Visual Memory Test Statistics

A distribution of levels behaves differently from one of times

Levels are whole numbers and there are not many of them, so this is a short discrete distribution rather than a continuous curve. Each bar is a level, and the difference between neighbouring bars is a genuine step in difficulty rather than a rounding of something smooth.

That makes the percentile jump in visible increments, and there is no honest way around it. Clearing one more level really does move you past everyone who stopped at the one below, and interpolating within a level would invent a precision the score does not have.

It also makes the mean slightly misleading in a different way from the reaction tests. An average level of nine point four does not describe any attempt that has ever happened, whereas the share of attempts clearing level nine describes something real.

Where the curve steepens, and why

Attempts cluster and then fall away sharply rather than tailing off evenly. The reason is that the test grows on two axes: most levels add one square, but some also enlarge the grid, and the levels where both happen at once are where a lot of runs end.

The first few levels are cleared by nearly everybody, so the left end of the distribution is nearly empty. A run ending at level three is much more likely to be a misclick or an abandoned attempt than a genuine limit, which is worth remembering when reading the bottom of the chart.

The upper tail is thin and long. A small number of attempts reach levels far beyond the bulk, and because three lives allow recovery from mistakes, a very high score generally reflects a real strategy rather than luck.

What the sample is and is not

It is attempts, not people. Somebody who plays ten times contributes ten results, and because scores improve over the first few attempts as strategy develops, the distribution mixes first tries with practised ones. No site of this kind can separate the two without accounts, and this one deliberately has none.

Mouse and touch attempts share a single distribution here, unlike the reaction tests. The score is a level rather than a latency, so the input device cannot shift it the way it shifts a reaction time, and keeping one pooled sample gets the chart to a usable size sooner.

The count of attempts behind the chart is printed beside it and is not rounded up or dressed as a bigger number. Until it passes two hundred there is no curve here at all, only the count and a note saying so.

Order and reverse sit under the pattern chart

Sequence and reverse write the second and third figures on this page. They share the board with pattern and they do not share its axis, because a length recalled in order is not a level of squares recalled in any order.

Each playlist has its own two-hundred-attempt gate. Reverse will lag the other two and the caption will not pretend otherwise.

Mouse and touch stay pooled on all three, as they do on the other memory tests. Input latency cannot move a level the way it moves a reaction time.

This is a measurement toy, not a medical or psychological test. No score here says anything about your health, attention or cognition, and it is not a screening tool of any kind.

Questions

What percentage of people beat level ten on visual memory?

The chart on this page gives the current figure from the attempts recorded here, and it changes as more arrive. Any fixed number quoted without a sample size behind it is a guess.

Why are there so few results at the very low levels?

Because almost everyone clears the first few, so a run ending there usually means an abandoned attempt or a misclick rather than a genuine limit.

Do repeat attempts inflate the distribution?

They shift it upward somewhat, since scores improve over the first few tries and the sample counts attempts rather than people. There is no account system here to separate them.

Where did the sequence curve go?

It is the second figure on this page, with reverse underneath it. Both are playlists on the visual memory test, not separate URLs.