The practical outcome
- A matrix that keeps reported evidence separate from your reading of it.
- A worked synthesis of three fictional studies.
- A checklist for avoiding unsupported gap claims.
This original exercise uses fictional study cards. They are not publications and must not be cited as evidence. Use them to practise the reasoning, then replace them with verified sources from your own reading. You can recreate the matrix in a spreadsheet or a plain document.
Begin with a question that makes comparison possible
“Feedback in language teaching” is a topic. “Under what task conditions do learners use peer feedback to revise an argument?” is a question that suggests what to extract: the task, the feedback process, and evidence of revision. It also helps you decide whether a study is relevant even when its title uses different vocabulary.
Choose the columns before collecting large quantities of notes. Start with bibliographic details and a page or section locator, then add context, design, reported result, limitation, and relevance to your question. Keep direct quotations visibly marked. A source locator saves time when a promising sentence later needs verification.
Try the matrix with three study cards
| Card and context | Evidence described | What remains uncertain |
|---|---|---|
| A: A class revises essays after exchanging comments. | Later drafts contain more reasons supporting the writer’s position. | There is no comparison condition; additional drafting time may matter. |
| B: Learners compare two feedback formats on short paragraphs. | More comments are incorporated with a checklist. | Incorporation is counted; the quality of the resulting argument is not assessed. |
| C: Learners discuss feedback in interviews. | They describe trust in a peer as a reason for accepting advice. | Accounts of decisions are not direct observations of revision. |
Now add a column headed My interpretation. For A, you might write “revision improved after the activity, but this design does not isolate the contribution of peer feedback.” For B, note that using a comment and improving an argument are different outcomes. For C, trust is a possible explanation to investigate, rather than a measured effect.
This separation matters when you return to the matrix weeks later. A source’s result and your inference can otherwise merge into one apparently established statement. Record uncertainties while reading, rather than trying to reconstruct them during drafting.
Write across the rows
A source-by-source paragraph would retell A, then B, then C. A synthesis instead groups them around a distinction. Here, the distinction is between taking up feedback, improving a text, and explaining a decision.
The three study cards treat feedback use as different kinds of evidence. A examines changes in argument support, B counts incorporated comments, and C elicits learners’ accounts of their decisions. Together they suggest a need to distinguish uptake from improvement and to investigate how learners explain their choices. They do not establish that a checklist produces stronger arguments or that trust causes revision.
The final sentence is doing analytical work: it prevents a relationship assembled by the reviewer from becoming a causal conclusion. In a real review, attach verified citations to the relevant claims and check that the paragraph still represents each source fairly.
Turn a possible gap into a bounded contribution
Avoid moving from “I have not found a study” to “No research exists.” Your search coverage may be limited by databases, language, terminology, or dates. Keep a separate search log: where you searched, the query, the date, the inclusion rules, and how you followed relevant references.
A contribution based on this exercise might be: “The proposed study will examine feedback uptake and changes in argument quality within the same revision sequence.” That describes what the design will connect. It still needs a real literature search to justify why the connection matters and whether it has already been examined.
Before you draft from your matrix
- Can you locate every extracted claim in its original source?
- Have you recorded what an outcome measure actually captures?
- Are contradictory results linked to relevant differences rather than dismissed?
- Can a reader distinguish the source’s interpretation from yours?
- Does the proposed contribution follow from the comparison?
If you use AI to organise notes, compare every resulting claim with the source and retain the source locators. See our task-by-task guide to AI in research. For the next stage, our guide to writing for international journals considers how to frame a contribution for readers.