Just War or Just Politics? The Determinants of Foreign Military Intervention Averyroughdraft.Thankyouforyourcomments. Shannon Carcelli UC San Diego scarcell@ucsd.edu January 22, 2014 1 Introduction Under what conditions do states employ military force abroad? Researchers have posited various explanations for foreign military intervention, ranging from humanitarian to security objectives. To date, no author has empirically tested the role of economic interdependence in the decision to intervene in a foreign state. Further, no research on foreign military intervention has used a panel dataset with directed dyad-year units. Studies have focused mainly on either the United States or civil wars as the unit of analysis, both of which ignore important cases of intervention. I will examine the role of a series of factors on outcomes of foreign military intervention. 2 Hypotheses Ihypothesizethatthedecisiontointerveneunilaterallywillbedrivenmainlybyeconomic and domestic political determinants. Although interventions in general will be more likely in a humanitarian crisis (during a civil war or a great deal of civilian killing) within a target state, the best determinant of intervention in a dyad will be political characteristics of that dyad, specifically trade flows. 3 Data 3.1 Dependent Variable: Foreign Military Intervention I draw the dependent variable of interest from a foreign military intervention dataset compiled by Pickering and Kisangani (2009). This dataset defines foreign military intervention as the movement of regular troops or forces (airborne, seaborne, shelling, etc) of one country inside another, in the context of some political issue or dispute (Pearson and Baumann, 1
1993, 1). It excludes accidental troop movements, military bases, and non-governmental militaries or militias. It defines the actor and target states of each intervention as well as attempting to denote the purpose of the intervention (economic, humanitarian, territorial, etc.), the direction of the intervention (e.g. siding with or opposing the target state), as well as several other variables. It sources this information from various news archives. This dataset contains instances of military intervention from 1948 to 2006, but I limited my analysis to 1989 and onward. I restrict the universe of data for reasons of data availability as well as the fact that the world witnessed a great increase in such intervention in the post-cold War period. My main humanitarian measure, civilian casualties, is only available from 1989 onward. This makes it di cult to rule out humanitarianism as a determinant of intervention before 1989. This unfortunate restricting of the data is not as problematic as it first may seem. As Figure 1 indicates, worldwide incidence of military intervention spiked after the Cold War, and the type of intervention has shifted over time as well. Restricting my analysis to post-cold War interventions allows me to ensure that I study only the contemporary phenomenon of foreign military intervention. As a robustness test, I will also drop the humanitarian variable and carry out statistical tests for all years in the intervention dataset. Insert Figure 1 about here Along with excluding early years from the dataset, I also exclude interventions by the United Nations. The research I intend to pursue has nothing to say about United Nations interventions, as my focus is on the domestic political and economic determinants of the decision to intervene. Because the United Nations makes intervention decisions in such a di erent manner from individual states, these data would only add random noise to the model. This is because actors voting upon a UN intervention (member states of the UN Security Council) are not necessarily the ones expending the resources to carry out the intervention. This complicates the costs of intervention in a way inconsistent with the mechanisms of my theory. Because the theory I will be testing with these data only makes predictions about interventions in which the actors deciding upon the intervention are the same ones paying the costs of intervention, excluding United Nations interventions will allow me to focus only on the cases relevant to my theory. A similar argumet could be made (albeit less strongly) for many multilateral interventions, and I will therefore also test my data on subsets of only unilateral or only multilateral dependent variables. I recode this dataset to report the presence or absence of a military intervention for every directed dyad-year. Dyad-years that appear at least once in the dataset are coded as 1, and all other directed dyad-years are coded as 0. In very few cases, a directed dyad was coded to have experienced more than one intervention in a given year. I drop this variation in favor of a dichotomous dependent variable. Because the intervention variable is an arbitrary magnitude anyway it does not distinguish large from small interventions IargueIamnotlosinganyvariationfromdoingthis. ThevariableIntervention is coded as a 1 if the first state in the directed dyad, the Intervener, begins a military intervention directed toward the second state, the Target, in the year in question. It is coded as a 0 otherwise. Coding only the onset of an intervention has a number of advantages. First, it allows me to determine which factors were present when the intervening state made the final 2
