A-Level Statistics: The Complete Hypothesis Testing Guide — A-Level 数学统计：假设检验完全指南

一、假设检验的本质：从”猜测”到”证据”的统计推理 | The Nature of Hypothesis Testing: From Guesswork to Statistical Evidence

假设检验是 A-Level 统计学（AQA 国际大纲 9660 MA04 单元）中最核心的推理工具。它的基本问题是：当我们观察到一组数据时，这个结果究竟只是随机波动，还是背后真的存在某种规律？例如，一家奶茶店声称自家大杯奶茶平均容量是 500 ml，你随机买了 40 杯称量，发现平均只有 495 ml。这 5 ml 的差距，是抽样碰巧偏小，还是店家真的缺斤少两？假设检验就为回答这类问题提供了一套严格的数学程序。

Hypothesis testing is the most important inferential tool in A-Level Statistics (AQA International Syllabus 9660, MA04 unit). Its fundamental question is: when we observe a set of data, is the result merely random fluctuation, or does a real pattern lie behind it? For example, a bubble tea shop claims its large cups contain 500 ml on average. You buy 40 cups at random and weigh them, finding an average of only 495 ml. Is that 5 ml gap just a small sampling fluctuation, or is the shop really short-changing customers? Hypothesis testing provides a rigorous mathematical procedure for answering exactly this kind of question.

整套方法的核心思想是”先假设，再检验”。我们先把一个需要质疑的陈述当作”零假设”（记为 H₀），同时提出一个与之对立的”备择假设”（记为 H₁）。然后计算：如果零假设真的成立，那么观察到当前数据（或更极端的数据）的概率有多大？如果这个概率小到令人难以置信，我们就认为数据提供了反对零假设的强有力证据，从而拒绝它。

The core idea of the whole method is “assume first, then test.” We first treat a claim that needs questioning as the “null hypothesis” (denoted H0), and propose an opposing “alternative hypothesis” (denoted H1). We then calculate: if the null hypothesis is really true, how likely is it to observe the current data, or data even more extreme? If this probability is so small that it strains belief, we take the data as strong evidence against the null hypothesis and reject it.

需要注意的是，假设检验永远无法”证明”某个假设为真。它只提供两种结论：拒绝零假设，或没有足够证据拒绝零假设。这种表述上的严谨性是考试评分的重要依据，也是初学者最容易丢分的地方。

Note that hypothesis testing can never “prove” that a hypothesis is true. It only offers two conclusions: reject the null hypothesis, or do not have enough evidence to reject it. This precision of wording is an important basis for exam marking, and it is also where beginners lose marks most easily.

二、零假设与备择假设：符号、写法与判定规则 | Null and Alternative Hypotheses: Notation, Wording and Decision Rules

零假设 H₀ 永远包含等号。它描述的是”现状”或”原声称”：总体参数等于某个具体数值。例如检验奶茶店平均容量的声称时，H₀: μ = 500。备择假设 H₁ 描述的是我们怀疑的”另一面”，它可以是单侧的（μ < 500 或 μ > 500），也可以是双侧的（μ ≠ 500）。

The null hypothesis H0 always contains an equals sign. It describes the “status quo” or the “original claim”: the population parameter equals a specific value. For example, when testing the bubble tea shop’s claim about mean volume, H0: μ = 500. The alternative hypothesis H1 describes the “other side” we suspect; it can be one-sided (μ < 500 or μ > 500) or two-sided (μ ≠ 500).

正确的写法是考试的基本功。H₀ 和 H₁ 必须使用总体参数（μ、p、σ），而不是样本统计量（x̄、p̂）。常见错误是把 H₁ 写成 x̄ < 495，这是概念性错误：假设检验针对的是总体，样本均值只是一个观测值。此外，H₀ 和 H₁ 必须穷尽所有可能，且互不重叠。

Correct notation is basic exam craft. H0 and H1 must use population parameters (μ, p, σ), not sample statistics (x̄, p̂). A common error is writing H1: x̄ < 495; this is a conceptual mistake: hypothesis testing concerns the population, and the sample mean is just one observation. In addition, H0 and H1 must cover all possibilities and must not overlap.

