Peak-End Rule: Why the Worst and Last Moments Decide

An experience that lasts is lived moment by moment and remembered as a summary, and the summary is not a total. When people rate a past episode of discomfort, their ratings are predicted well by two numbers: how bad the worst moment was and how bad the last moment was. The length of the episode, which would dominate any sum of the discomfort, barely registers. That is the finding, and duration neglect is the name for the missing piece.

Varey and Kahneman showed this with described episodes in 1992. Participants read series of discomfort ratings said to have been made at five-minute intervals by someone undergoing an unpleasant experience, then judged the episode as a whole. An unweighted combination of the peak rating and the final rating accounted for 94 percent of the variance in those judgments, and adding duration raised that by only 3 percentage points. The consequence is a reversal that looks like an error in arithmetic: the series 2-5-8-4 was judged much less aversive than the series 2-5-8, although the first contains the second plus five more minutes of discomfort.

Why memory works this way is not settled, but the shape of the explanation is agreed. Memory keeps representative moments rather than totals. A peak is representative because it is extreme; an ending is representative because it is where things landed. Adding the moments up is work that nothing in memory does automatically. The distinction that follows is between the discomfort as it is lived and the discomfort as it is summarized afterward: they are different quantities, and choices about the future are made from the second one.

The rule is a good description, not a law. In the 1993 cold-water experiment the majority who preferred more pain was 69 percent rather than everybody, and 11 of the 32 participants reported almost no relief during the extra 30 seconds, which is the part the effect depends on. Kemp, Burt and Furneaux tested the rule on extended autobiographical events, such as vacations, and found it was not an outstandingly good predictor of how those were remembered. Duration is neglected, not ignored: it reappears when an episode is long enough for the length itself to become the story.

Daniel Kahneman, Barbara Fredrickson, Charles Schreiber and Donald Redelmeier published "When more pain is preferred to less: Adding a better end" in Psychological Science in November 1993. Thirty-two male students at the University of California, aged 19 to 39, were paid for a one-hour session and told the study concerned differences between the hands in the experience of discomfort.

Each participant had two experiences. In the short trial, one hand was immersed for 60 seconds in water held at 14.1 degrees Celsius. In the long trial, the other hand was immersed at the same temperature for 60 seconds and then kept in the water for 30 seconds more while the temperature was raised to about 15.2 degrees, still uncomfortable but less so. Throughout each trial participants reported discomfort on a dial that ran from 0 to 14. The order of the trials and the hand used were counterbalanced, and seven minutes of filler work separated them.

The real-time records showed that the two trials were the same while they overlapped: mean discomfort at the 60-second mark was 8.44 in the short trial and 8.34 in the long one. The gradual warming at the end of the long trial produced a clear drop in the dial. Then participants, who expected a third immersion, were asked which of the two trials they would choose to repeat. Twenty-two of the 32, or 69 percent, chose the long trial, the one that included every second of the short trial and 30 seconds of additional cold.

The same group took the finding into a hospital. Redelmeier and Kahneman recorded real-time pain every 60 seconds from 154 patients undergoing colonoscopy and 133 undergoing lithotripsy in 1996, and found that retrospective judgments of total pain tracked the peak intensity and the intensity of the final three minutes, while procedures that ran much longer, from 4 to 67 minutes, were not remembered as worse. In 2003 Redelmeier, Katz and Kahneman ran a randomized trial with 682 colonoscopy patients in which half had a short, deliberately uneventful interval added at the end. Those patients rated the final moments as less painful, 1.7 against 2.5 on a ten-point scale, and the whole experience as less unpleasant, 4.4 against 4.9. Over a median follow-up of 5.3 years, return rates for a repeat colonoscopy averaged 50.4 percent and were slightly higher in the extended group, with an odds ratio of 1.41.

The rows below are illustrations of where a remembered summary stands in for an experience, not findings from the studies cited on this page.

The rule describes how memory summarizes, so there are two jobs: making better decisions from your own memories, and designing experiences honestly.

• Record while it happens. A note at the time, or a photograph, is the only record that keeps duration in it.
• When choosing whether to repeat something, ask how long it was, not only how it ended. The length is the part your memory left out.
• Separate the questions when you collect feedback. How was the ending and how long did it take are different measurements, and only one of them survives in a single satisfaction score.
• Design endings deliberately. The 2003 colonoscopy trial shows that a better ending genuinely improves the memory of a procedure, which is useful for anything a person has to come back to.
• For your own hard stretches, taper rather than finishing at the worst point. Nothing about the earlier work is lost, and the memory you keep is the one you will act on.

The peak-end rule is about the retrospective evaluation of an extended experience. Several neighbors sound similar and measure something else.

• It is not duration neglect on its own, which is the missing half of the same finding: how long an episode lasted barely affects how bad it is remembered as being. Peak and end describe what does the work instead.
• It is not rosy retrospection, the tendency to recall the past as better than it was rated at the time. That is a shift in the overall tone of a memory; this is a rule about which moments the memory is built from.
• It is not the serial-position effect or the recency effect, which are about which items on a list are recalled. Those concern remembering items; this concerns evaluating an experience.
• It is not the impact bias, which is about predicting how strongly a future event will affect you. Peak-end looks backward at an episode already lived.