Wednesday, 12 June 2013

Averaging opinions on Nate Silver's The Signal and The Noise

I always write my book reviews before looking at anyone else's opinion. In the case of Nate Silver's The Signal and The Noise, though, I couldn't think of much to say.

But given that a substantial part of Silver's election prediction methodology was to take the average of all the polls he could get his hands on, what better way to consider his book than to take the average of everyone else's reviews of it?

The best place I know to find book reviews is Goodreads - the same place I do my reviewing, in case you didn't bother to click the above link. The mean Goodreads score for TSTN is currently 3.95 out of 5, based on 5,695 ratings. Not bad at all.

There's something a bit strange about the 'rating details' button for this particular book, as it is only providing a breakdown of how 77 of those 5,695 raters scored the book (The first 77? The most recent 77? Why 77???). This is not a very large sample, but it may still be informative.

The breakdown of those 77 ratings is as follows:

35% 27
50% 39
11% 9
2% 2
0% 0

The mean score of this sample of 77 ratings is 4.18, so it seems fairly representative, and the standard deviation, assuming I've done my calculations correctly, is 0.74.

By comparison, Malcolm Gladwell's The Tipping Point, another popular science-esque book to go stratospheric, has a mean rating of 3.79 out of 5 based on 226,252 ratings. The 'rating details' button works better for TTP, although still not perfectly, and the data from its sample of 177,026 ratings breaks down like this:

26% 46663
37% 65831
24% 44192
6% 12276
4% 8064

You can see by eye that there is more variation in Gladwell's ratings, but to give you the hardcore data, the standard deviation of his sample ratings is 1.06.

So Silver's TSTN has a very repectable mean score of 3.96 out of 5, and a very low proportion of people (if we take The Tipping Point as a comparitor) deviated much from thinking it was at least this good. I suppose you would expect a higher mean score to have less deviation, but it makes you warm and fuzzy to have the detail, doesn't it?

So that's the quantitative data. Now for the qualitative.

In general only the first page of reviews on Goodreads (the 'best' 30 reviews) is worth reading, and that's assuming the book has even garnered 30 reviews. However, TSTN is a very popular book and has so far collected 853 reviews, and the quality of these (judging purely by length) doesn't seriously drop off until the 17th page, i.e. after about 500 reviews. But given that, as we've already established, there was very little deviation between the ratings, we can safely assume that there will be little deviation between the reviews too, right? Let's stick to page 1.

The top-rated review has attracted 19 likes and 13 comments. Charles rated the book 4 out of 5, although he says he would have liked to give it 3.5 (Goodreads doesn't allow half-point scoring). Interestingly, Charles also found TSTN difficult to review, and he therefore presents what he says are the consensus views of his book group. So here we have an averaged review inside an averaged review!

Charles' book group found the book generally interesting, but thought some chapters, including those on baseball and terrorists, were dull. Charles also mentions an 'unacceptable number of typographical errors' - something I noticed too, although I only found it mildly annoying.

Stuart's review, the second best, has 6 likes and 1 comment. Stuart has experience in earthquake prediction, which one of the chapters of TSTN takes on. He found 'a lot to like in terms of tone' but thought that 'Silver isn't a good writer' and that the book's organization was 'haphazard'. The most interesting point here is that Stuart felt Silver should have included Monte Carlo simulations in the book. Ultimately, he thought it was in need of more robust editing.

Here are some illustrative excerpts from the other 28 reviewers:
  • 'Silver is a great writer'
  • 'Silver is not the best writer ... His casual style ... diminishes the impact'
  • 'This is a really amazing book'
  • 'there is not much coherence to the thing'
  • There was plenty of good stuff in here, and little to actively disagree with'
  • I don't imagine that a lot of this material is going to stick with me, in no small part because ... Silver is often arguing for common sense things'
  • 'I'm not sure the chapters on baseball and chess especially added all that much'
  • Some good insights here, as well as a very good chapter on IBM's chess playing computer ... The section on rating a shortstop's fielding abilities was equally excellent.'
  • 'Bayes's theorem, however, requires us to know the probability of an event before we weigh in the new evidence. ... Silver doesn't really get into this discussion.'
  • 'considering its centrality to his book, I didn’t feel he did enough to explain how one goes about arriving at a prior probability in the first place'
  • 'I sadly did not feel like I had gained a very deep understanding of Bayesian thinking by the end'
  • 'Indeed his explanation of Bayes Theorem which has eluded me for years finally made the penny drop.'
So, not as little deviation there as we'd expected: Silver is either a great writer or a poor one, and chess and baseball either interest you or they don't.

However, on the key points there does seem to be good agreement: the book is largely interesting and highly readable, some chapters are more interesting than others, and some of the key arguments could be more deeply explored or better explained.

And I think that brings me back to where I started. Usually when I struggle to review a book, I find when I look at other people's reviews that they aren't terribly interesting or inspired, and that there is largely agreement on the key points and disagreement only about things that don't matter. That was very much the case here.

Having now looked at everyone else's reviews of The Signal and The Noise, I'm glad I didn't waste a lot of time trying to think of something to say.

No comments:

Post a Comment