Two columns of scores sit beside each other. One learner sees two groups; another sees the same people measured twice. Both reach for a statistical test, and only one has read the design correctly. The difference lies in what the rows represent, not in how the table happens to look.
For AQA A-Level Psychology inferential-test questions, identify the investigation's aim, the relationship between observations and the level of measurement before choosing a test. Repeated measures and deliberately matched pairs create related observations; two separate groups do not become paired because their scores appear on the same row or their sample sizes are equal.
This guide uses original fictional scenarios to practise that decision. It concerns AQA 7182, with the updated specification first assessed at A-level in summer 2027, checked on 24 October 2026. The aim follows the Education parent guide on learning that transfers into life: explain the design in a new scenario before relying on a familiar test name.
Find your next learning step
Recognise repeated measures, matched pairs and independent groups before choosing an inferential test, with original AQA Psychology practice cases and clear limits.
ROUTE 1 · CHAPTERS 1–3
Read the design
Identify what one score and one row represent.
ROUTE 2 · CHAPTERS 4–6
Work through comparisons
Separate paired observations, independent groups and correlation.
ROUTE 3 · CHAPTERS 7–9
Repair the reasoning
Preserve identities and justify the test from the scenario.
ROUTE 4 · CHAPTERS 10–12
Transfer and practise
Change the design and adapt the learner support.
ROUTE 5 · CHAPTERS 13–15
Check and continue
Try independent cases, FAQs and current primary sources.
Full chapter index · Independent learning check · Education parent guide
Full chapter index
Read the design · Chapters 1–3
Work through comparisons · Chapters 4–6
Repair the reasoning · Chapters 7–9
Transfer and practise · Chapters 10–12
Check and continue · Chapters 13–15
CHAPTER 1 OF 15 · Read the design
1. Ask what a single recorded value represents
Before comparing columns, explain one value. Is it one person's task score, a count of people in a category, a rank, or a rating? What condition or variable does it refer to? A numerical display can conceal several different kinds of information.
Suppose a fictional study gives each participant a score after using one of two instruction formats. “Score” alone does not tell us whether the two columns contain different participants or repeated observations from the same participants. We need the method, participant identifiers or an explicit design description.
Treat the table heading as evidence, but read it with the scenario. A heading such as “Condition A” identifies a condition, not a participant group. The same people could complete both conditions. Conversely, a before-and-after-looking layout could compare different samples measured at different times.
A useful first annotation is, “One value is __ for __ under __.” A second is, “The corresponding value in the other column comes from __.” If the second blank cannot be filled from the question, do not quietly invent a pairing.
This is not an extra calculation. It is the interpretation that tells you which calculation could make sense. A test applied to the wrong data structure can produce a precise number that answers an inappropriate question.
AQA's current Psychology research-methods content includes experimental design and level of measurement among the factors affecting test choice. Start with those factors in the actual scenario. Memorising a decision grid becomes more useful when its categories refer to something you can identify and justify.
CHAPTER 2 OF 15 · Read the design
2. Distinguish the research aim from the table's shape
Two columns do not automatically mean a test of difference. They might record the same variable under two conditions, or two different variables measured for each person. The investigation's aim determines which relationship is being examined.
Imagine a fictional task comparing performance after a diagram instruction and after a written instruction. If the aim concerns whether performance differs between those conditions, the question is about a difference. Now imagine recording each person's reading-time score and task-performance score to investigate whether they vary together. That is a relationship question.
Both datasets could place one participant on each row and two numbers beside their identifier. Their visual similarity does not make the research aims identical. A learner who chooses a test from the number of columns has skipped the substantive question.
Write the aim as a sentence before naming a test: “The researcher wants to compare __ with __” or “The researcher wants to examine the relationship between __ and __.” Use the actual variables. If the sentence becomes unclear, return to the scenario's aim and hypothesis.
The distinction also prevents a common misuse of the word “related.” In test-choice teaching, a related design refers to linked observations such as repeated measures or matched pairs. It does not merely mean the topic is interestingly connected to another topic, or that a researcher is investigating a correlation.
