Visual working memory, and what an average score on a visual memory test means
· 6 min read · Tyamka team
Visual memory versus visual working memory, what Luck and Vogel measured in 1997, and why their four items are a research result, not a target for you.
People search for the average score on a visual memory test after they have clicked through a grid of flashing squares and want to know where they stand. The research answer is that visual working memory holds about four simple objects at a time, but that number comes from one specific lab task and does not translate into a score on a web game, including ours. This article explains what visual working memory is, what the four-item result actually measured, and how our visual memory test online differs from it.
Visual memory and visual working memory
"Visual memory" in everyday speech covers a lot. Recognising a face from school, finding your car in a car park, remembering the cover of a book you read years ago. Most of that is long-term memory, and there is no single test for it.
Visual working memory is narrower. It is what holds a picture in mind for a few seconds after it has gone, so you can compare it with what you see next or act on it. You use it when you glance at a map and then look at the street, or when you copy a pattern from one part of a page to another. Its content fades within seconds unless you keep attending to it, and it holds only a few things at once.
Most web tests called "visual memory" measure some version of this short, limited store. That is why research on it is the useful comparison, and also why a result from one task says little about another.
What Luck and Vogel measured
The best known study is by Steven Luck and Edward Vogel, published in Nature in 1997 under the title "The capacity of visual working memory for features and conjunctions".
The task is called change detection. You see an array of coloured squares for 100 milliseconds. The screen goes blank for 900 milliseconds. Then a second array appears for 2 seconds. It is either the same as the first, or one square has changed colour, and you say which. The researchers varied how many squares were in the array, from 1 up to 12.
With 1 to 3 squares, people were almost perfect. From 4 squares on, accuracy fell steadily as the arrays got larger. Using a standard way of turning these answers into a capacity estimate, the authors concluded that people held the colours of roughly four items.
The same paper has two more results that are quoted less often.
- Objects, not features. When the items were coloured bars at different angles, people could remember both the colour and the angle of about four bars, not two of each. With objects that combined four features, they held about four objects, so sixteen features in total. The limit seemed to be counted in objects.
- Not words. In some blocks, people had to hold two digits in mind and say them at the end, which keeps the verbal system busy. The visual capacity estimate did not change, so people were not getting by on naming the colours to themselves.
Slots or a shared pool
Not everyone reads the result as four fixed slots. Some researchers, among them Paul Bays and Masud Husain (Science, 2008) and Ronald van den Berg, Wei Ji Ma and colleagues (PNAS, 2012), argue that visual working memory is more like a pool of resource spread across everything you are holding. Each added item gets a smaller share, so it is remembered less exactly, and some changes are missed because the memory is too vague, not because the item is gone.
The debate is still open. For a player it has one practical meaning. The limit is not a wall at exactly four. Memory gets less exact as you add items, and where it fails depends on how different the items are and how small a change you have to notice.
Why four is not a score for you
Four items is a result for one task under strict timing. It is a capacity estimate calculated from many trials, not a number of squares you should reach in a game.
Web tests differ from this in several ways.
- All at once or one by one. In change detection all the squares appear together. In many web tests, and in ours, the squares light up one after another, so time and order come into play.
- Same or different, or tap them back. Saying "changed" or "not changed" is easier than recreating a whole pattern by tapping. Guessing also works differently in each.
- Timing. In the lab the whole array was on screen for 100 milliseconds, and showing it for 500 milliseconds did not change the results. In a sequence game each square has its own flash, and the whole show can take several seconds, so you also have to hold the first squares while the last ones appear.
- Many trials or one run. A lab estimate averages dozens of trials per person. A game ends when you run out of lives.
So the level you reach or the number of squares you repeat in a web test cannot be compared with Luck and Vogel's four. We could not find a published average for the popular web formats that comes with a clear method, so we do not quote one.
What our game asks
Grid Trail is closer to a different lab task. Cards on a grid light up one at a time, and then you tap the same cards in the same order. The length of the sequence grows from 3 cards on level 1 to 8 on level 15, the grid grows from 3×3 to 5×5, and the flash of each card depends on the level, from 700 milliseconds on level 1 to 420 at the fastest. On level 14 you tap the cards in reverse, starting from the last one shown. A wrong tap costs one of three lives, and the game outlines the right card.
Remembering where things were and in what order is what the Corsi block test measures, a spatial memory test from the early 1970s. In one adult sample reported by Kessels and colleagues in 2000, the average span was 6.2 blocks, with a standard deviation of 1.3. We kept every level to 8 cards or fewer. The article on the sequence memory test average goes into Corsi in more detail.
Our levels are a guess until playtests, and we have not measured how far players get. We do not show a norm after a level. You see points, the time, your mistakes and your previous best on that level.
How to read your own score
Your earlier scores on the same level are the only fair comparison. The sequence of cards is new each attempt, some sequences are easier to see as a shape than others, and a short game like this is sensitive to sleep and distraction. Play a level three times and look at the middle score. If it rises over a couple of weeks, you have got better at this game. We do not claim that this changes anything beyond it. The research on that question is in does brain training work.
If you are curious how the same limit shows up with numbers, read the article on the number memory test average. The wider idea behind all of this is covered in what working memory is. Other games are on the all games page.