What the test measures
Visual working memory holds what you have just seen for a second or two while you use it. The question here is how many objects it holds at once. You see 2, 3, 4, 6 or 8 coloured squares for a tenth of a second, and then one of them comes back. If you kept that square in mind, you know whether its colour changed; if not, you can only guess. How many squares you keep is your capacity, K.
How a trial runs
- A cross appears in the centre. Keep your eyes on it.
- 500 ms later the squares flash for 100 ms, each in a different one of 9 colours, at random places with centres at least 2.4 square widths apart.
- For 900 ms only the cross stays.
- One square returns at one of the places: in half of the trials in a colour that was not in the display, in the other half in its old colour. You answer Same or Changed within 3 seconds.
A run has 24 trials for each number of squares, 12 changed and 12 the same, 120 in all, in random order, with a short break in the middle. It takes 5 to 6 minutes. Before your first run there are 8 practice trials that tell you whether you were right; they do not count.
The design follows an experiment by Luck and Vogel that Cowan (2001) describes: 1 to 12 coloured squares for 100 ms, a 900 ms blank, then the same display or one with a single square changed. Performance was nearly perfect with 1 to 3 squares, slightly worse with 4 and much worse with more. Here the test display shows only the one square, so you know which square to check.
A screen draws whole frames, so the flash is timed in frames: 6 frames at 60 Hz. The page measures the frame rate before each run, records how long every flash was really on, and repeats a trial later in the run when a frame was dropped, for example because the computer was busy.
How K is calculated
For each number of squares N the page counts the hit rate h (the share of changed trials where you answered Changed) and the false alarm rate f (the share of unchanged trials where you also answered Changed).
This is Cowan's formula. Rouder et al. (2011) derive it from a slot model: you hold K of the N squares, so the tested square is in memory with probability K/N and then you answer correctly; otherwise you guess Changed at some rate u. So h = K/N + (1 − K/N)u and f = (1 − K/N)u, and the guessing cancels out in h − f. An example: with 4 squares, 11 of 12 changes caught and 2 false alarms in 12, K = 4 × (11/12 − 2/12) = 3.0. Pashler's formula, K = N(h − f)/(1 − f), is for tests where the whole display returns; Rouder et al. show that the two are not interchangeable, so a single-square test uses Cowan's.
Kmax, the main result, is one K and one guessing rate u fitted by maximum likelihood to all your answers with 4, 6 and 8 squares: the values under which your answers were most probable. With 2 or 3 squares K cannot come out above N, and a single slip there pulls it down, so those trials draw the start of the curve but stay out of Kmax, as Rouder et al. suggest for set sizes below your capacity.
The 95% interval of Kmax holds every K under which your answers are not much less probable than under Kmax: twice the difference in log likelihood stays below 3.84, the 95% point of the chi-squared distribution with one degree of freedom. For K at each number of squares the page draws your trials again at random with replacement 2,000 times, recomputes K and keeps the middle 95%. Shown values stay within 0 to N; when every answer at a set size is right, the table says all correct.
In the first chart the dashed line is the slot model with your Kmax, K = min(N, Kmax). In the second it is the share of correct answers the model predicts, (1 + min(1, Kmax/N))/2: 100% while N is at most Kmax, then falling towards 50%.
What to look for
- The points rise with N and level off. Where they level off is your capacity.
- Luck and Vogel (2013) write that the typical college student holds only 3 to 4 objects' worth of simple, clearly different coloured squares. The shaded band in the K charts marks that range.
- The guessing rate u shows whether you lean towards Changed or Same when you do not know. A lean moves hits and false alarms together and leaves K as it is.
Things to try
- Do two runs on different days on the same screen and compare their intervals in History. When they overlap, the difference may be chance.
- Do one run while you repeat a word aloud, "the, the, the", so that you cannot name the colours to yourself, and compare it with a quiet run.
- Look at the false alarms with 8 squares: how often you answer Changed when the square kept its colour.
- Run the test on a phone and on a computer, then switch History to all runs.
Comparing your result
History compares runs on the same input whose refresh rates are within 10% (or both below 50 Hz), because screens differ in size, brightness and timing. Luck and Vogel (2013) report that a 10-minute change localisation task, a version where a square always changes and you point to it, had a test-retest reliability of 0.77 over 1.5 years. This page uses a different task and 120 trials, so read the interval of a run rather than its decimal. There is no percentile or score against other people: one run is too short for that, and the published figures come from other procedures.
Limits
- The task is about colour. With red-green colour blindness or another colour vision difference some of the 9 colours look alike, and K comes out lower without saying anything about memory. Screens also differ in brightness and colour, so a run on another screen is not quite the same test.
- The flash lasts whole frames: 100 ms is 14 frames = 97.2 ms at 144 Hz and 8 frames = 106.7 ms at 75 Hz. The page shows the real value for your screen. Below 50 Hz or with uneven frames, which remote desktop and power-saving modes often cause, it offers Run anyway: the flash then lasts whole slower frames (3 frames = 92 ms at 32.6 Hz), and the run is marked less precise and kept apart in History.
- With 8 squares only one of the 9 colours is left out of the display, so a change always brings that colour.
- A moment of inattention counts as a guess and lowers K a little. A capacity above 4 shows only with 6 and 8 squares, so it is measured less precisely.
- The slot model is a simplification. Luck and Vogel (2013) review evidence for fixed slots and for a resource shared among all items that gets less precise as there are more of them.
- This is not a medical test. It cannot show memory problems, dementia or colour blindness.
Controls
- Keyboard: F = same colour, J = changed, Esc = pause; Enter or Space presses the focused button. Only the first key after the square returns counts.
- Mouse and touch: the Same and Changed buttons. After a pause, break or result, buttons react only after 0.7 s, so a double press cannot skip a step.
- Run anyway (less precise): see Limits.
- Language: the small link at the end of History switches between English and Slovak.
FAQ
Why is my K with 2 squares exactly 2?
Because K cannot be larger than N: answer every trial with 2 squares correctly and h = 1, f = 0, K = 2 × 1 = 2. Less than 2 there means you missed some.
Why does the test say the flash cannot be timed?
It measures your screen first. Below 50 frames a second, or with uneven frames, a 100 ms flash cannot be shown exactly; remote desktop and power-saving modes are common causes. Closing other tabs or plugging in a laptop often helps; otherwise press Run anyway, and the result is marked less precise.
Sources
- Cowan N. (2001). The magical number 4 in short-term memory: a reconsideration of mental storage capacity. Behavioral and Brain Sciences 24(1):87–114. doi:10.1017/S0140525X01003922 (the experiment of Luck and Vogel is described on p. 95).
- Rouder J. N., Morey R. D., Morey C. C. & Cowan N. (2011). How to measure working memory capacity in the change detection paradigm. Psychonomic Bulletin & Review 18(2):324–330. PMC3070885
- Luck S. J. & Vogel E. K. (2013). Visual working memory capacity: from psychophysics and neurobiology to individual differences. Trends in Cognitive Sciences 17(8):391–400. PMC3729738