3.1.2 AS research methodology

Exam analysis

Most tested topics

Research methods

An experiment deliberately manipulates an independent variable (IV), measures an operational dependent variable (DV), and compares conditions while controlling alternatives. Setting (laboratory/field) and participant design (independent, matched or repeated) are separate choices.

Type	Defining feature	Typical strength	Typical cost
Laboratory experiment	IV manipulated in an artificial/high-control setting	Standardisation, replication and internal validity	Lower mundane/ecological validity; demand characteristics; imposed-task ethics
Field experiment	IV manipulated in a natural everyday setting	Natural behaviour and lower awareness/demand	Less control, harder replication/allocation; consent/debrief/withdrawal problems
Not an experiment	No manipulated IV/control comparison	Can describe/associate rich real behaviour	Cannot claim the measured factor caused the outcome

Design	Allocation	Main advantage	Main threat and remedy
Independent measures	Different participants in each condition; random allocation if possible	No order effects	Participant differences; use random allocation, sufficient sample/control
Matched pairs	Different people matched on relevant variable(s), one per condition	Reduces selected participant differences without order effects	Matching is slow/imperfect; loss of one may lose pair
Repeated measures	Same participants complete every condition	Controls participant differences; fewer participants	Practice/fatigue/carry-over; counterbalance condition order

Novel scenario workflow: write an operational IV with at least two levels; write a measurable DV with units/scoring window; name design and allocation; hold instructions/materials/time/setting constant; add control condition/group; counterbalance if repeated; state consent, withdrawal, harm and debrief safeguards.

Evaluate in context: reliability asks whether the standardised procedure/measure could reproduce; internal validity asks whether IV rather than confounds changed DV; ecological validity asks whether task/setting represents the target behaviour; ethics asks what manipulation, deception or allocation does to participants.

Random sampling selects people from a population; random allocation assigns recruited participants to conditions. A control group receives no/alternative manipulation; a control condition can be completed by the same repeated-measures participants. Laboratory location alone does not make a study experimental.

A self-report asks participants to report their own behaviour, cognition, emotion or experience. Questionnaires and interviews can both use open and closed questions; interview structure and delivery technique must be specified separately.

Choice	Defining feature	Gain	Cost
Paper/online questionnaire	Participant completes written items	Large/cheap/anonymous; standardised	Misunderstanding, low return, no probing
Structured interview	Same questions/order, usually fixed prompts	Comparable and reliable	Limited depth; interviewer presence/bias
Unstructured interview	Flexible conversation guided by participant	Rich, unexpected detail	Low standardisation, hard replication/analysis
Semi-structured interview	Core standard questions plus planned/contingent probes	Comparability plus depth	Requires interviewer skill; probing can vary
Telephone/face-to-face	Remote voice versus co-present delivery	Reach/lower visual pressure versus rapport/non-verbal cues	Identity/rapport versus interviewer/social desirability effects

Item type	Example construction	Data/evaluation
Closed	'In the last 7 days, on how many days did you sleep ≥8 hours? 0-7'	Quantitative, fast/reliable comparison; options may force/omit answers
Likert/rating	'I felt anxious before the test: 1 strongly disagree … 5 strongly agree'	Operational score; midpoint/response-set meaning must be clear
Open	'Describe one situation in which the test changed how you felt.'	Qualitative depth/validity; coding is slower and less reliable

Write one idea per neutral item; define time/context; make response options exhaustive and mutually exclusive; avoid leading, loaded, double-barrelled, ambiguous, jargon and double-negative wording; pilot for interpretation; standardise interviewer prompts; plan coding before collection.

Quality question	Contextual check	Improvement
Reliability	Would wording/order/interviewer/coding yield similar scores?	Standard script, pilot, coding scheme, inter-rater/test-retest check
Validity	Does answer reflect target construct rather than memory/demand?	Anonymous/private response, multiple items, open probe/triangulation
Bias	Social desirability, acquiescence, recall, interviewer expectations?	Neutral wording, balanced items, confidentiality, trained interviewer
Ethics	Sensitive disclosure, privacy, storage and right to skip?	Informed consent, skip/withdraw option, safeguarding and secure anonymisation

Interview does not mean qualitative: closed structured questions produce quantitative data, while questionnaire open questions produce qualitative data. An anonymous response may reduce social desirability but cannot guarantee honesty, accurate memory or construct validity.

