Research methods — AQA A-Level Psychology
Test yourself on Research methods with AQA A-Level practice questions.
7 days Premium · Then the free plan continues · No card, no charge
Research methods explained
Research methods, scientific processes, and techniques of data handling and analysis are examined across all topics.
Read the full explanation
A central theme is the trade-off between control and realism. A laboratory study increases internal validity, but risks low mundane realism. Conversely, a naturalistic observation increases ecological validity but sacrifices control. Students must demonstrate knowledge and understanding of these methods, be familiar with their use, and apply strengths and limitations to specific scenarios. Data analysis requires selecting statistical tests based on design and level of measurement. Graph choice depends on data type and whether testing a difference or correlation. Central tendency depends on the level of measurement and data distribution, such as skewness or outliers.
Experimental method. Types of experiment, laboratory and field experiments; natural and quasi-experiments.
An experiment manipulates one variable and measures another while holding the rest steady, so a difference in the dependent variable can be pinned on the independent variable. In a laboratory the researcher controls the setting as well as the manipulation: Loftus and Palmer got mean speed estimates of 40.8 mph for "smashed" against 31.8 mph for "contacted", a difference only that control makes interpretable, though artificial tasks invite demand characteristics. A field experiment moves the same deliberate manipulation into the participants’ own setting, gaining realism and losing control. In a natural experiment the variable changes without the researcher, as with Rutter’s Romanian adoptees; in a quasi-experiment it is a fixed characteristic such as age or sex. Neither allows random allocation, so participant variables may confound the result.
Observational techniques. Types of observation: naturalistic and controlled observation; covert and overt observation; participant and non-participant observation.
Observation records what people do rather than what they say they do. It is a technique, not a design. Naturalistic observation means behaviour is watched where it normally happens, so realism is high and extraneous variables are uncontrolled. Controlled observation means the setting is structured, like Ainsworth's Strange Situation, buying control at the cost of realism. Covert watching removes demand characteristics but cannot take informed consent; overt keeps the ethics but invites reactivity. Participant observation gives insight while risking the researcher going native; non-participant maintains objectivity. Whichever is used, categories must be operationalised and two coders should agree at around 0.80 or better.
Self-report techniques. Questionnaires; interviews, structured and unstructured.
Self-report asks people about themselves, accessing thoughts and feelings. A questionnaire is written and fixed, so hundreds can be distributed cheaply with identical wording. Closed items, like a five-point Likert scale, produce quantitative data that can be summarised and tested, while open items produce richer material that is slow to analyse. Spoken methods differ in what is decided in advance: a structured interview is a questionnaire read aloud, replicable but with no room to follow up, while an unstructured interview starts from a topic and follows the respondent, gaining depth but losing standardisation. Social desirability bias is the standing weakness, and leading or double-barrelled items make it worse.
Correlations. Analysis of the relationship between co-variables. The difference between correlations and experiments.
A correlation measures how two co-variables move together. Both are measured, neither is manipulated and nothing is controlled, separating it from an experiment and explaining why it cannot show cause. The coefficient runs from minus one to plus one: the sign gives the direction and the size gives the strength, so -0.76 is stronger than 0.31. Plot the pairs on a scattergram and the pattern is visible at a glance. Because nothing was manipulated, an untested third variable may be driving both, and a curved relationship can produce a coefficient near zero even where the link is strong. The pay-off is a quick, ethical first look at variables nobody could manipulate, and a source of hypotheses for future experiments.
Content analysis.
Content analysis turns existing media—adverts, diaries, tweets—into data, observing artefacts rather than people. The procedure involves sampling material, agreeing coding units in advance, and tallying appearances. Coding yields quantitative frequency counts that can be graphed. Thematic analysis instead lets recurring qualitative ideas emerge as themes, which are re-applied to fresh material for verification. For example, Manstead and McCulloch coded 170 British TV adverts, finding men often in autonomous roles and women in dependent ones. The material is real and unaffected by being studied, eliminating demand characteristics, but coding relies on subjective judgement, meaning inter-rater reliability must be established.
Case studies.
This is a detailed investigation of one person, family, institution or event, usually followed over a long period and usually using several methods at once: interviews, observation, psychometric testing, medical records and accounts from family, so the data are mainly qualitative and unusually rich. The rare instances are the point. Removal of HM’s hippocampus left him able to learn new motor skills while unable to form new episodic memories, which is direct evidence that memory is not one store, and KF’s poor digit span alongside intact long-term recall contradicted the multi-store model’s single route. The strength is also the weakness: one person cannot be generalised from, nothing can be replicated, interpretation leans on the researcher, and family history is collected from memories that may be wrong.
