Scientific processes — AQA A-Level Psychology

A prediction is directional, or one-tailed, when it says which way the results will fall: participants recall more words in silence than with music. It is non-directional, or two-tailed, when it predicts a difference or a relationship but not which group does better. The choice is not free. Where previous research points consistently one way, a direction is justified; where findings are mixed, contradictory, or the area is new, leave the direction out, because a one-tailed prediction that goes the wrong way cannot be supported however large the effect. Alongside sits the null hypothesis, which states there is no difference beyond chance, and it is the null a statistical test allows you to reject at the 0.05 level. Either type must name both conditions and an operationalised DV.

The target population is the whole group findings are meant to describe; the sample is the handful actually tested, and each technique trades representativeness against effort. Random sampling gives everyone an equal chance, usually by numbering a complete list and drawing lots or using a random number generator, so it is unbiased, but it needs that full list and can still produce an odd sample by chance. Systematic takes every nth name from a sampling frame. Stratified identifies subgroups and samples each in proportion to the population, making it the most representative. Opportunity takes whoever happens to be available and volunteer relies on people answering an advert, as Milgram's newspaper advert did: both are quick, both attract an atypical group, and generalisation suffers.

A trial run is a small-scale version of the whole procedure, carried out on a few participants before the real investigation, and its point is to find problems while they are still cheap to fix. It checks that instructions are understood, that the task is neither so easy that everyone is at ceiling nor so hard that nobody scores, that the timings work, and that the manipulation of the IV actually produces a change. With a questionnaire it exposes ambiguous, leading or intrusive items; in an observation it tests whether the behavioural categories can be used consistently by two observers. It also saves money and is an ethical gain, because fewer people are put through a flawed design. The data collected are not reported as results and the hypothesis is not tested on them.

With repeated measures the same people take part in every condition, so participant variables are held constant and half as many people are needed. The price is order effects such as practice, fatigue and boredom, which counterbalancing addresses, and a higher risk of demand characteristics. In an independent groups design a different set of people does each condition: no order effects, but individual differences now sit between the groups and could produce the result on their own, so allocation is randomised and twice as many participants are recruited. A matched pairs design tests different people who have been paired on variables relevant to the DV, such as age or IQ, splitting each pair across conditions, which takes time and is never perfect outside identical twins.

Behavioural categories turn a vague construct into a checklist an observer can tick without interpreting anything: aggression becomes pushing, kicking, biting and shouting, with every category observable, clearly defined and not overlapping. Event sampling counts each occurrence of the target behaviour across the whole session, which suits behaviour that happens rarely but is hard to keep up with when several things happen at once. Time sampling records only what is happening at fixed intervals, for example every 15 seconds, as Ainsworth did in the Strange Situation: it makes the workload manageable, but the sampled moments may misrepresent everything in between. Two observers coding the same session allow inter-observer reliability to be calculated and agreement checked.

Closed questions fix the response options, such as yes or no, a rating or a Likert scale, and produce numbers that are quick to analyse and easy to compare, but they force people into a box and say nothing about why. Open questions invite a free answer, which is richer and less researcher-led, though the replies then need qualitative analysis or coding through content analysis. Well-built items are short, unambiguous and free of jargon, leading wording, double-barrelled questions and emotive language, and fillers can disguise the aim. Interviews sit on the same axis: a structured interview follows a fixed schedule in a fixed order, so it is replicable but cannot follow anything up; an unstructured interview is flexible and detailed but vulnerable to interviewer bias; a semi-structured one does both.

The IV is what the researcher changes, the DV what is measured, and both must be operationalised, which means written as the exact procedure used: memory cannot be measured, but the number of words correctly recalled from a list of 20 within two minutes can. An extraneous variable is any other factor that could affect the DV, such as noise, time of day, or differences between participants; these are nuisances, and standardisation and randomisation keep them quiet. A variable becomes confounding only when it varies systematically with the IV, so its effect cannot be separated from the IV's: if everyone in the silent condition was tested first thing and everyone in the noisy condition after lunch, the study cannot say which produced the result.

These four control techniques are distinct. Random allocation decides by chance which condition each person enters, spreading individual differences evenly across independent groups. Counterbalancing tackles order effects in a repeated measures design: half do condition A then B, half do B then A. This shares practice or fatigue equally between conditions rather than removing them. Randomisation uses chance in designing materials, such as presenting a word list in a different order for every participant, preventing researcher bias. Standardisation makes the procedure and instructions identical for everyone, stopping the procedure acting as an extraneous variable and allowing replication.

