Boy! Was this a disappointing read! I thought Nate Silver knows what he was doing when I saw FiveThirtyEight give 30% chance for Donald Trump to win the election. His interviews after the election were also refreshingly sane. He pointed out that many polls gave Hillary a high chance of winning without actually worrying about the margin of errors in polls. He also mentioned that sometimes these margins could be correlated. I remember him say that if you have a coin where tales can come up with a probability of 0.3, you will see that the coin does result in a tail sometimes. I had high hopes of understanding what Nate can teach me about predictions. After reading this book, I am really disappointed. The book can be summed up in one sentence: always think of predictions as probability distributions and be mindful of uncertainties.
The book did not have a proper beginning and ending. The chapters are all ad-hoc. The book talks about prediction in the following disciplines: Economy, Politics, Sports, Weather, Earthquakes, Epidemics, Games, Gambling, Climate Change, and finally Terrorism. The order I wrote is the order of chapters discussed in the book. This order doesn't make any sense. Other than the fact that all of these are avenues where people make predictions, there is absolutely nothing common among these disciplines.
There is no discussion about Science in this book at all. Science is basically about generating models and abstractions of the natural world so that we can make testable predictions. There are a lot of things humans are damn good at predicting. Weather we can see a complete solar eclipse from a given place and time, weather a missile can reach a target, weather we can generate enough energy from a given fuel source, weather a medicine works or not, weather a given amount of silicon is sufficient to make a given number of computations, etc. Humans are good at making these predictions because we understand the underlying principles (physics, chemistry, biology, and their corresponding mathematical foundations) are very well understood. There is ZERO mention of this in the book. This book only makes economic arguments about making predictions. Of course, I should not have expected anything else from a semi-professional gambler, and ex-KPMG employee, and economics graduate with zero scientific background.
As an economics student, Nate is stuck about thinking every prediction as a probability distribution. Yes, thinking explicitly about uncertainties and probability distributions is a useful exercise. Probably, I could have learned something if Nate exclusively focused on separating signal from noise in politics and baseball and gave several examples of careful systematic analysis. However, by writing about the general topic of prediction, and rambling about several topics, the book becomes incoherent. The discussions on topics do not have any closure at all. This book should be renamed: All about noise.
For example, in the chapter on Basketball, Nate introduces the graph of average performance vs age and then introduces several players where the curve is not fully valid. There is no further discussion about it. To someone who wants to predict the scores/understand the prediction method that Nate developed, this cursory discussion was useless. Similarly, the discussion about how Kasparov was defeated by Deep Blue does not have any relations with the rest of the book. I learned nothing from the chapter on weather prediction. The chapter on earthquake predictions was basically redundant. Sure, as an analyst, I can keep drawing curves and perform regressions, but the fundamental question that we do not understand the fundamental model behind the earthquakes is not communicated clearly. The discussion about predicting nuclear and bio terrorism is a perfect example of unknown unknown. At that time, Nate thought that these are the two big things. He fails to include anything about Cyber-terrorism or disinformation campaigns. The discussion about Bayesian inference and why it is important as opposed to Fisher statistics and p-values is basically nonexistent. Each of the chapters seems like a buzz word soup with ideas primarily derived from other books.
I imagine the following conversation between Nate and Publisher.
N: I want to capitalize on my exact predictions of 2012 elections, I want to write a book on predicting politics and sports.
P: Sure. Do you want to give details of some models that you use? How can people make such predictions by themselves?
N: No. I think the readers are too stupid to understand the complexities of all that. I just want to say that general predictions are bad and my predictions are good.
P: In that case, why don't you write about predictions in general. Since all the "predictions" are bad, you can just list them all and ramble about them.
N: Good idea, I will read a few books and meet with a few people. Let me make a list and I will send you chapter by chapter.
P: Sounds good!
The book did not have a proper beginning and ending. The chapters are all ad-hoc. The book talks about prediction in the following disciplines: Economy, Politics, Sports, Weather, Earthquakes, Epidemics, Games, Gambling, Climate Change, and finally Terrorism. The order I wrote is the order of chapters discussed in the book. This order doesn't make any sense. Other than the fact that all of these are avenues where people make predictions, there is absolutely nothing common among these disciplines.
There is no discussion about Science in this book at all. Science is basically about generating models and abstractions of the natural world so that we can make testable predictions. There are a lot of things humans are damn good at predicting. Weather we can see a complete solar eclipse from a given place and time, weather a missile can reach a target, weather we can generate enough energy from a given fuel source, weather a medicine works or not, weather a given amount of silicon is sufficient to make a given number of computations, etc. Humans are good at making these predictions because we understand the underlying principles (physics, chemistry, biology, and their corresponding mathematical foundations) are very well understood. There is ZERO mention of this in the book. This book only makes economic arguments about making predictions. Of course, I should not have expected anything else from a semi-professional gambler, and ex-KPMG employee, and economics graduate with zero scientific background.
As an economics student, Nate is stuck about thinking every prediction as a probability distribution. Yes, thinking explicitly about uncertainties and probability distributions is a useful exercise. Probably, I could have learned something if Nate exclusively focused on separating signal from noise in politics and baseball and gave several examples of careful systematic analysis. However, by writing about the general topic of prediction, and rambling about several topics, the book becomes incoherent. The discussions on topics do not have any closure at all. This book should be renamed: All about noise.
For example, in the chapter on Basketball, Nate introduces the graph of average performance vs age and then introduces several players where the curve is not fully valid. There is no further discussion about it. To someone who wants to predict the scores/understand the prediction method that Nate developed, this cursory discussion was useless. Similarly, the discussion about how Kasparov was defeated by Deep Blue does not have any relations with the rest of the book. I learned nothing from the chapter on weather prediction. The chapter on earthquake predictions was basically redundant. Sure, as an analyst, I can keep drawing curves and perform regressions, but the fundamental question that we do not understand the fundamental model behind the earthquakes is not communicated clearly. The discussion about predicting nuclear and bio terrorism is a perfect example of unknown unknown. At that time, Nate thought that these are the two big things. He fails to include anything about Cyber-terrorism or disinformation campaigns. The discussion about Bayesian inference and why it is important as opposed to Fisher statistics and p-values is basically nonexistent. Each of the chapters seems like a buzz word soup with ideas primarily derived from other books.
I imagine the following conversation between Nate and Publisher.
N: I want to capitalize on my exact predictions of 2012 elections, I want to write a book on predicting politics and sports.
P: Sure. Do you want to give details of some models that you use? How can people make such predictions by themselves?
N: No. I think the readers are too stupid to understand the complexities of all that. I just want to say that general predictions are bad and my predictions are good.
P: In that case, why don't you write about predictions in general. Since all the "predictions" are bad, you can just list them all and ramble about them.
N: Good idea, I will read a few books and meet with a few people. Let me make a list and I will send you chapter by chapter.
P: Sounds good!
Comments