The Science of Studying: Encoding

The one idea

Most of us were handed material, told to learn it, and left to invent methods that felt like learning: reread, highlight, repeat. The problem is that fluency and retention are different variables, and rereading reliably produces the first while doing almost nothing for the second. The whole series is an attempt to replace feel-good methods with ones that have a mechanism.

The opening technique states the thesis better than any summary: abandon willpower to remember. Mandler’s sorting experiment found that people told to memorize did about as well as people merely told to sort words into categories — because both groups processed meaning. The intention to learn has no power of its own. What determines retention is the operations you actually ran on the material, which means a study session should be judged by what you did, not by how hard it felt.

The engine

The sixteen rules in Part I group into four mechanisms.

Process for meaning, and give it structure. Depth-of-processing work says semantic encoding leaves a richer trace than surface encoding. But meaning alone isn’t enough if it’s left unarranged — Bower’s 1969 hierarchies produced dramatically better recall than the same words randomized, because structure gives spreading activation something to travel along. Hence: draw the skeleton of a chapter before taking notes into it, rather than producing a flat bulleted dump that follows the lecture’s sequence.

Generate before you receive. The densest cluster and the most counterintuitive. Self-explanation, the Feynman pass (“explain it like they’re twelve,” then hunt for the spots where you stall), and elaborative interrogation — asking why must this be true given what I already know — all force you to rebuild connections rather than recognise them. Then two that feel wrong and aren’t: generate before you read (produce your own definition first), and test before you study — Kornell and Bjork found that failed retrieval followed by feedback beats spending the same time reading. The failure is doing the work.

Exploit the second channel. Paivio’s dual coding: verbal and visual systems run independently but linked, so an item encoded in both has two routes back. A crude sketch beside a term outperforms a paragraph. The method of loci is the extreme case, borrowing spatial navigation circuitry as a scaffold — Dresler’s novices went from 26 to 61 items in six weeks.

Respect the four-slot ceiling. Working memory holds about four chunks, not four items, so expertise doesn’t add slots — it enlarges what fits in one. Chase and Simon’s chess masters recalled 90% of a real board and dropped to novice level on a random one, which is the cleanest demonstration available that pattern knowledge is capacity.

The format deserves credit: every rule carries a mechanism, an application, and a common mistake — and the mistakes are the sharpest part. Expecting to teach without producing anything. Peeking at the source during the Feynman pass, which turns it into paraphrasing and discards the diagnostic. Skipping feedback after a pretest, which just consolidates the wrong answer.

My take

The common-mistake field is the innovation here. Most technique lists tell you what to do and leave you to discover that you’ve been doing a hollow version of it for a year — the failure modes are where the actual difficulty lives, and putting them beside each rule is worth more than the rules.

Rule 8 is the one I’d argue with a colleague about: handwriting over laptop, because keyboards are fast enough to stenograph and handwriting forces paraphrase. The finding is contested. The mechanism is not, and it suggests the real instruction isn’t “use a pen” — it’s “make capture slow enough that you have to compress.” A slow keyboard method works fine.

Rule 12 — test before you study — is the highest-yield thing in the list and the one nobody does, because deliberately failing feels like wasting time.

Where it gets thin

This is Part I of three. Sixteen of a promised 63 techniques; storage and retrieval are the sections where spacing and testing effects live, which is where the largest and best-replicated gains are. The piece is a foundation, not a protocol, until the other two parts exist.

The citations are more mixed than the confident tone suggests. Mueller and Oppenheimer’s longhand advantage has largely failed to replicate at scale. The effect sizes quoted for elaborative interrogation (d = 0.85 to 2.57) come mostly from short-retention laboratory studies on factual sentences and should not be expected to survive contact with a real syllabus.

There’s also a load problem the piece doesn’t address: running sixteen encoding techniques on one chapter is not possible, and no guidance is offered on which to pick for which material. And Rule 5 quietly contains the sharpest caveat in the whole series — elaborative interrogation on material you have no foundation in produces confident, wrong explanations that then get encoded as false memories. That deserves to be a section, not a line.

The distilled principle

Judge a study session by the operations you ran on the material, never by how hard it felt — difficulty is a poor proxy and fluency is a worse one.