knowingknowledge.com — how memory actually works

An adult alone at a desk mid-recall with the book shut

Fig. 01 · The shut book

The entry to read first

The testing effect

Roediger and Karpicke found that students who re-read felt more confident and remembered far less than students who tested themselves — retrieval is a learning event, not a measurement of one.

Retrieval—Replicates broadly

The claim: Retrieval practice beats re-reading on any delayed test of the same material.
Where it breaks: Immediate tests, multiple-choice formats, and far transfer.

Some of what follows has been replicated for a century. Some of it is one meta-analysis away from collapsing. Reporting them at the same volume is the most common thing done wrong with this literature.

The register, in part

Eight of the twenty-one entries: what each effect predicts, the condition under which it stops working, and how well it holds.

All 21 entries ↗

Sec.	Entry	What it predicts	Where it breaks	Evidence
01	The testing effect	Retrieval practice beats re-reading on any delayed test of the same material.	Immediate tests, multiple-choice formats, and far transfer.	Replicates broadly
02	Distributed practice	The same total study time spread across occasions beats the same time massed.	Very short retention intervals, where massing can win.	Replicates broadly
03	Mixing problem types	Mixed practice lowers accuracy during practice and raises it on a later test.	Types with nothing to confuse, and very short practice sets.	Robust, bounded
04	Split attention	A diagram separated from the text that explains it worsens comprehension of both.	Readers already fluent in the material.	Robust, bounded
05	The Ebbinghaus curve	Retention falls steeply at first and then progressively more slowly.	As a source of exact intervals: nonsense syllables, one subject.	Shape replicates
02	Expanding intervals	Review gaps that lengthen as material is relearned outperform fixed gaps.	Comparisons with equal-interval schedules, which often tie.	Mixed
06	Fluency	Ease of processing raises confidence in later recall without raising recall.	When judgements follow a delayed practice test rather than a reading.	Robust, bounded
07	Near and far transfer	Gains from practice extend to genuinely novel problems.	Most tests of far transfer; near transfer is the safer claim.	Contested

Replicates broadly — holds across labs, materials and adult populations. Robust, bounded — reliable inside the stated condition, not outside it. Mixed — direction agreed, size disputed. Contested — results conflict; no settled estimate.