Decompose the unknown into knowable sub-problems

Break the question you cannot answer directly into smaller questions you can.

Why it works

Direct intuition about large, unfamiliar quantities is poorly calibrated — people cannot reliably estimate the number of gas stations in a country, but they can estimate the population, the number of cars per person, typical refueling frequency, and a station’s daily throughput. Decomposition replaces one impossible estimation task with several easier ones, each of which benefits from more available knowledge. Independent errors in sub-estimates also tend to cancel rather than compound.

How to do it

1. Write the unknown quantity as a product or sum of sub-quantities.
2. For each sub-quantity, ask: do I have any anchors for this from everyday knowledge?
3. Estimate each sub-quantity independently, then combine.
4. Do a sanity check: does the combined estimate feel plausible? If wildly implausible, re-examine the biggest input.

Evidence

Decomposition is a validated forecasting technique. Research in judgment and decision-making shows that decomposing estimation tasks into sub-problems reduces systematic bias in aggregate estimates, because errors in components are partially independent. MacGregor, Lichtenstein, and Slovic (1988) demonstrated experimentally that structuring an estimate as decomposed sub-judgments improved accuracy over holistic guessing, with the gain largest for quantities the estimator knew least about directly. (observational)

Decomposition helps when sub-estimates are genuinely independent; when errors are correlated (e.g., all influenced by the same wrong assumption), they do not cancel.

Sources

• MacGregor et al. (1988), "Decomposition as a strategy for judgmental forecasting", Journal of Forecasting
• MacGregor, D. G., Lichtenstein, S., & Slovic, P. (1988). Structuring knowledge retrieval: An analysis of decomposed quantitative judgments. Organizational Behavior and Human Decision Processes, 42(3), 303-323.

Common mistake

Decomposing into sub-problems that all depend on the same unknown variable, so errors correlate rather than cancel.