← Thinking, Fast and Slow
第 6 章 · 20 分钟
Framing, mental accounting and the two selves
From judgement to evaluation: who is scoring the episode.
概念地图
Framing
Equivalent descriptions change choices because they change the baseline.
深色为本章概念，浅色为其他章
陪练模式
不想只是读？让它带你走一遍。
导师把这一章拆成小步，每一步先问你，再讲；你用自己的话答，它先认下对的部分再纠偏。用你自己的 AI，对话只存在这台设备上。
Locus01
延伸阅读
Tversky and Kahneman · 1981 · the framing of decisions
Used the Asian disease problem to show equivalent descriptions changing choices, against treating preference as a stable order independent of description. The cost is that in every such experiment the frame is set by the experimenter, so it says nothing about where frames come from in natural settings.
Foundational
Thaler · 1980s · mental accounting
Extended the reference-point idea to how money is sorted: the same sum is booked to different accounts by source and purpose and treated differently. It carried prospect theory from single choices into everyday financial behaviour, at the cost that the accounts have no criterion independent of the behaviour.
Turn
Redelmeier and Kahneman · 1990s · pain ratings in colonoscopy
Recorded momentary pain and retrospective ratings during a real medical procedure and found the retrospective rating set by peak and end, and barely by duration. Rare in this book for having both real-time measurement and real consequences.
Revision
Kahneman · 2011 · the chapters on the two selves
Draws from this the distinction between the experiencing and the remembering self and concludes that we choose for memory rather than for experience. The book states that normative conclusion more strongly than the evidence allows; the third crack lands there.
Subject of this course
机制02
Separate accounts
The same money labelled 'bonus' rather than 'salary' gets spent quite differently; lose a ticket to a show and most people will not buy another, lose the same amount in cash and they buy anyway. Thaler named this mental accounting. Its place in the argument is as chapter 5's reference point generalised: each account is its own baseline, and money is not fully fungible across them.
机制The same sum of money, through being sorted into separate accounts, is evaluated against each account's own baseline rather than against the total.
可迁移性测试
Move it to engineering capacity: the same engineer-day is guarded when booked to 'new features' and spent freely when booked to 'tech debt', though both come out of one team's capacity. The isomorphism breaks in that an engineering budget is written down and can be challenged, while a mental account never is.
机制03
Peak-end bookkeeping
After an episode ends, what is retained is not the sum of moment-by-moment feeling but mainly the most intense moment and the final one. This is the peak-end rule. Its implication is hard: lengthening an unpleasant episode while softening the ending improves the retrospective rating, even though more momentary discomfort was experienced. The derivation and the test below are both built on that implication.
机制An episode, through entering memory only as its peak and its ending, has its overall rating decoupled from its duration.
可迁移性测试
Move it to a software incident: a two-hour outage that recovers smoothly in its last ten minutes is rated better in the post-mortem than a forty-minute one that ends abruptly. The isomorphism breaks in that an incident has an objective impact duration on record, while the duration of a personal experience is available only by self-report.
Derivation04
From peak-end bookkeeping to 'a longer pain is rated milder'
What is being derived is the least intuitive and most testable prediction of this mechanism.
The retrospective rating is a function of peak intensity and end intensity
Chosen, and the heaviest line: it puts duration outside the rating entirely.
Cumulative momentary discomfort is the integral of intensity over time
Chosen, and the operational definition of the experiencing self.
Appending a stretch of lower intensity does not raise the peak
Chosen: the appended stretch must be below the original peak, or the peak moves.
推导 · 0 / 4
裂缝05
Cracks in this framework
争议地形06
本书主张
The remembering self and the experiencing self conflict systematically: we make choices by peak-end bookkeeping and so keep choosing episodes that sound better in recollection and are harder to live through.
另一种看法
Another line holds that memory evolved to guide future choice, so encoding by peak and end is a compression useful for the future rather than an error. Using the integral of momentary feeling as the correct answer is a norm imposed by the analyst.
分歧扎在
The disagreement is rooted in which measure counts as correct for an episode, not in the data. Both accept the peak-end findings.
什么证据能裁决
What would settle it is predictive validity: compare how well each measure — the retrospective rating and the integral of momentary feeling — predicts whether the person repeats the episode, and how they actually feel on the repeat. If the rating predicts better, it is a useful compression rather than a distortion.
The evidence points to the rating predicting repeat choice better and actual re-experienced feeling worse: the two measures explain different things. What is missing is a combined utility measure — current practice forces a choice between them, so the argument cannot converge.
Falsification07
Peak-end bookkeeping as a preregistrable test
The mechanism under test: an episode, through entering memory only as its peak and its ending, has its overall rating decoupled from its duration.
Data source: newly collected real-time intensity series with retrospective ratings, intensity self-reported at fixed intervals
Both momentary data and the rating are needed, or duration cannot be separated from peak and end.
Sample: a pre-computed sample size analysed in full, with no exclusion on 'reporting consistency'
Consistency checks correlate with reporting style, so post hoc exclusion introduces selection.
Window: a single episode, with the appended stretch fixed in advance as a proportion of the original
The proportion is fixed so the appended length cannot be adjusted after seeing results.
Threshold: controlling for peak and end intensity, the partial coefficient of total duration on the rating lies inside a pre-registered equivalence interval
The claim being tested is 'no effect', so this is an equivalence test rather than a non-significant result.
Failure condition: the duration coefficient falls outside the equivalence interval, or peak and end are not significant predictors
Either counts as refutation.
推导 · 0 / 3
接口08
Which slot it hangs on
挂在哪个槽位The conclusion you already hold is probably 'how good an episode was is how it went overall' — the rating equated with accumulated feeling. This chapter goes at that equation.
Does this chapter (a) replace 'the rating equals accumulated feeling', or (b) constrain it to cases where duration is explicitly recorded?
慢变量Register two slow variables: how many stages of your process have had their ending deliberately designed, and whether you hold any record of momentary experience at all. Look at the second first — without a record, all you ever get is the rating.
小结09
本章小结
01Mental accounting generalises the reference point: money is not fully fungible across accounts.
02'Appending a low stretch improves the rating and raises cumulative discomfort' is an arithmetic consequence of the peak-end assumption.
03The step carrying empirical risk is that lengthening with a gentler ending improves ratings in a real procedure.
04Duration neglect depends on duration not being explicitly recorded; a clock defeats it.
05Which self to decide for is a normative question that data cannot settle.
提取练习 · 合上书，先自己答一遍。
?Is 'the appended stretch improves the rating while raising cumulative discomfort' (a) an identity or (b) an empirical claim?
?Is 'the remembering self overriding the experiencing self is an error' (a) refutable by data or (b) a normative claim needing a stipulated measure?
Forced choice10
C · Locate the crack
In which situation does this framework give a confident and wrong answer?
二选一
Forced choice11
D · Judge the interface
Which one hangs on a slot in your existing structure?
二选一
读到这里，把它记为已读。