A case study investigates one bounded participant, family, group, institution or event in depth and context. It is a research strategy, not a single data-collection technique: interviews, questionnaires, observations, tests, records, biological measures and follow-up can be combined.

Evidence layer	Possible technique	Contribution
History/context	Records, informant interviews, timeline	Explains onset and environmental meaning
Current experience	Interview/questionnaire/diary	First-person cognition/emotion
Observable behaviour	Naturalistic/structured observation or task	Behavioural evidence beyond self-report
Objective/standard measure	Test, diagnostic scale, physiological/brain measure	Repeatable comparison or mechanism evidence
Change over time	Treatment phases and follow-up	Trajectory, maintenance and alternative explanations

Novel scenario: define the bounded case and why it is theoretically/clinically informative; collect several independent evidence sources; create a dated sequence; operationalise repeated measures; record contradictory evidence; obtain consent from the participant and relevant informants; anonymise identifying detail; avoid treating intervention response as controlled causal proof.

Strength	Limitation
Rich holistic/contextual detail gives high ecological/construct insight	Unique history and tiny unit prevent statistical population generalisation
Triangulation can test convergence across methods/data types	Researcher interpretation and retrospective memory may bias the narrative
Rare/unethical-to-create cases generate hypotheses and practical learning	No control group, random allocation or isolated IV means weak causal inference
Repeated follow-up captures change and unusual processes	Time/cost, attrition and changing measures reduce reliability
Participant voice can preserve meaning	Privacy, stigma, consent capacity and deductive identification are acute ethical risks

Case evidence can show that a phenomenon is possible, reveal a mechanism candidate and generate/refine theory. Transfer analytically by asking whether the new case shares relevant mechanisms/context—not by claiming one case represents everyone.

One participant in a laboratory experiment is not automatically a case study; intensive contextual study is required. A case can contain quantitative data and standardised tests. Depth improves understanding, not automatic validity, causality or generalisability.

Observation systematically records behaviour. Classify it on four independent axes, then specify behavioural categories, sampling and observers. Observation is a technique; it becomes part of an experiment only when an IV is manipulated.

Axis	Option A	Option B	Main trade-off
Awareness	Overt: participants know	Covert: do not know	Consent/transparency versus reactivity/demand
Observer role	Participant: joins group	Non-participant: remains separate	Insider access versus detachment/role bias
Coding	Structured: predefined categories/checklist	Unstructured: open narrative	Reliability/comparison versus depth/unexpected behaviour
Setting/control	Naturalistic: normal setting	Controlled: arranged setting/task	Ecological validity versus standardisation/control

Design step	Required operational decision
Define behaviour	Mutually exclusive, observable categories—not inferred motives (e.g. 'raises voice above conversational level for ≥2 s')
Sample behaviour	Event sampling counts every target event; time sampling records at fixed intervals
Record	Frequency, duration, latency, sequence and/or contextual field notes
Train observers	Examples/non-examples, blind coding where possible, pilot ambiguous cases
Check reliability	Two observers code same segment; compare agreement/correlation and revise categories
Protect participants	Consent where feasible, public/private expectation, debrief deception, anonymise and stop if harm occurs

Strength	Limitation and contextual remedy
Direct behaviour avoids memory/self-report bias	Cannot directly access thought/emotion; triangulate with self-report/measure
Naturalistic/covert designs can capture spontaneous behaviour	Low control and ethical problems; standardise window/context and debrief where possible
Structured categories yield quantitative comparison and replication	Category reduction may miss meaning; pilot and retain field notes
Unstructured records discover unexpected patterns	Observer bias/low reliability; reflexive notes, second coding and clear later scheme

Scenario answer sequence: state all four axes; operationalise at least two categories; choose event/time sampling; say exactly who observes, where and for how long; describe inter-observer reliability; identify reactivity/confounds; add consent/privacy/debrief safeguards.

Naturalistic does not imply covert, unstructured or participant observation. Controlled observation does not automatically manipulate an IV. High observer agreement means consistent coding, not that categories validly represent aggression, empathy or another inferred construct.

A correlation measures how two co-variables vary together; neither is manipulated. Each participant/unit contributes a paired score plotted as one point. Direction and strength describe association, not cause.

