📚 Hypothesis Testing | 假设检验
Hypothesis testing is a powerful statistical method used to decide whether there is enough evidence to support a particular belief about a population parameter. It forms a core topic of the Edexcel A-Level Mathematics specification, appearing in both pure statistical contexts and applied problems involving binomial distributions, normal distributions, and correlation coefficients. Through this article, you will learn how to formulate null and alternative hypotheses, interpret significance levels, calculate p‑values, determine critical regions, and appreciate the risks of Type I and Type II errors. Clear worked examples and exam‑focused guidance are provided to build your confidence for Paper 3.
假设检验是一种强大的统计方法，用于判断是否有足够证据支持关于总体参数的特定信念。它是 Edexcel A-Level 数学大纲的核心内容，涉及二项分布、正态分布以及相关系数等应用问题。通过本文，你将学会如何设定原假设和备择假设、解释显著性水平、计算 p 值、确定临界区域，并理解第一类和第二类错误的风险。文中提供了清晰的例题和考试导向的指导，以增强你应对试卷三的信心。
1. Introduction to Hypothesis Testing | 假设检验简介
A statistical hypothesis is a statement about a population parameter, such as a probability of success p, a population mean μ, or the population correlation coefficient ρ. Hypothesis testing provides a formal framework to evaluate whether sample data contradicts a stated assumption. The process begins by temporarily assuming that a given claim (the null hypothesis) is true. If the observed sample result is extremely unlikely under this assumption, we reject the null hypothesis in favour of an alternative statement.
统计假设是关于总体参数的一个陈述，例如成功概率 p、总体均值 μ 或总体相关系数 ρ。假设检验提供了一个正式的框架，用于评估样本数据是否与所陈述的假设相矛盾。这个过程首先暂时假定某个主张（原假设）为真，如果在该假设下观察到的样本结果极其不可能发生，我们就拒绝原假设而支持备择陈述。
The reasoning behind hypothesis testing mirrors a court trial: the defendant is presumed innocent (null hypothesis) until the evidence is so strong that the jury rejects this presumption. In statistics, “strong evidence” is quantified by probabilities. Edexcel A-Level requires you to perform such tests for binomial probabilities, normal means, and correlation coefficients using critical values, p-values, or confidence interval arguments.
假设检验背后的推理类似于法庭审判：被告被假定无罪（原假设），直到证据足够有力，陪审团才推翻这一假定。在统计学中，“有力证据”是通过概率量化的。Edexcel A-Level 要求你能对二项概率、正态均值和相关系数进行这样的检验，可以使用临界值、p 值或置信区间的论点。
2. Setting Up Hypotheses | 设定假设
alternative hypothesis, denoted H₁ or Hₐ, represents what we seek evidence for. For a binomial test with parameter p, typical structures are H₀ : p = 0.4, H₁ : p ≠ 0.4 (two‑tailed), H₁ : p > 0.4 (right‑tailed), or H₁ : p < 0.4 (left‑ relation coefficient ρ, the null is usually H₀ : ρ = 0, indicating no linear relationship.
每个假设检验都涉及两个对立的陈述。原假设，记作 H₀，代表现状或怀疑立场。备择假设，记作 H₁ 或 Hₐ，代表我们要寻找证据支持的主张。对于参数为 p 的二项检验，典型结构是 H₀ : p = 0.4，H₁ : p ≠ 0.4（双尾），H₁ : p > 0.4（右尾），或 H₁ : p < 0.4（左尾）。对于正态均值 μ，我们写为 H₀ : μ = 50 对 H₁ : μ ≠ 50、μ > 50 或 μ < 50。在检验总体相关系数 ρ 时，原假设通常为 H₀ : ρ = 0，表示没有线性关系。
ative reflects the direction of interest. In Edexcel examinations, you will often be given a question that implies a direction; you must then decide whether a one‑tailed or two‑tailed test is appropriate before proceeding.