Your focus
- Explain the trade between control and realism by contrasting internal validity with ecological validity.
- Apply the level of measurement and data distribution to select the correct measure of central tendency.
- Select appropriate graphs based on data type and whether investigating a difference or correlation.
Show all 24 objectives
- Evaluate a study described in a stem by tying each strength and limitation to its details and conclusion.
- Describe the defining features of an experiment, naming the manipulated variable, the measured variable and the controls.
- Distinguish laboratory, field, natural and quasi-experiments by who manipulates the variable and where it happens.
- Identify the type of experiment used in a stem and justify the choice from how the variable came to vary.
- Evaluate natural and quasi-experiments, which allow no random allocation, so participant variables may confound results.
- Describe what changes between naturalistic and controlled, covert and overt, and participant and non-participant observation.
- Apply the three pairs to a described study by naming which type it used and quoting the detail in the stem that shows it.
- Evaluate a chosen observation type by trading its control and its ethics against realism and reactivity.
- Outline how questionnaires, structured interviews and unstructured interviews differ in how much is fixed before data collection.
- Write closed and open items on a given topic that avoid leading, double-barrelled and emotive wording.
- Evaluate a self-report study by explaining how social desirability and interviewer effects could have shaped its answers.
- Explain why a correlation has co-variables rather than an IV and a DV, and what that costs it in causal claims.
- Identify from coefficients such as -0.76 and 0.31 which relationship is the stronger and which way each one runs.
- Evaluate a reported correlation by naming a plausible third variable from the scenario that could produce both results.
- Explain why a coefficient near zero does not prove that two variables are unrelated.
- Describe the order of a content analysis from sampling the material through operationalising coding units to counting.
- Apply coding to given material such as adverts by proposing categories and saying what the tallies would show.
- Evaluate content analysis by weighing the absence of demand characteristics against observer bias in choosing categories.
- Describe what a case study involves, naming the methods combined and the kind of data each one produces.
- Explain how HM and KF each forced a change to memory theory that a large-sample study could not have delivered.
- Evaluate the case study by setting its unique richness against the impossibility of generalising or replicating it.
Research methods exam tips
Marking Points
- Identifying the method used in the stem and justifying that identification from a detail in the stem, demonstrating knowledge and understanding.
- Internal validity: whether the effect was really caused by the manipulated variable rather than by an extraneous or confounding one.
- Reliability as consistency, established by test-retest, inter-observer agreement or split-half, and kept distinct from validity.
- Strengths and limitations applied to the scenario, with the consequence stated: what the weakness does to the conclusion.
- Data handling: graph choice depends on data type and whether testing a difference or correlation, while central tendency depends on the level of measurement and data distribution (e.g., skewness).
- The defining features: an independent variable manipulated, a dependent variable measured, other variables controlled, so cause can be inferred.
- Laboratory experiment: manipulation in a controlled, often artificial setting, giving high internal validity, standardisation and easy replication.
- Field experiment: the researcher still manipulates the variable, but in the participants’ natural environment, raising realism and reducing control.
- Natural experiment: the independent variable varies for reasons outside the researcher’s control and would have varied anyway, as in Rutter’s Romanian orphan study.
- Quasi-experiment: the independent variable is a pre-existing characteristic of the participants, such as age, sex or a diagnosis, and cannot be manipulated at all.
- Neither natural nor quasi-experiments permit random allocation, so participant variables are uncontrolled and causal conclusions are weaker.
- Evaluation: laboratory control against low mundane realism and demand characteristics; field realism against uncontrolled extraneous variables and consent problems.
- Observation defined as recording behaviour directly, and distinguished from a design: it can be used within an experiment or on its own.
- Naturalistic observation: behaviour observed in its usual setting with nothing manipulated, high external validity, low control, hard to replicate.
- Controlled observation: some variables regulated and the setting structured, as in Ainsworth's Strange Situation, so replication is easier and realism lower.
- Covert observation: participants unaware, so behaviour is natural and demand characteristics are removed, but informed consent and privacy are compromised.
- Overt observation: participants know, which is ethically defensible but produces reactivity and social desirability in what is displayed.
- Participant observation: the researcher joins the group, gaining insight, risking loss of objectivity and going native; non-participant keeps distance and objectivity but loses depth.