decision to intervene. While the decision to maintain an already-existing intervention is an interesting one, I expect that the criteria with which a state chooses to initiate an intervention di er from those with which a state chooses to continue an existing intervention. Second, Iexpectsomeofmyindependentvariablesofinteresttochangebothasacauseandasan e ect of military intervention. Focusing only on the onset of an intervention minimizes the endogeneity of these variables. Third, it reduces problems of serial auto-correlation, which would be rampant if a state continued an intervention within the same target for multiple years. 3.2 Independent Variables To examine the role of humanitarian factors on foreign military intervention, I include a variable indicating the number of fatalities due to one-sided civilian killings, according to the Uppsala Conflict Data Project (UCDP) One-Sided Killing dataset. This dataset codes the number of civilian killings in each country-year from 1989 to 2011, based on a variety of news sources (). Because much of the humanitarian military intervention literature focuses on interventions targeted toward civil wars (e.g. ), I also include a binary indicator of whether or not the potential target state is engaged in a civil or interstate war in the given year, according to PRIO data (). I predict that all three of these factors will play a role in any state s decision to intervene, but the domestic factors within intervening states will be the final deciders of which states are the ones to carry out an intervention. In other words, humanitarian factors give states an opportunity to intervene, whereas their political position gives them the motivation to do so. In order to examine the role of domestic political motivations on foreign intervention, Iexaminebilateraltradebyincludingavariableindicatingthelevelofunexpected(either positive or negative) bilateral trade flows between the intervening and target state in the given year. I determine this by first using a list of dyad-year level variables compiled by CEPII () that often predict trade, and regressing these variables on trade flows of both states, in both directions. In doing so, I am essentially running a gravity model, commonly utilized in international trade literature and generally found to be very predictive of bilateral trade. It includes such elements as the size and GDP of states, whether or not they are members of the General Agreement on Trades and Tari s, their linguistic and colonial similarities, and their physical distance from each other. Whatever bilateral trade is not predicted by such a model can be supposed to derive from other idiosyncrasies of either the states themselves or the states relations with each other. These idiosyncrasies are precisely what I am attempting to measure. Ipositthatstatesthattradewitheachotherinunpredictedamounts whetheritbe more or less than predicted have a special political relationship that will a ect their relationship in other ways as well. States that do not trade with each other as much as they otherwise would have may present a certain amount of unobserved cultural, historic, or economic emnity, which could increase their probability of conflict. On the other hand, states that trade with each other more often than expected may have domestic actors who rely upon each other enough to dampen their aggressiveness. It is important, however, to consider trade flows in both directions. The dependent variables of the gravity models I run are the level of trade in each direction. Separating out the two directions of trade flows 3
allows me to better understand the direction of trade that is spurring or dampening monadic conflict. Further, because the military intervention dataset defines events monadically, it is important that I di erentiate the trade data accordingly. I therefore create two separate variables, res1 andres2, to define the residuals of the gravity model in each direction. res1 denotes the unexpected level of exports from the target to the intervening state, and res2 denotes the unexpected amount of imports from the intervening to the target state. I collect these trade data from the Correlates of War (COW) project (). In addition to trade, I also examine the alliance status of the two states. I predict that a military alliance will make an intervention more likely, because states generally create alliances with, and intervene in, partners whose actions they care about. I will later disaggregate these data to determine whether states are more likely to intervene on the side of the state, rather than against it, when they have signed an alliance with that state. I also examine the di erence between the two states military capability scores. This ratio represents a state s material ability to fight a war, accounting for GDP, certain natural resource deposits, population, possession of weapons technology, and other relevant variables. Because my unit is the directed dyad, I am not interested in these raw capability scores but in the di erence between the relative capabilities for the two states. I therefore take the di