📚 Case Study Combat: Pre-U Edexcel Statistics in Action | 案例分析实战演练：Pre-U Edexcel 统计学实战
In the Pre-U Edexcel Statistics examination, the ability to dissect a real-world scenario, select the appropriate statistical tool, and communicate findings lucidly is what distinguishes a top-grade candidate. This article walks you through a full-scale case study involving a pharmaceutical trial, forcing you to apply everything from sampling techniques to non-parametric tests. We will confront messy data, justify assumptions, compute test statistics manually where meaningful, and interpret results in context. By the end, you will have a model for structuring your own responses under timed pressure.
在 Pre-U Edexcel 统计学考试中，能够剖析真实情境、选择合适的统计工具并清晰地传达结论，是取得高分的分水岭。本文将通过一个完整的药物试验案例，带你运用从抽样技术到非参数检验的全套知识。我们会面对杂乱的数据，论证假设，有意义地进行手工计算，并在上下文中解释结果。最终，你将掌握一个在限时压力下组织答案的范本。
1. Setting the Scene – The Hypotensive Drug Trial | 背景设定——降压药试验
A pharmaceutical company claims that its new drug reduces systolic blood pressure by more than 5 mmHg on average compared to a placebo. A random sample of 35 patients is assigned to the treatment group, and another 32 to the placebo group. After eight weeks, the reduction in systolic blood pressure (mmHg) is recorded. Before the main analysis, we must scrutinise the study design: is it completely randomised, double-blind? Are there confounding variables (age, diet, baseline BP)? The examiner expects you to identify potential sources of bias and suggest remedies, even if the question does not explicitly ask.
某制药公司声称，其新药相比安慰剂能将收缩压平均多降低 5 mmHg 以上。一项试验中，35 名患者被随机分配至治疗组，32 名分配至安慰剂组。八周后，记录收缩压下降值 (mmHg)。在主分析之前，我们必须审视研究设计：是否为完全随机、双盲？是否有混杂变量（年龄、饮食、基线血压）？考官期望你找出潜在偏倚来源并提出改进方法，即使题干未明确要求。
2. Exploratory Data Analysis – Shapes and Summaries | 探索性数据分析——形态与汇总
We are given summary statistics: treatment group mean reduction x̄₁ = 8.4 mmHg, standard deviation s₁ = 3.2, n₁ = 35; placebo group x̄₂ = 2.1 mmHg, s₂ = 2.8, n₂ = 32. A quick box-plot sketch would reveal overlap, but the treatment group appears to have a higher centre. Normality is questionable for small samples; we check skewness coefficients or use a probability plot. In a Pre-U answer, stating that we proceed with a two-sample t-test but will consider a Mann-Whitney U test if the normality assumption is violated shows maturity.
我们获得汇总统计：治疗组平均降压 x̄₁ = 8.4 mmHg，标准差 s₁ = 3.2，n₁ = 35；安慰剂组 x̄₂ = 2.1 mmHg，s₂ = 2.8，n₂ = 32。快速绘制的箱线图会显示重叠，但治疗组中心值似乎更高。小样本的正态性值得怀疑；我们可以检查偏度系数或使用概率图。在 Pre-U 答案中，陈述我们将进行双样本 t 检验，但如果正态性假设不成立会考虑 Mann-Whitney U 检验，这体现了思维的成熟度。
3. Two-Sample t-Test – Does the Drug Work? | 双样本 t 检验——药物是否有效？
The null hypothesis H₀: μ₁ – μ₂ = 5 (the claimed margin), against H₁: μ₁ – μ₂ > 5 (one-tailed). We use the unequal variance (Welch) t-test because sample sizes differ and we cannot assume equal population variances. The test statistic is
零假设 H₀: μ₁ – μ₂ = 5（声称的优势），备择假设 H₁: μ₁ – μ₂ > 5（单尾）。我们使用不等方差的 Welch t 检验，因为样本量不同且不能假设总体方差相等。检验统计量为
t = ((x̄₁ – x̄₂) – 5) / √(s₁²/n₁ + s₂²/n₂)
Plugging numbers: (8.4 – 2.1 – 5) / √(3.2²/35 + 2.8²/32) = 1.3 / √(0.2926 + 0.2450) = 1.3 / √0.5376 = 1.3 / 0.7332 ≈ 1.773. Degrees of freedom via Welch-Satterthwaite equation:
代入数值：(8.4 – 2.1 – 5) / √(3.2²/35 + 2.8²/32) = 1.3 / √(0.2926 + 0.2450) = 1.3 / √0.5376 = 1.3 / 0.7332 ≈ 1.773。自由度通过 Welch-Satterthwaite 公式计算：
df ≈ (s₁²/n₁ + s₂²/n₂)² / [ (s₁²/n₁)²/(n₁-1) + (s₂²/n₂)²/(n₂-1) ]
This yields df ≈ 62.4 (say 62). The critical value for α = 0.05 one-tailed is about 1.67. Our t = 1.773 > 1.67, so we reject H₀. There is sufficient evidence at the 5% level that the drug reduces systolic BP by more than 5 mmHg over placebo. But always check p-value: for t=1.773 with df=62, p ≈ 0.0406, confirming significance.