判定方向取决于问题的措辞。”是否低于””是否下降””是否减少”对应单侧检验；”是否改变””是否等于””是否不同”对应双侧检验。读题时先圈出这些关键词，再决定 H₁ 的方向，这是标准化答题的第一步。

The direction of the test is determined by the wording of the question. “Is it lower”, “has it decreased”, “has it been reduced” correspond to one-tailed tests; “has it changed”, “is it equal to”, “is it different” correspond to two-tailed tests. When reading a question, first circle these keywords, then decide the direction of H1; this is the first step of a standardised answer.

三、显著性水平与 p 值：显著性到底意味着什么 | Significance Levels and p-Values: What “Significant” Actually Means

显著性水平 α（通常取 0.05 或 0.01）是一个预先设定的概率阈值，表示我们愿意承受的”冤枉”风险：即使零假设为真，我们仍然可能错误地拒绝它，这个错误的概率上限就是 α。5% 的显著性水平意味着：如果 H₀ 为真，我们允许自己大约每 20 次检验中错误拒绝 1 次。

The significance level α (usually 0.05 or 0.01) is a pre-set probability threshold that represents the risk of a “false accusation” we are willing to bear: even when the null hypothesis is true, we may still wrongly reject it, and the upper bound of this error probability is α. A 5% significance level means: if H0 is true, we allow ourselves to wrongly reject it about once in every 20 tests.

p 值是与样本数据直接相关的量：它是在 H₀ 为真的假设下，观察到当前检验统计量及更极端值的概率。判定规则非常简洁：p 值 ≤ α 时拒绝 H₀（结果显著）；p 值 > α 时没有足够证据拒绝 H₀（结果不显著）。在 A-Level 考试中，p 值通常通过查统计表获得，而不是用软件计算。

The p-value is a quantity directly linked to the sample data: it is the probability, assuming H0 is true, of observing the current test statistic and values more extreme. The decision rule is very simple: reject H0 when p ≤ α (the result is significant); when p > α, there is not enough evidence to reject H0 (the result is not significant). In A-Level exams, the p-value is usually obtained from statistical tables rather than computed by software.

一个常见的理解误区是”p 值越小，效应越大”。p 值衡量的是证据的强度，而不是效应的大小。一个非常大的样本可以把一个微小的、实际无意义的差异检验为”显著”。因此考试中遇到”解释显著性水平的含义”这类题，要答”在 H₀ 为真时错误拒绝 H₀ 的概率”，而不是笼统地说”犯错的概率”。

A common misconception is that “the smaller the p-value, the larger the effect.” The p-value measures the strength of evidence, not the size of the effect. A very large sample can make a tiny, practically meaningless difference test as “significant”. Therefore, when an exam asks you to “explain the meaning of the significance level”, answer “the probability of wrongly rejecting H0 when H0 is true”, rather than vaguely saying “the probability of making a mistake”.

四、单尾检验与双尾检验：方向决定一半分数 | One-Tailed vs Two-Tailed Tests: The Direction Decides Half the Marks

单尾检验的拒绝域只位于分布的一侧。若 H₁: μ > μ₀，拒绝域在分布右尾，临界值 z* 满足 P(Z > z*) = α；若 H₁: μ < μ₀，拒绝域在左尾，临界值满足 P(Z < z*) = α。双尾检验的拒绝域分居两侧，每侧概率各为 α/2，临界值满足 P(Z > z*) = α/2。

The rejection region of a one-tailed test lies on only one side of the distribution. If H1: μ > μ0, the rejection region is in the right tail, and the critical value z* satisfies P(Z > z*) = α; if H1: μ < μ0, the rejection region is in the left tail, and the critical value satisfies P(Z < z*) = α. In a two-tailed test the rejection region is split across both sides, each side carrying probability α/2, and the critical value satisfies P(Z > z*) = α/2.

选择错误的方向是致命的：用单尾检验的临界值去判双尾问题（或反过来），结论很可能完全颠倒。判定的依据永远来自题目语境：题目问”是否有证据表明平均重量低于声称值”，就是左尾检验；问”平均重量是否不同于声称值”，就是双尾检验。

Choosing the wrong direction is fatal: using the critical value of a one-tailed test for a two-tailed problem (or vice versa) can completely reverse the conclusion. The basis for the decision always comes from the context of the question: if the question asks “is there evidence that the mean weight is below the claimed value”, it is a left-tailed test; if it asks “whether the mean weight differs from the claimed value”, it is a two-tailed test.