Do not assume that a difference automatically establishes causation. The aim and design help identify a suitable comparison; they do not remove confounding variables or supply experimental control that the scenario lacks. Keep test selection, statistical inference and causal interpretation as connected but separate decisions.
CHAPTER 3 OF 15 · Read the design
3. Recognise repeated measures through participant identity
In a repeated-measures design, the same participants provide observations in the conditions being compared. The relationship is built into who produced the data. The second score is linked to that person's first score, rather than to whichever number happens to appear beside it after sorting.
For a fictional classroom example, the same learners complete a short visual-search task with two display layouts. Each learner supplies a score under Layout A and a score under Layout B. The question concerns a difference between layouts. That gives a related comparison because each person's two observations belong together.
Explain the design without relying on its label: “Each learner completed both layouts, so their Layout A score corresponds to their own Layout B score.” This sentence demonstrates the relationship more clearly than “it is repeated measures because it repeats.”
The design may also create order concerns. If everyone completes A first, an observed difference could reflect practice or fatigue as well as layout. Recognising pairing does not solve that concern. Counterbalancing, if described, addresses an order issue while preserving the paired relationship.
A useful practice variation changes only the participants. Keep the same task and layouts, but let one group complete A and a separate group complete B. The variables and numbers might remain similar, yet the design is now independent groups. The test-choice reasoning must change accordingly.
Students often find this reassuring. They do not need to guess a test from a topic such as memory or perception. They need to read a small number of consequential facts accurately. Participant identity is one of those facts, and it remains relevant even when the subject matter or wording changes.
CHAPTER 4 OF 15 · Work through comparisons
4. Follow four invented paired records
This miniature table illustrates structure only. It is not a real study, a representative sample or sufficient evidence of an effect. Assume the same fictional participants completed both display layouts, and that higher scores indicate more correct responses.
| Participant | Layout A score | Layout B score | B minus A |
|---|---|---|---|
| P1 | 12 | 14 | 2 |
| P2 | 15 | 13 | -2 |
| P3 | 8 | 11 | 3 |
| P4 | 20 | 18 | -2 |
P1's comparison is 14 with 12, because both values belong to P1. P2's comparison is 13 with 15. Those links remain meaningful even though the scores differ in direction. Pairing is not a claim that everybody improved or that the two scores are similar.
The difference column preserves the chosen subtraction order. Positive values mean a higher B score; negative values mean a lower B score. Reversing the subtraction would reverse every sign without changing who is paired with whom. State the convention so the interpretation remains clear.
The table also shows why counting all displayed values as independent participants is wrong. Four people supplied eight scores. The number of observations, the number of participants and the number of pairs describe different quantities. A learner should be able to name each one without using them interchangeably.
No inferential result has been calculated here. We have not established that the outcome should be treated as interval data, checked a sampling model or specified a test decision. The table's job is to make the within-person link visible before those later decisions.
For practice, cover the difference column and reconstruct it. Then explain why subtracting P3's B score from P1's A score would answer no intended within-person question. The arithmetic could be performed, but the relationship would have been invented.
CHAPTER 5 OF 15 · Work through comparisons
5. Understand why separate sorting destroys the comparison
Suppose someone sorts the A scores from smallest to largest, then independently sorts the B scores from smallest to largest and places the lists side by side. The columns still contain exactly the same values. Their totals and separate averages have not changed. Yet the original within-person pairings are gone.
In the fictional table, P1's A score of 12 belongs with P1's B score of 14. In independently sorted columns, 12 could sit beside 13. That position does not turn the two values into measurements from the same person.
This matters because a paired analysis uses information about the differences within the genuine pairs. Reassigning partners changes those differences and can change the variability of the differences. Preserving separate column summaries is therefore not enough to preserve the intended analysis.
The repair is to sort complete records together, keeping the participant identifier and both scores on the same row. If the identifiers have been lost, the original correspondence may need to be recovered from the source records. Do not reconstruct it by choosing pairs that make a preferred result look stronger.
