There is no good dataset for all wars throughout the world since the start of recorded history, and we wouldn't know how to interpret it if there were. Societies have undergone such radical and uneven changes over the centuries that a single death toll for the entire world would sum over too many different kinds of societies. But the political scientist Jack Levy has assembled a dataset that gives us a clear view of the trajectory of war in a particularly important slice of space and time.

Richardson reached two broad conclusions about the statistics of war: their timing is random, and their magnitudes are distributed according to a power law. But he was unable to say much about how the two key parameters -- the probability of wars, and the amount of damage they cause -- change over time. His suggestion that wars were becoming less frequent but more lethal was restricted to the interval between 1820 and 1950 and limited by the spotty list of wars in his dataset. How much more do we know about the long-term trajectory of war today?

The countries that Levy focused on are the ones that belong to the great power system -- the handful of states in a given epoch that can throw their weight around the world. Levy found that at any time a small number of eighthundred-pound gorillas are responsible for a majority of the mayhem. The great powers participated in about 70 percent of all the wars that Wright included in his half-millennium database for the entire world, and four of them have the dubious honor of having participated in at least a fifth of all European wars. (This remains true today: France, the U. K., the United States, and the USSR/Russia have been involved in more international conflicts since World War II than any other countries.) Countries that slip in or out of the great power league fight far more wars when they are in than when they are out. One more advantage of focusing on great powers is that with footprints that large, it's unlikely that any war they fought would have been missed by the scribblers of the day.

The time span is the era that began in the late 1400s, when gunpowder, ocean navigation, and the printing press are said to have inaugurated the modern age (using one of the many definitions of the word modern). That is also the time at which sovereign states began to emerge from the medieval quilt of baronies and duchies.

As we might predict from the lopsided power-law distribution of war magnitudes, the wars among great powers (especially the wars that embroiled several great powers at a time) account for a substantial proportion of all recorded war deaths. According to the African proverb (like most African proverbs, attributed to many different tribes), when elephants fight, it is the grass that suffers. And these elephants have a habit of getting into fights with one another because they are not leashed by some larger suzerain but constantly eye each other in a state of nervous Hobbesian anarchy.

Levy set out technical criteria for being a great power and listed the countries that met them between 1495 and 1975. Most of them are large European states: France and England/Great Britain/U. K. for the entire period; the entities ruled by the Habsburg dynasty through 1918; Spain until 1808; the Netherlands and Sweden in the 17th and early 18th centuries; Russia/USSR from 1721 on; Prussia/ Germany from 1740 on; and Italy from 1861 to 1943. But the system also includes a few powers outside Europe: the Ottoman Empire until 1699; the United States from 1898 on; Japan from 1905 to 1945; and China from 1949. Levy assembled a dataset of wars that had at least a thousand battle deaths a year (a conventional cutoff for a "war" in many datasets, such as the Correlates of War Project), that had a great power on at least one side, and that had a state on the other side. He excluded colonial wars and civil wars unless a great power was butting into a civil war on the side of the insurgency, which would mean that the war had pitted a great power against a foreign government. Using the Correlates of War Dataset, and in consultation with Levy, I have extended his data through the quarter-century ending in 2000.

One indication of the impact of war in different eras is the percentage of time that people had to endure wars between great powers, with their disruptions, sacrifices, and changes in priorities. Figure 5-12 shows the percentage of years in each quarter-century that saw the great powers of the day at war. In two of the early quarter-centuries (1550-75 and 1625-50), the line bumps up against the ceiling: great powers fought each other in all 25 of the 25 years. These periods were saturated with the horrendous European Wars of Religion, including the First Huguenot War and the Thirty Years' War. From there the trend is unmistakably downward. Great powers fought each other for less of the time as the centuries proceeded, though with a few partial reversals, including the quarters with the French Revolutionary and Napoleonic Wars and with the two world wars. At the toe of the graph on the right one can see the first signs of the Long Peace. The quarter-century from 1950 to 1975 had one war between the great powers (the Korean War, from 1950 to 1953, with the United States and China on opposite sides), and there has not been once since.

Let's start with the clashes of the titans -- the wars with at least one great power on each side. Among them are what Levy called "general wars" but which could also be called "world wars," at least in the sense that World War I deserves that name -- not that the fighting spanned the globe, but that it embroiled most of the world's great powers. These include the Thirty Years' War (1618-48; six of the seven great powers), the Dutch War of Louis XIV (1672-78; six of seven), the War of the League of Augsburg (1688-97; five of seven), the War of the Spanish Succession (1701-13; five of six), the War of the Austrian Succession (1739-48; six of six), the Seven Years' War (1755-63; six of six), and the French Revolutionary and Napoleonic Wars (1792-1815; six of six), together with the two world wars. There are more than fifty other wars in which two or more great powers faced off.

