Abstract
Major powers often deploy troops abroad with the consent of host states. The stated aim of these deployments is often both to protect the host state and to foster stability in the region. Drawing from an extension of Palmer and Morgan’s two-good theory of foreign policy, the authors explore some of the (perhaps unintended) effects of troop deployments abroad on the foreign policies of the host states. In particular, the authors focus on the effects of US deployments. The authors argue that as the number of US troops deployed to a host state increases, we should expect the host state to reduce its own troop levels, be more likely to initiate militarized interstate disputes, and be less likely to be the target of interstate disputes. The authors test these hypotheses using data on US troop deployments abroad from 1950 to 2005 and discuss implications that their findings may have for US foreign policy.
Major powers often deploy troops abroad. 1 While some of these deployments are part of an invading force, they are more often deployed with the consent of the host state. These troops may be personnel in a permanent base (such as Soviet troops deployed to East Germany during the Cold War), or part of a temporary mission in a noncombat role (such as providing disaster relief, as in the case of US troops deployed to Haiti in 2010). Though their form may vary, the stated aim of most of these noncombat deployments is to provide protection and economic benefits to the host state as well as to bring stability to the region. 2 While this may be the stated aim of a deployment, the effects of stationing troops abroad may not always be the expected ones.
This article seeks to identify the effects, both intended and unintended, of noncombat troop deployments in foreign countries. The general theoretical basis for this explanation is drawn from Palmer and Morgan’s two-good theory of foreign policy, 3 which has been extended by Kobayashi and Morgan 4 to explain how one state’s policies will affect policy decisions in another state. This theory leads to a number of hypotheses associating the stationing of troops abroad with other states’ behavior. Specifically, the theory provides an explanation for how troop deployments influence host countries’ decisions regarding their own troop levels, their decisions regarding whether to initiate militarized disputes, and other countries’ decisions regarding whether to initiate a militarized dispute against the host state. This general explanation is then applied to the development of specific hypotheses regarding the effect of overseas US troop deployments. These hypotheses are tested empirically using data on US deployments from 1950 to 2005. The following section contains a brief review of related literature, which is followed by the theoretical argument and the hypotheses. The next sections outline the research design and present and discuss the empirical results. The article concludes with a discussion of the results and their implications for major power troop deployments abroad.
The Study of Troop Deployments
While the troop deployments that are most noticed and discussed are those that are part of a war-fighting force, it is also important to note that most troop deployments happen under friendly (and consensual) circumstances. Noninvasion deployments are a common method for deterring third-party states from attacking host states. By placing their troops in the territory of these nations, sender states are aiming to send a credible signal that they would defend the host if it were attacked. The fact that these troops are placed “in harm’s way” sets up a trip-wire mechanism such that any attack on the host state would also be an attack on the troops of the sender state. Thus, countries use their troop deployments as a form of extended deterrence. 5 Deployed troops send a credible signal not only because of this trip-wire mechanism, but because troops stationed abroad tie the sending state’s prestige to its commitment, thus making it more likely that it will intervene in favor of its ally, if attacked. 6
This type of mechanism was at work in the case of American troop deployments in the second half of the twentieth century, when the United States began to expand its global military presence. In this time period, the location of US troops often reflected where they were at the end of the last war in which they were involved. Some examples are the cases of Germany and Japan after World War II, South Korea after the Korean War, or the Balkans after the Bosnian War. 7 In these cases, one aim of these troops was often to keep the host state from remilitarizing, and to do so by providing for the security of the host state. Thus, host countries would not have to militarize, as the US troops were already serving the function of protection for them. 8 As the US troops were expected to provide security for the host state by deterring invasion, it was also expected that these troop deployments would lead to regional stability in the area in which they were stationed. 9
The deployment of US troops abroad has also been used as a signal of American foreign policy interests and as a strong indicator of American commitment to a particular region. The recently updated official National Military Strategy of the United States refers to global deployments as the United States’ “most powerful form of commitment.” 10 This commitment derives in part from the trip-wire mechanism of putting American troops in harm’s way, but even deployments in which the troops are not directly at risk are used by the US as an expression of commitment. 11 The American goal is that by maintaining US troops in a region of other countries, host nations and third parties alike, will be convinced of US commitment, even if the deployment is not militarily or strategically essential to any particular campaign. 12
Much of what has been written in recent years on the subject of troop deployments abroad under noncombat circumstances (particularly the deployment of US troops) deals with either the original aims of the troop deployments or with the perception of these troops by their host states (or even by the US public). For example, a vast number of articles have been written on Japanese perceptions of US troops stationed in Okinawa and their diplomatic effects. 13 These articles focus on Japanese distaste for the expenses associated with high levels of US troops, as well as for the perceived special treatment that these troops receive. In contrast, the focus here is on the effect that troop deployments have on the international interactions of their host states.
