The world “Ally” carries a positive connotation that blinds our rational evaluation of its actual value. We presume that allies must share the same interests and goals as us and will stand by our side in times of need. Some think of allies as friends.

Now, what if I told you that your allies will abandon you in a war about 75% of the time?

There is an ancient Greek Proverb: After the war is over, make alliances

There is a good deal of truth to this. Allies are often as dangerous as enemies. No two states have the exact same goal, so caution is needed.

Sabrosky studied alliances in conflicts between 1816 and 1965 and found that alliances were not honored in conflicts 73% of the time (1980). In the remaining quarter of the cases, alliances are honored partially or fully. More often than not, an ally will defect on the terms of its treaty.

Furthermore, Alliances are short-term agreements. Morrow’s 1991 study found that the average Asymmetrical Alliance lasts 15.7 years and the average Symmetrical Alliance lasts 12.2 years. Client States make more durable and reliable allies than equal partners. States which are equal in power are more likely to pursue their own interests without consulting their ally.

Do allies even contribute much in war, given how much hassle they are?

From Bruce Bueno de Mesquito’s Dataset of 598 armed conflicts in Europe between 1816 and 1984
-Unilateral attack on a lone state: Attacker wins 74% of the time (328)
-Alliance attacks a lone state: Attackers win 80% of the time (138)
-Unilateral attack on an alliance: Attacker win 40% of the time (60)
-Alliance attacks an alliance: Attackers win 44% of the time (72)

Alliances have the greatest utility as a defensive agreement. They do not add much value in offensive wars. Given the probability of allies abandoning allies, it is more sensible and cheaper to separate your enemy from his allies than to build up your own alliance.

There is another fact that is important. Offensive action outperforms defensive action. Those who initiate the wars have a very significant advantage which exceeds their relative advantage in strength.

Correlates of War data (Small and Singer) shows that attacking state wins 68% of the time. Some of this is due to self-selection (after all, you attack if you believe you will win). 59% of wars are intitiated by the stronger party. Even weaker states gain an advantage if they attack first. And alliances contribute little to victory.

Here’s another fact. Democratic allies are not any more reliable than autocratic allies. In fact, Democracies are potentially less reliable and shorter in duration. There is a higher turnover of leadership and there are greater difficulties with special interest groups whose desires conflict with allied states.

The entire Correlates of War project produced a number of interesting empirical studies.

So what’s the point of alliances?
War is about politics and Alliances are based on common political interests and goals against a common enemy. Many alliances are based on the principle that the enemy of my enemy is my friend. Sometimes the enemy of your enemy is also your enemy, so you must be careful trusting such an ally.

No two individuals, much less states, can have the exact preferences. Everyone is in a different situation and will disagree on how to allocate scarce resources. There is always friction and disagreement between allies. All that can be said is that allies are at least temporarily closer to each other than to the common enemy. The alliance terminates when they politically drift apart.

Interests and power in international relations is often intransitive. So preferences A>B>C>A, make an infinite preference loop. This makes concrete agreements and cooperation on a single preference difficult if not impossible.

There is an added complication that alliances are webs of nation-states. Bilateral alliances are difficult enough, multi-lateral alliances have little cohesion.

Multi-sided Conflicts are more complicated than two-sided conflicts. There are diverging political goals where allies are unable to agree or coordinate on a common objective. Indeed, they stab each other in the back, as is the case of the First and Second Balkan Wars.

The relationship between multi-sided politics and two-sided war thought is nicely illustrated by an episode from the First Balkan War (1912-1913). Crown Prince Constantine, who commanded the Greek army in the field, wanted to use his forces to help the much-larger Bulgarian army defeat the main Turkish army. Bulgaria and Greece, after all, were allies, and shared the common goal of expelling the Turks from the Balkan peninsula. King George of Greece, however, had a different idea. While just as eager as anyone else to gain territory at the expense of the Turks, he was also worried about the territorial ambitions of his allies. The king therefore ordered his son to let the Bulgarians worry about the main Turkish army while the Greek field army took the shortest possible route to the choicest bit of real estate at issue, the city of Salonika.

Constantine made all sorts of noise about the necessity of keeping faith with allies and how the main Turkish army was the center of gravity of the campaign. In the end, however, he obeyed his orders. As a result, the relatively small detachments that the field armies of Bulgaria and Serbia (which was also taking part in the war against Turkey) sent to Salonika found themselves marching into in a city that was already occupied by a much larger Greek force.

As might be imagined, the Bulgarians, who were fighting alone against the main Turkish army, were not happy about this turn of events. They were even less happy when they discovered that the Serbian army had done something similar in central Macedonia… Indeed, the Bulgarians, who had done the lion’s share of the work of driving the Turks out of the Balkans, and who had consequently taken the lion’s share of the casualties, were greatly annoyed. Thus, when it became clear that neither the Greeks nor the Serbs were willing to give up the territory they had snatched while Bulgaria had been distracted, Bulgaria declared war on its former allies.

Every state in the Balkan alliance acted in its own interest. They wanted to reduce their own casualties and maximize the valuable land captured. So, of course, states will gladly let their allies bleed while they take the undefended spoils.

