BrainFinch
Articles

Subitizing: why you can see four dots without counting

Subitizing is knowing how many items are in a small group the moment you see them, without counting. It works for up to about four items. Past that, people either count one by one, which is slow, or estimate, which is fast but approximate, with errors that grow as the group gets bigger.

The game above is Dot Estimator. It flashes a field of dots for about a second and asks how many there were.

What is subitizing?

It is the quick, exact recognition of a small number of things. Three coins on a table look like three. Nobody counts them.

The word comes from a 1949 paper by Kaufman, Lord, Reese and Volkmann in The American Journal of Psychology, titled “The discrimination of visual number”. They built it from the Latin subitus, meaning sudden.

How many items can you subitize?

About four, though sources draw the line in slightly different places.

Trick and Pylyshyn (1994) summarized the timing evidence in Psychological Review. Below four items, enumeration is rapid, accurate and takes little effort, with each extra item adding roughly 40 to 100 milliseconds. Above four it is slow and error-prone, and each extra item adds roughly 250 to 350 milliseconds. Plotted on a graph, response time against number of items is a shallow line that bends sharply upward.

Revkin and colleagues (2008) give the range as up to about 3 or 4 items.

Those per-item times explain why counting fails on a brief display. At 250 milliseconds an item, 20 dots would take five seconds to count, and Dot Estimator shows them for one.

What happens beyond four items?

If there is no time to count, you estimate. An estimate comes quickly and is approximately right, and how approximate depends on the number.

The rule usually cited here is Weber’s law. In its general form it says that the smallest difference you can notice between two stimuli is proportional to their size. Applied to number, the usual reading is that the ratio between two quantities is what matters, and that the spread of people’s estimates should widen in step with the number being estimated. An error of 3 is large for a field of 10 dots and small for a field of 60.

Cordes, Gelman, Gallistel and Whalen (2001) found this pattern in a task without dots. Participants pressed a button a target number of times while saying “the” with each press, which kept them from counting in words. The variability of their responses, measured relative to the target, stayed constant across targets. The authors call that scalar variability.

Estimates also drift in one direction. Izard and Dehaene (2008) asked adults to estimate arrays of dots and found that, without any reference, the estimates rose with the true number but came in under it. Showing a few example displays labeled as 30 dots was enough to recalibrate estimates across the whole range tested.

Is subitizing just very accurate estimation?

The evidence we checked does not settle it. The idea is reasonable, because Weber’s law predicts high precision for small numbers anyway.

Revkin and colleagues (2008) tested the idea directly. Adults named the number of items in briefly shown, masked displays of 1 to 8 items and of 10 to 80 items, with the two sets matched for how hard their members were to tell apart. If one estimation system handled both, precision should have been similar across the matched sets. It was much higher for 1 to 4 items than for 10 to 40, and the authors concluded that small numbers have a dedicated mechanism.

The button-press study by Cordes and colleagues points the other way. Its variability measure stayed the same inside and outside the range of 1 to 4 or 5, which the authors read as small numbers sitting on the same continuum as large ones.

The two studies used different tasks, one visual and one based on producing a count, so they do not flatly contradict each other.

How does Dot Estimator work?

Dots flash, hide, and you report the count. Early trials offer four numbers to tap. Later trials ask for a typed two-digit estimate, which scores if it is within 15 percent of the true count.

A round lasts 45 seconds and each correct answer is one point. After every answer the dots come back with the real count printed, for 0.5 seconds after a right answer and 0.95 seconds after a wrong one.

We read the table below from the game’s generator. The difficulty moves up a step after every three fields, and a new player starts at step 0.

Step Dots On screen for How you answer What scores
0 3 to 8 1.2 s tap one of four numbers the exact count
1 5 to 12 1.1 s tap one of four numbers the exact count
2 8 to 18 1.0 s tap one of four numbers the exact count
3 10 to 25 0.9 s tap one of four numbers the exact count
4 12 to 24 1.0 s type two digits within 15%
5 12 to 30 0.95 s type two digits within 15%
6 12 to 36 0.9 s type two digits within 15%
8 12 to 48 0.8 s type two digits within 15%
10 12 to 60 0.7 s type two digits within 15%
12 and up 12 to 60 0.6 s type two digits within 15%

Between steps 4 and 12 the top of the range rises by 6 dots a step until it reaches 60, and the flash shortens by 0.05 seconds a step. Steps 7, 9 and 11 are left out of the table for space.