计算得自由度约 62.4（取 62）。α = 0.05 单尾临界值约为 1.67。我们的 t = 1.773 > 1.67，因此拒绝 H₀。有充分证据在 5% 显著性水平下表明该药物相比安慰剂多降低收缩压超过 5 mmHg。但还需检查 p 值：t=1.773，df=62 时 p ≈ 0.0406，确认为显著。
4. Assumption Checking and Sensitivity | 假设检验与敏感性
A Pre-U examiner will reward a thorough discussion. The Welch test relaxes the equal variance assumption, but we still assume independent observations and approximately normal sampling distributions of means. Given moderate sample sizes (both >30), the Central Limit Theorem provides some robustness. However, we should check for outliers: suppose the treatment group has a maximum of 15.1 mmHg reduction. A Grubbs test could be applied, but visual inspection of a box-plot would suffice to note that data are reasonably symmetric. If not, a log-transformation or Mann-Whitney test is advised.
Pre-U 考官会对充分的讨论给予奖励。Welch 检验放松了方差相等假设，但我们仍假定观测独立，且均值的抽样分布近似正态。由于样本量中等（均大于 30），中心极限定理提供了一定的稳健性。但我们应检查离群值：假设治疗组最大值降幅为 15.1 mmHg。可应用 Grubbs 检验，但通过箱线图目视检查足以说明数据基本对称。若非如此，建议进行对数变换或 Mann-Whitney 检验。
5. Confidence Interval for the Difference | 差值的置信区间
Beyond hypothesis testing, estimation is vital. A 95% confidence interval for μ₁ – μ₂ is given by
除假设检验外，估计至关重要。μ₁ – μ₂ 的 95% 置信区间由下式给出：
(x̄₁ – x̄₂) ± t\* × √(s₁²/n₁ + s₂²/n₂)
With t* (df=62, two-tailed) ≈ 2.000. So CI = (8.4 – 2.1) ± 2.000 × 0.7332 = 6.3 ± 1.466, i.e., (4.834, 7.766). This interval does not contain 5? Wait, it does: 5 lies within (4.834, 7.766). How do we reconcile with the hypothesis test that rejected 5? The test was one-tailed for >5, while the CI is two-tailed. The lower bound is 4.834 < 5, meaning with 95% confidence the difference could be below 5. But the one-tailed test focused on values greater than 5; the 95% lower confidence bound (one-sided) would be (6.3 - 1.67×0.7332) = 5.075, which excludes 5. This is a subtle point that shows deep understanding.
取 t*（df=62，双尾）≈ 2.000。因此 CI = (8.4 – 2.1) ± 2.000 × 0.7332 = 6.3 ± 1.466，即 (4.834, 7.766)。该区间包含 5 吗？等等，确实包含：5 在 (4.834, 7.766) 内。如何与拒绝了 5 的假设检验协调？该检验是单尾检验（>5），而 CI 是双尾的。下限 4.834 < 5，意味着在 95% 置信度下差值可能低于 5。但单尾假设检验关注大于 5 的值；单侧 95% 置信下限为 (6.3 - 1.67×0.7332) = 5.075，不含 5。这个微妙之处展现了深刻理解。
6. Power Analysis – Did the Study Have Enough Patients? | 功效分析——研究是否有足够的患者？
Examiners often slip in a power question. Suppose the clinically meaningful difference is 3 mmHg (beyond the 5 margin). With the observed pooled standard deviation estimate of about 3.0, and total sample size 67, what was the power to detect a true difference of 6.3 mmHg (our observed) if α = 0.05? Using non-central t distributions is beyond Pre-U, but you can reason that the magnitude matters: the effect size d = 6.3/3 = 2.1 is large, so power is likely high. Conversely, if the true difference were only 3 mmHg, power might be inadequate, suggesting a larger trial is needed for regulatory approval. This shows strategic thinking.
