📚 Case Study: Hypothesis Testing and Confidence Intervals in Action | 案例研究：假设检验与置信区间实战
Statistical inference lies at the heart of the Cambridge Year 13 Statistics syllabus, and nothing sharpens understanding like a real‐world case study. In this walkthrough, we will apply hypothesis testing and confidence interval construction to a practical scenario involving battery lifetimes. You will see how to state hypotheses, compute a test statistic, interpret p‑values, and link the test outcome to a confidence interval. The aim is to build a robust analytical mindset for exam‑style data analysis questions.
统计推断是剑桥 Year 13 统计大纲的核心，而最能加深理解的方法莫过于一个真实案例的演练。在本文中，我们将把假设检验和置信区间构建应用到一个有关电池寿命的实际场景中。你将看到如何陈述假设、计算检验统计量、解读 p 值，以及如何将检验结论与置信区间联系起来。目的是为应对考试风格的数据分析问题培养扎实的分析思维。
1. The Battery Lifetime Scenario | 电池寿命情景
A battery manufacturer claims that the mean lifetime of its latest AA battery is 1000 hours. A consumer watchdog suspects that the true mean is lower and decides to conduct a statistical test. A random sample of 50 batteries is selected and tested, yielding a sample mean lifetime of 985 hours. From historical data, the population standard deviation is known to be σ = 50 hours. The sample size of 50 allows us to invoke the Central Limit Theorem, so we can safely use a z‑test even if the underlying distribution is not perfectly normal.
一家电池制造商声称其最新款 AA 电池的平均寿命为 1000 小时。消费者监督机构怀疑真实均值低于该数值，决定进行统计检验。随机抽取 50 节电池进行测试，得到样本平均寿命为 985 小时。历史数据表明总体标准差 σ 为 50 小时。样本容量 50 使我们能够调用中心极限定理，因此即使总体分布不完全正态，我们也可以放心地使用 z 检验。
2. Formulating the Hypotheses | 建立假设
We are testing whether the mean lifetime is less than the claimed 1000 hours. The null hypothesis represents the status quo or the claim, while the alternative hypothesis captures our suspicion. This is a left‑tailed test.
我们要检验平均寿命是否低于声称的 1000 小时。零假设代表现状或声称值，备择假设则反映我们的怀疑。这是一个左侧检验。
H₀: μ = 1000 h
H₁: μ < 1000 h
The parameter μ denotes the true population mean lifetime. Notice that the alternative is directional, so we use a one‑sided test. In exam contexts, always specify both the parameter and its value clearly.
参数 μ 表示真实的总体平均寿命。注意备择假设是有方向的，因此我们使用单侧检验。在考试情境中，务必清晰地写明参数及其取值。
3. Selecting the Test Statistic | 选择检验统计量
Because the population standard deviation σ is known and the sample size n = 50 is large, the appropriate test statistic is the z‑statistic:
由于总体标准差 σ 已知且样本容量 n = 50 较大，合适的检验统计量是 z 统计量：
Z = (X̄ – μ₀) / (σ / √n)
Here X̄ is the sample mean, μ₀ = 1000 is the hypothesised mean, σ = 50, and n = 50. Under H₀, Z follows a standard normal distribution, N(0,1). This choice is justified by the Central Limit Theorem, which guarantees approximate normality of the sampling distribution of the mean for sufficiently large n.
这里 X̄ 是样本均值，μ₀ = 1000 是假设的均值，σ = 50，n = 50。在零假设下，Z 服从标准正态分布 N(0,1)。这一选择的依据是中心极限定理，它保证了当 n 足够大时，样本均值的抽样分布近似正态。
4. Setting the Significance Level and Critical Region | 设定显著性水平与拒绝域
We adopt a significance level of α = 0.05, which is common in many industrial quality checks. The critical value for a left‑tailed test at the 5% level is zα = –1.645. The decision rule is summarised in the table below.
我们采用显著性水平 α = 0.05，这在许多工业质量检查中很常见。左侧检验在 5% 水平下的临界值为 zα = –1.645。下表总结了决策规则。
| Significance level α | Critical value (one‑tailed) | Decision rule |
|---|---|---|
| 0.05 | –1.645 | Reject H₀ if Z < –1.645 |
If the computed z‑score falls in the critical region (i.e., Z < –1.645), we will reject the null hypothesis. Otherwise, we will not have sufficient evidence to reject the claim.
