You are viewing a free preview of this lesson.
Subscribe to unlock all 10 lessons in this course and every other course on LearningBro.
The experiment is the single most powerful tool psychology has for establishing cause and effect, and it is the reason the discipline can call itself a science at all. Every other method you will meet — observations, self-report, correlations, case studies — can reveal that two things go together; only the experiment, by actively manipulating one variable while holding everything else constant, can show that one variable produces a change in another. That causal power is precisely what makes experiments so heavily weighted in Edexcel Paper 3, and it is why the examiner expects you to be fluent not only in defining experimental terms but in applying them: identifying and operationalising variables in an unfamiliar scenario, naming an appropriate design and justifying it, and evaluating the trade-offs a researcher has made. This lesson builds that fluency, from the logic of manipulation and control through the four types of experiment, the three experimental designs, and the control techniques that protect a study's internal validity.
Key Definition: An experiment is a research method in which the researcher manipulates an independent variable (IV), measures a dependent variable (DV), and controls extraneous variables in order to test whether the IV causes a change in the DV.
By the end of this lesson you will be able to:
Edexcel 9PS0 — Paper 3: Psychological Skills (Research Methods). This lesson develops the methods content assessed in Section A of Paper 3, where research-methods questions are answered in the context of a described study (a "novel-scenario stem"). Our teaching sequence groups the material by reasoning task — first the logic of causation and variables, then experiment type, then design, then control — rather than reproducing the specification's own ordering.
| Our lesson covers | Edexcel 9PS0 research-methods area |
|---|---|
| Laboratory, field, natural and quasi-experiments | Experimental methods and the settings in which experiments are conducted |
| IV, DV, operationalisation; extraneous and confounding variables | Variables and their control |
| Independent groups, repeated measures, matched pairs | Experimental (participant) designs |
| Randomisation, counterbalancing, standardisation, blind procedures | Techniques for controlling variables and reducing bias |
Assessment Objectives. Research-methods items are dominated by AO2 (applying methodological knowledge to a described study — identifying and operationalising the IV/DV, selecting and justifying a design) and AO3 (evaluating design decisions and the validity of a study), resting on a foundation of AO1 (accurately defining experiment types, designs and controls). Pure-recall AO1 marks are comparatively few; most marks reward applied and evaluative reasoning, so a scenario-free answer scores poorly.
Connects to…
| Term | Definition |
|---|---|
| Independent variable (IV) | The variable the researcher deliberately manipulates to observe its effect |
| Dependent variable (DV) | The variable the researcher measures to detect the effect of the IV |
| Extraneous variable (EV) | Any variable other than the IV that could affect the DV if it is not controlled (a "nuisance" variable) |
| Confounding variable | An extraneous variable that has actually varied systematically with the IV, so it is impossible to tell which one changed the DV |
| Operationalisation | Defining a variable precisely in terms of how it will be manipulated or measured |
| Hypothesis | A testable, precise prediction about the relationship between the IV and DV |
The unique power of the experiment comes from combining three things: manipulation, measurement and control. By deliberately changing only the IV while holding everything else constant, the researcher engineers a situation in which — logically — any resulting change in the DV can have just one source, the IV. No other method achieves this. An observation or correlation can show that two variables move together, but only an experiment, by intervening, can show that one variable produces the change in the other. The price of this causal power is artificiality: the control required often means studying behaviour in conditions unlike everyday life. That trade-off between internal validity (causal certainty) and ecological validity (real-world relevance) is the single most important idea in the whole topic, and it runs through every evaluation you will write.
To operationalise a variable is to state exactly how it will be manipulated (the IV) or measured (the DV). "Memory" is not measurable as it stands; "the number of words correctly recalled from a 20-item list after a two-minute filled delay" is. Operationalisation matters for two reasons the examiner cares about. First, replication: another researcher can only repeat a study — and so check its findings — if the variables are defined in concrete, measurable terms. Second, objectivity: a precisely defined measure leaves less room for the experimenter's judgement to creep in. A poorly operationalised variable threatens both validity (we may not be measuring what we think we are) and reliability (different researchers measure it differently), which is why a fully operationalised hypothesis is such a common exam demand.