A student can test their understanding by asking what would happen if the second column were randomly rearranged. For an independent-groups comparison, arbitrary row alignment was not part of the design. For a repeated-measures comparison, the rearrangement breaks meaningful links.
This example connects statistics to ordinary data handling. A spreadsheet operation can preserve every number while damaging the information those numbers represent. In a written examination, the same mistake appears as a test chosen from neat columns rather than the method. Keep the identity link visible before worrying about how tidy the display looks.
CHAPTER 6 OF 15 · Work through comparisons
6. Distinguish matched pairs from similar-looking groups
A matched-pairs design links different participants according to relevant characteristics before comparing their outcomes. The members of a pair are not the same person, but the design specifies which participant corresponds to which other participant. That is a different route to related observations.
Imagine a fictional comparison of two instruction formats. Participants are deliberately paired using an appropriate pre-task measure described in the scenario. One member of each pair uses Format A and the other uses Format B. The analysis needs to respect the intended pairings.
Now change the description. Two existing classes use the formats, and their average ages happen to be similar. That fact alone does not establish individual matched pairs. Similar group summaries are not the same as a documented participant-to-participant correspondence.
Ask, “Could I identify the partner of this particular participant from the design?” If the answer is yes, explain how the matching was established. If the answer is only that both groups contain twelve people, the pairing has not been justified.
Matching also does not make two people identical. They can differ on characteristics that were not matched. The design may address some sources of variation while leaving others. Do not turn a recognised matched-pairs structure into a claim of perfect control.
Consider a fictional pair labelled M1, consisting of one participant under each format. Their link comes from the matching process recorded in the method. If the higher-scoring person is moved into another pair after outcomes are seen, the original design is no longer being represented. Keep the matching record stable even when the resulting difference is inconvenient for the prediction. This is a data-interpretation principle, not an instruction to conduct a study.
A useful error-repair exercise gives two almost identical scenarios: one with explicit pairing and one with only equal sample sizes. Ask for the design classification and the exact phrase that supports it. This keeps the learner from treating “same number” as a shortcut for “related.”
The key habit is to preserve the relationship supplied by the method. Never create matches after seeing the outcome merely because a paired test seems convenient.
CHAPTER 7 OF 15 · Repair the reasoning
7. Recognise independent groups without adding a relationship
In an independent-groups design for a two-condition comparison, different participants provide the observations for each condition, without the repeated or matched relationship used in the earlier examples. Two columns can still be displayed side by side for convenience.
Suppose one fictional group sees a diagram instruction and another group sees a written instruction. No person completes both conditions and no matching is described. The fact that the participants attend the same institution does not, by itself, create paired scores.
For example, a table might place the first ten volunteers for one condition beside the first ten volunteers for another. The row positions only organise the display. A learner can demonstrate the distinction by rearranging one list on paper: no intended participant-to-participant link is lost because none was established. That differs from the repeated-measures table, where rearrangement would disconnect a person from their own second score.
The same applies to matching labels in a table. “Participant 1” in Group A and “Participant 1” in Group B may simply mean the first entry in each list. Unless the method establishes a relationship, those numbering positions are not evidence of a matched pair.
A concise explanation is, “Different participants completed the two conditions, and the scenario does not describe matching between them.” That justifies an unrelated comparison more directly than “the numbers are different,” since repeated-measures scores can differ too.
Be careful with the word “independent” beyond the school classification. Real statistical independence can be affected by additional structure, such as participants working in shared groups. This guide practises the two-condition designs specified in examination scenarios; it is not a universal analysis plan for every clustered or complex dataset.
If the question explicitly gives a more complex structure, use that information rather than forcing it into a simplified decision grid. The purpose of the grid is to organise relevant facts, not to make inconvenient facts disappear.
For learning, change the order of presentation. Read a method paragraph before seeing the table, classify the design, and predict how participant identifiers should be recorded. This tests whether the learner understands the design rather than recognising one familiar page layout.
