The role of variable retrieval in effective learning

Effective learning in educational settings is strongly supported by two evidence-based strategies: retrieval practice and spacing. The present research asks whether adding variability across practice sessions further enhances the benefits of these techniques. The encoding variability principle proposes that experiencing different contextual facets of the same information across learning opportunities creates richer contextual representations, increasing the likelihood of successful later retrieval. The authors hypothesize that varying retrieval cues (rather than simply restudying material) will boost memory by enriching episodic context representations, that variability will particularly benefit learning when practice involves retrieval (versus restudy), and that variability will magnify the spacing advantage. They also examine whether learners accurately appraise the effectiveness of variable retrieval.

Encoding variability has long been theorized to aid long-term retention by increasing the breadth of contextual features associated with target information. Despite its inclusion in memory models, empirical findings on variable encoding have been mixed: several studies found null or detrimental effects, sometimes favoring constant over varied encoding, whereas positive effects often required constrained conditions (e.g., specific encoding processes or particular retrieval conditions). Prior work on varying incidental contexts during retrieval found benefits for retrieval practice that did not extend to restudy, but those manipulations used incidental cues that did not directly support retrieval. Studies that varied the wording of questions sometimes found no benefit when variability was limited to rephrasing, while other work with lecture materials demonstrated benefits of variable questions for both retrieval and restudy. The present study aims to provide systematic support for variable cueing under conditions that closely align with educational practice, directly testing its interaction with retrieval practice and spacing and assessing learners’ metacognitive judgments.

Across seven experiments, participants learned with either constant or varied cues, primarily using foreign vocabulary translations and, in the final experiment, lecture content. Experiments 1a and 1b: Polish-speaking participants without Finnish knowledge learned 40 Finnish–Polish word pairs across five spaced retrieval cycles. Cue sentences in Polish either remained constant or varied across cycles. In 1a, an initial study phase was followed by five retrieval cycles without feedback; in 1b, there was no initial study and each retrieval attempt was followed by feedback (correct translation). Final cued-recall tests probed translations with only the Finnish word as the cue. Experiments 2a and 2b: A within-participants 2 (learning condition: varied vs. constant) × 2 (learning mode: retrieval practice vs. restudy) design contrasted retrieval practice with restudy (translations presented outright below the cue sentence for 13 s). Both experiments used practice-with-feedback; 2a tested immediately, 2b tested after 24 h. Experiment 3: Tested spacing by manipulating lag. One group had long lags (average ~40 items between repetitions, as in 1b), and another had short lags (mini-blocks of two items, average lag ~0.5 item). Learning condition (constant vs. varied) was within participants; lag was between participants. Test occurred immediately after practice. Experiment 4: Replicated the variable retrieval advantage and collected metacognitive judgments: global predictions before study (expected performance for constant vs. varied), item-by-item judgments of learning (JOLs) on cycle 5 after feedback, and global postdictions after the final test. Experiment 5: English-speaking participants studied five short geology lecture segments (12 concepts total). For each concept, practice involved either repeated (constant) or varied questions (three unique questions) with feedback; transfer was assessed via novel test questions. Before and after the final test, participants chose which method they believed was more effective (repeated questions, different questions, or equally effective). Participants were recruited via Prolific or from university pools, met language criteria, and provided informed consent; procedures were conducted online. Counterbalancing and randomization were used for condition assignment and order.