To the extent that scholars have addressed the effects of US troops abroad, their work tends to focus on the economic effect that US troops have on host states. The general idea behind much of this work is that although US troops are often deployed to conflicted states that one would consider unattractive to US investors, these troops might, by maintaining stability in the host state, actually make foreign countries more attractive to US investors. 14 Thus, through US agencies such as the Overseas Private Investment Corporation, which offers insurance to US firms investing in countries in which the United States maintains a military presence and through the general “security umbrella” provided by this military presence, US troops have been suggested to lead to greater foreign direct investment, trade, and economic growth. 15
All these previous works share the assumption that US troops make the host states more secure and then explain how this security leads to economic benefits. This article does not make that assumption; rather, it seeks to determine whether US troops really do make host states more secure by reducing the chance they are attacked by others or by decreasing their own militaristic behavior. It aims to develop an explanation of how foreign troop deployments affect the decision process of the host state when it is developing its foreign policy.
General Theory
The theoretical argument is taken directly from the two-good theory of foreign policy developed by Morgan and Palmer, 16 and extended by Kobayashi and Morgan. 17 This theory, and particularly the Kobayashi and Morgan extension, explains how the external shock of US troop presences may lead to changes in the foreign policy actions of the host states. The theory is based on a formal, mathematical, model; but, since the formalization can be found in the original works, the mathematics is kept at a minimum, presenting instead the intuition behind the argument. This section begins by discussing the original theory and then introduces the extensions that allow for the derivation of hypotheses relating troop deployments by one state to behaviors of other states.
The two-good theory of foreign policy was developed as a general theory of foreign policy. It requires the adoption of a somewhat different view of foreign policy than is common in a number of respects. First, and foremost, the theory is based on the notion that states seek to produce two goods, called maintenance and change, through their foreign policy behaviors. This stands in contrast to the common, neorealist perspective on foreign policy, which presumes that states seek only security. Change and maintenance are viewed as composite goods that actually incorporate all of the specific things states can seek through their foreign policies. Maintenance-seeking actions refer to all actions taken to preserve aspects of the status quo that are in accordance with a state’s preferences. Alternatively, change-seeking actions refer to all actions taken to alter those aspects of the status quo with which the state is dissatisfied. 18
While state preferences play an important role in determining foreign policy decisions, states are constrained by their capabilities and by their ability to translate these capabilities into change and maintenance. This ability is affected by other states and their behaviors as well as by the “technology” available to the state. It is assumed that the ability to produce both goods increases with capabilities. However, a state’s ability to produce maintenance increases at a decreasing rate while its ability to produce change increases at an increasing rate. 19
Given a state’s preferences and the constraints placed on it by the environment, an equilibrium (in terms of the amount of change and maintenance it will produce) foreign policy can be identified. The actions that scholars typically observe and try to explain (military spending, joining alliances, initiating conflicts, providing foreign aid, implementing economic sanctions, etc.) are the instruments that translate capabilities into change and maintenance. Each of these behaviors can be characterized by the marginal efficiency with which it can produce one, or both, of the goods. In this theory, “foreign policy” is conceptualized as a portfolio of policies and “explaining” foreign policy involves demonstrating how various factors affect the allocation of resources across the instruments that make up the foreign policy portfolio. 20
Again, the assumptions of this theory are more completely justified and this theory is explained more fully, as well as more formally, in other sources. 21 For the purposes here, the key point of the theory is that there are three factors that determine a state’s foreign policy: its relative preferences for change and maintenance, the resources it has to devote to foreign policy, and the relative efficiency with which the instruments available can translate resources into change and maintenance. One of the major shortcomings of the theory, however, is that it does not account for strategic interaction. That is, it does not allow us to evaluate how the behaviors of one state will affect the behaviors of others.
