The Curious Science of Counting a Crowd

On June 4, a huge crowd gathered in Hong Kong for a vigil to commemorate the 22nd anniversary of the Tiananmen Square massacre in Beijing. But just how huge? In some stories 77,000 people showed up. Another story, though, listed the attendance as nearly double that: 150,000.

There's a reason for the disparity. The first figure?77,000?is a police estimate. The second is from the event's coordinators, who probably had some motivation to pad their numbers. To find out which crowd size was correct, two professors?Paul Yip at the University of Hong Kong and Ray Watson at Melbourne University?ran the numbers. To fit 150,000 people into that space, they'd have to cram together at about one person per 2.7 square feet (four per square meter), so that estimate is unrealistic. That would be "mosh-pit density," the researchers write in a new paper on crowd estimation techniques published in the journal Significance.

This story of competing head counts is not uncommon. Estimating large numbers is difficult even with the best of intention. If you count the number of jellybeans in a jar three times, you'll probably have three different numbers, because people simply cannot count very large numbers without some error. Now, imagine trying to count a shifting mass of heads, some stooping to tie shoes, some sharing the same umbrella, some arriving late or leaving early. Plus, this is one field in which good intentions are rare. Crowd-size estimation is a murky science, positioned at the intersection of statistical precision and political sleight-of-hand, and plenty of people are motivated to either exaggerate or low-ball an event's attendance.

"Almost everyone who has tried to make a crowd estimate has a vested interest in what the outcome of the estimate is," Charles Seife says. Seife is a journalism professor at New York University who writes about math and physics. [Disclosure: I had a class with Seife at NYU.] His newest book Proofiness tackles the ways that people try to fool others (and sometimes fool themselves) with numbers. "Whenever you see a crowd estimate," he says, "you have to wonder where it's coming from." Nevertheless, Seife says, if you do your math carefully, it is possible to count a large crowd to within a couple of tens of thousands. And researchers like Yip and Watson are now applying new strategies to find out whether it is indeed possible to get a more accurate count of a teeming mass of humanity.

Crowd-Counting 101

Herbert Jacobs, a journalism professor at the University of California, Berkeley, in the 1960s, is credited with modernizing crowd-counting techniques. From his office window, Jacobs could see students gathered on a plaza below protesting the Vietnam War. The plaza's concrete was poured in a grid, so Jacobs counted students in a few squares to get an average of students per square, then multiplied by the total squares. He derived a basic density rule that says a light crowd has one person per 10 square feet, a dense crowd has one person per 4.5 square feet, and Yip and Watson's mosh-pit density would have one person per 2.5 square feet.

Fifty years after Jacobs, the tools for counting crowds have improved but the principle is the same: area times density. Steve Doig, a journalism professor at Arizona State University, used a photo from a GeoEye-1 military satellite to count people at President Barack Obama's inauguration speech in 2009 (he estimated 800,000 people). The New York Police Department counts the people in the fenced crowd-control barricades that it places, then multiplies by the number of barricades. Yip and Watson applied the basic formula to the candlelight vigil in Hong Kong.

But a simple area times density calculation has its limits. Crowds are not uniform?they clump in some places and spread out in others. To account for this, estimation methods are becoming more sophisticated. Companies such Digital Design and Imaging Service are now adapting the formula for multiple densities. The firm has counted attendance at major events on the National Mall in recent years and claims it can count the crowd to within 10 percent. So CBS hired DDIS to count heads at Glenn Beck's rally at the Lincoln Memorial in August 2010.

To get its figure for the Beck rally, the design firm first cased the venue, created 3D maps marked with probable high-density spots and cross-referenced those with historical photos of similar events. The result was a prediction of how people would congregate. "Our goal is to find out where we anticipate the crowds will gather. If it's in the winter, we look for the wind breaks, and if it's in the heat of summer, we look for the shade," Curt Westergard, the company's president, says. Crowds press toward the stage, but also toward the Jumbotron screens, and they shy from loudspeakers, he says.

Knowing what to expect, Westergard chose his observation point and launched a tethered balloon at the height of the rally. The balloon lifted a suite of remote-control cameras that, within seconds, had captured 360-degree crowd shots at various heights: 200 feet, 400 feet and 800 feet. The different heights allowed for shots of people under trees and in hard-to-see places. He laid a composite of the images over the 3D model and counted heads. His team counted heads in grid squares that represented different densities. Then, for each density (such as lightly populated or very heavily populated) they multiplied the number of people per square by the number of squares of that category, finally arriving at an estimate of 87,000 people for the Beck rally.

"Unquestionably, that's what it was. As a benefit of the doubt, we gave it a 10 percent rate of error," Westergard says. "We go in pixel by pixel and put a dot on every head that we see. If a lady is there holding a baby, we put two dots there. We counted this thing three times and got an outside guy [Steve Doig] to count it, and he got back to us with a number that was similar to ours, which was 80,000." But, unsurprisingly, Westergard's certainty didn't satisfy everyone: News reports about the rally reported a smattering of different numbers. Rep. Michele Bachmann, announced from the stage that there must be a million people present, while NBC counted 300,000, and Glen Beck himself estimated the crowd at 300,000 to 650,000. In a summer of competing rallies, such as Jon Stewart's, there were no shortage of swipes at DDIS's methodology and its relatively low number. (For its part, the National Park Service tries to stay above the fray by not estimating crowd sizes. It stopped providing head counts after the organizers of the 1995 Million Man March accused the service of underestimating their crowd.)

The Future of Head Counts

Photos are good proxies for static crowds, but to count a crowd on the move, Yip and Watson say you've got to get down into the mass. That's the focus of their newest research, and they've come up with new methods for doing it. They've come up with two methods so far: a strategy that uses one crowd inspection point, and another that uses two.

In the one-point method, counters positioned near the focal point of a march or a parade tally the number of people who pass their station in a given time interval. But it's not exactly ideal: Some people may have ducked out before reaching the station and others may leave soon after passing it. Not ideal: You'd have to do a phone survey of the marchers to find out how long they stayed with the march, and that introduces a whole new set of survey-bias problems. A faster, more accurate method is to set up two counting stations, suitably spread out. The counters would tally people passing and also randomly survey them to ask if they also passed the other station (or planned to). Any more than two inspection points would increase the cost, but not appreciably increase the accuracy, the researchers write in their study.

At this point, it seems like getting an estimation so exact might be more trouble than it's worth. But technological help may be on the way, this time from the Web. Westergard has plans to crowd-source head-counting aerial photos to Amazon's Mechanical Turk. The Turk is a network of people around the world who do tasks online for a fee. Westergard can send a photo to 20 people, quickly receive 20 different head counts, throw out the outliers and average the rest.

If this all sounds like an academic exercise, remember that accurate crowd counting can have practical applications such as preparing emergency responders. If a fire, terrorist attack, stage collapse or other calamity happened at a large event, Westergard figures that within 20 minutes he could provide first responders with the location of the threat and rough estimates of the number of people who might need treatment.

And, as Yip said in a statement about his study, a good way to count crowds could cut through the politically motivated stats we put up with now. "In the absence of any accurate estimation methods, the public are left with a view of the truth colored by the beliefs of the people making the estimates. The public would be better served by estimates less open to political bias."

Source: http://www.popularmechanics.com/science/the-curious-science-of-counting-a-crowd?src=rss

311 ben bernanke good morning america sesame street colin farrell trigeminal neuralgia trigeminal neuralgia