximal approval would lead to qualitatively identical results. 18 Again assuming that the marginal drop in approval is bigger the further the distance of initial approval with minimal approval would lead to qualitatively identical results.

Gambling for resurrection
The foreign government is willing to intervene in a civil war abroad if and only if the benefits outweigh the costs. Formally, the foreign government intervenes if and only if where resources r F are chosen to optimize Eq. (3). Any interior r F has to satisfy the following first order condition: Substituting the expressions for r F * and p f * into Eq.
(3) and simplifying yield that the foreign government will intervene if and only if After inspection of Ψ i , we obtain the following result: Proposition 1. The foreign politician's willingness to sponsor a civil war abroad is increasing in Π F and r O and decreasing in β, r I , c i and u.
Proof. The comparative static results for Π F , r O , r I and c i are immediate from condition Eq. (6). Simple calculations show that the left hand side of Eq. (6) decreases in β . The change with respect to u is given as ■ To keep the model tractable, we impose the following assumption Assumption 1. For the low cost type L with (c L ,u L ) condition (6) is satisfied. For the high cost type H with (c H ,u H ) condition (6) is violated.
Assumption 1 ensures that the two types of foreign government behave differently in equilibrium. Only a low intervention cost type will intervene if the intervention is accepted. We therefore need to show when bargaining between the domestic players does indeed break down due to the existence of a potential intervening country.
Importantly, the gambling for resurrection stage micro-founds the following reduced form of our infinitely repeated bargaining game: in each period, the incumbent and opposition bargain how to split the country's spoils where war is the failure of the bargaining process and the opposition's win probability should the war occurs is p with probability 1 − λ and p f with probability λ. We now show under which condition these potential shifts in power lead to a foreign-induced civil war due to commitment problems.

The commitment problem
The existence of the potentially intervening country might cause war both in the presence and absence of a foreign intervention today. We first study the situation when there is no foreign intervention in the current period, namely Fi t = 0. In this case, the incumbent government is momentarily strong but there is the possibility of a future power shift in the opposition's favor. Can the opposition induce the incumbent not to exploit his temporarily advantageous position? The value of war today for the incumbent is given by where δ is the common discount factor. The maximum that the opposition can credibly promise the incumbent in the future is the expected entire spoils minus what the opposition can lock in tomorrow by going to war tomorrow. The value of war tomorrow for the opposition is given by Today, the maximum that the opposition can give to the incumbent is the entire spoils. Hence, the maximum possible transfer to preserve peace from the opposition to the incumbent is given by The incumbent cannot be appeased if π I t war Fi t ¼ 0 , namely if the destructiveness of war is limited to Observe that the left hand side of Eq. (7) is decreasing in λ, so the condition is most easily satisfied if there is a shift in power for sure tomorrow. On the other hand, when no shift in power is expected, λ = 0, the condition can never be satisfied and peace prevails. Indeed, condition (7) can be rewritten as where p is given by Eq.
(1) and p f is given by Eq. (5). Expression (8) gives us a lower bound on the probability of a low intervention cost foreign government so that there are no credible transfers to maintain peace and hence the bargaining range for peace is empty. We now study the opposite situation where there is a foreign intervention in period t, namely Fi t = 1. The momentarily weak player is the incumbent and we need to check whether it can buy off the opposition in order to preserve peace. The value of war today for the opposition is The value of war tomorrow for the incumbent is Hence, the maximum transfer it can credibly offer to appease the opposition is the entire pie today plus the discounted value of the pie from tomorrow minus Eπ t + 1 I (war|Fi t = 1): The opposition prefers war if π O t war Fi t ¼ 1 This expression is increasing in λ. It is easiest to satisfy when there is a power shift for sure in the future, namely for λ = 0 and will never be satisfied in the absence of a power shift λ = 1 Condition (9) can be rewritten as an upper bound on the probability of a low intervention cost foreign government so that there are no credible transfers to maintain peace: We have therefore shown that: Proposition 2. In the absence of a foreign intervention today, the existence of a potentially intervening foreign government causes the incumbent to go to war today if λ N λ I since the opposition cannot credibly promise sufficiently high transfers to the incumbent to preserve peace. If, on the other hand, the opposition receives foreign support today, the incumbent cannot credibly promise sufficiently high transfers to the opposition to preserve peace if λ N λ O . For λ I N λ b λ O both commitment problems strike: a currently strong incumbent and a currently strong opposition will choose war. 19

