Hypothesis Testing | 假设检验

📚 Hypothesis Testing | 假设检验

Hypothesis testing is a core concept in A-Level CCEA Mathematics that allows statisticians to make inferences about population parameters based on sample data. It provides a structured framework for deciding whether observed evidence is strong enough to reject a stated belief. Mastering this topic is essential for tackling examination questions that involve binomial, Poisson, and normal distributions.

假设检验是 A-Level CCEA 数学中的一个核心概念，它让统计学家能够基于样本数据对总体参数做出推断。它为判断观察到的证据是否足够有力地推翻某个已有观点提供了一个结构化的框架。掌握这一主题对于解决涉及二项分布、泊松分布和正态分布的考试题目至关重要。

1. Introduction to Hypothesis Testing | 假设检验简介

A hypothesis test begins with a claim about a population, such as ‘the coin is fair’ or ‘the mean height has decreased’. We collect a sample, compute a test statistic, and decide whether the sample result is sufficiently unlikely under the original claim. If it is, we reject that claim in favour of an alternative.

假设检验从一个关于总体的主张开始，比如“硬币是公平的”或“平均身高下降了”。我们收集样本，计算检验统计量，并判断在原主张下样本结果是否足够不可能。如果是，我们就拒绝原主张，转而支持备择主张。

The formal process involves setting up two competing statements, choosing a significance level, determining the rejection region, and then interpreting the p‑value or comparing the test statistic to critical values. This logical procedure is used across every statistical distribution you will encounter in CCEA exams.

正式流程包括设立两个相互竞争的陈述，选择显著性水平，确定拒绝域，然后解读 p 值或将检验统计量与临界值进行比较。这个逻辑程序将贯穿你在 CCEA 考试中遇到的所有统计分布。

2. Null and Alternative Hypotheses | 零假设与备择假设

The null hypothesis, written as H₀, is the default position that there is no effect or no change. For a binomial test, H₀: p = 0.5 means ‘the probability of success is 0.5’. For a normal mean test, H₀: μ = 50 states that the population mean equals 50. We always assume H₀ is true unless the data provide strong evidence against it.

零假设，记作 H₀，是默认的立场，即没有效应或没有变化。对于二项检验，H₀: p = 0.5 表示“成功的概率为 0.5”。对于正态均值检验，H₀: μ = 50 表明总体均值为 50。除非数据提供了强有力的证据反对它，否则我们始终假定 H₀ 为真。

The alternative hypothesis, H₁ or Hₐ, is what we conclude if H₀ is rejected. It can be one‑sided, such as H₁: p > 0.5 or H₁: μ < 50, or two‑sided, e.g. H₁: p ≠ 0.5. The direction of H₁ determines whether the test is one‑tailed or two‑tailed.

备择假设，记作 H₁ 或 Hₐ，是当 H₀ 被拒绝后我们得出的结论。它可以是单边的，如 H₁: p > 0.5 或 H₁: μ < 50，也可以是双边的，例如 H₁: p ≠ 0.5。H₁ 的方向决定了检验是单尾还是双尾。

3. Significance Level and Critical Region | 显著性水平与临界域

The significance level, denoted by α, is the probability of rejecting H₀ when it is actually true. In CCEA exams the most common values are 5% (0.05) and 1% (0.01). Choosing a smaller α makes it harder to reject H₀, reducing the chance of a Type I error.

显著性水平，用 α 表示，是当 H₀ 实际为真时拒绝它的概率。在 CCEA 考试中最常见的取值是 5% (0.05) 和 1% (0.01)。选择较小的 α 会使拒绝 H₀ 变得更困难，从而减小发生第一类错误的机会。

The critical region (or rejection region) is the set of values of the test statistic for which we reject H₀. For a one‑tailed binomial test with H₁: p > 0.5 and α = 0.05, the critical region might be X ≥ 8 if n = 10. Every outcome in the critical region has a cumulative probability ≤ α under H₀.

临界域（或拒绝域）是检验统计量取值的集合，当它落入该区域时我们拒绝 H₀。对于 H₁: p > 0.5 且 α = 0.05 的单尾二项检验，若 n = 10，临界域可能是 X ≥ 8。在 H₀ 下，临界域中每一个结果的累积概率都不超过 α。

4. One-tailed vs Two-tailed Tests | 单尾与双尾检验

A one‑tailed test is used when the alternative hypothesis specifies a direction, for example ‘the new drug increases recovery rate’. The entire significance level α is placed in one tail of the distribution. The critical value is found so that the probability in that tail is as close as possible to α without exceeding it.