双尾检验还有一个常见陷阱：有些人把”小于”和”大于”两种单尾检验各做一遍，然后取其中一个显著的结果作为结论。这是错误的，因为它把总错误率翻倍了。双尾检验必须用 α/2 在两侧分别划定拒绝域，一次性得出结论。

There is also a common trap in two-tailed tests: some people perform both one-tailed tests (“less than” and “greater than”) separately, then take whichever result is significant as the conclusion. This is wrong because it doubles the overall error rate. A two-tailed test must allocate α/2 to each side and reach a single conclusion in one pass.

五、正态分布下的总体均值检验：Z 检验的完整步骤 | Testing a Population Mean with the Normal Distribution: The Complete Z-Test Procedure

当总体方差已知（或样本足够大，可用样本方差近似），检验总体均值 μ 使用标准正态分布。检验统计量为 z = (x̄ − μ₀) / (σ / √n)，其中 μ₀ 是 H₀ 中的假设值，σ 是总体标准差，n 是样本容量。这一步是 A-Level 统计的必考点，公式必须默写无误。

When the population variance is known (or the sample is large enough for the sample variance to be used as an approximation), testing the population mean μ uses the standard normal distribution. The test statistic is z = (x̄ − μ0) / (σ / √n), where μ0 is the hypothesised value in H0, σ is the population standard deviation, and n is the sample size. This step is a guaranteed examination point in A-Level Statistics, and the formula must be reproduced from memory without error.

完整的答题流程共六步：第一步，写出 H₀ 和 H₁；第二步，确定显著性水平 α 与检验方向；第三步，计算检验统计量 z 的数值；第四步，查表得到临界值（或 p 值）；第五步，比较并作出判定（拒绝或不拒绝 H₀）；第六步，用一句完整的中文/英文陈述结论，回扣题目语境。

The complete answering procedure has six steps: first, write down H0 and H1; second, fix the significance level α and the direction of the test; third, compute the value of the test statistic z; fourth, look up the critical value (or p-value) in tables; fifth, compare and decide (reject or not reject H0); sixth, state the conclusion in a complete sentence that links back to the context of the question.

举一个完整例子：某厂商声称电池平均寿命为 120 小时，σ = 8 小时。随机抽取 36 节电池，平均寿命 x̄ = 117.5 小时。在 5% 显著性水平下，是否有证据表明平均寿命低于声称值？检验统计量 z = (117.5 − 120) / (8 / √36) = −2.5 / 1.333 = −1.875。左尾 5% 的临界值为 −1.6449。因为 −1.875 < −1.6449，落在拒绝域内，所以拒绝 H₀，有充分证据表明平均寿命低于 120 小时。

Here is a complete example: a manufacturer claims its batteries last 120 hours on average, with σ = 8 hours. A random sample of 36 batteries gives a mean life of x̄ = 117.5 hours. At the 5% significance level, is there evidence that the mean life is below the claimed value? The test statistic is z = (117.5 − 120) / (8 / √36) = −2.5 / 1.333 = −1.875. The 5% left-tail critical value is −1.6449. Since −1.875 < −1.6449, the value lies in the rejection region, so we reject H0 and conclude there is strong evidence that the mean life is below 120 hours.

答题时最容易扣分的是最后一步的结论表述。必须明确写出”拒绝 H₀”或”没有足够证据拒绝 H₀”，并且把结论翻译回实际背景（电池、奶茶、考试成绩等），不能只写统计术语。

The conclusion statement in the final step is where most marks are lost. You must explicitly write “reject H0” or “there is insufficient evidence to reject H0”, and translate the conclusion back into the practical context (batteries, bubble tea, exam scores, etc.), rather than writing statistical jargon alone.

