Writing

Guides and templates for writing up your research.

The introduction

An introduction has two parts. The opening tells the whole story of your study in plain language. The rest steps back and builds the theoretical case, narrowing from the broad research area down to your research questions.

The shape of an introduction: a short plain-language opening, then a theory section that narrows from the research area down to the research questions. The opening Context · Problem · Solution then step back and narrow down Research area Research topic Research problem What we know What we don’t know Current study Questions

Each section below sits beside an excerpt from an example introduction on the reactivity of metacognitive ratings. Read the excerpts in order to see how the pieces fit together.

  1. Part 1

    The opening

    Three short paragraphs that give the reader the whole study before any theory.

    • No jargonA reader should be able to follow the opening without any definitions.
    • A mini-abstractA reader who stops after the opening should still know the problem, why it matters and what you did.

  2. Context

    Situate the reader and set the context of the problem you are trying to solve.

    When psychologists want to know how well people judge their own thinking, they usually just ask them. How confident are you in that answer? How likely are you to remember this later? These ratings appear in hundreds of studies of learning, memory and decision-making.

  3. The problem

    Specify the problem, and explain why it is a problem.

    But asking the question may change the answer. A growing body of work suggests that making these ratings can change how people learn and perform. If so, the measure is altering the very thing it is meant to measure.

  4. Your solution

    Explain how you are going to solve the problem in this study.

    In this study, we compared people who rated their confidence with people who did not, to test whether rating changes performance on [task].

  5. Part 2

    The theory

    Step back, then narrow down. Each section is a little more specific than the one before, until the reader arrives at your research questions.

  6. Research area

    Now step back and start with very general background. How far you step back depends on the length of the piece, but it is usually the research area.

    Metacognition refers to the ability to monitor and control one’s own cognitive processes (Flavell, 1979).

  7. Research topic

    Narrow to your research topic within that area.

    Metacognition is most often measured with self-reported ratings, such as judgements of learning or confidence ratings, collected during or after a task.

  8. Research problem

    Introduce the specific research problem your study addresses.

    These measures rest on an assumption that is rarely tested: that asking for a rating does not change the processes being measured. When it does, the measure is said to be reactive.

  9. What we know

    Review what we already know about the research problem.

    Organise this by idea, not by paper. Group studies around a finding or argument rather than giving each paper its own paragraph (“Smith (2019) found… Jones (2020) found…”).

    Judgements of learning are now known to be reactive. Making them improves memory for related word pairs, but has little effect on unrelated pairs (Double et al., 2018; Soderstrom et al., 2015). Confidence ratings can also change performance on reasoning tasks (Double & Birney, 2017).

  10. What we don’t know

    Set out what we don’t yet know about the research problem, and what we might expect to find.

    Make the gap concrete. “No one has studied this” is not enough on its own. Say why it matters that we don’t know, and what the possible results would tell us.

    However, it is unclear whether ratings are reactive in [context]. This matters because [why it matters]. If reactivity reflects a change in how people approach the task, we would expect [prediction].

  11. Current study

    Summarise the current study and your approach: how it will answer what we don’t yet know.

    In the current study, participants completed [task] either with or without confidence ratings. This allowed us to test whether [gap].

  12. Research questions and hypotheses

    Finish with your research questions, your hypotheses, or both.

    We asked whether making confidence ratings changes [outcome], and whether any effect depends on [moderator].

  13. General principles

    • Lead with topic sentencesOpen each paragraph with a sentence that states its point. Reading only the first sentence of every paragraph should give the whole argument.
    • Each paragraph sets up the nextThe research problem creates the need to review what we know, the gap creates the need for your study, and so on. If a paragraph doesn’t lead anywhere, cut it or move it.
    • Define terms once and keep themIntroduce each technical term when it first appears in the theory section, then use the same word throughout rather than switching between synonyms.

The method

The method should give another researcher everything they need to run your study again. Use these subheadings.

Participants
Who took part, how many, where they were recruited and what they received for taking part. Give the age and gender breakdown, how you decided on the sample size, and how many people you excluded and why.
Design
The independent and dependent variables, and whether each manipulation was between or within participants.
Materials
The tasks, measures and stimuli, in enough detail to rebuild them. Link to the materials if you can share them.
Procedure
What participants did, in order, and how long it took. Include the ethics approval and, if you preregistered, a link to the preregistration.
Analysis
Your planned analyses and the software and packages you used. Say where the data and analysis script are available.

Claude or ChatGPT can write a good first draft of the method, especially if you ask in the same chat you used to build the task. It already knows the design, the trials and the exclusions. Check every detail against the experiment itself.

The results

The results report what you found, without interpreting it. Save what it means for the discussion.

  • Start with the data. Say how the data were screened, how many participants were excluded, and give the descriptive statistics.
  • Follow your research questions. Report the analyses in the same order as the questions or hypotheses at the end of your introduction.
  • Report each test in full. Say what was tested, give the test statistic, p value and effect size with its confidence interval, then say in one plain sentence what the result shows.
  • Let tables and figures carry the detail. Use the text to point out what matters rather than repeating every number.
  • Keep exploratory analyses separate. Label any analysis you did not plan, after the planned ones. This doesn’t apply to basic descriptives such as means, standard deviations and correlations between your measures. You don’t need to have planned those, and you don’t need to call them exploratory.