当备择假设指明了方向时，例如“新药提高了康复率”，就使用单尾检验。整个显著性水平 α 被放在分布的一个尾部。临界值的确定应使该尾部概率尽量接近 α 而不超过它。

In a two‑tailed test, H₁: p ≠ 0.5, we split α equally between the two tails, so each tail has α/2. Consequently, the critical region consists of two parts, usually the lower and upper extremes. Two‑tailed tests are more conservative because a larger deviation from H₀ is required to reach significance.

在双尾检验中，H₁: p ≠ 0.5，我们将 α 平分到两个尾部，因此每侧有 α/2。于是临界域由两部分组成，通常是下侧和上侧的极端值。双尾检验更为保守，因为需要偏离 H₀ 更远才能达到显著性。

CCEA exam questions often require you to decide whether the context implies a one‑tailed or two‑tailed test. Look for keywords like ‘increased’, ‘decreased’, ‘changed’, or ‘different’ to guide your choice.

CCEA 考试题目经常要求你根据上下文判断应采用单尾还是双尾检验。注意像“增加”、“减少”、“改变”或“不同”这样的关键词，以引导你的选择。

5. Test Statistic and p-value | 检验统计量与p值

The test statistic is a numerical summary calculated from the sample that measures how compatible the data are with H₀. For a binomial test, the test statistic is simply the observed number of successes, X ∼ B(n, p). For a normal mean test, we use the z‑statistic: z = (x̄ − μ₀) / (σ/√n).

检验统计量是从样本计算出来的一个数值摘要，衡量数据与 H₀ 的相容程度。对于二项检验，检验统计量就是观察到的成功次数 X ∼ B(n, p)。对于正态均值检验，我们使用 z 统计量：z = (x̄ − μ₀) / (σ/√n)。

The p‑value is the probability, under the null hypothesis, of obtaining a result at least as extreme as the observed one. If the p‑value ≤ α, we reject H₀. For example, in a binomial test with H₁: p > 0.3, observed X = 12 out of n = 25, the p‑value is P(X ≥ 12 | p = 0.3).

p 值是在零假设下，获得与观测结果同样极端或更极端结果的概率。如果 p 值 ≤ α，我们拒绝 H₀。例如，在 H₁: p > 0.3 的二项检验中，观测到 n = 25 时 X = 12，p 值就是 P(X ≥ 12 | p = 0.3)。

In CCEA mark schemes, both the critical‑value method and the p‑value method are accepted. Showing the p‑value explicitly alongside the significance level can make your reasoning clearer.

在 CCEA 的评分标准中，临界值法和 p 值法均被接受。在显著性水平旁边明确写出 p 值可以使你的推理更加清晰。

6. Type I and Type II Errors | 第一类错误与第二类错误

A Type I error occurs when H₀ is true but we reject it. The probability of a Type I error is exactly the significance level α. If a test is carried out at the 5% level, there is a 5% risk of falsely claiming an effect when none exists.

第一类错误发生在 H₀ 为真但我们却拒绝了它的情况下。发生第一类错误的概率恰好就是显著性水平 α。如果在 5% 的水平上进行检验，就有 5% 的风险在没有效应时错误地宣称有效应。

A Type II error happens when H₀ is false but we fail to reject it. The probability of a Type II error is denoted by β, and it depends on the true value of the parameter. Increasing the sample size is the most effective way to reduce β without raising α.

第二类错误发生在 H₀ 为假但我们未能拒绝它的情况下。第二类错误的概率用 β 表示，它取决于参数的真实值。增大样本量是在不提高 α 的前提下减小 β 最有效的方法。

Exam questions may ask you to explain these errors in context, for instance: ‘Describe a Type I error in the context of testing whether a machine produces defective items correctly.’ Clear contextual descriptions are expected.

考试题目可能要求你结合具体场景解释这些错误，例如：“在检验一台机器是否正常生产缺陷产品的背景下描述第一类错误。” 清晰的语境描述是得分所必需的。

7. Hypothesis Testing with Binomial Distribution | 二项分布假设检验

When the population consists of independent trials with two outcomes, we model the number of successes using X ∼ B(n, p). The null hypothesis usually states a specific value for p, e.g. H₀: p = 0.4. The test uses individual binomial probabilities or cumulative tables.

当总体由每次有两种结果的独立试验组成时，我们用 X ∼ B(n, p) 对成功次数建模。零假设通常会指明 p 的具体取值，例如 H₀: p = 0.4。检验使用个别的二项概率或累积表格。

Suppose a manufacturer claims that at most 10% of items are defective. We test H₀: p = 0.1 against H₁: p > 0.1. A sample of 20 items yields 5 defectives. The p‑value is P(X ≥ 5 | p = 0.1). If this probability is less than 0.05, we reject H₀ and conclude the defect rate has increased.

