6.2.1, 6.2.2, 6.2.3, 6.2.4, 6.2.5 - Methods in criminological research

Experiments and case studies in context

A method should match the question: whether wording causes a change in recall requires a different design from understanding one offender's history. In a laboratory experiment, a researcher might randomly allocate viewers of the same harmless film to differently worded questions and count accurate details. The manipulated wording is the independent variable; the operationalised recall score is the dependent variable. Keeping the film, delay and question order constant helps isolate the wording effect. The film's known content provides a reference for accuracy.

A field experiment manipulates a variable in an everyday setting. For example, consenting volunteers could observe a planned, harmless lost-property interaction in a campus café and then receive different neutral interview procedures. Compared with a laboratory, incidental attention and environmental distractions may be more realistic. Noise, viewing distance and interruptions are harder to control. If different groups also sit at different distances, distance is a confound rather than merely background variation. A real event observed without manipulation is not automatically a field experiment.

A case study investigates an individual or small bounded case in depth, using sources such as interviews, records and observations. A forensic formulation might relate a person's learning history, triggers and protective factors to offending. Multiple sources permit triangulation: agreement can strengthen credibility, while disagreement needs explaining. Case evidence can generate hypotheses and reveal complexity. It rarely isolates one causal factor and cannot establish that all offenders share that history. Confidential records require lawful permission and secure handling, not copying into a classroom exercise.

Selecting participants without overstating representativeness

Sampling determines who can support a conclusion about eyewitnesses or potential jurors. A random sample gives each member of an identified target population an equal selection chance, for example randomly selecting names from a complete volunteer-eligible list. A complete sampling frame and responses from selected people are needed; non-response can still distort the achieved sample. Random sampling is different from randomly allocating already recruited participants to conditions.

In stratified sampling, the researcher divides the population into relevant categories and samples in proportion to their population sizes. If an eligible population of 200 contains 120 younger and 80 older adults, a proportional sample of 20 contains 12 and 8 respectively, ideally randomly chosen within each category. This represents those age proportions; it does not guarantee representation of every other characteristic.

A volunteer sample responds to an advertisement. Volunteers for a memory study may be unusually interested or confident. An opportunity sample uses accessible people, such as students available after a class. It is practical but may omit people with different ages, occupations or experiences. Calling either sample random because the researcher did not know who would agree confuses unpredictability with equal selection probability. State the actual recruitment, target population and resulting limits, rather than claiming that a large accessible sample represents all jurors.

Reliability validity objectivity and credibility

Research quality involves several different questions, not a single label of validity. Reliability concerns consistency: two coders should score the same recalled detail in the same way. A clear scoring key and independent coding of a subset can test inter-rater reliability. Repeating an interview is a poor simple test-retest solution if repetition itself changes the memory.

Internal validity asks whether a recall difference is attributable to the manipulated interview rather than unequal delays or different viewing conditions. Construct validity asks whether a score represents the intended concept: number of details alone measures amount, not necessarily accuracy. Record errors as well as correct details. Ecological validity concerns whether the task resembles the relevant real activity; population validity concerns generalising beyond the sample. Improving realism can reduce experimental control, so quality requires explaining the trade-off.

Objectivity means decisions are not driven by the researcher's preferences. Predetermined detail categories and a coder unaware of the interview condition reduce expectancy effects. They do not make ambiguous descriptions disappear; the key must state accepted equivalents. Credibility concerns whether an interpretation is believable and well supported by the evidence. In qualitative work, retaining the context of quotations, an audit trail of coding decisions and consideration of contradictory accounts strengthen credibility. Triangulation is useful but not infallible: several sources may repeat the same initial error.

Describing recall and combining research

Quantitative analysis starts with a defined unit: a correct distinct detail, an error, or a categorical verdict. For illustrative recall scores 2, 4, 4 and 6, the mean is 4, the median is 4, the mode is 4 and the range is 4. A frequency table gives one score of 2, two of 4 and one of 6. Unlike the range, which uses only extremes, standard deviation uses deviations across all observations.

The deviations from 4 are −2, 0, 0 and 2; their squares total 8. Thus $s=\sqrt{8/3}\approx1.63$ details. A smaller spread means more similar scores, not automatically more accurate scores. Means and spread for each condition should be considered alongside errors and the design, not used as proof that an interview caused improvement.

A correlation describes how paired measures vary together. A study might relate delay in hours to recall accuracy; a negative association means longer delays tend to accompany lower accuracy. It does not establish causation because other factors, such as event difficulty, may differ. A scatter plot should place delay on one axis and accuracy on the other, with one point per participant; joining these independent people into a time series would misrepresent the data.

Meta-analysis combines comparable numerical findings from several studies, usually through effect sizes rather than averaging raw scores with different scales. A publicity meta-analysis can examine whether effects differ between student and potential-juror samples. It increases the evidence base, but results depend on inclusion rules, study quality, publication bias and differences between procedures. An overall effect does not imply that every included study found the same effect.