How did the U.N. come up with its new Syria death toll?

Syria’s suffering now has an official number: 59,648.

That’s the death toll that Navi Pillay, the U.N. high commissioner for human rights, assigned to the Syrian government’s bloody political crackdown and the resulting civil war, over a period ranging from March 15, 2011, to November 30, 2012.

The precise number is, of course, an educated guess, but that figure has almost certainly passed the 60,000 mark in the new year, Pillay said.

The real number, according to Pillay, is probably even higher than that, given the fact that much of the Syrian carnage has played out in dark places, beyond the prying eyes of witnesses. "The number of casualties is much higher than we expected, and is truly shocking," she said Wednesday.

In fact, the number — which is significantly higher than previous, informal U.N. estimates of about 40,000 dead — has caught many top U.N. officials by surprise.

So, how then, did the U.N. human rights office, which has virtually no presence on the ground in Syria, come up with that figure?

They commissioned a team of statistical wizards at Benetech, a West Coast non-profit that runs a human rights program that crunches data to unlock hidden patterns of mass killing around the world.

The team was headed by the group’s lead statistician, Megan Price, and included Patrick Ball — chief scientist and vice president of the firm’s human rights data analysis group — whose computer models have been used to identify patterns of human rights violations from Guatemala to South Africa, and whose numbers aided in the prosecution of the alleged Serbian war criminal Slobodan Milosevic. (Read the excellent profile of Ball by Tina Rosenberg here.)

Applying a data mining technique called an alternating decision tree, Price, Ball and Jeff Klinger compiled basic fatality figures — such as victims’ ages, time and place of death — from seven separate data sets, including those maintained by the Syrian government and opposition groups, including the oft-cited Syrian Observatory for Human Rights.

The names and vital details of 147,349 reported killings were then run through a computer program that is designed to detect duplicate references to individuals. The model was refined by a native Syrian Arab speaker who went through a sample of about 8,200 pairs of killings.

The figure was then whittled down to 59,648 "unique" deaths, though Benetech notes that it "was not able to differentiate clearly between combatants and non-combatants." The seven data sets used ranged from the Syrian government’s record of 2,539 dead to more than 38,120 counted by the Violations Documentation Center, an opposition group. The larger number included in Pillay’s estimate reflected the fact that the analysis was drawn from seven separate data sets.

Price, the lead statistician, said that counting the dead in a war zone is a “really hard problem,” particularly given the fact that there are many other “things that feel more pressing than figuring out mortality figures in an active conflict.”

Price objected to the characterization of her group’s numbers as estimates, saying she and her colleagues simply enumerated “documented, verifiable deaths.”

“We in fact don’t know how many people have been killed in Syria,” she told Turtle Bay. “What we know is how many deaths have been documented by these seven groups.”

Price said she recognizes that the fog of war leaves open the possibility of errors creeping into her team’s count; for instance, an automobile accident victim counted as conflict related death. Or a single victims name is spelled differently on different data sets, leading to a single death counted as two.

But she said her team sought to anticipate some of these mistakes through a variety of computer procedures with names like “fuzzy matching” and “rejection rules.”  An example of fuzzy matching could involve the identification of variations on a single name –like Bob, Bobby, Rob and Robert – that would be read by the computer as the same name. Rejections rules are designed to prevent the computer from eliminating a potential fatality because they share a similar attribute—say a name – with another victim, but are not likely the same person. “Rejections rules are hard boundaries you are going to define to say those records cannot match,” Price said. “A common rejection rule is gender: any two records that have different genders are not likely the same individual.”

The decision by U.N. officials to assign a death toll for a given conflict can be highly controversial, and invariably provokes challenges by governments and sometimes other U.N. officials. In 2009, Pillay encountered intense pushback from top U.N. officials before publishing an account of the number of civilians who were slaughtered during the final months of Sri Lanka’s civil war.

This time around, Pillay’s deputy, Ivan Simonovic, faced little opposition when he informed U.N. Secretary General Ban Ki-moon and other top U.N. officials before Christmas that Pillay’s figure was going to be high– though he didn’t cite a number. One U.N. official said the figure turned out to be significantly higher than most of Ban’s aides had anticipated.

Rupert Colville, Pillay’s spokesman, told Turtle Bay that while this is the first time that the high commissioner has commissioned Benetech to estimate a conflict death toll, she has previously offered guesstimates of death tolls in Egypt and Tunisia.

Pillay released a Syrian death toll estimate in 2011, but resisted subsequent pressure to release an update because of uncertainty about the numbers. She was persuaded by Benetech’s analysis, according to Colville.

Colville acknowledged that there "is a bit of a risk" in basing the high commissioner’s estimate on raw data collected by independent groups. "It’s not a perfect number," he said. "But given the level of research that went into this, it’s far better than what we had before."

Benetech’s analysis showed a steady increase in the rate of killing — from 1,000 per month in the summer of 2011 to more than 5,000 per month since July 2012. The vast majority of those killed were male — over 76 percent. Just 7.5 percent were female. (The gender was unclear for 16.4 percent of cases.)

As for the geography of this grim toll, the largest numbers of killings were in Homs (12,560), rural Damascus (10,862), and Idlib (7,686), followed by Aleppo (6,188), Daraa (6,034) and Hama (5,080).

"While many details remain unclear, there can be no justification for the massive scale of the killing highlighted by this analysis," Pillay said. "The failure of the international community, in particular the Security Council, shames us all."

Follow me on Twitter @columlynch