It can seem odd to test the null rather than the experimental hypothesis directly, but the logic is sound. We can never prove an experimental hypothesis true — there might always be an unseen exception — but we can gather enough evidence to make the "no effect" explanation implausible. So we assume the null (no effect) and ask: if that were true, how likely is the result we obtained? If that probability is very low (at or below the chosen significance level), we reject the null and, by elimination, accept the experimental (alternative) hypothesis.
graph TD
A[Does previous research<br/>predict a direction?] -->|Yes| B[Directional<br/>one-tailed hypothesis]
A -->|No / contradictory| C[Non-directional<br/>two-tailed hypothesis]
B --> D[Plus a NULL hypothesis<br/>no difference / due to chance]
C --> D
Worked example — turning an aim into a hypothesis. Suppose the aim is "to investigate whether caffeine affects reaction time".
Notice that every acceptable version specifies how the IV is manipulated (200 mg caffeine vs placebo) and how the DV is measured (mean reaction time in ms on a stated task), and that only the directional version contains a direction word ("faster").
Exam Tip: Always operationalise both the IV and the DV inside the hypothesis. A vague prediction such as "sleep affects memory" gains no marks. If asked for a directional hypothesis, you must include a direction word (more, fewer, faster, higher); a null hypothesis should always include "no significant difference" and "due to chance".
The four types differ along two dimensions: where the study takes place (setting) and, crucially, who controls the IV — the researcher, or nature.
| Feature | Description |
|---|---|
| Setting | Controlled laboratory environment |
| IV | Manipulated by the researcher |
| Control | High — extraneous variables can be tightly controlled |
Strengths: high internal validity, because tight control means changes in the DV can confidently be attributed to the IV; highly replicable, because standardised procedures can be repeated to test reliability; precise, calibrated measurement.
Limitations: low ecological validity — the artificial setting and tasks may not reflect everyday behaviour; demand characteristics — participants who know they are being studied may guess the aim and change their behaviour; investigator effects — the researcher's expectations or manner may unconsciously bias responses.
| Feature | Description |
|---|---|
| Setting | Natural, real-world environment |
| IV | Still manipulated by the researcher |
| Control | Lower than the lab — harder to control extraneous variables |
Strengths: higher ecological validity — behaviour is observed in its natural context; reduced demand characteristics, because participants are often unaware they are in a study.
Limitations: lower internal validity — extraneous variables are harder to control; ethical concerns — covert manipulation makes informed consent impossible; harder to replicate precisely.
| Feature | Description |
|---|---|
| Setting | Varies — lab or field |
| IV | NOT manipulated by the researcher — it occurs naturally (e.g. a natural disaster, a policy change, institutionalisation) |
| Control | Variable |
Strengths: allows the study of variables that cannot ethically or practically be manipulated (e.g. the effect of early institutional deprivation on children — Rutter et al., 1998); high ecological validity in real-world settings.
Limitations: the researcher cannot infer causation with confidence, because the IV was not directly manipulated and confounds abound; random allocation to conditions is impossible; difficult to replicate, because the natural event cannot be reproduced on demand.
| Feature | Description |
|---|---|
| Setting | Varies |
| IV | A pre-existing characteristic of the participants (e.g. age, sex, a clinical diagnosis) — it cannot be assigned |
| Control | Variable |
Strengths: allows comparison of groups that differ on an inherent characteristic (e.g. people with vs without a diagnosis of depression); the study around the IV can often be otherwise well-controlled.
Limitations: no random allocation, because participants already belong to their groups; cannot establish causation, as group differences may be due to confounds linked to the characteristic; participant variables are built in and cannot be eliminated.
Exam Tip: To classify an experiment in a scenario, ask "who controls the IV?" Researcher manipulates it in a lab → laboratory; researcher manipulates it in a natural setting → field; the IV is a naturally occurring event the researcher merely takes advantage of → natural; the IV is a pre-existing participant characteristic → quasi. Natural and quasi experiments cannot demonstrate cause and effect, because there is no manipulation and no random allocation.
Applying the "who controls the IV?" question to familiar studies makes the distinctions concrete:
| Study scenario | Type | IV | DV |
|---|---|---|---|
| Participants given word lists to learn under loud vs quiet conditions in a soundproofed lab | Laboratory | Noise level (loud/quiet) | Number of words recalled |
| Researchers vary how a confederate is dressed (smart/scruffy) while asking strangers for directions in a street | Field | Confederate's dress | Whether help is given |
| Comparing the development of children adopted before vs after the closure of Romanian orphanages | Natural | Age at adoption (naturally occurring) | Developmental / IQ scores |
| Comparing memory-test scores of participants with vs without a dyslexia diagnosis | Quasi | Presence of dyslexia (pre-existing) | Memory-test score |
In the first two cases the researcher manipulates the IV, so causation can in principle be inferred; in the last two the IV is not manipulated, so only an association can be claimed. A common exam task is to take a scenario like these, identify and operationalise the IV and DV, and name the experiment type with justification — always tie your justification to who controls the IV and where the study takes place.