The same applies here. In the chat where you analysed the data, Claude or ChatGPT already has every number and can draft a solid results section. Check each number against your own R output before you use it.

The discussion

The discussion mirrors the introduction. The introduction narrows from the broad research area down to your questions. The discussion starts with your answers and widens back out to what they mean for the field.

The discussion widens out: from a summary of your findings to their broader implications, ending with a conclusion.Summary of findingsInterpretationImplicationsLimitationsFuture directionsConclusion
Summary of findings
Restate the aim of the study, then answer each research question in plain language, in the same order as the introduction. Leave out the statistics.
Interpretation
For each finding, say what it means and how it fits with the work you reviewed in the introduction. Where your results disagree with past work, offer possible reasons.
Implications
Return to the problem from your opening and say what your findings mean for theory, measurement or practice.
Limitations
Consider other explanations for your results. Be specific: say how each limitation could have affected your conclusions. Skip generic limitations, such as “the sample was mostly students”, unless they matter for this question.
Future directions
Suggest concrete next studies that follow from your findings or limitations, rather than “more research is needed”.
Conclusion
End with a short paragraph giving the take-home message.
  • No new results. Everything you discuss should already be in the results section.
  • Interpret, don’t repeat. The results say what you found. The discussion says what it means.
  • Don’t overclaim. Match the strength of your conclusions to the strength of your evidence, especially for null or exploratory results.

Review your draft

Once you have a full draft, use Claude or ChatGPT to review it with the prompts below. If you write in Word, the easiest way is the Claude plugin for Microsoft Word, which runs the prompts on your document and can leave its feedback as comments or make edits directly. Run the prompts one at a time, in this order. Doing them separately, rather than in one giant “check everything” prompt, gets you sharper, less generic feedback.

  1. 1Logic and reasoningChecks whether every claim is supported, and finds gaps in the argument. 
    Read this draft as a critical reviewer focused only on the argument, not the writing style. For each claim I make, check whether it is actually supported by the evidence or reasoning I provide. Identify:
- Any claims that are unsupported, overstated, or not backed by what I've cited
- Gaps in the logical chain (where a conclusion doesn't follow from the premises)
- Circular reasoning or unstated assumptions I'm relying on
- Places where I've ignored an obvious counterargument or alternative explanation
- Any internal contradictions between different parts of the draft

For each issue, quote the specific sentence and explain the problem in one or two sentences. Do not comment on grammar, style, or formatting.
  2. 2Clarity and organisationReads it as a newcomer and lists the issues that hurt readability most. 
    Read this draft as someone unfamiliar with the topic. Assess:
- Whether the overall structure makes sense (does each section build logically on the last?)
- Whether the purpose of each paragraph is clear within the first sentence or two
- Any paragraphs that try to do too much or wander off topic
- Sentences that are unclear, ambiguous, or require rereading to understand
- Whether transitions between sections/paragraphs actually connect the ideas, or just sit there

Give me a short list of the 5-8 highest-impact issues, ordered by how much they hurt readability. For each, quote the passage and suggest a fix. Do not comment on grammar or formatting.
  3. 3ConsistencyFinds terms, numbers and claims that don’t match across the draft. 
    Check this draft for internal consistency only. Look for:
- Terminology used inconsistently (e.g. switching between two terms for the same concept)
- Numbers, statistics, or facts stated differently in different places
- Claims made early in the draft that are contradicted or quietly abandoned later
- Inconsistent tense, voice, or level of formality across sections
- Inconsistent citation style or referencing format

List each inconsistency with both locations quoted side by side.
  4. 4Copy editSentence-level fixes, given as before-and-after pairs you can accept or reject. 
    Copy edit this draft at the sentence level. Fix:
- Grammar, punctuation, and spelling errors
- Awkward or clunky phrasing
- Unnecessary wordiness or redundancy
- Passive voice where active would be stronger

Do not change my arguments, structure, or content. Give me the edits as a list of "original -> revised" pairs, not a full rewritten draft, so I can see exactly what changed and decide whether I agree.
  5. 5Formatting checkHeadings, lists, references, figures and tables. 
    Check this draft for minor formatting issues only:
- Inconsistent heading levels or heading style
- Inconsistent spacing, bullet/numbering style, or indentation
- Citation and reference list formatting errors (missing details, inconsistent style, references not matching in-text citations)
- Figure/table numbering or caption issues
- Any obvious layout problems (e.g. orphaned headings, inconsistent font/size cues if described in the text)

List each issue with its location. Do not comment on content, argument, or wording.
  • Logic first, formatting last. There’s no point polishing sentences around a broken argument.
  • One job per prompt. If you combine them, the model tends to default to surface-level copy edits and under-reports structural or logical problems.
  • Diagnose, then decide. The prompts flag problems; they don’t fix them. Read every flagged issue and decide whether to act on it. Don’t paste the output straight into your draft.