On Thursday, President Obama confessed to a terrible mistake. An American drone strike on al-Qaida operatives in Pakistan, carried out in January, accidentally killed two hostages—Warren Weinstein of the United States and Giovanni Lo Porto of Italy—who were concealed at the site.

The drone program’s enemies, and even some of its friends, are aghast. “Current and former U.S. counterterrorism officials said that Thursday’s disclosures undercut years of U.S. claims about the accuracy of the drone program,” says the Washington Post. The new fatalities “add to an increasingly dismal set of statistics on U.S. citizens. Since 2002, at least eight Americans have been killed in U.S. drone strikes in Pakistan and Yemen.”

The New York Times takes the same grim view. “Every independent investigation of the strikes has found far more civilian casualties than administration officials admit,” says the paper. It calls the latest news “a devastating acknowledgment for Mr. Obama, who had hoped to pioneer a new, more discriminating kind of warfare.”

The outrage is understandable. But these two deaths, tragic as they are, don’t change the fundamental truth: For civilians, drones are the safest form of war in modern history. As I’ve documented before, they’re more discriminating and more accurate. If you want to minimize civilian casualties, getting rid of drones—and steering warfare back to bombing and shelling—is the worst thing you could do.

Look at the record in Pakistan. The harshest tally of drone strikes, maintained by the Bureau of Investigative Journalism, says drones have killed 2,449 to 3,949 people there, including 423 to 962 civilians. If you work with the low-end figures, that’s a civilian casualty rate of 17 percent. If you use the high-end figures, it’s 24 percent. In Yemen, the bureau counts 436 to 646 deaths by drone, of whom 65 to 96 were civilians. That’s a rate of 15 percent. If you factor in other incidents classified as possible but unconfirmed drone strikes, the rate in Yemen drops to somewhere between 8 percent and 14 percent.

The New America Foundation keeps a different tally. Its figures imply a civilian casualty rate of 8 percent to 12 percent in Pakistan and 8 percent to 9 percent in Yemen. A third count, maintained by the Long War Journal, indicates a 5 percent civilian casualty rate in Pakistan (once Weinstein and Lo Porto are added to the tally) and 16 percent in Yemen.

Compare those numbers with any other method of warfare. Start with an apples-to-apples comparison: the Bureau of Investigative Journalism’s analysis of “other covert operations” in Yemen. According to BIJ’s methodology, this category consists of nondrone attacks by U.S. forces, “including airstrikes, missile attacks and ground operations.” BIJ counts 68 to 99 civilian deaths in these operations, among 156 to 365 total casualties. That’s a civilian casualty rate of 27 percent to 44 percent: three times worse than drone strikes in the same country. Or look at the bureau’s data from Somalia. For drones, the BIJ counts 23 to 105 casualties, of whom zero to five were civilian. For other covert operations, the BIJ counts 40 to 141 casualties, of whom seven to 47 were civilian. If you go with the low-end numbers, drones have a perfect record in Somalia. If you go with the high-end numbers, drones are seven times safer than the alternatives.

In the past month, hundreds of civilians have died in Yemen. But the culprit isn’t drones. It’s old-fashioned airstrikes and artillery fire, courtesy of Saudi Arabia and its Arab partners. The campaign got off to a roaring start, with attacks on schools, hospitals, houses, mosques, a market, a dairy factory, and a refugee camp. As of April 14, the U.N. reported at least 364 civilian deaths. During this time, the BIJ counted four drone strikes in Yemen, resulting in 13 to 22 fatalities. None of them were civilian.

Before the emergence of drones and other precise weapons, war was far more dangerous for ordinary people. In World War II, an estimated 40 to 67 percent of the dead were civilians. In Korea, the estimate was 70 percent. In Vietnam, it was about one civilian for every two enemy combatants. In the Persian Gulf War, it may have been no better. In Kosovo, it seems to have been worse. In Afghanistan, civilian deaths have been estimated at 60 to 150 percent of Taliban deaths. In Iraq, civilians account for more than 80 percent of the casualties. To be fair, these were full-blown wars. You can argue that the better alternative to drone strikes is diplomacy, not invasion. But you ought to credit drones, conversely, for providing a military alternative to all-out war.

Last summer, Israel took extraordinary measures to avoid killing innocent people in Gaza. But the results were still horrific. According to a postwar investigation by the Associated Press, Israel’s 247 airstrikes on residential buildings killed 844 people. Of these, 508 were women, children, or men aged 60 or older, “all presumed to be civilians.” If, in exchange for that presumption, you posit that every dead man between the ages of 16 and 59 was an enemy combatant, that’s still a 60 percent civilian casualty rate. A broader U.N. tally, counting 1,483 Palestinian civilians among 2,205 total casualties of the war, puts the rate at 66 percent.

In the current war against ISIS, the U.S. and its allies claim to have carried out more than 3,000 airstrikes and killed more than 8,000 ISIS fighters with zero confirmed civilian casualties. But nobody trusts those numbers. As of April 10, Airwars, a website dedicated to investigating the casualties, had compiled claims of 470 to 565 civilian deaths, of which 240 to 300 were credible based on “reasonable indications.” The site also says at least 110 deaths, “with varying levels of certainty,” have been attributed to friendly fire.

If you look at long-term data from Pakistan, you’ll see a clear trend. Since 2012, drone strikes have declined. But civilian fatalities, at a far more acute rate, have virtually disappeared. A year ago, BIJ reported, “In the past 18 months, reports of civilian casualties in attacks on any targets have almost completely vanished … despite a rise in the proportion of strikes that hit houses.” By contrast, BIJ noted that in the previous six months, “the Pakistan military has carried out several large-scale bombings on suspected militant targets, including in urban areas. Scores of civilians have reportedly been killed.”

Drones aren’t more dangerous to hostages, either. Two men held by al-Qaida, including American Luke Somers, died last December during a commando raid in Yemen. Another American hostage, Kayla Mueller, was confirmed dead in February after a Jordanian airstrike on an ISIS facility. Officials involved in the strike insisted “they had conducted detailed surveillance to make sure that no hostages were seen going in or out” of the building, according to the Times. But one official “acknowledged that they had not been able to survey the building around the clock.”

With drones, you can watch the target around the clock. You might miss something, as in the Weinstein case. But you’re far more likely to make that mistake in a conventional airstrike or a ground assault. In the air campaign against ISIS, we’re dropping bombs 25 minutes after our allies on the ground call in targets. Would you rather be a civilian there or in Pakistan?

I’m sorry Weinstein and Lo Porto are dead. I’m even sorrier that our own government killed them. But they didn’t die because the weapon we sent to watch that building was a drone. They died despite it.