amazon.co.ukA-Level Psychology revision on Research Methods and Statistics: experiments, designs, sampling, ethics, descriptive statistics, distributions, hypothesis testing, all seven inferential tests, type I/II errors, peer review and features of science.
Experiments test cause and effect by manipulating an independent variable (IV) and measuring a dependent variable (DV). Both must be operationalised (defined precisely so they can be measured). Types of experiment: lab (artificial setting, high control, high internal validity but low ecological validity and demand characteristics), field (natural setting, higher ecological validity but less control and ethical issues - consent), natural (the IV occurs naturally, e.g. a disaster - high ecological validity, no manipulation, rare events, confounding variables) and quasi (the IV is pre-existing and not manipulated, e.g. gender, age - no random allocation, so causation is weaker).
Variables and controls: extraneous variables (any variable other than the IV affecting the DV - situational: noise, time of day; participant: age, intelligence) must be controlled or they become confounding variables (variables that systematically vary with the IV, making it impossible to tell which caused the DV change). Control techniques: standardisation (same procedure/materials for all), randomisation (random order of tasks/conditions), random allocation (assigning participants to conditions by chance, spreading participant variables evenly), counterbalancing (ABBA technique for order effects) and single/double-blind procedures (participants/researchers unaware of condition, reducing demand characteristics and investigator effects). Demand characteristics - cues letting participants guess the aim (Orne: please-you effects) - and investigator effects (researcher's behaviour influencing results, e.g. Rosenthal and Fode's 'clever Hans' effect) both threaten internal validity; single-blind and double-blind designs control them.
Experimental designs: independent measures (different participants in each condition - no order effects, but participant variables confound and more participants are needed; controlled by random allocation); repeated measures (same participants in all conditions - participant variables controlled, fewer participants needed, but order effects (practice, fatigue, boredom) confound - controlled by counterbalancing or leaving a gap between conditions); matched pairs (different participants matched on key variables such as age or IQ - controls participant variables without order effects, but matching is time-consuming and imperfect).
Evaluation (AO3):
Observations: watching and recording behaviour. Naturalistic (real environment, high ecological validity, low control) versus controlled (structured setting, control over variables); covert (participants unaware - no demand characteristics but ethical issues) versus overt (aware - consent obtained but reactivity); participant (researcher joins the group - rich insight but loss of objectivity) versus non-participant (researcher observes from outside - objective but less insight). Behavioural categories must be operationalised (specific, observable actions) and event sampling (counting every occurrence) or time sampling (recording at intervals) used; inter-rater reliability checks consistency between observers.
Self-report techniques: questionnaires (open questions - rich qualitative data, hard to analyse; closed questions - quantitative, easy to analyse but restricted; social desirability bias and response bias/acquiescence are key weaknesses) and interviews (structured - standardised questions, replicable; unstructured - flexible, rich data but interviewer bias and hard to replicate). Correlations: measuring the association between two co-variables (no IV manipulation, so no cause and effect - a third variable may explain the relationship); positive (both increase), negative (one increases, other decreases), zero (no relationship); presented as scattergrams with a correlation coefficient (-1 to +1). Case studies: in-depth study of one individual or small group (HM, Clive Wearing) - rich, detailed, unusual cases that cannot be generated experimentally, but small samples, researcher bias and poor generalisability.
Sampling techniques for selecting a representative sample from the target population: random (every member has an equal chance - unbiased and generalisable, but needs a full list and may still be unrepresentative by chance); stratified (subgroups in proportion to the population, randomly selected - most representative, but time-consuming); systematic (every nth person from a list - objective, but may coincide with a pattern); opportunity (whoever is available - quick and easy but biased towards the researcher's location and unrepresentative); volunteer (self-selecting - willing and engaged, but volunteer bias: more motivated, extroverted, may respond differently).