CHAPTER 8 OF 15 · Repair the reasoning
8. Add measurement level before naming the candidate test
Once the aim and design are clear, inspect the level of measurement and any assumptions supplied in the question. Numerical labels do not automatically make data interval-level. A ranking or ordered rating can use numbers without establishing equal distances between successive values.
In an AQA-style choice exercise, a difference comparison with related observations and data treated as ordinal points towards Wilcoxon. A difference comparison with unrelated observations and ordinal data points towards Mann-Whitney. The participant relationship changes which of these candidates fits the scenario.
AQA’s published answers and commentaries on an older assessment illustrate the repeated-measures, ordinal-data and difference rationale. That resource supports the underlying distinction; use the updated specification for the actual 2027 course.
For interval-level data with the relevant parametric assumptions satisfied, the related or unrelated t-test may be appropriate respectively. The words “with assumptions satisfied” matter. A repeated-measures design alone does not justify a related t-test for every possible outcome measure.
For a paired t-test, the distribution relevant to the usual normality assumption is the distribution of within-pair differences. It is not enough to declare that a column contains numbers. The NIST explanation of paired observations shows how the analysis is built from those differences, while JMP's paired t-test guidance states the assumptions more explicitly.
For the fuller statistical machinery, the existing specialist guide to nonparametric statistics and rank-based tests explains signs, ranks and the assumptions those procedures retain. Keep that technical question with its specialist owner; the task here is recognising the examination scenario’s design before selecting a candidate.
This page is deliberately focused on recognising the relationship between observations. It is not a complete reference for every test in the specification. Nominal data, sign-test decisions, associations and correlations require their own appropriate reasoning. Do not replace that reasoning with the four candidate names used in these examples.
A further boundary matters when moving beyond a short examination scenario. A non-parametric test still has conditions; it is not a device that makes any data suitable for inference. The observations, measurement and proposed comparison must support the chosen procedure. If the question gives the necessary conditions, use them. If a real dataset leaves them uncertain, recognise the missing information rather than selecting a test solely because it appears more forgiving.
A good short justification joins the three facts: the aim concerns a difference, the observations are related or unrelated for a stated reason, and the measurement level and assumptions fit the proposed test. The name comes at the end of the reasoning, even if you write it first in your answer.
CHAPTER 9 OF 15 · Repair the reasoning
9. Avoid converting every paired dataset into a correlation
A learner notices that each person has two values and says, “The scores are paired, so use a correlation.” This confuses data structure with research aim. Paired observations can support a comparison of conditions; two variables measured for each person can support a relationship question. The intended question determines the analysis.
Use a fictional example with two instruction formats. The same participants complete both formats, and the researcher wants to know whether their performance differs across formats. The within-person link supports a related test of difference. The aim is not simply whether high scorers under A tend also to score highly under B.
Those are genuinely different questions. It is possible for people to retain a similar ordering across conditions while scores shift. A high relationship between two measurements would not, by itself, establish that the measurements have the same average level.
For a concrete illustration, imagine one fictional condition gives every participant exactly five more points than the other. Their ordering is preserved, but the two condition scores are not equal. This arithmetic illustration shows why a relationship and a difference should not be used as synonyms.
Now change the aim to whether time spent on a task is associated with performance on it. The variables have changed, and so has the question. Appropriate correlation reasoning may now be relevant, subject to the actual measurement and assumptions.
Practise stating the null question in ordinary language before choosing a test. Are you investigating a difference between conditions or an association between variables? This pause is especially useful when a table looks familiar. A familiar shape can tempt the learner to answer a previous question rather than the one currently being asked.
CHAPTER 10 OF 15 · Transfer and practise
10. Write a justification that belongs to the scenario
A weak justification reads, “Wilcoxon is used because the data are related.” That may identify one relevant fact, but it leaves the aim and measurement level unstated. A stronger answer links the candidate test to the supplied investigation.
