Evidence audit
Thinking, Fast and Slow, Rebuilt on the Research
Audited at the level of the book's five Parts. The core work that won a Nobel Prize holds. One chapter did not — and the person who said so first was Kahneman himself.
Published 8 July 2026 · Last checked 14 July 2026 · 5 chapters audited
The quick verdict
The Nobel Prize-winning core — anchoring, framing, the planning fallacy, prospect theory — survived the replication crisis and is still the best guide to how judgment goes wrong. The borrowed social psychology did not survive: Kahneman himself later wrote of the priming chapter, "I placed too much faith in underpowered studies." Read the decision-making parts as science; read Chapter 4 as retracted.
3 chapters hold up strongly · 2 mixed or overstated
Source book: Thinking, Fast and Slow · Daniel Kahneman (2011)
Daniel Kahneman's Thinking, Fast and Slow runs to 38 chapters, so we audit it at the level of its five Parts rather than inventing granularity the book doesn't reward. This is the most distinguished book in this section and it gets the most respectful audit — but also the most striking one, because its most important correction was issued by the author. In 2017 Kahneman publicly accepted that the priming research in Chapter 4 was far weaker than he had claimed, writing: "I placed too much faith in underpowered studies." The judgment-and-decision-making core of the book — the work with Amos Tversky that won the Nobel Prize — has largely survived. The borrowed social psychology has not. A research-grounded companion to the book, not a replacement for reading it.
How we audited this book
We kept the book's own chapters and checked every claim underneath against the original studies. Each chapter gets one of two verdicts:
- Holds up
- The core idea is backed by findings that different research teams have confirmed independently. Safe to build habits on.
- Mixed
- A real effect exists, but it's smaller, more conditional, or more contested than the popular version suggests.
- ⚑ Where the book overstates it
- Marks the specific place the book's language runs ahead of the science — with the correction right next to it.
The research terms, in plain language
- Meta-analysis:
- One study that pools all previous studies on a question — the closest thing science has to a final answer.
- Replication:
- Re-running a study to see if the result appears again. Many famous findings failed this test in the 2010s.
- Correlation (r or ρ):
- How tightly two things move together, from 0 (no link) to 1 (lockstep). In psychology, .18 is weak, .50 is strong, and .84 means two questionnaires are nearly measuring the same thing.
- Effect size (d):
- How big a difference an intervention makes. Rough guide: 0.2 is small, 0.5 medium, 0.8 large — so d = 0.08 is close to nothing.
- Variance explained:
- The share of the differences between people that one factor accounts for. “1%” means the other 99% is something else.
- Pre-registered:
- The analysis plan was locked in before the data was collected, so researchers can't quietly pick the flattering result.
Chapter by chapter
- 01System 1 and System 2 — a Useful Fiction, and the Author Said So
- 02Heuristics and Biases — the Nobel Core, and It Holds
- 03Overconfidence — Strong, With One Famous Study Now in Doubt
- 04Choices — Prospect Theory Stands, Loss Aversion Is Being Renegotiated
- 05Two Selves — and Chapter 4, Which the Author Retracted
holds up · mixed or overstated
System 1 and System 2 — a Useful Fiction, and the Author Said So
In the book: “Part I: Two Systems”
The book's organising metaphor: a fast, automatic System 1 and a slow, effortful System 2. Enormously useful for thinking about thinking — and, as Kahneman explicitly warns in the book itself, not two actual things in your head.
What the research says
Kahneman's own caveat
He states plainly that System 1 and System 2 are fictional characters, not brain structures — a narrative device to make cognitive processes memorable. Credit where due: the book is honest about this, and most people who quote it are not.
Dual-process theory in cognitive science
The broad distinction between automatic and controlled processing is well established. The clean two-system architecture is contested, and many cognitive scientists regard it as an oversimplification rather than a model of how cognition is actually organised.
