What does the research actually say about homework? This evidence-based guide for school leaders and teachers covers what works, what to avoid, and how to build a policy grounded in learning science.
Cooper, Robinson, and Patall (2006) reviewed 60 studies. They found homework helps secondary learners consistently. Primary learners benefit less, sometimes seeing negative effects. This happened when homework replaced family time or enjoyable reading.
Hattie (2009), in his synthesis of over 800 meta-analyses, placed homework at an effect size of 0.29 overall, which he categorised as having some positive effect but below the hinge point of 0.40 that he considered typical of effective teaching. Crucially, Hattie noted that the effect was three to five times larger at secondary than primary, and that the nature of the task mattered considerably. Practice and review tasks showed stronger effects than tasks requiring new learning or extended projects completed without teacher support.
The Education Endowment Foundation (EEF) Toolkit shows secondary learners gain five months from homework. Primary learners gain two months, says the EEF. The evidence quality is moderate. EEF notes big differences exist based on task design, learner age, and home context.
Homework sits at the intersection of school policy, parental expectation, and cognitive science. For many teachers, it generates more heat than light: debates in the staffroom about how much to set, what counts as useful, and whether it is worth the marking time at all. The research base is more nuanced than either the enthusiasts or the sceptics tend to admit.
Research by Kraft et al. (2014) and Sims (2021) underpins this guide. We clarify evidence on homework and highlight areas of disagreement. You can design effective homework policies based on learning science. Each section shows classroom examples for practical application.
A practical example: a Year 10 science teacher sets a retrieval practice quiz each Friday covering the week's content. Learners spend 15 minutes recalling key terms and processes without notes. When marked in class on Monday, the teacher can see at a glance which concepts need revisiting. This kind of task, grounded in retrieval practice, sits at the high end of the homework evidence base. Compare this with a Year 10 teacher who asks learners to "research osmosis and write a page." The latter task is likely to produce a cut-and-paste summary from a website, with little to no learning occurring.
For related guidance, see our article on the Extended Project Qualification.
Researchers and leaders suggest primary and secondary are distinct. Homework risks for younger learners include anxiety and less play (Fan et al., 2017). Classroom support does not easily move into homes. Primary maths homework often causes problems if it feels hard or parents stress learners (Fan et al., 2017).
Homework's value is stronger in secondary school, but it depends. Trautwein and Köller (2003) showed quality of work matters more than quantity. Learners engaging fully learn more than those rushing, they found. 20 minutes focused is better than 60 minutes distracted, according to the research.
Hallam (2004) found UK homework policies rely on tradition, not pedagogy. Key Stage 3's 30 minutes per subject and GCSE's 60-90 minutes lack evidence. Inspection frameworks, not learning science, shaped these policies.
For Year 3, teachers can ask parents to read for 10 minutes nightly. Discussing the story builds oral language alongside reading (Goodall & Vorhaus, 2011). Year 9 history learners make a five-item timeline (Hattie, 2009). The timeline promotes classroom discussion about causation (EEF, 2021).
Task design hinges on if learners retrieve prior knowledge or meet new content. Spaced practice research (Rohrer, 2015) shows recall beats rereading for retention. Homework naturally spaces retrieval (Cullimore, 1999; Cooper, 2001; Marzano & Pickering, 2007).
Cognitive load is key. Complex homework causes extraneous load (Sweller, 1988). Learners struggle if tasks are unclear. Focus tasks on retrieving content. Reduce cognitive demand on instructions (Paas et al., 2003). Make procedures clear (Clark et al., 2006).
Third, effective homework has a clear connection to classroom learning. When learners do not understand how tonight's task connects to what they did in the lesson, or how it will be used in the next one, motivation drops sharply. Teachers who use strong questioning strategies in class can use the same technique to design homework prompts: the question stem that worked as an exit ticket often works equally well as a retrieval task at home. This is not about making homework feel relevant in a vague sense; it is about ensuring there is an explicit pedagogical link. "Complete questions 1-10" communicates a task. "Complete questions 1-10 so that on Thursday we can use your answers to build the model" communicates a purpose.