An extension of the theory proposed by Kobayashi and Morgan is intended to address, at least partially, this shortcoming. This extension focuses on providing expectations for how a foreign policy action taken by one state will alter the foreign policy portfolio (i.e., how resources are allocated across policies) of other states. This is particularly appropriate for deriving expectations regarding how troop deployments to another state should affect that state’s policies.
The underlying premise of the extension of the two-good theory rests with the fact that the initial theory states that there are three factors that can alter any state’s foreign policy. First, its policy portfolio can be affected if the state’s preferences regarding trade-offs over change and maintenance are altered. Second, its policy portfolio will change if there is a change in the resources 22 the state can devote to foreign policy. Finally, its policy portfolio can change if there is a change in the relative efficiency with which the foreign policy instruments it has available can translate resources into foreign policy goods. Thus, the actions of one state will affect the foreign policy of another by influencing the other’s preferences, resources, or the relative efficiency of its policies. In order to examine these relationships, Kobayashi and Morgan 23 modify the two-good theory in two principal ways.
First, the model is generalized to an N-good model and the “goods” are permitted to be domestic or foreign. The motivation for this comes partly from a desire to be more general; but, more importantly, from recognizing that many foreign policy actions are intended to influence domestic matters in the target state. Development assistance, for example, is intended to provide economic infrastructure and growth in the recipient. One important consideration is that such aid can free resources that the recipient can devote to other goods. So, a donor that evaluated its aid policy and found that its funds do, in fact, lead to economic development in recipient states might well view its policy as a success. The theory implies that the donor should also determine how its aid affects other recipient policies. If, for example, aid frees up the recipient’s resources so that they can then be used to repress minorities or attack neighboring states, the evaluation of the policy might be very different. Extending the model to N goods allows for the capture of such possibilities.
Second, a parameter is added to the model that captures the effect on a state’s policies of an external shock. An “external shock,” simply refers to something coming from outside the state that provides (or destroys) some quantity of one of the goods it values. If, for example, one state bombs components of another’s economic infrastructure (say a railroad, a dam, or a nuclear facility), the target would lose something of value and would probably devote resources to replacing what was lost. Conversely, if one state provides something to a recipient (perhaps protection from attack through the stationing of troops), the recipient should at least consider altering its policy portfolio.
The full mathematical development of this theory is beyond the scope of this article and can be found elsewhere. 24 Still, we can derive a variety of hypotheses based on the general arguments of the theory. 25 First, the theory suggests that as the resources available for policy increase, resources devoted to all policies will increase. Second, a shock will have an inverse effect on the policy at which it is directed and a positive effect on all other policies. This suggests that a positive shock (i.e., one that increases some instrument available to the recipient) will lead to an increase in resources devoted to all other instruments of policy and a decrease in the resources devoted to the shocked policy. Note that this predicts an overall increase in the shocked policy, but the increase that comes from external sources will be partly offset by a decrease in the recipient’s own resources devoted to the affected policy instrument. Third, the theory suggests that increasing the efficiency with which one instrument produces some good will lead to an increase in the resources devoted to that instrument and a decrease in the resources devoted to all other policy instruments. Finally, the theory implies that increasing the salience of a good will lead to an increase in resources devoted to all policies that produce that good and a decrease in resources devoted to all other policies. It should be obvious that the converse of each general hypothesis also holds; that is, if an increase in x leads to an increase in y, then a decrease in x will lead to a decrease in y.
