📚 Case Study Walkthrough in AQA Year 12 Statistics | AQA Year 12 统计案例分析实战演练
A solid understanding of statistics goes far beyond memorising formulas – it requires the ability to apply concepts to real-world scenarios, interpret data meaningfully, and justify every step of an investigation. This article presents a complete case study walkthrough tailored to the AQA Year 12 Statistics specification, demonstrating how each topic from sampling to hypothesis testing can be integrated into a single, coherent analysis. You will see how a customer satisfaction survey for a coffee chain evolves from a research question through data collection, summary statistics, probability modelling, and finally a formal hypothesis test. Every technique is linked directly to the skills examined in AQA papers, making this an ideal revision resource for building confidence in statistical problem-solving.
扎实的统计学理解远不止于记忆公式 —— 它要求你能够将概念应用到真实情境中，有意义地解读数据，并为调查的每一步提供依据。本文提供了一个契合 AQA Year 12 统计考试大纲的完整案例分析推演，展示了从抽样到假设检验的每一个主题如何融入一个连贯的分析。你将看到一家咖啡连锁店的顾客满意度调查如何从一个研究问题逐步展开，经历数据收集、概括统计量、概率建模，最后进入正式的假设检验。每一步技巧都直接关联 AQA 试卷所考查的能力，使本文成为提升统计问题解决信心的理想复习资源。
1. Setting the Scene: A Customer Satisfaction Survey | 场景设定：顾客满意度调查
A regional manager of a coffee chain wants to know whether the proportion of customers who rate their experience as ‘satisfied or above’ has increased since a new training programme was introduced for baristas. Historical records show that 72% of customers used to be satisfied. The manager decides to conduct a survey in ten randomly selected stores over one week, asking customers to rate their visit on a scale from 1 to 5, with 4 and 5 indicating satisfaction. The research question is: ‘Is the current proportion of satisfied customers greater than 72%?’ This sets the stage for a one-tailed hypothesis test, but before any inference can be made, the team must carefully plan the data collection.
一家咖啡连锁店的区域经理想知道，自从为咖啡师引入新的培训项目后，将体验评为“满意及以上”的顾客比例是否有所上升。历史记录显示，过去有 72% 的顾客感到满意。经理决定在十家随机选择的门店进行为期一周的调查，请顾客以 1 到 5 分评价其光顾体验，其中 4 分和 5 分表示满意。研究问题是：“当前满意的顾客比例是否大于 72%？”这为单尾假设检验搭建了舞台，但在进行任何推断之前，团队必须仔细规划数据的收集。
2. Sampling Methods and Data Collection | 抽样方法与数据收集
The manager uses stratified sampling to select ten stores: the chain’s 80 outlets are divided into four strata by region, and within each stratum, stores are selected in proportion to the number of outlets in that region. This ensures geographical representation. At each chosen store, every 10th customer exiting between 8am and 8pm is invited to complete the questionnaire, which is an example of systematic sampling. The combination of these methods balances practical convenience with a reasonable coverage of the target population – all customers visiting the chain during the study period. Potential biases are considered, such as time-of-day effects and non-response from busy customers.
经理使用分层抽样选择十家门店：该连锁的 80 家门店按地区划分为四个层，在每个层内按该地区门店数量比例选取门店，这就确保了地理代表性。在每个被选中的门店里，上午 8 点到晚上 8 点之间，每第 10 位离开的顾客被邀请填写问卷，这是系统抽样的一个例子。这些方法的组合在实际便利性与对目标总体（研究期间光顾该连锁的所有顾客）的合理覆盖之间取得了平衡。同时考虑了可能的偏差，例如时间段效应以及忙碌顾客的无回应问题。
3. Types of Data and Data Cleaning | 数据类型与数据整理
The raw data collected includes the store location, the date and time, the satisfaction rating (1–5), and an optional comment. The rating is treated as discrete numerical data but can also be viewed as ordinal categorical data because the numbers represent ordered categories. Before analysis, the team cleans the data by removing incomplete responses – any form where the rating was left blank is discarded, leaving a sample of 250 fully completed questionnaires. Out of these, 195 customers gave a rating of 4 or 5. The variable of interest is thus binary: satisfied (score ≥4) or not satisfied. This transformation is crucial because the hypothesis test will focus on a population proportion.