- Behavioural categories must be operationalised, observable, mutually exclusive and cover all likely behaviour, then be sampled by event or by time.
- Inter-observer reliability: two observers code the same behaviour independently and their totals are correlated, with about 0.80 taken as acceptable agreement.
- Self-report defined as the participant supplying the data about their own thoughts, feelings or behaviour, reaching what cannot be observed.
- Questionnaire: a fixed set of written items, delivered without a researcher present, so large samples are cheap and wording is identical for everyone.
- Closed questions give quantitative data that is quick to analyse and compare but restricts the answer; open questions give qualitative depth that is harder to summarise.
- Structured interview: pre-set questions asked in a fixed order, so it is replicable, comparable across respondents and needs little training.
- Unstructured interview: a topic rather than a schedule, with follow-up questions, producing rich qualitative data but low standardisation and difficult analysis.
- Design points that earn marks: no leading, double-barrelled, emotive or jargon-laden items, a sensible mix of closed and open, filler items to disguise the aim, and a pilot study.
- Evaluation: social desirability bias, acquiescence and response set, interviewer effects in spoken methods, and low response rates for postal questionnaires.
- Co-variables: both are measured as they are found, neither is manipulated, so there is no independent or dependent variable.
- Direction read from the sign and strength read from the magnitude, on a scale running from minus one through zero to plus one.
- The scattergram as the display, with one co-variable on each axis and one point per participant.
- No causal conclusion: without manipulation and control of extraneous variables, the association may run either way or come from a third variable.
- The third-variable problem named with a plausible example from the scenario rather than as an abstract caution.
- A curvilinear relationship can give a coefficient near zero, so a weak coefficient does not prove the variables are unrelated.
- Strengths: uses existing data economically, is ethically possible where manipulation is not, and generates hypotheses for later experimental work.
- Defined as an indirect observational technique applied to communications and artefacts rather than to people behaving in front of a researcher.
- Procedure in order: sample the material, decide and operationalise the coding units or categories, code systematically, count or summarise.
- Coding produces quantitative frequency data; thematic analysis produces qualitative themes that are then tested against new material.
- A named example with its finding, such as Manstead and McCulloch’s coding of 170 television adverts showing men in autonomous and women in dependent roles.
- Strengths: the material was produced for real purposes, so external validity is high, and studying it creates no demand characteristics.
- Strengths: publicly available material raises few ethical problems, and clearly stated coding units make the study replicable.
- Limitations: observer bias in deciding what counts, material analysed outside the context that produced it, and categories that carry the researcher’s own culture.
- Reliability established by a second researcher coding the same material independently and the two sets of tallies being correlated.
- Defined as an in-depth, usually longitudinal study of a single individual, group, institution or event.
- Multiple methods used together - interview, observation, psychometric tests, records and reports from others - so the account is triangulated.
- Data are mainly qualitative, although quantitative measures such as IQ or memory-span scores are routinely included.
- A named example with what it demonstrated, such as HM retaining procedural learning while losing the ability to form new episodic memories.
- Strength: the only way to study cases too rare or too unethical to create, which is where much of what is known about brain damage comes from.
- Strength: a single case can falsify a general theory, as KF’s intact long-term memory alongside impaired short-term memory did to the multi-store model.
- Limitation: the case is by definition unusual, so generalising to a population is unsafe.
- Limitation: the researcher’s interpretation shapes the account, replication is impossible, and retrospective family reports are of uncertain accuracy.
Examiner Tips
- 💡Quote the stem. Naming the participants, the setting or the measure from the scenario is what turns a generic point into an applied mark.
- 💡For 'explain one limitation' write three parts: name it, say how it operates in this study, say what it does to the conclusion.
- 💡When asked to identify the type of experiment, justify it by saying who or what set the independent variable, quoting the stem.
- 💡Operationalise both variables as they would be recorded: the two conditions for the independent variable, the units or score for the dependent variable.
- 💡For "explain one limitation of this natural experiment", go straight to the absence of random allocation and name a plausible participant variable in the scenario.
- 💡Do not evaluate laboratory work as automatically weak; say which conclusion in this scenario is threatened, and why.
- 💡When asked to design an observation, give at least three operationalised categories drawn from the stem and name the sampling method with its interval.
- 💡Explain inter-observer reliability as a procedure: two observers, same behaviour, independent recording, correlate the tallies.
- 💡Compare on the dimension the question names; a question about covert versus overt gets nothing for a paragraph about naturalistic settings.
- 💡If the stem gives a tally chart, say what type of data it produces before choosing a graph or a test.
