TL;DR
- A results section has two jobs — report what happened, then say what it means — and most drafts only do the first.
- Quantitative findings need actual numbers behind them: a test called "significant" with no p-value or effect size isn't a finding yet.
- Qualitative findings need to stay inside what the interviews actually confirmed, not stretch into claims nobody in the data ever made.
- The discussion section gets checked hardest, since that's where a claim can quietly contradict the very results it's supposed to explain.
Reviewers don't send a paper back because a table is formatted wrong. They send it back because a finding got reported and nobody ever said what it meant. That's the reason behind most revision requests on a results section — not bad data, just data nobody explained. Knowing how to write the results section of a research paper isn't about picking the right statistical test or the right software output. It's about finishing the sentence every finding leaves hanging: so what does this actually mean?
Are you reporting results, or explaining them?
A results section in a research paper has two jobs. Report what you found, and say what it means. Most drafts only do the first one — every table labeled correctly, every test named right — and stop there, as if the numbers will explain themselves to the reader. They won't.
Here's the test that catches it.
After every finding — a statistic, a theme, a percentage — try finishing this sentence: "this means…" or "this suggests…" If you can't finish it without just repeating the number you already gave, the finding isn't done yet.A finding without an interpretation isn't a result yet — it's just a number waiting to be explained. A p-value tells you something happened. It doesn't tell you how much it matters — that's what effect size actually measures, and it's why "significant" on its own is never the finish line.
Take something small: a survey finds 62% of participants preferred online classes over in-person ones. Reporting stops there. Interpreting adds one line — this suggests convenience now outweighs the classroom experience for most of this group — and suddenly the number connects back to whatever the research question actually asked.
The test works the same way on a theme, not just a number. Say five interview participants all mention feeling "unsupported" during a transition period. Reporting that theme by name isn't enough on its own. Interpreting suggests the support structure in place wasn't visible to the people it was meant to reach — a claim the interview data can actually carry, stated plainly instead of left as a label.
That's the test in miniature — report the number, then finish the sentence it leaves hanging. Next is where it actually breaks down. Results sections often call a test "significant" and never once give a number to back it up. Without that number, a reader has no way to check the claim for themselves.
What's missing when a finding is only "significant"?
A test runs. A verdict gets reported. The actual number never shows up. This is where a results section of research paper most often breaks down, and it hides well, because "significant" sounds like a finding even when it isn't one.
The verdict with no number behind it
One paper I reviewed ran a Chi-square test, a Mann-Whitney U, a logistic regression, and a Wilcoxon signed-rank test. Every one of them got the same treatment in the results section: "significant," "positive," "revealed a significant association." Not a single p-value. Not one chi-square statistic, odds ratio, or z-score, anywhere in the section. Four separate tests, and not one number a reader could check.
"Significant" is a verdict, not evidence. A p-value only tells you one thing — how likely your result is to be chance — which is a different question from how much it matters. Report the actual value first, every time. Then say what it means.
The number you can't trace
The paper states "one in four children was subject to the development of socio-emotional difficulties," citing Brown et al. (2012) right after it. Reading Brown et al. directly, the citation supports the general concept — early capacity-building in children under five — not the one-in-four figure itself. That specific ratio doesn't have its own source anywhere in the paper.
Elsewhere, "screens including mobile & TV activities increased during covid… in Portugal," citing Pombo et al. (2021), with no percentage or magnitude given at all. Increased by 5%? By 50%? The citation backs a direction, not a size.
The sample nobody explained
The same paper reports a quantitative sample of 225, a qualitative sample of 10 parents, and a counselling-intervention subsample of 25 — with no power analysis or sampling rationale anywhere in the methodology.
A sample size stated without justification invites the same question as an unreported p-value: why this number, and how do you know it's enough to support the claim built on top of it? None of these three are math errors. Each one is the same move: a number-shaped claim, stopped before the number does any work.
The next section runs the identical test against qualitative findings — the same discipline behind how to write a findings section of a research paper — where the missing piece isn't a number at all, but a claim the interviews never actually confirmed.
Why doesn't the interview data back up what it is claiming?
The same test applies here, just aimed at something other than a number. Knowing how to write a good results for a research paper means applying it to qualitative findings too — not just statistics. A qualitative finding fails the test not by hiding a number, but by claiming more than the interviews can actually support.
"the child… makes the parents role models by imitating…" Ten parents were interviewed, but no child was. Read as a claim, this describes something nobody actually observed — it's an inference from what a parent said, dressed up as a direct finding.