六、总体比例的假设检验：二项分布与正态近似 | Hypothesis Testing for a Proportion: The Binomial Distribution and the Normal Approximation

检验总体比例 p 时，样本中的”成功次数”X 在 H₀ 下服从二项分布 X ~ B(n, p₀)。当 n 足够大（通常要求 np₀ ≥ 5 且 n(1−p₀) ≥ 5）时，可以用正态近似 X ~ N(np₀, np₀(1−p₀))，检验统计量 z = (X − np₀) / √(np₀(1−p₀))。

When testing a population proportion p, the number of “successes” X in the sample follows a binomial distribution X ~ B(n, p0) under H0. When n is large enough (usually requiring np0 ≥ 5 and n(1−p0) ≥ 5), the normal approximation X ~ N(np0, np0(1−p0)) can be used, with test statistic z = (X − np0) / √(np0(1−p0)).

二项分布情形下的精确检验需要小心处理”≥”和”>”的边界。例如 H₀: p = 0.4，H₁: p > 0.4，样本 n = 20，观察到 X = 12。则 p 值 = P(X ≥ 12 | p = 0.4) = 1 − P(X ≤ 11)。查二项分布表时，必须确认表格给的是 P(X ≤ x) 还是 P(X ≥ x)，用错方向会直接判错。

Exact tests with the binomial distribution require careful handling of the boundaries between “≥” and “>”. For example, with H0: p = 0.4, H1: p > 0.4, sample n = 20, and observed X = 12, the p-value is P(X ≥ 12 | p = 0.4) = 1 − P(X ≤ 11). When using binomial tables, you must check whether the table gives P(X ≤ x) or P(X ≥ x); using the wrong direction is an immediate error.

正态近似的连续性修正（continuity correction）是进阶考点。当 n 不大时，用 P(X ≥ 11.5) 代替 P(X ≥ 12) 可以显著提高近似精度。AQA 国际大纲的 MA04 试卷中，连续性修正常以”说明为什么需要修正”的形式出现，答案要点是”二项分布是离散的，正态分布是连续的，修正用于弥合离散与连续之间的差距”。

The continuity correction for the normal approximation is an advanced examination point. When n is moderate, replacing P(X ≥ 12) with P(X ≥ 11.5) markedly improves the accuracy of the approximation. In AQA International MA04 papers, the continuity correction often appears as “explain why the correction is needed”; the key point of the answer is that “the binomial distribution is discrete while the normal distribution is continuous, and the correction bridges the gap between discrete and continuous”.

七、第一类错误与第二类错误：理解检验的风险边界 | Type I and Type II Errors: Understanding the Risk Boundaries of a Test

第一类错误（Type I error）是在 H₀ 实际为真时错误地拒绝了它，其概率恰好等于显著性水平 α。第二类错误（Type II error）是在 H₀ 实际为假时未能拒绝它，其概率记为 β。两类错误像跷跷板的两端：在样本容量不变时，减小 α 会使 β 增大，反之亦然；唯一的出路是增大样本容量 n，才能同时压低两者。

A Type I error occurs when H0 is actually true but we wrongly reject it; its probability is exactly the significance level α. A Type II error occurs when H0 is actually false but we fail to reject it; its probability is denoted β. The two errors are like the two ends of a seesaw: with a fixed sample size, decreasing α increases β, and vice versa; the only way out is to increase the sample size n, which reduces both at once.

计算第二类错误的概率是考试中的高阶题。以 Z 检验为例：设 H₀: μ = 100，H₁: μ > 100，σ = 10，n = 25，α = 0.05。临界值 z* = 1.6449，对应样本均值临界点 x̄* = 100 + 1.6449 × (10/5) = 103.29。若真实均值 μ₁ = 105，则 β = P(x̄ < 103.29 | μ = 105) = P(Z < (103.29 − 105)/2) = P(Z < −0.855) ≈ 0.196。

Computing the probability of a Type II error is an advanced question in exams. Take a Z-test as an example: let H0: μ = 100, H1: μ > 100, σ = 10, n = 25, α = 0.05. The critical value is z* = 1.6449, which corresponds to the sample-mean cut-off x̄* = 100 + 1.6449 × (10/5) = 103.29. If the true mean is μ1 = 105, then β = P(x̄ < 103.29 | μ = 105) = P(Z < (103.29 − 105)/2) = P(Z < −0.855) ≈ 0.196.

