Uncertainty, probability and double yoked eggs

Episode Summary

Executive Summary: Tim Harford interviews statistician Sir David Spiegelhalter about his new book, The Art of Uncertainty, exploring how to think about ignorance, risk, luck, and chance. The conversation uses egg coincidences, failed financial models, Bay of Pigs, and crisis communication during BSE, foot and mouth, and COVID to show that uncertainty should be handled with humility, provisional advice, and honest communication.

Main Topics: Uncertainty as broader than the future (Priority: 5/5): Spiegelhalter argues uncertainty includes not only what will happen, but also what is happening now and what happened in the past. He frames ignorance as the conscious awareness of not knowing. Coincidences and the double-yolked egg example (Priority: 4/5): The book cover and opening example challenge misleading probability calculations around six double-yolked eggs, showing how media stories can ignore sorting and selection effects. Limits of models in decision-making (Priority: 5/5): He warns that complex models can be useful guides but are not reliable truths, especially under changing or unstable conditions, as seen in the 2008 financial crisis. Communicating probability clearly (Priority: 4/5): The Bay of Pigs example shows how changing a quantified probability into vague language like 'a fair chance' can distort decision-making, while intelligence agencies standardize probability language. Best practices for crisis communication (Priority: 5/5): Spiegelhalter praises John Krebs's approach during food crises: state what is known, what is unknown, how to find out more, what the public can do, and that advice may change. Political difficulty with revision (Priority: 4/5): The discussion criticizes politicians for resisting changes in advice, even when new evidence emerges, because they fear accusations of a 'U-turn.'

Key Arguments: Uncertainty is not just about predicting the future; it also includes ignorance about the present and the past. Probability calculations can be misleading if they ignore how events are selected, sorted, or reported. Complex models are caricatures of reality: useful, but only within limits and under stable conditions. Good modellers do not 'believe' their models; they use them cautiously as guides. Quantified probability language is better than vague terms like 'likely' or 'a fair chance' because it reduces ambiguity. Effective crisis communication should be honest, provisional, and explicit about knowledge gaps and next steps. Public authorities should be willing to update advice when evidence changes, even if that means admitting previous guidance was incomplete or wrong.

Data Points: Chance of a single double-yolk egg: 1 in 1,000 - Mentioned by the Egg Council in the anecdote about six double-yolk eggs. Chance of a box with six double-yolk eggs: 1 in a million, million, million - The naive calculation presented as an example of misleading probability reasoning. Waiting time for such an event: 500 million years - Estimated from the naive probability of finding a box of six double-yolk eggs. CIA original estimate of Bay of Pigs success: about 30% - The initial military intelligence report before wording was softened. UK intelligence meaning of 'likely': 55% to 70% - Standardized probability yardstick described for UK intelligence communication.

Pivotal Quotes: "all models are wrong and some are useful" — David Spiegelhalter: Explaining why models should be treated as guides rather than truths, especially after discussing financial crisis models. "the conscious awareness of ignorance" — David Spiegelhalter: Defining uncertainty as including not just unknown futures but also unknown facts about the present and past. "we will come back to you and our advice may change" — David Spiegelhalter: Describing John Krebs's crisis communication principle of provisional advice during food safety crises.

Implications: Listeners are encouraged to treat probabilities and expert advice as provisional, demand clearer communication, and accept uncertainty as normal. For institutions, humility, transparency, and willingness to revise guidance are essential for trust.