i

Description

The Eurovision Song Contest stage

Source

MrSilesian · original page

Licence

CC0 Public domain

Original photo, unretouched. The background is a version of it stylised by the site.

The recency effect

We judge a year, a project, or a person mostly by their last few weeks.

In brief

• What : the most recent information has disproportionate weight in judgment or memory.
• Why : what just happened is still in working memory, so it's more accessible than the rest.
• Detect : an evaluation that cites only facts from the last two or three months; a project retrospective centered on the final phase.
• Avoid : take notes continuously, review the entire period, evaluate criterion by criterion, date the facts.

On this page

1. Definition
2. The mechanism
3. Where you encounter it
   1. In project management
   2. In management
4. Famous cases
   1. Lengthening an exam so it's remembered better 2003
   2. Going last gives an advantage at Eurovision 2005
5. Detection signals
6. Avoidance reflexes
7. With AI: detect, correct, mitigate
   1. Detect
   2. Fix
   3. Mitigate
   4. Prompts to integrate into your practices
   5. How to integrate them
   6. Precautions
8. Limitations
9. Sources and verification
10. See also
    1. Combines with, or reinforces
    2. Related biases
    3. Counterweights and related reflexes

Definition#

Fictitious illustration

The recency effect is the tendency to give more weight to the most recent information or events than to those that came before. It comes from memory psychology: when people are asked to remember a list of words, the ones at the end are recalled best. This serial position has been studied since the 1960s, notably by Bennet Murdock (1962).

In the workplace, the same phenomenon distorts judgments made about a period: the annual review is mostly a recap of the last quarter, a project retrospective is mostly about acceptance testing, and a two-hour meeting gets summed up by its last ten minutes.

The mechanism#

• A short-term store. In memory models, the last items perceived are still available in working memory at recall time. Glanzer and Cunitz (1966) showed this: a short 30-second filler task makes the advantage of the last items disappear, without erasing the advantage of the first ones.
• Accessibility mistaken for importance. What comes to mind easily is judged representative. That's the mechanism of the Availability heuristic.
• The ending sets the tone of memory. Kahneman and colleagues (1993) described a peak-end rule: memory of an experience is shaped mostly by its most intense moment and by its ending, and very little by its duration.
• Lack of records. Without notes taken continuously, the evaluator has only memory left—and memory favors the recent.

Where you encounter it#

In project management#

• Retrospective : the end-of-project retrospective talks mostly about acceptance testing and go-live, because that's what the team remembers. Architecture decisions made ten months earlier—often the source of the difficulties—fall through the cracks.
• Milestone review : one failed sprint right before the committee overshadows six steady sprints; the reverse is also true: a strong final demo masks a drift.
• Judgment of a vendor : the latest support incident, or the latest quickly delivered fix, makes people forget the overall quality of the service.
• Review of a meeting or workshop : minutes retain the points covered at the end of the session, when attention is already low.

In management#

• Annual review : an incident in November weighs more than a year of solid work; a good December delivery makes up for a difficult first half. Studies on performance evaluation (for example Steiner and Rain, 1989) have observed order effects depending on how information is presented and the delay before judgment.
• Evaluation of candidates or presentations : in singing competitions, performance order influences scores. Bruine de Bruin (2005) analyzed the Eurovision contest and found an effect of performance order on points awarded by juries.
• Feedback : people react to what just happened without placing it back in the person's trajectory.

Famous cases#

Detection signals#

• All examples cited in an evaluation date from the last two or three months.
• “This year, he really…” followed by a November event.
• The yearly score is very close to the impression from the past two weeks.
• A project retrospective covers the final phase at length and the design phase in a few lines.
• Evaluators prepare their reviews the day before, from memory, without notes.
• A ranking result depends on the order in which files were reviewed.