FIGURE 5-12: Percentage of years in which the great powers fought one another, 1500-2000

Now let's zoom out and look at a wider view of war: the hundred-plus wars with a great power on one side and any country whatsoever, great or not, on the other. With this larger dataset we can unpack the years-at-war measure from the previous graph into two dimensions. The first is frequency. Figure 5-13 plots how many wars were fought in each quarter-century. Once again we see a decline over the five centuries: the great powers have become less and less likely to fall into wars. During the last quarter of the 20th, only four wars met Levy's criteria: the two wars between China and Vietnam (1979 and 1987), the UNSANCTIONED war to reverse Iraq's invasion of Kuwait (1991), and NATO's bombing of Yugoslavia to halt its displacement of ethnic Albanians in Kosovo (1999).

Source: Graph adapted from Levy & Thompson, 2011. Data are aggregated over 25-year periods.

FIGURE 5-13: Frequency of wars involving the great powers, 1500-2000

The second dimension is duration. Figure 5-14 shows how long, on average, these wars dragged on. Once again the trend is downward, though with a spike around the middle of the 17th century. This is not a simpleminded consequence of counting the Thirty Years' War as lasting exactly thirty years; following the practice of other historians, Levy divided it into four more circumscribed wars. Even after that slicing, the Wars of Religion in that era were brutally long. But from then on the great powers sought to end their wars soon after beginning them, culminating in the last quarter of the 20th century, when the four wars involving great powers lasted an average of 97 days.

Sources: Graph from Levy, 1983, except the last point, which is based on the Correlates of War InterState War Dataset, 1816-1997, Sarkees, 2000, and, for 1997-99, the PRIO Battle Deaths Dataset 1946-2008, Lacina & Gleditsch, 2005. Data are aggregated over 25-year periods.

Sources: Graph from Levy, 1983, except the last point, which is based on the Correlates of War InterState War Dataset, 1816-1997, Sarkees, 2000, and, for 1997-99, the PRIO Battle Deaths Dataset 1946-2008, Lacina & Gleditsch, 2005. Data are aggregated over 25-year periods.

What about destructiveness? Figure 5-15 plots the log of the number of battle deaths in the wars fought by at least one great power. The loss of life rises from 1500 through the beginning of the 19th century, bounces downward in the rest of that century, resumes its climb through the two world wars, and then plunges precipitously during the second half of the 20th century. One gets an impression that over most of the half-millennium, the wars that did take place were getting more destructive, presumably because of advances in military technology and organization. If so, the crossing trends -- fewer wars, but more destructive wars -- would be consistent with Richardson's conjecture, though stretched out over a fivefold greater time span.

FIGURE 5-14: Duration of wars involving the great powers, 1500-2000

We can't prove that this is what we're seeing, because figure 5-15 folds together the frequency of wars and their magnitudes, but Levy suggests that pure destructiveness can be separated out in a measure he calls "concentration," namely the damage a conflict causes per nation per year of war. Figure 5-16 plots this measure. In this graph the steady increase in the deadliness of great power wars through World War II is more apparent, because it is not hidden by the paucity of those wars in the later 19th century. What is striking about the latter half of the 20th century is the sudden reversal of the crisscrossing trends of the 450 years preceding it. The late 20th century was unique in seeing declines both in the number of great power wars and in the killing power of each one -- a pair of downslopes that captures the war-aversion of the Long Peace. Before we turn from statistics to narratives in order to understand the events behind these trends, let's be sure they can be seen in a wider view of the trajectory of war.

Sources: Graph from Levy, 1983, except the last point, which is based on the Correlates of War InterState War Dataset, 1816-1997, Sarkees, 2000, and, for 1997-99, the PRIO Battle Deaths Dataset 1946-2008, Lacina & Gleditsch, 2005. Data are aggregated over 25-year periods.

FIGURE 5-15: Deaths in wars involving the great powers, 1500-2000

FIGURE 5-16: Concentration of deaths in wars involving the great powers, 1500-2000

Sources: Graph from Levy, 1983, except the last point, which is based on the Correlates of War InterState War Dataset, 1816-1997, Sarkees, 2000, and, for 1997-99, the PRIO Battle Deaths Dataset 1946-2008, Lacina & Gleditsch, 2005. Data are aggregated over 25-year periods.