Reporting a result is saying what the data showed. Interpreting it is explaining what it means. Mixing the two in the same paragraph is the most common structural error in a results chapter, and examiners are trained to catch it.
Sign in to mark this article as read and track your progress.
The results section and the discussion section have distinct jobs. In many student reports they blur into each other, and the blur costs marks.
Results report what the data showed: the value, the test statistic, whether the result was statistically significant. No interpretation, no comparison to prior work, no explanation of why the result came out as it did.
Discussion explains what the results mean: why they came out the way they did, what they have in common with or contradict in prior work, what alternative explanations exist, and what the result cannot claim.
The test is simple: every sentence in a results chapter should be answerable from the data alone. "The new rule set had a mean detection time of 3.4 seconds (SD = 0.6), compared with 4.9 seconds (SD = 0.8) for the baseline, a difference that was statistically significant (paired t-test: t(24) = 6.1, p < 0.001, d = 1.2)." That sentence is a result. "The improved detection time is consistent with prior work suggesting that rule-set specificity reduces alert queue depth, which in turn reduces the time an analyst spends triaging before acting on a genuine alert." That sentence is discussion.
A table is for presenting multiple precise values so the reader can compare them. A figure is for showing a pattern or distribution that would take many sentences to describe in prose.
A table of mean and standard deviation for three classifiers across five datasets is a table: the reader needs the precise numbers to judge significance. A plot of detection time over 30 days showing a steady improvement that accelerates after a software update is a figure: the trend is the point, not the individual values.
Every table and figure must be self-contained: a caption that describes what it shows, column headers or axis labels that give units, and a reference in the body text before it appears. A reader who looks only at the figure — not the surrounding text — should understand what it is displaying.
A well-structured discussion paragraph has three moves:
"The personalised aggregation strategy achieved 4.7% higher F1 than the FedAvg baseline on non-IID data (Table 3). This is consistent with Li et al.'s (2020) finding that client-drift in non-IID settings systematically degrades FedAvg, and with Karimireddy et al.'s (2020) theoretical bound on FedAvg's convergence rate. One alternative explanation is that the performance gain reflects the specific heterogeneity structure in this study's ten-client simulation rather than a general advantage of personalisation; a study using a broader range of heterogeneity parameters would test this."
Overclaiming is reporting a finding in language stronger than the evidence supports. A study of fifty ransomware samples from a single family cannot support "ransomware universally..." A pilot study with thirty participants cannot support "employees consistently..." The fix is hedging that accurately reflects scope: "in this sample," "under the experimental conditions described," "for the malware family examined." Hedging is not weakness; it is accuracy.
A results chapter contains this sentence: "The detection rate improved to 96%, which is better than prior systems and shows that our approach is more effective because it uses more features." Identify every problem with that sentence and rewrite it as a results statement followed by a separate discussion sentence.
