Research Paper Sample: Retrieval Practice in Secondary Mathematics

Model Answer

Abstract

This study examines the effect of structured retrieval practice on Year 10 mathematics attainment and retention within two comprehensive schools in the North West of England. Using a quasi-experimental pre-test/post-test design, 124 students (intervention n = 64; comparison n = 60) completed an eight-week programme in which the intervention group undertook weekly ten-minute low-stakes retrieval quizzes covering previously taught number and algebra content, while the comparison group continued standard homework-based review of the same material. Attainment was measured with a standardised 40-mark assessment administered before the intervention, immediately after, and at a six-week delayed follow-up. Analysis of covariance, controlling for pre-test score, showed the intervention group significantly outperformed the comparison group at post-test, F(1, 121) = 14.62, p < .001, partial η² = .108, with the advantage widening at delayed retention, t(121) = 4.87, p < .001, d = 0.87. These findings support retrieval practice as a low-cost, classroom-ready strategy for improving durable mathematics learning and are discussed in relation to desirable-difficulties theory and their implications for secondary mathematics pedagogy in England.

Introduction

Secondary mathematics attainment in England remains a persistent policy concern, with a substantial minority of students failing to secure a grade 4 or above in GCSE Mathematics each summer (Department for Education, 2023). Much of the underperformance is attributed not to a lack of initial instruction but to forgetting: topics taught early in Key Stage 4, or even Key Stage 3, are frequently not revisited until formal revision begins, by which point substantial decay has occurred (Sumeracki and Weinstein, 2022). Cognitive psychology offers a well-evidenced remedy to this problem in the form of retrieval practice: the act of actively recalling information from memory, rather than passively re-reading or reviewing it, which has repeatedly been shown to produce more durable learning than equivalent time spent restudying (Roediger and Karpicke, 2006).

Since the 2019 Ofsted Education Inspection Framework explicitly referenced cognitive science principles, including retrieval practice, as markers of effective curriculum design, English secondary schools have adopted low-stakes quizzing at scale (Ofsted, 2019). However, much of the underlying evidence base derives from laboratory studies or from subjects such as science and history, where factual recall is more obviously central to the discipline. Mathematics presents a distinct case: success depends not only on remembering facts and procedures but on flexibly applying them to unfamiliar problems, and it remains unclear whether the benefits of retrieval practice observed for declarative knowledge transfer equally well to procedural mathematical skill (Agarwal et al., 2021).

This study addresses that gap by evaluating an eight-week, classroom-embedded retrieval practice intervention in Year 10 mathematics across two comprehensive schools. The research asks: (1) does weekly low-stakes retrieval quizzing improve post-intervention attainment relative to standard homework-based review, and (2) does any advantage persist at a six-week delayed retention test? It was hypothesised that the intervention group would outperform the comparison group at both post-test and delayed retention, with the size of the advantage expected to be larger at delay, consistent with the theoretical account that retrieval practice chiefly protects against forgetting rather than boosting immediate performance (Bjork and Bjork, 2011).

The choice of Year 10 as the target cohort is deliberate. By this stage of Key Stage 4, students are expected to draw on number and algebra content first introduced in Year 7 or 8, often without any structured opportunity to revisit it in the intervening years, and teachers frequently report that forgotten prior content, rather than unfamiliarity with the current topic, is the principal obstacle to progress (Ofsted, 2021). A classroom-embedded intervention at this stage therefore has clear practical relevance: if retrieval practice can be shown to meaningfully slow the forgetting of previously taught procedures, schools have a low-cost lever available well before the higher-stakes revision period preceding GCSE examinations. The study is also positioned to speak to a live methodological debate in education research, namely whether effects demonstrated under controlled laboratory conditions, often using artificial stimuli such as word lists or short prose passages, generalise to authentic curriculum content delivered by classroom teachers operating under normal time pressures (Dunlosky et al., 2013). By using real scheme-of-work content, teacher-delivered quizzes, and an assessment format aligned with GCSE-style questioning, the design prioritises ecological validity alongside experimental rigour.

