Update beliefs by degrees, not wholesale
Treat new information as evidence that shifts probabilities, not as proof that changes everything.
Key takeaways
- What it is: Treat new information as evidence that shifts probabilities, not as proof that changes everything.
- Why it works: When people update beliefs in response to new information, they tend to either ignore it (conservatism bias) or overcorrect to near-certainty (representativeness). Bayesian updating — the mathematically correct approach — requires multiplying the prior probability by the likelihood ratio of the new evidence. Even an informal version of this (how much more likely would this evidence be if my hypothesis is true, versus false?) produces more calibrated beliefs.
- Evidence: Backed by observational / correlational evidence.
- Avoid: Treating a single confirming data point as near-proof — which is the representativeness heuristic exactly: a story that matches your hypothesis feels like it proves the hypothesis.
Why it works
When people update beliefs in response to new information, they tend to either ignore it (conservatism bias) or overcorrect to near-certainty (representativeness). Bayesian updating — the mathematically correct approach — requires multiplying the prior probability by the likelihood ratio of the new evidence. Even an informal version of this (how much more likely would this evidence be if my hypothesis is true, versus false?) produces more calibrated beliefs.
How to do it
- 1State your current probability estimate before looking at new evidence.
- 2Ask: "How much more (or less) likely is this evidence if my hypothesis is correct vs incorrect?"
- 3Shift your probability in proportion to that ratio — not to 0% or 100% unless the evidence is truly definitive.
What the evidence says
ObservationalBayesian reasoning is the normative standard in probability theory. People systematically deviate from it in both directions (over- and under-updating) depending on whether the evidence is vivid or abstract. Training in Bayesian reasoning improves calibration.
Honest caveat: Formal Bayesian computation is not practical for everyday decisions; the goal is the habit of proportional updating, not precise calculation.
- — Kahneman (2011), Thinking, Fast and Slow — review of representativeness and Bayesian deviation research
Common mistake
Treating a single confirming data point as near-proof — which is the representativeness heuristic exactly: a story that matches your hypothesis feels like it proves the hypothesis.
IX Coach asks you to state your current probability and the strength of new evidence separately before updating, preventing the automatic representativeness override.
Practice this with IX Coach →