Evaluation (AO3):
Further non-experimental techniques: a pilot study is a small-scale trial run before the main study, checking the procedure works, questions are understood and timings are right - saving money and time by catching design flaws early; it also trains observers for inter-rater reliability. Content analysis systematically converts qualitative material (documents, media, transcripts) into quantitative data using coding units (categories of content counted across the material) - non-invasive and high in external validity, but the researcher's interpretation of coding units risks subjectivity; thematic analysis (identifying themes rather than counting) keeps the qualitative richness but is harder to replicate. Case study defence: unusual cases (HM, Clive Wearing) generate data experiments cannot, but generalisability is limited - the standard A* trade-off in methods essays.
Ethics (BPS Code of Ethics and Conduct): informed consent (participants must know the aim and agree; alternatives: presumptive consent - asking similar people, prior general consent, retrospective consent); deception (misleading participants - only acceptable if justified by the cost-benefit ratio and followed by debriefing); protection from harm (physical and psychological; participants must leave as they arrived); right to withdraw (during and after the study); confidentiality (no names, use numbers; the Data Protection Act); privacy (no observations in private spaces without consent).
Reliability (consistency): test-retest (same test on two occasions, high correlation โฅ 0.8 = reliable), inter-rater/inter-observer (two observers agree), split-half (two halves of a test correlate). Improved by: standardised procedures, operationalised behavioural categories, trained observers and pilot studies. Validity (accuracy): internal (did the IV cause the DV change? - threatened by demand characteristics, confounding variables, investigator effects); external (generalisable to other settings/populations/eras); ecological (real-life applicability); temporal (applicable across eras); construct (does the test measure the concept it claims to?); face (does it look valid?); concurrent (agrees with an established measure). Improved by: control groups, standardisation, single-blind procedures, natural settings and operationalisation.
Features of science (AQA 4.2.3): theory construction (hypotheses derived from theories), falsifiability (Popper - theories must be testable and capable of being disproven; psychology's theories must generate testable predictions), replicability (findings must be repeatable - the basis of reliability), objectivity (bias-free measurement, empirical methods), empirical methods (evidence from observation/experiment, not speculation) and paradigms and paradigm shifts (Kuhn - a science has a shared framework; psychology may be a 'pre-science' because it lacks a single paradigm). Peer review: research is assessed by independent experts before publication to check validity, significance and originality; it allocates research funding and maintains quality, but it is imperfect - reviewers may be biased (against unconventional findings, or in favour of their own field), anonymity can hide personal attacks, and publication bias favours positive results (the 'file drawer problem').
Psychological report structure (AQA 4.2.3): abstract (~150-word summary of aim, method, results, conclusions), introduction (literature review leading to aims and hypotheses), method (design, sample, materials/apparatus, procedure - detailed enough to replicate), results (descriptive and inferential statistics, tables/graphs; qualitative analysis if applicable), discussion (interpretation, evaluation, implications, further research), references (full APA format: Author, A. A. (Year). Title of article. Journal Name, Volume(Issue), pages.)
Descriptive statistics summarise data. Measures of central tendency: mean (sum รท number - uses all data, most sensitive, but distorted by outliers); median (middle value - unaffected by outliers, best for skewed data, but ignores some values); mode (most frequent - usable for nominal data, but may be bimodal and unrepresentative). Measures of dispersion: standard deviation (average spread around the mean - precise, uses all values, but distorted by outliers and harder to calculate) and range (highest minus lowest - quick and easy, but crude: distorted by a single outlier and ignores the distribution in between). Percentages and ratios are also used for summarising frequencies.
Distributions: a normal distribution is symmetrical and bell-shaped, with the mean, median and mode all at the midpoint (e.g. height, IQ); a positive skew has a long tail to the right (most scores at the low end - the mean is pulled above the median; use the median); a negative skew has a long tail to the left (most scores at the high end - the mean is pulled below the median). Exam questions often ask you to identify the distribution and choose the appropriate measure of central tendency.