- 💡Asked to write two questions, give one closed item with its response options printed out and one open item, both clearly about the stem’s topic.
- 💡Justify each design choice against the aim in the stem rather than in general, for instance why depth matters more than comparability here.
- 💡Explain a limitation in context: say what a participant in this particular study would be tempted to under-report or exaggerate.
- 💡If asked to improve an item, rewrite it in full rather than describing the improvement.
- 💡Phrase a directional hypothesis as "there will be a positive correlation between" the two co-variables, each operationalised as it was measured.
- 💡When asked why a conclusion is not justified, name a specific third variable that fits the scenario rather than saying correlation is not causation.
- 💡On a scattergram question, label both axes with the co-variables and their units, and plot one point per pair of scores.
- 💡State the level of measurement before choosing between the correlation tests available in the statistical tables.
- 💡Answer "how would you carry this out" as an ordered procedure, with the coding units taken from the material named in the stem.
- 💡Say which type of data the method will produce, because that decides the graph and the statistical test in later parts of the question.
- 💡Evaluate by comparison: contrast it with interviewing the same people, where demand characteristics would appear.
- 💡If the question says analyse qualitatively, describe themes emerging and being verified, not a tally chart.
- 💡Name a real case and what it showed; a specific example turns a generic evaluation sentence into a credited one.
- 💡Where the stem describes one participant studied in depth over time, identify the method and quote the detail from the stem that proves it.
- 💡Balance the answer, because an evaluation that is all limitations cannot reach the top band on a discuss question.
- 💡If asked how the findings could be checked, suggest further cases with the same profile rather than replication of this one.
Common Mistakes
- Reciting a memorised list of strengths and limitations that never mentions the study in the stem, which caps an application answer at the lower mark bands; always link to the scenario.
- Using reliability and validity as synonyms, so that 'the questionnaire was unreliable' is offered as evidence that it measured the wrong thing; keep consistency and accuracy distinct.
- Assuming central tendency depends solely on the level of measurement, ignoring that extreme outliers make the median more appropriate than the mean for interval data.
- Calling any study done outdoors a natural experiment; the test is who set the independent variable, and in a field experiment the researcher still sets it.
- Using "quasi-experiment" for anything not in a laboratory, when its defining feature is that the variable is a characteristic the participant already had.
- Claiming a laboratory experiment removes extraneous variables, when control reduces them and demand characteristics remain.
- Writing an unoperationalised independent variable, for instance "noise", instead of the two conditions a participant would actually experience.
- Treating naturalistic and covert as the same thing; they are independent choices, and a controlled observation can be covert behind a one-way mirror.
- Writing behavioural categories that are inferences, such as "being aggressive" or "showing anxiety", instead of countable acts such as "hits the doll with the mallet".
- Saying covert observation is simply unethical, without the qualification that observing public behaviour where privacy is not expected is normally acceptable.
- Swapping event and time sampling, or describing time sampling without stating the interval that makes it time sampling.
- Writing that questionnaires are "unreliable" when the point being reached for is social desirability bias or acquiescence, which are validity problems.
- Calling any spoken method unstructured; a fixed list of questions read out is a structured interview however informal it sounds.
- Producing example items that are themselves leading or double-barrelled, such as asking about stress and sleep in one question.
- Assuming all questionnaire data is quantitative, then choosing a statistical test for answers that are open-ended prose.
- Reading a negative coefficient as a weak one, so -0.8 is ranked below 0.4 when it is the stronger relationship.
- Writing the hypothesis as a difference between conditions instead of a relationship between two operationalised co-variables.
- Sliding into causal language in the conclusion, saying the data show that one variable causes the other.
- Labelling one co-variable as the independent variable, which throws away the very distinction the question is testing.
- Describing it as an experiment, or as simply reading through the material and writing an impression of it.
- Offering coding categories that are interpretations, such as "sexist", instead of countable units such as "the number of adverts with a male voice-over".
- Losing the distinction between coding, which is quantitative, and thematic analysis, which is qualitative.
- Asserting there are no ethical issues at all, when private material such as diaries or therapy notes still needs consent.
- Describing it as one method, usually an interview, and losing the mark for the range of techniques combined in it.
- Saying it has low validity without distinguishing: the detail about the individual can be very valid, and it is generalisation that fails.
- Confusing the method with an illustrative example used to support a theory in an essay.
- Evaluating only negatively, when falsifying a theory and reaching otherwise unreachable cases are creditable strengths.