You are viewing a free preview of this lesson.
Subscribe to unlock all 10 lessons in this course and every other course on LearningBro.
Two decisions shape the architecture of almost every experiment: how you allocate participants to conditions (the experimental design) and how you recruit participants in the first place (the sampling method). Both are heavily examined in Component 01, and both carry consequences that ripple through the whole study — a poor design lets participant variables masquerade as effects; a poor sample makes the findings ungeneralisable.
This lesson covers the three experimental designs (repeated measures, independent measures, matched participants), the problems each is designed to solve or creates — chiefly order effects and participant variables — and the techniques used to manage them, especially counterbalancing. It then covers the difference between a target population and a sample, the four sampling methods OCR names (random, snowball, opportunity, self-selected), and the ever-present risk of sampling bias. These are exactly the tools you deploy in your own Practical Activities, and "identify the design and explain one weakness" is a standard exam item.
| This lesson covers | OCR H567 Component 01 sub-area | AO focus |
|---|---|---|
| Repeated measures, independent measures, matched participants | 1.2 Planning & conducting — experimental designs | AO1; AO3 evaluation |
| Order effects and counterbalancing | 1.2 — control of extraneous variables | AO1; AO2 |
| Participant variables | 1.2 — designs; control | AO3 |
| Target population and sample | 1.2 — target population & sample | AO1 |
| Random, snowball, opportunity, self-selected sampling; sampling bias | 1.2 — sampling | AO1; AO2; AO3 |
Referenced descriptively; see the official OCR H567 specification document for exact wording. This lesson develops AO1 (defining designs and sampling methods), AO2 (selecting an appropriate design/sample for a scenario) and AO3 (evaluating the trade-offs, especially order effects vs participant variables, and representativeness vs practicality).
An experimental design is the way participants are allocated to the different conditions of the IV. There are three.
In an independent-measures design, different participants take part in each condition — one group experiences condition A, a separate group experiences condition B. Their scores are then compared.
Strengths. No order effects, because each participant does only one condition. The same materials can be reused across conditions without practice contaminating results. Participants cannot guess the aim by comparing conditions, reducing demand characteristics.
Weaknesses. Participant variables — differences between the people in each group (ability, mood, motivation) — may confound results, because the groups are made of different individuals. More participants are needed to get the same amount of data. Random allocation to conditions is used to spread participant variables evenly.
The signature weakness of independent measures — participant variables — deserves unpacking, because the phrase is often written without being understood. Since different people populate each condition, the two groups may differ before the IV is even applied: one group might, by chance, contain more able, more motivated, or better-rested individuals. If that group then scores higher, we cannot tell whether the IV or the pre-existing difference produced the effect. The remedy is random allocation — using chance to decide who goes into which condition — which does not equalise the groups exactly but ensures that participant variables are, on average, spread evenly rather than clustered. Random allocation is thus the independent-measures analogue of counterbalancing: both use chance to stop a nuisance factor from lining up systematically with the IV. It is worth stressing that random allocation (assigning existing participants to conditions) is distinct from random sampling (selecting participants from the population); a study can use one without the other, and confusing the two is a common error.
In a repeated-measures design, the same participants take part in all conditions, and each person acts as their own control.
Strengths. Participant variables are controlled, because the same people appear in every condition — any difference cannot be due to who is in which group. Fewer participants are needed.
Weaknesses. Order effects — doing one condition first can affect performance on the next through practice (improvement) or fatigue/boredom (deterioration). Participants are more likely to guess the aim across conditions, increasing demand characteristics. Different but equivalent materials may be needed for each condition.
The great virtue of repeated measures is worth appreciating fully: by using each participant as their own control, it eliminates participant variables as a source of difference between conditions, because the very same people — with the very same abilities, personalities and moods — appear in both. This makes it a powerful and economical design, needing far fewer participants than independent measures to obtain the same quantity of data (every person yields two scores instead of one). But it purchases this power at the cost of two threats absent from independent designs. The first is order effects, already discussed. The second is heightened demand characteristics: a participant who experiences every condition has far more opportunity to notice what is being manipulated, infer the hypothesis, and — consciously or not — adjust their behaviour to fit it. A participant who revises once in silence and once with music can hardly fail to guess that the study concerns music and memory. Countermeasures include using a cover story, disguising the true purpose, or spacing the conditions apart in time, but the vulnerability is real and is a legitimate evaluative point whenever a repeated-measures study is described. Choosing repeated measures therefore means judging that the elimination of participant variables outweighs the risks of order effects and demand characteristics for the particular study in hand.
In a matched-participants design, different participants are used in each condition (as in independent measures), but they are matched in pairs on variables relevant to the study (e.g. age, IQ, sex), with one member of each pair assigned to each condition.
Strengths. Reduces participant variables (the matching controls the key confounds) and avoids order effects (different people in each condition). A compromise capturing benefits of both other designs.
Weaknesses. Matching is time-consuming and difficult — you can never match on every variable, and finding matched pairs is laborious. If a participant drops out, their matched partner's data may be lost too.
