List-Length Effect: More Words Recalled, a Smaller Share of the List

Two things happen at once when a list grows. More items are available to be recalled, so the raw count rises; and each item gets a smaller share of a limited resource, so the chance that any particular item survives falls. The first fact is obvious and the second is the effect. A person who recalls six of eight words and nine of twenty has done better in absolute terms on the long list and much worse in proportion, and both descriptions are true.

The resources that run out are of two kinds. During presentation, attention and rehearsal are spread across whatever has been shown so far; the more items there are, the less each one is rehearsed and the less distinctive each one becomes among its neighbors. At recall, the items compete: every word brought to mind is a cue that leads to some of the others and away from the rest, and a longer list means more competitors for every retrieval. Models of free recall explain the pattern with some mix of these two, and the most durable account is that longer lists dilute both the encoding of each item and the usefulness of the list itself as a cue at recall.

The effect interacts with the better-known serial position curve. In Murdock's data, the last few words of a list were recalled well whatever its length, because they were still in immediate memory when recall began, and the first few were recalled well because they were rehearsed most. What list length changes is the middle: a longer list has a longer, flatter middle in which each item has a low chance of coming back. Ward's 2002 overt-rehearsal study traced the effect to rehearsal recency: when recall was replotted by when each word had last been rehearsed, the curves for 10-, 20- and 30-word lists lay on top of one another, because a longer list leaves a smaller proportion of its words rehearsed and leaves the unrehearsed words further from the end. Where recall starts shifts with length as well: as Cortis Mack and colleagues summarize the later work from Ward's group, people are less likely to begin with the first item and more likely to begin with one of the last few as the list grows.

One caution about what the effect is not claiming. It does not say that long lists are useless or that memory has a fixed number of slots. The count keeps rising with list length; it just rises more slowly than the list does. And the proportion recalled depends on much besides length: the rate of presentation, the familiarity and relatedness of the items, whether rehearsal is possible, and how recall is tested all move it, which is why one list on one occasion says little about any individual.

Bennet Murdock's 1962 paper "The serial position effect of free recall" in the Journal of Experimental Psychology is remembered for the U-shaped curve on its cover, but its design was a study of list length and presentation rate. One hundred and three students heard lists of common words read aloud, with six list conditions combining lengths from 10 to 40 words with presentation rates of one or two seconds a word, and after each list wrote down as many words as they could in any order. The serial position curves showed the familiar shape in every condition: a steep primacy effect over the first three or four words, an S-shaped recency effect over the last eight, and a flat middle between them.

Across conditions, the longer lists produced more recalled words and a lower proportion recalled, with the extra length going almost entirely into the flat middle section of the curve. The recency portion was about the same size whatever the list length, which Murdock and later writers took as the signature of a short-term component that holds a few items regardless of how many preceded them. The paper's figures have been reanalyzed many times since; they remain the standard illustration of free recall curves in textbooks.

William Roberts's 1972 paper "Free recall of word lists varying in length and rate of presentation: A test of total-time hypotheses," also in the Journal of Experimental Psychology, took the two variables apart more systematically. By crossing several list lengths with several presentation rates, Roberts could ask whether recall depended only on the total time spent on a list, as total-time hypotheses predicted, or on length and rate separately. The results reproduced Murdock's pattern, more words and a smaller proportion from longer lists, and the paper's title records what it set out to test: whether total time alone could account for recall. Ward's 2002 paper gives the standard one-sentence summary of both: as the number of words in a list increases, the proportion recalled decreases even though the number recalled slightly increases.

The pattern has held up in later work far from the original laboratory. Cortis Mack and colleagues presented 40 students with daily lists of two to ten words on their smartphones in 2017, one word an hour, with free recall of the day's words each day. Because some scheduled words were never opened, they scored recall against the words each person had actually seen and treated the list lengths as nominal. On that footing the list length effect was clear at that timescale too: nominal two-word lists returned 1.4 words on average, 78 percent of the words seen, and ten-word lists 3.5 words, 45 percent of the words seen. They described the effect as well established: as list length increases, the number of words recalled increases, but the proportion recalled decreases.

The effect shows up wherever more is asked of memory than a handful of items.

The counter is to stop asking memory to carry long lists, and to shape the lists it must carry.

• Write it down. A list of more than a handful of items belongs on paper or a screen. The effect is a property of unaided recall, and the aid removes it.
• Chunk long lists into short ones. Sixteen grocery items in four aisles of four behave like four short lists, each of which is recalled nearly whole.
• Put what matters at the ends. The first and last items of any list survive best; the middle is where length does its damage. Open and close with the point you most need remembered.
• Slow down. Presentation rate matters alongside length. A list read at two seconds a word is recalled better than the same list at one, so give listeners time between items.
• Cut the list. When you cannot write it down, the cheapest fix is fewer items. Three instructions delivered will beat eight delivered and five forgotten.
• Test rather than repeat. Recalling a list once, a day later, does more for the middle items than hearing it again.

The list-length effect is one of several regularities in free recall, and its neighbors describe different parts of the same curves.

• It is not the magical number seven, Miller's estimate of how many items immediate memory holds at once. The span is a ceiling on a single immediate recall; the list-length effect describes what happens to count and proportion as lists grow past any span.
• It is not the serial-position effect, the U-shaped curve of recall by position. The two come from the same experiments, and list length changes the curve by stretching its flat middle, but the serial-position effect is about where in the list an item sits and the list-length effect about how many items there are.
• It is not the recency effect on its own. Recency holds the last few items whatever the length; it is the reason the list-length effect falls mostly on the middle.