这类题的解题关键是先算出”临界点”（在 H₀ 的尺度下），再把它放到 H₁ 的真实分布里计算概率。许多同学把两个分布混在一起算，导致 β 计算错误。记住：α 在 H₀ 的分布里定义，β 在 H₁ 的分布里定义，两套分布必须分开使用。

The key to solving such questions is first computing the “cut-off point” (on the H0 scale), then placing it in the true distribution under H1 to calculate the probability. Many students mix the two distributions together and get β wrong. Remember: α is defined in the distribution under H0, while β is defined in the distribution under H1; the two distributions must be used separately.

八、积矩相关系数的假设检验：从样本相关到总体相关 | Hypothesis Testing for Correlation: From Sample Correlation to Population Correlation

样本积矩相关系数 r 描述的是样本中两个变量的线性相关程度，但它是否代表总体中真的存在相关关系，需要假设检验来回答。检验的零假设是 H₀: ρ = 0（总体相关系数为 0，即两变量总体无关），备择假设可以是 ρ > 0、ρ < 0 或 ρ ≠ 0，取决于题目问的是正相关、负相关还是”是否存在相关”。

The sample product-moment correlation coefficient r describes the strength of the linear relationship between two variables in the sample, but whether it represents a genuine relationship in the population must be answered by hypothesis testing. The null hypothesis is H0: ρ = 0 (the population correlation coefficient is 0, meaning the variables are unrelated in the population); the alternative can be ρ > 0, ρ < 0, or ρ ≠ 0, depending on whether the question asks about positive correlation, negative correlation, or “whether any correlation exists”.

检验方法非常直接：查”积矩相关系数临界值表”，表中给出不同样本容量 n 和显著性水平 α 下的临界值。若 |r| 大于临界值，则拒绝 H₀，认为存在显著的线性相关；否则没有足够证据认为总体存在相关。注意：临界值表通常按自由度（n − 2）或直接按 n 列出行，读表前先确认行、列的含义。

The test method is very direct: consult the “critical values table for the product-moment correlation coefficient”, which lists critical values for different sample sizes n and significance levels α. If |r| exceeds the critical value, reject H0 and conclude that there is significant linear correlation; otherwise there is insufficient evidence of correlation in the population. Note: critical value tables are usually organised by degrees of freedom (n − 2) or directly by n; confirm the meaning of the rows and columns before reading the table.

一个常被忽略的细节：样本容量 n 越小，临界值越大，需要更强的样本相关才能判定总体相关显著。例如 n = 10、α = 0.05 时临界值约为 0.632，而 n = 50 时临界值降至约 0.279。这解释了为什么小样本下”看似很强的相关”也可能不显著。

A detail that is often overlooked: the smaller the sample size n, the larger the critical value, and the stronger the sample correlation needed to declare a significant population correlation. For example, with n = 10 and α = 0.05 the critical value is about 0.632, while with n = 50 it drops to about 0.279. This explains why an “apparently strong correlation” from a small sample may still be insignificant.

九、完整例题解析：从写假设到写结论的满分示范 | Worked Example: A Full-Mark Demonstration from Hypotheses to Conclusion

下面用一道 AQA 风格的完整例题串联全部步骤。题目：某校声称学生平均每周学习时间为 15 小时，总体标准差为 3 小时。随机抽取 49 名学生，样本平均学习时间为 15.8 小时。在 5% 显著性水平下，检验”平均学习时间是否高于声称值”。

The following complete example in AQA style ties all the steps together. Question: a school claims its students study 15 hours per week on average, with a population standard deviation of 3 hours. A random sample of 49 students gives a sample mean study time of 15.8 hours. At the 5% significance level, test whether the mean study time is higher than claimed.

第一步，写假设：H₀: μ = 15，H₁: μ > 15（右尾检验，因为问题问”是否高于”）。第二步，确认 α = 0.05，右尾临界值 z* = 1.6449。第三步，计算检验统计量：z = (15.8 − 15) / (3 / √49) = 0.8 / 0.4286 = 1.8667。第四步，比较：1.8667 > 1.6449，检验统计量落在拒绝域内。第五步，判定：拒绝 H₀。第六步，结论：在 5% 显著性水平下，有充分证据表明学生平均每周学习时间高于 15 小时。

Step one, state the hypotheses: H0: μ = 15, H1: μ > 15 (right-tailed test, because the question asks “whether higher”). Step two, confirm α = 0.05 and the right-tail critical value z* = 1.6449. Step three, compute the test statistic: z = (15.8 − 15) / (3 / √49) = 0.8 / 0.4286 = 1.8667. Step four, compare: 1.8667 > 1.6449, so the test statistic lies in the rejection region. Step five, decide: reject H0. Step six, conclude: at the 5% significance level, there is strong evidence that the mean weekly study time of students is higher than 15 hours.