The four choices are spaced by ratio. Each wrong choice is the count multiplied or divided by a set factor, then the same again for the next one out. The factor is 1.6 at step 0 and falls to 1.5, 1.4 and 1.3. A field of 6 dots at step 0 can come with the choices 6, 10, 16 and 26. No wrong choice is ever within 15 percent of the true count, and the true count can sit in any of the four positions. So the task in the early steps is to tell apart numbers that differ by a known ratio.

Only step 0 reaches into the subitizing range, with its fields of 3 and 4 dots. Everything else in the game is estimation.

You also cannot judge the count from how much color is on screen. Each dot’s radius is set to 0.27 divided by the square root of the count, as a fraction of the field’s width, with a cap for small counts. For 13 dots or more that makes the dots cover about 23 percent of the field in total, whatever the count. More dots means smaller dots. The dots also never touch. If a field is too crowded to place them all, the code shrinks every dot by a tenth and tries again.

How close does a typed estimate have to be?

Within 15 percent of the true count, checked with whole numbers. We ran the game’s scoring function over every two-digit guess to get the accepted range for each count.

True count Accepted estimates Furthest miss allowed
12 11 to 13 1
20 17 to 23 3
30 26 to 34 4
40 34 to 46 6
50 43 to 57 7
60 51 to 69 9

The window is a fixed ratio, so it widens with the count. That is a design choice modeled on the way estimation error is described above, with tolerance growing in proportion to the number. It also means a guess of 34 scores for both 30 and 40 dots.

Given the finding that uncalibrated estimates tend to come in low, the count shown after each answer is the useful part of the screen. It is the only feedback that tells you which way you missed.

What does a Dot Estimator score tell you?

It tells you how many fields you judged correctly in 45 seconds, and nothing beyond that. It is a game score, and it does not measure number sense or any ability.

People do get better at a task they repeat, so scores tend to rise with play. Whether practice of this kind carries over to anything else is a separate matter, and our article on what the research shows covers it.

Two other games on the site involve quantities seen quickly. Count Only drifts shapes across the screen and asks you to count the ones that match a target, which is exact counting under distraction. Quick Compare shows two numbers and you tap the bigger. The rest of the math lane is described in the mental math games guide.

Common questions

What is subitizing?

Subitizing is recognizing how many items are in a small group at a glance, without counting them one by one. Kaufman, Lord, Reese and Volkmann coined the term in 1949 from the Latin word subitus, meaning sudden.

How many objects can you subitize?

About four. Trick and Pylyshyn (1994) describe enumeration as fast and accurate below four items and slow and error-prone above four. Revkin and colleagues (2008) put the range at up to about three or four.

What is the difference between subitizing and estimating?

Subitizing gives an exact count of a small group almost at once. Estimating gives an approximate count of a larger group, and the size of the typical error grows with the number of items.

Sources

  1. Kaufman, E. L., Lord, M. W., Reese, T. W. & Volkmann, J. (1949). The discrimination of visual number. The American Journal of Psychology, 62(4), 498.
  2. Trick, L. M. & Pylyshyn, Z. W. (1994). Why are small and large numbers enumerated differently? A limited-capacity preattentive stage in vision. Psychological Review, 101(1), 80-102.
  3. Revkin, S. K., Piazza, M., Izard, V., Cohen, L. & Dehaene, S. (2008). Does subitizing reflect numerical estimation? Psychological Science, 19(6), 607-614.
  4. Izard, V. & Dehaene, S. (2008). Calibrating the mental number line. Cognition, 106(3), 1221-1247.
  5. Cordes, S., Gelman, R., Gallistel, C. R. & Whalen, J. (2001). Variability signatures distinguish verbal from nonverbal counting for both large and small numbers. Psychonomic Bulletin & Review, 8(4), 698-707.
  6. Weber-Fechner law. Wikipedia.

Games in this article

Take all 62 games with you

The BrainFinch app plays offline, keeps your streak and reminds you when the daily challenge is ready.

Get it on Google Play