The discussion section explains what your results mean: how they answer your research question, how they compare with earlier studies, why any unexpected findings might have occurred, what limits the conclusions, and what follows for theory and practice. It interprets the results; it does not repeat them.
This guide is for scholars writing either the discussion section of a journal paper or the discussion chapter of a thesis. It gives a paragraph order that works, shows the difference between reporting and interpreting with worked examples, and covers limitations and implications, which are the two parts examiners and reviewers read most critically. The overall structure of a paper is covered in our research paper guide, and where the discussion sits in a thesis is covered in the thesis structure guide.
Last reviewed September 2026.
What should the discussion section do?
Examiners and reviewers read the discussion to answer one question: does this scholar understand what their findings mean? A results section can be accurate and still leave that question open. The discussion is where you close it.
It has five jobs. It answers the research question in plain words. It places the answer in the existing literature. It deals openly with anything unexpected. It sets out the limits of what can be concluded. And it says why any of this matters, to theory, to practice, or to the next researcher.
| Results section says | Discussion section says |
|---|---|
| Engagement was higher among employees with 2–3 office days a week (M = 3.9) than among fully remote employees (M = 3.4), p < .01. | A moderate amount of office time appears to support engagement, which fits the argument that some face-to-face contact maintains team identity (earlier studies). Fully remote work may remove informal contact that engagement depends on. |
| The moderation effect of supervisor support was not significant (β = 0.06, p = .21). | Contrary to H3, supervisor support did not change this relationship. One possible reason is that the scale measured general support rather than support specific to remote work; a second is limited variation, since most respondents rated their supervisors highly. |
| Accuracy of the proposed model was 94.2% against 91.8% for the best baseline. | The gain of 2.4 points is modest but was consistent across all five folds. It came mainly from fewer false positives on minority attack classes, which matters more in practice than overall accuracy. |
Look at what changes between the two columns. The numbers mostly disappear. Words like “appears”, “may” and “one possible reason” appear. Each result is now tied to an explanation and to other work.
What order should the discussion follow?
There is no single required structure, but the order below suits most empirical papers and thesis chapters, and reviewers find it easy to follow.
State the main finding
One or two sentences answering the research question, without statistics.
Compare with earlier work
Where your result agrees with published studies, where it differs, and why it might differ.
Explain the surprises
Unexpected or non-significant results, with plausible explanations you label as such.
Limitations
What limits the conclusions, and in which direction each limitation might push the results.
Implications
What follows for theory, for practice or policy, and for future research.
Start with the answer, not the background
The first paragraph should state your main finding in a sentence or two, without statistics. Many scholars open with a paragraph restating the aim and the method. The reader has just read those. Start with what you found.
One finding per block
If you have three or four main findings, give each its own block of one to three paragraphs: the finding, how it compares with earlier work, and what explains it. Take them in order of importance, or in the order of your research objectives. Examiners like the second, because they can tick off each objective as they go.
Limitations and implications at the end
Some journals want a separate “Limitations” heading, and some want “Theoretical implications” and “Practical implications” as sub-headings. Check the author guidelines and two or three recent articles in your target journal.
How do you compare your findings with earlier studies?
This is the part that shows whether you know your field. Three patterns cover most cases (the author names below are placeholders).
Agreement. “This is consistent with Sharma and Rao (2021), who found a similar effect among bank employees.” Useful, but do not stop there. Agreement with a study in a different setting extends what we know; say how. If your finding simply repeats what five earlier studies found, your contribution lies elsewhere, so say where.
Disagreement. “In contrast, Lee et al. (2020) reported no relationship.” Then try to explain the difference: a different population, a different measure, a different period, a different analysis. This is often the most interesting paragraph in the discussion. Do not hide it.
Nothing to compare with. If you have genuinely found no earlier study on this relationship, say so carefully (“We are not aware of earlier studies that…”). Then compare with the nearest related work. Examiners are suspicious of “no study has ever…”, so check before you write it.
Cite studies you have actually read. Pulling citations from another paper’s discussion without reading them is how errors spread, and viva examiners do ask what a cited study found.
How do you discuss unexpected or non-significant results?
Every real study has some. Handled well, they make the discussion more credible, not less.
- Report them plainly. “H3 was not supported.” Do not bury it in a subordinate clause.
- Offer one or two plausible explanations, labelled as possibilities: measurement, sample characteristics, context, timing, or the theory itself being wrong for this setting.
- Do not invent a new theory to rescue the hypothesis. If you find yourself writing a page to explain one non-significant path, that is a sign to step back.