On the basis of these general, abstract conclusions, we can specify a number of specific, empirically testable, hypotheses regarding the topic of interest—the effect of troop deployments abroad (particularly, for this article, the effect of US troops). In general terms, troops can be deployed to serve two purposes. They can afford some protection against attack to the host state by augmenting its own forces and they can serve to provide a means by which the sender state can exercise influence over the host state. In terms of the theory, the former effect is captured through the shock parameter, in that the deployed troops would be seen as producing the same good through the same instrument as the host state’s troops. The latter affect is captured through the policy efficiency parameters, since the deployed troops serve to reduce the relative efficiency of the host state’s own troops. Interestingly, these have the same effect on the host state’s policies. Whether deployed troops serve to supplement the host state’s forces or to reduce their efficiency, the result should be a reduction in the size of the host state’s armed forces (either a “shock” consisting of an addition to a particular input or something that reduces the efficiency of a particular input leads the host state to reduce its own resources devoted to that input). Deployed troops should also lead to an increase in the resources the recipient state devotes to all other policy inputs.
26
In principle, this could lead to a large number of empirical hypotheses. Two specific ones are relevant here:
Hypothesis 1: As the number of US troops deployed to a state increases, the number of troops in that state’s armed forces will decrease, all else equal.
Hypothesis 2: As the number of US troops deployed to a state increases, the probability that the host state will initiate a militarized dispute increases, all else equal.
These particular empirical hypotheses are the focus of this article for two reasons. First, they are especially interesting in that they address important policy decisions. This is particularly true given the likely motivation for the sending state to deploy its troops. As stated before, at least in the case of the United States, one of the primary aims of deploying troops abroad is to make the host state more secure. If the sender state is seeking to enhance the security of the recipient, it probably does not intend for the recipient to reduce its own troop levels.
27
Also, the review of previous literature on troop deployments shows that states often deploy troops abroad to bring stability to the region they are stationed in and that noncombat troops are often there for defense purposes. If, as the theory predicts, these troops were actually to lead to an increase in dispute initiation by the host state, then this would go against the aim of the sender state.
28
Second, these hypotheses taken together are somewhat counterintuitive. That is, a state initiating a dispute would not be expected to decrease the size of its forces, even if it is able to pass off some of its security concerns to another. Thus, testing these propositions constitutes a particularly tough test for the theoretical argument.
Troop deployments are also likely to affect the behavior of third-party states. If deployments serve to enhance the security of the recipient, presumably they are intended to reduce the propensity of other states to attack the recipient. Focusing the theory on potential attackers, this effect should reduce the efficiency of their military forces. This leads to the expectation that the probability that some other state initiates a dispute against the recipient of deployed troops would decrease. Thus, there is one additional hypothesis:
Hypothesis 3: As the number of US troops deployed to a state increases, the probability that a large-scale militarized dispute is initiated against it decreases.
This hypothesis is specifically about large-scale disputes being initiated against the host state. This is the case because of the fact that troops will often be deployed to states that are already under threat. The argument then is that the presence of the troops should prevent potential attackers from escalating their threats to large-scale conflicts.
The rest of this article is devoted to testing these hypotheses. While the theory and the hypotheses can be generalized to any troop deployment by any state in the international system, this particular article will test the hypotheses using data on US troops deployments from 1950 to 2005. This is not because there is anything particular about US troop deployments that should lead us to expect that these relationships will hold. Rather, the decision to use US troop deployments to test the hypotheses is based on data availability.
Data
To analyze US troop deployments, we use a data set created by Tim Kane for the Heritage Foundation, which includes US troop deployments by state, by year, for all countries in the world from 1950 to 2005. Each observation is a state-year, and the data set gives the actual number of troops stationed in that particular state-year. The data set was created by Kane from disaggregated records from the Department of Defense, collected by the Statistical Information Analysis Division of the Directorate for Information Operations and Reports. 29 The troop count represents the number of “billets” or troop-years for that particular state. In other words, it represents the number of troops at a specific moment in time of that year, not the total number of troops that came in and out of the state in that particular year. For example, 100 soldiers serving one-year tours and 200 soldiers serving half-year tours would both be counted as a troop level of 100. The troop level represents US troops ashore, though Kane admits that for some of the older data it is difficult to disaggregate the ashore and afloat troop levels. The data for the years 1950 and 1951 are estimated, as there are no original records of US troop deployment for these two years (Kane speculates that they were possibly destroyed for security reasons during the Korean War). 30
The analysis also includes capabilities data from the Correlates of War (COW) Project National Material Capabilities data set, version 3.02. 31 In particular, the military personnel variable from this data set can be used to evaluate Hypothesis 1. In the COW data set, military personnel is measured, in thousands, as the number of “troops under the command of the national government, intended for use against foreign adversaries.” 32 This data set also includes variables that can be used as controls, such as total military expenditures.
