📚 Edexcel GCSE Statistics: Practical Investigation Key Points | Edexcel GCSE统计: 实验与实践考核要点
Mastering practical investigation skills is essential for success in Edexcel GCSE Statistics. The specification emphasises the ability to design, carry out and critically evaluate statistical enquiries. Whether you are planning an experiment or analysing survey data, understanding the underlying principles will help you score highly in both exam papers and any coursework-style tasks. This article covers the key concepts you need, from the Statistical Enquiry Cycle to experimental design, sampling, analysis and evaluation.
掌握实践调查技能对于在 Edexcel GCSE 统计考试中取得好成绩至关重要。考试规范强调设计、实施并批判性评估统计调查的能力。无论你是在规划一项实验还是在分析调查数据，理解其基本原则都能帮助你在试卷和课程作业类任务中获得高分。本文涵盖你需要掌握的关键概念，从统计探究循环到实验设计、抽样、分析与评估。
1. The Statistical Enquiry Cycle (PPDAC) | 统计探究循环 (PPDAC)
The Statistical Enquiry Cycle, often abbreviated as PPDAC, structures a statistical investigation into five stages: Problem, Plan, Data, Analysis and Conclusion. In the exam you may be asked to name the stages, place a given activity at the correct stage, or explain why the cycle is iterative rather than linear.
统计探究循环通常缩写为 PPDAC，将统计调查分为五个阶段：问题（Problem）、计划（Plan）、数据（Data）、分析（Analysis）和结论（Conclusion）。考试中你可能会被要求说出这些阶段、把给定活动归入正确阶段，或解释为什么这个循环是迭代的而非线性的。
Problem: Clearly define the research question and the population of interest. A well-worded question should be specific, measurable and achievable. For example, 'Do students who eat breakfast perform better in morning tests?' is better than 'Is breakfast important?'.
问题： 明确定义研究问题和感兴趣的目标总体。措辞得当的问题应当具体、可测量且可实现。例如，"吃早餐的学生在上午测试中表现更好吗？"就优于"早餐重要吗？"。
Plan: Decide on the sampling method, sample size, data-collection instrument and timescale, and state a hypothesis before any data are gathered.
计划： 确定抽样方法、样本量、数据收集工具和时间安排，并在收集任何数据之前先写出假设。
Data: Collect the data using the planned method, record raw values accurately, and note any missing values or unusual readings.
数据： 按照既定的方法收集数据，准确记录原始数值，并标注任何缺失值或异常读数。
Analysis: Calculate summary statistics, draw appropriate charts, and compare distributions to look for patterns, trends and relationships.
分析： 计算汇总统计量，绘制合适的图表，并比较各分布以寻找规律、趋势和关系。
Conclusion: Return to the original question, answer it using the evidence, and state how far the conclusion can be trusted.
结论： 回到最初的问题，用证据作出回答，并说明该结论的可信程度。
| Stage | Typical exam activity | 阶段 | 常见考查活动 |
| Problem | Write a hypothesis or define the population | 问题 | 写假设或定义总体 |
| Plan | Choose a sampling method and justify it | 计划 | 选择抽样方法并说明理由 |
| Data | Criticise a questionnaire or a recording sheet | 数据 | 评价问卷或记录表 |
| Analysis | Calculate statistics and draw a comparison | 分析 | 计算统计量并作图比较 |
| Conclusion | Evaluate the validity and reliability of findings | 结论 | 评估结果的有效性与可靠性 |
2. Defining the Problem, Population and Hypothesis | 定义问题、总体与假设
Every good investigation starts with a precise question. Vague aims such as 'find out whether students are happier' cannot be measured, so they cannot be tested statistically. A usable question names the variable, the group being studied and the comparison being made.
每一项好的调查都始于一个精确的问题。诸如"调查学生是否更快乐"这类模糊目标无法测量，因此无法进行统计检验。可用的研究问题会指明变量、被研究的群体以及所比较的对象。
The population is the whole group you are interested in; the sampling frame is the list you actually sample from; the sample is the subset you measure. If the sampling frame does not cover the whole population, coverage bias appears before a single value is recorded.
总体是你感兴趣的完整群体；抽样框是你实际抽取样本所依据的名单；样本是你实际测量的子集。如果抽样框无法覆盖整个总体，那么在记录第一个数值之前就已经产生了覆盖偏倚。
A hypothesis should be a testable statement, not a question. 'The mean reaction time of students who drink water is lower than that of students who do not' can be supported or contradicted by data; 'water is good for you' cannot.