erence between this score for the intervening state and that of the target state. A larger number indicates that the potentially intervening state is more powerful, and a small (negative) number indicates that the potential target state is the more powerful in the dyad. Iexpectthat,allelseequal,stateswillchoosetointerveneintargetsthatarelesspowerful than they are, which will give me a positive coe cient for the variable, which I call cincdiff. I take all of the preceding variables from the Correlates of War dataset (). Finally, I include a variable for the dyad s S-score, or the similarity of their UN General Assembly voting records in the given year. I use Gartzke s () data for this variable and expect to find that states that vote against each other in the UN are more prone to interventions, which would be apparent a negative coe cient. 3.3 Controls Icontrolforfactorsthatmayfurthera ectastate sopportunityorwillingnesstoengageina military intervention in another state s territory. These include colonial heritage, contiguity, oil production in the target and intervening states, GDP, and per-capita GDP for both states. IcollectthesedatafromWorldBankandCorrelatesofWardatasets.Ialsocontrolforthe regime type of each state, based on their polity score, as the international relations literature has witnessed a strong empirical regularity involving regime type and war (). Although these variables will no doubt have some e ect on outcomes of military intervention, it is beyond the scope of this paper to hypothesize on the role of these factors. To minimize the possibility that a state may be a particularly strong target for intervention, I also control for any other intervention that may have taken place within the target state in the previous year. 3.4 Table of Means Insert Table of Means about here 4
4 Methodology 4.1 Statistical Models To determine the factors leading to a foreign military intervention, I create a time-series cross-sectional dataset with a directed dyad unit. Because the number of interventions in this dataset is very small compared to the number of dyad-years, I estimate a rare events logit following King et al (). I run five separate models, whose outcomes can be viewed in Table 2 below. Model 1 includes only the trade variables of interest. Model 2 includes these trade variables, along with a subset of independent variables that were not found in the gravity model. Model 3 includes all control and other independent variables. Model 4 includes all of these controls but adds dyad-level fixed e ects. Model 5 estimates Model 4 with robust standard errors, clustered at the level of the intervening state (or at least it will once I figure out why my standard errors are lower than when I did not cluster them. For now Model 4 is there again as a placeholder). 5 Results As evident in Table 2, trade and military intervention have a complicated relationship. An increase in trade flows in one direction from target to intervener predicts an increase in military intervention. On the other hand, an increase in trade flows in the other direction from intervener to target state decreases the probability of a military intervention. This could be the case for a variety of reasons, which I will discuss below. While the coe cient on trade flows to the intervening state is highly robust, the coe cient on trade in the other direction fails to surpass the p = 0.05 level in one model. However, its robustness in all other models indicates with relatively high confidence that trade is in fact having an e ect. 5.1 Robustness I carried out several additional tests to ensure the robustness of these data. First, there is a danger that trade flows are endogenous to the decision to intervene. In other words, if I am looking at trade during the year of a military intervention, the intervening state s decision to trade with the target state may be driven by the presence of a military intervention. I account for this danger by running a separate test replacing all independent variables with the same variables from year preceding the intervention, as well as their mean values for the five years preceding the intervention. The trade variables of interest to me remain significant throughout these models. I also carried out the same tests using all dyad-years instead of limiting them to 1989 and onward, and found no di erence in my models. Finally, I reran these tests but replaced the dependent variable with only unilateral and only multilateral military interventions. As would be expected, I find that unilateral intervenion is driving these findings, and that the e ects of the political variables especially are muted in the multilateral model. This is as expected, especially as it relates to the trade variables. The mean of the residuals from the gravity model is (by definition) 0. Therefore, the more states are involved with a target state, the more muted I would expect the e ect to become. 5