Levels of measurement: nominal (categories/frequencies - e.g. gender, blood type; mode only; chi-square, sign test), ordinal (ranked/ordered data, unequal units - e.g. Likert scales, rankings, ratings; median, range; non-parametric tests) and interval (equal units on a scale with no true zero - e.g. seconds, temperature in ยฐC, test scores; mean, SD; parametric tests). The level of measurement determines which inferential test is appropriate - the standard first question when choosing a test.
Graphs and presentation: bar charts (categories, bars separated - nominal data), histograms (continuous data, bars touching - interval), scattergrams (correlations), line graphs (trends over time), tables (raw and summary data). The choice of graph matches the level of measurement - bar charts for nominal, scattergrams for correlations, histograms for interval distributions.
Hypotheses: the null hypothesis (H0) predicts no difference/relationship ('there will be no difference...'); the alternative/experimental hypothesis (H1) predicts a difference/relationship - directional (one-tailed) if it states which way ('music will improve recall' - only used when prior research justifies the direction) or non-directional (two-tailed) ('music will affect recall'). Hypotheses must be operationalised and testable.
Hypothesis testing and significance: the researcher sets a significance level (p โค 0.05 by convention - the probability of results occurring by chance is 5% or less), calculates the test statistic (calculated value) and compares it with the critical value from statistical tables. For most tests (t-tests, Mann-Whitney, Wilcoxon, Spearman, Pearson), results are significant if the calculated value equals or exceeds the critical value; for the sign test (and chi-square with one degree of freedom), significance requires the calculated value to be equal to or less than the critical value. The decision depends on: significance level (0.05 or 0.01), one- or two-tailed test, degrees of freedom (df = N-1 for related tests, N1+N2-2 for unrelated t-tests, (rows-1) ร (columns-1) for chi-square) and N (sample size). Conclusions must state whether the null hypothesis is rejected or retained and report p (e.g. 'p โค 0.05, so the null hypothesis is rejected - music significantly improved recall').
Type I and type II errors: a type I error is a false positive - wrongly rejecting a true null hypothesis (claiming an effect exists when it does not); it occurs when the significance level is too lenient (e.g. using p < 0.10 instead of 0.05). A type II error is a false negative - wrongly accepting a false null hypothesis (missing a real effect); it occurs when the level is too stringent (e.g. using p < 0.01 when 0.05 would have shown significance). The 0.05 level balances the two risks; exam questions give a calculated value near the critical value and ask which error could be made.
Choosing the inferential test - three questions decide which of the seven tests to use: (1) test of difference, correlation or association? (2) related (repeated measures/matched pairs) or unrelated (independent measures) design? (3) level of measurement: nominal, ordinal or interval?
| Test | Purpose | Design | Data level |
|---|---|---|---|
| Chi-square (ฯยฒ) | Association/difference | Independent (unrelated) | Nominal |
| Sign test | Difference | Related (repeated measures) | Nominal (direction of difference) |
| Mann-Whitney U | Difference | Unrelated (independent measures) | Ordinal |
| Wilcoxon (signed ranks) | Difference | Related (repeated measures/matched pairs) | Ordinal |
| Unrelated t-test | Difference | Unrelated (independent measures) | Interval (normal distribution) |
| Related t-test | Difference | Related (repeated measures/matched pairs) | Interval (normal distribution) |
| Spearman's rho | Correlation | Pairs of scores | Ordinal |
| Pearson's r | Correlation | Pairs of scores | Interval (normal distribution) |
Parametric tests (related and unrelated t, Pearson's r) assume interval data and a normally distributed population - they are more powerful (more likely to detect a real effect) when assumptions are met; non-parametric tests (chi-square, sign test, Mann-Whitney, Wilcoxon, Spearman) make no distribution assumptions and suit ordinal/nominal data. The sign test (the simplest, and the one you must be able to compute): (1) record the direction of difference (+/-) between each pair of scores (ignore zeros, reducing N); (2) count the number of less frequent sign (S = the smaller total); (3) compare S with the critical value at p โค 0.05 for one/two-tailed; (4) significant if S โค critical value.