Literature Review

The empirical case for retrieval practice, often termed the testing effect, is one of the most replicated findings in educational psychology. In a foundational study, Roediger and Karpicke (2006) found that students who took repeated recall tests on studied prose passages retained substantially more information a week later than students who repeatedly restudied the same material, even though the restudy group had performed better on an immediate test. This counter-intuitive pattern, strong initial performance for restudying but superior long-term retention for testing, has since been replicated across age groups, subject domains, and assessment formats in multiple meta-analyses (Adesope, Trevisan and Sundararajan, 2017; Yang et al., 2021), with average effect sizes in the medium range (d ≈ 0.5 to 0.7).

Several theoretical accounts explain the effect. The elaborative retrieval hypothesis holds that the effortful act of recall strengthens and diversifies the memory traces and retrieval routes associated with the target information (Carpenter, 2011). Bjork and Bjork’s (2011) “desirable difficulties” framework situates retrieval practice alongside spacing and interleaving as strategies that feel harder and produce more errors during learning but yield superior long-term retention precisely because that difficulty prompts deeper processing. Both accounts predict that retrieval practice should be particularly valuable when learning is intended to be durable rather than merely sufficient for an imminent test, which aligns closely with the demands of a cumulative subject like mathematics.

Mathematics-specific evidence, while smaller in volume than the literature on verbal or factual learning, is broadly supportive. Agarwal et al. (2021) found that weekly low-stakes quizzing on previously taught content improved unit test scores among middle-school mathematics students relative to a no-quizzing comparison group, with the advantage most pronounced for problems requiring recall of procedures rather than novel problem-solving. Similarly, a classroom-based trial by Yeo and Fazio (2019) reported that retrieval practice improved performance on both repeated and transfer items in secondary algebra, suggesting some benefit extends beyond rote procedural recall. However, other work has been more equivocal: Lyle et al. (2020) found that retrieval practice improved fluency with practised problem types but produced no measurable advantage on items requiring students to combine multiple procedures, raising the possibility that the benefit of retrieval practice in mathematics is concentrated in component-skill fluency rather than higher-order problem-solving.

Within the English policy context, the 2019 Ofsted Education Inspection Framework’s explicit endorsement of cognitive-science-informed curriculum design (Ofsted, 2019) has driven widespread adoption of low-stakes quizzing, often under the banner of “retrieval starters” or “do now” tasks. Sumeracki and Weinstein (2022) note, however, that much classroom implementation departs from the conditions under which the strongest laboratory effects were demonstrated, for instance using very short delays between initial learning and retrieval, or providing minimal corrective feedback, both of which are theorised to attenuate the benefit. There remains, therefore, a need for classroom-based, UK-context evaluations that test retrieval practice under realistic implementation conditions and report both immediate and delayed outcomes; this study is designed to contribute such evidence in the specific case of secondary mathematics.

A further strand of relevant literature concerns the role of feedback in retrieval practice. Dunlosky et al. (2013), in a widely cited review of learning techniques, rated practice testing among the small number of strategies with high utility across age groups, subject areas, and learning conditions, but stressed that the benefit is contingent on retrieval attempts being followed by feedback that corrects errors rather than simply reinforcing them; without such feedback, incorrectly retrieved information can itself become more strongly entrenched. This has direct implications for classroom implementation: a retrieval quiz that is administered but not promptly and accurately reviewed may fail to reproduce the effects observed in tightly controlled studies, and several UK-based process evaluations of retrieval-practice rollouts have flagged inconsistent feedback provision as a common weakness of real-world adoption (Sumeracki and Weinstein, 2022). The intervention protocol used in the present study was therefore designed to build in brief, whole-class corrective feedback immediately following each quiz, a feature intended to preserve fidelity to the conditions under which the strongest effects have previously been demonstrated.