原假设必须包含等号形式（=、≤ 或 ≥），而备择假设反映感兴趣的方向。在 Edexcel 考试中，题目通常会暗示一个方向，你必须在进行检验之前判断应该使用单尾还是双尾检验。
3. Significance Level and Rejection Region | 显著性水平与拒绝域
treme the test statistic must be before we doubt the null hypothesis. Common choices are 5% (α = 0.05) or 1% (α = 0.01). For instance, if α = 0.05, we are willing to accept a 5% chance of making a Type I error – rejecting a true null hypothesis.
显著性水平 α 是当 H₀ 为真时拒绝 H₀ 的概率。它作为一个门槛，决定检验统计量要多么极端我们才会怀疑原假设。常见的选择有 5%（α = 0.05）或 1%（α = 0.01）。例如，若 α = 0.05，我们愿意接受 5% 的风险犯第一类错误——即拒绝了一个正确的原假设。
The rejection region (or critical region) is the set of values of the test statistic that leads to the rejection of H₀. Its size is determined by α. For a binomial test with H₁ : p > 0.3, the critical region lies in the upper tail of the binomial distribution; for a two‑tailed test on a normal mean, the critical region is split equally between the two tails. Any test statistic falling inside the rejection region results in a “statistically significant” outcome.
拒绝域（或临界区域）是导致拒绝 H₀ 的检验统计量的取值集合，其大小由 α 决定。如果 H₁ : p > 0.3，则二项检验的临界区域落在二项分布的上尾；对正态均值进行双尾检验时，拒绝域平分在两个尾部。任何落入拒绝域的检验统计量都会得出“统计显著”的结论。
4. One-tailed vs Two-tailed Tests | 单尾与双尾检验
A one‑ p > 0.25. In this case, we reject H₀ only if the test statistic is unusually large. A two‑ is then divided between both tails. The choice directly affects the critical values and p‑value calculations; a two‑tailed test at α = 0.05 places 2.5% in each tail.
当备择假设指明了方向时，比如 H₁ : p > 0.25，应使用单尾检验。此时只有当检验统计量异常大时才拒绝 H₀。双尾检验适用于我们只关心是否存在差异的情形，例如 H₁ : p ≠ 0.25，此时拒绝域分布在两个尾部。这一选择直接影响临界值和 p 值的计算；在 α = 0.05 的双尾检验中，每尾各分配 2.5%。
In a binomial context with X ~ B(n, p), a right‑tailed test finds the smallest r such that P(X ≥ r) ≤ α. For a left‑tailed test we look for the largest r with P(X ≤ r) ≤ α. In a two‑ ates otherwise. With normal distribution tests, symmetry about the mean simplifies the two‑tailed case: the critical z‑values are ±z(α/2).
在二项分布的背景下，X ~ B(n, p)，右尾检验寻找最小的 r 使得 P(X ≥ r) ≤ α。左尾检验寻找最大的 r 使得 P(X ≤ r) ≤ α。在二项双尾检验中，通常将显著性水平对半分给每个尾部，除非题目另有说明。对于正态分布检验，均值两侧的对称性简化了双尾情形：临界 z 值为 ±z(α/2)。
5. The p-value Approach | p值方法
The p‑ a test statistic at least as extreme as the one observed, assuming H₀ is true. For a right‑tailed binomial test, the p‑value is P(X ≥ observed value). If this p‑value is less than or equal to α, we reject H₀. Otherwise, we fail to reject H₀. The p‑value approach provides a continuous measure of evidence against the null; a very small p‑value (e.g. p < 0.001) indicates strong evidence against H₀.
p 值是在 H₀ 成立的条件下，获得至少与观测值同样极端的检验统计量的概率。对于右尾二项检验，p 值是 P(X ≥ 观测值)。如果这个 p 值 ≤ α，我们就拒绝 H₀，否则不能拒绝 H₀。p 值方法提供了一个连续衡量反对原假设的证据强度；非常小的 p 值（如 p < 0.001）表明反对 H₀ 的证据很强。
To compute a p‑ a normal mean test with known variance, we first standardise the sample mean to obtain a z‑score. For H₁ : μ > μ₀, the p‑value is P(Z ≥ zₒ𝒃𝒔). For a two‑tailed test, the p‑value is 2 × P(Z ≥ |zₒ𝒃𝒔|). Many Edexcel questions expect you to compare the p‑ e level rather than directly finding critical regions, so this technique must be mastered.
