How to Write a Discussion Section of a Research Paper (With a Worked Example)

how to write a discussion section
Table of Contents

Learning how to write a discussion section is hard when your results are ready but your argument isn’t.

A good discussion explains what your findings mean without repeating numbers or claiming too much. This guide gives you seven moves, a full worked example, sentence frames, and a checklist.

Turn your results into a clear argument.

Key Takeaways

  • Interpret, Don’t Repeat: The discussion explains what results mean and what limits them, with no new results.
  • Follow Seven Moves: Answer the question, interpret, compare, handle surprises, limit, apply, then point to future research.
  • Match Claims to Evidence: Use “showed” for measured results, “suggests” for inferences, and “may” for untested explanations.
  • Tie Limitations to Claims: Each limitation should name the claim it weakens and the likely direction of its effect.
  • Lock Meaning Before Editing: Protect numbers, terms, citations, and hedges before any person or AI tool edits for clarity.
Writing help

Struggling with Clarity in your writing?

You're not alone. Many writers face this exact challenge.

Orwellix provides you with advanced writing tools specifically designed to overcome common writing hurdles. Our AI-powered platform helps you craft clearer, more engaging content with less effort.

Start for Free

What a Discussion Section Does: Discussion vs Results vs Conclusion

The discussion section explains what your results mean, how far they reach, and what limits them. The Results section reports what you found. The conclusion gives the short take-home message.

Most weak discussions mix these jobs up. They repeat the numbers from the Results, or they jump straight to big claims the data can’t carry.

SectionIts jobWhat it sounds like
ResultsReport what you found“Quiz scores were higher on the delayed test.”
DiscussionExplain what it means and what limits it“This suggests retrieval practice helps recall last.”
ConclusionGive the short take-home“Short quizzes are a low-cost way to aid recall.”

The USC Libraries discussion guide puts two rules plainly: don’t introduce new results in the discussion, and don’t simply restate the results you already reported. If a number appears in the discussion for the first time, it belongs in the Results.

Journals don’t all split these sections the same way. Some combine Results and Discussion into one section. Others want a separate, short Conclusion after the discussion. Check your target journal’s author guidelines or your course brief before you plan.

If you do need a separate Conclusion of a paragraph or two, the free Conclusion Generator can draft one from your own text. It doesn’t write discussion sections, which need your interpretation of your own data.

The most useful way to think about a discussion is as a set of claims. Each claim has a strength, and that strength must match the evidence behind it. The rest of this guide builds on that idea.

Plan Your Discussion as a List of Claims

Before you write a sentence of the discussion, list the claims you want to make. Next to each one, note the result that supports it and how strong that support is.

This takes ten minutes and saves hours of rewriting. It shows you which claims are solid, which are guesses, and which you can’t make at all.

Use three strength levels:

  • Measured: You tested it directly, and the result is in your tables or figures.
  • Inferred: It follows from what you measured, but you didn’t test it directly.
  • Speculated: It’s a possible explanation that your study didn’t test.

Here is a claim list for the fictional study used throughout this guide. First-year students either took short retrieval quizzes or re-read their notes, then sat an immediate test and a test one week later.

ClaimEvidenceStrength
Quizzes improved delayed test scoresDelayed test, Table 2Measured
Quizzes did not change immediate scoresImmediate test, Table 2Measured
Retrieval strengthens long-term memoryPattern across both testsInferred
Quizzing would help in other coursesNone in this studySpeculated

Put the claims in the same order your Results section used. Then move the one that answers your research question to the top. If you already planned the paper with a research paper outline, add this claim list under its Discussion heading.

The Seven Moves of a Discussion Section

A strong discussion starts with the answer to your research question and works outward to its wider meaning: comparisons, limits, implications, and next steps. Seven moves cover that path, usually one paragraph each.

The moves aren’t a new invention. In a 1999 BMJ editorial, Docherty and Smith proposed a structured discussion in five parts, from principal findings to future research. They called the discussion “often the weakest part of the paper.” The APA JARS–Quant reporting standards ask psychology authors to cover the same ground: support for hypotheses, comparison with others’ work, interpretation, generalizability, and implications.

Seven moves

From Your Answer Out to the Field

Specific · your study
01Answer the question
02Interpret the finding
03Compare with prior studies
04Explain unexpected results
05State the limitations
06Spell out implications
07Point to future research
General · the field
Orwellixorwellix.com
Each move widens the view, from what your study found to what it means for the field.

