Evidence status: relatively stable Peaks and endings reliably influence retrospective global evaluation, with a 2022 meta-analysis providing strong cumulative support. Duration neglect also has substantial evidence. The popular claim that memory “only” uses the peak and end is an exaggeration: average experience, meaning, narrative, expectation and the timing of recall still matter. This is a version 0.1 research draft.
How Everyday Psychology Takes Shape · Season One: Why Do We Think This Way? · Article 3
A holiday consists mostly of transfers, queues and bad weather. The final day is unexpectedly clear. Months later, photographs and stories increasingly centre on that day, and the trip becomes “actually quite good”. In the other direction, a satisfying dinner ends with a dispute over the bill, and the ending darkens the whole episode.
Everyday intuition might say that overall evaluation should work like an average: add the quality of every moment and account for duration. Psychological memory does not preserve a lossless time series. When people look back, the most intense episode and the period near the ending often receive disproportionate weight. This is known as the peak–end effect or peak–end rule.
The word rule can imply that only two moments survive. The evidence supports unequal weighting, not a memory made solely of peak and ending. That difference determines whether the concept remains a useful psychological finding or becomes a marketing formula for manipulating experience.
Two meanings of “how good was it?”
Evaluation during an experience and evaluation after it are different tasks. The first asks how a person feels now and can be sampled repeatedly. The second compresses a temporally structured event into one global judgment.
Barbara Fredrickson and Daniel Kahneman’s 1993 work asked participants to view pleasant and unpleasant film clips varying in length and intensity. Participants reported ongoing affect and later rated the clips globally. Duration had little direct influence on retrospective evaluation after its association with momentary affect was considered, while feelings near the end mattered.
A cold-water experiment published in the same year produced a particularly counterintuitive choice. Participants experienced a shorter trial and a longer trial that repeated the same discomfort before adding a period that was slightly less cold but still unpleasant. Many chose to repeat the longer version. Additional discomfort did not reduce the cumulative pain, but the improved ending produced a better remembered evaluation.
This violates a temporal-monotonicity intuition that adding pain must make an episode worse. It cannot be used to justify adding pain. The result shows that retrospective choice can depart from cumulative experience. Ethical design should be alert to that departure rather than exploit it.
What the meta-analysis supports
In 2022, Balca Alaybek and colleagues synthesised 174 effect sizes. They reported a large association between peak–end information and retrospective global evaluation, approximately r = .58, robust across the boundary conditions they examined. Peak–end information was more predictive than the beginning, trough and duration. Its predictive strength was comparable to the overall mean of the experience.
The last comparison is crucial. Popular accounts often say people forget the average and remember only peak and end. The meta-analysis does not support that absolute replacement. The average remained a powerful predictor. Peak and ending provide an efficient compression, not the only route by which an experience informs judgment.
“Peak” also varies across designs. In a negative episode it may mean greatest pain; in a positive episode it may mean greatest pleasure. Whether the experience is continuous or divided into episodes, whether people expect a later evaluation, and how the outcome is measured can matter. A multiday life event carries narrative, goals and social meaning that a short laboratory stimulus cannot reproduce fully.
The cumulative evidence is therefore strong at the level of uneven retrospective weighting. It is weaker as a universal numerical recipe for predicting every memory.
Why might endings receive special weight?
One account begins with accessibility. Material near the end has less later interference and may be easier to retrieve during a global evaluation. An ending also tells a person where events led: whether danger passed, a goal was achieved, or conflict was repaired. It can change the meaning of earlier episodes as well as supply the final moment.
Peaks have salience. The most intense period may become a prototype for the event, giving later memory a representative scene. Compressing a long sequence into a small amount of information lowers judgment costs.
These explanations can coexist. The peak–end effect need not arise from one dedicated “brain rule”. Memory accessibility, attention, narrative completion and decision strategy can all produce unequal weights.
The influence of an ending may also depend on whether the event feels complete. An arbitrary interruption and a meaningful resolution occupy the same temporal position but do different interpretive work. A useful explanation needs the structure of the experience, not merely a final data point.
Holidays, medicine and customer service are not the same experiment
A holiday extends across days and includes places, photographs, shared narration and later changes of meaning. Brief film or cold-water tasks cannot establish that arranging a good final day will make any trip successful. A terrible ending may dominate recall, but persistent serious problems do not disappear after a polite farewell.
Medical applications require greater caution. Improving comfort near the conclusion of a procedure may alter memory and willingness to return. This does not permit unnecessary prolongation of pain for the sake of a better end. The first obligations are effective treatment and harm reduction. Remembered experience is one important outcome among others.
Service designers often present the rule as a practical technique. A surprise at the end may improve satisfaction. If the preceding service involved unfairness, delay or a defective product, a decorative finale is not repair. Psychological evidence gives no automatic legitimacy to manipulative design.
The most defensible use is less theatrical: avoid an unnecessarily confusing or abrupt final stage, communicate what has been completed, and leave a genuine route for unresolved problems. This improves the actual ending rather than merely staging an impression.
Which self should count?
Kahneman later popularised a distinction between the experiencing self and remembering self. It is illuminating, but people should not be divided into two literally independent subjects. Momentary feeling, later memory and prospective choice belong to one person answering different questions.
Remembered global judgment affects whether a person repeats a holiday, treatment or job. It is not false memory to be discarded. Endings can contain recovery, completion and meaning, all of which genuinely belong to experience.
If institutions optimise only the later rating, however, moment-to-moment suffering loses representation. Care, employment and education should gather process and retrospective evidence rather than allowing one version to monopolise “the real experience”.
Duration is a moral as well as measurement issue. A person exposed to a lower-intensity burden for months may give a moderate global score while accumulating substantial cost. A summary rating cannot by itself settle whether the arrangement was acceptable.
A simple way to observe temporal weighting
For an important experience, separate several records: the most common state during it, the most intense moment, the ending, duration, and the global judgment several days later. If they differ, it is unnecessary to decide immediately which is fake.
The comparison can reveal what a future decision uses. If willingness to repeat depends mainly on the ending, process notes can be reconsidered. If the process was difficult but later judged worthwhile, ask whether the value came from achievement, relationship, purpose or memory compression.
Recording itself changes attention, so this is not neutral scientific measurement. It is a way to keep more than one time scale visible.
For routine pleasures, constant tracking can damage the experience it is meant to understand. The method is most useful when a consequential decision otherwise depends on one global recollection.
How much can an ending rewrite?
An experience is not a fixed object completely sealed when it ends. Memory reorganises episodes, and the ending provides a powerful cue for that organisation. It can change overall tone without cancelling every moment that occurred.
The version 0.1 judgment is that a good ending can improve memory for an entire process, while an intense peak often receives disproportionate weight. The evidence does not reduce global evaluation to the arithmetic average of peak and end. Understanding a consequential experience requires process, intensity, ending, duration and later meaning to remain visible together. An ending changes the story a person carries away. It cannot make the earlier cost cease to have happened.
Primary research sources
- Fredrickson, B. L., & Kahneman, D. (1993), “Duration Neglect in Retrospective Evaluations of Affective Episodes”.
- Kahneman, D. et al. (1993), “When More Pain Is Preferred to Less”.
- Alaybek, B. et al. (2022), “All’s Well That Ends (and Peaks) Well?”.
- Fredrickson and Kahneman paper, archival full text.
Series navigation: How Everyday Psychology Takes Shape — Season One overview
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.