Observable implications
Besides micro-founding our stylized infinitely repeated bargaining model Proposition 1 allows us to derive two testable predictions concerning the political motives of intervention which we will use to identify the effects of U.S. foreign influence. It states that a foreign intervention is more attractive the lower the personal cost of going to war and the lower the foreign politician's popularity leading to more pronounced power shifts and therefore a higher likelihood of a foreign induced commitment problem.
Prediction 1. Ideology matters: the probability of civil war should increase if the head of the foreign government has a more pro-war ideology and hence lower personal costs c i to initiate a civil war.
Prediction 2. Approval matters: The probability of civil war decreases with the approval of the foreign government within its own country.
These predictions are important since they relate politics in the potentially intervening foreign country to the probability of civil war around the world. Moreover, it is easy to reinterpret our model so that it generally applies to civil war incidence and not only to civil war onset. Our model can describe a foreign caused prolongation of a civil war where a potential intervening country interrupts the peace negotiations in an ongoing civil war.

Empirical exercises
Our analysis shows that ideology and popularity both affect the incentives to intervene in conflicts abroad. We focus now on the case of the U.S. as the source of foreign intervention. If U.S. foreign intervention is a determinant of civil war abroad, and the U.S. propensity to intervene depends on domestic political factors, then we should observe that political changes in the U.S. are systematically associated with the incidence of civil war around the world. As long as political changes in the U.S. constitute an exogenous variation from the perspective of the country potentially in conflict, our predictions provide a way to identify the effect of foreign intervention on the incidence of civil war.
Of course there might be other countries willing to intervene in conflicts abroad. However, the reasons for focusing on the U.S. are obvious. As discussed in the introduction, (i) the U.S. is a global leader with massive economic and political interests all over the world; (ii) there is an extensive record of U.S. interventions; (iii) the Democratic and Republican government may differ in their foreign policies, and thus in their willingness to intervene in foreign conflicts; 20 and (iv) it is very likely that the U.S. citizens vote without these interventions in mind. 21