Methodology

Design. A quasi-experimental, non-equivalent groups pre-test/post-test/delayed-retention design was used. True random allocation of individual students was not feasible within an intact school timetable, so allocation occurred at the class level: two Year 10 mathematics classes at School A were assigned to the intervention condition and two comparable classes at School B, matched on prior attainment banding, served as the comparison condition. Both schools followed the same scheme of work and used the same core textbook series.

Participants. The final sample comprised 124 Year 10 students aged 14–15 (intervention n = 64; comparison n = 60) following parental opt-out consent procedures approved by both schools’ senior leadership and in line with British Educational Research Association (BERA, 2018) ethical guidelines. Six students originally enrolled did not complete all three assessment points due to absence and were excluded from analysis, leaving complete data for the reported sample. Groups did not differ significantly on prior Key Stage 3 mathematics attainment at baseline, t(122) = 0.61, p = .544.

Intervention. For eight consecutive teaching weeks, intervention classes began each mathematics lesson with a ten-minute low-stakes retrieval quiz comprising six short-answer questions drawn from content taught two to six weeks previously, spanning number, algebra, and ratio topics from the current scheme of work. Quizzes were self-marked against a displayed answer key immediately afterwards, with the teacher briefly addressing any items where more than a third of the class made the same error. Scores were not recorded for grading purposes. Comparison classes spent an equivalent amount of lesson time reviewing the same underlying content through teacher-led worked examples and independent practice questions, without a retrieval-based quiz format, reflecting typical existing practice at School B.

Measures. Mathematics attainment was assessed using a 40-mark test constructed from past-paper style items covering number, algebra, and ratio, calibrated for approximate equivalence in difficulty across the pre-test, post-test, and delayed-retention versions by an experienced head of mathematics not otherwise involved in the study. The same test format, but with different surface-level numbers and contexts, was used at each of the three time points to reduce direct item-recognition effects. Internal consistency for the assessment was acceptable (Cronbach’s α = .81 at post-test).

Procedure and analysis. The pre-test was administered in the week before the intervention began, the post-test in the week immediately following the final quiz, and the delayed-retention test six weeks after the intervention ended, with no further targeted revision of the covered content in either condition during the delay period. Data were analysed using analysis of covariance (ANCOVA) to compare post-test scores between groups while controlling for pre-test score, and an independent-samples t-test to compare delayed-retention scores. Effect sizes are reported as partial eta-squared for the ANCOVA and Cohen’s d for the t-test. All analyses were conducted with an alpha level of .05.

Sample size and assumptions. An a priori power analysis using G*Power indicated that a total sample of approximately 110 would provide 80% power to detect a medium effect (f = 0.25) at α = .05 for the planned ANCOVA, so the achieved sample of 124 was judged adequate. Prior to the main analyses, standard ANCOVA assumptions were checked: residuals were approximately normally distributed by visual inspection of Q-Q plots, Levene’s test indicated homogeneity of variance was not violated, F(1, 122) = 1.14, p = .288, and the assumption of homogeneity of regression slopes was satisfied, as the interaction between group and pre-test score was non-significant, F(1, 120) = 0.87, p = .353, confirming that ANCOVA was an appropriate analytic choice.

Results

Descriptive statistics for both groups at each assessment point are presented in Table 1. Both groups improved from pre-test to post-test, but the intervention group’s gain (+5.4 marks) was substantially larger than the comparison group’s gain (+1.2 marks). At delayed retention, six weeks after the intervention ended, the comparison group’s mean score had fallen back close to its pre-test level (-1.5 marks from post-test), whereas the intervention group retained the great majority of its gain (-0.9 marks from post-test).

An ANCOVA on post-test scores, with group as the independent variable and pre-test score as a covariate, confirmed a significant effect of group, F(1, 121) = 14.62, p < .001, partial η² = .108, indicating a medium-to-large effect once baseline differences were accounted for. The covariate itself was also a significant predictor of post-test score, F(1, 121) = 68.34, p < .001, confirming the appropriateness of controlling for prior attainment. At delayed retention, an independent-samples t-test found a significant difference favouring the intervention group, t(121) = 4.87, p < .001, d = 0.87, a larger standardised effect than that observed at post-test (d equivalent ≈ 0.62), consistent with the prediction that the benefit of retrieval practice would be more pronounced after a delay. Figure 1 illustrates the mean scores for both groups across the three assessment points.

