Research
The Evidence Base
A cited research base on how people actually learn and forget — compiled to decide what Intellura builds, and published because the citations are the useful part.
Every claim on this site carries an author, a year, a sample size or effect size where one exists, and — where the underlying study says so — an explicit statement that the evidence is weak, contested, unreplicated, or absent. Nothing is rounded up into a cleaner story. Nothing is averaged into a false consensus where the literature disagrees with itself.
This is not marketing copy. It is a working research reference, drawn from meta-analyses and randomized trials in cognitive psychology and educational research, organized around the questions that determine how a study tool should actually be built: what makes retrieval practice work, why people cannot accurately judge their own learning, how spacing and interleaving affect retention, what feedback formats work and which don't, and why most retention mechanics used by consumer apps have little or no supporting evidence.
Sections
The Testing Effect →
Retrieving information from memory produces more durable learning than restudying it — but the size of the advantage depends entirely on what it's compared against.
g = 0.499 overall (Yang et al. 2021, N = 48,478)
Why You Cannot Judge Your Own Learning →
Foresight bias, stability bias, and the fluency illusion distort self-assessment in a consistent, measurable direction — and the distortion is correctable, but not by introspection.
delaying a judgment of learning: g = 0.93 (Rhodes & Tauber 2011)
Spacing and Interleaving →
The gap between study sessions has a measurable optimum that depends on how long you need to remember something — and interleaving only helps when the material is genuinely confusable.
+111% recall at the measured optimum (Cepeda et al. 2008)
Feedback →
A failed retrieval attempt with no feedback teaches almost nothing. Elaborated feedback beats correct-answer-only feedback, which beats right/wrong feedback — and delaying feedback does not help.
no feedback + <50% initial success: g = 0.03, CI includes zero (Rowland 2014)
Habitual vs. Episodic Retention →
Daily-habit retention is not achievable for a learner with no external stake. The evidence points to episodic, need-triggered return instead — and most retention tactics in consumer software don't survive scrutiny.
MOOC completion: 3.13% (Reich & Ruipérez-Valiente 2019, Science)
How this is compiled
The source document behind these pages obeys five rules, which any page derived from it must also obey:
- Every substantive claim carries author, year, and a number. Claims without numbers are labeled as such.
- Where researchers disagreed, or the literature disagrees with itself, the disagreement is stated — not averaged away.
- Where evidence is weak, contested, unreplicated, or absent, that is written in the sentence making the claim, not in a footnote.
- Effect sizes are transferred with their boundary conditions attached. An effect size stripped of its boundary conditions is folklore.
- Real-world effects are smaller than published effects. DellaVigna & Linos (2022, Econometrica, 126 RCTs, 23 million people) found the average nudge effect was 8.7 percentage points in academic journals but 1.4 percentage points in government nudge units — a gap fully explained by publication bias.