再给一道比例检验例题。题目：某品牌薯片包装上写着”每袋 30% 的概率抽中限量卡片”。一位顾客买了 40 袋，只抽中 6 张卡片。在 5% 显著性水平下，检验”中卡概率是否低于 30%”。H₀: p = 0.3，H₁: p < 0.3。np₀ = 12 ≥ 5，可用正态近似。z = (6 − 12) / √(12 × 0.7) = −6 / 2.898 = −2.070。左尾临界值 −1.6449。因为 −2.070 < −1.6449，拒绝 H₀：有证据表明中卡概率低于 30%。

Here is a second worked example on proportion testing. Question: a brand of crisps states on its packaging “each bag has a 30% chance of containing a limited-edition card”. A customer buys 40 bags and gets only 6 cards. At the 5% significance level, test whether the card probability is below 30%. H0: p = 0.3, H1: p < 0.3. Since np0 = 12 ≥ 5, the normal approximation is valid. z = (6 − 12) / √(12 × 0.7) = −6 / 2.898 = −2.070. The left-tail critical value is −1.6449. Since −2.070 < −1.6449, we reject H0: there is evidence that the card probability is below 30%.

十、常见失分点：考生最容易踩的五个坑 | Five Common Mark-Losing Traps in Hypothesis Testing

第一个坑是把样本统计量写进假设。H₀ 和 H₁ 必须使用总体参数 μ、p、ρ，写 x̄ 或 p̂ 一律扣分。第二个坑是方向选错：把”是否低于”做成双尾检验，或者把”是否不同”做成单尾检验，结论随之全错。第三个坑是查表方向错误：二项分布表有 P(X ≤ x) 和 P(X ≥ x) 两种，正态表有左侧面积和右侧面积两种，用前必须确认。

The first trap is putting sample statistics into the hypotheses. H0 and H1 must use the population parameters μ, p, ρ; writing x̄ or p̂ always loses marks. The second trap is choosing the wrong direction: turning “whether lower” into a two-tailed test, or “whether different” into a one-tailed test, which makes the whole conclusion wrong. The third trap is reading tables in the wrong direction: binomial tables come in P(X ≤ x) and P(X ≥ x) forms, and normal tables in left-tail and right-tail forms; confirm before use.

第四个坑是结论表述不规范。只写”拒绝 H₀”而不回扣题目背景，或者写”证明 H₀ 为假””接受 H₁ 为真”这类绝对化表述，都会被扣分。规范的写法是”有（充分）证据表明……”或”没有足够证据表明……”。第五个坑是忽略连续性修正的使用条件：题目明确要求说明何时需要修正、为什么修正，答不出要点等于放弃整道小题。

The fourth trap is a non-standard conclusion statement. Merely writing “reject H0” without linking back to the context of the question, or using absolute wording such as “prove H0 false” or “accept H1 as true”, will lose marks. The standard wording is “there is (strong) evidence that…” or “there is insufficient evidence that…”. The fifth trap is ignoring the conditions for using the continuity correction: when a question explicitly asks when and why the correction is needed, failing to state the key points means abandoning the whole sub-question.

最后一个隐藏陷阱是”显著性水平 α 与 p 值的换算”。有的题目给出的是 p 值而不是临界值，例如 p = 0.023 与 α = 0.05 比较时，0.023 < 0.05，拒绝 H₀。很多同学只会比临界值，遇到 p 值就不知所措。两种判定路径都要熟练掌握。

A final hidden trap is converting between the significance level α and the p-value. Some questions give a p-value instead of a critical value; for example, when p = 0.023 is compared with α = 0.05, since 0.023 < 0.05, we reject H0. Many students only know how to compare critical values and are at a loss when faced with a p-value. You must be fluent in both decision paths.