Also from the COW Project (from the militarized interstate dispute [MID] 3.10 data set) it is possible to derive a dichotomous variable that is coded 1 if the state in question initiated an MID in that particular year and 0 otherwise. 33 This variable is intended to evaluate Hypothesis 2. MIDs (instead of wars) represent conflict initiation for two basic reasons: first, this increases the number of observations in order to provide a more meaningful statistical analysis, and second, this provides a better indicator of how aggressive the host state is, which is the subject of the hypothesis. Using higher levels of conflict, like war, would measure not only the aggressive behavior of the host state but also how the target responds to it (because in order to have a war both sides need to be willing to fight).
The data set also includes a control variable that is coded 1 if the host state was at war with the United States during that particular year and the war was being fought in the host state’s territory, and 0 otherwise. Another variable is coded 1 if the host state was involved in a war at the same time as the United States (regardless of which side it fought on). These variables are derived from the COW Interstate War Participants data set, version 3.0. 34 The first variable will make it possible to discard cases in which the US troops were in the host state as attackers, and therefore would not be helping protect the host state’s interests. In these cases, the host state is not expected to behave in the ways that the theory would predict, as the motive for the US troops being present would be different from the other cases. We use this variable to drop from the analysis all cases in which US troops were present as part of an invading force. The second variable makes it possible to also discard cases in which the United States and the host state were fighting a war together and attacks were jointly launched by both states.
Analysis
Hypothesis 1 states that as the number of US troops deployed to a state increases, the number of military personnel of the host state will decrease. To test this hypothesis, the dependent variable will be the change in the number of troops under the command of the host government (measured in thousands), over the previous five years. The change in military personnel is a better measure of the dependent variable than present value because this measure appears to represent more accurately the expected changes. Once American troops are stationed in a host state, that state cannot immediately alter the size of its military. The government must go through some bureaucratic process to expand or shrink its military (even autocrats face some roadblocks when implementing large-scale policy changes), which can take time. In the case of an expansion, the government must increase its officer corps, train new recruits, purchase equipment, set up accommodations for the troops, and so on. In the case of a decrease of the military, most governments have to wait for the contracts of their troops to run out. It is also plausible that a host state would wait some time to observe the behavior of US troops in its territory (e.g., to make sure they are reliable) before choosing to reduce the size of its own military. Thus, the null hypothesis would be that an increase (decrease) in US troops would lead to no change, or to an increase (decrease), in the level of military personnel of the host state.
The main independent variable will be the natural log of the number of US troops stationed in the host state, lagged by five years. 35 This lag is present because, as explained in the previous paragraph, it should be the amount of troops five years ago that would have an effect on the given year of an observation. Because troop levels in the COW data set are reported in thousands, the Kane 36 troop levels are standardized with the COW data set and reported in thousands. Because the troop deployment data are skewed (with very few observations on the higher end of the distribution), the logged version of it keeps data outliers from having too large an influence on the results. Model 1.1 in Table 1 presents the results.
Dependent Variable: Change in Military Personnel of Host State (in Thousands) over Five Years
Note: Cell entries are coefficient estimates, standard errors are in parentheses.
***Significant at .01 level. **Significant at .05 level.
That same table also presents the results of a regression which includes a variety of control variables that could affect the troop levels of the host states. The first control variable is the level of defense expenditures in the host state, given that the level of defense personnel should be positively related to military expenditures. 37 In particular, the variable measures the change of military expenditures from five years ago, since the dependent variable is the change in troop levels from five years ago. 38
It can also be expected that if a state is involved in an interstate rivalry or MID, that state will feel threatened and will increase its troop levels. 39 Thus, the analysis includes a dummy variable that is coded 1 if the host state was involved in an MID the previous year. 40 It also includes a second variable that measures the number of interstate rivalries that the given state was involved in during that particular year. 41 Model 1.2 presents the results of this analysis that includes the control variables.
