How Many People Are Here?

weirdly accurate.

How Many People Are Here?

How good are people at estimating crowd size?

This crowd estimation test asks how well your eye judges a crowd at a glance: seven crowds on a plaza, seen from high above for 1.5 seconds each, from about 15 people to 2,000. You guess each one and see the true count straight away. Psychologists study the same skill with brief flashes of dots, as Véronique Izard and Stanislas Dehaene did in 2008; people who count crowds for a living use a method Herbert Jacobs described in 1967.

This test turns the lab task into crowds of people on a plaza, seen from high above. Seven crowds appear for 1.5 seconds each, starting at about 15 to 30 people and roughly doubling every time up to 1,000 to 2,000; you answer on a logarithmic dial from 10 to 10,000 and see the true count straight after each guess. Your result is how many of the seven you got within ×1.25, how far off you typically were, and a ratio, guess ÷ truth, written as how many people you see for every 100 who are there, for the small crowds and the big ones.

It is not a test of intelligence or eyesight, and there is no percentile, because no survey of crowd guesses is big enough to rank against. The studies used dots rather than people, a shorter look and smaller numbers, so their figures are reference points. Screen size changes how a crowd looks, and seven guesses are a small sample.

What the test measures

One thing: how many people your eye sees for every 100 who are really there, and how that holds up as crowds get bigger.

The task

  • Seven crowds on a white stone plaza, seen from straight above, each for 1.5 seconds
  • Every count is exact, and each crowd’s density is set at random, so neither the area covered nor how dark it looks gives the number away
  • Answer on a logarithmic dial from 10 to 10,000, where each ×10 is the same length; tap roughly, then drag or use the Fewer / More bar to fine-tune
  • The crowds go from small to big, each drawn from a size band twice the last (15–30 up to 1,000–2,000), so each one is bigger than the one before, but by how much varies
  • After every guess the crowd comes back and its true count appears beside your answer on the dial, with how far off you were

Your result

  • Your score: how many of the seven crowds you got within ×1.25
  • Your typical miss (how many times off a guess usually was) and an overall ratio: how many you see for every 100 people there
  • A verdict: Underestimator, Well calibrated (within ×1.25) or Overestimator
  • Your seven crowds in a table, in the order you saw them: how many there were, what you said, and the ratio
  • Your pattern: the small crowds against the big ones, and whether your guesses kept pace as crowds grew (a rough power exponent from seven crowds)
  • Your ratios on the same scale as lab reference points from Izard & Dehaene, not a ranking
  • How crowds are really counted, Jacobs’s area × density method, and why official crowd estimates disagree
  • A shareable card with your overall score only, and a challenge link that sends a friend the same seven crowds

Before you start

  • About a minute and a half; free, no sign-up
  • Works on a phone or a computer; the 3D plaza needs WebGL. A phone and a big monitor are not directly comparable
  • Sound effects are optional and can be muted
  • Your guesses are not uploaded; a challenge link carries only your score, typical miss and ratio, the name you choose and which crowds you saw

Crowd size estimation FAQ

Is this test free? Do I need to sign up to see my result?

Yes, it’s free, and there’s nothing to sign up for: no account, no email, no payment. Your result appears on this page the moment you finish, and you can share it or play again straight away.

How do you estimate the size of a crowd?

Professionals multiply area by density, the method Herbert Jacobs described in 1967. As statisticians Ray Watson and Paul Yip summarise it, a loose crowd holds about 1 person per square metre, a solid one about 2 and a very dense one about 4. A careful estimate from a good photo is good to about ±10%.

How is the crowd estimation test scored?

The crowd estimation test divides each guess by the true count. A guess within ×1.25 either way counts as close, and the score is how many of the seven crowds were close. It also reports your typical miss and how many people you see for every 100 who are there, overall and on the small and big crowds.

Do people guess crowds too high or too low?

The result sheet answers that after your seventh crowd, because knowing first would change your guesses. It sets your ratios beside lab reference points: Izard and Dehaene’s 2008 adults, who saw brief flashes of up to 100 dots, and Portley and Durgin’s 2019 study of 113 people. Those studies used dots, not people, so they are reference points rather than a ranking.

Does seeing the true number help you estimate crowds?

Often, yes. In the second experiment of Izard and Dehaene (2008), 24 adults began each block with one display they were told held 30 dots, and their estimates lined up with it across the whole range, up to about 100 dots; the result sheet shows how far that one number moved them. This test shows the true count after every crowd, but each next crowd is bigger, so you still have to stretch the last answer.

Why do crowd estimates for the same event differ so much?

Crowds are hard to count, and people have reasons to get the count wrong, as Watson and Yip (2011) put it. Estimates for the 1995 Million Man March in Washington ranged from 400,000 (US Park Police) to 870,000 ±25% (a Boston University team using aerial photos) to 1.5–2 million (the organisers).

Is there a percentile or an average for crowd guesses?

No. No survey of crowd guesses is big enough to rank against, so the crowd estimation test shows lab reference points on the same ratio scale instead. Those studies used dots, shorter looks and smaller numbers, and the site does not publish its own players’ data.

How long does the crowd test take, and does it work on a phone?

The crowd estimation test takes about a minute and a half and works on a phone or a computer, with sound optional. Screen size changes how a crowd looks, so a phone result and a big-monitor result are not directly comparable.

Where the numbers come from

Your ratios are shown on the same scale as reference points from Izard & Dehaene (2008), whose participants estimated dots, not people, so there is no percentile. Jacobs’s densities are as Watson & Yip (2011) give them; Jacobs’s own 1967 article was not seen. Portley & Durgin (2019) was checked from its abstract. The site does not publish its own players’ data.

How the score works

Ratio

Your guess divided by the true count. 0.5 means you saw half the people who were there; it is shown as 50 for every 100. Your overall ratio is the geometric mean over the seven crowds; the small and big figures use the three smallest and the three biggest.

Close and typical miss

A guess within ×1.25 either way of the true count counts as close. The typical miss is how many times off a guess was, too low or too high, averaged geometrically: ×1.4 means a typical guess was off by 40% in ratio terms.

Power exponent

The slope of your guesses against the true counts on log–log axes. 1 means your guesses grow in step with the crowds; below 1 they fall behind as crowds grow, above 1 they run ahead. From seven crowds it is a rough figure.