Substack cometh, and lo it is good. (Pricing)

The survivorship bias in book ratings

Just finished The Fortunes of Africa: A 5000-Year History of Wealth, Greed, and Endeavor by Martin Meredith, and gave it 4 out of 5 stars on GoodReads. At nearly 700 pages of narrative text The Fortunes of Africa is not a small book, but it’s pretty dense with fact and a “quick read”. The author is good at balancing narrative flow with packing a lot of information into any given page.

But as I rated the book I realized that the vast majority of my ratings are 4 out of 5. I reserve 5’s for really good books. But why so few ratings less than 4? Obviously, this is due to survivorship bias: in general, I’m not going to finish a book that I don’t like, and I won’t rate books that I don’t finish.

Additionally, the longer a book is, the better it probably has to be for me to finish it. If it is a short book (less than 200 pages) I may just push all the way through, but in general anything longer and I won’t read “cover-to-cover.” When I was younger I would sample chapters and such, but for whatever reason as I’ve gotten older I generally adhere to the sequential structure as envisioned by the author.

Of course there are exceptions. Stephen Jay Gould’s The Structure of Evolutionary Theory is very long. In general I did not really enjoy reading it (though it has its moments, in particular when it comes to history of science), but finish it I did. I read The Structure of Evolutionary Theory in full because it was Gould’s magnum opus, and the best place to get a sense of his thought without ploughing through his whole oeuvre. Though I did not think much of Gould’s ideas personally (few people with an evolutionary genetics orientation do), he was objectively an intellectual of some standing and influence, so it is useful to understand his thought. He mattered, for better or worse.

The Structure of Evolutionary Theory was Stephen Jay Gould’s last book (he died two months after it was published). He had become such an enormous public intellectual that he was clearly beyond the power of any editors to control his prose flourishes. It’s a prolix repetitious work (I read Wonderful Life more recently, and it benefited from being more tightly written).

In contrast when I read The Twilight of Atheism Alister McGrath I thought it was a decently well written book, but totally unpersuasive on the merits of the substance. But after ten years I think descriptively McGrath was right in some deep ways. So I’d probably change my rating of this book between then and now.

So categories of books I read all the way through:

  • Books I enjoy. I’ve read The Fall of Rome three times. I’ve only read The Genetical Theory of Natural Selection front to back a few times, but some chapters (especially the earlier ones) I’ve read many times.
  • Books which are important. I’ve read probably two dozen translations of Genesis in my life (it’s a short book when standalone, so not a big achievement). A lot of the religious stuff I read is because religion is so important to people, even if it isn’t important to me. Honestly, the same with a lot of philosophy.
  • Books which challenge my viewpoints in a substantive sense. The Structure of Evolutionary Theory and The Twilight of Atheism fall into these categories. I was a much more doctrinaire libertarian when I read Michael Parenti’s Blackshirts and Reds, which engaged some apologia for Marxist-Leninism.

4 thoughts on “The survivorship bias in book ratings

  1. Most of my ratings are pretty high as well, but that’s mostly because I’ve sort of developed a sense if I’m going to like something before I read it.

    I need to read more non-fiction, though. I feel bad that 90% of what I read is fiction and poetry, even if half of it is either “canonical” or at least “serious”.

  2. i’m tellin’ you bro, Amazon Audible sub is $15/mo and you get free book credit every month + 30% off all selections. Buy more credits for $11 a piece and you get any priced book/lecture for 1 credit (some normally cost $40 for a 25 hr lecture.) Listen at 2x speed and I can get through 2 or 3 books on a good week *while I’m doing boring chores and stuff.* Life is good. Point being that i run across many bad books but just plow through them anyway in a few hours or also Amazon will let you return them if you don’t like them. Tried the regular Kindle audio book option too.
    The AI text to speech on the Pocket app is pretty good now. Do all my articles with that now and I almost never have a big backlog of articles and books to read cuz you can listen at warp speed. Tyler Cowen said he does this too.

  3. I always rate books. If I drop them, then I’ll give them a 2-star “it was ok” if I simply couldn’t finish, or 1-star “bleh” if I dropped them because they were genuinely not good.

    Most “modern” recommendations algorithms note how you rate books and adjust based on your median or average. If you don’t rate low because you don’t rate things that are not good, they’ll consider you an optimist reader and will downgrade the value of your advice.

    (not that I know how goodreads use ratings for recommendations, if they do rather than simply books read)

  4. You’d also expect the ratings to skew high because you are picking which books to read partly based on whether you think you’ll like them (and presumably you have some sense of your own tastes).

Comments are closed.