As you probably already know, today is the 100th anniversary of the assassination of Archduke Franz Ferdinand, which would set off the July Crisis and then World War I. For the next few months, the media will undoubtedly bombard us with World War I history and attempt to teach us something based on it.
This is not a good idea. World War I was exceptional. To make generalizations based on it would be like making generalizations about basketball players based on LeBron James. LeBron is so many standard deviations away from the norm that anything you learn from him hardly carries over to basketball players in general. At best, it carries over to maybe one player per generation. The same is true about World War I. It was so many standard deviations away from the norm that anything you learn from it hardly carries over to wars in general. At best, it carries over to maybe one war per generation.
Anyway, the impetus for this post was a piece on Saturday’s edition of CBS’s This Morning, where a guest said something to the effect of “the lesson of World War I is that sometimes it is difficult to stop fighting once you’ve started.” (Apologies I don’t have the exact quote. They didn’t put the piece on their website, but I will update this post if they do.) I suppose this is true in the strictest sense–sometimes countries fight for a very long time. However, such long wars are rare events. Most armed conflicts between countries are very, very short.
To illustrate this, I did some basic analysis of militarized interstate disputes (MIDs)–armed conflicts that may or may not escalate to full scale war from 1816 to 2010. If we are interested in whether fighting really begets further fighting, this is the dataset that we are interested in analyzing since it represents all instances in which states started butting heads, not just the most salient ones.
So what do we find in the data? Well, the dataset includes a measure of length of conflicts. If fighting begets further fighting in general, we would expect to see very few instances of short conflicts and a much larger distribution of longer conflicts. Yet, looking at a histogram of conflicts by length, we find the exact opposite:
(I used the maximum length measure in the MIDs dataset to create this histogram. Because the length of a conflict can vary depending on who you ask, the MIDs dataset includes a minimum and maximum duration measure. By using the maximum, I have stacked the deck to make it appear that conflicts last longer.)
Each bin in the histogram represents 50 days. A majority of all MIDs fit into that very first bin. More than 90% fall into the first seven bins, or roughly a year in time. Conflicts as long as World War I are barely a blip in the radar. Thus, fighting does not appear to beget further fighting. If anything, it appears to do just the opposite.
One potential confound here is that these conflicts are ending because one side is being completely destroyed militarily. In turn, fighting begets further fighting but stops rather quickly because one side cannot fight any longer. But this is not true. Less than 5% of MIDs result in more than 1000 causualties and even fewer destroy enough resources to prohibit one side from continuing to fight.
So why doesn’t fighting beget further fighting? The simple answer is that war is a learning process. Countries begin fighting when they fail to reach a negotiated settlement, often because one side overestimates its ability to win a war or underestimates how willing to fight the opposing side is. War thus acts as a mechanism for information transmission–the more countries fight, the more they learn about each other, and the easier it is to reach a settlement and avoid further costs of conflict. As a result, we should expect war to beget less war, not more. And the histogram shows that this is true in practice.
Do not take this post to mean that World War I was unimportant. Although it was exceptional, it also represents a disproportionately large percentage of mankind’s casualties in war. It was brutal. It was devastating. It was ugly. But for all those reasons, it was not normal. Consequently, we should not be generalizing based on it.