The Decline of Conventional War: 1800 - 2011

Whither War

“Conflicts of interest are inevitable and continue to persist within the developed world. But the notion that war should be used to resolve them has increasingly been discredited and abandoned there. War is apparently becoming obsolete, at least in the developed world:…” John Mueller (1990)

Whither War

“If the parameters that govern the mechanism by which wars escalate hasn’t (sic) changed—and there’s no evidence to indicate that they have—it’s not at all unlikely that another war that would surpass the two World Wars in lethality will happen in your lifetime. And if it is bigger than the two World Wars, it could easily be a lot bigger.” (Braumoeller 2019)

Mooooo

The Correlates of War (COW), Inter State database covering 1816 - 2007 has been the main workhorse in the international relations literature

COW may be useful for some purposes but it manifestly unfit for addressing the decline of war question:

Mooooo

1. For the 19th and early 20th century CoW only includes wars between belligerents that are all recognized as legitimate States by England or France - so there were no wars in, e.g., Sub-Saharan Africa or South Asia because these were view as anarchy zones that were there for the taking.

Mooooo

2. It starts after the massively violent Napoleonic Wars and doesn’t even discover a war until 1823.

3. Coding of death counts is scatter shot

The New Project Mars Dataset Solves the COW Problem

The Mars database was created by Jason Lyall of Darthmouth University.

1. It is not compromised by a colonial perspective - it covers all conventional wars 1800 - 2011.

The New Project Mars Dataset Solves the COW Problem

2. It begins in 1800 - earlier would be better but still this is an improvement

3. There are low and high estimates of Killed in Action (KIA) and good documentation of coding decisions.

And What a Difference Mars Makes

The next table summarizes the Mars data and the one after that summarizes the CoW data

Focus on the rightmost columns

For Mars the 19th century has the highest rate of war arrivals

For COW the 19th century has the lowest rate of war arrivals.

And What a Difference Mars Makes

Versus this Table for COW

The Peaceful 19th Century?

The two tables also show that:

For Mars KIA per 100,000 of world population is more than twice as high in the 19th century as it was after 1950.

For COW in population adjusted terms it’s a dead heat between the two periods.

Focus on 1950 (the year the Korean War started)

In the paper we cite three quantitative studies that all point to 1950 as a year when decline of war may have started although all relied on COW or COW-related data.

We also survey work of Cold War historians and political scientists who make a qualitative case for the Korean War as a turning point after which the Great Powers settled into vigorous, and often violent, but restrained competition.

Focus on 1950

The picture on the next slide shows a bunch of forward means

For each year, t, the point above t shows the mean number of KIA for year t through 2011.

The upper picture is for the low estimate of KIA and the lower picture is for the high estimate of KIA.

Focus on 1950

Why not just Stop Right Here?

Obviously, the trend is down so you may wonder why there are more slides?

The answer is that some influential critics of the decline-of-war thesis have pointed out that the distribution of war sizes has a fat tail so it’s possible that the underlying distribution hasn’t changed but we’ve just been lucky in recent decades to avoid drawing really huge wars.

An Overblown Critique

The Big Extrapolation

The belief of no change in the distribution of war sizes since World War 2 requires super human confidence in the extrapolation of curves far beyond all post-WW2 data.

Multiple fits to the data are equally plausible but, once extrapolated, have very different implications for the probabilities of really huge wars.

Some Hypothesis Tests

The following graph computes, for each KIA count on the X axis, the pre-1950 arrival rate for wars with at least that KIA count and then graphs:

1. The predicted number of post-1950 wars in this size range if wars arrive at the pre-1950 rate during the post-1950 period.

2. The actual number of post-1950 wars in this size range

Some Hypothesis Tests

3. The bottom of a 95% confidence interval for the predicted number of post-1950 wars in this size range assuming a poisson arrival process with rate equal to the pre-1950 rate

4. The bottom of an 80% confidence interval for the predicted number of post-1950 wars in this size range assuming a poisson arrival process with rate equal to the pre-1950 rate.

Some Hypothesis Tests

Some Hypothesis Tests

There are always fewer actual wars than predicted

We get statistically significant rejections of the hypothesis that the pre-1950 rate continues to govern war arrivals in the chosen range up until the bottom of the range rises a bit over the median war size (4,200 for the low estimates and 7,300 for the high estimates)

Some Hypothesis Tests

And here’s the same picture but using KIA counts per 100,000 of world population.

For orientation, a contemporary KIA count of 1 per 100,000 translates into nearly 80,000 KIA with a current world population near 8 billion.

Some Hypothesis Tests

These Results are the Best we can Do with these Methods

If we set a high minimum war size then our sample of wars above that size will be small - the higher the threshold the smaller the sample.

This means that the post-1950 true arrival rate of wars within a high range can be substantial slower than the pre-1950 arrival rate for wars in this size range but we are unlikely to detect this change with statistical significance.

A Bayesian Approach

Finally please have a look at these two online apps:the raw counts app and the population adjusted app.

These allow you to choose a prior probability distribution over the post-1950 arrival rate within a size range you also choose.

The apps then combine the prior with the post-1950 data and displays the posterior along with a marker for the pre-1950 arrival rate within your chosen range so you can gauge at a glance the probability of post-1950 slowdown.

A Bayesian Approach

The posterior always has a hump to the left of the pre-1950 rate, even if we start by a very flat prior and/or a prior with a mean above the pre-1950 rate.

In the raw counts app, the evidence is quite strong for ranges that begin below the median war size and grows progressively weaker as the bottoms of the ranges rises.

The evidence for slowdown in overwhelming in the population adjusted app.

Last Slide

This was meant as a highlight real - please read the paper.

But hopefully this was enough to convince you that there really was a post-1950 decline in conventional war