Edexcel心理 研究方法 变量 抽样 统计检验
Edexcel 的 A-Level 心理学（9PS0）把研究方法集中在 Paper 3，卷面代号 9PS0/03。这一卷两小时、80 分，占整个资格 30%，Section A 的 24 分全部是研究方法题。很多学生以为研究方法只是背名词，真到卷面上却要在陌生情境里判断该用哪种检验、该写哪一种假设。考纲把 Methods 与 Synoptic review of studies、Issues and debates 并列放在 Topic 9，说明它是整门课的方法底座。本文按一项研究的实际搭建顺序梳理：从目标与假设起步，经过变量、设计、抽样，走到统计与结论。每一节都给出可以照着写的句式，而不是只列术语。
In the Pearson Edexcel A-Level Psychology specification, research methods sit in Paper 3, coded 9PS0/03, which runs for two hours and carries 80 marks, 30% of the qualification, with Section A alone worth 24 marks of methods questions. The methods content is not a vocabulary list to memorise but a set of decisions to justify against unfamiliar stimulus material, so the useful revision question is always why a particular design, sample or test fits the scenario. The sections below follow the order in which a study is actually built, from aims and hypotheses through variables, design and sampling to the statistics and the final conclusion.
目标、方向性假设与零假设
研究的起点是目标，用一句话说清要检验什么关系。假设把目标变成可测的陈述句，并且必须能被数据推翻。方向性假设写出差异或相关的方向，例如睡足七小时的学生回忆成绩更高。非方向性假设只写存在差异，不预设哪一边更大。很多学生以为假设写得越笼统越安全，其实阅卷要的正是可检验的具体表述。零假设写的是没有差异或没有相关，它是统计检验真正比较的对象。常见的错误是把零假设当成客套话，答题时干脆不写。真正决定后面选哪种检验的，是假设里有没有方向、变量属于什么测量水平。
A study opens with an aim, one sentence naming the relationship to be examined, and the hypothesis turns that aim into a testable statement that data could disprove. A directional hypothesis states which group or direction is expected to be higher, while a non-directional one predicts only that a difference or correlation exists. The null hypothesis claims there is no difference and no correlation, and it is the statement the statistical test actually evaluates. Students often write a vague hypothesis because it feels safe, yet marks come from operational clarity rather than caution.
自变量、因变量与操作化定义
自变量是被操纵或被比较的那一项，因变量是被测量的结果。操作化是把抽象概念变成可观察的指标，例如把记忆成绩定义成二十个词里正确回忆的个数。这一步最容易丢分，因为学生常把定义写成同义反复，把焦虑解释成感到焦虑。额外变量会干扰结果但不随自变量系统变化，混杂变量与自变量一起变动，因此能提供另一套解释。需求特征让参与者猜出研究目的并改变行为，研究者效应则来自施测者的期待。标准化流程与随机分配是控制这两类问题的常用手段。
The independent variable is what the researcher manipulates or compares, and the dependent variable is what is measured as the outcome. Operationalisation converts an abstract construct into an observable measure, such as defining memory performance as the number of words correctly recalled out of twenty. Extraneous variables add noise without moving with the independent variable, whereas confounding variables change alongside it and offer a rival explanation for the result. Demand characteristics and investigator effects both distort behaviour through expectation, and standardised procedures with random allocation are the usual defences.
独立组、重复测量与配对设计
独立组设计让不同参与者分别接受一种条件，同一人不会做两次。重复测量设计让同一批人做完所有条件，参与者差异被消掉，顺序效应随之而来。配对设计先按关键特征配对再分到两组，试图兼顾两者。抵消平衡把条件顺序的一半对一半倒过来，用来压住练习与疲劳。很多学生以为重复测量一定更好，其实题目一旦涉及学习或记忆任务，顺序效应会让结论变脆。判断设计时先看参与者是否重复，再看有没有配对过程。
An independent measures design gives each condition to a different group of participants, so nobody completes the task twice. A repeated measures design uses the same people in every condition, which removes participant variables while introducing order effects such as practice and fatigue. Matched pairs pairs participants on a relevant characteristic before splitting them across conditions. Counterbalancing reverses the order for half the sample, and the choice between designs depends on how susceptible the task is to practice.
随机、机会、志愿、系统与分层抽样
随机抽样给总体中每个人相等的入选机会，理论上的代表性最好。机会抽样取手边最容易找到的人，速度快但样本容易偏向某一类学生。志愿抽样由参与者自己报名，得到的人往往更愿意配合。系统抽样按固定间隔从名单里取人。分层抽样先按性别或年级分组，再按比例各取一部分。样本偏倚不会因为人数变多而自动消失，扩大规模只缩小随机误差。评估抽样时要说清偏的是哪一类人，以及这个偏倚怎样影响结论的推广。
Random sampling gives every member of the target population an equal chance of selection, which is the strongest theoretical claim to representativeness. Opportunity sampling takes whoever is available, so it is quick but tends to over-represent one kind of student. Volunteer sampling relies on self-selection and therefore attracts cooperative participants, while systematic sampling takes every nth person from a list. Stratified sampling divides the population into groups and samples each in proportion, and a biased sample stays biased however large it grows.
