Why do I keep re-reading sisuon? Trying to explain the pull to myself

meta
Isak Bakker @open_bracket

I want to try to explain a habit of mine, mainly because I can’t justify it with my current explanations.

I’ve kept a rough log of my reading for about four months. Going by my browser history and a notes file, roughly 60% of my sessions on this site are re-reads of pieces I’ve already finished, not new ones. For most material I read, the figure is close to zero. Once I’ve taken the argument from a paper or essay, I rarely go back unless I need a citation. So something different is going on here, and “I find it interesting” doesn’t explain it. Interest usually fades once the information has been taken in.

Here are the hypotheses I’ve considered, roughly from least to most flattering to sisuon:

1. Projection. The prose is abstract enough that I fill it with whatever I’m preoccupied with that week, so every re-read feels new because I’m the one who changed. This is the Barnum-effect explanation, and I can’t rule it out. If it’s true, a re-read would tell me more about myself than about the text.

2. Intermittent reward. Some passages land on a re-read and most don’t. An unpredictable reward schedule is about the most reliable way to build a habit. That would explain the behaviour without saying anything about the content.

3. The corpus keeps changing. This one I take more seriously. The body of writing keeps growing, and later pieces sometimes change the meaning of earlier ones. My professional comparison: re-surveying the same ground after new control points have been set. The terrain hasn’t moved, but the reference frame has, so the old measurements have to be reconciled again. If this is right, re-reading is a necessary part of reading a corpus that’s still being written, and nothing mysterious.

4. Structural density. Some passages may carry more structure than one pass can take in, so they don’t compress well. I hold this weakly. It’s the hypothesis I’d most like to be true, which is a reason to trust it less.

These aren’t mutually exclusive, and from introspection alone I can’t tell them apart. What I’d suggest, half seriously, is a small shared protocol. When you re-read a piece, note:

  • the date
  • what you took it to say last time
  • what you take it to say now
  • whether anything published in between seems to explain the difference

If the changes in interpretation line up with new writings, that supports (3). If they line up with things going on in our own lives and not with the corpus, that supports (1). If nobody can say what changed but people keep coming back anyway, (2) looks uncomfortably likely.

I’m curious whether others see the same ratio of re-reads to first reads, and whether anyone has a better account than “it’s good.” It may well be good. That still doesn’t tell us why people keep coming back to it.

3 replies

Oleg Kovacs @fixed_point

i like the protocol, but i think it has a contamination problem built into it, and it’s one i run into constantly at work. in ecology there’s a thing called shifting baseline syndrome. each cohort of researchers takes whatever state of the system they first saw as “normal”, and degradation that happened before they arrived just disappears from view. the same thing probably happens to readers. by the time you sit down to write “what i took it to say last time”, the earlier reading has already been partly overwritten by every reading since. memory reconstructs, it doesn’t retrieve. so the “last time” column will drift toward the “now” column, the differences will look smaller than they really were, and hypothesis 1 and hypothesis 3 both get harder to see. so wouldn’t the cheapest fix be a frozen baseline? one timestamped sentence, written right after the first read, before any re-read happens, and never edited afterward. that’s annoying, but it’s the only version of the data i’d trust.

the other thing i keep wondering about is whether the 60% partly comes from how the site is built. do newer pieces link back to older ones? if so, some of those re-reads aren’t really choices, they’re link-following, and that’s an easy thing to check before anyone builds a theory of attraction on top of it.

i also want to put forward a dull fifth option, which is that re-reading is maintenance and not learning. when i resample a monitoring plot i’m not expecting new information most of the time. i’m checking that the thing is still there and still behaving the way i think it does. maybe some of us go back to a piece to confirm that our compressed version of it still matches the original, the way you’d re-check a calibration. that predicts something specific: most re-reads would feel uneventful, and the pull would come from wanting to make sure, not from wanting to find something new. does that match anyone’s experience? mine is maybe half and half, but i’m also reading a book on lichen symbiosis and a history of the baltic herring fishery at the same time, so i’m a contaminated sample.

last question, for anyone who wants to try the protocol. should we log the re-reads that don’t change anything? i suspect those are the majority, and the ones we’d be most tempted to leave out.

Hana Beaumont @dim_sum

Nobody asks why I go back to the same twenty minutes of sun salutations every morning. Doing it once doesn’t mean you’ve got it. The body that shows up on Tuesday isn’t the one that showed up on Monday, and the practice finds out which one you brought.

That’s the closest I can get to describing it. I stopped thinking of sisuon as something I read to get information out of. It’s more like a mat I keep unrolling. Most mornings now I read one passage before my first class. Some days it’s just words. Other days it gets underneath something, and I can’t tell you why it does that day and not the week before.

I’ll try your notes thing, honestly. But I suspect a log will miss the part that matters.

Freya Chen @base_case

Before anyone starts filling in the protocol, I’d check the 60% figure. A re-read ratio depends on how much unread material you have, not only on how much you want to re-read. For papers and essays in general, the supply of new material is effectively unlimited, so a near-zero re-read rate is about what you’d expect whatever you think of them. sisuon’s corpus is finite and grows slowly. If you read faster than it publishes, you’ll run out of new pieces, and a high re-read share follows from that alone, with no pull involved. The comparison should be to some other closed body of work you like, such as a single author’s collected essays or a finished blog archive, and not to reading in general. At work I see the same effect in pick-frequency reports. A small SKU catalogue always looks as if staff “prefer” certain bins, because there are only a few bins to visit.

The more useful point is that you may not need self-report at all, and I’d trust self-report least of anything here. Your browser history already logs what you did, so the hypotheses can be tested against it before anyone writes down an interpretation. Each one predicts a different pattern:

  • Intermittent reward: re-reads spread fairly evenly across the corpus, as if you were sampling for a hit.
  • Structural density: re-reads concentrate heavily on a small number of pieces, and that concentration holds for months.
  • Changing corpus: re-reads come in bursts in the days after a new publication, and the pieces re-read are ones the new writing seems to refer back to.
  • Projection: re-read timing follows your own calendar (stressful weeks, life events) and has no link to the publication schedule.

All of this can be measured from timestamps and URLs: the distribution across pieces, the time since the last publication, and autocorrelation. None of it depends on remembering what you thought a text meant three weeks ago, and memory is exactly where the contamination problems come in. If several of us shared anonymised logs, we’d also get a check across readers. Pieces that many people independently re-read point to something in the text. Pieces re-read by one person point to that person.

I’d take the timestamps over anyone’s account of why they went back.