Move 1: Answer the Research Question

Open with a direct answer to the question your paper asked. This closes the gap your introduction set up, so the reader sees the paper’s argument come full circle. For a refresher on that setup, see how to state the gap your introduction set up.

Frame: “This study asked whether [X affects Y]. The results suggest that [answer], at least for [population or setting].”

Usual failure: Opening with a statistic or a summary of methods. The reader already has those.

Move 2: Interpret the Main Finding

Explain why the main result might have happened. Offer the most likely explanation and say what kind of evidence supports it.

Frame: “One explanation for [finding] is that [mechanism]. This would account for [specific pattern in your data].”

Usual failure: Restating the result in new words instead of explaining it.

Move 3: Compare With Previous Studies

Show where your findings agree with earlier work, where they differ, and why. Name a reason for each difference, such as a different sample, measure, or time frame. A synthesis matrix from your literature review makes this move much faster.

Frame: “This result is consistent with [Author, Year], who found [similar result]. It differs from [Author, Year], possibly because [difference in method or sample].”

Usual failure: A list of citations that “support” your finding, with no comparison at all.

Move 4: Explain Unexpected or Null Results

Address the results you didn’t predict. A null result is still a result, and readers trust a discussion that faces it.

Frame: “Contrary to our hypothesis, [result]. This may reflect [possible reason], which future studies could test by [method].”

Usual failure: Skipping the result, or explaining it away with a reason you never checked.

Move 5: State the Limitations

Name the weaknesses that affect specific claims, and say how much they matter. The limitations section below shows a full method for this paragraph.

Frame: “Because [limitation], the finding that [claim] may not extend to [group or setting].”

Usual failure: A generic list (“small sample, one site”) that never says which claim it weakens.

Move 6: Spell Out the Implications

Say what your findings mean for practice, theory, or policy. Keep the reach of each implication inside the reach of your sample.

Frame: “If confirmed in [wider population], these findings would suggest that [practical or theoretical implication].”

Usual failure: Implications that cover every student, patient, or company when you studied one group.

Move 7: Recommend Specific Future Research and End With a Take-Home

Propose the next study your results point to: who, what, and how. Then close with one sentence that states the paper’s main contribution.

Frame: “Future studies could test [specific question] with [sample or design]. Overall, [one-sentence take-home].”

Usual failure: “More research is needed.” It tells the reader nothing about what to study next.

For more frames for each of these moves, including hedging and limitations, the free Academic Writing Phrasebook groups 301 sentence frames by purpose.

Discussion Section Example: One Study Through All Seven Moves

Here is a complete short discussion built with the seven moves. The study is invented for teaching, and so are its numbers. Citations appear as “[Author, Year]” placeholders where real sources would go.

The fictional study: 84 first-year students in one introductory psychology course volunteered for a four-week study. Half answered short retrieval quizzes after each lecture, and half re-read their notes. Both groups took an immediate test and a delayed test one week later. The quiz group scored higher on the delayed test only.

Move 1, answer the question: This study asked whether short retrieval quizzes improve exam performance more than re-reading. The results suggest that quizzing improved delayed recall, but not immediate recall, among first-year students in one course.

Move 2, interpret the main finding: One explanation is that retrieving information strengthens memory more than re-exposure does. This benefit may show up only after forgetting has had time to act. This would account for the absence of a difference on the immediate test.

Move 3, compare with previous studies: The delayed benefit is consistent with laboratory studies of the testing effect [Author, Year]. It is smaller than the effect reported in a similar classroom study [Author, Year], possibly because our quizzes were optional and some students skipped them.

Move 4, address the unexpected result: We expected quizzing to help on the immediate test as well. The null result there may reflect a ceiling effect, since both groups averaged above 85 percent. A harder immediate test could show whether a short-term benefit exists.

Move 5, state the limitations: Because students chose to take part, the sample may over-represent motivated learners. This limits the claim that quizzing helps first-year students in general and may inflate the delayed benefit. Random assignment to groups protects the comparison between conditions, but it cannot change who volunteered.

Move 6, spell out the implications: If replicated in required courses, these findings would suggest that brief weekly quizzes offer instructors a low-cost way to support long-term retention.

Move 7, future research and take-home: Future studies could test the effect with a whole-class sample, a longer delay, and the course’s final exam as the outcome. Overall, retrieval practice appears to help first-year students remember course material a week later, even when it shows no benefit right away.

Look at where the hedges sit. The measured fact uses a plain verb (“both groups averaged above 85 percent”). The interpretation uses “suggest” and “one explanation is.” The untested ideas use “may” and “could.” Each claim gets one hedge that fits its evidence, which is the skill the next section teaches.

