Quantization of Books

In response to my post about Amazon sales-rank tracking, Matthew Beckler created just such a tracker:


That’s the last few days’ worth of hourly rankings of How to Teach Physics to Your Dog, as of 10am Eastern. Enormous dork that I am, I find this really fascinating, and not just in an absolute-number sense (because, really, these numbers don’t mean much of anything).

The big thing that jumps out at me is the quantization of books. You see a bunch of sharp, downward-going jumps in the graph, which presumably correspond to discrete book purchase events. In between jumps, there’s a slow upward trend, taking 6-8 hours to erase the jump from a single book (lower numbers are better, remember), which presumably says something about the rate at which Amazon’s ranking algorithm depreciates recent purchases. The jump is also fairly consistently smaller when starting at lower numbers, which also makes sense– the vast majority of books at Amazon haven’t sold anything recently, so you can leapfrog a lot of them with a single sale, but once you enter the ranks of books that are currently selling copies, the relative effect of a single sale becomes much less.

And so on. I am inordinately fond of looking at graphs of data, even relatively meaningless data like this. Consider yourselves lucky that I don’t have the data as a table of numbers, thus sparing you from curve fits…


  1. #1 MRW
    December 14, 2009

    I thought that was pretty interesting, too.

    “Consider yourselves lucky that I don’t have the data as a table of numbers, thus sparing you from curve fits…”

    Ah, but you could – There’s a link to the numbers at the bottom of the page

  2. #2 Matt Leifer
    December 14, 2009

    Ha, one of those drops is me buying the book for my girlfriend. I’m not sure exactly which one, but it is definitely one of the last three on the right hand side. You missed a small trick by not instructing your blog readers to all buy the book at the same time. It could have got you onto the bestsellers list for science.

  3. #3 bmanders
    December 14, 2009

    It’s like the frank hertz experiment. but a lot COOLer. nyuk nyuk nyuk.

    you know, cause the books aren’t being heated…

    :waits for laughter:

    wow. tough crowd.

  4. #4 Janne
    December 14, 2009

    Invert the Y-axis and you got the behaviour of a simple conductance-based leaky integrate-and-fire neuron model. Which makes me wonder what the “reversal potential” of the sales input is, what the spike-triggering level is and what happens when you reach it (interview on Oprah?).

  5. #5 Ron
    December 15, 2009

    Is there a blip at 15 Dec 2009 01:49:17 +0000 (UTC)?

New comments have been temporarily disabled. Please check back soon.