The experimental design determines how participants are allocated to the conditions of the IV.
Different participants take part in each condition.
| Strengths | Limitations |
|---|---|
| No order effects — each participant does only one condition | Participant variables — individual differences between the groups may confound the results |
| Demand characteristics reduced — participants see only one condition | Needs roughly twice as many participants for the same power |
Control technique — random allocation: every participant has an equal chance of being placed in each condition (e.g. drawing names, using a random-number generator). This does not eliminate participant variables but should distribute them evenly across conditions. With n participants split equally between two conditions, the number of possible allocations is the binomial coefficient
(n/2n)=(2n)!(2n)!n!
so even a modest group yields a very large number of possible random splits — which is why random allocation is the standard defence against systematic bias.
The same participants take part in all conditions.
| Strengths | Limitations |
|---|---|
| Eliminates participant variables — each person acts as their own control | Order effects — performance may improve (practice) or worsen (fatigue/boredom) across conditions |
| Needs fewer participants | Demand characteristics — doing both conditions makes the aim easier to guess |
Control technique — counterbalancing: half the participants complete condition A then B, the other half B then A (an "ABBA" arrangement), so order effects are balanced across conditions rather than acting on one condition alone. For an experiment with k conditions, the number of possible orders is
k!=k×(k−1)×(k−2)×⋯×2×1
so two conditions have 2!=2 orders, three conditions have 3!=6, and four conditions have 4!=24 — which is why full counterbalancing becomes impractical as the number of conditions grows.
Order effects in detail. When the same participants do every condition, completing one condition can change performance on the next in two opposing ways. A practice effect improves later performance (participants become familiar with the task, more relaxed, or simply better at it), while a fatigue (or boredom) effect worsens it. Either way, the order in which conditions are completed becomes confounded with the IV. Counterbalancing does not remove order effects — they still occur — but it balances them, so that any practice or fatigue advantage is shared equally across both conditions and cancels out at the group level. Where even this is unsatisfactory (for example, if repeating the task is impossible because the participant now knows the answers), a matched pairs or independent groups design avoids order effects altogether.
Different participants take part in each condition, but they are paired on key variables (e.g. age, IQ, sex) likely to affect the DV.
| Strengths | Limitations |
|---|---|
| Reduces participant variables with no order effects | Time-consuming and costly — participants must be pre-tested and matched |
| No order effects — each person does one condition | Impossible to match on every variable — unmatched differences may still confound |
| Fewer demand characteristics than RMD | Needs a large pool of participants to find good matches |
Matched pairs is, in effect, a compromise that tries to capture the best of both worlds: like independent groups it uses different people in each condition (so no order effects), but like repeated measures it reduces participant variables (by pairing people on the variables most likely to matter). Its practical drawback is that matching must be done on a pre-test or on known characteristics, which is time-consuming, and matching can only ever be partial — there will always be some relevant variable on which a pair differs (you might match on age and IQ, but not on motivation). Identical twins are sometimes used as a "perfect" matched-pairs design in nature–nurture research, since they are matched on genotype.
Exam Tip: To recommend a design for a scenario, weigh three questions: would individual differences seriously distort results (→ RMD or MPD)? Would order effects be a problem, e.g. because the same test cannot be reused (→ IGD or MPD)? Is the participant pool small (→ RMD)? Always justify your choice with explicit reference to the study.
Protecting internal validity means minimising the influence of extraneous variables and preventing them from becoming confounds.
| Control method | Description | Example |
|---|---|---|
| Standardisation | Every participant experiences the same procedure, instructions and conditions | Identical scripted briefing and debriefing |
| Randomisation | Using chance to allocate participants or order stimuli, removing experimenter bias | Randomising the order of items in a word list |
| Counterbalancing | Varying the order of conditions across participants | Half do A→B, half do B→A |
| Single blind | The participant does not know which condition they are in | A placebo in a drug trial |
| Double blind | Neither the participant nor the data-collecting researcher knows the condition | Controls demand characteristics and investigator effects |
| Standardised instructions | Identical written or recorded instructions for all | Removes a source of experimenter bias |
Key Definition: A confounding variable is an extraneous variable that varies systematically with the IV, providing a rival explanation for any change in the DV. The whole point of control is to stop extraneous variables from becoming confounds.