Pattern	Meaning	What it does not mean
Positive	Higher X tends to accompany higher Y (and lower with lower)	X is beneficial or causes Y
Negative	Higher X tends to accompany lower Y	Weak, harmful or no relationship
Strong	Points cluster closely around a monotonic trend; coefficient nearer ±1	Valid measurement or causality
Weak/zero	Points are dispersed/no monotonic trend; coefficient nearer 0	No non-linear relation or no subgroup pattern

Name both co-variables and give units/scoring window: for example, 'minutes of phone use from device log between 20:00-24:00 averaged over seven days' and 'total sleep minutes from actigraphy on the same nights'. 'Phone use and sleep' is not operational.

Step	Decision
Sample paired scores	Same units/people measured on both variables
Inspect scatterplot	Direction, form, outliers and possible subgroups
Select coefficient	Match scale/distribution/monotonic assumptions
Describe	State direction and strength, preferably with coefficient
Evaluate	Reliability/validity of both measures, range restriction, sampling and outliers
Infer	Predict/identify association; do not assign causal direction

Causal threat	Example for phone use ↔ sleep
Directionality	Phone use may reduce sleep, or inability to sleep may increase use
Third variable	Stress, workload or caffeine may increase use and reduce sleep
Selection/measurement	A narrow student sample or inaccurate self-report can create/distort association

Strengths: studies naturally occurring variables that cannot ethically/practically be manipulated, quantifies prediction and generates hypotheses. Limits: no causal conclusion, vulnerable to third variables/directionality, and the correlation cannot be more valid or reliable than its two operational measures.

Do not call co-variables IV and DV. A coefficient sign gives direction, while absolute size gives strength. Even a perfect association cannot by itself show which variable causes which or rule out a common cause.

A longitudinal study repeatedly measures the same participant(s), case(s) or cohort across a meaningful period to investigate change, continuity or delayed effects. It can be observational/correlational or experimental if an IV/control comparison is added.

Form	Structure	Claim boundary
Descriptive longitudinal	Same measures at several time points	Describes within-unit trajectory, not its cause
Correlational longitudinal	Earlier co-variable predicts later outcome	Establishes time order but still has third-variable/confounding threats
Longitudinal experiment	Manipulated condition/control followed over time	Stronger causal and maintenance inference if allocation/control/attrition remain sound
Pre-post only	Same unit measured before and after	Change is visible, but history, maturation, testing and regression remain alternatives

Specify target interval and measurement schedule; keep operational measures equivalent; record baseline and relevant confounds; preserve participant IDs securely; standardise contact; plan retention and missing-data rules; add comparison/control where causal inference is intended; predefine follow-up outcome and stopping/safeguarding procedures.

Strength	Why it matters
Within-person change	Separates individual trajectory from one-time age-group differences
Temporal order	Shows predictor preceded outcome, narrowing but not eliminating causal explanations
Delayed/maintenance effects	Tests whether learning, treatment or brain/behaviour change persists
Rich repeated data	Reveals turning points and individual differences hidden by group averages

Threat	Consequence	Mitigation
Attrition	Smaller sample and systematic survivor bias	Retention plan, compare dropouts, transparent missing-data analysis
Practice/testing	Repetition itself changes scores	Alternate forms, spacing, appropriate control
Historical/maturation change	Time-related events/development mimic effect	Comparison group, repeated baseline/context measures
Measure drift	New tools/raters change apparent score	Calibrate/equate methods and document changes
Cost/privacy	Long commitment and sensitive linked records	Proportionate schedule, renewed consent, secure pseudonymous linkage

Novel scenario answer: name the same participants, at least two dated waves, an unchanged operational outcome, expected change, retention method, attrition/practice/history control and ethical re-consent/confidentiality. If treatment is manipulated, also specify allocation and control condition.

Repeated measures compares conditions using the same participants; longitudinal follows the same units over a meaningful time course. A study can be both, but they are not synonyms. Time order improves causal reasoning yet does not remove third variables, history or maturation.

Methodological concepts

An aim states what a study intends to investigate. An alternative hypothesis predicts a difference/relationship; a null predicts no difference/relationship beyond chance. Every hypothesis must identify the population and operational variables.

Statement	Experiment form	Correlation form
Aim	Investigate whether condition X affects measured Y in population P	Investigate the relationship between measured X and Y in P
Directional alternative	P in X1 will score higher/lower on Y than P in X2	As X increases, Y will increase/decrease in P
Non-directional alternative	There will be a difference in Y between X1 and X2 for P	There will be a relationship between X and Y in P
Null	There will be no difference in Y between X1 and X2 for P	There will be no relationship between X and Y in P

Use a directional hypothesis only when prior theory/evidence justifies the direction before data collection. Use non-directional when an effect is expected but its direction is uncertain. Write scores/behaviour, not vague 'better' or causal wording for a correlation.