实验室、现场、自然与准实验
实验室实验在受控环境中操纵自变量，因果推断最干净，生态效度却常被质疑。现场实验把操纵搬到真实场景，行为更自然，控制力随之下降。自然实验利用本来就会发生的变化，例如政策调整或灾害，研究者不操纵自变量。准实验比较现成的组别，例如两个班或两类诊断，参与者没有被随机分配。考纲把这几类放在同一节，要的是能在情境里辨出操纵与分配这两条线索。判断时问两句：自变量是谁定的，参与者是怎么进组的。
A laboratory experiment manipulates the independent variable in a controlled setting, which supports causal claims but invites criticism over ecological validity. A field experiment moves the manipulation into a natural setting, gaining realism at the cost of control. A natural experiment exploits a change the researcher did not create, such as a policy shift, and a quasi-experiment compares existing groups without random allocation. Two questions separate them in the exam: who set the independent variable, and how did participants end up in their groups.
行为类别与观察者间信度
观察法记录行为本身而不去问感受，适合不便打扰的场景或难以自我报告的对象。自然观察不干预环境，控制观察事先安排好情境。参与观察者进入群体，非参与观察者留在外面记录。公开观察取得知情同意，隐蔽观察不告知，伦理与真实性在这里互相拉扯。行为类别必须先定义清楚，否则两个观察者记下的东西无法合并。观察者间信度靠同时记录再比对，一致率低就回去改类别定义。时间抽样与事件抽样决定什么时候记，漏掉这一点会让记录变成随手笔记。
Observation records behaviour directly instead of asking people to report on it, which suits settings where interference would ruin the data or participants who cannot self-report. Naturalistic observation leaves the setting untouched, whereas a controlled observation stages the situation. Participant observers join the group and non-participant observers stay outside it, and covert observation trades informed consent for realism. Behavioural categories must be defined before recording begins, and inter-rater reliability is checked by comparing two observers who score the same session.
问卷与访谈中的社会赞许性
问卷用封闭题拿到可比的数据，用开放题拿到意料之外的回答。访谈从结构化走到半结构化再到无结构，深度增加，可比性下降。社会赞许性让人往体面的方向答，匿名与中性措辞能压低这种偏差。访谈者效应来自语气、期待与追问方式，同一份提纲由不同人执行会得到不同结果。很多学生以为问卷比实验省事，其实题项措辞、题目顺序与选项设计每一步都能制造偏差。评估自我报告时，先问参与者有没有动机、有没有能力准确回答。
Questionnaires use closed items for comparable data and open items for unexpected answers, and interviews run from structured through semi-structured to unstructured as depth rises and comparability falls. Social desirability pushes respondents towards the respectable answer, so anonymity and neutral wording reduce it. Interviewer effects arise from tone, expectation and probing style, which means one schedule delivered by two researchers can produce two different sets of data. Before judging a self-report measure, ask whether participants had both the motive and the ability to answer accurately.
相关系数、个案研究与内容分析
相关研究测量两个变量一起变化的程度，系数从负一到正一，符号表示方向，绝对值表示强度。相关不等于因果，因为反向因果与第三变量都能给出同样的散点图。很多学生以为负相关就是没有关系，负号只说明一边升高时另一边下降。个案研究深入描述一个人或一个小群体，常用来研究罕见现象，样本只有一人时推广要格外小心。内容分析把文本或影像转成类别计数，主题分析则归纳反复出现的模式。选方法时看研究问题要的是关系、差异，还是过程描述。
Correlation measures how far two variables move together, with a coefficient between minus one and plus one where the sign gives direction and the size gives strength. Correlation is not causation, because reverse causation and a third variable can produce the same scattergram. Case studies describe one person or a small group in depth and are useful for rare phenomena, though generalising from a single case is risky. Content analysis turns text or images into category counts, while thematic analysis groups recurring patterns into themes.
