How to Write a Discussion Section (Without Over-Claiming)

Everything up to this point in your paper has been reporting. The Discussion is where the article finally speaks — where your findings meet the literature and the paper takes a stand. It's also where a lot of good research quietly falls apart: careful methods, clean results, and then a Discussion that over-reaches, wanders off into new literature, or simply argues before it has told the reader what happened.

I'm a clinical psychologist and statistician. I finished my PhD on schedule, I've published around twenty papers, and I'd describe myself as a fairly average researcher. That last part matters, because the method I'm about to give you isn't built on talent. It's built on doing two steps in a fixed order and refusing to do them out of order.

Here's the order, and it doesn't move:

1. Summarise the findings in plain language, before interpreting anything.
2. Match every discussion point to a claim your introduction already made.

That's the whole Discussion. Two moves. Once they're done, the section is essentially built — and the limitations, conclusion, and final polish that follow (I cover those in the companion post) become the easy part. Let's walk through both.

Step 1: Summarise before you argue

The first paragraph of the Discussion is not the place to argue. It's the place to orient the reader.

Before you interpret anything, compare it to the literature, or acknowledge what your study can't tell you, the reader needs one thing: a clear, plain-language statement of what you found. Not the statistics — those are already in the Results. A readable summary of what happened, in terms a non-specialist could follow, that gives the reader a map before you start discussing what it all means.

Here's what that looks like in practice:

This study evaluated the effectiveness of an enhanced group CBT treatment for adolescent anxiety disorders, including intensive therapist/family/peer-assisted exposure therapy with family member and school personnel involvement. At post-treatment and at the 12-month follow-up, 41.6% and 85.9%, respectively, of those who completed treatment were free of all anxiety diagnoses. This substantial increase in effectiveness from post-treatment to follow-up was not due to receiving additional therapy — only 12.5% of those who achieved remission at follow-up received any additional treatment. Benchmarking against a meta-analysis indicated equivalence on symptom measures but inferiority on remission at post-treatment; however, at 12-month follow-up, there was a 99.99% probability that the treatment was superior to the benchmark on remission. Attrition was lower than the benchmark, and adolescents indicated they would recommend the treatment to a peer. Overall, the results indicate that the treatment was effective and acceptable for adolescents with a range of anxiety disorders.

(The numbers above are from an illustrative worked example, not a live result — use them as a shape to copy, not a finding to cite.)

Notice what this paragraph does not do. It doesn't interpret. It doesn't say why remission increased from post-treatment to follow-up. It doesn't argue that the results support a particular mechanism. It doesn't hedge, apologise, or compare to the literature. It states what happened — in human terms, with the most important numbers, in the order they matter.

The most common mistake is skipping this step and jumping straight into argument. "Although results found X, this may reflect…" is an interpretation sentence. It belongs in the next step, not the first paragraph of the Discussion. When a Discussion opens with interpretation, you're asking the reader to evaluate a claim about findings they haven't been reminded of yet. Your Results section may be ten pages back. Readers — especially reviewers — need the map before the argument.

The second mistake is treating this summary as a copy of the abstract or a repeat of the Results. It's neither. The abstract is written for people who won't read the paper. The Results section is written with statistics and technical precision. This summary is written for someone who has just finished reading your paper and needs one paragraph to consolidate it before the interpretation begins.

I started my very first Discussion the wrong way — launching straight into "although the results indicate X, it is important to consider…" before I'd told the reader what the results even were. My supervisor caught it immediately: ground the reader first, then discuss. I've written it that way ever since.

Summarise before you argue. One paragraph, plain language, no interpretation. Then proceed.

📄 Free: Not sure how you'll get from a pile of notes to a first draft in the first place? Grab From Notes to First Draft — a short guide to the exact sequence I use to turn scattered reading and results into a paper skeleton you can write from. Download it here.

Step 2: Interpret — one finding, one theory, one paragraph at a time

Here's the good news: the Discussion writes itself — if you do the matching exercise first.

Before you write a single interpretive sentence, make a simple table. On one side, list every claim you made in the introduction: the background facts, the gap, the hypothesis. On the other side, list every finding from your results. Then draw the lines. Each matched pair — one background claim, one result — becomes one Discussion paragraph. When the matching is done, the structure is done.

Here's what one matched paragraph looks like:

In line with expectations and previous research, the current sample had a higher average age and higher rates of Social Anxiety Disorder as primary diagnosis than the benchmark. Thus, it is not unexpected that treatment did not outperform the benchmark at post-treatment, given that higher age and SAD are associated with poorer outcomes. However, outcomes on remission were better than those expected from studies that only included adolescents, and results were comparable to effectiveness studies targeting a younger population with SAD. Despite this comparability, age and diagnostic composition may explain why the treatment did not show an enhanced effect relative to the benchmark at post-treatment.

Follow the structure:

• The topic sentence connects the finding to something the reader already knows from your introduction — a background fact, an established association, a prior expectation.
• The middle sentences use prior literature to explain or contextualise the finding.
• The closing sentence draws the interpretive point: what does this mean, given everything above?

Two failure modes are common.

Over-claiming — using the results to argue for something the data can't support. A finding that is "consistent with" a theory is not evidence the theory is correct. A non-inferior result is not a superior one. Reviewers spot over-claiming instantly, and it damages credibility across the whole paper. Once a reviewer catches you stretching one claim, they start re-reading every other claim looking for the same move.

Under-claiming — writing "consistent with prior research" and stopping there. That adds nothing to the field. If your finding merely replicates what's known, you have to explain why the replication is valuable: different population, different setting, different measure. If you genuinely can't explain what the finding adds, that's a signal to go back to your introduction and reconsider the gap.

The single most reliable warning sign: a study you want to cite in the Discussion that was not in the introduction. If a new reference appears here, your introduction needs it too — and if the introduction has no room for it, the finding may not belong in the Discussion at all. The Discussion is a closed loop. It interprets findings using the framework the introduction built. New literature opens a new loop, which means rewriting the introduction — so catch it now.

This claims-and-findings table enforces the discipline. Every discussion paragraph traces back to something in the introduction. Every introduction claim has a corresponding finding — or an acknowledged absence. The loop is closed.

List the claims. List the findings. Match them. Then write.

The Discussion is done. Now finish the paper.

You've done the two moves that make a Discussion work: you summarised in plain language before you argued, and you interpreted one matched finding at a time inside a closed loop. That's the hard intellectual work of the section, and it's behind you.

What's left is the part that decides whether the paper survives review and gets found: the limitations that disarm your reviewers, a conclusion that takes a stand instead of sagging into hedges, and the keywords, title, and abstract you should be writing last. I walk through all of that — with the same worked example — in the companion post.

None of this depended on inspiration. It's a fixed order of small, checkable steps — which is exactly why your next paper will be faster than this one, and the one after that faster still.

One question before you go: when you sit down to write a Discussion, do you argue first or summarise first? Be honest — I did it backwards for my whole first paper. Hit reply and tell me which camp you're in. I read every reply.

📄 Free download: From Notes to First Draft — the short sequence I use to turn scattered reading and results into a paper skeleton you can actually write from. Start here, and the Discussion has something to close the loop on.