Ego depletion — 'The Lazy Controller' (Chapter 3)
⚠️ Failed. The claim that System 2 runs on a depletable glucose-like resource rests on ego depletion, which was tested in a multi-lab pre-registered replication (Hagger et al., 2016, ~two dozen labs) and produced an effect estimated at approximately **zero**. The 'mental energy tank' framing in this Part did not survive.
⚑ Where the book overstates it
Two things in Part I have aged badly: ego depletion (failed replication) and the glucose-restores-willpower story attached to it. The two-systems metaphor itself is fine as long as you hold it as loosely as Kahneman told you to.
Do this
Use the metaphor where it earns its keep: on important decisions, deliberately slow down and write the reasoning out. Don't believe you have a willpower fuel gauge that a biscuit refills — that specific claim is dead.
Heuristics and Biases — the Nobel Core, and It Holds
In the book: “Part II: Heuristics and Biases”
This is the material Kahneman and Tversky built over thirty years, and it is the reason the book matters. When the replication crisis came for psychology, this is the part that stood up.
What the research says
Anchoring — replicated (Many Labs, Klein et al., 2014)
An arbitrary number you've just seen contaminates your subsequent numerical judgement — even when you know it's arbitrary. Anchoring replicated consistently and with a substantial effect in the large multi-lab replication project. This one is solid.
Framing — replicated at roughly half strength (Many Labs, 2014)
The same choice described as '200 saved' versus '400 will die' produces different decisions. The effect reproduced — but at about **half** the size of the original. Real, and smaller than the book implies.
Availability and representativeness (the Linda problem, base-rate neglect)
The conjunction fallacy and base-rate neglect are robust and have been demonstrated repeatedly across decades and populations. This is genuinely established science.
Do this
Anchoring is the one to actually use. In a salary negotiation, whoever says a number first moves the whole conversation — so know your number, and don't let an employer's first offer become the anchor you argue down from.
Overconfidence — Strong, With One Famous Study Now in Doubt
In the book: “Part III: Overconfidence (chapters 19–24)”
That we construct confident stories from too little information, and systematically underestimate how much we don't know, is well supported. One celebrated illustration in this neighbourhood has since been undermined.
What the research says
The planning fallacy and the illusion of validity
People consistently underestimate how long tasks will take and overestimate the reliability of their own judgement — and experts are often worse, not better. Well replicated, and among the most practically useful findings in the book.
The 'hungry judges' parole study (Danziger et al., 2011)
⚠️ Seriously contested. The famous result — parole grants around 65% at the start of a session, falling to near zero before a meal break — depends on the assumption that cases arrive in random order. Weinshall-Margel and Shapard showed case ordering is **not** random: boards finish all cases from one prison before breaking, and unrepresented prisoners (much less likely to get parole) tend to go last. Control for the scheduling and most of the 'hunger' effect disappears.
⚑ Where the book overstates it
The hungry-judges study is quoted everywhere as proof that decisions are hostage to blood sugar. Its central assumption was wrong. The overconfidence research around it is fine — this particular headline is not.
Do this
Assume your plan will take longer than you think, because it reliably will. When you estimate your job search will take a month, plan for three and you'll be close to right.
Choices — Prospect Theory Stands, Loss Aversion Is Being Renegotiated
In the book: “Part IV: Choices”
Prospect theory is the work that won the Nobel Prize in Economics, and it remains one of the great achievements of behavioural science. Its most quotable component is nonetheless under active challenge.
What the research says
Prospect theory (Kahneman & Tversky, 1979)
People evaluate outcomes as gains and losses relative to a reference point, not as absolute states of wealth, and are risk-averse for gains and risk-seeking for losses. This reshaped economics and remains well supported.
Gal, Rucker & Shavitt, 2018 (Journal of Consumer Psychology) — 'The Loss of Loss Aversion'
⚠️ Contested. The familiar claim that losses hurt about **twice** as much as equivalent gains please is challenged: the authors review the evidence and argue that losses do not, on balance, reliably loom larger than gains, and that the 2-to-1 ratio is not a universal law. This is a live scientific dispute — other researchers (e.g. Mrkva et al., 2020) defend loss aversion under specified conditions. Treat the 2:1 figure as contested, not settled.