Data
In this paper, we mainly use the Correlates of War (CoW) measure of civil conflict. More specifically, we identify the conflicts from the "intrastate wars" database. Domestic conflicts are defined as intra-state war every time that there is sustained combat, involving organized armed domestic forces and resulting in a cumulated death toll of more than 1000 battle-related people over a year. The beginning and end dates of the conflict are both included in the CoW database. A conflict is assumed to be over when sustained combat ends or when the number of fatalities declines below the threshold of 1000 deaths. An additional requirement for a domestic conflict to be coded as civil war is the "effective resistance" on both sides. This criterium is used to differentiate wars from massacres and other episodes of one-side violence (e.g. government repression), which are not considered to be the same phenomenon as a civil war. Decolonization wars are not coded as a civil conflict because they imply at least one foreign party involved in the conflict (the colonizer). 22 According to this definition, we construct a civil war dichotomic variable (conflict jt ), which takes the value of 1 (0) if a country j was (not) engaged in a civil war in a given year. Otherwise, confilct jt = 0. 23 Fig. 1 shows the incidence of civil conflicts between 1937 and 1997. As Fearon and Laitin (2003) suggest, the secular increase in civil wars during this period can be explained by the accumulation of unresolved conflicts.
To explore the importance of the U.S. political situation, we use the annual presidential approval rates (PA t ) constructed by Gallup. This variable goes from 0 to 100, reflecting the percentage of surveyed people approving the government. We also define a dichotomic variable indicating whether the U.S. incumbent party is Republican (REP t = 1) or Democratic (REP t = 0). Table 1 reports some statistics describing selected relevant features of the data. According to CoW, there were 93 new civil wars between 1937 and 1997, which lasted on average 6 years. 53 countries (30% of the sample) suffered from at least one civil war episode in the 1937-1997 period. Most of these countries do not belong to the OECD (94%). Spain , Greece (1944-45), and Turkey (1991-97) were the OECD cases. There are around 9 (8.6) average conflicts per year and 19 Observe that the latter condition can only hold if σ N 2 . 20 Republican and Democrats differ in their position vis-a-vis isolationism as well. It is important to clarify that isolationism refers to the U.S. reluctance to be involved in European inter-state conflicts, which is no indicator of their willingness to intervene in foreign conflicts. In fact, and perhaps, paradoxically, isolationism is strongly associated with the origin of U.S. interventions in Latin America, as exemplified by the Monroe Doctrine where it is established that Europe should refrain from influencing the Americas. See Sexton (2011). 21 We refer the reader to the introduction for a justification of these claims. 22 Additional specifications about the "intra-state wars" database can be found at Correlates of War website: http://www.correlatesofwar.org/. 23 As a robustness check, we use the UCDP/PRIO civil-war incidence measure. Although fairly equivalent (the correlation between both measures is very high, about 75% at country-year level) and their use make no difference in terms of our results, which are qualitatively the same and quantitatively very similar. We prefer using the CoW measure as it covers a larger time span, which includes years before and after the Cold-War.
the Cold War seems not to make any substantial difference. The average presidential approval of the U.S. government is 56.3% with a standard deviation of 12.4. Republican administrations represented about 48% of the period. We also control for oil prices and other country specific economic variables. Oil prices are taken from BP world energy statistics, which are based on key crudes quotes from Brent, West Texas Intermediate (WTI), Nigerian Focados and Dubai expressed in US$ per barrel. As in Besley and Persson (2011), we proxy income shocks by exploiting the (arguably) exogenous variation generated by natural disasters. They construct their measure from the EM-DAT database. 24 We define this variable as Natural Disaster jt , which takes the value of 1 if a country j suffers in year t from any of the following calamities: extreme temperature events, floods, slides and tidal-waves. If none of these events take place, Natural Disaster jt takes a value of 0. Statistics on World Population, GDP and Per Capita GDP are taken from Angus Maddison's database. 25 Table 2 reports the average number of ongoing and outbreaking civil wars (based on the CoW data set) under Democratic and Republican administrations for the period 1937-1997. We also distinguish years according to whether the incumbent had low (below the median) or high (above the median) presidential approval rates. The incidence of civil war is about 51% higher under Republican administrations. It is also 32% higher when only the number of outbreaking conflicts is considered. In the second panel, we observe a similar increase in the number of ongoing and outbreaking civil wars in years when the U.S. incumbent suffers from low approval rates. The equality of means tests suggests that the differences are statistically significant.

Time series evidence
Another way to show the association between the political situation in the U.S. and the number of conflicts around the world is to regress the number of ongoing and emerging civil conflicts (in logs) at t on REP t and PA t . In Table 3, we display the results. In columns (1), (2) and (3) we observe that the number of conflicts is significantly higher under Republican governments and negatively associated with the level of presidential approval. In columns (4), (5) and (6), we observe a similar result, although smaller in magnitude, for the number of civil war outbreaks in a year. Observe that in columns (2) and (5) we control for the growth of gross world product (ΔlogGWP t ) and the results remain unchanged. Finally, we control for shocks in oil prices (Δlog Oil Prices t ) in columns (3) and (6). In this way, we control for potential global demand and productivity shocks that might be associated with the incidence and onset of civil war around the world.
While this analysis is suggestive, a more serious test of our theory requires exploiting within-country variations in panel data to which we turn next.

Panel data evidence
We estimate the incidence of civil war; that is, the probability of observing civil war in country j in year t (conflict jt ). To put our results in context, we replicate the empirical strategy developed in Besley and Persson (2011). Consequently, we will mainly rely on Natural Disaster j,t , as an exogenous source of variation in per capita income. 26 As discussed in Section 2, most of the empirical civil war literature fails to exploit within-country variation in panel data, which leads to biased estimates. To avoid this problem, we only exploit within country variations. Thus, country fixed effects (γ j ) are used in all of our main estimations as in Besley andPersson (2011), Brückner andCiccone (2010) or Miguel et al. (2004). To this specification, we add our REP t and PA t variables.
The main difficulty with our empirical strategy is that both REP and PA are year (country-invariant) variables, which makes it difficult to distinguish the effects of Republican governments or presidential approval from any other country invariant year effect, like, for example, aggregate shocks taking place at the world level in a given year. In principle, this should not be a serious source of concern as long as the processes followed by the political cycle or the evolution of preferential approval in the U.S. are independent from the process governing the evolution of the other relevant year fixed effects, like global and U.S. productivity or demand shocks or oil prices. In any case, to mitigate the risk of this unlikely but potential problem, we include the growth of gross world (ΔlogGWP t ) product to capture aggregate demand or productivity shocks. Furthermore, we also include in some specifications the U.S. gross domestic product to control for economic shocks specific to the U.S. (ΔlogGWPus,t). Finally, we also control for changes in oil prices 24 EM-DAT refers to the International Disaster Database, created by the Centre for Research on the Epidemiology of Disasters (CRED). For more information follow: http://www.emdat.be/database. 25 http://www.ggdc.net/maddison/. 26 Alternatively, we could use a direct measure of growth in per capita GDP. As shown in Table 4, the effect of the variables of interest remains unchanged.