假设应当是可供检验的陈述，而不是问句。"喝水学生的平均反应时间低于不喝水学生"可以被数据支持或反驳；而"喝水有益健康"则不能。
3. Planning and Sampling Methods | 计划与抽样方法
The sampling method determines how much you can trust the conclusion. Edexcel expects you to choose a method, describe how to carry it out, and state one advantage and one disadvantage.
抽样方法决定了结论的可信程度。Edexcel 要求你选择一种方法、描述其操作步骤，并说出一个优点和一个缺点。
Simple random sampling: every member of the sampling frame has an equal chance of selection; use a random number generator or a table of random digits. It is unbiased, but a small sample may still miss whole sub-groups.
简单随机抽样： 抽样框中每个个体被选中的机会均等；可使用随机数生成器或随机数表。它无偏倚，但样本量较小时仍可能遗漏整个子群体。
Systematic sampling: select every k-th item after a random start. It is quick and covers the list evenly, but hidden patterns in the list (for example, rows arranged by ability) can introduce bias.
系统抽样： 随机确定起点后每隔第 k 个个体抽取一个。它速度快、能均匀覆盖名单，但名单中隐藏的规律（例如按能力排列的行）可能引入偏倚。
Stratified sampling: split the population into non-overlapping strata that share a relevant characteristic, then sample proportionally within each stratum. The number taken from stratum i is
分层抽样： 将总体划分为互不重叠、且具有相关特征的分层，然后在各层内按比例抽样。第 i 层应抽取的人数为
nᵢ = (Nᵢ ÷ N) × n
Stratified sampling guarantees representation of every stratum and usually gives a more precise estimate for the same sample size, but it requires reliable information about the strata before sampling begins.
分层抽样能保证每一层都被代表，在相同样本量下通常给出更精确的估计，但在抽样前需要掌握关于各层的可靠信息。
Cluster sampling: randomly select whole clusters (for example, tutor groups) and measure everyone inside them. It is cheap and quick, but individuals within a cluster tend to be similar, so sampling error is larger.
整群抽样： 随机抽取整个群（例如班级），并测量群内所有人。它成本低、速度快，但同一群内的个体往往相似，因此抽样误差更大。
Quota and opportunity sampling: non-random methods that are convenient but cannot support generalisations to a wider population and are easily biased.
配额抽样与方便抽样： 属于非随机方法，操作方便，但无法支持推广到更广的总体，且容易产生偏倚。
4. Data Collection and Data Types | 数据收集与数据类型
Choosing the right data type keeps the rest of the investigation consistent. Primary data are collected first-hand for your own question; secondary data already exist and may have been gathered for a different purpose.
选择正确的数据类型能让后续调查保持一致。一手数据是为你自己的研究问题而亲自收集的；二手数据已经存在，当初可能是为其他目的收集的。
Primary data are tailored to the question and you control the quality, but collection is time-consuming and the sample is usually small. Secondary data are quick, cheap and often large-scale, but definitions may not match your question and the data may be out of date.
一手数据贴合研究问题且质量由你掌控，但收集耗时，样本量通常较小。二手数据获取快、成本低、往往规模大，但其定义可能与你的问题不匹配，数据也可能过时。
Qualitative data describe qualities such as colour or opinion in words; quantitative data are numerical. Quantitative data may be discrete (counted, such as number of siblings) or continuous (measured on a scale, such as height in cm).
定性数据用文字描述颜色、观点等性质；定量数据则是数值。定量数据可分为离散型（通过计数得到，例如兄弟姐妹数量）或连续型（通过刻度测量得到，例如以厘米计的身高）。
| Level | Feature | Example |
| Nominal | Categories with no order | Eye colour |
| Ordinal | Categories with a clear order | 1 = dislike, 5 = like |
| Discrete | Counted whole numbers | Books read in a month |
| Continuous | Any value on a scale | Time to run 100 m |
5. Experimental Design: Variables, Control and Randomisation | 实验设计：变量、控制与随机化
In an experiment you deliberately change something and measure the effect. The variable you change is the independent variable; the variable you measure is the dependent variable; everything else that could affect the outcome should be a controlled variable, kept the same for every participant.
在实验中，你主动改变某个因素并测量其影响。你改变的那个变量是自变量；你测量的变量是因变量；其他可能影响结果的因素都应作为控制变量，对所有参与者保持一致。
A control group receives no treatment or a placebo, so the effect of the treatment can be isolated. Random allocation of participants to groups spreads unknown confounding variables evenly and prevents the experimenter choosing who gets which condition.
