This is a useful, well-written and wonderful book. This praise does not mean it is an easy read. Many topics are covered and these are often difficult to understand. You must concentrate but that concentration will be paid back. The author acknowledges this towards the end (p. 425) by congratulating the reader for finishing the book (assuming you had not jumped to that point).

We cannot avoid uncertainty. Whether it is assessing whether the latest report about a medical breakthrough is tenuous or realistic, whether it is forecasting the possible path of interest rates or whether it is considering whether to buy a property in a “low risk” flood zone, we all must make decisions with incomplete information. A theme that runs through the book is that uncertainty is personal.

To understand this, consider certainty. This can be defined as a “firm conviction, with no doubts, that something is the case”. This clearly expresses the idea that certainty is a personal thing. Therefore, so is uncertainty, which occurs when someone does not have a firm conviction and does harbour doubts”. As a result “Uncertainty [can be defined as]: the conscious awareness of ignorance. … we should not be thinking of uncertainty as a property of the world but our relationship with the world. This means that individuals or groups can, quite reasonably, have different degrees of uncertainty about exactly the same thing, due to them having different knowledge or perspectives.” (p. 19)

There are different forms of uncertainty. A basic split is between epistemic and aleatory uncertainty. Epistemic uncertainty can be reduced by more knowledge. Aleatory uncertainty, meanwhile, cannot be reduced by more knowledge. It involves randomness “about the future we cannot know.” (p.17)

The preceding words in this review should alert the reader that the book is not solely about the statistics and probability. It is as much philosophical as technical. Statistics play a major role in understanding uncertainty but they are not enough.

The book has insightful chapters on using statistics to understand uncertainty, a chapter on communicating uncertainty (a crucial issue in the age of disinformation, misinformation and bullshitting. Yes, that can be technical term – see page 425) and a chapter on living with uncertainty. On communicating uncertainty, Spieglehalter praises the Bank of England’s fan charts. In a footnote, he notes that the Bernanke review of the Bank’s forecasting suggests replacing the fan charts with central projections, alternative scenarios, and statements of uncertainty. In Spiegelhalter’s view he “personally feel(s) it is unfortunate to lose a strong visual representation of modelled uncertainty.” (p. 314)

For the audience this review is written for (i.e. economists), there is particular relevance in chapter 13 which covers deep uncertainty. This occurs when “we are facing circumstances we don’t understand well, we are unwilling to assign probabilities, and have low confidence in any models. We might be considering events that have not happened before, when we might not even be able to imagine what shocks may be in store.” (p. 356)

Deep uncertainty incorporates the concept of radical uncertainty discussed by John Kay and Mervyn King (see a review of Radical Uncertainty here). Chapter 13 discusses the views of John Maynard Keynes and Frank Knight when it comes to situations where we may be unwilling to assign probabilities and have low confidence in any models. Essentially, Keynes and Knight consider that one cannot and should not attempt to enumerate risk when events in the future cannot be known. Spiegelhalter considers these view “were rendered out-of-date when ideas of subjective probability became both respectable and widespread” (p. 360) On Keynes, Spiegelhalter argues “He says there is no basis for calculating a probability, but that does not mean there is no way of assessing probability.” (p. 360). On Knight, he is “only focussing on situations where there is no ‘measurable’ probability and ignores the use of subjective argument. The unfortunate term ‘Knightian uncertainty’ has come to be used for situations when people ‘don’t know the probability distribution’, but this inappropriately implies that probability is an objective property of the world which we happen not to know.” (p. 361)

As mentioned earlier, this book is not an easy read. It forces one to think a lot about what we mean when we use terms such as likely or unlikely or how to evaluate whether a model is useful. These topics are not new. They have been covered increasingly over the past 20 or 30 years. John Kay and Mervyn King’s book “Radical Uncertainty” gives one view – which seems to be contested by this book. Philip Tetlock and Dan Gardner’s 2016 book “Superforecasting” and Tetlock’s earlier 2005 “Expert Political Judgement” are also insightful. Spiegelhalter’s book builds wonderfully upon these and other studies.