A-Level Further Mathematics: Type I and Type II Errors in Hypothesis Testing | 进阶数学：假设检验中的第一类与第二类错误

📚 A-Level Further Mathematics: Type I and Type II Errors in Hypothesis Testing | 进阶数学：假设检验中的第一类与第二类错误

Hypothesis testing is a cornerstone of statistical inference, and in Edexcel A-Level Further Mathematics, a deep understanding of Type I and Type II errors is essential for both problem solving and statistical reasoning. These errors quantify the risks we take when we make decisions based on sample data rather than complete information.

假设检验是统计推断的核心内容。在 Edexcel 进阶数学中，深入理解第一类错误与第二类错误不仅是解题的关键，更是培养统计思维的重要环节。当我们只能基于样本数据而非完整信息做出判断时，这两类错误量化了我们所承担的风险。

1. Understanding Hypotheses: Null and Alternative | 理解假设：原假设与备择假设

Every hypothesis test begins with two competing statements. The null hypothesis (H₀) represents the status quo or the claim we assume to be true until evidence suggests otherwise. The alternative hypothesis (H₁) is the claim we wish to support; it may be one-tailed (using > or <) or two-tailed (using ≠).

每一次假设检验都始于两个相互对立的陈述。原假设 H₀ 代表现状或我们默认成立的命题；备择假设 H₁ 则是我们希望支持的命题，既可以是单侧（使用 > 或 <），也可以是双侧（使用 ≠）。

For example, suppose we suspect a coin is biased towards heads. We set H₀: p = 0.5 (the coin is fair) against H₁: p > 0.5 (the coin is biased towards heads), where p is the probability of heads. The test uses the number of heads in n tosses as the test statistic.

例如，我们怀疑一枚硬币偏向正面。我们设定 H₀: p = 0.5（硬币是公平的），备择假设 H₁: p > 0.5（硬币偏向正面），其中 p 为正面朝上的概率。检验使用 n 次投掷中正面出现的次数作为检验统计量。

2. The Decision Matrix: Truth vs Conclusion | 决策矩阵：真实情况与检验结论

When we perform a hypothesis test, the unknown truth is either H₀ true or H₀ false. Our decision is either to reject H₀ or not to reject H₀. Combining these gives four possible outcomes. Two are correct decisions; two are errors.

当我们进行假设检验时，未知的真实情况要么是 H₀ 成立，要么是 H₀ 不成立。我们的决策要么是拒绝 H₀，要么是不拒绝 H₀。将两者组合，共有四种可能的结果：两种是正确决策，两种是错误。

	H₀ is true \| H₀ 成立	H₀ is false \| H₀ 不成立
Reject H₀ \| 拒绝 H₀	Type I error (α) 第一类错误 (α)	Correct decision 正确决策（检验功效）
Do not reject H₀ \| 不拒绝 H₀	Correct decision 正确决策	Type II error (β) 第二类错误 (β)

Notice the technical wording: we say “do not reject H₀” rather than “accept H₀”. This reminds us that failing to find evidence against H₀ does not prove it is true.

请注意我们使用“不拒绝 H₀”而非“接受 H₀”这一技术性表达。这提醒我们：未能找到反对 H₀ 的证据，并不等于证明了 H₀ 为真。

3. Type I Error: Rejecting a True Null | 第一类错误：拒绝真实的原假设

A Type I error occurs when we reject H₀ even though H₀ is actually true. Its probability is denoted by α, the significance level of the test. In practice, we choose α before conducting the test, typically α = 0.10, 0.05 or 0.01.

第一类错误是指当 H₀ 实际上为真时，我们却拒绝了 H₀。其概率用 α 表示，即检验的显著性水平。在实际操作中，我们会在检验之前选定 α，通常取 α = 0.10、0.05 或 0.01。

For example, if we test a coin at the 5% significance level and H₀: p = 0.5 is true, then α = 0.05 means there is a 5% chance that we will wrongly conclude the coin is biased. This risk is entirely chosen by the researcher.

例如，若我们在 5% 显著性水平下检验硬币，且 H₀: p = 0.5 真实成立，则 α = 0.05 表示我们有 5% 的概率错误地得出硬币有偏的结论。这一风险完全由研究者事先选定。

Key fact: By definition, P(Type I error) = α, provided that the test statistic’s distribution is exactly as specified under H₀. The critical region of a test is constructed so that its total probability under H₀ equals α.

关键事实：根据定义，第一类错误的概率 P(第一类错误) = α，前提是检验统计量在 H₀ 下的分布与规定完全一致。检验的拒绝域正是按照其在 H₀ 下的总概率等于 α 来构造的。

4. Type II Error: Failing to Reject a False Null | 第二类错误：未拒绝错误的原假设

A Type II error occurs when we do not reject H₀ even though H₀ is actually false. Its probability is denoted by β. Unlike α, β is not a single fixed value; it depends on the true value of the population parameter under H₁.

