Research methodology

A method that manipulates an IV and measures its effect on a DV, controlling other variables.

The variable the researcher changes on purpose.

The variable the researcher measures.

One that randomly allocates participants to conditions.

One where the IV is a pre-existing feature (e.g. age), so no random allocation is possible.

It spreads individual differences evenly, so DV changes are more likely caused by the IV.

Control and random allocation isolate the IV as the likely cause of DV changes.

Best method for showing cause and effect, with high control and replicability.

Control can make them artificial, and participants may show demand characteristics.

Causality — they test whether the IV causes a change in the DV.

A testable prediction about the effect of the IV on the DV.

The experimental group gets the treatment (IV); the control group does not and is the comparison.

Extraneous = any other variable that could affect the DV; confounding = one that changes with the IV and ruins the cause-claim.

Repeated measures (same people), independent measures (different people, random allocation), matched pairs (pre-tested and paired).

Practice or fatigue from doing conditions in a set order; reduced by counterbalancing.

The IV is manipulated in a real-life setting: more natural, less control.

The IV changes naturally (e.g. a lockdown); no random allocation; useful when manipulation would be unethical.

Longitudinal follows the same people over time; cross-sectional compares different groups at one moment.

To state exactly how it will be changed or measured, for example "memory" as "words recalled out of 20".

Ecological validity: the setting is like real life. Mundane realism: the task is like everyday life.

How sure we can be that the IV, and nothing else, caused the change in the DV.

A method that studies behaviour by watching and systematically recording what people do.

Naturalistic = real setting (high ecological validity); controlled = set-up situation (more control).

Covert = people don't know they're watched; overt = they know.

Participant = the researcher joins the group; non-participant = watches from outside.

Actions defined so exactly that they can be seen and counted, such as "playing alone" or "speaks in Spanish".

People changing their behaviour because they know they are being observed.

Captures real behaviour directly, often with high ecological validity.

It shows what people do, not why, and can suffer observer bias.

Use a clear coding scheme and a second observer (inter-rater reliability).

A participant observer gets an insider's view of the group. A non-participant keeps an outsider's view, which is more objective.

A list of operationalised behavioural categories that observers tally.

Time = record at fixed intervals; event = record each time the behaviour occurs.

The observer notices and records what they expect to see. Clear categories and a second observer reduce it.

Usually only in public places where they expect to be seen, with no identification and no harm.

Watching one person, recording what they are doing, then moving on to the next person.

How closely two observers' separate records of the same behaviour agree.

If the researcher sets something up and changes it on purpose, it is a field experiment. In an observation nothing is changed.

An in-depth investigation of a single person, group or event, usually using several methods.

For rare, complex or unrepeatable cases too unusual for other methods.

Through other methods combined on the same case: interviews, observations, tests and records, often over time.

Rich, detailed, realistic data on complex cases; high ecological validity.

Findings may not generalise and they can't establish cause and effect.

A striking case can reshape a theory, even though it can't be generalised.

A case study gives depth on one case. A survey or an experiment gives breadth across many people.

No — that requires a controlled experiment.

Researcher subjectivity, and distortion when relying on memory of the past.

Measurement — many kinds of data build one rich picture.

Using two or more methods, data sources, researchers or theories to study the same thing and cross-check findings.

Method, data, researcher and theoretical.

The researcher reflecting on how their own beliefs and role may have shaped the study.

How far the findings truly reflect participants' experiences; improved by triangulation, reflexivity and credibility checks.

Using two or more methods on the same case, such as interviews, tests and observation, so each can check the others.

Generalise: apply findings to a whole population. Transfer: apply them to a similar case or context. Case studies transfer.

The case is unique, so the study cannot be repeated to check the result.

A method that measures the relationship between two variables without manipulating them.

Both variables rise together (e.g. more study, higher grades).

As one variable rises, the other falls (e.g. more screen time, less sleep).

A third variable could drive both, or the causal arrow could run the other way.

An unmeasured factor that drives both correlated variables (e.g. heat behind ice cream and drowning).

They can study variables that can't be manipulated (e.g. stress, trauma).

They can't show cause and effect; open to third-variable and reverse-causation problems.

A correlation spots a pattern; an experiment can then test whether it's causal.

Ice-cream sales and drowning rise together because hot weather raises both.

When the direction of cause and effect between two variables is unclear: A may affect B, or B may affect A.

A number from -1 to +1 giving the direction and strength of a relationship.

Dots tightly clustered along a line sloping upwards from left to right.

Two variables that are related by chance or through a third variable, not because one affects the other.

A correlation finds a pattern in real life. An experiment, where one is possible, tests whether it is causal.

r = -0.8. Strength is how close r is to 1 or -1; the sign only gives the direction.

It only describes the range that was measured. Outside that range the pattern may be different.

Gathering data by asking people about their own thoughts, feelings or behaviour.

A set of written questions, the same for everyone, given on paper, online or face to face. The two words mean the same thing.

An interview with fixed questions asked the same way each time — easy to compare.

A guided conversation started by one or two open questions. Rich data, but hard to compare and analyse.

Answering to look good rather than truthfully.

A question that pushes the respondent towards a particular answer.

It reaches thoughts and feelings that cannot be observed.

Answers may be dishonest (social desirability) or inaccurate.

Use anonymous questionnaires and neutral wording; build trust in interviews.

Measurement — it turns private experience into comparable data.

Closed = fixed answers, quantitative; open = own words, qualitative.

A rating scale from, e.g., strongly disagree to strongly agree.

Fixed questions in fixed order; prepared open questions that can change; only a topic and a time.

Natural conversation sparks ideas; dominant voices and conformity can distort answers.

Finding repeated ideas in transcripts and grouping them into themes that answer the aim.

Prepared open questions, with freedom to change the order, skip some and ask follow-ups.

The interviewer's wording, tone or reactions shaping what the person says.

The three kinds of focus-group question: to get people talking, to discuss the topic, and to check nothing was missed.

To find unclear or leading questions before the real study.

The group actually studied, chosen to represent a larger population.

The whole group the researcher wants their findings to apply to.

Everyone in the population has an equal chance of being chosen — most representative.

Using whoever is available and willing at the time. Also called convenience sampling. Quick, but often unrepresentative.

People choose to take part, for example by answering an advert. Also called volunteer sampling.

Choosing sub-groups in the same proportions as the population.

Participants recruit others — useful for hard-to-reach groups.

A biased sample limits generalisability — findings may only apply to that group.

Random/stratified = more representative; opportunity/self-selected = more practical.

Bias — an unrepresentative sample biases the findings.

Target population = everyone the study is about; sample = those who take part.

A complete list of the target population, needed for random or stratified sampling.

When the people chosen are not representative, because they self-select, the researcher picks them, or groups are excluded.

Population (other people), ecological (real-life settings), temporal (other times).

Generalisability = applying results to the population (experiments); transferability = applying findings to a similar setting (qualitative).

Samples from Western, Educated, Industrialised, Rich, Democratic societies, which may not represent most people.

Every subgroup is represented in proportion; needs detailed knowledge of the population to build the strata.

Reaches hidden groups and builds trust; participants know each other, so the sample is not representative and privacy is harder.

Everyone has an equal chance, so the sample tends to be representative; but it needs a full sampling frame, and a small random sample can still be unrepresentative by chance (sampling error). People who refuse cause non-response bias.

Probability: chance decides who takes part (random, stratified). Non-probability: it does not (opportunity, self-selected, snowball, purposive).

Choosing participants on purpose because they have the characteristics or experience the study needs.

When the people who refuse or do not reply differ from those who take part, so the sample is tilted.