假设某制造商声称次品率最多为 10%。我们检验 H₀: p = 0.1 对 H₁: p > 0.1。一个包含 20 件产品的样本中有 5 件次品。p 值为 P(X ≥ 5 | p = 0.1)。如果这个概率小于 0.05，我们就拒绝 H₀，并得出结论：次品率已经上升。

CCEA questions frequently provide cumulative binomial probability tables. You must be able to find P(X ≤ k) or P(X ≥ k) correctly, remembering that P(X ≥ k) = 1 − P(X ≤ k − 1).

CCEA 的试题经常提供累积二项分布概率表。你必须能正确地找出 P(X ≤ k) 或 P(X ≥ k)，并记住 P(X ≥ k) = 1 − P(X ≤ k − 1)。

8. Hypothesis Testing with Poisson Distribution | 泊松分布假设检验

When events occur randomly and independently at a constant average rate, the Poisson distribution is appropriate. The test statistic is Y ∼ Po(λ). We test hypotheses about the population mean λ, for example H₀: λ = 3 against H₁: λ < 3.

当事件随机、独立地以恒定平均速率发生时，泊松分布是合适的。检验统计量为 Y ∼ Po(λ)。我们检验关于总体均值 λ 的假设，例如 H₀: λ = 3 对 H₁: λ < 3。

For a lower‑tail test, the p‑value is P(Y ≤ observed | H₀). For an upper‑tail test, it is P(Y ≥ observed | H₀). Because the Poisson distribution is discrete, the actual significance level may be slightly less than the nominal α. This is acceptable and should be noted.

对于下尾检验，p 值为 P(Y ≤ 观测值 | H₀)。对于上尾检验，p 值为 P(Y ≥ 观测值 | H₀)。由于泊松分布是离散的，实际显著性水平可能略低于名义上的 α。这是可以接受的，并且应予以说明。

Imagine a call centre claims to receive 4 calls per hour on average. A monitoring period of 2 hours might use Y ∼ Po(8) under H₀. If only 2 calls are recorded, the p‑value P(Y ≤ 2 | λ = 8) is extremely small, leading to rejection of H₀.

设想一个呼叫中心声称平均每小时接到 4 通电话。在 2 小时的监测期间中，H₀ 下可以使用 Y ∼ Po(8)。如果只记录到 2 通电话，p 值 P(Y ≤ 2 | λ = 8) 极小，从而导致拒绝 H₀。

9. Hypothesis Testing for Normal Mean (Variance Known) | 正态分布均值的假设检验（方差已知）

When testing the mean of a normally distributed population and the population variance σ² is known, the test statistic follows a standard normal distribution: Z = (x̄ − μ₀) / (σ/√n) ∼ N(0, 1). This is the foundation of z‑tests.

当检验正态分布总体的均值且总体方差 σ² 已知时，检验统计量服从标准正态分布：Z = (x̄ − μ₀) / (σ/√n) ∼ N(0, 1)。这是 z 检验的基础。

For a two‑tailed test with α = 0.05, the critical values are z = ±1.96. If the calculated |z| > 1.96, we reject H₀. For one‑tailed tests, the critical value is 1.645 (upper tail) or −1.645 (lower tail) at the 5% level.

对于 α = 0.05 的双尾检验，临界值为 z = ±1.96。如果计算得到的 |z| > 1.96，我们拒绝 H₀。对于单尾检验，在 5% 水平下临界值为 1.645（上尾）或 −1.645（下尾）。

In CCEA exams you may be given the sample mean and told to assume the central limit theorem applies for large samples, even if the population distribution is unknown. Always check whether σ is given or must be estimated; if only a sample standard deviation is available, a t‑test may be needed, but at A‑Level it is usually specified.

在 CCEA 考试中，你可能得到样本均值，并被告知对于大样本可应用中心极限定理，即使总体分布未知。始终要检查 σ 是已知还是必须估计；如果只提供了样本标准差，可能需要 t 检验，但在 A‑Level 阶段通常会有明确说明。

10. Critical Values and Using Tables | 临界值与表的使用

Statistical tables are your primary tool in a CCEA hypothesis testing exam. For binomial and Poisson distributions, cumulative tables give P(X ≤ k). You must identify the correct row and column for n, p, or λ. Always read the table carefully to avoid picking the wrong tail.

