On Thursday, more than 700 people were killed in a stampede outside the holy city of Mecca. The disaster took place during the annual Hajj pilgrimage, which draws about 2 million Muslims to Saudi Arabia each year.

Although this was the deadliest Hajj episode in a quarter-century, it is a story that is sadly familiar. In Mina, where Thursday’s disaster took place, stampedes killed more than 360 people in 2006 and 244 in 2004. In the worst Hajj stampede, 1,426 pilgrims were crushed in a pedestrian tunnel leading to Mecca in 1990.

Worldwide, human stampedes are so common—and so confounding—that they’ve inspired their own body of academic research within the larger field of study on crowd behavior. According to one 2010 study led by Edbert Hsu of the Johns Hopkins School of Medicine, 215 human stampedes took place worldwide between 1980 and 2007, leading to more than 7,000 deaths and 14,000 injuries. Stampedes have been evaluated as a public health issue, and as a sociological phenomenon. Others have asked whether the right algorithm could help identify dangerous crowd surges before they turn deadly.

Large religious gatherings are a particular stampede danger in the developing world. A 2013 paper out of India, for example, found that 79 percent of stampedes in that country have taken place at religious events, as opposed to political or entertainment-related events. In 2014, 178 people were killed in various Indian stampedes, and the country’s annual total death toll from stampedes has topped 300 four times in the past decade. According to Hsu’s research, the deadliest stampedes are concentrated in Southeast Asia and in Africa, and at religious events. But they can and do occur anywhere, as evidenced by the notorious New York stampede at a 2008 Wal-Mart Black Friday sale, which killed a store employee. With a growing global population concentrated in crowded cities, Hsu and his team found in 2010 that stampede incidents were on the rise.

Stampedes are captivating in part because, though somewhat common, they are also unpredictable. Here’s how John Seabrook described them in a harrowing 2011 article in The New Yorker:

The transition from fraternal smooshing to suffocating pressure—a “crowd crush”—often occurs almost imperceptibly; one doesn’t realize what’s happening until it’s too late to escape. Something interrupts the flow of pedestrians—a blocked exit, say, while an escalator continues to feed people into a closed-off space. ... At a certain point, you feel pressure on all sides of your body, and realize that you can’t raise your arms. You are pulled off your feet, and welded into a block of people. The crowd force squeezes the air out of your lungs, and you struggle to take another breath.

For remote observers, the terror of the phenomenon is compounded by the fact that it’s so hard to know whom to blame. In Saudi Arabia, the country’s health minister chalked up the latest incident to a failure to follow instructions, and the head of the Central Hajj Committee blamed “some pilgrims from African nationalities.” But others pointed to the Saudi government’s failure to manage the event. (It doesn’t help the Saudis’ claims of competence that a crane collapse at Mecca’s Grand Mosque killed 109 people just a few weeks ago.)

The current discussion echoes what Seabrook concluded: Crowds often take the blame for what are actually failures of planning and logistics. Press accounts, he found, often characterize stampedes as “panics,” with a frenzied mob surging forward with no regard to whom they trample. But one recent analysis of crowd disasters, which focused on 2010 stampede that killed 21 people at the Love Parade music festival in Germany, found that the disaster had more to do with physics than psychology: “Video recordings show that people stumbled and piled up due to a ‘domino effect’, resulting from a phenomenon called ‘crowd turbulence’ or ‘crowd quake.’ This was the consequence of amplifying feedback and cascading effects, which are typical for systemic instabilities.”

Blocked exits, overcrowded spaces, and insufficient security and emergency services all exacerbate the dangers. Event organizers, managers, governments, promoters, designers, and other entities defend themselves vigorously in the aftermath of these disasters. But the crowd, so brutally cooperative in one moment, rarely speaks in such unison afterward. With no one to defend it, the crowd can be personified as violent or dumb.

And although it’s easy to assume that stampedes are caused by panicked crowds running away from something in fear, Seabrook found that, in general, that’s only true in fires. In most stampedes, the crowd is churning toward something. In the United States and Europe, stampedes are rarer than they are in the developing world, and they don’t tend to happen on religious occasions. Americans and Europeans stampede for other causes: Black Friday sales, rock concerts, and sporting events. No one person decides to stampede. But if there’s a connection between what attracts a crowd and what a society holds dear, then stampedes are a deadly illustration of those values.