Recency Effect: Why the Last Thing You Heard Stays Freshest

The classic explanation is that the last few items on a list are still in short-term memory when recall begins. They have not been rehearsed into long-term memory the way the first items have, and they do not need to be: they are simply still available, and most people write them down first. The middle items have been pushed out of the short-term store by the items that followed and did not get enough rehearsal to be stored durably, so they are the ones that fade.

The decisive evidence is what a delay does. If recall is postponed for a few seconds while the person does something that occupies short-term memory, such as counting aloud, the end of the list loses its advantage and the last items are recalled no better than the middle, while the first items are unaffected. A delay that is simply empty does much less, because people keep rehearsing the last items through it. The demo above uses a filled delay of 30 seconds, the condition in which Glanzer and Cunitz found no trace of the end peak, so for most visitors the right side of the drawn curve will not rise.

The two-store account is not the only one. Bjork and Whitten showed in 1974 that a recency effect reappears in long-term memory when the items are separated by filled intervals and the test comes after a further interval, a pattern a short-term store cannot produce. Later models explain recency as a matter of temporal context: the most recent items were encoded in a context most similar to the context at the moment of recall, so they are the easiest to retrieve, at whatever time scale. On that view the filled delay works by changing the context rather than by emptying a store. The curve and the effect of the delay are agreed on; the mechanism is still discussed.

In judgment rather than recall, recency names a different but related pattern: when evidence arrives in sequence, the latest piece often moves the final judgment most. Hogarth and Einhorn's 1992 belief-adjustment model predicts when this happens, mainly when items are evaluated one at a time and the sequence is short, and when a primacy effect appears instead, mainly when a judgment is formed once at the end from a long series.

Murray Glanzer and Anita Cunitz published "Two storage mechanisms in free recall" in the Journal of Verbal Learning and Verbal Behavior in 1966. The paper set out to show that the two ends of the serial-position curve come from two different memory stores, by finding one manipulation that moved only the beginning of the curve and another that moved only the end.

The first experiment varied the rate at which words were presented and whether each word was repeated. Slower presentation raised recall of the first words and the middle of the list and left the last words alone, which fit the idea that the beginning of the curve depends on rehearsal into long-term storage. Repetition did not add anything beyond the extra time it took.

The second experiment is the one the demo follows. Forty-six Army enlisted men were shown fifteen lists of 15 common one-syllable nouns, each word shown for one second with two seconds between words and read aloud by the experimenter. After the last word, either a symbol appeared and the man began writing at once, or a digit appeared and he counted aloud from it until told to write, for 10 or 30 seconds. Each man had five lists in each delay condition, in random order. A pilot study had already shown that an empty 30-second delay reduced the end peak only a little, which is why the counting task was added.

With immediate recall the curve had its usual shape, a raised start and a steeply raised end. A 10-second delay filled with counting removed most of the end peak. A 30-second delay left no trace of it, and the men recalled the last words no better than the middle ones. The start of the curve did not move in either delay condition. A year earlier, in 1965, Postman and Phillips had reported a parallel result with a short counting task after lists of different lengths.

Outside the laboratory the word covers both the memory advantage of recent items and the extra weight of recent evidence. The rows below are illustrations, not findings from the studies cited on this page.

The recall effect can be used, and the judgment effect can be planned around.

• When something must be remembered for more than a few seconds, write it down or rehearse it. Short-term memory holds the last items only until something else occupies it, which is what the counting in the demo did.
• When you are presenting, put the most important point first and repeat it last. If a listener will be judging right after you finish, the end is the best position; if they will judge later, the beginning is.
• When judging a sequence of people, options or results, record a judgment on each one as it arrives. Deferred judgments weight whatever came last.
• Rotate the order where fairness matters: interview slots, auditions, closing statements. Where the order cannot be changed, record it, so that ratings can be checked against position afterward.
• For records that cover a long period, read the record before forming a view. The most recent quarter, month or incident will come to mind first whether or not it is representative.

The recency effect is one end of a curve and one sense of a word that is used loosely, so its neighbors matter.

• It is not the whole serial-position effect, which is the U-shaped curve made of a raised start and a raised end. Recency is the raised end on its own, and it is the half of the curve that a short filled delay removes.
• It is not the primacy effect, the raised start, which on the classic account comes from rehearsal into long-term memory and survives the delay that removes recency. The two respond to different manipulations, which is the reason they are given different names.
• It is not the peak-end rule, which concerns how an experience is evaluated afterward, from its most intense moment and its ending. The peak-end rule is about the pleasantness of an episode; the recency effect is about which items are recalled or which evidence weighs most.
• It is not the availability heuristic, judging frequency by how easily examples come to mind. Recent examples are often the available ones, so the two overlap in practice, but availability also runs on vividness and personal relevance, with no sequence involved.