统计表是你在 CCEA 假设检验考试中的主要工具。对于二项分布和泊松分布，累积表格给出 P(X ≤ k)。你必须根据 n、p 或 λ 确定正确的行和列。务必仔细阅读表格，以免选错尾部。

For the normal distribution, the table gives Φ(z) = P(Z ≤ z). To find an upper‑tail probability, use P(Z > z) = 1 − Φ(z). To find a critical z‑value for α = 0.05, locate the z that gives Φ(z) = 0.95 or 0.975 for two‑tailed tests. Some papers provide percentage points tables directly.

对于正态分布，表格给出 Φ(z) = P(Z ≤ z)。要获得上尾概率，使用 P(Z > z) = 1 − Φ(z)。要找到 α = 0.05 对应的临界 z 值，找到使 Φ(z) = 0.95 或（对于双尾检验）0.975 的 z 值。有些试卷直接提供百分比点表格。

In questions where the exact significance level cannot be achieved due to discreteness, you must state the actual significance level of your critical region. This is a common mark in CCEA mark schemes.

在由于离散性而无法达到精确显著性水平的题目中，你必须给出你的临界域的实际显著性水平。这是 CCEA 评分标准中常见的一个得分点。

11. Interpreting Results and Drawing Conclusions | 结果解读与得出结论

Your conclusion must be written in the context of the problem, not just ‘reject H₀’. Use phrases like ‘There is sufficient evidence at the 5% level to suggest the new treatment is effective’ or ‘We do not reject H₀. The data do not support the claim of a decrease.’

你的结论必须结合问题背景来写，不能只说“拒绝 H₀”。使用诸如“在 5% 水平下有充分证据表明新疗法有效”或“我们不拒绝 H₀，数据不支持下降的说法”这样的表述。

If the test statistic falls just outside the critical region, do not claim the result is ‘almost significant’—exam boards expect a clear accept/reject decision based on the predetermined significance level. A p‑value just above 0.05 does not mean the null hypothesis is true; it simply means the evidence is insufficient to reject it.

如果检验统计量恰好落在临界域之外，不要声称结果是“几乎显著”——考试委员会希望基于预设的显著性水平给出明确的接受或拒绝决定。p 值略高于 0.05 并不意味着零假设为真；它仅仅意味着证据不足以拒绝它。

Double‑check that you have stated the hypotheses in symbols and in words where required, as CCEA sometimes asks for both. Always use the exact wording ‘there is (or is not) sufficient evidence to reject H₀’.

请仔细检查你是否按要求用符号和文字陈述了假设，因为 CCEA 有时会同时要求这两者。始终使用精确的措辞“有（或没有）充分证据拒绝 H₀”。

12. Exam Tips and Common Pitfalls | 考试技巧与常见错误

Always define the random variable and state the distribution under H₀ before calculations. E.g. ‘Let X be the number of successes; X ∼ B(20, 0.3) under H₀.’
在进行计算前，务必定义随机变量并给出 H₀ 下的分布。例如“设 X 为成功次数；在 H₀ 下 X ∼ B(20, 0.3)”。

Write H₀ and H₁ clearly using correct notation. A common mistake is switching the inequality direction: H₁ must match the suspicion in the question.
使用正确的符号清晰地写出 H₀ 和 H₁。一个常见错误是弄错了不等号的方向：H₁ 必须与题目中的怀疑相符。

When finding P(X ≥ k), remember to subtract the lower tail correctly: P(X ≥ k) = 1 − P(X ≤ k − 1). Many errors arise from forgetting the ‘−1’.
在求 P(X ≥ k) 时，记住要正确减去下尾：P(X ≥ k) = 1 − P(X ≤ k − 1)。许多错误就是因为忘了“−1”而产生的。

For a two‑tailed binomial test, find the critical values such that P(X ≤ c₁) ≤ α/2 and P(X ≥ c₂) ≤ α/2. Do not simply double a one‑tailed p‑value unless the distribution is symmetric.
对于双尾二项检验，找出临界值 c₁ 和 c₂，使得 P(X ≤ c₁) ≤ α/2 且 P(X ≥ c₂) ≤ α/2。除非分布是对称的，否则不要简单地将单尾 p 值翻倍。

Check that the sample size is large enough when using a normal approximation, though CCEA will usually guide you if that is required. If using a continuity correction, apply it carefully.
当使用正态近似时，要检查样本量是否足够大，不过如果需要，CCEA 通常会给出指导。如果使用连续性校正，要谨慎应用。

Finally, always relate your conclusion back to the original claim. Marks are awarded for contextual interpretation, not just numerical results.
最后，务必将你的结论与原始主张联系起来。评分是根据情境化的解读来给分的，而不仅仅是数值结果。

Published by TutorHao | Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

Hypothesis Testing | 假设检验