Audit: population named; IV levels or both co-variables operationalised; DV measure named; comparison/relationship word present; direction only when justified; null is exact logical counterpart.

One-tailed means one predicted direction, not one condition. The null is not 'nothing happened'; it is a population statement of no difference/association. Results support or fail to support a hypothesis—they do not rewrite it after seeing data.

The IV is deliberately changed in an experiment; the DV is the measured outcome. An operational definition states exactly how each variable is created or scored so another researcher could reproduce it.

Concept	Weak label	Operational version
IV: background sound	Music vs silence	70 dB instrumental track through headphones versus identical headphones with no audio during a 10-minute task
DV: memory	Memory score	Number of 20 nouns correctly freely recalled in 2 minutes, duplicates/intrusions excluded
Correlational co-variable	Stress	Total score on named 10-item scale completed after school

A DV can be frequency, duration, latency, accuracy/error, test/scale score, choice, physiological value or coded category. State unit, observation window, scoring rules and direction. Pilot floor/ceiling effects and ambiguity.

Operational gain	Possible cost
Precision and replicability	Narrow measure may omit the construct
Quantitative comparison	Score may reward speed/strategy rather than target ability
Standardisation	Artificial task may reduce ecological validity

Predictors/co-variables are not IVs unless manipulated. 'Aggression', 'happiness' and 'learning' are constructs, not complete DVs. Operational clarity improves replicability but does not automatically establish validity.

A control holds, removes, measures or balances a variable so conditions differ mainly in the IV. Standardisation gives every participant the same procedure. An uncontrolled variable becomes a confound when it systematically co-varies with the IV and could change the DV.

Source	Example	Feasible control
Participant	Baseline memory, age, sleep	Repeated measures, matching, random allocation, baseline measurement
Situational	Room noise, time, device, experimenter tone	Same setting/time/material/script; randomise/balance sessions
Order	Practice, fatigue, carry-over	Counterbalance, rest/washout, alternate forms
Demand/experimenter	Guessing aim, cueing responses	Blind/double-blind, cover story where ethical, standard script

For each threat: name it; explain how it differs between IV levels; explain how it could alter the operational DV; specify an exact control. A generic 'keep everything the same' earns less than a mechanism-linked control.

Controls strengthen internal validity/reliability but may make tasks artificial, restrict natural variation, increase cost or create ethical issues. Measure rather than eliminate important real-world factors when ecological validity is central.

Not every uncontrolled variable is a confound; it must vary systematically with the IV and explain the DV. Standardisation cannot remove participant differences by itself. Random allocation balances conditions; random sampling addresses population recruitment.

Quantitative versus qualitative describes data form; objective versus subjective describes dependence on personal judgement/experience. The two axes can combine in four ways.

	Objective/externally verifiable	Subjective/judgement-based
Quantitative	Reaction time, heart rate, correct-answer count	Participant's 0-8 distress rating; observer category score requiring judgement
Qualitative	Verbatim audio transcript/recorded words as raw event	Participant's interpretation or researcher's thematic account

Type	Main value	Main limit
Quantitative	Compact comparison, graphs, replicable scoring	Reduction may omit meaning/context
Qualitative	Rich explanations and unexpected themes	Slow coding, interpretation and lower inter-rater reliability
Objective	Less affected by self-presentation/interpretation	May measure proxy rather than lived construct
Subjective	Direct access to private experience	Social desirability, memory and perspective bias

Choose data to fit the question, then triangulate: e.g. combine sleep minutes from actigraphy (quantitative/objective), rating of sleep quality (quantitative/subjective) and diary account (qualitative/subjective). Agreement strengthens confidence; disagreement is evidence to explain, not delete.

Numbers are not automatically objective: a Likert score quantifies a subjective judgement. Words are not automatically subjective: an exact recorded utterance is an observable event, though coding its meaning may be subjective. No data type is universally best.

The population is the full group to which a researcher wants to generalise; the sample is the participating subset. Sampling technique controls who gets an invitation, but non-response and eligibility still shape the final sample.

