Estimating the alliance effect: a synthetic control approach

ABSTRACT

In a time of rising threat, recurring discussions about burden sharing within NATO and twenty years after the start of the “war on terrorism,” I explore a novel idea in the field of alliance and defense spending: the effect of alliance membership on defense spending in response to a threat. Instead of estimating the determinants of defense spending or burden sharing among members, this paper focuses on two types of states (small and large) and how alliance membership shapes their response to threat. Using the synthetic control method, I create a comparison unit for each type consisting of a weighted average of non-NATO, European countries. 2001 is considered as the starting year of widespread threat, caused by a sudden increase in transnational terrorism. This way, I can estimate the “alliance effect.” I find that both types of states have stronger (positive) response to threat as NATO members, compared to if they would not have been part of the alliance.

KEYWORDS:

• Defense spending
• NATO
• threat
• alliance

Acknowledgments

I would like to thank David Levine, Thomas Crossley, Andrea Mattozi and Aschkan Mery for their helpful comments and suggestions, as well as seminar and workshop participants at the European University Institute.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed here

A.1. Robustness tests A.1.1. Inclusion of predictor variables

The inclusion of predictor variables is important. For the main analysis I have chosen to adhere to the general consensus that less is more, and only included the average defense spending trend as well as the last pre-treatment annual value of defense spending. Below I show the results for the analysis with inclusion of additional variables - GDP, population growth, trade and a measure for war.

The first change is with respect to the weights placed on the donor pool countries to create the synthetic control. Whereas the weights for the small type do not change much, the synthetic control for the large type is now made up of Austria, Finland and Sweden (see table 4). The values of the predictor variables are shown in table 5.

The effect for the small type is almost identical as to the main analysis (see figure 2). The RMSPE is 26.971, slightly larger than in the baseline estimation. For the large type, however, the positive alliance effect is much smaller than previously found. Notably, the RMSPE is smaller (though marginally) than in the baseline estimation - it is 15.53. This result suggests that it might be likely that the alliance effect for the large type is very small.

A.1.2. In-time placebo

One other test for the robustness of our results is to change the time of the intervention to a moment before 2001. Finding a large effect would decrease our confidence that the effect found are indeed due to alliance membership and not due to the lack of predictive power. For more information on this placebo test, see Abadie et al. (2015) and Abadie (2021).

The results can be seen in figure 3 below. Naturally, as the pre-treatment period is shortened the number of data points used to “fit” the synthetic control decrease, and thereby the pre-treatment fit is less good. However, the results for the both types are positive. The divergence of the trends takes place at 2001, which is what the main estimation showed.

A.1.3. In-space placebo

Finally, I run the in-space placebo test. The idea behind this test is to check whether, if one was to assign treatment “randomly” to each of the units in our sample and estimate the alliance effect, one would find a similar or larger effect than found in the main analysis. For example: assume Austria is a NATO country and the other countries in the sample (Switzerland, Sweden, Ireland and Finland and our small or large type, respectively) are not. A synthetic control unit for Austria is created following the exact same method as before. Next, the RMSPE of the post-intervention period and the pre-intervention period RMSPE are calculated, and the first is divided by the latter.

In the case of the large NATO type, the post-intervention gap is about seven times as large as the pre-intervention gap. If one was to pick a country at random and assign the treatment to it, the likelihood of obtaining a ratio as high as the above would be¹. For the small time, the post-intervention gap is almost seven times as large as the pre-intervention gap and the odds are ¹. There is one donor pool unit (Finland) that has a larger post-intervention gap, but only by 0.03 points (see figure 4).

Figure 4. In-space placebo estimation.

The two graphs show the post-intervention RMSPE/pre-intervention RMSPE ratios for the small and large type and countries from the donor pools. In the left graph we see that Finland has an almost equally large post- intervention gap as the treated unit. In the graph on the right we see clearly that the treated unit has the largest post-intervention gap.

Notes

1. According to Becker and Malesky (2017): “‘Atlanticism’ shapes decisions about resource allocation through three mechanisms: a shared normative understanding of a Western-led international order; a belief in the importance of the United States in European security; and a preference for NATO as a platform for coordinating force planning and operational deployment.” (p.163). The authors note furthermore that “‘Atlanticist’ strategic cultures allocate a greater share of their defense resources to Alliance priorities than those exhibiting ”Europeanist” strategic cultures” (p.163).

2. After the attacks, the threat of terrorism was acutely felt across Europe, affecting both NATO and non-NATO countries (Europol 2021).

3. As noted in Oma (2012) there is a need to understand the strategic interest with respect to decision making dynamics of small states specifically (p. 570).

4. Data on defense spending per capita in current USD is obtained from the Stockholm International Peace Research Institute (SIPRI).

5. During the time period considered the United States pressured NATO members to increase their spending. It is tempting that this pressure is behind the increase in defense spending of the large and small type. However, Becker et al. (2019) has shown that this “transatlantic shakedown” has no effect on defense spending, and this therefore cannot explain our finding.

6. The weaker response of non-allied states to threat is interesting, but even more curious is that these states (Ireland and Sweden) contributed to the NATO mission proactively. Of course, these states are NATO-partners, but their participation was completely voluntary. This fact suggests that the idea that NATO partners have the best of both worlds does not hold up: it is not possible for these states to enjoy the spill-overs of the collective defense good of the alliance (due to their geographical proximity it is likely that any threat they face is also confronted by NATO) while not having to contribute to the cost of the good. If anything, this suggests that the tendency to be considered a reliable partner and ally extends beyond alliance membership and into the realm of shared ideology and views on world order, relating to three mechanisms through which Atlanticism affects defense spending decisions (Becker and Malesky 2017). This might be an interesting topic for future research.