The matched-participants design deserves careful thought because it is the most conceptually interesting of the three. Its ambition is to enjoy the participant-variable control of a repeated-measures design without incurring order effects, and it achieves this by making the two groups equivalent through deliberate matching rather than through being the same people. But the ambition is only partly realisable. First, you can only match on the variables you anticipate mattering and can measure — match on age and IQ, and an unmeasured difference in motivation or prior experience may still confound the groups. Second, matching is genuinely laborious: it requires measuring the matching variables for a large pool of potential participants and then pairing them, which is costly in time and often in participant goodwill. Third, the design is fragile — because the analysis treats each pair as a unit, the loss of one member (through withdrawal or missing data) usually means discarding their partner too, shrinking the sample. For these reasons matched participants is used when the participant variable is known, important and measurable and order effects genuinely cannot be tolerated; otherwise a simpler design is often preferred. A special and elegant case is the use of identical twins as matched pairs in studies of nature versus nurture, where the "matching" on genotype is as close to perfect as biology allows.
| Independent measures | Repeated measures | Matched participants | |
|---|---|---|---|
| Participants per condition | Different | Same | Different (matched pairs) |
| Order effects | None | Present (need counterbalancing) | None |
| Participant variables | A problem (random allocation helps) | Controlled | Reduced by matching |
| Participants needed | More | Fewer | More (plus matching effort) |
Order effects are the central weakness of repeated measures. Counterbalancing controls them by varying the order in which participants experience the conditions. In the simplest ABBA approach, half the participants do condition A then B, and the other half do B then A. Practice and fatigue effects then cancel out across the sample rather than favouring one condition.
Counterbalancing does not eliminate order effects for any individual — it balances them across the group so they do not systematically advantage one condition. That distinction is a reliable AO3 discriminator.
It is worth understanding why order effects are so corrosive and why balancing (rather than eliminating) is the best available remedy. In a repeated-measures study, every participant contributes to both conditions in some sequence, and the mere act of doing a task once changes the person who then does it again — they may be quicker through practice, slower through fatigue, or simply less naive about the study's purpose. If everyone does condition A first and B second, then all of the practice benefit lands on B, and a difference favouring B could be entirely an artefact of order rather than any real property of the B condition. Counterbalancing breaks this systematic link: with half the sample doing A–B and half doing B–A, whatever practice or fatigue accrues is distributed roughly equally across the two conditions, so it no longer masquerades as a treatment effect. The residual order effects for individuals remain, but they cancel at the group level, which is what the statistical comparison cares about. A subtle point that impresses examiners: counterbalancing controls order effects but does not control asymmetrical (differential) transfer, where doing condition A specifically changes how you respond to B in a way that doesn't happen in reverse — in such cases even a repeated-measures design with counterbalancing can be compromised, and an independent-measures design may be safer.
The choice between designs is therefore rarely obvious, and the best answers weigh the specific costs against the specific study. Independent measures trades the elimination of order effects against exposure to participant variables; repeated measures trades control of participant variables against exposure to order effects; matched participants attempts to secure both benefits but pays in the time, difficulty and imperfection of matching, and in the loss of a whole pair's data if one partner withdraws. When a question asks you to choose or evaluate a design, the marks come from naming which threat matters most for this study and justifying the design that best manages it.
graph TD
A["How are participants allocated<br/>to the conditions?"] --> B["Same people in every condition"]
A --> C["Different people in each condition"]
B --> D["Repeated measures<br/>+ controls participant variables<br/>− order effects → counterbalance"]
C --> E{"Matched on key variables?"}
E -->|No| F["Independent measures<br/>+ no order effects<br/>− participant variables → random allocation"]
E -->|Yes| G["Matched participants<br/>+ no order effects, fewer participant variables<br/>− matching is laborious"]
style D fill:#2980b9,color:#fff
style F fill:#27ae60,color:#fff
style G fill:#8e44ad,color:#fff
The target population is the entire group of people the researcher is interested in and wishes to draw conclusions about (e.g. "UK secondary-school students"). Because it is almost never possible to study everyone, the researcher studies a sample — a smaller subset — and, if the sample is representative, generalises the findings back to the population.
The sampling method is how that subset is selected. A good sampling method maximises representativeness so that findings generalise; a poor one introduces sampling bias, where certain subgroups are over- or under-represented and the sample no longer mirrors the population.
The stakes of sampling are easy to underestimate but genuinely high. A study can be immaculately designed, tightly controlled and correctly analysed, yet if its sample does not represent the population, the conclusions cannot be safely extended beyond the particular people studied — a limitation of external (population) validity that no amount of internal rigour can repair. Much of psychology's most-cited research has been criticised precisely here: samples drawn overwhelmingly from WEIRD populations (Western, Educated, Industrialised, Rich, Democratic — and disproportionately university undergraduates) may not generalise to humanity at large, and findings assumed to be universal have sometimes turned out to be culturally specific. Milgram's original obedience participants, for instance, were all American men recruited through a newspaper advertisement, which raises questions about generalising to women, to other cultures, and to people who would not volunteer for a psychology study. Recognising that the quality of the sample directly bounds the reach of the conclusions is a central evaluative theme of the whole subject.
There is also an unavoidable practical dimension. Researchers operate under real constraints of time, money and access, and the "best" sampling method in principle (usually random) is often the most demanding to execute. The result is a constant tension between methodological ideal and practical feasibility, and much real-world psychology uses opportunity or volunteer samples not because researchers are ignorant of their weaknesses but because a perfect random sample of the target population is simply unattainable. A mature evaluation acknowledges this trade-off rather than treating opportunity sampling as a naive mistake: the honest question is not "did they use the ideal method?" but "how much does the method they could realistically use limit what we can conclude?"
OCR names four sampling methods. Know the procedure, a strength and a weakness of each.
In random sampling every member of the target population has an equal chance of being selected — for example, by assigning everyone a number and using a random number generator or drawing names from a hat.
Strengths. Free from researcher bias in selection; the most likely of the four to be representative, so findings generalise.
Weaknesses. Requires a complete list of the population (often unavailable); time-consuming; even a random sample can, by chance, be unrepresentative. Chosen participants may decline, reintroducing bias.
Subscribe to continue reading
Get full access to this lesson and all 10 lessons in this course.