United States military assistance is an important component
of U.S. relationships abroad and U.S. foreign policy. In this analysis, I
examine the effects of U.S. military assistance on political stability within
countries. Understanding the effects of U.S. military assistance can aid policy
makers in allocating funds to best utilize the United States military and its
budget. Additionally, because a fundamental aspect of international relations
is forming partnerships with stable countries in key regions of the world,
exploring the effects of U.S. military assistance on these countries’ stability
can offer insights into optimal U.S. foreign policy.
Much of the literature on U.S. military assistance isolates
specific types of security assistance, such as counterterrorism or police force
training. For instance, Dube and Naidu examine how paramilitary and guerrilla
groups respond to U.S. military aid in Colombia. Additionally, a RAND report observes the effects of security cooperation, but still focuses on authoritarian
regimes and states with pre-existing fragility. My analysis looks at overall
effects of military assistance, providing a general foundation for more
specific analyses.
In this analysis, I use cross-sectional time-series model,
focusing on countries receiving U.S. military assistance from 2003-2012.[i]
I chose a cross-sectional time-series analysis over a quasi-experimental design
because finding two similar countries, one of which received military assistance,
to serve as treatment and control groups is extremely rare.[ii]
In the model, I also include three control variables that might affect the
relationship between military assistance and stability. Rule of law, GDP, and
level of democracy might relate to both the explanatory and outcome variables;
an increase in any of these variables might decrease the need for military
assistance and increase stability. While I am concerned about selection bias,
or whether the sample is representative of the population, with these cases, the
lack of an available comparison case limits my ability to overcome this issue.
I first performed tests to determine whether the variables
were normally distributed and if the model’s effects were consistent.[iii]
Because the effects of military assistance are not instantaneous, I lagged
political stability by two years. I regressed the lagged political stability variable
on the independent variables, using a fixed effects estimator. Figure A is a
visual depiction of the effect of U.S. military assistance on political
stability. The graph shows a decline in political stability with increased
military assistance; however, this negative relationship may be exaggerated by the
extreme instability of a few cases receiving high levels of military assistance
(seen in the lower right-hand side of the graph).
In contrast, the estimates of the variables depict a
slightly positive relationship between U.S. military assistance and political
stability (see Appendix A for a table of regression estimates). In a bivariate
model (Model 1, Appendix A), a one-unit increase in U.S. military assistance is
associated with a 0.055-unit increase of political stability within countries,
on average. This effect is statistically significant with a p-value of less than
0.05. This value indicates that there is less than a 5% chance that the observed
relationship is due to chance. The effect of U.S. military assistance is
substantively similar across the models, but only retains its significance when
GDP is not included in the model. In the multivariate model (Model 4, Appendix
A), both level of democracy and rule of law have a positive effect on political
stability within countries, with a one-unit increase corresponding with a
0.007-unit increase and 0.691-unit increase respectively, on average and
holding all else constant. These effects are not statistically significant. GDP
also has a positive effect; a one-unit increase in GDP is associated with a
0.215-unit increase in political stability within countries, on average and
holding all else constant. This effect is statistically significant at the 0.1%
level—the only statistically significant explanatory variable in the full
model.
An examination of the changes in political stability in
Serbia and in Montenegro from 2006-2012 supports these results. Montenegro
gained its independence from Serbia in 2006, so it is reasonable to assume that
these countries serve as a good comparison. After independence, Montenegro’s
level of political stability increased sharply, then declined in 2009 and
remained constant (Figure B). Montenegro did not receive military assistance
from the United States. In contrast, Serbia, which began receiving U.S.
military assistance in 2007, experienced an upward trend in political stability
from 2007-2012. As the amount of military assistance to Serbia increased,
political stability in Serbia rose, as well (Figures C and D). 
Thus, both the cross-sectional time-series analysis and the
case study of Montenegro and Serbia suggest that increased levels of U.S.
military assistance are associated with an increase in political stability. However,
I remain cautious about confirming causality; the military assistance variable
lacked statistical significance in the multivariate model, and the level of
increase was not substantively large. Despite these limitations, the analysis
indicates that the use of U.S. military assistance may increase political
stability, potentially contributing to U.S. relationships abroad and foreign
policy.
[i]
Data availability limited the observations to a ten-year span.
[ii]
Discontinuity design is in appropriate because there are no clear criteria for
entry. Difference-in-differences would require similar treatment and control
groups, but there are few, if any, scenarios in which one of two nearly
identical countries received military assistance while the other did not.
Finally, I was unable to find an instrument for instrumental variable design
(2SLS) that only influenced the dependent variable through the independent
variable.
[iii]
I found that the military assistance, rule of law, and GDP variables were
positively skewed and logged these variables to correct skewness. The Hausman
test yielded a p-value of 0.000, which indicated that the difference in
coefficients is not systematic and the fixed effects estimator is preferred. I
also tested for multicollinearity, but found no strong collinearity among the
variables.
No comments:
Post a Comment