Additionally, Table 1 includes a regression using a moving average of the US troop level with five lagged terms. This is simply a different way to go about representing the independent variable, and which is included as a robustness check. The model with the moving average of the logged value of the US troop level is model 1.3 and that same model with control variables is model 1.4.
The results of Table 1 support Hypothesis 1. The coefficient on the variable representing US troop levels five years ago is negative and significant as it is expected to be in all four models. 42 This implies that higher levels of US troops are correlated with decreases in domestic troops in the following five years. Note that the values of the R 2 in these models are very small. This is in accordance with the theory, as it does not propose that US deployments are a major determinant of troop levels, but rather that they have a noticeable effect at the margin. 43
Having found these results, we now test Hypothesis 2, which states that as the number of US troops stationed in the host state increases, so should MID initiation by the host state. Because the dependent variable, whether the host state initiated an MID in that particular year (considering only the original initiators, not joiners), is a dichotomous variable the analysis will use a Bernoulli logistic (logit) model. The main independent variable will again be the natural log of the number of US troops stationed in the host state. In this case, the independent variable will not be lagged, as the host state’s choice to initiate an MID will be more likely driven by US troop levels in the present than in the past. This should be the case because initiating lower-level conflict such as MIDs usually does not require as much planning as a restructuring of the armed forces.
Model 2.1 presents the results of the logit analysis using no control variables. Then, model 2.2 presents the results including control variables that could lead to higher MID initiation. The first control variable is the Composite Index of National Capability (CINC) score of the host state, to represent its capabilities and thus its capacity to initiate an MID in that given year. 44 Given that rivals are also considered more conflict prone than other states, 45,46 the second control variable is the rivalries variable, which measures the number of rivalries the host state was involved in during that year. 47 The results are presented in Table 2.
Dependent Variable: Dispute Initiation
Note: Cell entries are coefficient estimates, standard errors are in parentheses.
***Significant at .01 level.
The results also show support for Hypothesis 2. In the first model, the one with no control variables included, the effect of the number of American troops stationed in the host state is positive, as expected, and it is significant at the .01 level. Once model 2.2 includes the control variables, the coefficient on the troops variable remains significant and in the expected direction, though its coefficient is somewhat smaller. Again, the Pseudo R 2 is small in both models. As explained before, this is consistent with the theory, since US troop levels are not expected to be a perfect predictor for conflict initiation by the host state; rather, they should have an effect at the margins.
Finally, we turn to testing Hypothesis 3, which states that as the number of US troops stationed in the host state increases, the host state should be less likely to be the target of an MID. The dependent variable will be whether the host state was the target of an MID level 4 or 5 (referring to uses of force or war) initiation in that given year (cases in which the host state joined an MID on the side of the original target are not included). The analysis is limited to only the higher-level MIDs (levels 4 and 5) because if US troops are deployed to a particular state it is for a reason, and that reason is likely to be that the host is facing some form of threat. Thus, if US troops are present, there will likely be some level of hostilities. Those low-level hostilities would not necessarily be expected to go away. Rather, the theory predicts that the US troops would prevent the escalation of low-level conflict to more aggressive and conflictual MIDs.
Because the dependent variable is again a dichotomous one, the analysis will once more use a Bernoulli logistic (logit) model. The main independent variable remains the natural log of the number of US troops stationed in the host state. The independent variable will again not be lagged, as it should be the number of US troops stationed in the present that would serve to make the option of MID initiation more inefficient (all cases in which the host state was involved in a war with the United States are dropped).
