Your Honours Thesis Results Section in Four Weeks: The Eight Paragraphs You Actually Need

You have the data, you have four weeks and you have a blank document. Results sections almost never sink because of a badly chosen analysis: they sink because ten days out you decide to change the model, and that one change forces you to rewrite eight paragraphs, two tables and half the discussion when there is no slack left. What follows is an operational index —which paragraphs to write, in which week, in what shape— rather than a statistics lesson. If what you need is the reference for which numbers APA 7 demands from each test, that lives in how to report results in APA 7.

The eight paragraphs, in this order

A results section is not an open-ended chapter: it is a fixed sequence of eight paragraphs and two tables. Your marker reads it in that order and looks for each item in its place, so writing it out of sequence costs you twice the time and loses you marks for navigation alone. What matters today is that five of the eight do not depend on your main analysis working: you can write them this week from the dataset you already have.

1. Participants and screening. The final N, how many cases you removed and on what criterion, decided before you looked at the outcome: “Of the 168 questionnaires returned, 18 were removed for not completing the primary scale, leaving a final sample of N = 150.” 2. Missing data. What percentage was missing, whether the pattern looked random, and what you did about it. Two sentences will do, but they have to be there: say nothing and the marker assumes you never looked.

3. Descriptives. Means, standard deviations and ranges for every study variable, pointing to Table 1. 4. Reliability. Alpha or omega for each scale in your own sample, not the value from the original validation paper; one sentence per instrument. 5. Assumptions. One paragraph, sometimes two sentences: what you checked, what failed and what you did about it. If you are not sure which ones your test requires, the list is in how to check statistical assumptions.

6 and 7. Your hypothesis tests. One paragraph per hypothesis, in the same order you stated them in the introduction. These are the only two paragraphs that change with your design, and you have a template for each below. 8. Secondary analyses and close. Anything exploratory you ran, labeled as exploratory rather than dressed up as a hypothesis, plus one sentence summarizing which hypotheses were supported. No interpretation: that is the discussion.

The four-week schedule

One rule organizes the whole calendar: write the parts that do not depend on the outcome first. And before you write a single line, freeze the dataset. Save a copy with the date in the filename, write down the exclusion criteria you applied, and commit to every number in the document coming out of that one file. The thing that most often blows up an honours timeline is not a badly chosen test, it is a re-run with a slightly different exclusion whose N no longer matches the page before.

Week 1: freeze the data and write the easy five. Dataset locked, paragraphs 1 to 5 drafted, Table 1 built. By Friday you have a page and a half of results written without having run a single hypothesis test, and that changes how the next three weeks feel. Send that page and a half to your supervisor straight away: this is the week their comments are still cheap to act on.

Weeks 2 and 3: analyses, tables, prose. In week 2 you run the main tests once, using the analysis you preregistered or agreed with your supervisor, and build the results table. In week 3 you write paragraphs 6, 7 and 8 and check the APA formatting: italics, decimals, degrees of freedom, exact p values. Do not swap the order. Writing prose while the numbers are still moving is how three versions of the same value end up in the same document.

Week 4: buffer. Submission week is not for writing, it is for what always happens: your supervisor takes four days to reply because they are marking six other theses, the table splits across two pages, the contents page renumbers itself, the repository wants a specific file format. If you reach that week still writing results, you submit something nobody has read. Plan as though the deadline were seven days before the real one, whether you are due on the 2nd or the 8th of October.

Templates for comparing groups and correlating

The five templates below all use the same worked example, so you can see how they fit together: 150 undergraduates, test anxiety as the outcome, academic self-efficacy as the predictor and procrastination as the mediator. Change the names, change the numbers, and the paragraph is written.

Independent-samples t-test. “Women (n = 82) reported higher test anxiety (M = 3.42, SD = 0.71) than men (n = 68; M = 3.11, SD = 0.74). The difference was statistically significant, t(148) = 2.61, p = .010, and small to moderate in size, d = 0.43. The mean difference was 0.31 points, 95% CI [0.08, 0.54].” Note that the confidence interval belongs to the difference, not to each mean. The full set of rules is in how to report a t-test.

One-way ANOVA. “Test anxiety differed across the three year levels, F(2, 147) = 6.84, p = .001, partial η² = .09. Tukey-corrected pairwise comparisons showed that first-year students scored higher than third-year students (difference = 0.52, p < .001, d = 0.74), with no difference between first and second year (p = .21).” The classic error here is reporting the omnibus F and never saying which groups differ: the F tells you a difference exists somewhere, not where. The long version, repeated measures included, is in how to report an ANOVA.

Correlation. “Academic self-efficacy correlated negatively with test anxiety, r(148) = -.34, p < .001, 95% CI [-.47, -.19]. Correlations among all study variables are presented in Table 2.” With more than three variables, do not narrate them one by one: they belong in a matrix, and the text highlights only the two or three that answer your hypotheses. The APA correlation table generator builds the matrix already formatted, and the canonical wording is in how to report a correlation.

Templates for predicting and explaining

Hierarchical regression. “A two-step hierarchical regression was estimated. Age and gender were entered at Step 1 as covariates and explained 5% of the variance, R² = .05, F(2, 147) = 3.87, p = .023. Academic self-efficacy was added at Step 2 and increased the explained variance, ΔR² = .08, ΔF(1, 146) = 13.43, p < .001. The final model explained 13% of the variance, R² = .13, adjusted R² = .11, F(3, 146) = 7.27, p < .001. Self-efficacy was the only significant predictor, B = -0.33, SE = 0.09, β = -.30, t(146) = -3.67, p < .001, 95% CI [-0.51, -0.15].” Without a ΔR² for each block you have run a multiple regression with decorative steps; the reasoning is in how to report a hierarchical regression.

