The discussion is where it shows most clearly whether you understand your own results. The numbers are already in the previous section; here you have to say what they mean, and it is easy to get it wrong in both directions: either you repeat the results in other words, or you launch into claiming more than your data allow. The second is one of the most frequent criticisms from reviewers.
A good discussion works the other way round from the introduction: it opens out from your study toward the field. It starts from what you found, sets it against what was known, offers explanations and warns of what you cannot conclude, before ending with what it implies and what remains to be done.
This guide gives you a six-step outline with the function of each paragraph, a rough length, an example and the mistakes manuscripts repeat most. For limitations, future directions and conclusions there is a separate page for each.
It has three tasks: to interpret your results, to place them in the literature and to say how far they reach. It is the section where you are allowed to opine, and that is also why a reviewer is most demanding here: every claim must have behind it a datum of yours or a reference.
In length it is usually the longest section of the paper alongside the introduction, often between two and four pages. But it is not measured in pages: it is measured by the ratio between what you say and what you can support. A long, speculative discussion is worse than a short, tight one.
Each step is usually a paragraph, although some take two. This order works for almost any empirical study.
Open by answering the question of your study, without repeating figures. “The cognitive restructuring programme reduced anxiety about oral exams relative to the waitlist, with a medium effect.” The reader needs to know up front what you found before you discuss it. If you had hypotheses, say which were supported and which not.
Set the result against what was known: does it agree with earlier studies, contradict them, qualify them? Cite those who found the same and those who did not, and if there is a difference, try to explain it (sample, measure, design). It is the step most tempting for citing only what agrees with you; a reviewer who knows the field will miss what you omit.
Here you interpret: why you think it came out this way. But offer at least one plausible alternative explanation, and say why your study favours it or cannot rule it out. A result with a single explanation sounds like a certainty the design rarely allows. And the unexpected result deserves its own paragraph: do not leave it unexplained.
What changes in theory or practice if your results are true. Be proportionate: an implication is not a firm recommendation, and it must follow from what the data showed, not from what you would like. Distinguish between what it implies for research and what it implies for practice.
Those that truly affect what you can conclude, with their effect and what you did about them. There is a full guide on how to write the limitations of a study. Place them before the end, so that the last paragraph is not a list of problems.
What it would take to go further, concretely, and a final sentence that returns the reader to the starting question. In some journals the conclusions are separate; in others this last paragraph acts as the conclusion. There is a page on future research directions and another on conclusions.
“This result agrees with earlier studies in health-science students (Author, year; Author, year), which found reductions of similar size after cognitive restructuring programmes. It differs, however, from what Author (year) found, who observed no effect in a sample of engineering students. One possible explanation is the exam format: our sample was dominated by oral tests, where social anxiety weighs more and restructuring may be more relevant. We cannot rule out that the difference is due to the measure used, since Author (year) employed a general anxiety instrument rather than one specific to exams.”
The paragraph says what agrees, what does not, offers an explanation and acknowledges another it cannot rule out. That is the combination a reviewer looks for: it compares, explains and restrains itself.
They recur so much that they are almost predictable:
Underline each claim in your discussion and note in the margin where it comes from: a datum of yours, a reference or an opinion. Those left as “opinion” need one of the other two, or a more cautious verb (“suggests”, “is consistent with”, “could”). It is what the reviewer will do, and better that you do it first.
And check the discussion against the objectives too: each objective should have its answer in step one, and each claim in the discussion should be traceable to a result. The guide on how to write research objectives has the traceability table that makes this easy.
Paste your discussion and you will see the first objection a reviewer would raise, with the exact sentence from your text and what it would take to fix it. It looks at the same things as this guide: whether the interpretation goes beyond what the data support, whether causal language appears with a design that does not allow it, whether you generalize to populations you did not sample, whether the limitations admit the real problem or only the harmless ones, and whether you discuss hypotheses that were never tested.
It returns one objection, the one that weighs most, so you know where to start. It needs at least 350 characters and what you paste is not stored. Because it only sees the discussion, if deciding something would require the results it says so separately rather than guessing.
The discussion interprets the results, compares them with the literature and covers limitations and implications. The conclusions, which are sometimes a separate section and sometimes the last paragraph of the discussion, answer the objectives briefly without adding anything new.
Between two and four pages in a typical article, although it depends on the journal. What matters is that each paragraph has a function and that nothing is claimed beyond what can be supported.
Only the minimum: a brief reference to the main finding to open, and the figures needed for a specific comparison. Repeating all the results in other words is one of the commonest mistakes.
Report it clearly and explain it. A null or contrary result is information, and a reviewer values your discussing it honestly more than your dodging it. Offer possible reasons, including that the hypothesis was wrong.
Normally in the discussion, in the second-to-last paragraph. Some journals ask for a separate section with that heading; check the author instructions.
Yes, and it is common: the introduction justifies the study and the discussion may need literature to explain the specific results. What is not advisable is for them to appear only to back what you had already decided to conclude.