Why writing down your reasoning makes a chess lesson stick

You lose. You open the game review, see a red mark on move 25, look at the move the engine wanted, nod, and close the tab. Next Tuesday you play the same move in the same kind of position. And you're surprised.

That's normal. Memory researchers have been predicting it for about fifty years. Reading an answer is one of the weakest ways to learn anything, and the studies agree on that more than they agree on most things. Here's the short version of the research, and what it means for the ten minutes after a game.

In 1978, Norman Slamecka and Peter Graf ran five experiments with a simple setup. One group read pairs of words. The other group got a rule and a cue and had to come up with the second word themselves. They tested recognition, free recall and cued recall, and on every one of those tests the people who generated the words beat the people who only read them.¹

This became known as the generation effect, and it's been replicated hundreds of times. One meta-analysis pulled together 86 studies and 445 effect sizes and found that generating beats reading by roughly 0.4 standard deviations on average. That's a moderate effect, but a reliable one, and it holds up across different materials and different kinds of tests.²

Compare that with an engine review. It shows you Qe6 and you read it. You didn't come up with anything, so the generation effect has nothing to work with. You're in the “read the word pair” group.

Finding two

18. Nxf7 Ne3 19.Bxe3

MovesSummary

You spent a real think on Bxe3. What was your evaluation of the resulting trades, and what other candidate moves were you comparing it against?

Explaining it to yourself is where the understanding happens

Michelene Chi and her colleagues had physics students think aloud while they studied worked examples. The students who later did well on new problems were the ones who explained each step to themselves on their own as they went, and who caught it when something didn't make sense. The weaker students read the same examples and said things like “okay, that makes sense.”³

But that first study only observed what students did. So Chi's group ran an experiment. Eighth-graders read a text about the circulatory system, and half of them were prompted to explain every line to themselves. The prompted kids learned more, and the biggest gap showed up on the hard questions, the ones where you had to apply the idea to a situation the text never covered.⁴ A 2018 meta-analysis of 64 studies put the effect of prompting self-explanation at about half a standard deviation.⁵

Someone has even tried this with chess. Novices watched a computer play a king-and-rook endgame. Some had to predict its moves and explain why they were played, some only predicted, some only watched. The predict-and-explain group ended up with a better grasp of the principles and played the endgame better than both of the others.⁶ It's a small study, beginners on a single endgame, so we wouldn't lean on it too hard. Still, it lines up with everything else here.

Researchers call this one elaborative interrogation, which is a fancy name for asking “why” about something you're trying to learn. The original experiment was almost funny in how simple it was. Adults got a list of arbitrary sentences to remember. The ones asked to explain why each sentence might be true remembered more than the ones who just read them. The more precise their explanation, the better they did.⁷

We should be upfront about how strong this evidence is, though. The most-cited review of study techniques rates elaborative interrogation and self-explanation as “moderate utility,” not “high.” The reason is that there's less classroom-scale evidence for them than for practice testing and spaced study.⁸ Rereading, highlighting and summarizing are at the bottom of that same table. Going through an engine line is rereading.

Why it feels wrong

18. Nxf7 Ne3 19.Bxe3

MovesSummary

You spent a real think on Bxe3. What was your evaluation of the resulting trades, and what other candidate moves were you comparing it against?

You

This seemed to be the simplest reply to prevent the fork on the rook and the queen by exchanging pieces. I get rid of that dangerous attacker, and I could not see an immediate mating sequence that Black is threatening.

Why the engine line feels like learning

Robert Bjork has spent decades studying something counterintuitive. The practice conditions that feel easy and quick often lead to the weakest long-term learning. The ones that feel slow and awkward (generating answers, getting tested, spacing your study out) tend to produce more. He calls these desirable difficulties.^9,10

An engine review is about as easy as it gets. The answer shows up instantly, it's correct, and you come away feeling like you learned something. In one well-known study, students who reread a passage beat students who tested themselves on it when both groups were checked five minutes later. A week later it flipped. The rereaders remembered about 40 percent and the self-testers about 61.¹¹ When the engine line “makes sense” to you right after the game, that's the five-minute check.

Put all of this together and you get a pretty good description of what a decent coach already does. Before telling you anything, they ask what you were thinking. You have to come up with a reason (generation) and explain your own move (self-explanation), and the question itself is a why question (elaborative interrogation). The whole thing is also slower and more uncomfortable than just being shown the answer, which is the desirable difficulty part.

Here's one moment from a real game analyzed on Chessalyz, shown the engine way and the coach way.