Table 1
Descriptive statistics   (Δlog Oil Price t ). This way we control for the most plausible potential sources of civil war that may be omitted behind our REP t or PA t variables.
We also carry out a great variety of robustness checks tackling specific concerns and showing that our predictions do not hold in situations where we expect them to fail. To summarize, we estimate different variations of the following regression equation: where x′ is a vector of additional (country invariant) year variables like the mentioned ΔlogGWP t , ΔlogGWP us,t or ΔlogOil Price t . As we follow Besley and Persson (2011), we expect α 1 to be significantly positive. More importantly for our purposes, Predictions 1 and 2 imply a positive α 2 and a negative α 3 .
Finally, we have to deal with the potential problem of cross-sectional and time-series correlation (Bertrand et al., 2004). To do this, we implement multi-way clustering at the year and country levels, which accounts simultaneously for autocorrelation with a country, as well as for correlation within-year across countries in presence of potential geographic-based correlation (Cameron et al., 2011). Table 4 reports our baseline results. In column 1, we report the most basic specification. Reassuringly, negative shocks in the wage rate or income triggered by a natural disaster raise the probability of observing civil war in a similar way and order of magnitude than in Besley and Persson (2011). Importantly, the coefficient associated with REP t is positive and significant. The magnitude of the estimated effect is far from trivial: with an unconditional probability of conflict of around 5%, this corresponds to an increase of about 60%. We can also observe the significantly negative coefficient associated with our U.S. presidential approval variable (PA t ). This coefficient implies that a fall of 1 percentage point in the presidential approval rating increases the incidence of civil war by 2%.

Main results
The effects of these two variables are robust to any modification we perform on the basic specification. In column 2, we replace Natural Disaster j,t by Δlog pcGDP j,t , which appears to be negatively associated with the incidence of civil war. In the remaining specifications (columns 3, 4 and 5), we include ΔlogGWP t . This way we control for aggregate productivity or demand shocks, which may be correlated with the U.S. political party in office. The associated coefficient is negative but insignificant. In the following estimation (columns 4), we add ΔlogGWP US,t , which controls for GDP growth in the U.S. Including these additional country invariant year variables has qualitatively no effect on neither the way in which Natural Disaster (as a proxy of wage rate or income shocks) or our main variables. Finally, we control for changes in oil prices. The reason is that oil prices may affect both the political situation in the U.S., through its effects on U.S. inflation, and the incidence of conflict via inflation or, for oil producer countries, its effects on national income or revenues. Although we find a statistically significant positive effect of variations in oil prices, a result interesting in itself, the inclusion of this additional year (country-invariant) variable does not affect our main results.

Robustness
We perform a multiplicity of robustness checks, which we expose according to different potential concerns.

