Edexcel心理学 研究方法 实验设计 统计检验
研究方法这一块在 Edexcel 的 9PS0 里不是背了名词就能拿分的部分，Paper 3 的 Section A 直接用一份陌生的研究做背景，让你当场判断设计、抽样、效度与统计检验。下面十二节按卷面顺序排开，从 24 分怎么分配，到抽样与实验设计，再到测量水平和四类推断检验的选用，每一节都写成能在答案里落笔的句子。文中的分值、卷面结构与统计口径来自 Pearson 官方 spec（9PS0，Issue 4，2026 年 3 月）与近年 9PS0/03 真题。
Research methods in Edexcel 9PS0 is less about naming techniques than applying them inside an unfamiliar study; Section A of Paper 3 hands you a scenario and asks you to judge its design, sampling, validity and statistical testing. The twelve sections below follow the order of the paper, from the mark split through sampling and design to levels of measurement and the choice of inferential test.
Section A 的 24 分：研究情境题的卷面结构
Paper 3 Psychological Skills 考试 2 小时、满分 80 分、占 A-level 的 30%，Section A 的名字就是 Research Methods，值 24 分。June 2022 那套 9PS0/03 把 24 分拆成 16 分与 8 分两道题，两道题共用一份陌生研究做背景。小问从「一种设计的一个优点与一个缺点」这类 4 分题，一路排到「看图判断偏态类型」这种 1 分题。同一卷的 Section B 是 24 分的研究评价，其中一道 16 分题要评价两项经典研究；Section C 的议题与争论占 32 分，由 12 分情境题与 20 分论述题组成。卷首一并印出标准差、Spearman's rho、chi-squared 与 Mann-Whitney U 的公式和临界值表，计算器允许使用。Pearson 的评分方案在多个小问后面重复同一句：答案要与情境相关，Generic answers score 0 marks，阅卷时丢分最多的正是只堆方法名词却没换回情境变量的答案。
Paper 3 Psychological Skills runs for two hours and carries 80 marks, 30% of the A-level, with Section A titled Research Methods and worth 24 marks; the June 2022 9PS0/03 paper split that into a 16-mark and an 8-mark question built on one unfamiliar study. The rest of the paper is a 24-mark review of studies with one 16-mark evaluation, plus an issues and debates section worth 32 marks as a 12-mark scenario answer and a 20-mark essay. Formulae and critical value tables for standard deviation, Spearman's rho, chi-squared and Mann-Whitney U are printed at the front, and Pearson repeats after most parts that generic answers score zero.
假设与变量：IV、DV 与操作化
实验里被操纵的变量是自变量，被测量的是因变量，答案里只写这两个缩写不给分。要把概念落成情境里的可测指标，这一步叫操作化，比如自变量是提示说谎或说真话的颜色，因变量是反应时间与错误率。方向明确时写方向性假设，方向不确定时写非方向性假设，两者都要写成一句能被统计检验检验的完整预测。零假设写的是变量之间没有效应的版本，它是推断检验的对象，结果出来后要么拒绝、要么保留。很多学生以为假设里可以写「记忆会变得更好」这样笼统的一句，评分方案只认能落到数字上的预测。错得最多的是把零假设写成「实验不会成功」，那样写出来的句子没有统计含义。
The independent variable is the one manipulated and the dependent variable is the one measured, but the scenario's own wording earns credit where the abbreviations do not; operationalisation turns a concept such as lying ability into measurable indices such as error rate and response time. A directional hypothesis predicts the direction of an effect and a non-directional one predicts only that an effect exists, while the null hypothesis states that there is no effect and is the statement an inferential test evaluates. Many students write a vague prediction that memory will improve, when the mark scheme rewards a prediction carrying numbers.