收集到的原始数据包括门店位置、日期和时间、满意度评分（1–5）以及可选评论。评分被视为离散数值型数据，但也可看作有序的分类数据，因为数字代表有序的类别。在分析之前，团队通过移除不完整的回复来清洗数据 —— 任何评分留空的表格都被丢弃，最终剩下 250 份完整填写的问卷。其中，195 位顾客给出了 4 或 5 分。因此，感兴趣的变量是二值的：满意（分数 ≥4）或不满意。这一转换至关重要，因为假设检验将针对一个总体比例进行。
4. Summarising Data: Central Tendency and Spread | 数据概括：集中趋势与离散程度
For the rating scores treated as numerical values, the mean score is calculated as 4.12, and the median is 4. The close alignment of mean and median suggests that the distribution is fairly symmetric. The standard deviation is 0.85, indicating that most ratings fall within roughly one point of the mean. For the binary ‘satisfied’ variable, the sample proportion is p̂ = 195/250 = 0.78. This is higher than the historical 0.72, but we need to test whether this difference is statistically significant. A summary table of basic statistics helps clarify the picture:
将评分视为数值时，计算得到的平均分为 4.12，中位数为 4。均值与中位数接近，表明分布相当对称。标准差为 0.85，意味着大多数评分落在均值上下约 1 分范围内。对于二值的“满意”变量，样本比例为 p̂ = 195/250 = 0.78。这高于历史值 0.72，但我们需要检验这一差异是否具有统计显著性。一个基本统计量的汇总表格有助于澄清情况：
| Statistic | Value |
|---|---|
| Sample size n | 250 |
| Number satisfied X | 195 |
| Sample proportion p̂ | 0.78 |
| Mean rating (1–5) | 4.12 |
| Standard deviation | 0.85 |
Using these summaries, we can already see that customer satisfaction appears to be high, but the formal test must account for sampling variability.
借助这些概括统计量，我们已经能看出顾客满意度似乎较高，但正式的检验必须考虑抽样变异性。
5. Visualising the Data | 数据可视化
A bar chart of the rating frequencies shows a concentration at ratings 4 and 5, with very few low scores. Plotting a histogram for the 1–5 scores (using intervals 0.5–1.5, 1.5–2.5, etc.) confirms a left-skewed shape, typical of satisfaction data where most customers are satisfied. A box plot can also be used to display the five-number summary: minimum 1, Q1 4, median 4, Q3 5, maximum 5. The interquartile range (IQR) is only 1, reinforcing the observation of low variability among satisfied customers. These visualisations are not just about creating graphs; they help the analyst detect outliers, assess symmetry, and communicate findings to non-statistical stakeholders.
评分频数的条形图显示分数集中在 4 和 5，极低分很少。将 1–5 分绘制成直方图（区间为 0.5–1.5、1.5–2.5 等），证实了左偏态形状，这在大多数顾客满意的满意度数据中是典型的。箱线图也可用来展示五数概括：最小值 1，下四分位数 4，中位数 4，上四分位数 5，最大值 5。四分位距（IQR）仅为 1，强化了满意顾客中变异性低的观察。这些可视化并不仅仅是为了画图；它们帮助分析者检测离群值、评估对称性，并向非统计专业的利益相关者传达发现。
6. Probability Foundations: Defining Events | 概率基础：定义事件
Probability theory underpins all the inference that follows. Let S be the event that a randomly selected customer is satisfied. From the historical data, we assume P(S) = 0.72 under the null hypothesis. In the sampling process, we consider the 250 responses as trials. If we assume customers respond independently, the count of satisfied customers, X, can be modelled by a binomial distribution if the null proportion is true. The independence condition is approximately met because customers were sampled systematically and the sample is small relative to the total number of customers visiting the chain that week. Understanding the structure of this random variable allows us to calculate how likely or unlikely a sample proportion of 0.78 is under the null hypothesis.
概率论为所有后续推断奠定了基础。设 S 为随机抽取的顾客表示满意的事件。根据历史数据，我们假定在零假设下 P(S) = 0.72。在抽样过程中，我们将 250 份回答视为试验。如果假设顾客的回答相互独立，那么在零假设比例成立的情况下，满意顾客的数量 X 可以用二项分布建模。独立性条件大致满足，因为顾客是系统抽样抽取的，且样本量相对于该周光顾该连锁店的顾客总数来说很小。理解这一随机变量的结构，使我们能够计算在零假设下样本比例 0.78 出现的可能性有多大或多小。
7. Binomial Distribution in Action | 二项分布实战
Under the null hypothesis H₀: p = 0.72, the number of satisfied customers in a sample of 250 follows the binomial distribution B(250, 0.72). The probability mass function is given by:
在零假设 H₀: p = 0.72 下，容量为 250 的样本中满意的顾客数量服从二项分布 B(250, 0.72)。其概率质量函数为：
P(X = k) = C(250, k) × (0.72)ᵏ × (0.28)²⁵⁰⁻ᵏ
We are interested in the probability of observing a result as extreme as, or more extreme than, 195 satisfied customers, given H₀. This is P(X ≥ 195). Direct calculation by hand is cumbersome, so either a statistical software or a normal approximation is used. The expected number under H₀ is np = 250 × 0.72 = 180. The standard deviation is √(np(1 – p)) = √(250 × 0.72 × 0.28) ≈ 7.10. The observed 195 is about 2.11 standard deviations above the mean. This suggests that the result might be statistically significant, but we must apply the formal test to be sure.