考官常会加入功效问题。假设临床有意义的差值为 3 mmHg（超出 5 的边界）。使用观测到的合并标准差估计约 3.0，总样本量 67，当 α = 0.05 时，检测到真实差值为 6.3 mmHg（我们观测的）的功效是多少？使用非中心 t 分布超出了 Pre-U 范围，但你可以推理：效应量 d = 6.3/3 = 2.1 很大，因此功效可能很高。反之，若真实差值仅为 3 mmHg，功效可能不足，表明需要更大规模试验才能获得监管批准。这体现了战略思维。
7. Non-Parametric Alternative – Mann-Whitney U Test | 非参数替代方案——Mann-Whitney U 检验
When normality is severely violated, we rank all 67 observations. Suppose the sum of ranks for the treatment group is R₁ = 1420. The expected sum under H₀: no difference between distributions, is n₁(N+1)/2 = 35×68/2 = 1190. The standard error is √(n₁n₂(N+1)/12) = √(35×32×68/12) = √(76160/12) = √6346.67 ≈ 79.67. The test statistic z = (1420 – 1190) / 79.67 ≈ 2.886, giving p ≈ 0.002 (two-tailed). We conclude the distributions differ in location. Crucially, you must state that the Mann-Whitney test examines whether one distribution is stochastically larger, not a direct comparison of medians unless distributions are identically shaped.
当正态性严重违背时，我们对全部 67 个观测值排序。假设治疗组秩和 R₁ = 1420。在 H₀（两分布无差异）下，期望秩和为 n₁(N+1)/2 = 35×68/2 = 1190。标准误为 √(n₁n₂(N+1)/12) = √(35×32×68/12) = √(76160/12) = √6346.67 ≈ 79.67。检验统计量 z = (1420 – 1190) / 79.67 ≈ 2.886，p ≈ 0.002（双尾）。我们得出结论：分布在位置上不同。关键要说明，Mann-Whitney 检验考察的是一个分布是否随机地更大，而非直接比较中位数，除非分布形状相同。
8. Regression Twist – Does Age Influence Efficacy? | 回归转折——年龄影响疗效吗？
The examiner may introduce patient-level data: for the treatment group, we have age (x) and blood pressure reduction (y). A scatterplot suggests a positive linear trend. We fit y = a + b x. Given Σx = 1890, Σy = 294, Σx² = 104650, Σy² = 2650, Σxy = 16110, n = 35. The slope b = (nΣxy – ΣxΣy) / (nΣx² – (Σx)²) = (35×16110 – 1890×294) / (35×104650 – 1890²) = (563850 – 555660) / (3662750 – 3572100) = 8190 / 90650 ≈ 0.0904. Intercept a = ȳ – b x̄ = 8.4 – 0.0904×54 = 8.4 – 4.8816 = 3.5184. So predicted reduction = 3.52 + 0.0904×Age. To test if age is significant, we compute the t-ratio for b. First, residual sum of squares RSS = Σ(y – ŷ)². You would need to compute SSE or use given ANOVA table. Suppose the standard error of b is 0.028, then t = 0.0904/0.028 ≈ 3.23 with df=33, p < 0.01, indicating age significantly contributes. A 95% CI for slope: 0.0904 ± 2.03×0.028 = (0.0335, 0.147). This shows older patients tend to have a larger reduction – a finding with clinical implications.
考官可能引入个体数据：治疗组提供年龄 (x) 与血压降幅 (y)。散点图显示正向线性趋势。我们拟合 y = a + b x。给定 Σx = 1890, Σy = 294, Σx² = 104650, Σy² = 2650, Σxy = 16110, n = 35。斜率 b = (nΣxy – ΣxΣy) / (nΣx² – (Σx)²) = (35×16110 – 1890×294) / (35×104650 – 1890²) = 0.0904。截距 a = 8.4 – 0.0904×54 = 3.5184。预测降幅 = 3.52 + 0.0904×年龄。检验年龄是否显著，需计算 b 的 t 比。假设 b 的标准误为 0.028，则 t = 3.23，df=33，p < 0.01，表明年龄有显著贡献。斜率的 95% CI：(0.0335, 0.147)。这表明年长患者降幅更大——具有临床意义的发现。
9. Chi-Squared: Adverse Events Analysis | 卡方检验：不良事件分析
Another facet: a 2×2 contingency table of adverse event (AE) occurrence by group. Suppose 8 out of 35 in treatment experienced nausea versus 3 out of 32 in placebo. The expected frequencies under independence: Treatment AE: (35×11)/67 ≈ 5.75, No AE: 29.25; Placebo AE: 5.25, No AE: 26.75. χ² = Σ (O-E)²/E. χ² = (8-5.75)²/5.75 + (27-29.25)²/29.25 + (3-5.25)²/5.25 + (29-26.75)²/26.75 ≈ 0.878 + 0.173 + 0.964 + 0.189 = 2.204. With 1 df, critical value at 5% is 3.841. Thus, no significant association. However, some Pre-U questions ask for Yates’ correction: χ²_corrected = 1.42 even less significant. Always discuss the caveat of small expected frequencies; Fisher’s exact test might be more appropriate if any cell <5.
