IB Diploma Programme Psychology HL (first assessment 2021; current guide). This topic covers the CORE's "approaches to researching behaviour" in full — the quantitative methods (experiments and their designs, correlational studies, descriptive statistics) and the qualitative and mixed toolset (case studies, observations, interviews, questionnaires), with sampling, ethics and the evaluation concepts (generalizability, credibility, bias) — and then the INTERNAL ASSESSMENT, the experimental study and its report. The worked numbers are an original rehearsal experiment whose arithmetic is carried through the page.

Must-know for research methods:

The EXPERIMENT is psychology's causal instrument: manipulate the independent variable 自变量 (IV), control the rest, measure the dependent variable 因变量 (DV). LAB experiments control maximally at the cost of ecological validity; FIELD experiments trade control for realism; NATURAL experiments exploit existing differences the researcher cannot manipulate; QUASI-experiments use pre-existing groups without random allocation. DESIGNS: INDEPENDENT MEASURES (one group per condition — no order effects, but participant variables between groups; random ALLOCATION balances them), REPEATED MEASURES (everyone does both conditions — no participant variables, but order effects, managed by counterbalancing) and MATCHED PAIRS (a compromise costing matching time). Demand characteristics 需求特征 — participants guessing the study's aim — are attacked with standardised instructions, single-blind designs and cover stories that stay within ethics.

SAMPLING: Random sampling 随机抽样 gives everyone an equal chance (representative in principle, hard in practice); OPPORTUNITY sampling takes who is available (cheap, biased toward the accessible); VOLUNTEER (self-selected) samples are cooperative but unrepresentative — the self-selection bias. Sampling chooses WHO you test; ALLOCATION assigns them to conditions — random sampling is not random allocation, and confusing the two is the classic IA error. Descriptive statistics 描述统计: the MEAN uses every score but is dragged by outliers; the MEDIAN resists them; the MODE fits categories. Dispersion: the RANGE is quick but rests on two scores; the STANDARD DEVIATION uses them all. CORRELATION measures association without manipulation — the correlational study 相关研究 gives a coefficient $r$ from $-1$ through 0 to $+1$, and never a direction of cause: sleep and grades correlate; so may the third factor behind both.

Worked example (the rehearsal experiment). Does semantic processing beat acoustic processing for word recall? Repeated-measures design, counterbalanced (AB/BA), 10 participants, standardised word lists. Group means: acoustic 6.0 words, semantic 8.7 words. The mean difference (2.7 words) is DESCRIPTIVE — it says what happened in this sample, not whether the difference exceeds chance; variability (one acoustic participant recalled 11, above the semantic median) shows why ranges and SDs travel with means. Only an INFERENTIAL test (HL) settles significance.

CASE STUDIES examine one person or event in depth (H.M., Gage) — richness against generalizability. OBSERVATIONS may be NATURALISTIC or CONTROLLED, OVERT or COVERT, PARTICIPANT or NON-PARTICIPANT; they need BEHAVIOURAL CATEGORIES and inter-rater reliability 评分者间信度 (two observers agreeing, checked on a coefficient). INTERVIEWS run structured (fixed questions, comparable), semi-structured (pursue the interesting answer) or as FOCUS GROUPS (interaction itself is data — and a source of conformity). QUESTIONNAIRES scale cheaply but suffer social-desirability bias and item ambiguity. TRIANGULATION — combining methods, data sources or researchers — is the qualitative answer to single-lens error.

ETHICS (and the IA is marked on it): Informed consent 知情同意, deception only when unavoidable with full DEBRIEF, protection from harm, the RIGHT TO WITHDRAW at any time, CONFIDENTIALITY, and for animals the 3Rs. The evaluation concepts: GENERALIZABILITY (do findings travel beyond this sample and setting?), CREDIBILITY (do the findings reflect the participants' reality — in qualitative work: credibility, transferability, dependability, confirmability?) and BIAS (researcher expectancy, participant reactivity, sampling bias). SIGNIFICANCE (HL): a result is significant at $p < 0.05$ if it would occur by chance less than 5% of the time; the sign test 符号检验 on direction-of-change counts illustrates the logic — with $N = 8$ participants and 2 negative signs, $S = 2$ exceeds the critical value, so the change is NOT significant at the 5% level.

The IA (SL and HL) is a SIMPLE EXPERIMENT the student designs, runs and reports on a small sample — original data, ethically conducted. The report's structure mirrors the method. INTRODUCTION: background theory (one or two studies), the rationale, and an operationalised AIM. EXPLORATION: the design (which of the three, and why), variables identified and operationalised, CONTROLS (standardised instructions and materials, counterbalancing, environment), the sampling method and sample with its limits, ETHICS addressed point by point (consent forms in the appendix), and materials. ANALYSIS: descriptive statistics with correctly labelled graphs; at HL an INFERENTIAL TEST chosen to match the design (sign test, Mann–Whitney U, Wilcoxon) with the statistical decision stated against a 0.05 level. EVALUATION: strengths, LIMITATIONS tied to the design and sample, and modifications that follow from them. APPENDICES: raw data, sign-up sheets, standardised instructions 标准化指示, consent forms. The marking criterion rewards a study whose design choices are EXPLAINED, not just listed — every choice carries a because.

Worked example (the report's spine). Aim: to investigate the effect of processing level (shallow/acoustic vs deep/semantic) on word recall in a repeated-measures experiment. Design choice: repeated measures removes participant variables; counterbalancing (half do A then B, half B then A) removes order effects. Controls: same word list lengths, same 60-second recall window, written standardised instructions read aloud. Sample: 10 volunteers from one school year (opportunity sampling — limited generalizability, stated). Ethics: informed consent from participants, right to withdraw, no deception, anonymised data. Analysis: means 6.0 vs 8.7 with ranges and a bar chart of means; HL adds the Wilcoxon test decision. Evaluation: volunteer bias and single-year sample limit transfer; a replication across year groups and a larger sample are the named modifications.

• The four experiment types with their control–realism trade-offs; the three designs with order effects and counterbalancing (quantitative).
• IV and DV operationalised; demand characteristics and the standardised countermeasures (quantitative).
• Random, opportunity and volunteer sampling with allocation distinguished from sampling (quantitative).
• Mean, median, mode, range, SD — what each is good and bad at; correlation's $r$ and its no-causality rule (quantitative).
• Case studies, observations with inter-rater reliability, interviews and focus groups, questionnaires, triangulation (qualitative).
• The six ethical principles and the 3Rs (ethics).
• Generalizability, credibility and the bias family; HL significance at $p < 0.05$ and the sign-test logic (evaluation).
• The IA report's five sections and the spine: aim → design → controls → ethics → statistics → evaluation with modifications (IA).

Boundary notes: Paper 3 (HL) examines exactly this topic on unseen research scenarios — methodological and ethical analysis; Paper 1 Section A may ask for method definitions. The IA's marking criteria reward explained choices and correctly computed statistics; all worked numbers are original to the rehearsal experiment.