Calibrate Every Claim to Its Evidence

A calibrated claim says exactly as much as the evidence supports, no more and no less. Hedging language in academic writing is how you set that level, with words like “suggests,” “may,” and “indicates.”

The UNC Writing Center’s guide to scientific reports notes that “supported,” “indicated,” and “suggested” are better ways to judge a hypothesis than “proved.” Middlebury’s Write Like a Scientist guide says hedging is used most often in the discussion section, and advises never using “prove” there at all.

Claim calibration

Say Exactly What Your Data Show

Overclaimed

“This proves quizzes improve learning.”

Calibrated

“These results suggest retrieval practice improved delayed recall in this sample.”

Certaintyproves → suggest
Outcomelearning → delayed recall
Scope+ in this sample
Orwellixorwellix.com
A calibrated claim names what was measured, who it applies to, and how certain the evidence is.

Overclaimed vs Calibrated Sentences

Overinterpretation of results is the most common problem in a discussion. Each pair below fixes one kind of overclaim from the fictional study.

  1. Certainty: “This proves quizzes improve learning.” becomes “These results suggest retrieval practice improved delayed recall in this sample.”
  2. Scope: “First-year students benefit from weekly quizzes.” becomes “First-year students in one introductory course benefited from weekly quizzes.”
  3. Untested cause: “Retrieval strengthened students’ memory traces.” becomes “One likely explanation is that retrieval strengthened memory for the tested material.”
  4. Recommendation: “Instructors should replace re-reading with quizzes.” becomes “If replicated, these findings would support adding short quizzes to existing study methods.”
  5. Null result: “A ceiling effect explains the immediate test result.” becomes “The null immediate result may reflect a ceiling effect.”

Hedging can also go too far. Stacked hedges make a claim sound weaker than your data are.

Before: “These results may possibly suggest that quizzing could perhaps help recall.” After: “These results suggest that quizzing helped delayed recall.” The second version uses one hedge, which is all a measured result with an inferred meaning needs.

The Hedging Ladder

Match each verb to the kind of evidence behind the claim. The word list draws on the hedging words in Middlebury’s guide.

Evidence levelTypical wordingExample
Measured resultshowed, found, was higher“The quiz group scored higher at one week.”
Inferencesuggests, indicates, supports“This suggests retrieval aids later recall.”
Untested explanationmay, might, could“Motivation may have inflated the effect.”
Speculationlabel it as speculation“One speculative possibility is that…”

Two rules keep the ladder honest. Use one hedge per claim, placed on the verb that carries the claim. Never use “prove,” because a single study can support a claim but can’t settle it.

If your thesis has claims your method can’t support, fixing the wording is only half the job. Check that the claim matches your design before you choose its verb.

Write Limitations That Name the Claim They Limit

A useful limitation names the claim it weakens. Write each one in four parts: the limitation, the claim it constrains, the likely direction of its effect, and what you did about it.

Limitations

Every Limitation Points at a Claim

LimitationOne-week delay
Claim it limitsLong-term retention
Likely directionMay overstate it
What you didSame delay for both
Orwellixorwellix.com
A limitation earns its place when it tells the reader which claim to read more cautiously.

To find candidates, start with the APA JARS–Quant standards. They list the usual sources: potential bias, imprecise measures, the number of tests run, sample size, and how far the findings generalize. Go through that list and keep only the limitations that touch a specific claim.

Limitation: [what was limited in the design, sample, or measures]
Claim it constrains: [which finding or interpretation it weakens]
Likely direction: [would it inflate, shrink, or blur the effect?]
What you did: [design choice, check, or analysis that reduces it]

Here is that template filled in for the fictional study:

Example: The delayed test came only one week after the last lecture. This limits the claim that quizzing improves long-term retention, since one week is short compared with a full term. Over a full term, the gap could narrow as both groups forget, so the one-week result may overstate the lasting benefit. We used the same delay for both groups, so the comparison between conditions remains fair.

Notice the tone. The paragraph doesn’t apologize, and it doesn’t list every flaw a study could have. It gives the reader exactly what they need to judge one claim.

Limitation vs delimitation: A limitation is a weakness you didn’t choose, such as a low response rate. A delimitation is a boundary you set on purpose, such as studying one course.

Two or three well-built limitations are enough for most papers. A long list of vague ones (“small sample, short study, one site”) tells reviewers you haven’t thought about which ones matter.

Tense, Voice, and Order in the Discussion

Use the past tense for what happened in your study and in specific earlier studies. Use the present tense for established knowledge and for what your results mean now.