对照组不接受处理或接受安慰剂，从而可以分离出处理本身的效果。随机分配参与者到各组，能把未知的混杂变量均匀分散，并防止实验者自行决定谁接受哪种条件。
Blinding means the participant does not know which group they are in; double blinding means the person recording the results does not know either. Both reduce the risk that expectations change behaviour.
盲法指参与者不知道自己属于哪一组；双盲则指记录结果的人也不知道。两者都能降低预期改变行为的风险。
A matched pairs design pairs participants who are similar in a relevant characteristic and gives one of each pair the treatment, which reduces natural variation but requires twice the screening effort.
配对设计把在相关特征上相似的参与者配成一对，每对中一人接受处理，这能减少个体差异，但需要双倍的筛选工作量。
Ethical points that examiners look for include informed consent, the right to withdraw, confidentiality of responses and avoiding any physical or psychological harm.
考官关注的伦理要点包括：知情同意、退出权利、答复的保密性，以及避免任何身心伤害。
6. Questionnaire Design and Sources of Bias | 问卷设计与偏倚来源
Most practical investigation questions in Paper 2 ask you to criticise a questionnaire. Learn the standard faults and the standard fixes.
Paper 2 中大多数实践调查题都要求你评价一份问卷。请掌握常见的典型缺陷及其改进方法。
Leading questions push the respondent towards an answer, such as 'Do you agree that the new canteen is much better?'. Rewrite neutrally: 'How does the new canteen compare with the old one?'.
诱导性问题会把受访者推向某个答案，例如"你同意新食堂好得多吗？"。应改为中性表述："新食堂与旧食堂相比如何？"。
Double-barrelled questions ask two things at once, for example 'Do you eat breakfast and exercise regularly?'. Split them into two separate questions.
双重问题一次问了两个内容，例如"你经常吃早餐并锻炼吗？"。应拆分为两个独立问题。
Response boxes must be mutually exclusive (no overlap) and exhaustive (covering every possible answer, including 'other' or 'prefer not to say'). Overlapping age bands such as 10-15 and 15-20 make the data unusable.
选项框必须互斥（不重叠）且穷尽（涵盖所有可能答案，包括"其他"或"不愿透露"）。诸如 10-15 与 15-20 这类重叠的年龄段会使数据无法使用。
Watch for non-response bias (the people who reply differ from those who do not), volunteer bias (self-selected respondents are often more motivated) and social desirability bias (respondents give the answer they think is expected). A short pilot study on a small group usually exposes these problems before the main collection begins.
要留意无应答偏倚（回复者与未回复者不同）、自愿者偏倚（自荐受访者往往更有积极性）和社会期望偏倚（受访者给出他们认为被期待的答案）。在正式收集前先用小群体做预调查，通常能提前暴露这些问题。
7. Cleaning, Grouping and Recording Data | 数据清理、分组与记录
Before analysis, raw data must be checked. An outlier R. Do not delete an outlier automatically: investigate whether it is a recording error, an unusual but genuine case, or evidence of a faulty measuring instrument, and state your decision.
分析之前必须先检查原始数据。异常值是远离分布其余部分的数值，通常定义为低于 LQ − 1.5 × IQR 或高于 UQ + 1.5 × IQR。不要自动删除异常值：应调查它是记录错误、罕见但真实的情况，还是测量工具失灵的迹象，并说明你的处理决定。
Missing values should be reported, not silently ignored, because ignoring them reduces the effective sample size and may distort the mean if the missing cases share a characteristic.
缺失值应被如实报告，而不能悄悄忽略，因为忽略会降低有效样本量，并且如果缺失案例具有共同特征，还可能使均值失真。
For continuous data, group into classes of equal width where possible. Unequal class widths are allowed but then you must plot frequency density rather than frequency on a histogram:
对于连续数据，尽可能划分为等宽的组距。允许使用不等组距，但此时在直方图上必须绘制频数密度而非频数：
Frequency density = Frequency ÷ Class width
Keep a recording sheet with clear column headings, units and a consistent number of decimal places. In the exam, sloppy units are a frequent source of lost marks.
记录表应具备清晰的列标题、单位以及一致的小数位数。考试中，单位书写马虎是常见的失分原因。
8. Analysing Data: Averages and Spread | 数据分析：集中趋势与离散程度
A single average never tells the whole story. A full comparison of two distributions requires one measure of location and one measure of spread, both quoted in context.
