Practical investigation: Suitable examples — Edexcel A-Level Psychology

The effect being tested comes from the minimal group experiments, in which Tajfel sorted schoolboys by something trivial, a stated preference between two painters or the toss of a coin, and then had them allocate points to anonymous members of each group. They favoured their own group, and many chose the option that widened the gap between the groups even when a different choice would have given their own group more. A classroom version keeps that logic: create an arbitrary group, then measure allocation, liking or trait ratings for an in-group and an out-group target. Because the groups are meaningless, no real competition, history or contact can explain a difference, which is exactly why the finding supports social identity theory rather than realistic conflict theory.

This design puts the working memory model on trial. Baddeley and Hitch argued in 1974 that short-term memory is not one store but a central executive directing a phonological loop and a visuo-spatial sketchpad, so two jobs drawing on the same subsystem should collide while two jobs drawing on different ones should not. Baddeley, Grant, Wight and Thomson found that concurrent visual tracking wrecked performance on an imagery based mnemonic but barely touched rote verbal learning, which is exactly the pattern predicted. In a classroom version the independent variable is usually a matched against a mismatched secondary task, the dependent variable is items correctly recalled or tracking errors, and the order of conditions is counterbalanced when the same participants do both.

Baddeley built this study in 1966 by handing participants lists that either sounded alike, such as cat, cab, can and cap, or sounded quite different, and then lists that shared meaning or did not. Immediate serial recall collapsed for the similar sounding list while meaning made almost no difference; when recall was delayed by twenty minutes of an interfering task the pattern reversed and the lists that shared meaning became the hard ones. Conrad had already shown that letters shown silently on a screen were misremembered as letters that rhyme with them. The conclusion is that the short-term store codes sound while the long-term store codes meaning. Your own version manipulates similarity and counts items recalled in the right position, with lists matched for length, frequency and familiarity.

The two co-variables are measured, never manipulated: age is taken in whole years and attitude is a total score on a set of rating items where a high total means a more permissive view. Because a total from a rating scale is ordinal and the question is about relationship rather than difference, the rank order test is the right one, and a negative coefficient would mean that as age rises permissiveness falls. The content is socially sensitive, so participants are asked what they think about drug use rather than what they have taken, scripts carry no names, the topic is disclosed before consent is given, and anyone under sixteen needs parental consent as well as their own agreement. The design also confounds age with generation, which belongs in the discussion.

This pairing puts an objective physical measure beside a subjective rating, and the contrast is the teaching point. Height can be measured in centimetres to the same standard for everyone, while a self-rating of aggressive tendencies depends on insight and honesty, so any weak coefficient may say more about the measure than about people. Using a published instrument such as the Buss and Perry Aggression Questionnaire, which sums items on physical aggression, verbal aggression, anger and hostility, improves that measure without curing it. The theoretical rationale for expecting any link is thin, which is why a non-directional hypothesis is defensible, and a coefficient close to zero means the null hypothesis is retained rather than that the study went wrong.

The learning topic's practical must be an observation, so this one runs at the roadside as a covert, non-participant, structured observation of naturally occurring driving. Speed is operationalised either by a radar reading or by timing each vehicle across a measured gap between two fixed markers and converting, while driver sex and an agreed age band are recorded on a tally sheet using categories written before collection begins. Two observers record the same vehicles independently and their totals are correlated to establish inter-observer reliability. The behaviour belongs in this topic because speed choice is shaped by reinforcement, arriving sooner and peer approval, and by models watched from the passenger seat in childhood.

This second observation works for the learning topic because helping and courtesy are exactly the behaviours that differential reinforcement and modelling are said to build. A field setting is chosen, a shop doorway or a corridor, and the categories are written first: door held open, dropped item retrieved, thanks offered, directions given. Each passer-by is tallied once into a table of sex against helped or did not help, which yields frequencies in separate categories and points straight at chi-squared. Piliavin, Rodin and Piliavin's subway field experiment reported that almost every first helper was male, while Eagly and Crowley's meta-analysis found men helped more in public, risky or chivalrous episodes and women more in caring, sustained ones.