Model 3.1 presents the results of the logit analysis using no control variables. Model 3.2 presents the results including control variables that could make a state more likely to be the target of an MID initiation. One of these is the CINC score of the host state, as it could be the case that weaker states make more attractive targets (also, it may be the case that major powers, for which CINC score is an adequate proxy, are more conflict prone). 48 Another control variable is the number of interstate rivalries that the state is involved in, given that a rivalry would make a state more conflict prone (as argued in the rivalry literature that was referenced earlier). The third control variable is a dummy variable for whether the state was involved in an MID in the past five years. 49 This is because whether a state was previously involved or targeted in an MID may be a strong indicator of whether it will be involved in an MID in the present. 50 The results are presented in Table 3.
Dependent Variable: Dispute Targeting
Note: Cell entries are coefficient estimates, standard errors are in parentheses.
***Significant at .01 level. *Significant at .10 level.
As Table 3 shows, in the model in which no control variables are used, the US troops variable is actually significant in the opposite direction that the theory would predict. Of course, this model may be reflecting very strong selection effects, as the United States is expected to deploy more troops to countries that are more likely to be the targets of MIDs. Model 3.2 attempts to control for some of the factors that would make a state more likely to be the target of an MID. In model 3.2 (which has a higher pseudo R 2 value), the US troops variable is negative (the expected direction), but is not statistically significant. This may reflect the extent to which the theory can be extended. The discussion section will further discuss these results.
Another option to consider, as an additional robustness check, would be whether there is something about particular states that makes them more likely to either initiate or be the target of MIDs or decrease their domestic troop levels, something that is independent of US troop levels but that may be confounding the results. One way to check for this is to run a fixed-effects (within) regression on the data, which is essentially the same as including dummy variables for each state in the data set. This will control for the effect of the countries on MID initiation and changes in troop level, to observe the effect of US troops net of the state effect. The regressions for MID initiation, targeting, and troop levels, are run again, using the different measures of US troops deployed abroad.
Table 4 shows mixed results when running the fixed effects model. The support for Hypothesis 1 (as levels of US troops increase, domestic troops in the host state decrease) holds after this analysis. Regardless of the measure of the US troops variable, the coefficient on it remains negative and significant, as expected. As in the previous models, the R 2 s remain small, both in the cases of within and between countries.
Robustness Checks
Note: Cell entries are coefficient estimates, standard errors are in parentheses. dv = dependent variable.
***Significant at .01 level.
When it comes to Hypothesis 2 (higher levels of US troops are likely to lead to higher levels of MID initiation by the host state), the results are less clear. In this case, the coefficient on the troops variable actually switches to the opposite direction from what the theory predicts. This result may be due to the fact that there is a very large variation in levels of MID initiation between countries. This seems to imply that an examination of other forms of change-seeking behavior may be needed in order to further test the theory.
Finally, and interestingly, the model that includes fixed effects supports Hypothesis 3 (as levels of US troops increase, host states are less likely to be the targets of MID targeting). As evidenced in the fourth column of Table 4, the coefficient on the US troops variable is negative and significant, as expected. This likely happens because in the previous analysis there was a very strong selection effect, with states that were more likely to be targeted being the ones that received US troop deployments. Controlling for the state effects shows the expected effect of troops on MID targeting.
One more issue to consider is the effect of time on the dependent variables. After all, this is time-series data, and it may be the case that the results are being driven by time. It would not seem implausible to imagine that there would be trends in time when it came to either MID initiation and targeting or domestic troop levels. When dealing with change in host states’ domestic troop levels, the variety of lags take time into account.
We also control for the effects of time in the case of our second hypothesis (as US troop levels increase, so does MID initiation). Model 5.1 first assumes linear time dependence, but because the MID initiation variable is a binary one, the Beck, Katz, and Tucker (BKT) 51 method can be implemented to analyze it. This means that the model will include dummy variables for every year in the data set and drop the constant (though, of course, the coefficients of the BKT dummies are not included in Table 4). As in previous regressions, the main independent variable is the natural log of US troop levels (model 5.2). We then do the same to test our third hypothesis (as US troop levels increase, MID initiation against the host state will decrease) and present the results in models 5.3 and 5.4 in Table 5.
Temporal Effects
Note: Cell entries are coefficient estimates, standard errors are in parentheses. dv = dependent variable.