⚑ Where the book overstates it
Loss aversion is quoted with a precision it no longer deserves. The reference-point insight is robust; the specific 'losses hurt twice as much' multiplier is the subject of an open and serious argument.
Do this
Watch the reference point in your own job decisions. A Rs. 40,000 offer feels like a loss if you were hoping for 50,000, and a win if you were expecting 30,000 — the offer is identical. Decide on the number, not on the gap from your hope.
Two Selves — and Chapter 4, Which the Author Retracted
In the book: “Part V: Two Selves”
The final Part — the experiencing self versus the remembering self — is a genuinely original contribution. But this is the place to record the correction that defines this book's afterlife, and it is one Kahneman made himself, with a candour almost no author manages.
What the research says
The peak-end rule and duration neglect
We judge past experiences by their most intense moment and their ending, largely ignoring how long they lasted — demonstrated in Kahneman's own experimental work, including the ice-water (cold-pressor) and colonoscopy studies. Holds up reasonably well, and it is a real insight.
Kahneman, 2017 — his response to the Replicability-Index analysis of Chapter 4
✅ Verified, and remarkable. After Schimmack and colleagues analysed the social-priming studies cited in Chapter 4 ("The Associative Machine") and estimated the average chance of a successful replication at around **14%**, Kahneman replied: "What the blog gets absolutely right is that I placed too much faith in underpowered studies." He added that "the experimental evidence for the ideas I presented in that chapter was significantly weaker than I believed when I wrote it. This was simply an error."
What the book had said about that same research
⚠️ The sentence that has not aged well, from the book itself: "disbelief is not an option. The results are not made up, nor are they statistical flukes. You have no choice but to accept that the major conclusions of these studies are true." That was written about the social-priming literature — much of which has since failed to replicate.
⚑ Where the book overstates it
This is the most instructive episode in this entire section, and it is worth more than any single statistic. A Nobel laureate told his readers they had no choice but to believe a body of research. He was wrong, the research was underpowered, and — when the evidence came in — he said so publicly and without excuses. That is what intellectual honesty looks like, and it is exactly why the rest of the book still deserves your trust.
Do this
Take the lesson the author himself demonstrated: when good evidence contradicts something you've said confidently, change your mind out loud. That is a better career skill than anything in the book — and, incidentally, the reason this section exists.
Questions people ask
Is Thinking, Fast and Slow still worth reading?
Yes — with a map. The judgment-and-decision-making core (anchoring, framing, overconfidence, prospect theory) held up when psychology's replication crisis hit. What fell was mostly borrowed: the social-priming studies of Chapter 4, the "mental fuel tank" theory of willpower, and the famous hungry-judges story. The book's own author flagged the biggest correction himself.
Did Kahneman really retract part of his book?
In effect, yes. In 2017, after an independent analysis estimated the priming studies in Chapter 4 had roughly a 14% chance of replicating, Kahneman publicly replied: "I placed too much faith in underpowered studies… This was simply an error." The book had told readers they had "no choice but to accept" those findings.
Are System 1 and System 2 real parts of the brain?
No — and Kahneman says so in the book itself: they're fictional characters, a storytelling device for automatic versus effortful thinking. The broad distinction between the two kinds of processing is well established; the clean two-system machinery is not. Use the metaphor for what it's for: a reminder to slow down and write out your reasoning on decisions that matter.
Which ideas from the book should I actually use?
Anchoring and the planning fallacy. Whoever says the first number in a salary negotiation moves the whole conversation — know your number before you walk in. And plans reliably run long: when you estimate your job search will take a month, plan for three. Both effects replicated robustly in large multi-lab projects.
Compiled as a research-grounded companion to the book — not a replacement for reading it. Citations are author/year pointers to real, findable studies rather than formatted references; verify specifics before relying on them for high-stakes decisions.