It helps to classify the variables that threaten internal validity, because different types call for different controls:
The general principle is that an extraneous variable is only dangerous if it ends up varying with the IV. Random allocation, counterbalancing and standardisation all work by ensuring such variables are either spread evenly across conditions or held constant, so that the IV remains the only systematic difference between conditions. Note too that not every control suits every design: counterbalancing only makes sense in a repeated measures design (there is no order to counterbalance if each participant does one condition), whereas random allocation only applies when different participants are assigned to conditions, as in independent groups. Matching the control to the design — and explaining which threat it neutralises — is exactly the applied reasoning that distinguishes a strong answer.
Key Definition: Demand characteristics are cues in a study that may reveal its purpose to participants, leading them to change their behaviour in ways that do not reflect their genuine responses.
Orne (1962) showed that participants actively try to work out the aim of a study and behave accordingly — either helping the researcher (the good-participant effect) or deliberately sabotaging it. Either way, behaviour reflects the participant's interpretation of the study rather than the variable of interest.
Key Definition: Investigator effects are any influence — through expectations, manner, appearance or interpretation of data — that the researcher has on the outcome of a study.
Rosenthal and Fode (1963) demonstrated this powerfully: students told their lab rats were "maze-bright" obtained better maze performance than students told theirs were "maze-dull", even though the rats had been allocated at random. This experimenter-expectancy effect is the rationale for double-blind procedures. A pilot study — a small-scale trial run before the main study — is used to spot such problems in advance, along with ambiguous instructions or a DV that does not respond to the IV, saving time and money before the full study runs.
The whole purpose of careful design and control is to produce findings that are valid (accurate) and reliable (consistent). The design choices above feed directly into these qualities.
| Type of validity | Question it asks | How experiments protect it |
|---|---|---|
| Internal validity | Was the change in the DV genuinely caused by the IV (not a confound)? | Control of extraneous variables, randomisation, single/double blind, standardisation |
| External validity | Do the findings generalise beyond the study? | Representative sampling; realistic tasks and settings |
| Ecological validity | Do they generalise to everyday settings and behaviour? | Field experiments; mundane realism in tasks |
| Temporal validity | Do they still hold across time and eras? | Replication in later cohorts |
There is a characteristic trade-off: the tight control that gives a laboratory experiment its high internal validity simultaneously lowers its ecological validity, because the controlled setting and artificial task are unlike everyday life. Field and natural experiments sit at the opposite end. Recognising and discussing this trade-off is central to AO3, and it is the natural spine for almost any evaluation of an experimental study.
Reliability is assessed largely through replication: because experiments use standardised, operationalised procedures, another researcher can repeat them exactly and check whether the same results emerge. Consistent replication indicates the findings are reliable rather than a one-off fluke — which is precisely why operationalisation and standardised instructions matter so much, and why the experimental method is often held up as psychology's most scientific.
The laboratory experiment's outstanding strength is its high internal validity, which makes it uniquely able to establish cause and effect. Because extraneous variables are rigorously controlled and the IV is deliberately manipulated, a change in the DV can be confidently attributed to the IV rather than to confounds. The implication is that, when the research question is genuinely about causation, the lab experiment is the gold standard — no other method licenses causal claims so securely. Its weakness is the mirror image: the very control that buys internal validity creates artificial conditions and tasks, lowering ecological validity, so findings may not generalise to everyday behaviour.
Field and natural experiments trade internal validity for ecological validity, which makes them stronger for real-world relevance but weaker for causal inference. Studying behaviour in its natural setting (field), or exploiting a naturally occurring IV (natural), means findings are more likely to reflect real life and reduces demand characteristics. However, because extraneous variables are harder to control — and, in natural experiments, participants are not randomly allocated — rival explanations cannot be ruled out. The implication is that these designs are well suited to questions that cannot ethically be brought into a lab (e.g. the effects of deprivation), but their conclusions must be framed as associations rather than proven causes.
Quasi-experiments allow the study of inherent participant characteristics but cannot demonstrate causation, which is their defining limitation. Comparing, say, people with and without a diagnosis lets psychologists investigate clinically important groups, yet because participants are not randomly allocated, any difference may stem from confounds bound up with the characteristic rather than the "IV" itself. The implication is that quasi-experimental findings are valuable for description and hypothesis-generation but should never be over-interpreted as showing that the characteristic causes the outcome.