第二类错误是指当 H₀ 实际上不成立时，我们却未能拒绝 H₀。其概率用 β 表示。与 α 不同，β 并不是一个固定不变的值；它取决于 H₁ 下总体参数的真实取值。

For instance, in a coin test with H₀: p = 0.5 against H₁: p > 0.5, if the true bias is p = 0.9, the test will almost surely reject H₀, so β is very small. But if the true bias is only p = 0.55, the test may frequently fail to detect it, giving a much larger β.

例如，在 H₀: p = 0.5 对 H₁: p > 0.5 的硬币检验中，如果真实偏差为 p = 0.9，检验几乎肯定会拒绝 H₀，因此 β 非常小。但如果真实偏差仅为 p = 0.55，检验可能经常无法察觉这一偏差，此时 β 会大得多。

Key fact: To calculate β, we must assume a specific alternative value of the parameter, then compute the probability that the test statistic falls outside the critical region under that alternative distribution.

关键事实：要计算 β，我们必须先假定参数的具体备择值，然后计算在该备择分布下检验统计量落在拒绝域之外的概率。

5. Significance Level α and the Critical Region | 显著性水平 α 与拒绝域

The critical region is the set of values of the test statistic that lead to rejection of H₀. Its size equals α. For a discrete distribution, we choose the critical region such that P(reject H₀ | H₀ true) ≤ α, while keeping it as close to α as possible without exceeding it.

拒绝域是使检验拒绝 H₀ 的检验统计量取值集合，其大小等于 α。对于离散分布，我们选择的拒绝域应满足 P(拒绝 H₀ | H₀ 成立) ≤ α，同时尽可能接近 α 但不超过 α。

For example, let X ~ B(20, p) and test H₀: p = 0.5 against H₁: p > 0.5 at α = 0.05. We search for a critical value c such that P(X ≥ c) ≤ 0.05.

例如，设 X ~ B(20, p)，在 α = 0.05 下检验 H₀: p = 0.5 对 H₁: p > 0.5。我们需要寻找临界值 c，使得 P(X ≥ c) ≤ 0.05。

P(X ≥ 14) = 1 − P(X ≤ 13) = 1 − 0.9423 = 0.0577 > 0.05

P(X ≥ 15) = 1 − P(X ≤ 14) = 1 − 0.9793 = 0.0207 ≤ 0.05

Hence the critical region is X ≥ 15, and the actual Type I error rate is α = 0.0207, not exactly 0.05. This is typical of discrete tests, where the exact significance level may be lower than the nominal level.

因此拒绝域为 X ≥ 15，实际第一类错误率为 α = 0.0207，而不是恰好 0.05。这是离散检验的典型现象：实际显著性水平可能低于名义水平。

6. Power of a Test and the Relationship with β | 检验功效及其与 β 的关系

The power of a test is the probability of correctly rejecting H₀ when H₀ is false. Power = 1 − β. A high-power test is more sensitive to detecting a false null hypothesis.

检验功效是指当 H₀ 不成立时，检验正确拒绝 H₀ 的概率，即功效 = 1 − β。功效越高的检验，对识别错误原假设越灵敏。

In the binomial example above, suppose the true probability of heads is p = 0.7. The Type II error is the probability that X falls in the acceptance region (X ≤ 14) when X ~ B(20, 0.7).

在上述二项分布例子中，假设正面概率的真实值为 p = 0.7。第二类错误就是在 X ~ B(20, 0.7) 时，X 落入接受域（X ≤ 14）的概率。

β = P(X ≤ 14 | X ~ B(20, 0.7)) ≈ 0.5836

Power = 1 − β ≈ 1 − 0.5836 = 0.4164

This means that if the true bias is p = 0.7, the test at the 5% significance level will correctly detect it only about 41.6% of the time. The remaining 58.4% of the time, the test incorrectly fails to reject H₀.

这意味着如果真实偏差为 p = 0.7，在 5% 显著性水平下，检验大约只有 41.6% 的概率能正确识别它。而其余 58.4% 的情况下，检验会错误地未能拒绝 H₀。

7. Trade-offs Between α and β | α 与 β 之间的权衡

For a fixed sample size, decreasing α (making the test more conservative) will increase β, and vice versa. There is an inherent tension between the two error probabilities. The only way to reduce both simultaneously is to increase the sample size.

在样本量固定的情况下，减小 α（使检验更保守）会增加 β，反之亦然。两类错误概率之间存在固有的矛盾。同时降低二者的唯一方法是增大样本量。

For instance, if we changed the critical region in the coin example to X ≥ 16, then α would drop to approximately 0.0059, but β at p = 0.7 would rise because the acceptance region becomes X ≤ 15. Conversely, using X ≥ 14 would raise α to 0.0577 but lower β.

例如，若将硬币检验的拒绝域改为 X ≥ 16，α 将降至约 0.0059，但在 p = 0.7 下的 β 会上升，因为接受域扩大为 X ≤ 15。反之，若使用 X ≥ 14，α 将升至 0.0577，但 β 会降低。

Exam tip: When asked to compare tests, always state the relationship: reducing the significance level shifts the critical region further into the tail, reducing Type I error but increasing Type II error for any given alternative.