十一、考试答题结构模板：按步骤稳稳拿满分 | Exam Answer Structure Template: Securing Full Marks Step by Step

把下面的模板背下来，考场上按顺序套用，可以避免绝大多数结构性丢分。第一步：写假设（H₀ 用等号，H₁ 用题目关键词确定方向）；第二步：写显著性水平与检验类型（如”5% 单尾检验”）；第三步：写出检验统计量公式并代入数值；第四步：给出临界值或 p 值，注明来源（”查正态分布表”）；第五步：比较并下判定（”由于……，拒绝 H₀”）；第六步：用实际背景语言陈述结论。

Memorise the template below and apply it in order in the exam room; it will prevent most structural mark loss. Step one: write the hypotheses (H0 uses an equals sign; H1 direction is fixed by the question keywords); step two: write the significance level and test type (for example “5% one-tailed test”); step three: write out the test statistic formula and substitute the values; step four: give the critical value or p-value, noting the source (“from the normal distribution table”); step five: compare and decide (“since…, reject H0”); step six: state the conclusion in the language of the practical context.

时间管理上，建议把”写假设”和”写结论”各控制在 30 秒内，把主要时间留给计算和查表。计算时保留至少 4 位有效数字，最终比较时统一保留 3 位小数，避免因舍入误差导致判定边缘出错。若计算出的统计量恰好等于临界值，按”落在拒绝域内”处理（临界值属于拒绝域）。

For time management, keep “writing the hypotheses” and “writing the conclusion” within 30 seconds each, and spend the bulk of your time on computation and table reading. Keep at least 4 significant figures during calculation, and round to 3 decimal places uniformly for the final comparison, to avoid borderline decision errors caused by rounding. If the computed statistic exactly equals the critical value, treat it as falling inside the rejection region (the critical value belongs to the rejection region).

复习建议：把近五年 AQA 国际大纲 9660 的 MA04 真题中所有假设检验题集中起来，按题型分类（均值检验、比例检验、相关检验）各练三遍。第一遍求做对，第二遍求步骤完整，第三遍限时模拟。错题整理成一张”失分点清单”，考前 24 小时只看清单。

Revision advice: gather all hypothesis-testing questions from the last five years of AQA International 9660 MA04 papers, classify them by question type (mean tests, proportion tests, correlation tests), and practise each type three times. The first pass aims for correctness, the second for complete steps, and the third is a timed simulation. Compile your mistakes into a single “mark-loss checklist” and review only that checklist in the final 24 hours before the exam.

Summary | 总结

假设检验是 A-Level 统计学中连接”数据”与”结论”的桥梁。它的完整链条是：写出总体参数的假设（H₀ 含等号，H₁ 由题目方向决定）→ 确定显著性水平 → 计算检验统计量 → 查表得临界值或 p 值 → 比较判定 → 用实际背景语言陈述结论。每一步都有固定的规范，结构性失分完全可以靠模板避免。

Hypothesis testing is the bridge connecting “data” and “conclusions” in A-Level Statistics. Its complete chain is: state the hypotheses about the population parameter (H0 contains the equals sign; H1 direction is fixed by the question) → fix the significance level → compute the test statistic → read the critical value or p-value from tables → compare and decide → state the conclusion in the language of the practical context. Every step has a fixed convention, and structural mark loss can be fully avoided with a template.

本单元的核心考点集中在五处：Z 检验的六步流程、二项分布与正态近似下的比例检验、第一类与第二类错误的概率计算、积矩相关系数的显著性检验，以及连续性修正的使用条件。把这五块内容练到”条件反射”的程度，配合真题限时训练，假设检验部分就能成为你 A-Level 数学考试中稳定拿分的板块。

The core examination points of this unit concentrate on five areas: the six-step Z-test procedure, proportion testing with the binomial distribution and normal approximation, the probability calculations of Type I and Type II errors, significance testing of the product-moment correlation coefficient, and the conditions for using the continuity correction. Drill these five blocks to the level of a conditioned reflex, combined with timed practice on real past papers, and hypothesis testing will become a reliable mark-earning section of your A-Level Mathematics exam.

更多咨询请联系16621398022（同微信）