- Say what would settle it. “A study using a remote-work-specific support scale would show whether…” turns a disappointment into a useful direction for future research.
A non-significant result is not the same as “no effect”. It means your data could not show an effect of that size with confidence. If your sample was modest, say that the study may have been underpowered for small effects.
- Repeating the results. If a paragraph could be moved into the results section without changing a word, it is in the wrong place.
- New results in the discussion. Every number you discuss must already appear in the results.
- Citing only studies that agree. Examiners know the literature. Leaving out the study that found the opposite looks either careless or selective.
- Overclaiming. A cross-sectional survey shows association, not cause. “X leads to Y” should usually be “X is associated with Y”.
- Explaining away every non-significant result. Some hypotheses are simply not supported. Say so, give one or two plausible reasons, and move on.
- A list of limitations with no consequence. “The sample size was small” is not enough. Say what that means for how far the findings can be trusted or generalised.
- Generic implications. “Managers should take note of these findings” gives the reader little to act on. Name who should do what differently.
How do you write limitations and implications?
Limitations
Every study has limitations, and examiners know that. What they look for is whether you understand yours. For each limitation, give three things: what it is, how it might affect the results, and what you did about it or what a future study could do.
Weak: The study was limited to Chennai.
Better: Data came from IT employees in Chennai, where most firms in the sample required at least two office days a week. Engagement patterns may differ in cities or firms where fully remote work remains common, so the findings on fully remote employees rest on a smaller subgroup (n = 64) and should be read with caution.
Three to five real limitations is typical. A page of them, including trivial ones, weakens the thesis. Do not list limitations that are really design choices you can defend; defend them in the methodology instead.
Implications
Theoretical implications say what your findings add to, qualify or contradict in the theory you used. If your framework predicted X and you found X only under condition Z, that is a theoretical contribution. State it.
Practical implications say who should do what differently. Name the actor: HR managers in mid-sized IT firms, district education officers, hospital infection control committees. Keep it proportionate. A survey of 300 employees in one city does not justify recommendations for national labour policy.
Future research should follow from your limitations and surprises, not be a generic list. Two or three specific suggestions are better than ten vague ones.
How is a thesis discussion chapter different from a paper’s discussion?
The logic is the same. The scale and placement differ.
- Placement. Some theses have a single discussion chapter after all results chapters. Others, especially in science and engineering, discuss results at the end of each results chapter and then have a short overall discussion or “summary and conclusions” chapter. Follow your university’s format and your department’s custom; look at recent theses on Shodhganga from your department.
- Length. A thesis discussion can run to 20–40 pages. That space is for depth on each objective, not for repeating the literature review.
- Integration. A thesis discussion should bring together findings across chapters or studies. If chapter 4 is a survey and chapter 5 is interviews, the discussion is where you show how they fit, or why they do not.
- Conclusion. The thesis conclusion chapter answers each research question briefly and states the contribution. It should not introduce new discussion.
Examiners often draft their viva questions from the discussion and limitations. Anything you claim there, expect to defend. The viva questions guide lists the ones that most often come from this chapter.
Language that helps
Use the past tense for what you found (“engagement was higher”) and the present tense for what it means and for established knowledge (“this suggests”, “earlier work shows”). Hedge in proportion to the evidence: “suggests” for a single survey, “shows” only for strong, replicated evidence. Scholars writing in a second language often over-hedge every sentence or hedge none. Aim for one level of caution per claim, not three.
Sources
FAQ
Questions scholars ask
Can I combine results and discussion?
Many journals allow a combined “Results and Discussion” section, especially in chemistry, engineering and materials science. Check the author guidelines. Even when combined, keep the interpretation clearly separate from the reporting within each paragraph.
Should I cite new literature in the discussion that was not in my literature review?
Yes, if an unexpected finding needs it. You may need studies you did not anticipate. Keep it limited, and if a source becomes central, consider adding it to the literature review too.
How long should the discussion be?
In a journal article, often about as long as the results, commonly 1,200 to 1,800 words. In a thesis, long enough to address each objective in depth; length for its own sake does not impress examiners.
What is the difference between the discussion and the conclusion?
The discussion interprets and argues. The conclusion states briefly what was found for each question and what the study contributes, without new argument or new citations. Our conclusion guide covers papers, literature reviews and the final thesis chapter.
My results contradict my supervisor’s earlier paper. What do I do?
Report it clearly and discuss possible reasons, as you would for any study. Talk to your supervisor early. Most supervisors respect a well-argued difference far more than a result adjusted to fit.