Different samples
Our first question is whether our results withstand changes in sampling. Reassuringly, this is not the case. We begin by restricting the sample to OECD countries. Of course, we do not expect U.S. influence to matter for these countries. Column 1 in Table 5 confirms this presumption. In column 2, we restrict the sample to non-OECD countries. Clearly, this results in higher coefficients associated with REP t and PA t . All the specifications control for country-fixed effects. Robust standard errors clustered by country in parentheses. Robust Multi-Way Clustering by country and year in brackets. All the specifications control for country-fixed effects. Robust standard errors clustered by country in parentheses. Robust Multi-Way Clustering by country and year in brackets.
In column 3, we display the results of a counterfactual. We restrict our sample to former French colonies where we should not expect strong U.S. intervention. 27 If anything, these countries are influenced by France. Thus, our results should not hold. As shown in column 3, neither REP nor PA are associated with significant coefficients, strengthening our argument.
Finally, in columns 4 and 5, we explore the possibility that REP is capturing something else rather than variations in the propensity of the U.S. to intervene abroad. Given the strong position of the U.S. president and the clear difference between the Democratic and Republican view on the role of the U.S. in the international arenas, we believe that party ideology should be more important for the case of the U.S. as a potentially intervening country than for other countries. That is, we should not observe that the probability of civil war is determined by which party is in office in countries like, for example, Sweden or even in the U.K. Interestingly, politics in those countries are also characterized by alternating political parties with different ideology so we can create variables like SOC t SW or CON t UK . These new variables take the value of 1 if the government is conservative in the U.K. and socialist in Sweden respectively, and 0 otherwise. Once we include these variables, the coefficients associated with REP and PA are still significant and positive (column 4). This reinforces the view that civil wars are influenced by the U.S. and that U.S. intervention is politically motivated. We see as well that political alternation in the U.K. and Sweden does not affect the incidence of civil war. The proper falsification test is displayed in columns 5 and 6 for the U.K. and Sweden, respectively, where the US domestic politics variables are dropped. 28

Different specifications
We explore different specifications in Table 6. In column 1, we report an estimation with decade fixed effects. We include a quadratic time trend in column 2. These modifications do not affect the qualitative results, although the coefficient associated with REP gets smaller and loses some significance once both decades fixed effects and the time trend are included. In column 3, we explore if the Republican effect is driven by any specific year of the presidential term. We do so by disaggregating REP in the first, second, third and fourth year of a Republican term. All the coefficients associated with REP t Y1 , REP t Y2 , REP t Y3 , and REP t Y4 are positive and significant as we expected. It is interesting to note that the coefficient associated with each year of a Republican term is similar. If anything, the coefficients of the last two years are slightly higher, 27 We thank Benjamin Cohen for suggesting this check. 28 Notice that this result does not imply that the U.K. did not intervene in foreign civil wars. It only implies that the political orientation of the party in power in the U.K. does not determine the probability of a U.K. intervention. The sign of the conservative dummy is positive but not significant, implying as expected that foreign interventions by the U.K. are not very sensitive to the party in power. which might suggest that politically motivated intervention is weaker during the first years in office. We finally explore whether our results are driven by the Cold War. In column (4), we add a dummy to differentiate this period. In line with the literature, we do not find a direct effect of the Cold War on civil war (Collier et al., 2005). 29 We also observe that our results remain unchanged. However, given that including decade fixed effects and time trends reduce the magnitude of the Republican and Approval effects, we aim at clarifying how these effects vary across the Cold War and Non-Cold War years. In columns (5) and (6) we restrict the sample to Cold War years and Non-Cold War years, respectively. Note first that the sign and the significance of REP and PA remain similar across both periods. Notice, however, that the magnitude of the Republican effect is relatively lower in Cold War era. Under the shadow of war with the USSR, it is possible that the U.S.'s decision to intervene abroad be less sensitive to domestic political changes. On the other hand, this fact, if true, does not imply the absence of nuance in the degree of anticommunism between Democrats and Republics. In fact, the sign associated with REP in the Cold War sample is still significantly positive. Moreover, the anticommunist threat might be overstated by a unpopular incumbent. Thus, there is no reason to expect the effect of PA to change. Our results are consistent with this view: the possibility that the appeal of foreign intervention during low approval years remains unchanged no matter if the U.S. be engaged in the Cold War or not.