The USC guide recommends the present tense for established facts and the past tense when you refer to specific prior studies. UNC’s report guide adds that your own data belong in the past tense.

  • Your results (past): “The quiz group scored higher on the delayed test.”
  • A specific study (past): “[Author, Year] found a similar effect in a lab setting.”
  • Established knowledge (present): “Retrieval practice is a well-studied learning strategy.”
  • Interpretation (present): “This pattern suggests that the benefit needs time to appear.”

Voice depends on your field and journal. Many sciences accept “we” in the discussion, and some prefer the passive. For the rules by section, see this guide to passive voice in academic writing.

Order matters as much as grammar. Discuss findings in the same order as your Results section, and use the same terms your introduction used. If the introduction said “retrieval practice,” don’t switch to “self-testing” in the discussion.

Thesis Discussion Chapters and Discipline Differences

The seven moves work in most fields, but their size and order shift with the type of document. Your department handbook or journal guidelines always win over general advice.

  • IMRaD papers: Most empirical articles follow the IMRaD structure (Introduction, Methods, Results, and Discussion). Here the discussion is usually one section of a few paragraphs.
  • Thesis discussion chapter: A thesis chapter is longer and answers every research question in turn. It often opens with a short recap of the aims and ties each finding back to the literature review chapter.
  • Psychology and APA journals: Check the JARS discussion items: hypotheses, comparison, interpretation, generalizability, and implications.
  • Qualitative studies: You interpret themes rather than effects. Many qualitative researchers discuss transferability instead of generalizability and reflect on how their own position shaped the analysis.
  • Multi-study papers and chapters: Give each study a short discussion, then close with a general discussion that compares them.

Whatever the format, the core logic stays the same. Each claim should match its evidence, and each limitation should name the claim it affects.

Revise Your Discussion for Clarity Without Changing Its Meaning

You can revise a discussion for clarity without changing its science if you decide in advance what no editor may touch. That applies to a supervisor, a paid editor, a co-author, or an AI tool.

Clarity edits are risky in a discussion because the meaning lives in small words. Swap “suggests” for “shows,” and the claim gets stronger. Swap “retrieval practice” for “testing,” and the variable changes.

Meaning-safe editing

Lock the Science, Edit the Sentences

LockedNever changes
  • Numbers & statistics
  • Technical terms
  • Citations
  • Hedge words
  • Claim direction
Free to improveClarity only
  • Sentence length
  • Repetition
  • A vague “this”
  • Transitions
Orwellixorwellix.com
Give any editor, human or AI, the locked list first, then let them work only on clarity.

Make a Lock List Before Anyone Edits

A lock list names every element that must come back unchanged. Write it at the top of the file or in your instructions to the editor.

  • Numbers and statistics: Every value, percentage, test statistic, and confidence interval.
  • Variable and technical terms: The exact names you used in the Methods and Results.
  • Citations: Every reference, in the same place and attached to the same claim.
  • Hedges: Every “suggests,” “may,” “indicates,” and “appears.”
  • Claim direction: What increased, decreased, or showed no difference.

Allowed and Forbidden Edits

Allowed: splitting long sentences, cutting repetition, fixing a vague “this,” removing stacked hedges you’ve approved, and smoothing transitions within a paragraph.

Forbidden: making a claim stronger or weaker, changing a term, adding findings or citations, and deleting or softening a limitation.

Before: “This could possibly be due to the fact that it was the case that the quizzes were optional, and this may have reduced the effect.” After: “This smaller effect may reflect the optional quizzes.” The edit cuts 17 words and fixes the vague “this.” It keeps one hedge, the cause, and the direction of the claim.

What Journals Allow for AI Language Help

Publishers generally allow AI help with language, with conditions. Elsevier’s generative AI policy lists improving the language, clarity, and readability of author-written text as appropriate use. It requires human oversight and a separate AI declaration, though basic grammar and spelling checks don’t need one.

The ICMJE recommendations say journals should require authors to disclose AI use at submission. Humans stay responsible for any submitted text, so authors must check AI output carefully.

In practice, that means three steps. Give the tool your lock list, review AI-generated edits one at a time, and follow your journal’s rules to disclose or cite AI use. If the problems go beyond the discussion, such as a gap between your question and your data, see these research paper writing mistakes first.

Discussion Section Checklist

Run this checklist on your draft before you send it to a supervisor, co-author, or journal. Each item takes a minute to check.