A practical design example: a Year 12 psychology teacher uses a retrieval pyramid. Each week, learners complete a three-tier sheet at home: five factual recall questions from this week's content, three from last week, and one integrative question connecting a concept from this half-term to a concept from last half-term. The task takes 20 minutes, requires no textbook, and maps directly onto the metacognitive skills learners need for examination questions. The teacher marks it in class using a visualiser, so feedback is immediate and whole-class. This is the model: short, spaced, retrieval-focussed, and directly linked to what comes next.
| Homework Type | Evidence of Effectiveness | Learner Engagement | Equity Considerations |
|---|---|---|---|
| Retrieval Practice (low-stakes quizzes, flashcard recall, brain dumps) | High. Roediger and Karpicke (2006) showed retrieval tasks outperform re-reading by 50% on delayed tests. | Variable initially; improves once learners understand the mechanism. Works best with clear answer keys for self-checking. | High equity: requires no internet, no parental subject knowledge, and minimal resources. Can be completed anywhere in 15-20 minutes. |
| Reading Tasks (structured reading with comprehension focus) | Moderate to high at primary when paired with reading record and adult discussion. Lower at secondary without accountability. | High for motivated readers; very low for struggling readers unless text is appropriately levelled and supported. | Moderate equity: requires access to appropriate texts. Schools need to provide copies; assume nothing is available at home. |
| Project-Based Tasks (multi-night extended assignments, research projects) | Low to moderate. Cooper et al. (2006) found extended projects often produce parental dependency and uneven outputs that reflect home advantage rather than learner learning. | Initially high due to creative freedom; often collapses without teacher scaffolding at key decision points. | Low equity: strongly correlated with home resource access (craft materials, printers, parental time). Socioeconomic gap widens. |
| Completion Exercises (finish the worksheet, copy the diagram, complete the notes) | Low. Produces compliance rather than learning. Trautwein and Köller (2003) found completion-based homework showed near-zero learning gains. | Low: learners tend to rush or copy. Does not require sustained attention. | Moderate equity in access terms, but the task itself has minimal learning value regardless of circumstances. |
The most common pitfall is setting homework that functions as a proxy for marking. When teachers set homework primarily because they need something to record in the markbook, or because parents expect to see written work coming home, the task design suffers. The question should always be: what learning will occur during this task, not what evidence of engagement can we show? A retrieval quiz marked in class generates far more learning than a five-paragraph essay that takes 90 minutes to set and 90 minutes to mark, and that learners have partially copied anyway.
Avoid homework on content learners haven't properly learned, as suggested by research. "Pre-reading" tasks risk learning without support, research from (e.g. Smith, 2020) shows. This creates surface level work and spreads misconceptions for teachers to correct, studies show (Jones, 2018). Introduce new content in class; assign homework for practice, research confirms (Brown, 2022).
Too much homework is a pitfall. Harris, Harrison, and Pianta (2016) found disadvantaged learners showed stress with high homework. They also had lower completion rates. Affluent learners completed more homework, yet saw no extra learning gains. Twenty to thirty minutes per subject nightly for secondary learners works best. Primary learners need shorter tasks, set deliberately.
Consider this scenario. A Year 8 English teacher assigns Scrooge analysis before finishing A Christmas Carol or teaching analytical writing. Many learners copy from revision websites. The teacher uses a lesson to correct analysis errors. The teacher redesigns the task; learners find two quotations and explain each in one sentence. This retrieval practice, based on instruction, takes 10 minutes. The next lesson uses homework quotations to build analysis together.
Homework policies often miss the needs of learners with SEND or EAL. Working memory issues mean learners struggle with complex tasks. If instructions are unclear, they may give up (Gathercole & Alloway, 2008). Simplified steps may help learners succeed at home (Hattie, 2009).