For an original fictional scenario, suppose the same people rate two display designs, the outcome is explicitly treated as ordinal and the researcher asks whether ratings differ. A useful justification identifies a difference comparison, the same participants in both conditions, and ordinal data. The answer should use the scenario's actual language rather than copy an abstract phrase without connection.
Do not invent an assumption to make a preferred test fit. If the question says the data are ordinal, do not call them interval merely because their labels are numerical. If no matching is described, do not declare the groups matched because they have similar average scores.
The justification can be concise. More words do not necessarily add evidence. A sentence about the topic's importance, for example, does not help distinguish Wilcoxon from Mann-Whitney. Select the facts that control the choice.
A useful revision method circles the test name and underlines each independent reason. Then draw a line from each reason to the phrase in the question that supports it. If a reason has no source in the scenario or established test conditions, inspect it before retaining it.
Our existing guide to Cambridge AS Psychology and the boundary between results and generalisations addresses a different decision after reading a study. It is a useful sibling because choosing an appropriate test does not, on its own, establish that a result applies to every population or situation.
A different limitation appears in the AP Biology guide to interpreting controls before a missing response. Recognising a paired design cannot repair a control that did not demonstrate its expected behaviour. Choosing a test and establishing what a measurement can reveal are distinct checks on an evidence-based explanation.
CHAPTER 11 OF 15 · Transfer and practise
11. Diagnose the student's first wrong branch
A wrong test name can arise from several different mistakes. The student may have misread the aim, lost the participant relationship, misunderstood the measurement level or memorised the wrong test association. Giving another mnemonic without finding the first wrong branch may leave the underlying difficulty unchanged.
Ask the learner to explain the scenario without naming any test. What is being compared? Who supplied the scores? What kind of data are they? This reveals whether the difficulty occurs before the test name becomes relevant.
If the learner says two separate groups completed both conditions, return to participant identity. Draw one symbol for each fictional person and connect it to the condition or conditions they completed. If the design is understood but the test name is confused, a small retrieval exercise may be enough.
If the learner treats all numbers as interval data, contrast a position in a ranking with a measured duration. Explain what the numbers mean before discussing test choice. The repair should concern the measurement claim, not the appearance of digits on the page.
If the learner correctly identifies repeated measures but automatically chooses correlation, use the same dataset with two different aims. Ask which question each analysis would address. This targets the precise confusion between pairing and association.
Keep feedback specific: “You found the ordinal outcome, but you treated equal group sizes as matching.” That sentence recognises the part that worked and identifies the part to repair. It is more useful than a general judgement that the learner is weak at statistics.
The Sengkang guide on observed, inferred and unknown learning evidence supports this wider teaching habit. A wrong final name is an observation; the reason for it still needs diagnosis before a suitable repair can be chosen.
CHAPTER 12 OF 15 · Transfer and practise
12. Practise changes that force the reasoning to move
Start with a scenario in which the same fictional participants complete two conditions and the scores are treated as ordinal. Ask for the aim, design, data level and candidate test. Then change only the participant arrangement to two separate groups.
The learner should preserve the aim and measurement reasoning while changing the relationship between observations and the corresponding candidate. If every part of the explanation changes, check whether they have understood which fact was altered.
Next keep the repeated-measures arrangement and change the outcome to interval-level data with the relevant parametric assumptions explicitly satisfied. Now the candidate can change while the participant relationship stays the same. This contrast prevents “repeated measures equals Wilcoxon” becoming a new oversimplification.
For another variation, keep the table but change the research aim from a difference between conditions to a relationship between two variables. The learner should recognise that the original condition-comparison question has been replaced, rather than simply attaching another test name to the same story.
A teacher can give students mixed scenarios without labelling the design. Ask them to underline the phrase that determines each branch. A parent or study partner can ask, “Which single change would make your answer different?” without needing to calculate the test themselves.
Use fictional written scenarios for this practice. There is no need to recruit participants or collect psychological data to learn how to read a design. Any real school research follows the teacher's ethical and practical arrangements.