Here the independent variable is the period a source comes from, so the design lives or dies on matching: the same publication, the same kind of item, comparable length, ideally the same season, so that only time differs. Language is the obvious coding target because terms drift. Nouns such as lunatic and the use of a diagnosis as an adjective fall away while recovery, wellbeing and support rise, and the manuals record the same drift, with homosexuality removed from the American classification in 1973 and hysteria vanishing altogether. Evaluations of the Time to Change campaign in England reported small measurable improvements in reported public attitudes across the years it ran, which gives you a benchmark for your own tallies. Watch for one news event inflating a single period's totals.

The comparison here is between outlets rather than eras, so the sampling frame is what you must defend: two source types matched for date and length, with a stated rule for which items counted. Philo and the Glasgow Media Group's analysis of British media found that the largest single category of coverage linked mental illness with violence, and their audience interviews found people citing media as the source of their beliefs, sometimes above their own experience. Useful categories are violence words, sympathy or recovery words, whether an expert or a person with lived experience is quoted, and whether a diagnosis is used as an adjective for a person. Remember that reporting guidelines, particularly those on suicide, mean some differences reflect editorial rules rather than attitude.

A workable version shows participants a short film of an incident, interviews one group with the cognitive techniques and the other with a fixed set of standard questions, and counts correct details against a checklist prepared from the film beforehand. Geiselman and colleagues, in the study that introduced the technique, found students interviewed with context reinstatement, report everything, reverse order and change of perspective recalled more correct items than those given a standard police style interview, and the meta analysis by Kohnken and colleagues put the gain in correct detail at roughly a third while also recording a rise in incorrect detail. That trade off is the thing your discussion should test, so score errors and invented details as well as hits.

This design produces words, and the marks come from turning them into numbers defensibly. Participants watch a fictional trial or crime story and are asked, in an interview or a questionnaire, why the accused person acted as he did, and the answers are coded by content analysis into categories fixed before coding begins, for example biological, learned and situational explanations. Those frequencies can then be tested with chi squared, which is exactly why the coding frame must be agreed in advance and checked by a second coder. The psychology behind it is attribution: Ross described the fundamental attribution error, the habit of explaining another person's conduct by disposition while underweighting the situation, and the blame attribution inventory of Gudjonsson and Singh separates external attribution, mental element and guilt feeling in offenders' own accounts.

This example is a correlation, not an experiment, so nothing in it can show that early bonds cause later relationship quality. The usual design is a structured interview built from rating scale items, one set about remembered closeness to a caregiver and one about a current or recent adult relationship, each giving every participant an ordinal total, which is why Spearman’s rho is the test. The psychology behind the prediction is Bowlby’s internal working model and the continuity hypothesis, supported by Hazan and Shaver, whose newspaper love quiz found adults recalling secure childhood bonds described trusting and longer lasting romances while avoidant adults reported fearing closeness. Retrospective report is the weakness to name.

The framing is the problem with this one, because asking only about the good parts invites agreement and leaves you no way of knowing what a parent would have said about the difficulties. The workable version is a semi structured interview of open questions, transcribed and then content analysed: agree coding categories in advance such as mixing with other children, warmth of staff, routine and parental peace of mind, tally how many parents mention each, and the result is frequency data in categories, which is chi squared territory. Have a second person code a sample of transcripts so inter rater reliability can be reported, and make consent, anonymity and the right to withdraw explicit, since parents of very young children can feel judged.

Coding media is a way of gathering quantitative data without going anywhere near a vulnerable participant, which is exactly why it suits a topic about drugs. The sampling frame is settled first, perhaps the songs in one week of the chart or an evening of broadcasting on one channel, and the coding categories are fixed before viewing, so that a mention of alcohol, tobacco, cannabis or another substance is counted the same way every time and tallied against its category. Two coders working the same material give inter-rater reliability, either as percentage agreement or as a correlation between their totals. Primack and colleagues did this with the most popular songs of a year and found roughly a third carried a substance reference, with rap far ahead of pop, which is a useful benchmark for a classroom version.

A practical of this shape turns printed text into numbers, so the marks sit in the coding frame rather than in the reading. Categories are agreed before coding starts, for example a health warning, a glamorising mention, a mention of cost or of the law, and each is tallied for drinking and for smoking across a matched sample of titles and dates. Two coders working the same articles give a reliability check, either as percentage agreement or as a correlation between their tallies. Counts in named categories are nominal data from independent sources, so chi squared tests the association, and the write up has to justify that choice, state the hypothesis and its direction, and say what the sample of papers allows anyone to claim.