What the test measures
Some dots flash, and then they move among identical ones. Nothing marks them any more, so only your attention keeps them apart from the rest. The question is how many you can hold on to, and how fast they can go before you lose them. Alvarez and Franconeri (2007) showed that the two belong together: "At slow speeds it is possible to track up to 8 objects, and yet there are fast speeds at which only a single object can be tracked." So this page measures capacity at a fixed speed and also measures the speed itself.
How a trial runs
- All the dots stand still for 1 second. There are always twice as many dots as targets.
- The targets flash three times, 200 ms on and 200 ms off, and stay marked for another 400 ms. A marked dot is red with a ring around it, so the difference does not rest on colour alone.
- All the dots move for 8 seconds in straight lines at one speed. They bounce off the frame and off each other, and their centres never come closer than 2 dot widths.
- They stop. You mark as many dots as there were targets; if you lost one, guess, because the guessing correction needs the same number of marks every time. The last mark ends the trial, and until then a second click takes a mark back.
- Correct picks turn green with a tick, wrong ones red with a cross, and a target you missed gets a green ring.
The three modes
Standard follows the test of Meyerhoff and Papenmeier (2020): 4 targets among 8 dots at one fixed speed, 5 practice trials and 5 blocks of 10, so 50 trials and about 12 minutes here with a pause after each block. Their dots were 1.3° wide in a 20° frame and moved at 8° per second. This page does not know how far you sit from the screen, so it keeps the proportions: the frame is 15.4 dot widths across and the speed is 8 / 1.3 = 6.15 dot widths per second.
Speed threshold uses the number of targets you choose, 2 to 6, and looks for the speed at which you get 75% of the targets right. After a trial with at least three quarters of the targets correct the speed goes up one step of about 16%; otherwise it goes down one step. A run has 60 trials and takes about 14 minutes. Run it at other numbers of targets on other occasions and History draws your threshold against the number of targets.
Training is a session of 20 trials, about 5 minutes, with the same up-and-down rule. It starts from the speed where your last session at that number of targets ended, and History plots your speed by day.
How the results are calculated
Standard reports your tracking capacity m, corrected for guessing, with the formula from Hulleman (2005) that Meyerhoff and Papenmeier use:
n is the number of targets and p the share of them you picked correctly. With half the dots being targets, blind guessing gives p = 0.5 and so m = 0. An example: 4 targets and 87.5% correct give m = 4 × (2 × 0.875 − 1) = 3.0. The 95% interval comes from drawing your trials again at random with replacement 2,000 times and keeping the middle 95% of the values of m.
Speed threshold fits one curve to all 60 trials. The share of correct targets at speed v falls from nearly 98% towards the 50% of guessing as a logistic function of ln v, and the threshold is the speed where the curve crosses 75%:
The steepness b is fitted too, and the 2% left at the top allows for slips that have nothing to do with speed. The 95% interval holds every threshold under which your answers are not much less probable than under the best one (twice the difference in log likelihood stays below 3.84).
Sixty trials give only a rough threshold, and the result says so. In simulated runs with a simulated person and this page's own scoring (scripts/cognitive-tests/multiple-object-tracking.mjs, 1 November 2026), half the width of the interval was typically ±23% of the threshold at 4 targets for a person with a fairly steep curve and ±36% with a flatter one, and it passed ±38% in the worst tenth of runs. Treat two thresholds whose intervals overlap as the same.
Comparing your results
History compares runs on the same input and a similar refresh rate, because screens and viewing distances differ. Meyerhoff and Papenmeier tested 50 students twice, about 15 minutes apart, with two sets of paths: after they excluded two people with m below 0.5, the correlation was r(46) = 0.87 and the intraclass correlation 0.86 (95% interval 0.76 to 0.92). There is no percentile and no score against other people here: this page's dots and speed are only proportional to theirs, and one sample of students is no norm.
The chart of threshold against targets is yours alone. In Alvarez and Franconeri's first experiment 14 people took part (12 were analysed), 16 circles were on screen whatever the number of targets, and each person set their own speed limit; it fell with each added target from 1 to 8. Here the number of dots changes with the number of targets, so that guessing stays at 50%. Fewer targets therefore also means less crowding, and your curve is not the same measure as theirs.
What Training can and cannot show
A rising line means you got faster at this task on this screen. It says nothing about anything else. A review of a commercial 3D tracking trainer (Vater, Gray and Holcombe, 2021) found no preregistered studies at all and called the evidence for improved working memory and sustained attention very weak. Only three studies looked at transfer to real-world tasks, and two of them found no effect. That review is about another product, and nothing shows this page does better, so use Training out of curiosity about the task and not as a way to improve your sport, driving or memory.
Things to try
- Do Standard twice with a break between and see whether the two intervals overlap.
- Fix your eyes on the middle of the arena for one run and follow the dots with your eyes in another.
Limits
- Speed is in dot widths per second, so the same number looks different on a phone and on a large monitor, and at a different distance. A very small arena also makes the dots hard to see apart.
- A dot may move at most half its width per frame, so speeds stop below half your refresh rate: under 30 widths per second at 60 Hz, under 16.3 at 32.6 Hz. If the staircase reaches that limit the result says so. A trial in which the screen stalled does not count and another takes its place, up to 12 times a run.
- Below 50 Hz, which remote desktop and power-saving modes often cause, the motion looks jerky. The page warns you and offers Run anyway; such a run is marked less precise and kept apart in History.
- Nothing checks where you look, and moving your eyes can change the result. The dots here keep 2 widths apart so that they can be tapped; in the original test they could overlap, so the task may be a little easier here.
- This is not a medical test. It is not a concussion, driving or sports-vision test.
Controls
- Mouse: click a dot to mark it, click again to take the mark back.
- Touch: tap a dot; the nearest dot within 1.5 dot widths counts (about 33 px on a 375 px phone), so you need not hit it exactly. A tap further away marks nothing: a ring shows where it landed and the status line tells you to aim closer.
- Keyboard: arrow keys move the ring to the next dot in that direction, Space or Enter marks it, Esc pauses. Enter or Space also presses a focused button.
- Buttons: Pause during a run. After a pause, a result or Stop, buttons react only after 0.7 s, and Pause ignores a second click within 0.7 s of a start, so a double press cannot skip a step.
- Language: the small link at the end of History switches between English and Slovak.
Sources
- Meyerhoff H. S. & Papenmeier F. (2020). Individual differences in visual attention: a short, reliable, open-source, and multilingual test of multiple object tracking in PsychoPy. Behavior Research Methods 52:2556–2566. doi:10.3758/s13428-020-01413-4
- Alvarez G. A. & Franconeri S. L. (2007). How many objects can you track? Evidence for a resource-limited attentive tracking mechanism. Journal of Vision 7(13):14. doi:10.1167/7.13.14
- Vater C., Gray R. & Holcombe A. O. (2021). Psychonomic Bulletin & Review 28(5), from p. 1458. A systematic review of a commercial 3D multiple object tracking trainer; cited without its title because the title names the product.