Players in the same shirt
Imagine keeping your eye on three players wearing the same shirt during a football match, remembering which lane a few cars moved into on the motorway, or watching two children at once in a playground. What these have in common is following several similar things that are all on the move. When no colour or shape sets them apart, the only thing that keeps each one 'itself' is the continuity of its position.
In this experiment there are ten identical balls. A few of them flash orange for a moment; then they all look the same and move around for about seven seconds. When they stop, you try to find the orange ones. The better you do, the more targets you get, and the balls speed up a little each round. It is a version of what researchers call the multiple object tracking (MOT) task.
Up to five objects at once
In 1988 Zenon Pylyshyn and Ron Storm showed people displays of ten identical objects moving at random. Participants could track up to five targets. The authors also tested a model in which attention visits the objects one at a time: speeds and distances were set so that even under generous assumptions such serial scanning would be right only about 40% of the time. Participants, however, did the task with 87% accuracy.
They concluded that the visual system has a parallel mechanism that 'tags' several objects at once and follows them as they move. In Pylyshyn's visual index (FINST) theory the number of these pointers is limited; later work commonly cited four or five as the typical limit.
Is four a magic number?
In 2007 George Alvarez and Steven Franconeri challenged the idea of a fixed limit. When they varied the speed of the objects, the picture changed: at slow speeds people could track up to eight objects, at very fast speeds just one. On their account, tracking relies less on an architecture with a fixed number of slots and more on a resource that is shared flexibly among the objects.
Franconeri, Jonathan and Scimeca went a step further in 2010. Increasing speed or tracking time usually also increases the number of moments when objects come close together. When they held the distribution of distances between objects constant, large changes in speed and tracking time had no effect on performance. In their view, what really limits tracking is how close targets come to distractors. In this game both speed and the number of targets increase, so we cannot separate the two; but as targets multiply, you may notice more moments when a target slips past a distractor.
Taking luck out of the score
If four of ten balls are targets and you picked four completely at random, you would still find 1.6 targets on average. So the number you get right is not directly the number you tracked. In 2005 Johan Hulleman showed that the 'mark all' method is equivalent to drawing without replacement: someone who really tracks m of K targets makes the remaining K − m choices among the N − m balls they did not track, and some of those turn out to be targets by luck.
That gives an expected number correct of m + (K − m)² ÷ (N − m). We find the m that matches the number you got right and score the round as 2.5 × m. For example, if you found three targets in a four-target round, the number you really tracked is about 2.8, because your fourth pick also had a chance of being a target. The correction assumes that only targets are tracked; as Hulleman also noted, a single method cannot distinguish tracking the targets from tracking distractors and leaving them out.
Why does the number of targets change?
Instead of giving everyone the same number of targets, we use a staircase: find every target and the next round has one more, miss one and it stays the same, miss more and it drops by one. That keeps the difficulty near your own limit, so you do not waste rounds that are too easy or get stuck on rounds that are far too hard.
The estimate on the results screen is the number of targets you really tracked in your best round. We treat it as a rough estimate of your tracking capacity. Because the balls speed up each round, holding the same number of targets gets harder towards the end; that is also a chance to see Alvarez and Franconeri's speed effect in your own game.
What your score does and does not tell you
Your score is 2.5 times the sum, over the five rounds, of the number of targets you really tracked once chance is taken out. Someone who plays every round perfectly sees 2, 3, 4, 5 and 6 targets and scores 50. Finding no targets in a round, or only as many as luck would give, earns zero for that round.
Because we do not know your screen size or viewing distance, speeds are defined relative to the width of the field rather than in degrees per second. The same game covers a smaller angle on a small phone and a larger one on a big monitor, which makes one-to-one comparison with laboratory values difficult. As crowd data comes in, we will also show accuracy by number of targets.