Your Assignment 2 analysis companion. Paste two paired conditions and it describes the data, runs one appropriate test, shows the working, and models the write-up, one step at a time.
Name the comparison before you touch the numbers.
The independent variable is what you deliberately changed (say noise: quiet vs noisy); the dependent variable is what you measured (say problems answered correctly). The other decisive question is whether the same people appear in both conditions, which is a paired, within-subjects design, or whether two different groups are compared. That single choice fixes which test is correct.
Before you paste anything, write down what you are comparing. This decides which test is correct, so it comes before the numbers.
Paired: every row is one person, and the two columns are the same people under two conditions, so you analyse each person's A − B difference. Independent: the two groups are different people, so there are no per-row differences to take and you compare the two group means directly. Counterbalancing condition order (some people quiet-first, some noisy-first) keeps a paired design valid; it does not make two separate groups paired.
One row per participant, one column per condition.
Paired data is tidy when each row is a single person measured in both conditions, so their two scores sit side by side and each person's A − B difference can be taken. Keep the columns in a consistent order and separate the two values with a space, tab, or comma; a text header row is detected and ignored.
Two columns, one row per participant (paired / within-subjects).
Describe the data before you test it.
The mean is the typical score in a condition; the standard deviation is how much scores scatter around that mean. Report both. A gap between two means means little until you know whether scores already vary by one point or by ten. Look hard at the figure before running any test.
Paste your two conditions on the Data step and choose Analyse. The descriptive statistics and figure appear here.
The mean is the typical score: add the values and divide by n. The standard deviation is how far scores sit from that mean on average: SD = sqrt( sum of (value − mean)² / (n − 1) ). The n − 1 (not n) is the sample correction. Report the mean and the SD for each condition, never the mean alone: a 3-point gap means little until you know whether scores already vary by 1 point or by 10.
One defensible test on one clear comparison.
A paired t-test asks whether the average difference is larger than the scatter in those differences would lead you to expect by chance; the Wilcoxon signed-rank is its rank-based counterpart for ordinal or skewed data such as Likert ratings. A p-value only says how surprising the gap would be if the conditions truly had no effect. It does not say how big the effect is. Cohen's dz reports that size, so always give both.
A paired-samples t-test is the default for paired numeric data. If your data are ordinal or clearly skewed (e.g. Likert comfort ratings), use the Wilcoxon signed-rankUse the Wilcoxon result when the outcome is ordinal (like a 1–5 Likert rating) or clearly skewed, where the t-test's assumptions do not hold. It compares the ranks of the differences rather than their sizes. result instead.
Paste your two conditions on the Data step and choose Analyse. The test statistics and working appear here.
Any of these gives the same paired result, so use whichever you know:
Excel: =T.TEST(A_range, B_range, 2, 1) returns the two-tailed paired p-value (the 2 = two tails, the 1 = paired). The Data Analysis add-in's "t-Test: Paired Two Sample for Means" gives the full table.
R: t.test(A, B, paired = TRUE) · ordinal/skewed: wilcox.test(A, B, paired = TRUE)
Python: scipy.stats.ttest_rel(A, B) · ordinal/skewed: scipy.stats.wilcoxon(A, B)
AI-assisted: you can ask an AI tool to run it, but then check the output yourself: is it a paired test, are the columns in the right order, is the p-value in a plausible range? Under the unit's Lane 2 AI stance any tool can run the test, but you own the result, not the tool.
Turn the numbers into paper-ready prose.
IMRaD Results state what you found in a sentence or two, giving the direction of the difference, its size, and the test statistic, then point to your figure. Methods describe your protocol so another researcher could reproduce it: prose that names the conditions, sample, instruments and the test, rather than a bullet list of ingredients.
A model Results sentence built from your numbers. Edit it into your own words; do not paste verbatim.
Run the analysis on the Data step to generate a model Results sentence from your numbers.
The tool cannot write your Methods, because they describe your experiment. Here is the shape a High Distinction Methods paragraph takes, on the worked-example scenario. Notice it lets another researcher reproduce the study.
The same facts as an AI-style bullet list are not what to submit:
Both hold the same facts. The bullets list ingredients; the prose lets someone reproduce your study, which is the whole point of a Methods section. It states the order was counterbalanced, defines the score, justifies why a paired test (each participant did both conditions), and names the tool. A2's Methods criterion rewards the prose version.
Check it holds together before you submit.
A strong submission names its design and reports means with their spread. It shows one well-labelled figure, runs one test matched to the design, and gives the effect size beside the p-value. Methods are written as reproducible prose. Acknowledge and validate any AI you used along the way.
Tick each item as you confirm it. Nothing is saved; the list resets on reload.