One advisor feedback post on this site calls this exact move "too descriptive": the narration is there, but what's actually being claimed as observed fact was never confirmed by anyone who could confirm it. The fix isn't to cut the claim — it's to name whose account it's actually built on. "Parents described their children as…" is honest. "The child does…" isn't, even if it's probably true.
The finding only one strand confirms
The same paper's Mann-Whitney U test found girls more affected by family crises than boys. Nothing in the ten parent interviews — no theme, no quote, no observed pattern — points at why, or even confirms it happened at all. The paper still describes its two strands as "consistent, with one enriching the other." That claim only holds if both strands actually say something.
Here, one of them says nothing on this point, and nobody checked before writing "consistent." Before calling two findings consistent, go back and confirm the second one actually exists — not just that it doesn't contradict the first.
| What the paper claimed | What the interview data actually showed |
|---|---|
| "The child… makes the parents role models by imitating…" | Ten parents were interviewed. No child was. The claim is an inference from a parent's account, not an observed fact. |
| The two strands are "consistent, with one enriching the other" | The Mann-Whitney U result found girls more affected than boys — but no theme, quote, or pattern in the ten parent interviews confirms it, or even addresses it. |
Results and discussion run into this same problem next, in a sharper form — a claim built directly on top of numbers that say the opposite of what's being claimed.
How to Write the Results Section of a Research Paper Without Contradicting Your Discussion
This section is the payoff of the whole test. Numbers and interviews are one thing to check. But the discussion is where those checks either hold up or quietly fall apart, because that's where a writer starts arguing instead of reporting — and it's easy to argue past what the data actually showed.
The misrepresented direction
A discussion states: "family involvement has a direct and positive effect on preschool children's development." Confident claim. Wrong. Go back to the same paper's own correlation table.
The only statistics that actually measure direction are Pearson's R at -.069 and Spearman's rho at -.080. Both negative. Neither significant, at p = .301 and p = .232.
The "positive" language comes from somewhere else — a Cramer's V of .297. That statistic measures how strong an association is. Not which way it points. Two different questions, answered as if they were one. Nobody caught it before the discussion went out the door.
USC's own guide to writing results sections warns against exactly this — smuggling interpretation into a section that's supposed to just report, and it works the same way in reverse: a discussion smuggling in a direction its own results never actually confirmed.
The phantom result
The same discussion cites a Mann-Whitney U result, a detailed logistic regression finding, and a Wilcoxon signed-rank result for a 25-parent counselling-intervention subsample. Three specific, citable-sounding findings. None of the three has a matching table anywhere in the results section. A reader can't check any of them — the discussion is asking to be trusted on evidence it never showed.
| What the discussion claimed | What the results actually showed |
|---|---|
| "A direct and positive effect" | Pearson's R -.069, Spearman's rho -.080 — both negative, non-significant |
| Mann-Whitney U, logistic regression, and Wilcoxon results, cited by name | No matching table for any of the three |
Your methods chapter runs into the identical mismatch — a description standing in where an argument should be. That's the full test, run three times. Numbers, interviews, and now the section that's supposed to make sense of both — which is exactly why this one gets checked last, and checked hardest.
FAQ
How to write the results section of a research paper so every number can be checked?
Read the citation directly, not just the sentence citing it. If the source only supports the general idea and not the specific figure attached to it, the number needs its own source — or it needs to go.
Is a significant p-value enough to claim a meaningful effect?
No. A significant result can still be small in practical terms. Report the effect size next to it, every time, so a reader can tell the difference.
Can I call two findings "consistent" if I never checked?
No. Confirm the second finding actually says something before writing that sentence. If it's silent on the point, it isn't confirming anything.
What's the fastest way to catch a table that contradicts itself?
Pull every variable that appears in more than one table in your results section and check the totals match before you submit. It takes about an hour.
Should interpretation go in results, or wait for discussion?
The one-line "this means" sentence goes in results, right after the finding. The longer argument — what it implies, how it fits other work — belongs in discussion.
Final thoughts
Every example in this post traces back to the same test. This means. Finish that sentence after a finding, or it isn't done yet. A results section of a research paper isn't hard because the statistics are hard. It's hard because reporting feels like enough, and it never quite is. The chapter that gets sent back and the one that doesn't usually contain the same numbers. The difference is one sentence, run consistently, after every single one of them. Run the test on your own draft before you submit it. Not just once. After every finding, every time.