Mediation. “A simple mediation model was estimated with 5,000 bootstrap resamples. The total effect of self-efficacy on test anxiety was c = -0.33, p < .001. The indirect effect through procrastination was significant, ab = -0.12, SE = 0.05, 95% bootstrap CI [-0.23, -0.04], and the direct effect remained, c′ = -0.21, p = .014.” The evidence for mediation is a bootstrap interval that excludes zero, not the significance of the direct effect and not the Baron and Kenny sequence, which has been superseded for years. The full walkthrough is in how to report a mediation analysis.

If you have the test statistic but no effect size —and at this stage that is the likely situation, because SPSS omits it from several menus— do not leave the slot empty and do not borrow a value from a similar paper: d can be recovered from a t and its degrees of freedom with a formula. Paste the sentence as it stands into the effect size from text calculator and it returns the value with its interval. A test without an effect size is the single most repeated correction markers and reviewers write.

The two tables, and what never gets repeated

Two tables are plenty. Table 1 carries descriptives and reliability for every variable: mean, standard deviation, range, and alpha or omega in one column. Table 2 is either the correlation matrix or the model table, depending on your main analysis. If your design is a regression, the model table replaces the matrix and the correlations move to supplementary material.

The rule for deciding what goes where is one line long: if it fits in a sentence it goes in the text; if it is more than three numbers per variable it goes in a table. And whatever appears in a table is not narrated again in the prose. Restating fifteen correlations from the matrix in sentences is the fastest way to make a marker stop reading and to inflate the section without adding anything. In the text you comment only on the numbers that carry a hypothesis.

Formatting is where marks are lost for free. M, SD, t, F, r, p, N, B, SE and d are italicized; Greek letters (beta, eta, alpha, omega) are not. The leading zero appears only on statistics that can exceed 1: M = 3.42 and d = 0.43 keep it, while p = .010, r = -.34 and R² = .13 do not. And p = .000 does not exist: that is written p < .001. To see a whole section assembled end to end with its tables, there is a complete one in the APA 7 results section with a worked example.

When the analysis is not working with two weeks left

A non-significant effect costs you neither marks nor days: a null result uses the same template and takes the same space. “No difference was found between groups, t(148) = 0.84, p = .402, d = 0.14, 95% CI [-0.19, 0.46].” The only section that changes is the discussion. What does cost you marks is dressing it up: calling p = .07 a trend toward significance, slicing the sample until a subgroup works, or presenting as a hypothesis something you thought of after seeing the data.

If the problem is technical —a model that will not converge, a broken assumption, a scale with an alpha of .58— pick plan B within 48 hours and then stop touching it. A violated normality assumption has its non-parametric alternative; an unreliable scale is reported with its real alpha and flagged as a limitation; a model that will not converge is simplified by dropping parameters. All three are defensible in an oral or in a marker's report. What is not defensible is arriving at submission week testing your fourth variant.

The last-week rule, and the only one here you will find genuinely hard to keep: after the end of week 2 the analysis does not change. Not because it could not be improved, but because every change drags eight paragraphs, two tables, the abstract and half the discussion with it, and that drag is exactly what makes people submit without rereading. A correct analysis reported cleanly always marks higher than a sophisticated one whose numbers do not line up.

Frequently asked questions

How long does it take to write a results section?

With the data collected and the analysis decided, roughly 15 to 20 working hours, which spread comfortably over three weeks alongside everything else. Writing is not the bottleneck: the decisions about exclusions, missing data and which model you estimate are. Close those decisions in week one and the text comes out in three or four sittings.

What should be in the results section of an honours thesis?

The final sample, how you handled missing data, descriptives, the reliability of your scales in your sample, assumption checks, a test of each hypothesis with its effect size, and any exploratory analyses labeled as exploratory. Not the interpretation, not the comparison with other studies, not the limitations, and not one causal sentence if your design is correlational. All of that belongs in the discussion.

How many tables should a results section have?

Two are usually enough for an honours thesis: descriptives with reliability, and either a correlation matrix or the model table. The working rule is that a number which fits in a sentence goes in the text, and more than three numbers per variable goes in a table. Anything that appears in a table is not narrated again.

Can you write a results section if nothing is significant?

Yes, and it takes exactly the same space. A null result is reported with the same structure, effect size and confidence interval included: t(148) = 0.84, p = .402, d = 0.14, 95% CI [-0.19, 0.46]. What loses marks is not the null result, it is talking about a trend toward significance or hunting for the subgroup where the effect appears.

Do you interpret results in the results section or the discussion?

In the discussion. The results section describes what came out; the discussion explains what it means. The practical boundary is simple: if a sentence contains because, suggests that or supports the idea that, it belongs in the discussion. Results only has room for the magnitude, direction and precision of the effect.

What order should you report analyses in?

In the same order you stated your hypotheses in the introduction, with the preliminaries first: participants, missing data, descriptives, reliability, assumptions. Exploratory analyses go last and are labeled as exploratory. A marker who can follow the correspondence hypothesis by hypothesis reads far faster and asks far less in the oral.

If building the schedule makes it clear that the problem is not the writing but that the analysis you need is not the one you know how to run, more weeks will not fix that. Tell me the design and the variables through statistical consulting and we can work out whether one session is enough to lock the analysis plan before it eats your calendar.