分半信度、重测信度与内外效度
信度是测量的一致程度，重复施测得到相近结果才算稳定。分半信度把量表拆成两半再比较得分，重测信度隔一段时间再测一次。观察者间信度要求两名观察者对同一段记录给出一致判断。效度问的是有没有测到想测的东西，内部效度关心因果推断是否站得住，外部效度关心结果能否推到别的人与场景。信度高是效度的前提，信度低时效度一定低。改良办法很具体：把题项写清楚、把流程标准化、把评分标准固定下来。
Reliability is consistency, shown when repeated measurement produces a similar result, and it can be checked by split-half comparison, by a test-retest interval or by agreement between two observers. Validity asks whether the measure captures the construct it claims to capture, with internal validity concerning causal inference and external validity concerning generalisation to other people and settings. High reliability is a precondition rather than a guarantee, since a stable measure can be stably wrong. The practical fixes are clear item wording, standardised procedure and a fixed scoring scheme.
知情同意、欺骗与事后说明
知情同意要求参与者在了解目的与流程后自愿加入，保密与匿名是两件不同的事。欺骗只在必要时使用，事后必须完整说明。保护参与者免受身心伤害，退出权贯穿整个过程而不是只在开头说一次。涉及未成年人或临床群体时，同意程序更严格，通常还要监护人同意。考纲把伦理判断放在情境题里考，答案要落到具体做法上，例如事后说明的形式、数据保存的方式、撤回申请的渠道。写伦理段时不要停在名词上，要说清这项研究里风险是什么、怎么降低。
Informed consent means participants join voluntarily after understanding the purpose and procedure, and confidentiality differs from anonymity because anonymised data cannot be traced back to a person. Deception is defensible only when necessary and must be followed by a full debrief, while protection from harm and the right to withdraw apply throughout the study rather than at its start. Work with minors or clinical groups usually adds guardian consent. Ethics answers score on concrete practice, so name the safeguard that the scenario would actually need.
平均数、中位数与众数的选择
平均数用上每一个数据，对极端值敏感。中位数只看中间位置，遇到偏态分布更稳。众数是出现最多的值，适合类别数据。全距只取最大减最小，容易被一两个离群值拉大。标准差描述数据围绕平均数的分散程度，比全距更充分利用信息。选哪个统计量取决于数据水平与分布形状，序数数据通常用中位数。画图之前先看清是频次、比例还是分布，图选错了后面的解释会跟着偏。
The mean uses every score and is therefore sensitive to extreme values, while the median relies on position and holds up better with skewed data. The mode is the most frequent value and suits nominal categories, and the range simply subtracts the lowest score from the highest. Standard deviation summarises spread around the mean and uses more of the data than the range does. Choosing between them depends on the level of measurement and the shape of the distribution.
卡方、斯皮尔曼、曼-惠特尼与威尔科克森
推断统计回答的是差异或相关有没有超出随机波动的范围。Edexcel 考纲列了四种检验：卡方用于名义数据与独立组，斯皮尔曼等级相关用于序数数据的相关，曼-惠特尼 U 用于序数数据与独立组的差异，威尔科克森用于序数数据与重复测量的差异。选检验要看三件事：数据是名义还是序数、要相关还是要差异、是独立组还是重复测量。算出观察值后与 Appendix 4 的临界值比较，观察值大于或等于临界值才拒绝零假设。显著性水平常用 0.05，单尾与双尾要看清用哪一列。水平放得太松会增加第一类错误，把没有的差异判成有；放得太紧则增加第二类错误。
Inferential statistics decide whether an observed difference or correlation is larger than chance would produce. The Pearson specification lists four tests: chi-squared for nominal data with independent groups, Spearman rank correlation for ordinal data, Mann-Whitney U for a difference between independent groups and Wilcoxon for repeated measures on ordinal data. Selection depends on the level of measurement, on whether the question asks about a difference or a relationship, and on whether the design is independent or repeated. The observed value is compared with the critical value printed in Appendix 4, and the null hypothesis is rejected when the observed value reaches or exceeds it.
情境题的判断句与改进写法
卷面题常给一段陌生研究，然后问该用什么设计、哪里可以改进、结论能不能下。答题时先定位题目问的是设计、抽样、统计还是伦理，再给一句判断加一句理由。改进类问题要写成具体做法，例如把机会抽样换成随机抽样，并说明这样能减少哪一类偏倚。统计题先报检验名称，再报观察值与临界值的比较，最后写出是否拒绝零假设。很多学生以为把术语堆满就能拿分，阅卷看的是判断与理由能否对上情境。练题时把每道题的判断句单独抄下来，考前只看这些句子最省时间。
Paper 3 questions usually present an unfamiliar study and ask which design was used, what could be improved, or whether the conclusion is justified. Locate what the question is really testing, then give one judgement followed by one reason tied to the scenario. Improvement answers need a concrete change plus the bias it removes, and statistics answers should name the test, compare the observed value with the critical value and state the decision about the null hypothesis.
牛津高阶英汉双解词典 第10版 商务印书馆¥168📱 去京东购买京东精选
A-Level / GCSE / IB / KS2-KS3 全科辅导
16621398022 同微信
Visit aleveler.com更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图