如果计算出的 z 得分落入拒绝域（即 Z < –1.645），我们就拒绝零假设。否则，我们将没有足够的证据否定该声称。
5. Computing the Observed Test Statistic | 计算观测检验统计量
Substituting the sample data into the formula gives:
将样本数据代入公式计算：
Zobs = (985 – 1000) / (50 / √50) = (–15) / (50 / 7.071) = –15 / 7.071 ≈ –2.12
Thus the observed z‑score is approximately –2.12. It is helpful to note that the standard error of the mean is σ/√n = 50/√50 = 7.071 hours, so the sample mean is about 2.12 standard errors below the hypothesised value.
因此，观测到的 z 得分约为 –2.12。注意到均值的标准误为 σ/√n = 50/√50 = 7.071 小时，因而样本均值比假设值低了约 2.12 个标准误。
6. Making a Decision: p‑value and Critical Value Approaches | 做出决策：p 值法与临界值法
Critical value approach: Since –2.12 < –1.645, the observed statistic lies inside the critical region. We therefore reject H₀ at the 5% significance level.
临界值法：由于 –2.12 < –1.645，观测统计量落在拒绝域内。因此我们在 5% 显著性水平下拒绝 H₀。
p‑value approach: The p‑value is the probability of obtaining a test statistic as extreme as, or more extreme than, the observed value under H₀. Here, p‑value = P(Z < –2.12) ≈ 0.0170. Because 0.017 < 0.05, we reject H₀. The p‑value quantifies the strength of evidence against the null; a small p‑value indicates that such an extreme result would be very unlikely if the claimed mean were true.
p 值法：p 值是在零假设下，获得与观测值同样极端或更极端检验统计量的概率。此处 p 值 = P(Z < –2.12) ≈ 0.0170。由于 0.017 < 0.05，我们拒绝 H₀。p 值量化了反对零假设的证据强度；一个很小的 p 值表明，如果声称的均值为真，出现如此极端结果的概率极低。
7. Constructing a 95% Confidence Interval | 构建 95% 置信区间
To complement the hypothesis test, we can build a 95% confidence interval for the true mean μ. Even though the test was one‑tailed, a two‑sided interval is often used to provide general information about the parameter. The formula for a symmetric z‑interval is:
为了对假设检验进行补充，我们可以为真实的均值 μ 构建 95% 置信区间。尽管检验是单侧的，但双侧区间常用来给出参数的一般信息。对称 z 区间的公式为：
X̄ ± zα/2 × (σ / √n)
With z0.025 = 1.96, the interval becomes 985 ± 1.96 × 7.071 = 985 ± 13.86, i.e. (971.14, 998.86) hours.
取 z0.025 = 1.96，计算得 985 ± 1.96 × 7.071 = 985 ± 13.86，即 (971.14, 998.86) 小时。
The interval does not contain the hypothesised value of 1000 hours, which is consistent with our decision to reject H₀. If 1000 were a plausible value for μ, we would expect it to lie within the interval. The upper bound being below 1000 further supports the conclusion that the true mean is likely less than the claim.
该区间未包含假设值 1000 小时，这与我们拒绝 H₀ 的决定一致。如果 1000 是 μ 的一个合理取值，那么它应当处于该区间内。上界低于 1000 进一步支持了真实均值很可能低于声称值的结论。
8. Interpreting the Results in Context | 在背景中解读结果
Statistically significant does not automatically mean practically important. Our analysis shows strong evidence that the true mean lifetime is below 1000 hours, but the estimated shortfall is only about 15 hours. Manufacturers and consumers must judge whether a 15‑hour difference (1.5% of the claimed value) matters for performance or cost. In many battery applications, such a small deviation may be acceptable, but the test signals that the production process may need monitoring.