5.2 Discussion While it is not an uncommon finding that trade decreases the probability of conflict (Russett and O Neal), and some have even found trade to increase the probability of conflict in some circumstances (Barbieri), no authors to my knowledge have found the direction of trade flows to be the determining factor. The fact that trade flows toward the intervening state increases the likelihood of intervention and trade flows in the other direction has the opposite e ect is puzzling. There are a variety of explanations for the complicated relationship outlined above. First, and most simply, perhaps the target state has something that the intervening state wants. If state-level actors are the ones importing these goods, we might expect them to tire of purchasing them and send the military into the target state. This is especially likely if the intervening state is dependent upon the target state for these goods. A closer look at the actors doing the trading and the goods in question will further our understanding of that possibility. The positive, but not robust, coe cient of the oil production control indicates that oil may be a driver of these findings, but a bit more in-depth statistical research, with better data, would be necessary to put that hypothesis to rest. A final hypothesis is that state actors within the intervening state are protecting actors within the target state upon whom they rely for certain resources and/or manufacturing. Disaggregating the interventions based on which actors the interveners support would be a good way to begin looking at this possibility. In an inital test, I found no significant result using disaggregated data, but there is more work to be done. 6 Conclusion Foreign military intervention depends more heavily upon trade dependence than previous research has suggested. This complex dependence has not been possible to predict by others studying the relationship between trade and conflict, as they have generally seen conflict as a dyadic, rather than a monadic, event. Because military intervention is monadic, this research project is uniquely able to di erentiate the type and direction of trade that leads to intervention decisions. There are several ways forward for this line of research. First, it will be important to determine the type of trade that is leading to these interventions. Are intervening states purchasing raw or manufactured goods from the states in which they intervene? Similarly, are these goods in which the intervening state has a comparative advantage or disadvantage? Finally, which actors within the intervening and target state benefit financially from these interventions? Future research must disaggregate these findings in order to further understandings of the mechanisms leading from trade to military intervention. 7 Appendix 6
Table 1 Statistic N Mean St. Dev. Min Max intervention 103,534 0.001 0.039 0 1 trade to intervener 103,534 58.660 2,114.476 8,622.356 259,918.500 trade to target 103,534 55.230 1,388.062 6,261.124 133,115.400 civilian death 103,534 1,121.203 35,666.580 0.000 1,252,672.000 pcgdp targ 103,534 9,216.675 14,061.120 50.042 80,925.220 pcgdp int 103,534 2,932.483 6,344.440 50.042 55,526.160 gdp targ 103,534 297,308,919,169.000 1,204,022,549,878.000 104,083,836.000 12,564,300,000,000.000 gdp int 103,534 97,766,595,391.000 231,049,605,610.000 104,083,836.000 2,256,900,000,000.000 oil targ 103,534 4.204 12.542 0.000 209.481 oil int 103,534 6.790 16.922 0.000 209.481 civil war targ 103,534 0.335 0.472 0 1 interstate war targ 103,534 0.0001 0.009 0 1 cincdi 103,534 0.002 0.029 0.184 0.184 sscore 103,534 0.681 0.284 1.000 1.000 polity2 targ 103,534 0.490 6.236 10 10 polity2 int 103,534 4.443 6.147 10 10 alliance 103,534 0.063 0.244 0 1 colonial 103,534 0.009 0.095 0 1 intervention.1yr.target 103,534 0.113 0.317 0 1 7
2013/Humanitarian Intervention/interventions.png Figure 1 8
Table 2 Military Intervention intervention (1) (2) (3) (4) (5) Constant 7.100 7.254 8.577 1.320 1.320 (0.048) (0.276) (0.392) (4.238) (4.238) trade to intervener 0.0003 0.0002 0.0001 0.0002 0.0002 (0.00002) (0.00003) (0.00003) (0.0001) (0.0001) trade to target 0.001 0.001 0.0002 0.001 0.001 (0.00004) (0.0001) (0.0001) (0.0002) (0.0002) civilian death 0.00000 0.00000 0.00000 0.00000 (0.00000) (0.00000) (0.00000) (0.00000) oil targ 0.011 0.012 0.015 0.015 (0.005) (0.006) (0.009) (0.009) oil int 0.004 0.005 0.013 0.013 (0.003) (0.004) (0.012) (0.012) civil war targ 0.915 0.893 1.780 1.780 (0.157) (0.175) (0.296) (0.296) interstate war targ 3.875 2.434 3.464 3.464 (1.096) (1.192) (1.878) (1.878) cincdi 12.605 12.651 4.781 4.781 (2.770) (3.998) (27.265) (27.265) sscore 0.338 0.015 0.487 0.487 (0.290) (0.396) (0.699) (0.699) polity2 targ 0.035 0.047 0.098 0.098 (0.013) (0.015) (0.030) (0.030) polity2 int 0.011 0.009 0.188 0.188 (0.016) (0.020) (0.064) (0.064) alliance 0.355 1.323 1.323 (0.252) (0.513) (0.513) contiguity 2.899 4.778 4.778 (0.317) (0.596) (0.596) pcgdp targ 0.00003 0.00005 0.00005 (0.00001) (0.0001) (0.0001) pcgdp int 0.00000 0.001 0.001 (0.00002) (0.0004) (0.0004) gdp targ 0.000 0.000 0.000 (0.000) (0.000) (0.000) gdp int 0.000 0.000 0.000 (0.000) (0.000) (0.000) colonial 2.285 1.899 1.899 (0.253) (0.365) (0.365) intervention.1yr.target 0.928 0.809 0.300 0.300 (0.178) (0.201) (0.215) (0.215) fixed e ects intervener, target intervener, target Observations 531,390 111,032 103,534 103,534 103,534 Log Likelihood 3,990.387 1,159.288 880.197 587.742 587.742 Akaike Inf. Crit. 7,986.774 2,344.576 9 1,808.394 1,693.484 1,693.484 Note: p<0.1; p<0.05; p<0.01