Corrections to the hardback

We want to stick to our own Stats Style Guide, specifically number 11: “If you get it wrong, admit it”. So here we are, admitting the bits we got wrong.

Chapter 3: Sample Sizes

P25. We said: “If you roll 100 times, your 95 per cent probability interval is

between eleven and twenty-five: just 10 per cent of the possible

outcomes.” It should be 15% of the possible outcomes.

We also said: “If you roll 1,000 times, then there’s a 95 per cent chance that

you’ll roll a 7 between 145 and 190 times. Your range of outcomes

has narrowed to just 4 per cent of the total.” It should be 4.6%.

Chapter 8: Causality

P56. We wrote: “causes usually have to come after effects”. We meant before effects. D’oh.

Chapter 10: Bayes’ theorem

P71. We got our sums wrong. If you have a million people, 100 of whom have some rare disease, and if you run a test on those million people which is 99% specific and 99% sensitive, then it would correctly tell 989,901 healthy people that they don’t have the disease; it would incorrectly tell 9,999 healthy people that they do have the disease; it would correctly tell 99 people that they do have the disease, and it would incorrectly tell one person that they don’t.

So the odds of having the disease, if you get a positive test, are just under 1%. We got several of those numbers wrong in the book, and ended up with a result of about 8%, which was incorrect.

P73. We say Andrew Deen was convicted in 1994 but that his trial was overturned in 1993, which obviously can't be right - he was convicted in 1990. We should also note that he was convicted in the retrial, which we hadn't realised at the time of writing.

P74. We mistranscribed a line from this study, which talks about a test for breast cancer which has an incidence of 10 women in every 1,000, or 1%. We wrote that as 10% of women; it meant that the subsequent finding, that a positive result would mean you have about 10% chance of having the disease, made no sense. A 10% incidence would have meant about a 50% chance of the test being accurate.

Chapter 12: Has What We’re Measuring Changed?

p87. We said: "Then, on 26 June 2019, several states agreed to include ‘probable’ deaths". The year should have been 2020.

Chapter 13: Assumptions in Models

p128. We said the Home Office modelled the impact of Brexit. when it should have been the Treasury.

As far as we know, the paperback contains no errors...