Experiments 1a and 1b: Varied cues during spaced retrieval practice improved final cued recall relative to constant cues. 1a (study then retrieval without feedback): t(30) = 4.52, P < 0.001, d = 0.81. 1b (retrieval with feedback only): t(30) = 3.71, P < 0.001, d = 0.67. Experiments 2a and 2b: Variable learning magnified the benefits of retrieval practice over restudy. 2a (immediate test): Main effect of learning condition, F(1,51) = 15.51, P < 0.001, ηp² = 0.233; main effect of learning mode, F(1,51) = 11.28, P < 0.001, ηp² = 0.181; interaction, F(1,51) = 5.87, P = 0.019, ηp² = 0.103. In the constant condition, retrieval did not significantly outperform restudy, t(51) = 1.41, P = 0.17, d = 0.20; in the varied condition, retrieval outperformed restudy, t(51) = 3.60, P < 0.001, d = 0.50. 2b (24-h delay): Main effect of learning condition, F(1,68) = 25.30, P < 0.001, ηp² = 0.271; main effect of learning mode, F(1,68) = 18.42, P < 0.001, ηp² = 0.213; interaction, F(1,68) = 5.71, P = 0.026, ηp² = 0.071. Retrieval outperformed restudy in both conditions, with a larger effect in the varied condition, t(68) = 5.05, P < 0.001, d = 0.61, than in the constant condition, t(68) = 2.10, P = 0.040, d = 0.25. Experiment 3: Spacing benefits were larger under variable learning. Main effect of learning condition, F(1,78) = 39.39, P < 0.001, ηp² = 0.336; main effect of lag, F(1,78) = 9.97, P = 0.002, ηp² = 0.113; interaction, F(1,78) = 6.65, P = 0.012, ηp² = 0.079. Long vs. short lag benefit was greater for varied cues, t(78) = 3.89, P < 0.001, d = 0.87, than for constant cues, t(78) = 2.31, P = 0.024, d = 0.52. Experiment 4: Final test showed a variable-cue advantage, t(39) = 2.11, P = 0.041, d = 0.33, but metacognitive judgments favored constant cues: predictions showed no initial preference, t(33) = 0.50, P = 0.62, d = 0.09; JOLs favored constant cues, t(38) = 3.80, P < 0.001, d = 0.61; postdictions favored constant cues, t(31) = 4.16, P < 0.001, d = 0.73. Experiment 5: Transfer performance was higher after varied questions, t(37) = 3.30, P = 0.002, d = 0.53. Yet most participants judged constant questions as more effective both before and after the test (χ²(2, 38) = 6.52, P = 0.038; N = 20 for constant both times; varied: N = 8 before, 10 after; equal: N = 10 before, 8 after). Practice performance patterns (Table 2) indicated that varied cues generally made retrieval during practice more difficult, consistent with a desirable difficulty account.

The findings confirm that varying retrieval cues enhances learning, especially when practice involves spaced retrieval and when compared to restudy. Variable retrieval appears to enrich episodic contextual representations, aligning with episodic context accounts of retrieval-based learning. The superadditive interactions observed—between variability and retrieval practice (Experiments 2a–2b) and between variability and spacing (Experiment 3)—suggest that combining these techniques yields benefits greater than either alone. Framed within the desirable difficulties perspective, varied cues make retrieval attempts more challenging, encouraging reliance on episodic context and improving later retention. The results also help reconcile prior subadditive findings by highlighting the distinction between deliberate retrieval and spontaneous reminding: variability may hinder reminding during restudy but supports context updating during deliberate retrieval. Metacognitive data reveal a robust illusion: learners misinterpret fluency during practice with constant cues as a signal of better learning, even when varied cues produce superior final performance and transfer. This misalignment between effectiveness and appraisal underscores the need to inform learners and educators about the benefits of variable, spaced retrieval and to develop interventions that correct these metacognitive errors.

Variable retrieval—changing cues across spaced practice—reliably improves retention and transfer beyond constant cueing, and magnifies the benefits of both retrieval practice and spacing. The study establishes variable retrieval as a powerful, educationally relevant strategy that integrates encoding variability with retrieval-based learning. Despite clear performance advantages, learners tend to undervalue variable retrieval, favoring constant cues based on practice fluency. Future work should generalize these findings across diverse educational materials and settings, clarify mechanisms under conditions with and without feedback, and develop approaches to correct metacognitive illusions to promote adoption of variable, spaced retrieval in instruction.

The primary demonstrations used simplified materials (foreign vocabulary translations), limiting immediate generalizability to complex educational content, although Experiment 5 extended findings to lecture-based learning. Many experiments included feedback to equate exposure to correct answers, meaning that observed benefits reflect both retrieval attempts and subsequent encoding from feedback; this design feature complicates precise mechanistic attribution. Practice retrieval was generally harder with varied cues, which may differentially impact learner engagement. The metacognitive tasks revealed persistent misappraisals that could hinder real-world adoption. The studies focused on specific languages and populations (Polish and English speakers) and online testing contexts; broader demographic and instructional settings should be examined.