An exploratory analysis by prior-attainment band (split at the sample median pre-test score) suggested the intervention benefited both higher- and lower-attaining students, with no significant group-by-band interaction at post-test, F(1, 119) = 1.42, p = .236, indicating that the advantage of retrieval practice was not confined to a particular attainment subgroup within this sample. Attendance at the weekly quizzes was high across the intervention classes, with students missing on average 0.7 of the eight sessions, most commonly due to routine absence, and a sensitivity analysis excluding the eleven students who missed two or more sessions produced a materially unchanged pattern of results, F(1, 108) = 13.89, p < .001, suggesting the main findings were not driven by a small number of highly engaged students.

Discussion

The results support the hypothesis that weekly low-stakes retrieval practice improves mathematics attainment relative to standard homework-based review, and that this benefit is not only maintained but proportionally larger after a six-week delay. This pattern, a moderate advantage immediately after the intervention that widens at delayed retention, mirrors the classic testing-effect finding of Roediger and Karpicke (2006) and is consistent with the desirable-difficulties account (Bjork and Bjork, 2011): retrieval practice appears to protect mathematical knowledge against the forgetting that the comparison group experienced once formal instruction on the covered topics ended.

The magnitude of the effect observed here, partial η² = .108 at post-test and d = 0.87 at delayed retention, is broadly consistent with, if slightly larger than, the medium effects reported in general-education meta-analyses of retrieval practice (Adesope, Trevisan and Sundararajan, 2017). It is also in line with Agarwal et al.’s (2021) middle-school mathematics findings, adding UK secondary evidence to a literature that has been dominated by North American samples. The result is more optimistic than Lyle et al.’s (2020) finding of no transfer advantage for multi-step problems; however, the present assessment weighted heavily toward component procedures (number, algebra manipulation, ratio calculation) rather than extended multi-step problem-solving, and it would be premature to claim that retrieval practice enhances higher-order mathematical reasoning on the basis of this study alone. Future work incorporating a greater proportion of unfamiliar, multi-step transfer items would help clarify the boundary conditions of the effect within mathematics specifically.

Several limitations qualify these conclusions. First, allocation occurred at the class and school level rather than through individual randomisation, meaning that unmeasured differences between School A and School B, in teaching staff, school culture, or the wider curriculum, cannot be fully ruled out as alternative explanations, despite the absence of a significant baseline difference in prior attainment. Second, the study relied on a single subject-specific attainment measure rather than a validated standardised instrument, which may limit comparability with other studies. Third, the eight-week duration and six-week delay, while practically meaningful, cannot speak to whether the retention advantage would persist over a longer period, such as until GCSE examinations many months later. Finally, because comparison-group teachers were aware they were the “business as usual” condition, some diffusion of retrieval-style techniques into their teaching cannot be entirely excluded, which, if anything, would tend to understate the true size of the intervention effect.

Despite these limitations, the findings carry practical implications for secondary mathematics teaching. A ten-minute weekly quiz requiring no additional resourcing beyond teacher time to prepare and mark a brief answer key represents a low-cost intervention that classroom teachers can implement within existing lesson structures, aligning with the direction already encouraged by the current Ofsted framework. Departments considering adopting or extending retrieval practice may wish to prioritise consistency and adequate spacing between initial teaching and retrieval, both of which were features of the protocol used here, rather than assuming that any form of low-stakes quizzing will automatically confer the same benefit.