抽样五法：从随机到分层
随机抽样要求总体里每个人机会均等，抽签与随机数生成器是常见做法，前提是先有一份完整名单。机会抽样就近找得到谁就问谁，省时间但样本偏差最大，结论很难推广到总体。志愿抽样靠被试自己回应征集，愿意参加的人往往比总体更外向、更有空，June 2022 那套题里的约 1000 名被试就是自愿到科学中心参观的人。系统抽样按固定间隔从名单取人，起点随机才谈得上抽样误差可控。分层抽样先按总体特征分层，再按各层在总体里的比例取人，Pearson 的评分方案把「按比例」单列为一个给分点。评价抽样方式时写清样本偏向哪一类人，比写「不够随机」有用得多。
Random sampling gives every member of the population an equal chance through a draw or a random number generator, and needs a full list to work at all; opportunity sampling takes whoever is available, which saves time but biases the sample and limits generalisation. Volunteer sampling recruits those who respond to an advert, and they tend to be more extraverted and more available than the population they represent, as in the June 2022 study of roughly a thousand science-centre visitors. Systematic sampling takes every nth name from a random start, while stratified sampling selects in proportion to each stratum's share, a step the mark scheme awards separately.
三种实验设计与顺序效应
独立测量把不同的人分到不同条件，练习与疲劳效应的影响小，代价是需要的人数翻倍。重复测量让同一批人做完所有条件，人数省一半，顺序效应却会把成绩往某个方向推。平衡法把被试分成两半，一半先做 A 再做 B、另一半反过来，答案里写清这个安排才拿得到分。配对设计按关键特征把两个被试配成一组，个体差异被控制住，配对标准不可靠时控制就落空。控制组提供比较用的基线，随机分派把个体差异摊平到各条件。Edexcel 只考实验室实验与田野实验两类，自然实验不在课纲里，Pearson 的 Getting Started 指南在 2016 版写明过这条改动。
Independent measures assigns different participants to each condition, which keeps practice and fatigue effects out at the cost of needing twice as many participants; repeated measures uses the same people throughout and saves that number while risking order effects. Counterbalancing splits participants so half meet the conditions in one order and half in the reverse, and matched pairs places people with similar characteristics one per condition, controlling individual differences only as far as the criteria are reliable. Edexcel covers laboratory and field experiments only, with natural experiments outside the specification.
观察法：类别清单与观察者信度
自然观察在行为本来的场所记录，生态效度高、控制最少，情景难重现也就难重做。受控观察在实验室布置情景，变量受控但场所本身会改变行为。参与观察让研究者进入群体，隐蔽参与减少反应性偏差，却带来欺骗与退出权的问题。观察前要先做一份行为类别清单，类别互斥且覆盖全部目标行为，否则观察者当场临时判断，记录就不稳定。观察者间信度用两位独立观察者记录的一致程度衡量，一致性低往往不是被试有问题，而是类别定义太模糊。阅卷时看到只写「自然观察很真实」却没有类别清单与信度计算的答案，2 分题也只给 1 分。
Naturalistic observation records behaviour where it happens, supporting ecological validity with little control, while controlled observation sets up the situation in a laboratory and gains control while the setting itself changes behaviour. Participant observation places the researcher inside the group, and covert participation reduces reactivity at the price of deception and the right to withdraw. Observers need mutually exclusive and exhaustive categories, otherwise judgements are made on the spot, and inter-rater reliability measures how far two independent observers agree.
自我报告：问卷、访谈与试点研究
问卷的闭合题给出固定选项，数据能直接统计，选项之外的回答却进不来。开放题信息量大，编码要花工夫，两位编码者解释不一致还会拉低信度。访谈按结构分成结构化、半结构化和非结构化三类，结构化等于把问卷念出来，非结构化更接近一次谈话，考纲要求说清每一类适合什么目的。社会赞许性偏差让被试往「应该如此」的方向作答，匿名化与说明没有对错能减轻它。试点研究先用少量被试跑一遍材料，检查题目有没有被理解成另一个意思、时间够不够用，这一步在 4 分题里常被单独考。很多学生以为题目数量够多问卷就可靠，重复施测的信度要靠同一批人两次作答成绩的相关来算。
Closed questions produce data that can be counted directly but cannot capture answers outside the options, while open questions collect richer material that costs time to code and loses reliability when coders disagree. Interviews divide into structured, semi-structured and unstructured forms, with the structured version behaving like a spoken questionnaire; social desirability bias pushes responses towards what participants think is expected, and anonymity plus a statement that there are no right answers reduces it. A pilot study runs the materials with a few participants to check that questions are understood as intended and that timing works, and retest reliability depends on correlating the same people's scores across two sittings.