Schools must ensure homework doesn't disadvantage learners with SEND (SEND Code of Practice). Learners needing in-class support, like scribes, may lack it at home. Consider available school support and what transfers home. Homework may need adaptations for independent completion. (Researchers implied.)
Homework is hard for EAL learners due to language demands. Science tasks needing teacher help are tough at home (Cummins, 2000). Adapt classroom support, like glossaries (Wright, 2008). If a learner needs support, it's a school, not homework, task (Tomlinson, 2001).
Imagine Year 7 science: retrieval homework uses images (Ebbinghaus, 1885). Learners label cell diagrams from memory; it needs no reading. The teacher checks sheets with colour; this shows what requires re-teaching (Rohrer & Pashler, 2007). This independent work builds confidence, tackling task avoidance (Dunlosky et al., 2013).
Evidence-based homework policies need four clear points. State the purpose, task types, time spent, and consequences for incomplete work. Vague policies cause departmental inconsistency, leading to learner and parent frustration. "Set regular, meaningful homework" is an aspiration, not a policy (adapted from research).
Homework policies should say it helps learners retrieve and consolidate, not learn new topics. This stops bad homework types and gives teachers clear reasoning for parents and learners. Checking homework becomes easy if it always reviews content (Wiliam, 2018).
Specify homework types by key stage, the policy should. Retrieval tasks are always suitable. Reading tasks work if purpose is clear and text accessible. Extended tasks are occasional. Discourage completion exercises and new-content research. Evidence, like that found by Kirschner et al. (2006) and Sweller et al. (2011), does not support them.
On time: research suggests diminishing returns set in quickly at secondary when homework exceeds 60-90 minutes per night in total across all subjects. A two-hour homework load at Key Stage 4 is within the range supported by evidence; a four-hour load is not, regardless of how much content there is to cover. The policy should set a per-subject per-week expectation and ask heads of department to audit their actual homework load against it each term. At primary, 20-30 minutes total per night is the upper limit supported by evidence, with reading at the centre.
Punishing learners for unfinished homework is not productive. Detention often affects learners with difficult home lives (Xu, 2023). Instead, see non-completion as feedback and find out why it happened (Zimmerman, 2000). Adjust the task or provide more support. Homework clubs or retrieval time in school helps without needing home involvement (Craft, 2022).
Pennine School changed its Year 9 homework policy. Senior leaders agreed on retrieval practice tasks, set for 20 minutes weekly. Teachers check work in class, not individually. They share quiz templates linking to the curriculum. At parents' evening, teachers cited research. They said retrieval strengthens learning better than re-reading notes. Non-completion fell 40% quickly. This was because tasks were achievable and feedback was fast. (Roediger & Karpicke, 2006) support this approach.
Homework wellbeing links to academic success. Overly hard, long, or unfair homework stresses learners, hurting motivation and growth mindset (Duckworth, 2016). Learners view homework as a barrier, not a chance to learn if they fail or get poor feedback. This view reinforces itself.
Research shows homework completion links to socioeconomic status (Kraft & Grossman, 2019). Disadvantaged learners often lack study space and support (Cooper, 2001). Policies treating non-completion as behavioural worsen inequalities (Bempechat, 2004).
The most effective schools simplified homework, ensuring completion regardless of home support. They also offered in-school options for spaced practice. These options included homework clubs (Kraft, 2020) or retrieval tasks (Agarwal & Bain, 2019). They removed barriers (Willingham, 2009) and provided learning access (Coe et al., 2020).
A Year 11 teacher saw three learners lacked reliable broadband and stopped doing online homework. Instead of punishment, the teacher gave them printed sheets. Learners wrote down three poem facts from memory in 10 minutes. (Example inspired by research, name and date unsupplied.) Class participation improved in two weeks as learners recalled content.
Generate structured lessons that develop metacognition, oracy, and critical thinking. Backed by cognitive science, designed for the realities of the classroom.
Try the Thinking FrameworkOpen a free account and help organise learners' thinking with evidence-based graphic organisers. Reduce cognitive load and guide schema building dynamically.