统计显著并不自动意味着实际重要。我们的分析显示有充分证据表明真实平均寿命低于 1000 小时，但估计的缺口仅约 15 小时。制造商和消费者需要判断 15 小时的差异（占声称值的 1.5%）是否对性能或成本有影响。在许多电池应用中，如此小的偏差可能是可以接受的，但检验提示生产过程可能需要监控。
The confidence interval places μ somewhere between 971 and 999 hours with 95% confidence, providing a range of plausible values that can guide business decisions. Always contextualise numerical findings to avoid overstating the results.
置信区间以 95% 的置信水平将 μ 定位在 971 至 999 小时之间，这个合理取值区间可以指导商业决策。始终要将数值发现置于背景中，避免夸大结论。
9. Type I Error, Type II Error and Power | 第一类错误、第二类错误与功效
Any hypothesis test carries risk. A Type I error (α) would be rejecting the manufacturer’s claim when it is actually true—the probability is fixed at 0.05 by our choice of α. A more subtle concern is the Type II error (β): failing to reject H₀ when in fact the true mean is below 1000. The power of the test (1 – β) is its ability to detect a specific deviation from the null. Let’s compute β and power if the true mean were 990 hours.
任何假设检验都伴随风险。第一类错误（α）是指当制造商声称实际为真时拒绝该声称——由我们选择的 α 将概率固定在 0.05。更令人关注的是第二类错误（β）：即当真实均值实际上低于 1000 时未能拒绝 H₀。检验的功效（1 – β）是它检测特定偏离零假设的能力。如果真实均值为 990 小时，让我们计算 β 与功效。
Under μ = 990, the distribution of X̄ is N(990, 50²/50), so X̄ ~ N(990, 7.071²). The critical value for X̄ corresponding to z = –1.645 is:
在 μ = 990 下，X̄ 的分布为 N(990, 50²/50)，即 X̄ ~ N(990, 7.071²)。对应于 z = –1.645 的 X̄ 临界值为：
c = μ₀ – 1.645 × (σ / √n) = 1000 – 1.645 × 7.071 ≈ 1000 – 11.634 = 988.37 h
We reject H₀ if X̄ < 988.37. Now, if the true mean is 990, the probability of not rejecting (i.e., X̄ ≥ 988.37) is:
如果 X̄ < 988.37 我们就拒绝 H₀。现在，若真实均值为 990，未能拒绝的概率（即 X̄ ≥ 988.37）为：
β = P(X̄ ≥ 988.37 | μ = 990) = P(Z ≥ (988.37 – 990) / 7.071) = P(Z ≥ –0.23) ≈ 0.591
Thus the power is 1 – β ≈ 0.409. This relatively low power means that detecting a 10‑hour shortfall with 50 batteries is fairly difficult. The calculation illustrates the importance of sample size; a larger sample would reduce the standard error and improve power.
因此功效为 1 – β ≈ 0.409。这一相对较低的功效意味着用 50 节电池来检测 10 小时的短缺相当困难。该计算说明了样本量的重要性；更大的样本会降低标准误并提高功效。
10. Practical Recommendations and Exam Tips | 实践建议与考试技巧
Based on the case study, several practical lessons emerge:
- Always check whether σ is known or unknown; if unknown, use the t‑distribution.
- State the hypotheses in symbols and in words when the question requires it.
- Show the formula, substitution, and final value of the test statistic clearly.
- Link the conclusion to both the p‑value and the critical region—examiners look for complete reasoning.
- Remember that a confidence interval provides additional information; reference it to confirm the test decision.
- Comment on the practical significance, not just statistical significance.
- For power calculations, draw a diagram of the two sampling distributions to visualise β.
基于本案例，我们得出几条实用经验：
- 务必检查 σ 是否已知；若未知则使用 t 分布。
- 根据题目要求，用符号和文字陈述假设。
- 清晰地写出检验统计量的公式、代入过程和最终值。
- 将结论同时与 p 值和临界域联系起来——考官期望完整的推理过程。
- 记住置信区间可提供额外信息；引用它来确认检验结论。
- 不仅要评论统计显著性，还要讨论实际显著性。
- 对于功效计算，画出两个抽样分布的示意图以帮助理解 β。
Mastering the back‑and‑forth between testing and confidence intervals will equip you with a powerful framework for data‑driven decision‑making, both on the exam paper and beyond.
掌握检验与置信区间之间的相互配合，将为你提供在考试及未来数据驱动决策中的有力框架。
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