对于已知方差的正态均值检验，我们先将样本均值标准化得到 z 分数。如果 H₁ : μ > μ₀，p 值为 P(Z ≥ zₒ𝒃𝒔)。双尾检验的 p 值是 2 × P(Z ≥ |zₒ𝒃𝒔|)。许多 Edexcel 题目要求你将 p 值与给定的显著性水平进行比较，而不是直接寻找临界区域，因此这一技巧必须掌握。
6. Critical Region Method | 临界区域法
mial test, you use cumulative probabilities to find the boundary values. For example, if X ~ X ≥ 15 because 0.0207 ≤ 0.05. If the test statistic falls in this region, H₀ is rejected.
临界区域法要求在观测数据之前就确定好导致拒绝 H₀ 的取值集合。在二项检验中，利用累积概率找出边界值。例如，若 X ~ B(20, 0.5) 且 H₁ : p > 0.5，α = 0.05，计算 P(X ≥ 14) = 1 − P(X ≤ 13) = 0.0577，P(X ≥ 15) = 0.0207。故临界区域为 X ≥ 15，因为 0.0207 ≤ 0.05。若检验统计量落于此区域，则拒绝 H₀。
rmal distribution test of H₀ : μ = μ₀ with known variance σ², the test statistic is Z = (x̄ − μ₀) / (σ/√n). The critical region at α = 0.05 for a two‑tailed test is Z ≤ −1.96 or Z ≥ 1.96. The observed sample mean is converted to a z‑ ers, you are often asked to state the critical region clearly and interpret it in context.
对已知方差 σ² 的正态分布检验 H₀ : μ = μ₀，检验统计量为 Z = (x̄ − μ₀) / (σ/√n)。在 α = 0.05 的双尾检验中，临界区域为 Z ≤ −1.96 或 Z ≥ 1.96。将样本均值转换为 z 值后即可立即做出判断。在 Edexcel 试卷中，经常要求清晰地写出临界区域并结合实际情境进行解释。
7. Type I and Type II Errors | 第一类和第二类错误
A Type I error occurs when H₀ is true but the test rejects it. The probability of a Type I error is exactly the significance level α. A Type II error occurs when H₀ is false but the test fails to reject it; its probability is denoted β. In an ideal test, both α and β would be small, but decreasing α typically increases β for a fixed sample size. Understanding these errors helps you appreciate the limitations of a hypothesis test.
第一类错误发生在 H₀ 为真却被错误拒绝时，其概率恰好为显著性水平 α。第二类错误发生在 H₀ 为假却未能拒绝时，概率记作 β。在理想的检验中，α 和 β 都应很小，但在固定样本量下，降低 α 通常会增大 β。理解这些错误有助于你认识假设检验的局限性。
In courtroom terms, a Type I error is like convicting an innocent person, while a Type II error is like acquitting a guilty one. Edexcel questions may ask you to identify which error is more serious in a given context, or to calculate β for a specific alternative value using the distribution of the test statistic under H₁. Power of a test, defined as 1 − β, is the probability of correctly rejecting a false H₀.
用法庭术语来说，第一类错误就像将无辜的人定罪，而第二类错误就像将有罪的人无罪释放。Edexcel 的考题可能会让你结合具体情境判断哪种错误的后果更严重，或要求你在特定备择假设值下计算 β，其中需使用 H₁ 下检验统计量的分布。检验的功效定义为 1 − β，即正确拒绝错误 H₀ 的概率。
8. Hypothesis Testing for Binomial Distribution | 二项分布的假设检验
from a sample of n independent trials. The null hypothesis states a value for p, such as H₀ : p = 0.6. Depending on the alternative, you compute tail probabilities using the binomial distribution formula or cumulative tables. With small n, exact probabilities are used; for large n, a normal approximation may be employed, though the specification emphasises direct binomial calculations.
