Update beliefs frequently and in small increments

When new evidence arrives, adjust your probability estimate — even if the change is small.

Why it works

Superforecasters update more often and in smaller increments than average forecasters. The mechanism is Bayesian: each new piece of evidence has a diagnostic weight, and updating in proportion to that weight produces calibration. The two failure modes are under-updating (anchoring on the original estimate) and over-updating (overweighting recent salient events). Small, frequent updates resist both by keeping the estimate close to current evidence.

How to do it

1. Set a regular review cadence for active forecasts: weekly for short-horizon, monthly for long.
2. At each review, ask: "What evidence arrived since last time, and which direction does it push my estimate?"
3. Make a numerical update — even ±3 percentage points — to log the decision.
4. If evidence is consistent with the prior, record that explicitly rather than leaving the number unchanged.

Evidence

In the Good Judgment Project, superforecasters updated their predictions more than twice as frequently as the average participant and maintained better calibration as a result. Sequential Bayesian updating is the formal framework; the tournament data confirm it in practice. (observational)

The update frequency that improved accuracy in forecasting tournaments may not apply identically to personal decision-making contexts with lower signal density.

Sources

• Tetlock & Gardner (2015), Superforecasting

Common mistake

Waiting until you’re "sure" about a change before updating — waiting for certainty means updating too late and by too much, which is just the anchoring failure.