The finding that neither higher- nor lower-attaining students appeared to benefit disproportionately is also of practical relevance to departments concerned that a whole-class retrieval routine might advantage already-secure students at the expense of those who are struggling. While the present sample is too small to draw firm conclusions about differential effects, the absence of a significant interaction here is at least consistent with retrieval practice functioning as a genuinely inclusive strategy rather than one that widens existing attainment gaps, echoing similar null interaction findings in Agarwal et al.’s (2021) middle-school study. Given the minimal additional workload involved relative to existing homework-review practice, and the absence of any indication that the intervention disadvantaged particular groups of students, the balance of evidence from this study favours continued and, where feasible, wider departmental adoption of structured, feedback-supported retrieval quizzing as a routine feature of secondary mathematics teaching, pending confirmation from larger and more tightly controlled trials.

Conclusion

This study found that an eight-week programme of weekly, low-stakes retrieval quizzing produced significantly better mathematics attainment among Year 10 students than standard homework-based review, with the advantage growing rather than shrinking over a six-week delay. The findings extend testing-effect research into UK secondary mathematics and offer encouraging, if not unqualified, support for the classroom-level adoption of retrieval practice as a means of strengthening durable learning. Given the study’s quasi-experimental design and single-subject focus, replication with individually randomised allocation, a longer follow-up period, and a broader range of problem types, including multi-step transfer items, would strengthen confidence in both the size and the scope of the effect before wider curricular recommendations are made.

References

• Adesope, O.O., Trevisan, D.A. and Sundararajan, N. (2017) ‘Rethinking the use of tests: A meta-analysis of practice testing’, Review of Educational Research, 87(3), pp. 659-701.
• Agarwal, P.K., Nunes, L.D. and Blunt, J.R. (2021) ‘Retrieval practice consistently benefits student learning: A systematic review of applied research in schools and classrooms’, Educational Psychology Review, 33(4), pp. 1409-1453.
• British Educational Research Association (BERA) (2018) Ethical Guidelines for Educational Research. 4th edn. London: BERA.
• Bjork, E.L. and Bjork, R.A. (2011) ‘Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning’, in Gernsbacher, M.A. et al. (eds.) Psychology and the Real World. New York: Worth Publishers, pp. 56-64.
• Carpenter, S.K. (2011) ‘Semantic information activated during retrieval contributes to later retention: Support for the mediator effectiveness hypothesis of the testing effect’, Journal of Experimental Psychology: Learning, Memory, and Cognition, 37(6), pp. 1547-1552.
• Department for Education (2023) GCSE and Equivalent Results in England. London: DfE.
• Dunlosky, J., Rawson, K.A., Marsh, E.J., Nathan, M.J. and Willingham, D.T. (2013) ‘Improving students’ learning with effective learning techniques’, Psychological Science in the Public Interest, 14(1), pp. 4-58.
• Lyle, K.B., Bego, C.R., Hopkins, R.F., Hieb, J.L. and Ralston, P.A.S. (2020) ‘How the amount and spacing of retrieval practice affect the balance of showing versus telling in mathematics’, Applied Cognitive Psychology, 34(2), pp. 350-360.
• Ofsted (2019) The Education Inspection Framework. Manchester: Ofsted.
• Ofsted (2021) Coronavirus (COVID-19) Findings from Visits to English Schools. Manchester: Ofsted.
• Roediger, H.L. and Karpicke, J.D. (2006) ‘Test-enhanced learning: Taking memory tests improves long-term retention’, Psychological Science, 17(3), pp. 249-255.
• Sumeracki, M.A. and Weinstein, Y. (2022) ‘Retrieval practice in the classroom: Lessons from cognitive science for teachers’, Impact: Journal of the Chartered College of Teaching, 14, pp. 22-25.
• Yang, C., Luo, L., Vadillo, M.A., Yu, R. and Shanks, D.R. (2021) ‘Testing (quizzing) boosts classroom learning: A systematic and meta-analytic review’, Psychological Bulletin, 147(4), pp. 399-435.
• Yeo, D.J. and Fazio, L.K. (2019) ‘The optimal learning strategy depends on learning goals and processes: Retrieval practice versus worked examples’, Journal of Educational Psychology, 111(1), pp. 73-90.