Defensive value
The greatest value of an alliance is deterrence. Attacking an alliance is difficult, whether done alone or with your own alliance. The majority of attacks on allies fail.

The greatest reason to create an alliance is during peace time to deter attacks on your country. There have to be regular shows of support to demonstrate that the alliance is “reliable” if a war occurs. When a state is actually at war, alliances have fewer benefits (and they usually fall apart). In a way, an alliance is a bluff more than an actual threat.

Alliances rarely cooperate in wartime. They simply force a common enemy to fight on two fronts. A small number of very close alliances fight almost like confederacies, where the governments and militaries merged together for the duration of the war. This includes the Anglo-American alliance of World War II. These are more durable and beneficial than ordinary alliances.

Client states last longer and are more reliable than any other form of alliance. Client states sacrifice some of their autonomy for additional security by allying with a powerful state. The powerful state does not want to get drawn into local disputes, so it oversees the foreign policy of its clients. The clients tolerate this because they would be too weak to survive independently anyway.

Powerful states seek alliances even if this causes greater risk to their security. Why? Part of the reason is economics and politics, moreso than military security. A client state is a reliable trade partner. Powerful states use clients to secure overseas economic interests which keep their domestic economies wealthy.

Can we test the reliability of alliances? To a certain extent. Alliances are built with the intent of maximizing expected utility.

For instance, if two allies discuss which military strategy to pursue, they do not decide which is militarily better, but which produces the best political result for them. So if faced with a choice between three strategies, they will select the one with the highest payoff for their own state. If they select different strategies, the alliance fails.

The Signorino-Ritter score measures the distance between alliances. (From “Tau-b or Not Tau-b: Measuring Alliance Portfolio Similarity” 1997)

Consider alliances as a network of nations. Each nation is a node and the type of alliance determines the power of the connector. No alliance is 0. An Entente is 1. A Neutrality Pact is 2. An Alliance is 3. You can make this more complicated by adding free trade agreements, military technology sharing agreements, and other forms of cooperation.

The tools of network analysis measure clusters. Alliances may be discrete or they may overlap. Some allies are more distant or have conflicting interests between two rival states. We cannot just measure the bilateral agreements between states A and B. This is a deceptive measurement. You have to measure States A and B’s relationship with the entire web of nation-states in the system in order to understand the real relationship between A and B. Frequently A and B will sign an alliance but have more dissimilar interests than similar ones. It turns out they were only allies on paper and not in reality.

Buena de Mesquita, Signorino and Ritter figured out a way to give a rough measurement of alliance reliability. In short, they joined game theory’s use of expected utility and mapped it onto a network of states. This measures the political distance between states.
So here is the S-score formual,

Without going through the mathematical formulas (which I’m finding tedious to write on a blog), this is what this means. Assume there are N-number of states in the system. Think of them as a interconnecting web between an alphabet of 26 states. You calculate the political distance of State A to state i (each of the 26 states in the system). Then you calculate the distance of State B to the same state i. The States C and D. You do this for all N states to measure the distance between each and state i. You add up the ordinal values (0, 1, 2, 3) so the bilateral distances should each be between 0-3. Now you can graph out a network on paper or just study it statistically.

You follow a process (if you’re interested, read S&R’s paper) to create an index where all the paired states are ranked between -1 and 1, where -1 means their interests are completly dissimilar and 1 means their interests are the same. This is the S-Score of an alliance within a web of states. How much in common does each state really have?

States A and B may be allied with each other, but they may be part of different state clusters. State A is allied with or close to States C, D, and E. State B is allied with or close to States F, G, H, and L. You can visualize A and B being parts of two rival clusters and they serve as a linking bridge between the two. If States L and H go to war with States C and D, what does this do to the alliance between A and B? If you ranked them by S-Score, you realize that A and B had greater dissimilar than similar interests and in all probability the alliance will fail. A and B’s S-Score may have been -0.75 on the index. Their stronger ties with other states outweigh their bilateral alliance.

If you want, you can add more weight to the S-Score by factoring in military or economic strength. This differentiates powerful from weak allies. It also gives you a headache.

You will quickly find that most allies are unreliable. States may be militarily incapable or lack the political will to help you in wartime. All states are interconnected with a large number of other states which means their political interests will diverge from yours. In most cases, allies gain greater utility by defecting from an alliance with you than shedding blood with you on a battlefield.

What a state should do is create a portfolio to balance risk of defection. For starters, get at least 4 major allies, because 3 of them will probably defect in a crisis. Nation-states should carefully select allies based on common political and economic causes. If allies are economic competitors (such as the US and EU), then they are more likely to defect because of conflicting interests. If they share political interests (US and Australia) the alliance will be more durable.

The S-Scores for NATO were relatively good during the Cold War with the USSR around. Without the USSR, the S-Score of NATO has been declining as Europeans diversified their interests and politically drifted away from the US. Meanwhile, the S-Scores of the US and Asian allies are growing in value as their interests converge.

Allies are not friends and they most serve to prevent wars, not fight in wars.