我们感兴趣的是在 H₀ 成立时，观察到 195 或更多满意顾客的概率，即 P(X ≥ 195)。直接手算十分繁琐，因此通常使用统计软件或正态近似。H₀ 下的期望值为 np = 250 × 0.72 = 180。标准差为 √(np(1 – p)) = √(250 × 0.72 × 0.28) ≈ 7.10。观察到的 195 大约比均值高出 2.11 个标准差。这表明该结果可能具有统计显著性，但我们必须应用正式检验来加以确认。
8. The Normal Distribution and Standard Scores | 正态分布与标准分
Because n is large and p is not too close to 0 or 1, the binomial distribution B(250, 0.72) can be approximated by a normal distribution with the same mean and standard deviation: N(180, 7.10²). To compute the tail probability, we standardise the observed count using a continuity correction. Without the correction, the z-score for 195 is:
由于 n 很大且 p 不太靠近 0 或 1，二项分布 B(250, 0.72) 可用具有相同均值和标准差的正态分布 N(180, 7.10²) 来近似。为计算尾部概率，我们结合连续校正对观测计数进行标准化。未校正时，195 的 z 值为：
z = (195 – 180) / 7.10 ≈ 2.11
Using standard normal tables, the one-tailed p-value corresponding to z = 2.11 is about 0.0174. This means that if the true proportion were still 0.72, we would observe 195 or more satisfied customers out of 250 in only about 1.74% of all possible samples of this size. The concept of the standard normal distribution underpins many AQA questions: being able to move back and forth between raw scores, z-scores, and probabilities is essential. Students should also be aware of when continuity corrections are required for greater accuracy.
根据标准正态分布表，z = 2.11 对应的单尾 p 值约为 0.0174。这意味着如果真实比例仍是 0.72，在容量为 250 的所有可能样本中，仅有约 1.74% 的样本会出现 195 或更多满意的顾客。标准正态分布的概念支撑着许多 AQA 试题：能够在原始分、z 分数与概率之间来回转换是至关重要的。学生还应意识到何时需要连续校正以提高准确性。
9. Sampling Distribution and the Central Limit Theorem | 抽样分布与中心极限定理
Instead of working with the count X, we could directly consider the sampling distribution of the proportion p̂. By the Central Limit Theorem, for large n, p̂ is approximately normally distributed with mean p and standard deviation √(p(1 – p) / n). Under H₀, the standard deviation of p̂ is √(0.72 × 0.28 / 250) ≈ 0.0284. The observed p̂ = 0.78 yields a z-score of:
除了处理计数 X，我们也可以直接考虑比例 p̂ 的抽样分布。根据中心极限定理，当 n 很大时，p̂ 近似服从均值为 p、标准差为 √(p(1 – p) / n) 的正态分布。在 H₀ 下，p̂ 的标准差为 √(0.72 × 0.28 / 250) ≈ 0.0284。观测到的 p̂ = 0.78 产生的 z 分数为：
z = (0.78 – 0.72) / 0.0284 ≈ 2.11
This is exactly the same standardised value as before, confirming the equivalence. The Central Limit Theorem thus allows us to conduct inference about proportions without needing to use the binomial distribution directly, provided the sample size is sufficiently large. In the AQA Year 12 course, recognising the conditions for the normal approximation (np > 5 and n(1 – p) > 5) is a key skill.
这与之前的标准值完全相同，确认了等价性。因此，中心极限定理使我们能够对比例进行推断，而无需直接使用二项分布，前提是样本量足够大。在 AQA Year 12 课程中，识别正态近似的条件（np > 5 且 n(1 – p) > 5）是一项关键技能。
10. Hypothesis Testing: A One-Sample Test for Proportion | 假设检验：单样本比例检验
We now formalise the test following the AQA structure. Step 1: State hypotheses. H₀: p = 0.72, H₁: p > 0.72 (one-tailed). Step 2: Significance level α = 0.05. Step 3: Test statistic. We use the approximate normal test. Under H₀, z = (p̂ – 0.72) / √(0.72 × 0.28 / 250) ≈ 2.11. Step 4: Critical value or p-value. The critical z-value for a one-tailed test at α = 0.05 is 1.645. Since 2.11 > 1.645, or equivalently p-value ≈ 0.0174 < 0.05, the result falls in the rejection region. Step 5: Conclusion. There is sufficient evidence to reject H₀ at the 5% significance level. This suggests that the true proportion of satisfied customers has increased above the historical 72% after the training programme.