相关研究：co-variables 与 Spearman's rho
相关研究测量两个变量一起变化的程度，两个变量都只是被测量，官方叫它们 co-variables。相关系数的取值落在 -1 到 +1 之间，正负号说明方向，绝对值说明强度。Spearman's rho 适用于两个变量都是序数或以上、研究者要找关系的情形，9PS0 卷首给出公式与各样本量的临界值表。判断显著性在这里是「等于或大于」，June 2022 那道题给出观察值 110，要考生结合 5% 双尾水平与临界值表解释结论。相关系数不能推断因果，冰淇淋销量与溺水人数同向上升，真正推动两者的是气温，这类第三变量解释是评价题的标准答案。很多学生以为 r 等于 0.5 表示「有一半的相关」，它是强度指标而不是比例。
Correlational research measures how far two variables change together, and because neither is manipulated they are called co-variables; the sign gives direction and the size gives strength. Spearman's rho suits ordinal data or better when a relationship is the target, and the 9PS0 paper prints its formula and critical values, so the work lies in ranking the data, finding the differences and reading the table. A coefficient never licenses a causal claim, since ice cream sales and drowning figures rise together mainly because of temperature, which is the third-variable explanation evaluation answers are expected to give.
效度与信度：分开算的两套指标
内部效度问研究内部能不能把结果归给自变量，混淆变量与需求特征都会把它拉低。外部效度问结果能不能离开样本与场所，生态效度是其中关于场所的那一条。内容效度看测量工具是否覆盖要测的整体；并存效度把新测量与已有测量做对比，构念效度问测量是不是真在测那个心理构念。信度分外部与内部两条线，测试重测信度用同一批人前后两次成绩的相关，评分者间信度用于主观判断的场合，折半信度把一份测验劈成两半。标准化同时服务两个指标，程序脚本一致才谈得上重测，指导语与计分口径一致才谈得上两位评分者对得上。学生最常把效度与信度混着写，一句「这个研究很可靠」想把两种分一起拿走，评分方案里这两类分是分开给的。
Internal validity asks whether results can be attributed to the independent variable, and confounding variables or demand characteristics lower it; external validity asks whether findings survive beyond that sample and setting, with ecological validity covering the setting. Content validity concerns whether a measure covers the whole construct, concurrent validity compares a new measure with an established one, and construct validity asks whether it taps the intended concept. Reliability divides into test-retest, inter-rater and split-half routes, and standardised procedures serve both properties because identical scripts make retesting meaningful and identical marking rules make two raters comparable.
伦理：把原则写成情境里的动作
Pearson 的 spec 要求考生掌握 BPS 伦理准则里的基本要求，知情同意在答案里要落到研究目的、参与者要做什么、可能的伤害与随时退出的权利。涉及欺骗的研究要在结束后做事后解释，说明真实目的、说明当时为什么不能透露，并补上知情同意。欺骗只有在没有其他办法、伤害可控、事后能完整解释时才被接受。保护不受伤害覆盖心理与生理两类风险，保密既指数据存放也指报告里不出现可识别的身份信息。研究人群是儿童、认知障碍者或囚犯时，同意要由监护人给出，本人理解研究之后还要给出同意。很多学生以为伦理题列出名词就能拿分，2 分题给的是「在这种情境里怎么做」这个动作。
Pearson's specification expects candidates to work with the core requirements of the BPS ethical guidelines, and informed consent has to name the purpose of the research, what participation involves, possible harm and the right to withdraw. Studies involving deception require a debrief that reveals the true aim and restores consent, with deception acceptable only where no alternative exists and harm is controlled; protection from harm covers stress and embarrassment as well as physical risk, and confidentiality extends to how data are stored. Where participants are children, people with cognitive impairment or prisoners, a guardian gives consent and the participants themselves give assent.
