Treat forecasting accuracy as a skill that improves with practice

Take prediction errors as performance feedback, not as proof that forecasting is futile.

Why it works

Superforecasters treat their scoring results as diagnostics, not verdicts. When they are miscalibrated in a domain, they investigate why — was the reference class wrong, was the evidence weighting off, was there a structural dependency they missed? This feedback orientation is the same mechanism that distinguishes deliberate practice from mere experience. Without it, repetition produces confidence without accuracy.

How to do it

1. After a prediction resolves, score it: compare your stated probability to the binary outcome.
2. For predictions you got significantly wrong, run a brief post-mortem: what was the strongest factor you underweighted?
3. Track error patterns: do you over-predict in certain domains, under-predict in others?
4. Set a learning goal for calibration (not just accuracy) — "I want my 70% predictions to land 70% of the time."

Evidence

Tetlock’s research found that forecasters who engaged in deliberate learning from scoring — particularly those who practiced in teams with feedback — showed the most improvement over time. The growth trajectory is consistent with deliberate practice research. (observational)

Calibration improvement is slow and context-dependent; large improvements were seen in tournament conditions with dense feedback. Real-world decision-making with sparse feedback produces slower gains.

Sources

• Tetlock & Gardner (2015), Superforecasting, Chapter 10

Common mistake

Using a single dramatic miss to conclude that forecasting is unreliable rather than investigating the specific mechanism of the miss — one error is a datum, not a refutation.