仅凭一个平均数无法说明全部问题。要完整比较两个分布，需要同时给出一个位置度量和一散布度量，并结合具体情境说明。
For raw data, the mean is
对于原始数据，平均数为
x̄ = Σx ÷ n
The median is the middle value once the data are ordered; it is resistant to outliers. The mode is the most frequent value and is the only average usable for qualitative or nominal data.
中位数是数据排序后位于中间的值，对异常值不敏感。众数是出现次数最多的值，是定性或名义数据唯一可用的平均数。
The range uses only two values, so a single outlier can destroy it. The interquartile range is far more robust:
极差只使用两个数值，因此一个异常值就能使其失去意义。四分位距则稳健得多：
IQR = Upper quartile − Lower quartile = Q₃ − Q₁
Quartiles also reveal shape. If Q₃ − Q₂ is greater than Q₂ − Q₁ the distribution is positively skewed; if Q₂ − Q₁ is greater the distribution is negatively skewed; if the two differences are roughly equal the distribution is symmetric.
四分位数还能揭示分布形状。若 Q₃ − Q₂ 大于 Q₂ − Q₁，则分布为正偏；若 Q₂ − Q₁ 更大，则为负偏；若两者大致相等，则分布近似对称。
When only grouped data are available, the estimated mean uses midpoints:
当只有分组数据时，估计均值需使用组中值：
x̄ ≈ Σ(f × midpoint) ÷ Σf
9. Choosing and Comparing Charts | 选择与比较图表
Match the chart to the data type. Bar charts suit discrete or categorical data; pie charts show proportion but become unreadable with many categories; histograms are for continuous data with frequency density on the vertical axis; cumulative frequency graphs give medians, quartiles and percentiles directly.
图表要与数据类型匹配。条形图适用于离散或分类数据；饼图展示比例，但类别过多时难以阅读；直方图用于连续数据，纵轴为频数密度；累积频数图可直接读出中位数、四分位数和百分位数。
A box plot displays the minimum, LQ, median, UQ and maximum in one line, which makes comparing two groups very efficient. It does not, however, show the sample size or the shape of the distribution in detail, so a box plot alone is rarely enough for a full evaluation.
箱线图用一条线同时展示最小值、下四分位数、中位数、上四分位数和最大值，因此非常适合比较两组数据。但它不显示样本量，也无法细致体现分布形状，所以单凭箱线图通常不足以完成完整评估。
A scatter graph displays the relationship between two variables. Look at the direction, form and strength of the association, and remember to describe outliers too. Spearman's rank correlation coefficient is used in the extended content to measure the strength of a monotonic relationship:
散点图展示两个变量之间的关系。要观察相关的方向、形式和强度，并记得描述异常点。在延伸内容中使用斯皮尔曼等级相关系数来衡量单调关系的强度：
rₛ = 1 − (6Σd²) ÷ (n(n² − 1))
where d is the difference between the ranks of each pair and n is the number of pairs. Values close to +1 or −1 indicate strong association, values near 0 indicate little or none.
其中 d 为每一对数据等级之差，n 为对的数量。接近 +1 或 −1 的取值表示相关性强，接近 0 则表示几乎没有相关。
10. Evaluating an Investigation | 评估调查
Evaluation carries a large share of the marks, so treat it as a structured answer rather than a final sentence. Address three issues: validity, reliability and bias.
评估部分占分比重很大，因此要把它当作结构化作答，而不是收尾的一句话。应处理三个问题：有效性、可靠性和偏倚。
Validity asks whether you measured what you claim to measure. A questionnaire about 'screen time' that only counts television misses phone and laptop use, so the measurement lacks validity.
有效性追问的是你是否测量了声称要测量的东西。只统计电视时间的"屏幕时间"问卷遗漏了手机和笔记本电脑的使用，因此该测量缺乏有效性。
Reliability asks whether repeating the investigation would give similar results. Larger samples, standardised procedures and consistent instruments all improve reliability; a single reading of one participant does not.
可靠性追问的是重复该调查是否会得到相似结果。更大的样本、标准化的流程和一致的测量工具都能提高可靠性；仅对一名参与者进行一次读数则不能。
Common evaluation weaknesses to name include: a small sample, a non-random sampling method, a leading question, a biased sampling frame, confounding variables that were not controlled, and results
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