二项假设检验是 Edexcel A-Level 统计大纲的基石。检验统计量为来自 n 次独立试验的成功次数 X。原假设给出 p 的取值，如 H₀ : p = 0.6。根据备择假设，使用二项分布公式或累积表计算尾部概率。当 n 较小时采用精确概率；对于大 n，可以使用正态近似，但考纲强调直接进行二项计算。
A typical question: “A coin is tossed 25 times and 17 heads are observed. Test at the 5% significance level whether the coin is biased in favour of heads.” Here H₀ : p = 0.5, H₁ : p > 0.5. Under H₀, X ~ B(25, 0.5). The p‑ eject H₀; the data do not support the claim that the coin is biased towards heads at the 5% level.
常见题型：“一枚硬币抛掷 25 次，观察到 17 次正面朝上。在 5% 显著性水平下检验该硬币是否偏向正面。” 此时 H₀ : p = 0.5，H₁ : p > 0.5。在 H₀ 下 X ~ B(25, 0.5)，p 值为 P(X ≥ 17) = 1 − P(X ≤ 16)，查表得 0.0539。因为 0.0539 > 0.05，没有足够证据拒绝 H₀；在 5% 水平下，数据不支持硬币偏向正面的说法。
9. Hypothesis Testing for Normal Distribution (Known Variance) | 正态分布（已知方差）的假设检验
When testing a population mean μ with known variance σ², the sample mean x̄ is normally distributed: x̄ ~ N(μ, σ²/n). The test statistic Z = (x̄ − μ₀) / (σ/√n H₀ : μ = μ₀. The critical values come from the standard normal tables (e.g. ±1.96 for a 5% two‑ ation is normal or the sample size is large enough for the Central Limit Theorem to apply.
当已知方差 σ² 而检验总体均值 μ 时，样本均值 x̄ 服从正态分布：x̄ ~ N(μ, σ²/n)。在 H₀ : μ = μ₀ 下，检验统计量 Z = (x̄ − μ₀) / (σ/√n) 服从标准正态分布。临界值来自标准正态表（例如双尾 5% 检验对应 ±1.96）。该检验假定总体为正态或样本容量足够大，以使中心极限定理适用。
0 packets, the mean weight is 503 g, with a known standard deviation of 8 g. Test at the 1% level whether the mean fill has changed. α = 0.01, two‑tailed critical values are ±2.576. The test statistic is Z = (503 − 500) / (8/√50) = 2.651. Since 2.651 > 2.576, H₀ is rejected; there is evidence at the 1% significance level that the mean fill differs from 500 g.
例如，一台机器包装标称 500 g 的袋子。抽取 50 包的样本称得均重 503 g，已知标准差为 8 g。在 1% 显著性水平下检验均值是否发生变化。α = 0.01，双尾临界值 ±2.576。检验统计量 Z = (503 − 500) / (8/√50) = 2.651，因为 2.651 > 2.576，拒绝 H₀；在 1% 显著性水平下有证据表明平均填充量与 500 g 不同。
If the population variance is unknown, the t‑distribution is used, but Edexcel A-Level Mathematics (not Further Maths) typically restricts normal hypothesis testing to the known variance case, or provides the variance. Always check the question: if σ is given, use the z‑test.
如果总体方差未知，则使用 t 分布，但 Edexcel A-Level 数学（非进阶数学）通常将正态假设检验限制在已知方差的情形，或会在题目中给出方差。务必审题：只要给出了 σ，就使用 z 检验。
10. Testing Correlation Coefficient | 相关系数的假设检验
there is evidence of a linear relationship, i.e. H₀ : ρ = 0 against H₁ : ρ ≠ 0, ρ > 0, or ρ < 0. The test statistic is r itself, but its distribution under H₀ is not normal for small n. For the Edexcel A-Level, you are provided with a critical value table for r at various significance levels and sample sizes; you compare |r| with the tabulated value.