Technique	Exact procedure	Strength	Bias/limit
Opportunity	Recruit available eligible people at chosen place/time	Fast, cheap, practical	Place/time/researcher-access bias; often unrepresentative
Volunteer/self-selecting	Advertise and eligible people opt in	Consent/interest, reaches dispersed people	Volunteer traits: motivation, time, topic interest, payment response
Random	Build complete sampling frame; use random generator/lottery so each has equal selection chance	Reduces researcher selection bias; potentially representative	Frame may omit people; selected non-response; costly

Novel scenario: define target population with inclusion/exclusion; identify/construct sampling frame if random; state recruitment location/channel/time; describe exact selection—not just name; anticipate who is missed/refuses; compare achieved sample demographics with population; bound generalisation accordingly.

Evidence	Generalisation consequence
Narrow age/sex/culture/occupation	Findings may not transfer where relevant mechanism differs
Large but biased online volunteer sample	Precision can increase while representativeness remains poor
Small random sample	Less selection bias but greater sampling fluctuation
Attrition/non-response	Final sample may differ from those initially selected

Random sampling recruits from a population; random allocation assigns a recruited sample to conditions. Large does not equal representative. Opportunity sampling is based on availability, not deliberate quota matching. Generalisability also depends on task/setting/time, not sample alone.

Human guideline	Requirement and safeguard
Minimise harm/maximise benefit	Risk assess, monitor distress, stop/refer/support; use least harmful effective procedure
Valid informed consent	Capacity, understandable purpose/procedure/risks/data use; guardian consent plus assent where relevant
Right to withdraw	Leave and remove data without penalty; make route/reminder practical
Lack of deception	Disclose truth unless justified/minimal and impossible otherwise; never deceive about material risk
Confidentiality	Limit access, pseudonymise, secure storage/reporting
Privacy	Observe/collect only where people reasonably expect and consent; minimise intrusion
Debriefing	Reveal aim/deception, restore understanding, answer questions, offer data withdrawal/support

Animal guideline	Applied question
Minimise harm/maximise benefit	Is scientific/welfare value proportionate to pain, distress and lasting effects?
Replacement	Can non-animal, simulation, existing data or less sentient model answer it?
Species	Is species scientifically appropriate and welfare expertise available?
Numbers	Use minimum needed for valid evidence—too few also wastes animals
Procedures	Refine handling/anaesthesia/endpoints; suitable housing/social needs; justified reward/deprivation; avoid aversive stimuli

Exam chain: name guideline → cite exact procedure/sample/data feature → explain likely harm/autonomy/welfare or benefit → judge severity/probability/reversibility → propose feasible safeguard and its methodological trade-off. For animals, use the syllabus animal guideline rather than importing human consent/right-to-withdraw labels.

Method need	Ethical tension	Design response
Avoid demand characteristics	Deception/incomplete disclosure	Minimal deception, no risk deception, prior/retrospective consent, prompt debrief
Natural public behaviour	Consent/privacy	Public-expectation audit, anonymised low-risk recording, gatekeeper/debrief where possible
Stress/emergency simulation	Harm and withdrawal	Lower intensity, screening, stop rule, support and alternative task
Animal reinforcement	Deprivation/aversive welfare	Preferred reward, minimal restriction, voluntary participation and welfare endpoints

Signed consent is not automatically valid if information, capacity or freedom is missing. Debriefing mitigates deception but cannot undo severe harm. Confidentiality concerns data identity; privacy concerns access to the person/behaviour. Ethical acceptability is a reasoned balance, not a checklist score.

Validity is the extent to which a measure/study supports the intended interpretation. Ask separately whether the construct was measured, the IV caused the DV, the behaviour represents real life and the finding transfers to the target population/context.

Validity question	Main threats	Contextual improvement
Construct/measurement	Proxy score, subjective coding, social desirability	Operational/pilot/validated measure, blind coding, triangulation
Internal/causal	Confounds, allocation/order/experimenter effects	Control/standardise, random allocation, counterbalance, blind
Ecological	Artificial task/setting, low mundane realism	More representative task/context, field evidence—while retaining controls
Population/generalisation	Narrow/biased sample and cultural/time differences	Broader/stratified/random recruitment, replication across groups

Demand characteristics arise when participants infer the aim and alter behaviour; reduce through credible neutral instructions, unobtrusive/indirect measures, blinding or justified deception/debrief. Subjectivity can add insight but requires transparent coding/checks. Objectivity reduces judgement, not necessarily proxy invalidity.

