Notes on Test-Enhanced Learning (Roediger & Karpicke)

Testing effect: taking memory tests enhances long-term retention beyond restudying, even without feedback.
Testing vs restudy examined under educationally relevant materials (prose passages) and free-recall tests.
Desirable difficulties: testing introduces retrieval practice that boosts later memory; massed studying yields short-term gains but poorer long-term retention.
Transfer-appropriate processing & retrieval-route multiplication: practice retrieval during learning improves performance on retrieval-based tests later.
Boundary conditions: timing of tests after learning and level of retrieval success influence the magnitude of the effect.

Subjects: $n = 120$ undergraduates.
Materials: Two prose passages ("The Sun" and "Sea Otters"), each divided into $30$ idea units; passages were $256$ and $275$ words.
design: 2 within-subject learning conditions (restudy vs. test) x 3 between-subject retention intervals (5 min, 2 days, 1 week).
Phase 1: four 7-minute study periods; during testing periods, participants wrote recall for the specified passage; between periods: arithmetic tasks.
Phase 2: retention interval (5 min / 2 days / 1 week); Phase 2: recall task identical to Phase 1.
Final measure: proportion of idea units recalled on the final test.
Initial recall performance: mean recalled idea units ≈ $20.9/30 \, (\approx 0.70)$ .

5-minute retention: restudying > initial testing (≈ $0.81$ vs. $0.75$ ).
Delayed retention (2 days, 1 week): initial testing > restudying.
Statistical results (ANOVA):
- Main effect of learning condition: F(1,117) = 36.39,\
  u^2_p = 0.24
- Main effect of retention interval: F(2,117) = 50.34,\
  u^2_p = 0.46
- Learning Condition × Retention Interval interaction: F(2,117) = 32.10,\
  u^2_p = 0.35
Post hoc highlights:
- 5-min: recall after restudy (≈ $0.81$ ) > recall after initial test (≈ $0.75$ ), t(39) = 3.22, d = 0.52.
- 2 days: testing ≈ $0.68$ vs restudy ≈ $0.54$ , t(39) = 6.97, d = 0.95.
- 1 week: testing ≈ $0.56$ vs restudy ≈ $0.42$ , t(39) = 6.41, d = 0.83.
Figure 1 shows a crossover: testing boosts delayed retention; restudy helps immediate retention.

Design: 3 learning conditions x 2 retention intervals (between-subjects): SSSS (repeated study), SSST (three study periods, then one test), STTT (one study period, then three tests).
Materials and procedures largely replicate Experiment 1; additional measurements:
- Reading counts: participants read passages several times (means: SSSS ≈ 14.2, SSST ≈ 10.3, STTT ≈ 3.4 reads).
- Post-learning questionnaire: rated interest, readability, and predicted memory after 1 week.
Final recall test: identical to initial recall tests; retention intervals: 5 min or 1 week.

Reading through passage (study counts): SSSS > SSST > STTT in total reads, confirming exposure differences across conditions.
Questionnaire findings: SSSS rated passage as less interesting; yet predicted better memory after 1 week compared to SSST and STTT (though the prediction did not fully align with actual performance).

5-minute final test:
- SSSS recalled about $0.83$ of the idea units, SSST about $0.78$ , STTT about $0.71$ .
1-week final test:
- STTT recalled about $0.61$ , SSST about $0.56$ , SSSS about $0.40$ .
ANOVA: Retention interval main effect: F(1,174) = 122.53,\
u^2p = 0.41. Learning-condition main effect: $F(2,174) = 2.32,\ u^2p = 0.03$ . Interaction: F(2,174) = 18.48,\
u^2_p = 0.18.
Forgetting (proportional forgetting):
- SSSS: 52 ext{%} forgotten; SSST: 28 ext{%} forgotten; STTT: 14 ext{%} forgotten.
Overall pattern: repeated testing yields greater long-term retention than repeated studying; 5-minute results favor more study, but 1-week results favor more testing.

Testing effect robust across materials and retrieval modes; not solely due to re-exposure during tests.
Supports transfer-appropriate processing: retrieval practice aligns with the demands of final recall tests.
Elaboration and retrieval-route theories: testing enhances cue-target retrieval and memory representations more effectively than restudy.
Desirable difficulties: testing introduces a productive challenge that improves retention in the long term.
Miscalibration of confidence: repeated studying increases short-term confidence despite poorer long-term retention.
Comparisons with spaced learning: both retrieval practice and spaced presentation improve long-term retention, but retrieval practice does so via a different mechanism.
Practical implication: frequent, low-stakes testing can be a powerful learning tool across educational levels.

Use tests as learning tools, not just assessment tools; frequent testing encourages spacing and retrieval practice.
Tests without feedback still yield substantial long-term benefits; adding feedback could further enhance performance.
Encourage retrieval-based formats (free recall, short-answer) to improve long-term retention beyond recognition-based formats.
Design curricula to incorporate regular low-stakes testing to promote durable learning and ongoing knowledge monitoring.

Immediate testing after study helps long-term retention more than additional studying for prose materials, even without feedback.
Repeated testing can dramatically reduce forgetting over a week, outperforming repeated studying when the final test is delayed.
The benefit of testing persists beyond exposure: it improves the quality of memory traces and retrieval routes.
Confidence in memory can be misleading: people may feel more confident after rereading, yet recall may be poorer long-term.

Experiment 1:
- Main effect of learning condition: $F(1,117) = 36.39, \eta_p^2 = 0.24$
- Main effect of retention interval: $F(2,117) = 50.34, \eta_p^2 = 0.46$
- Interaction: $F(2,117) = 32.10, \eta_p^2 = 0.35$
Experiment 2:
- Retention interval main effect: $F(1,174) = 122.53, \eta_p^2 = 0.41$
- Interaction: $F(2,174) = 18.48, \eta_p^2 = 0.18$
Final-test forgetting (proportional): SSSS ≈ 52 ext{%}, SSST ≈ 28 ext{%}, STTT ≈ 14 ext{%}