How does "Update beliefs incrementally, not all at once" work?

Bayesian updating is proportional: strong evidence produces large updates; weak evidence produces small ones. Most people update either too little (anchoring) or too dramatically (representativeness: treating one vivid case as decisive). The proportional update — multiplying your prior by the likelihood ratio the evidence provides — is what the math actually prescribes, and it is almost never as large as an all-or-nothing swing.

How does "Evaluate evidence by its likelihood ratio, not by how it makes you feel" work?

Not all evidence is equal; its strength is determined by its likelihood ratio: how much more (or less) probable the evidence is in the hypothesis-true world versus the hypothesis-false world. Evidence that is equally likely under both hypotheses carries zero information. This framing cuts through the common mistake of treating "evidence I like" as strong evidence and "evidence I dislike" as weak.

How does "Practice calibration by tracking your confidence predictions" work?

Calibration is the match between stated confidence and actual accuracy. Most people are overconfident: they assign high probabilities to beliefs that are true far less often than claimed. Deliberate calibration practice — predicting, then tracking outcomes — has been shown to meaningfully improve the accuracy of confidence estimates over time, because it provides the feedback loop that intuition normally lacks.

How does "Actively seek evidence that would disconfirm your belief" work?

Confirmation bias causes people to search for, notice, and remember evidence consistent with existing beliefs and to ignore, forget, or discount contradictory evidence. The Bayesian antidote is deliberate disconfirmation search: actively seeking the evidence that would be most likely in the hypothesis-false world. This is cognitively uncomfortable precisely because it works.

How does "Use reference classes to ground personal estimates in base rates" work?

People generate predictions primarily from inside view — their own model of the specific case — which is systematically optimistic because it ignores how often similar plans fail in general. Reference class forecasting corrects this by grounding the prior in outside view: the statistical distribution of outcomes for similar cases. The two views are then combined rather than the inside view dominating.

How does "Translate beliefs into bets to reveal your true confidence" work?

People routinely claim higher confidence than they actually have when there is no cost to being wrong. The prospect of a concrete loss — even a small or imaginary one — activates a more accurate self-assessment. Framing a belief as a bet makes the implicit probability claim explicit and subjects it to an immediate reality check: you either want to take the bet or you don’t, and that preference reveals something real about your true confidence level.

rationality

What is Bayesian thinking, and how do you use it to make better decisions?

Holding beliefs as probabilities and updating them when evidence arrives

Short answer

Bayesian thinking is the practice of holding beliefs as probabilities and updating them systematically when new evidence arrives — rather than treating beliefs as simply true or false. The mathematical framework is well established; the challenge is building the habits of explicit probability estimation and honest belief updating that make it practical.

Bayes’ theorem is a rule in probability theory describing how to revise beliefs in light of evidence. As a thinking practice, it asks something harder: can you actually say how confident you are in a belief, track when evidence should change that confidence, and update proportionally rather than all-or-nothing? Research on judgment under uncertainty suggests most people do not do this naturally — we anchor, we confirm, and we update too little or too dramatically. The practices below train the Bayesian habits that counteract those defaults.

The practices (7)

Why it works

Naming a prior before seeing evidence does two things: it makes the subsequent update visible (you can see how far you moved), and it forces you to confront base rates — the background frequency of an event — rather than reasoning only from the case at hand. Research on base rate neglect shows that people systematically ignore background frequencies when case-specific information is available; a stated prior anchors them.

How to do it

1Before examining any new evidence, ask: "What is the base rate for this type of thing?"
2State a number: "I think there is roughly a 20% chance this is true before I look at any evidence."
3Write it down — a mental estimate evaporates under the pressure of incoming evidence.
4Only then look at the evidence, so you can measure how much it actually moves you.

Evidence

RCT / meta-analysis

Base rate neglect is one of the most replicated findings in judgment research — people reliably underweight background frequencies when presented with individuating case information.

Honest caveat: The prior-setting habit as a practical countermeasure is widely endorsed in rationality training but has fewer direct experimental tests than the bias it targets.

— Kahneman & Tversky (1973), on the psychology of prediction, Psychological Review

Common mistake: Setting a "prior" after glancing at the evidence, then believing you are reasoning from base rates when you are actually reasoning from the evidence twice.

Go deeper on this practice →

Practice this with IX Coach

Reading about a practice changes nothing on its own. IX Coach turns these into a guided, adaptive routine — discerning where you are in real time and walking the practice with you, session after session.

Start with IX Coach →