测量水平与描述统计：能算什么由数据决定
测量水平决定一组数据能做什么运算：名义数据只是分类，序数数据能排序但间隔不等，等距数据间隔相等。集中趋势里均值容易被极端值拉动，中位数对极端值稳健，众数是唯一能用在名义数据上的指标。离散程度用全距与标准差，9PS0 卷首给出的样本标准差公式分母是 n 减 1。条形图画类别频次，直方图画连续数据的分布，柱间不留空隙这个细节常单独考 1 分。正偏说明尾部在右侧、多数分数挤在低端，均值被高端的少量极端值拉大，中位数更能代表典型水平。June 2022 的 9PS0/03 用一张直方图考了 1 分的偏态判断，绝大多数参与者过去 24 小时说谎 0 到 1 次，分布因此是正偏。
The level of measurement decides which operations the data support: nominal data are categories, ordinal data can be ranked although intervals are unequal, and interval data carry equal intervals. Among averages the mean uses every score and is dragged by extremes, the median resists them, and the mode is the only option for nominal data, while range and standard deviation describe spread and the formula printed in 9PS0 divides by n minus 1. Bar charts display category frequencies and histograms display continuous distributions; in a positively skewed set most scores sit at the low end while a few high scores pull the mean upwards.
推断统计：四类检验的选用与显著性口径
选检验先问三个问题：找差异还是找关系，两组数据来自同一批人还是不同的人，数据落在哪个测量水平。找差异且是不同的人、数据为序数或等距时用 Mann-Whitney U，找差异且是同一批人时用 Wilcoxon，找关系用 Spearman's rho，数据只是名义类别时用 chi-squared，自由度按行数减 1 乘列数减 1 算。比较口径的方向要记牢，Mann-Whitney U 与 Wilcoxon 的计算值等于或小于临界值才算显著，Spearman's rho 与 chi-squared 要等于或大于。显著性水平常用 p≤0.05，Edexcel 的表还给出 p≤0.10 与 p≤0.01 两档，方向性假设对应单尾检验。第一类错误是把为真的零假设拒绝掉，第二类错误是零假设为假却没有拒绝它，把水平从 0.05 放宽到 0.10 会让第一类错误变多。June 2024 的 9PS0/03 评分方案里有一道 Wilcoxon 计算题，差值列与正负秩列做完之后 T 等于 4。
Choosing a test comes down to three questions: whether the study looks for a difference or a relationship, whether the scores come from the same people, and what level of measurement the data sit at. Mann-Whitney U tests a difference between independent groups on ordinal or interval data, Wilcoxon handles repeated measures, Spearman's rho tests a relationship, and chi-squared tests a difference in nominal data with degrees of freedom as rows minus one times columns minus one. Mann-Whitney U and Wilcoxon are significant when the calculated value is equal to or below the critical value, whereas Spearman's rho and chi-squared need a value equal to or above it; a Type I error rejects a true null and a Type II error fails to reject a false one.
质性分析与个案研究：编码、主题与 HM
主题分析把访谈或观察记录反复读、逐句编码、归类，最后提炼成几个有原始语料支撑的主题。编码表稳不稳定决定了分析能不能重复，报告里通常要给出主题名与一句代表性语料。扎根理论把编码与取样交替推进，一边分析一边决定下一位该找什么样的参与者，理论从数据里长出来。个案研究适合罕见或极端情形，HM 的脑损伤无法复制，他在短时与长时记忆上的分离表现给出两者独立的证据，代价是难以推广、也难重做。课纲要求考生完成一次实践调查，用观察法收集量与质两类数据，量化部分用 chi-squared，质性部分用主题分析。错得最多的是把个案研究的价值写成「很细」，评价题要的是细在哪一个环节以及这个代价是什么。
Thematic analysis reads interview or observation records repeatedly, codes them line by line and groups the codes into themes supported by the original material, so the stability of the coding frame decides whether the analysis can be repeated. Grounded theory alternates coding with sampling so that theory emerges from data rather than being tested against it, and the 9PS0 specification lists it alongside thematic analysis as a qualitative method. Case studies suit rare cases, and HM's damage cannot be replicated, so his dissociation between short-term and long-term memory supports their independence while the single case limits generalisation; the specification also expects one practical investigation analysed with chi-squared and thematic analysis.
TI-Nspire CX II CAS 图形计算器 AP/IB/A-Level¥1320📱 去京东购买京东精选
需要一对一辅导？Need One-on-One Tutoring?
A-Level / GCSE / IB / KS2-KS3 全科辅导，研究方法与统计部分可以单独约课
16621398022 同微信
aleveler.com 心理学复习资料
或截图后用「扫一扫 → 相册」选择本图