***Significant at .01 level. **Significant at .05 level.
Table 5 shows that Hypothesis 2 is supported by models 5.1 and 5.2. The US troops variable remains positive and significant in both the cases of assumed linear time dependence and when using the BKT (1998) dummies. As before, in this analysis there appears to be no support for Hypothesis 3. The next section discusses potential explanations.
Discussion
Overall, the results both support the theory and provide some novel insight on the effects, unintended or otherwise, of stationing US troops abroad. Contrasting these effects with the original aims of the United States when it deploys its troops abroad provides insight into the potential gap between the intended and observed effects of foreign deployments.
It is particularly interesting that throughout the analysis there is consistent support for the idea that US troops are correlated with a decrease in the troop level of the host state. This supports Jones and Kane’s 52 idea that US troop presences may lead to decreased military spending by the host state (which they view as a positive, as it would allow the host state to devote more resources to development), yet may contradict some of the original aims of the United States when deploying its troops. As stated before, one of the more common reasons why the United States (or any other major power) deploys its troops abroad is to preserve the security of the host state. Unless these troop reductions are previously agreed upon with the host state, this means that troop deployments may not be serving their purpose as effectively as the sender states would want them to. If the host state reduces its troop levels, then this means that the presence of US (or other major power) troops is not resulting in as large an increase in security (through deterrence) to the host state, as the host state is also weakening itself, thus dampening the effect of the US troops.
Another interesting aspect of this analysis is that there is some support for host states being more likely to initiate disputes as the level of US troops present there increases. This may be an unintended effect of stationing US troops abroad. It could of course be argued that the troops could indeed be placed in the host state to allow it to initiate MIDs, thus making this effect an intended one. It is worth noting, though, that the analysis drops all cases in which the United States and the host state are engaged in a conflict together. This thus makes this possibility much less likely to be the explanation for this effect. Thus, it might be that when the United States deploys its troops abroad in an effort to promote stability in the region, one effect is that host countries initiate more interstate conflicts, which actually leads to instability.
As stated previously in the article, the fact that these two effects happen simultaneously is particularly counterintuitive. Because states tend to arm themselves before a conflict, it is unexpected that the same stimulus would make a state more likely to both decrease its troop levels and initiate armed conflicts. This result is thus particularly strong support for the theory, which does predict this somewhat unexpected behavior from states.
Finally, it appears that the presence of US troops does not deter attacks against the host state. The results suggest that US troop deployments do not make it less likely that the host state will be the target of either war or lower-level uses of force. This might just be a selection problem for which the present analysis does not control. In other words, it might be the case that the United States is more likely to send its troops to states that are more likely to be the target of an MID. There is some evidence that this is the case, as the model that controls for the effect of the individual host states does indeed support the hypothesis.
Another reason why the deterrent effect of troops stationed abroad is not significant may be the change in behavior by the host state. The theory, when it predicts the effect that an external shock will have on the foreign policy of states, holds everything else constant. That is, the theory predicts that the probability of an attack decreases because the presence of US troops decreases the efficiency of an attack for achieving the third parties’ aims, all else being equal. The problem is that in this case all else is not equal. If the presence of US troops leads the host state to decrease the size of its military, to initiate more disputes and to change other aspects of its foreign policy, then these changes might very well increase the probability of the host state being the target of an MID, despite the deterrent effect of having US troops present.
We believe that these results, combined with the theory, lead to a better understanding of the effects, intended and unintended, of troop deployments abroad. While this analysis deals specifically with US troop deployments, the theory predicts that any other consensual troop deployment would have the same sort of effects on the foreign policy of the host state. This study has contributed to an area of foreign policy that is often understudied by political scientists but forms an important part of foreign policy for major powers and the host states (In 2008 alone, the United States spent US$15,899 million on troop deployments, which is a significant portion of the total US$48,923 million that it spent on foreign assistance in the same year). 53
We hope that this study leads to more quantitative research on the effects of troop deployments on the foreign policy of the host state. As stated before, changes in quantity of military personnel and conflict initiation are only two of many policies that the host state may alter when facing the exogenous shock of a deployment. The theory predicts that many other policies would also be affected and further studies are needed to obtain a complete picture of what the true effect of deployments will be.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.