The methods, and what each is good for

• Laboratory experiment — high control, causal inference, replicable; low ecological validity, demand characteristics. (Baddeley, Loftus and Palmer.)
• Field experiment — real setting, higher ecological validity; less control, consent problems.
• Observation — real behaviour, useful with children; observer bias, no causal inference. (Bandura, the Strange Situation.)
• Questionnaire / interview — large samples, or rich depth; social desirability and researcher effects.
• Correlation — measures a relationship between two co-variables; never establishes cause and effect.
• Case study — rich, in-depth, ideal for rare cases (HM); not generalisable, researcher bias.
• Others: twin and adoption studies, animal experiments, brain scanning (CAT, PET, fMRI), content analysis, longitudinal and cross-sectional designs, cross-cultural research and meta-analysis.

Sampling, designs and hypotheses

• Sampling: random (everyone has an equal chance — unbiased but hard), stratified (proportional subgroups — representative but time-consuming), volunteer (easy — but volunteers differ from non-volunteers), opportunity (quick — but unrepresentative).
• Designs: independent groups (no order effects; participant variables — control by random allocation), repeated measures (no participant variables; order effects — control by counterbalancing), matched pairs (best of both; time-consuming and never a perfect match).
• Hypotheses: the alternate/experimental hypothesis predicts the difference or relationship; the null predicts none. Directional (one-tailed) if past research points one way; non-directional (two-tailed) if it does not, or the evidence is mixed. Always operationalise.
• Control issues: extraneous vs confounding variables, order effects, experimenter effects, demand characteristics, social desirability, situational and participant variables.

Choosing a design

Describing the data

• Central tendency: the mean uses every score (but is distorted by outliers and needs interval data); the median is unaffected by extreme scores (good for skewed or ordinal data); the mode is the only option for nominal data.
• Dispersion: the range (highest minus lowest) is quick but is determined entirely by two scores; the standard deviation uses every score and tells you how far, on average, scores lie from the mean. A large SD means the data are spread out, so the mean represents the group less well.
• Graphs: bar chart for categories, histogram for continuous data, scatter diagram for correlations. Frequency tables summarise counts.
• Distributions: in a normal distribution the mean, median and mode coincide. In a skewed distribution they separate, and the median is usually the better summary.

Calculate the mean

Calculate the range

Choose the test

Tap a scenario, then tap the correct inferential test.

Nominal, ordinal and interval data

You cannot choose an inferential test until you know the level of measurement.

• Nominal — data in categories, counted as frequencies (e.g. how many participants obeyed vs disobeyed). No order.
• Ordinal — data that can be ranked, but the intervals between ranks are not equal (e.g. a 1-5 rating scale, or positions in a race).
• Interval — data on a scale with equal intervals and a standardised unit (e.g. time in seconds, number of words recalled, scores from a standardised test).

Rule of thumb: most self-report rating scales are treated as ordinal; counts of people in categories are nominal; scores measured with a standardised unit are interval.

Level of measurement

Choosing the right test

Ask three questions: (1) am I testing a difference or a relationship? (2) is the design independent or related? (3) what is the level of measurement?

• Chi-squared — difference/association, independent groups, nominal data.
• Mann-Whitney U — difference, independent groups, ordinal data or above.
• Wilcoxon signed ranks — difference, related (repeated measures or matched pairs), ordinal data or above.
• Spearman's rho — relationship (correlation), ordinal data or above.

Critical values — get the direction right. For Mann-Whitney U and Wilcoxon, the observed value must be equal to or LESS than the critical value to be significant. For chi-squared and Spearman's rho, the observed value must be equal to or GREATER than the critical value. Always check N (or degrees of freedom), the level of significance, and whether the test is one- or two-tailed.

Degrees of freedom

Convert to a percentage

Match the term to its meaning

Tap a term on the left, then its correct definition on the right.

Significance, p values and errors

• p ≤ 0.05 is the standard level in psychology: there is a 5% or smaller probability that a result this extreme occurred by chance. Only if the result is significant do we reject the null hypothesis.
• p ≤ 0.01 is a more stringent level, used when a false positive would be costly (e.g. testing a drug).
• Type I error — a false positive: we claim an effect that is not really there. More likely if the significance level is lenient (p ≤ 0.10).
• Type II error — a false negative: we miss a real effect. More likely if the level is too stringent (p ≤ 0.01).

Sense-check your data: before you trust a test, look at the descriptives. If the means barely differ but the test is 'significant', check for an error; if there is a huge difference but no significance, check the N — small samples make Type II errors likely.

Which error?

Reports, peer review and ethics

• Report conventions: abstract (a summary), introduction (previous research narrowing to the aim and hypotheses), method (design, participants, materials, procedure — detailed enough to replicate), results (descriptive then inferential), discussion (interpretation, limitations, implications, future research).
• Peer review: before publication, other experts assess the quality, originality and validity of the work. It allocates funding, protects the integrity of the discipline and filters out poor research — but it can be slow, reviewers may be biased against novel or contradictory findings, and true anonymity is hard.
• Human ethics — BPS Code of Ethics and Conduct (2009): respect (informed consent, right to withdraw, confidentiality), competence, responsibility (protection from harm, debriefing) and integrity (avoiding deception). Carry out a risk assessment.
• Animal ethics: the Scientific Procedures Act (1986) and Home Office regulation — licences, cost-benefit review and the 3Rs (replacement, reduction, refinement).

Validity

Which test?

How Topic 9 methods are assessed

Paper 3: Psychological skills is worth 30%. It has three sections: research methods, a synoptic review of the classic studies, and issues and debates.

• The formulae and statistical tables are given to you in the paper, and calculators are allowed — so marks come from choosing and interpreting the right test, not from memorising formulae.
• Expect unseen data-response: read the scenario, identify the design and level of measurement, choose and justify the test, compare observed with critical value, and state a conclusion in terms of the hypothesis.
• Expect to evaluate a described study for validity, reliability, generalisability, objectivity and credibility.

Always write the conclusion properly: "The observed value of U (12) is less than the critical value (23) for N = 10, p ≤ 0.05, two-tailed, so the null hypothesis is rejected." That single sentence pattern earns marks every time.

The big ideas to know

This is the methods half of Topic 9 — assessed in Paper 3, Psychological skills. Press Finish to see your score.