Participants are not passive. Cues in a study, from the consent form to the task itself, let them guess what is expected and act on the guess; these cues are demand characteristics. Some help by producing the result they think is wanted, others deliberately do the opposite. Orne showed participants working at pointless tasks simply because the setting demanded it. The researcher creates investigator effects through leading questions, tone, appearance and expectancy effects, changing the DV without the IV moving. A single blind hides the condition and aim from the participant to reduce demand characteristics, while a double blind hides them from the researcher too.

The British Psychological Society's Code of Ethics rests on four principles: respect, competence, responsibility and integrity. From these stem key ethical issues in study design: informed consent, deception, protection from harm, privacy, confidentiality, and the right to withdraw. Dealing with ethical issues requires specific remedies. Consent is gained via a form detailing the study, or through presumptive consent if full disclosure would ruin it. Deception and distress are addressed by a full debrief, offering to withdraw data, and counselling. Confidentiality uses numbers instead of names. An ethics committee weighs likely benefits against costs before approving research.

Before a study appears in a journal, or a grant is awarded, other experts in the same field read it anonymously and recommend acceptance, revision or rejection. They judge originality and importance, whether the method and the statistics really support the conclusions, and whether anything has been overstated. Three functions are worth naming: allocating research funding, validating the quality of what gets published, and improving the work through suggested amendments. The evaluation is the more valuable half. Anonymity can shelter a reviewer settling a rivalry; journals prefer positive, novel results, so replications and null findings go unpublished, which is the file drawer problem; and established reviewers may suppress work that threatens the theory their own careers rest on, slowing any paradigm shift.

The economy here means the money and working time a country gains or loses, so a creditworthy answer runs from a finding, through a change in behaviour or policy, to a cost. Attachment research showing that fathers can be primary caregivers underpins shared parental leave, which keeps both parents in the workforce rather than one leaving it. Absence caused by mental illness costs employers billions a year, so evidence that CBT and drug therapy shorten episodes of depression returns people to work sooner and reduces sick pay. Work on circadian rhythms explains why errors cluster on night shifts, as at Chernobyl, and lets employers build rotas that avoid expensive accidents. Weak answers describe the psychology in detail and never arrive at the money, which is where the marks are.

Consistency is the whole idea: measure the same thing twice and the answer should come out the same. The test-retest method gives the same questionnaire, test or interview to the same people after a gap long enough that they cannot recall their previous answers but short enough that nothing has genuinely changed, then correlates the two sets of scores; a coefficient of 0.80 or above is the accepted threshold. Agreement between observers is checked by having two people watch the same behaviour using the same categories and correlating their tallies against that same figure. Improvement is practical work: operationalise categories so nothing is left to judgement, train observers together beforehand, rewrite ambiguous or complex items, and keep the interviewer and the schedule constant.

Validity asks whether a measure captures what it claims and whether findings are legitimate inside the study and beyond it. Face validity is the eyeball test, that a measure looks as though it taps the construct, sometimes judged by an expert. Concurrent validity is checked by giving the new measure and a well-established one to the same people and correlating the results, again looking for 0.80 or above. Ecological validity concerns generalising to other settings and everyday life, and depends on the mundane realism of the task rather than simply on whether the room was a laboratory. Temporal validity asks whether a finding still holds: Asch's conformity work is called a child of its time, whereas Burger's replication of Milgram in 2009 found obedience close to the original.

Objectivity means keeping expectations out of data using standardised, blind designs. The empirical method insists claims rest on direct observation. Replicability requires methods detailed enough to repeat, establishing reliability. Popper's falsifiability means a theory earns scientific status by making predictions that could be proven false. Theory construction involves gathering empirical evidence to build a framework explaining phenomena. This enables hypothesis testing, generating specific, testable predictions to see if the theory holds or needs modification. Kuhn added the paradigm, the shared assumptions of a discipline; psychology's lack of a single one makes it a pre-science, and accumulated anomalies force a paradigm shift.

A report follows a fixed order because readers use it in a fixed way. The abstract is a short summary of the aim, hypothesis, method, results and conclusion, read first to decide whether the paper is worth reading at all. The introduction reviews relevant theory and studies, narrowing from the broad area to this particular aim and prediction. The method is written so that someone else could repeat the study exactly: design, participants and sampling, apparatus, procedure including the standardised instructions, and ethics. Results give descriptive statistics first, then the inferential test, its calculated and critical values, the significance level and the decision on the null hypothesis. The discussion interprets the findings, admits limitations and suggests further work; references credit every source.