Repeated measures and matched pairs control participant variables better than independent groups, but each carries its own cost, so design choice is always a trade-off. Repeated measures removes individual differences entirely but introduces order effects and stronger demand characteristics; matched pairs avoids order effects but is time-consuming and can never match on every relevant variable; independent groups is quick and order-effect-free but is vulnerable to participant variables. The implication is that there is no universally "best" design — the appropriate choice depends on whether participant variables or order effects pose the greater threat in a particular study, which is exactly the judgement examiners reward.
A general limitation of all experiments is their vulnerability to demand characteristics and investigator effects, which threaten validity even under tight control. Orne (1962) and Rosenthal and Fode (1963) showed that both participants' guesses and researchers' expectations can distort results. The implication is that good experimental practice must go beyond manipulating the IV: single- and double-blind procedures, standardised instructions and (where ethical) deception are needed to stop these subtle biases from masquerading as genuine effects.
Specimen question modelled on the Edexcel 9PS0 paper format.
A psychologist wanted to find out whether background music affects how well students concentrate. She asked 40 students to complete a 15-minute proofreading task once while listening to fast-tempo music and once in silence. She recorded how many spelling errors each student correctly identified in each condition.
(a) Identify the experimental design used and explain one limitation of this design in the context of this study. (3 marks)
(b) Write a fully operationalised directional hypothesis for this study. (3 marks)
AO breakdown. These items are entirely application and analysis — a realistic split is AO2 = 4 marks, AO3 = 2 marks, with no pure AO1. Part (a) requires you to identify the design (AO2) and apply a relevant limitation to this scenario (AO3); part (b) requires you to operationalise both variables in the context of the study (AO2). Generic, scenario-free answers are heavily penalised.
Mid-band response
(a) It is a repeated measures design because the same students do both conditions. One limitation is order effects — doing the proofreading task twice means they might get better with practice. (b) Students will find more spelling errors when listening to music than in silence.
Examiner-style commentary: The design is correctly named (AO2) and the type of limitation is right, but the practice effect is not tied to this proofreading study, so the AO3 application is thin. In (b) the hypothesis is directional but not operationalised — no measure and no task detail — so it earns partial credit only. To reach the next band: state how practice specifically distorts the second proofreading condition, and put the operationalised IV and DV inside the hypothesis.
Stronger response
(a) The design is repeated measures, as the same 40 students complete both the music and the silence conditions. A limitation in this study is order effects: because students proofread in both conditions, practice could mean they identify more errors in whichever task they do second, regardless of the music, which confounds the IV. (b) Students will correctly identify significantly more spelling errors during the 15-minute proofreading task when listening to fast-tempo music than when working in silence.
Examiner-style commentary: Part (a) now applies order effects to the scenario and explains the confound, securing the application marks; part (b) operationalises both variables and keeps the direction explicit. What separates this from top-band is precision on mechanism and consequence: naming both practice and fatigue, linking the confound explicitly to reduced internal validity, and offering the control (counterbalancing) that a top answer volunteers.
Top-band response
(a) The experimental design is repeated measures, because the same 40 students take part in both the fast-tempo-music condition and the silence condition. A limitation in this study is the risk of order effects: as each student proofreads a passage twice, a practice effect could improve error-detection in whichever condition they complete second, while fatigue could worsen it — either way, the second-condition scores are distorted by order rather than by the music, so the IV (music vs silence) is confounded and internal validity is reduced. This could be controlled by counterbalancing (half do music first, half silence first). (b) "Students will correctly identify significantly more spelling errors in a 15-minute proofreading task when listening to fast-tempo music than when completing the task in silence." This operationalises the IV (presence vs absence of fast-tempo music) and the DV (number of spelling errors correctly identified in a 15-minute proofreading task) and, by stating "more", is directional.
Examiner-style commentary: Full marks on both parts. Part (a) is fully contextualised — it explains how practice and fatigue distort the second condition specifically, links this to internal validity, and offers counterbalancing as a control, which is the discriminator between a Stronger and a Top-band answer on this item type. Part (b) operationalises both variables and keeps the direction explicit. Nothing further is required.
This content is aligned with the Edexcel A-Level Psychology (9PS0) specification.