25 April 2013

The Signal and the Noise


Cover image

I've just finished reading Nate Silver's The Signal and the Noise, with mixed impressions. On the one hand, it is overall a pleasant read and the first best-seller starring Bayes' theorem (I wrote about this point before). Any book that keeps professional statisticians busy guessing whether the author is a true Bayesian or a frequentist in disguise can't be all bad! On the other hand, there are some disappointing points (discussed below.) Concentrating on the negative side might seem unfair, but should also more interesting, since the reviews so far have mostly been elogious.



  1. The style is conversational and easy to follow, but a few details are annoying, such as Silver's love for adverbs:  "literally" (used on almost every page, sometimes more than once) "essentially" (for "almost") and "incredibly" (for "very"). Mixed metaphors and chained clich├ęs are not uncommon. I would have expected more thorough editing for what was clearly a potential best-seller.
  2. One gets the impression that Silver proposes a technical solution to a social (and political) problem. This is (very briefly) touched upon by a review in Science1 and, more incisively, here. In the Introduction, he says:
    ... a single lax assumption in the credit ratings agencies' models played a huge role in bringing down the whole global financial system. 
    (namely, that mortgage defaults will be uncorrelated). Although he does nuance his position throughout the book, he does not consider the stark fact that the ratings agencies have a strong financial incentive to make the most favorable prediction, not the most accurate one. In fact, they are paid to certify a given version of the future, without having to bet on it. This problem cannot be solved by pushing them to adopt more elaborate (and thus even easier to fudge) prediction techniques, but rather by changing the reward mechanism.
  1. The analogy with quantum mechanics (the so-called "observer effect") does nothing to enlighten the reader, since the problem of quantum measurement is much less intuitive than the fact that a prediction can affect the phenomenon it refers to. In particular, Silver's insistence that the observer effect is not the same as Heisenberg's uncertainty principle (true, but irrelevant) gives one the feeling that he learned something interesting and needs to share it, whether or not it is related to the topic of the book.
  2. The author also believes that "snow molecules are much less dense" (than water) and that "molecules are much too large to be discernibly impacted by quantum physics". Unfortunately, "he doesn't know what he doesn't know" (as Donald Rumsfeld might have said).
  3. After borrowing the hedgehog/fox distinction and citing Isaiah Berlin's text, Silver casually notes that "Unless you are a fan of Tolstoy —or of flowery prose— you'll have no particular reason to read Berlin's essay". I find this remark just a tiny bit condescending. He must not think much of his readers, or Berlin (or both).
  4. The discussion of climate change in Chapter 12 is rather short (42 pages in the paperback edition) and does not go into all the fine details. This is understandable given the space limitations, but the text also drew a pointed critique from Michael Mann, who accuses Silver of misrepresenting his position.
  5. Chapter 13  –on terrorism– is even shorter and seems a bit rushed.It consists of a couple of log-log graphs and an exhortation to keep our minds open (I'm paraphrasing, but only slightly). Of course, terrorism is a BIG ISSUE, so every popular book must mention it, but this chapter can be safely skipped (unless you like profound Donald Rumsfeld quotes, in which case you won't be disappointed.)
These are minor critiques: overall, the book is quite enjoyable and I recommend it, unless you have a solid background in maths and your goal is to learn more about the Bayesian approach, in which case I would go directly to Jaynes' book2.


    1. Mr. Bayes Goes to Washington, Sam Wang and Benjamin C. Campbell,
    Science 339(6121), 758-759 (2013). Free full text here.

    No comments:

    Post a Comment