  1. The first paragraph answers the research question directly.
  2. Each paragraph opens with a claim, not a number.
  3. No result appears for the first time in the discussion.
  4. Every claim traces back to a specific result, table, or figure.
  5. Each hedge matches its evidence level, with one hedge per claim and no “prove.”
  6. Comparisons with earlier studies say where you agree, where you differ, and why.
  7. Unexpected and null results are named and discussed.
  8. Each limitation names the claim it constrains and its likely direction.
  9. Implications stay inside the reach of your sample.
  10. Key terms match the ones in your introduction and Methods.
  11. Future research names a specific question, sample, or design.
  12. The section ends with a one-sentence take-home.

Edit Your Discussion Section in Orwellix Without Changing the Science

Orwellix is an AI writing agent that proposes edits to your document as tracked changes. You accept or reject each one, so nothing in your discussion changes without your approval. It won’t judge whether your interpretation is right, and it can’t promise acceptance or a grade. What it can do is speed up the clarity pass while your lock list stays intact.

  1. Bring in the whole manuscript. Import a DOCX, TXT, or Markdown file up to 15 MB, or paste your text. With the full paper in the document, the assistant can read your Results and Discussion together.
  2. Find overclaims in Ask Mode. Ask Mode answers in the chat and doesn’t edit your document. It costs 1 credit per request. Ask: “Which sentences in my Discussion claim more than my Results report?”
  3. Edit for clarity in Agent Mode. Agent Mode costs 2 credits per request. Paste your lock list into the prompt: “Edit only the Discussion for clarity and concision. Do not change any number, statistic, citation, technical term, or hedge word. Do not make any claim stronger or weaker, and do not add findings.”
  4. Review each tracked edit. Accept or reject every change against your lock list. Reject any edit that touches a locked item, even if it reads better.
  5. Check the final text. Use readability analysis to find hard sentences and wordy passages. The grammar checker covers 5,000 characters a month on Free, 50,000 on Pro, and 150,000 on Premium, so a full thesis chapter may need Pro. Plagiarism checks come with Pro (5,000 words a month) and Premium (15,000), not Free.
  6. Export and disclose. Download the document as DOCX or PDF, then add the AI disclosure statement your journal or university requires.

The Free plan includes 25 AI credits a month, with no card needed.

Write smarter with Orwellix

Orwellix's AI capabilities help you craft clearer, more effective content.

Start for Free

Conclusion

A strong discussion section does a few clear jobs. It answers the research question first, interprets the main finding, compares it with earlier studies, and faces unexpected results. It then names its limitations, states realistic implications, and points to a specific next study. Behind every move sits the same rule: each claim should be exactly as strong as the evidence that supports it.

That rule ties the whole process together. The claim list sets the order, the hedging ladder sets the wording, and claim-specific limitations set the boundaries. The lock list then protects all three during revision, so clearer sentences never come at the cost of changed meaning.

As more researchers use AI tools to polish their writing, that last step matters more each year. Journals now expect authors to disclose AI help and to stay responsible for every claim. Orwellix fits this workflow by proposing clarity edits as tracked changes that you approve one by one. The interpretation, and the judgment behind it, stays yours.

Frequently Asked Questions (FAQs)

1. How long should a discussion section be?

There’s no fixed length, but a journal article’s discussion usually runs several paragraphs, roughly one per move or main finding. A thesis discussion chapter runs much longer because it answers every research question. Check your journal’s word limits or your department handbook, and cut any paragraph that only repeats results.

2. How do I start a discussion section?

Start by answering your research question in one or two sentences. For example: “This study asked whether weekly quizzes improve exam scores. The results suggest they improved delayed recall but not immediate recall.” Avoid opening with a statistic or a summary of your methods.

3. What is the difference between a discussion and a conclusion?

The discussion interprets your results, compares them with other studies, and explains their limits and implications. The conclusion is the short take-home message, often a final paragraph or a brief separate section. Some journals want both, and others fold the conclusion into the discussion.

4. Can I cite new sources in the discussion?

Yes, you can cite sources you didn’t use in the introduction when they help explain or compare your findings. What you shouldn’t add are new results from your own study. Most of your key sources should still appear earlier in the paper.

5. Can I use AI to edit my discussion section?

Many publishers allow AI help with language and clarity, as long as you oversee every change and disclose the use where required. Give the tool a lock list of numbers, terms, citations, and hedges that must not change. Then review each edit yourself, because you remain responsible for every claim.

Try it yourself

Try Orwellix for Free

Experience Orwellix AI Agent's capabilities on the free forever plan, with 25 AI credits a month and no card needed. Need more? Try Pro or Premium free for 7 days.

Start for Free