Avoidance reflexes#

1. Keep a logbook. One dated line per notable event, throughout the year: delivery, difficulty, customer feedback, progress. The goal is to replace memory with a file.
2. Review the whole period. Before scoring, reread meeting notes, tickets, code reviews, and goals set at the beginning of the year, month by month.
3. Break the period into chunks. Score each quarter separately, then compare: the trend becomes visible, and the last quarter is only one fourth of the total.
4. Evaluate criterion by criterion. For each criterion, find one fact from the whole year before moving to the next.
5. Address the fact when it happens. Feedback given within the week prevents an old incident from resurfacing in the review or an old success from being forgotten.
6. Change the order. For candidates, proposals, or back-to-back demos, reverse the order from one session to the next and score before discussing.
7. Polish the ending without abusing it. Ending a meeting or project with a clear, positive conclusion is legitimate; judging someone on that conclusion alone is not.

With AI: detect, correct, mitigate#

Prompts to integrate into your practices#

1. Timeline of the period. On anonymized follow-up notes (names removed).

Here are my follow-up notes for a year, dated and anonymized: [notes].
Build a timeline by quarter. For each quarter: facts, achievements, difficulties, and decisions, with the date of each item.
Then state the number of facts per quarter and flag quarters where I noted almost nothing, so I can look for evidence before judging.
Do not make any judgment about the person; do not fill in gaps.

2. Recency check of an evaluation. Before submitting a written evaluation.

Here is the text of an evaluation (anonymized): [text].
List each cited fact or example and its date, or “date not specified.” Calculate the distribution by quarter.
Say whether more than half of the examples come from the last three months. If so, for each criterion, suggest the questions I should ask myself to retrieve older facts.

How to integrate them#

• Throughout the year: note one fact per week, even briefly; prompt 1 turns it into a timeline when needed.
• Before the interview: run your text through prompt 2; revise, then write the final assessment.
• For successive rankings: request a random presentation order for each evaluator.

Precautions#

• Personal data: do not send names or details that could identify a person; check that the tool is authorized by your organization.
• AI summarizes, it does not judge: in Europe, evaluating employees with AI falls under high-risk systems (Regulation (EU) 2024/1689). The decision remains human and reasoned.
• For a long document, check that the middle has not been neglected: work in sections.
• It may invent a missing fact: ask it to answer “not documented” when there is no trace.

Limitations#

• Recency can be legitimate: a lasting behavior change, a recent recovery, or a recent decline contains real information about the current situation. The bias is forgetting everything else, not taking the present into account.
• Results vary by context: in memory experiments, recency disappears with an intervening task; in social judgments, primacy often dominates. Both effects coexist and depend on delay, attention, and the presence of notes.
• Performance evaluation studies in companies are less clear-cut than laboratory experiments: the effect varies by conditions (Steiner and Rain, 1989), and one may assume—without these studies establishing it—that trained evaluators or regular notes reduce it.

Sources and verification#

• Murdock (1962), “The serial position effect of free recall”, Journal of Experimental Psychology — the serial position curve.
• Glanzer & Cunitz (1966), “Two storage mechanisms in free recall”, Journal of Verbal Learning and Verbal Behavior — recency disappears after an intervening task.
• Kahneman, Fredrickson, Schreiber & Redelmeier (1993), “When more pain is preferred to less: Adding a better end”, Psychological Science — the peak-end rule.
• Redelmeier, Katz & Kahneman (2003), “Memories of colonoscopy: a randomized trial”, Pain 104 — 682 patients.
• Steiner & Rain (1989), “Immediate and delayed primacy and recency effects in performance evaluation”, Journal of Applied Psychology 74 — order effects in performance evaluation.
• Bruine de Bruin (2005), “Save the last dance for me: Unwanted serial position effects in jury evaluations”, Acta Psychologica 118 — Eurovision.
• Liu et al. (2024), “Lost in the Middle: How Language Models Use Long Contexts”, Transactions of the ACL — attention to the ends of a long context.
• Regulation (EU) 2024/1689 on artificial intelligence — Annex III: employment and worker management.

See also#