对于双变量样本数据，积矩相关系数 r 是总体相关系数 ρ 的估计。我们常希望检验是否存在线性关系的证据，即 H₀ : ρ = 0 对 H₁ : ρ ≠ 0、ρ > 0 或 ρ < 0。检验统计量就是 r 本身，但在 H₀ 下，对于小样本 r 的分布不是正态的。在 Edexcel A-Level 考试中，会提供一张不同显著性水平和样本容量下的 r 临界值表；你需要将 |r| 与表中的临界值进行比较。
m for this test are n − 2. As n increases, the critical value decreases. If the calculated |r| conclude there is statistically significant correlation. For example, with n = 12, the two‑tailed 5% critical value from the table is approximately 0.576. If r = 0.62, |r| > 0.576, so H₀ is rejected; there is evidence of a non‑zero correlation in the population.
该检验的自由度为 n − 2。随着 n 增加，临界值减小。若计算出的 |r| 超过了临界值，则拒绝 H₀，得出存在统计显著相关性的结论。例如，当 n = 12 时，双尾 5% 临界值约为 0.576。若 r = 0.62，由于 |r| > 0.576，拒绝 H₀；有证据表明总体的相关系数非零。
Edexcel questions may provide the critical value directly. If not, you must select the correct table entry based on n and whether the test is one‑ or two‑tailed. Note that the test is valid only if the underlying variables are approximately bivariate normal, but exam questions will state this or imply it through context.
Edexcel 的题目可能会直接给出临界值，否则你必须根据 n 以及是单尾还是双尾检验选择正确的表内值。注意，该检验仅在变量近似服从二元正态分布时有效，但考题通常会说明或通过背景暗示这一点。
11. Common Pitfalls and Tips | 常见错误与技巧
• or correlation tests, H₀ is almost always ρ = 0, and a positive expected association gives H₁ : ρ > 0.
• Forgetting to state the distribution: In binomial tests, always write “Assume H₀ is true, X ~ B(n, p)”. In normal tests, “Under H₀, Z ~ N(0, 1)”. Specifying this earns method marks.
• Using a critical region incorrectly for discrete distributions: Because the binomial is discrete, the actual significance level may be less than α. State the nominal significance level and the actual probability when asked.
• Mixing up tails: Double‑check whether the question implies a two‑tailed test (e.g. “changed”, “different”) or a one‑tailed test (e.g. “increased”, “higher”).
• Unit conversions: In normal tests, ensure the units of x̄, μ, and σ are consistent.
• 混淆 H₀ 和 H₁：始终将包含等号的陈述作为原假设。备择假设应与调查问题的措辞相匹配。对于相关性检验，H₀ 几乎总是 ρ = 0，若预计正相关则 H₁ : ρ > 0。
• 忘记标明分布：二项检验中一定要写“假定 H₀ 成立，X ~ B(n, p)”；正态检验中写“在 H₀ 下，Z ~ N(0, 1)”。写明分布能获得方法分。
• 对离散分布错误使用临界区域：二项分布是离散的，实际显著性水平可能低于 α。若被问到，应说明名义显著性水平与实际概率。
• 混淆尾端：仔细检查问题是隐含双尾检验（如“变化”、“不同”）还是单尾检验（如“增加”、“更高”）。
• 单位换算：在正态检验中确保 x̄、μ 和 σ 的单位一致。
12. Summary | 总结
Hypothesis testing is a systematic way to draw conclusions from sample data. By defining clear hypotheses, choosing an appropriate significance level, and calculating a p‑value or comparing against the critical region, you can make informed decisions about population parameters. Edexcel A-Level requires fluency in binomial tests, normal tests with known variance, and product moment correlation tests. Practise interpreting the results in context, and be precise with your notation and conclusions. Remember: “reject H₀” is a strong statement, while “fail to reject H₀” acknowledges that the evidence is insufficient but does not prove the null is true.
假设检验是一种从样本数据中得出结论的系统方法。通过定义明确的假设、选择合适的显著性水平、计算 p 值或与临界区域比较，你可以对总体参数做出有依据的判断。Edexcel A-Level 要求熟练掌握二项检验、已知方差的正态检验，以及积矩相关系数检验。平时要练习如何在情境中解释结果，并确保符号和结论的准确。记住：“拒绝 H₀”是一个强有力的陈述，而“不能拒绝 H₀”承认证据不足，但并不证明原假设为真。
Published by TutorHao | Mathematics Revision Series | aleveler.com
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