另一层面：按组别划分的不良事件 (AE) 2×2 列联表。假设治疗组 35 人中 8 人恶心，安慰剂组 32 人中 3 人。独立性下的期望频数：治疗组 AE: (35×11)/67 ≈ 5.75，无 AE: 29.25；安慰剂组 AE: 5.25，无 AE: 26.75。χ² = Σ (O-E)²/E = 2.204。df=1，5% 临界值 3.841，不显著。但部分 Pre-U 问题要求 Yates 校正：校正后 χ² = 1.42 更不显著。务必讨论期望频数较小的注意事项；若有单元格小于 5，Fisher 确切检验更合适。
10. Communicating Findings and Limitations | 沟通结果与局限性
A polished answer synthesises statistical evidence into a coherent narrative: “The drug demonstrates a statistically significant additional reduction of 6.3 mmHg (95% CI: 4.83 to 7.77) over placebo; however, the lower confidence bound narrowly exceeds 5 in a one-sided sense. Age positively modifies the effect, and adverse event rates appear comparable. Limitations include the small sample size for a drug trial, potential volunteer bias, and the need for longer follow-up.” The ability to balance statistical significance with clinical significance, and to suggest future studies, is the hallmark of a top-tier candidate.
一个精炼的答案将统计证据综合为连贯的叙述：“该药物在安慰剂基础上显示出统计显著的额外 6.3 mmHg 降压效果（95% CI: 4.83 至 7.77）；但单侧置信下限刚超过 5。年龄正面修饰效应，不良事件率相当。局限性包括药物试验样本量较小、潜在志愿者偏倚以及需要更长随访。”能够平衡统计显著性与临床意义，并提出未来研究建议，是顶尖考生的标志。
11. Model Answer Blueprint | 答案蓝图
Structure your response: (1) Clarify objectives and hypotheses. (2) EDA with descriptive statistics and graphs. (3) Choose and justify the test, including assumption checks. (4) Perform calculations, showing formulas. (5) State conclusions in context, including effect size and confidence interval. (6) Discuss limitations and alternatives. (7) Where applicable, extend to regression or categorical analysis. Time management is crucial; allocate ~5 minutes per sub-question in a 1-hour case study. Practice writing concise, technically precise English and Chinese summaries to shine.
组织答案结构：(1) 阐明目标与假设。(2) 包含描述统计与图形的 EDA。(3) 选择并论证检验，包括假设检查。(4) 进行计算，展示公式。(5) 在上下文中陈述结论，包括效应量与置信区间。(6) 讨论局限性与替代方案。(7) 适用时扩展至回归或分类分析。时间管理至关重要；在 1 小时案例分析中，为每个子问题分配约 5 分钟。练习书写简洁、技术精准的中英文总结，将脱颖而出。
12. Final Exam Tips | 考试贴士
Edexcel Pre-U Statistics papers demand precision: use notation correctly (H₀, H₁, μ, σ, t, χ², p-value). Always specify significance level α. Round to sensible decimal places. When unsure about normality, quote the Central Limit Theorem. If variances are unequal, mention Welch’s correction. Never forget a concluding sentence that answers the original question in plain language. A case study is a story; you are the detective using numbers as clues.
Edexcel Pre-U 统计学试卷要求精确：正确使用符号（H₀, H₁, μ, σ, t, χ², p 值）。务必指定显著性水平 α。合理保留小数位。对正态性有疑问时，引用中心极限定理。若方差不齐，提及 Welch 校正。切勿忘记用通俗语言回答原始问题的总结句。案例研究如同故事；你是用数字作为线索的侦探。
Published by TutorHao | Statistics 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Cancel reply