Memory span

The first three rows measure a span directly; the fourth is a reading of many experiments; the last shows how far one person went with a technique, not what memory alone can do.

In 1956 the psychologist George Miller published an essay called "The magical number seven, plus or minus two". Looking across experiments of the time, he noted that people could repeat about seven digits, letters or words after one hearing, and that the number stayed oddly similar across kinds of material^[2]. The title became one of the best-known phrases in psychology.

Miller did not measure a group of people himself; his range is a summary of other experiments, not a limit measured in one study. It is still a fair description of a classic digit span test: most adults land in the middle of it, between five and nine.

A study published in 2011 tested 763 volunteers aged 18 to 65 in New Zealand with a computer-run digit span. The digits were spoken one at a time, each list was one digit longer than the last that was repeated correctly, and the test stopped after two misses at one length. Most people's span was 5 to 7 digits, and the middle value was six^[1].

A span for places works the same way. In a computer version of the Corsi task, squares on a screen light up one by one and you click them back in order. The control group of a study published in 2016, 187 people aged 18 to 82, averaged a span of 5.2 squares, with a spread of about 1.2^[4]. In these two studies, the span for places was a little shorter than the span for digits.

In 2001 the psychologist Nelson Cowan reviewed a wide range of experiments and argued that Miller's seven was a rough estimate rather than a real limit. When people cannot group items or rehearse them, he found, the limit is about three to five chunks, with a central capacity of about 4^[3].

The two figures fit together through chunking. A chunk is any group you hold as one thing. A phone number written as three groups is easier than the same digits in one run, because each group becomes a chunk. When you repeat seven digits, you are usually holding three or four small groups, and saying them over to yourself helps keep them alive.

This is why span tests are run fast and with random digits: a familiar sequence such as a year or a phone number would be one chunk and would make the span look longer.

A span is not fixed. In a study published in 1980, a student practised digit lists in the laboratory for more than 230 hours. His span grew from 7 to 79 digits, and on other memory tests with digits he matched people trained in memory all their lives^[5].

The authors put it down to a mnemonic system, a way of turning digits into something meaningful that long-term memory can hold, and wrote that with such a system there seems to be no limit to what practice can do^[5]. He did not grow a bigger short-term memory; he found a way around it. It shows how much a span test measures strategy as well as capacity.

Age matters too, more gently. In a large online study of people aged 10 to 69, working memory tasks peaked at around 30 years of age and declined slowly after that^[6].

A span is always a span on one particular test. The careful study above spoke the digits aloud, one after another, and the volunteers repeated each list back^[1]. Our test shows them on the screen instead: each digit appears for 0.7 sec, a new one every 1 sec. Hearing and seeing are different routes into memory, so the two are close relatives rather than the same test.

Speed matters as well. Show the digits slowly and there is time to say them over and group them; show them fast and there is not. That is why careful tests fix the pace, and why a span measured at a different pace, or with digits you could see all at once, should not be compared with another directly. The rule of the stopping point matters too: a test that stops after one miss gives lower spans than one that allows a second try.

• A span between five and seven digits is what most adults reach on a careful test; eight or nine is good.
• Tests differ: digits shown on a screen, digits spoken aloud and squares on a grid give slightly different spans.
• If you group the digits as you see them, you are chunking, which is what memory is meant to do; just know that it lifts the number.
• A tired or distracted day can cost a digit or two. Try again another time before drawing conclusions.

Our number memory test shows one digit at a time, starting with lists of 3, gives two lists at each length, and stops after two misses at one length, much as the careful test above did. A memory test in a browser is for curiosity and self-reflection, not a medical or psychological assessment.

For a memory test where a young chimpanzee once beat university students, see the chimpanzee memory test.