Alternative stories
Our results might be suffering from a reverse causality problem: could it be the case that American citizens feel in danger if there are too many civil wars around the world and seek safety by voting for a Republican candidate? We address this potential problem by controlling for the number of civil wars taking place during presidential election years (NCW EY ). As reported in column 1 of Table 7, the estimates of the Republican and Presidential Approval effects come out virtually the same.
We also control for the intensity of conflicts around the world by including the number of civil wars in the previous year (NCW t − 1 ). The results displayed in column (2) show that this has no effect on our results.
Importantly, the Republican effect might be driven by particularly interventionist presidents independently of their party ideology. If particularly aggressive presidents happened to be Republican, then we would be reflecting the spurious impression that Republicans are more prone to intervene in foreign civil wars. For example, the U.S. presidential term during which the world suffered the highest number of civil war took place under Ronald Reagan. To control for this, we run all the regressions excluding one US president at the time. We don't report all the regressions to save space. In any case, none of these exclusions affected the results. Columns 3 and 4 of Table 7 report the regressions excluding Ronald Reagan and Lyndon B. Johnson, respectively.
So far, we have not considered the actions of the USSR during the Cold War period. This is important for at least three reasons: first, it is possible that U.S. interventions be the reaction to an intensification of the USSR political activism across the world. Second, it is also plausible that harder USSR interventions raise the appeal of Republican candidates. Finally, the USSR may also directly induce conflicts around the 29 The Cold War has other important effects on civil war. In a recent paper, Balcells and Kalyvas (2010) show that the effect of the Cold War is to shape the form, not the incidence, of civil war. For example, they show that insurgence (guerrillas or irregular wars) is the dominant form of conflict only during the Cold War. world. Thus, USSR interventions may simultaneously affect both republican elections and civil conflict. If this is the case, there is a potential endogeneity problem that would bias the coefficient estimations. To mitigate this concern, we control for USSR influence during the Cold War by including a measure of KGB interventions. This variable KGB jt takes the value of 1 if country j suffers from successful KGB actions in year t. We borrow this variable from Berger et al. (2013). They use historical studies to identify KGB interventions and apply an analogous methodology to CIA interventions, which we explain in further detail in Section 4.7. In column (5) of Table 7, we report our baseline estimation restricting the sample to Cold War years and including KGB jt in the regression. The effect of KGB jt is significant and positive at the 10% level, which suggests that KGB operations did have direct effect on the incidence of civil in countries where they took place. Importantly, controlling for their influence does not affect the coefficients associated with either REP t or PA t in a relevant way. Furthermore, KGB interventions may reflect a climate of world conflict. To control for this, we define NKGB t as the total number of KGB interventions in year t. The result of including this variable is displayed in Column (6) of Table 7. The coefficient associated with NKGB t is significant, although not robust to multi-way clustering by year and country. Again, our key results resist the inclusion of controls for potential alternative stories.

Instrumented presidential approval
Presidential approval rates might be plagued by another source of reverse causality: Americans may perceive that a world with greater civil wars represents a failure of American foreign diplomacy and peacekeeping efforts and punish the American president with lower approval. To tackle this concern, we instrument presidential approval ratings by inflation and GDP growth in the U.S. (Inflation rate US,t and ΔlogGDP US,t ), which are clearly unrelated to international events. 30 In Table 8 we report the results for the 2SLS estimation. Inflation and economic growth appear as valid instruments for presidential approval in the U.S. Importantly, the second-stage estimates suggest that the instrumented PA variable is negatively associated with the incidence of civil war. Observe as well, that the effect of REP remains the same.

Other robustness checks
We have also run a great variety of unreported additional regressions. 31 These are as follows: (i) we use PA only but not REP to get more year variations (ii) we estimate conditional logits for all the specifications; (iii) we use the UCDP/PRIO measure of civil-war incidence; (iv) we try with different samples and run our regressions separately for Sub-Saharan and commodity exporters countries and (v) we replaced Natural Disaster j,t by the actual measure of GDP growth and (vi) as in Collier and Hoeffler (2004), we control for the type of political regime by adding a new variable that takes the value of 1 for democratic countries defined using the Polity IV measures of democracy. The effect of U.S. political factors on the incidence of civil war withstands any of these robustness checks.

The onset of civil war
Our theoretical analysis shows that foreign intervention increases the occurrence of civil war by triggering new conflicts and prolonging existing ones. For this reason, our main empirical investigation is on the incidence of civil war, which captures both dimensions of a civil war. We check now whether our insights persist once the onset of civil is considered instead. We report in Table 9 our basic specification (columns 1-4). To give an idea of robustness we control for the Cold War years (column 5) and for the number of conflicts around the world per year (column 6). 32 Although weaker, the effect of our variables is robust to considering the onset of civil war, which we interpret as evidence of the influence of U.S. politics on the emergence of civil conflicts abroad. 33