我们现在按照 AQA 的结构将检验正式化。第一步：陈述假设。H₀: p = 0.72，H₁: p > 0.72（单尾）。第二步：显著性水平 α = 0.05。第三步：检验统计量。我们使用近似正态检验。在 H₀ 下，z = (p̂ – 0.72) / √(0.72 × 0.28 / 250) ≈ 2.11。第四步：临界值或 p 值。单尾检验在 α = 0.05 时的临界 z 值为 1.645。由于 2.11 > 1.645，或等价地 p 值 ≈ 0.0174 < 0.05，该结果落入拒绝域。第五步：结论。在 5% 的显著性水平下有足够证据拒绝 H₀。这表明在培训项目开展后，满意顾客的真实比例已经上升至高于历史水平 72%。
11. Interpreting Results and Drawing Conclusions | 结果解读与结论
The statistical test gives a clear answer, but interpretation must go beyond ‘reject H₀’. The p-value of 0.0174 means there is only a 1.74% probability of obtaining a sample proportion as large as 0.78 purely by chance if the true satisfaction rate were still 0.72. This is strong evidence against H₀, but it does not prove that the training programme caused the increase. Other factors, such as seasonal promotions or a change in the customer base, could have contributed. In a full report, the manager should also consider the practical significance: an increase from 0.72 to 0.78 represents about 6 percentage points, which might translate into a meaningful uplift in customer loyalty and revenue. Confidence intervals can further enrich the interpretation. A 95% approximate confidence interval for the true proportion is p̂ ± 1.96 × √(p̂(1 – p̂) / n) = 0.78 ± 1.96 × 0.0262, giving [0.729, 0.831]. This interval lies entirely above 0.72, aligning with the test conclusion.
统计检验给出了清晰的答案，但解读必须超越“拒绝 H₀”本身。p 值为 0.0174 意味着，如果真实满意率仍是 0.72，仅凭偶然出现高达 0.78 的样本比例的概率只有 1.74%。这是反对 H₀ 的有力证据，但并不能证明培训项目就是比例上升的原因。其他因素，例如季节性促销或顾客群体的变化，也可能有所贡献。在一份完整报告中，经理还应考虑实际意义上的显著性：从 0.72 上升到 0.78 代表了约 6 个百分点的增长，这可能转化为顾客忠诚度和营收方面有意义的提升。置信区间可以进一步丰富解读。真实比例的大约 95% 置信区间为 p̂ ± 1.96 × √(p̂(1 – p̂) / n) = 0.78 ± 1.96 × 0.0262，得出 [0.729, 0.831]。该区间完全位于 0.72 以上，与检验结论相符。
12. Reflection: Strengths and Limitations | 反思：优点与局限性
This case study illustrates how AQA Year 12 Statistics topics are interconnected. Sampling design affects the validity of inferences; data cleaning ensures reliable calculations; descriptive statistics and visualisations set the scene; probability distributions provide the theoretical backbone for testing. The main limitation in this study is the potential for sampling bias – systematic sampling within stores might miss customers who visit only at specific times, and the exclusion of incomplete surveys could introduce non-response bias if dissatisfied customers were less likely to complete the form. Additionally, the normal approximation is valid here, but with smaller samples the binomial test or continuity correction would be mandatory. Reflecting on these issues is exactly what examiners look for in higher-level evaluation questions.
这个案例分析展示了 AQA Year 12 统计的各个主题如何相互联系。抽样设计影响推断的有效性；数据清洗确保计算可靠；描述统计和可视化奠定基础；概率分布为检验提供理论支撑。本研究的主要局限性在于可能的抽样偏差 —— 门店内的系统抽样可能会错过那些只在特定时间光顾的顾客，而排除不完整问卷则可能引入无应答偏差，如果不满意顾客填表的意愿较低的话。此外，正态近似在此有效，但面对更小的样本时，则必须使用二项检验或进行连续校正。对这些问题的反思，正是考官在高阶评估问题中所关注的内容。
Published by TutorHao | 统计 Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
屏轩国际教育cambridge primary/secondary checkpoint, cat4, ukiset,ukcat,igcse,alevel,PAT,STEP,MAT, ibdp,ap,ssat,sat,sat2课程辅导，国外大学本科硕士研究生博士课程论文辅导