Write: identify exact evidence feature → name the interpretation threatened/supported → explain mechanism → judge consequence for this conclusion → propose improvement and trade-off. 'Laboratory means low validity' is incomplete without showing why this task differs from the target behaviour.

A field setting can contain an invalid measure; a laboratory task can validly test a narrow mechanism. Reliability is necessary for many valid measurements but consistent bias remains invalid. Generalisability extends beyond sample to setting, task, culture and time.

Reliability is consistency of measurement/procedure. Replicability is whether documentation/materials are sufficient for another researcher to repeat the study and test whether the pattern recurs.

Check	Same/different element	Use	Improve when low
Inter-rater	Two raters score same response/product	Interviews, tests, qualitative coding	Rubric, examples, training, blind double-code
Inter-observer	Two observers code same behaviour/time	Structured observation	Operational categories, observer training, video recode
Test-retest	Same measure to same people at two suitable times	Stable traits/questionnaires/tests	Clarify items, standardise conditions; avoid interval/practice extremes
Procedural replication	Independent study repeats documented method	Tests result robustness	Full script/materials/operational rules, sample and analysis transparency

Plan: standardise instructions, timing, apparatus and scoring; define categories/units; pilot ambiguity; preserve versioned materials; report recruitment/allocation/exclusions; have independent coders; choose a reliability check that targets the likely source of inconsistency.

High reliability narrows random/observer inconsistency and makes differences interpretable. Low reliability can hide or create effects. Yet a consistently wrong scale, biased item or invalid category can be highly reliable.

Two researchers repeating a whole study is replication, not inter-rater reliability. Test-retest uses the same measure/construct at two times, not two experimental conditions. Agreement does not prove validity, objectivity or truth.

Descriptive analysis organises and summarises observed data. Central tendency gives a typical/central value; spread shows variability. Cambridge requires recognition/finding/interpretation, not calculations or unspecified statistical tests.

Measure	How found	Best feature	Main weakness
Mode	Most frequent value/category	Works with nominal categories; shows common response	May be multiple/none; ignores rest
Median	Middle ordered score (average two middles if even)	Resistant to extreme/skew; ordinal suitable	Ignores distances and much data
Mean	Sum divided by number	Uses every interval/ratio score; sensitive comparison	Distorted by extremes/skew; unsuitable for categories

Measure	Meaning	Interpretation
Range	Highest minus lowest (state convention if endpoints included)	Overall span; one extreme can dominate
Standard deviation	Typical dispersion around mean	Low SD = scores clustered/consistent; high SD = dispersed—not a high/low mean

Display	Data structure	Construction/reading rule
Table	Any organised categories/conditions	Clear title, labelled rows/columns, units, no ambiguous totals
Bar chart	Separate categories/conditions	Equal-width separated bars; axes/units; height = frequency/summary
Histogram	Continuous scores grouped into adjacent intervals	Bars touch; numerical ordered x-axis; frequency on y
Scatter graph	Paired co-variable scores	One dot per pair; both axes operational variables; inspect direction/strength/outliers

Compare centre and spread together: equal means can hide different consistency; lower mean may accompany wider overlap. Describe exact pattern with units, largest/smallest, difference and variability. Do not infer cause or significance from descriptive displays.

Bar charts display discrete categories with gaps; histograms display continuous intervals with touching bars; scatter graphs do not display group frequencies. Low SD means low variation, not low scores. A mean difference alone does not prove a reliable population effect.

3.1.2 AS research methodology

Exam analysis

Most tested topics

Research methods

Experiments test causal effects by manipulating an IV and controlling comparison

Self-reports turn private experience into answers through designed questions

Case studies build intensive, triangulated evidence about one bounded unit

Observations turn behaviour into evidence through explicit setting, role and coding choices

Correlations describe the direction and strength of association between measured co-variables

Longitudinal studies follow the same units to explain change and continuity over time

Methodological concepts

Aims ask the question; hypotheses state the testable outcome pattern

Operational variables make manipulation and measurement exact and replicable

Controls reduce alternative explanations by standardising or balancing relevant variation

Data form and data source are separate: quantitative/qualitative and objective/subjective

Sampling links a target population to the people who actually provide data

Ethical research converts risk-benefit principles into human and animal safeguards

Validity asks whether the evidence supports the intended measure, cause and transfer

Reliability checks consistency; replicability makes an independent repetition possible

Descriptive analysis matches centre, spread and display to the structure of data