A channel of influence
We have so far provided strong evidence of the empirical association between the political situation in the U.S. and the incidence and onset of civil war around the world. As this influence is often channeled by the CIA, we investigate in this section whether the occurrence of CIA operations in conflicts abroad varies across the ideology and the approval of the U.S. government. We rely on a measure of CIA interventions used in Berger et al. (2013), which is based on recently declassified CIA covert actions aiming at supporting or deposing foreign political leaders. 34 As this variable is only available for the Cold War period, 1947-1989, we restrict our sample accordingly. 35 Unfortunately, this variable captures only successful CIA operations. That is, CIA operations that succeeded in imposing the ally of the US. This is restrictive as our theoretical results do not depend on whether the party supported by the foreign power wins or loses the conflict. These caveats notwithstanding, the relationship between the incidence of successful CIA activities in 30 See Berlemann and Enkelmann (2012), for a survey of the determinants of U.S. presidential approval. conflicts abroad and the political situation in the U.S. supports the existence of politically-motivated foreign interventions abroad and, consequently, it is interesting in its own right. Table 10 provides suggestive evidence according to which the mean of (successful) CIA operations per year is larger for low presidential approval rates and under Republican administrations. Standard errors in parentheses.

Table 9
The onset of civil war.

Table 11
The incidence of successful CIA operations.
(1) Given that the number of (successful) CIA operations is higher under Republican administrations and in years with low U.S. presidential approval, we test this relationship further by estimating the probability for a country j to experience a successful CIA covert operation in a given year (CIA jt ). 36 Table 11 reports the results. As can be seen in column (1), the incidence of a CIA (successful) operation is higher under Republican governments. However, the effect of presidential approval is not significant. Recall that PA t is a continuous variable ranging from 0 to 100, reflecting different degrees in presidential approval. Notice as well that using CIA t reduces the number of observations by almost a half, reducing considerably the power of the estimation. For this reason, in the following estimations (columns 2 to 5) we use a dichotomous version of PA t that takes the value of 1 if the level of presidential approval is high (above the median) and 0 otherwise. The effect of this variable (High PA t ) turns out to be negative and significant. This is clear in all the remaining estimations. In column (2), we report the effect of REP t and High PA t alone. In column (3) we add ΔlogGDP US,t , ΔlogGWP t and Δlog Oil Prices t as controls. This result clearly relates the incidence of CIA operations around the world with the U.S. political situation. The remaining results show that this effect is stronger for developing countries. In columns (4) and (5), we restrict the sample to commodity exporters and non-OECD countries respectively. The coefficients associated with REP t and High PA t become stronger and gain significance.
All in all, the estimations displayed in Table 11 suggest that the incidence of CIA (successful) operations is politically motivated, especially for developing countries, in the same way as the incidence and onset of civil war.

Discussion and conclusion
In this paper we showed that the existence of a potentially intervening country might trigger or prolong a civil war due to commitment problems. We briefly discuss another equally interesting channel how a secretive intervention might affect civil war incidence in a bargaining framework, namely by introducing persistent information asymmetries. While information asymmetries are a central theme in the literature on rationalist explanations of war (see e.g. Jackson and Morelli, 2011) and accepted as causes of war, it is generally argued that asymmetric information cannot fully explain long lasting conflicts because both sides will learn the true information over time (Fearon (2004)). This insight is destroyed in the presence of a potentially intervening country. Whenever a deal is offered to one of the domestic parties, the side allied with the foreign country is likely to be better informed about future investment plans of the foreign government in case of victory and hence the final size of the spoils. It is also likely to have better information about the amount of resources the foreign country is willing to provide in case of a conflict. More importantly, the exact amount of foreign resources depends on political factors in the foreign country that are highly uncertain and better understood within an alliance since they are not directly observable from the domestic country. These fluctuations are exogenous to the domestic parties in conflict and might lead to long lasting information asymmetries, which change over time and cannot (rapidly and evenly) be learned. This way, foreign interventions generate persistent uncertainty over the fighting resources available for each party in conflict which might explain even long-lasting conflicts.
The main theoretical novelty of the paper was to explicitly analyze the incentives for a third party to intervene which leads to two clearcut predictions that provide an identification strategy for the relevance of foreign intervention on the incidence and onset of civil war. Both predictions are confirmed for the case of the U.S. as a potential intervening country: (i) civil wars are more likely to take place when the U.S. is under a Republican government and (ii) the probability of civil wars decrease with the U.S. presidential approval rates. These empirical results, relevant and novel in themselves, show that foreign influence is an important determinant of civil war around the world.