Keep the first independent response before discussing it. Later, use a new topic and different table organisation. Success on the changed presentation is stronger evidence of understanding than a perfect answer to the same scenario after its design has been explained several times.
First scenario: the same fictional participants complete a task with two instruction formats. The researcher asks whether scores differ. The question states that the scores are treated as ordinal. Identify the relationship between observations and a suitable candidate from the tests discussed here.
A sound answer identifies a related difference comparison and Wilcoxon, justified by the same participants supplying both sets of ordinal scores. It does not claim significance, since no test result or decision information has been supplied.
Second scenario: two separate groups complete those formats, with no matching described. The outcome remains ordinal and the aim still concerns a difference. The candidate becomes Mann-Whitney. Equal group sizes, if added, would not change that reasoning by themselves.
Third scenario: different participants are deliberately matched into pairs using a stated relevant pre-task characteristic, and one person from each pair completes each condition. The outcome is explicitly interval-level and the question says the relevant parametric assumptions are met. A related t-test is a suitable candidate for the difference comparison under those stipulated conditions.
Now remove the phrase about assumptions from the third scenario. The paired relationship remains clear, but the full justification for the parametric candidate needs more information. A learner should distinguish an identified design from an analysis decision that depends on additional conditions.
Finally inspect this sentence: “There are twenty scores, so there are twenty independent participants.” If ten people each supplied two scores, the sentence misstates the structure. Rewrite it with the correct numbers of participants, scores and pairs. This simple check often exposes the same misunderstanding that would otherwise reappear during test selection.
CHAPTER 14 OF 15 · Check and continue
14. Resolve common questions about paired observations
Are equal sample sizes evidence of matching?
No. Matching requires a meaningful correspondence established by the design. Two unrelated groups can contain the same number of participants, and a side-by-side display does not create a relationship between their rows.
Does “before and after” always mean repeated measures?
Check who was measured. The same individuals measured at both times create paired observations. Different samples measured before and after an event do not become the same people because the column headings suggest a timeline.
Does a related design always require Wilcoxon?
No. The aim, data level and relevant test assumptions also matter. Wilcoxon and the related t-test are different candidates under different conditions. This article focuses on recognising pairing, not replacing every other test-choice criterion.
Can I sort a paired table?
Keep each complete record together so the two scores remain attached to their participant or matched-pair identifier. Sorting columns separately can destroy the relationship even when every original number remains present.
Does an appropriate test prove the cause of a difference?
No. Test choice does not supply randomisation, control of alternative explanations or a representative sample. Explain what the actual design and result support, and keep causal or population claims within their evidence.
Is this a complete research-analysis manual?
It is an educational guide to a narrow examination skill. Complex datasets can involve more conditions, missing observations, clustering or other requirements outside these examples. Use the actual course instructions and appropriate statistical expertise for a real investigation.
CHAPTER 15 OF 15 · Check and continue
15. Continue with current sources and a precise next step
AQA's 7182 overview identifies the updated course for first A-level examinations in summer 2027. The current specification PDF linked by AQA sets out the research-methods content. The online research-methods section confirms the relevance of design and measurement to inferential-test selection.
For the statistical distinction, NIST's paired-observations explanation, the University of York's test-selection guide and JMP's paired t-test guidance provide primary institutional or technical references. Source checks were completed on 24 October 2026. The fictional numbers and classroom exercises here are original, not official assessment items or empirical findings.
For support with independent attempts, the Bukit Timah guide to solving unfamiliar mathematical problems explains how first attempts, hints and later transfer provide different learning evidence. Apply that distinction when reviewing a test-choice answer: a correct response after the design is revealed is useful supported work, then needs a fresh independent check.
Choose the next task from the first wrong branch in your own explanation. Practise identity links if the design was misread, measurement meanings if the scale was misclassified, or test-name retrieval if the reasoning was already sound. That is more efficient than repeating an entire topic indiscriminately.
Return to learning that transfers into life for the wider standard. A learner who can explain why two observations belong together, recognise when that link disappears and adapt the analysis question has acquired more than a remembered row in a decision table.