考试提示：当被要求比较不同检验时，务必说明这层关系：降低显著性水平会使拒绝域向尾部移动，从而减小第一类错误，但会增大对任意给定备择值的第二类错误。

8. Worked Example: Normal Distribution Test | 例题：正态分布检验

A manufacturer claims that the mean weight of a bag of flour is 500 g. A sample of 25 bags is taken, and it is known that the population standard deviation is σ = 12 g. Test H₀: μ = 500 against H₁: μ ≠ 500 at the 5% significance level.

某厂商声称一袋面粉的平均重量为 500 g。现抽取 25 袋样本，已知总体标准差 σ = 12 g。在 5% 显著性水平下检验 H₀: μ = 500 对 H₁: μ ≠ 500。

Under H₀, the sample mean follows a normal distribution:

在 H₀ 下，样本均值服从正态分布：

X̄ ~ N(500, 12²/25) = N(500, 5.76)

The two-tailed critical region at α = 0.05 is defined by |Z| > 1.96, so we reject H₀ if the z-score lies outside (−1.96, 1.96).

在 α = 0.05 下的双侧拒绝域由 |Z| > 1.96 定义，因此当 z 值落在 (−1.96, 1.96) 之外时拒绝 H₀。

If the sample mean is X̄ = 504.5, then:

如果样本均值为 X̄ = 504.5，则：

Z = (504.5 − 500) / √5.76 = 4.5 / 2.4 = 1.875

Since 1.875 < 1.96, we do not reject H₀ at the 5% level. The Type I error probability for this test is exactly 0.05.

因为 1.875 < 1.96，在 5% 水平下我们不拒绝 H₀。该检验的第一类错误概率恰好为 0.05。

Now suppose the true mean has shifted to μ = 503. The Type II error is the probability that X̄ falls between the two critical values under this alternative. The critical values for X̄ are 500 ± 1.96 × 2.4, i.e. 495.296 and 504.704.

现在假设真实均值偏移到 μ = 503。第二类错误就是在这个备择假设下 X̄ 落入两个临界值之间的概率。X̄ 的临界值为 500 ± 1.96 × 2.4，即 495.296 和 504.704。

β = P(495.296 < X̄ < 504.704 | X̄ ~ N(503, 5.76))

Z₁ = (495.296 − 503) / 2.4 = −3.21, Z₂ = (504.704 − 503) / 2.4 = 0.71

β = P(−3.21 < Z < 0.71) = 0.7611 − 0.0007 = 0.7604

Thus the power of the test to detect a shift of +3 g is only 1 − 0.7604 = 0.2396, or about 24%. This low power reflects both the small sample size and the small effect size.

因此该检验检测 +3 g 偏移的功效仅为 1 − 0.7604 = 0.2396，约 24%。功效偏低，既反映了样本量较小，也反映了效应量较小。

9. Minimising Errors in Practice | 实践中如何减小错误

Since researchers cannot eliminate statistical errors entirely, they must manage them through design. The most powerful tool is increasing the sample size n, because the standard deviation of the sample mean decreases as n grows. This makes the distribution under H₀ and H₁ more separated, reducing both α and β for a fixed critical region.

由于研究者无法完全消除统计错误，只能通过设计来管控。最有力的工具是增大样本量 n，因为样本均值的标准差会随 n 增大而减小。这使得 H₀ 和 H₁ 下的分布更加分离，在拒绝域不变时同时减小 α 与 β。

Choose α based on the consequences of a Type I error. If a false positive is expensive, select a smaller α, e.g. 0.01.
根据第一类错误的后果来选择 α。如果误判成本高，就选择更小的 α，例如 0.01。
If failing to detect an effect is costly, increase sample size or accept a larger α to raise power.
如果漏检效应的代价高，则增大样本量或接受更大的 α 以提高功效。
In A-Level problems, always calculate β under the specific alternative value provided in the question; never guess.
在 A-Level 题目中，始终根据题目给出的具体备择值计算 β，切勿凭空猜测。

10. Exam Tips and Common Mistakes | 考试技巧与常见错误

Many students lose marks in hypothesis testing questions due to careless wording or incomplete calculation of errors. Here are the most important points to remember.

许多学生在假设检验题目中失分，往往是因为措辞不严谨或错误计算不完整。以下是需要记住的最重要要点。

Always state H₀ and H₁ using the correct symbols, including the parameter (e.g. p, μ) and the hypothesised value.
始终用正确的符号写出 H₀ 与 H₁，包括参数（如 p、μ）和假设值。
Remember that P(Type I error) = P(Reject H₀ | H₀ true) = α exactly when no discreteness is involved; for discrete distributions, use the nearest value not exceeding α.
记住 P(第一类错误) = P(拒绝 H₀ | H₀ 成立) = α；对离散分布，使用不超过 α 的最接近值。
For Type II error, replace the parameter with the alternative value and recompute the probability of the acceptance region under that distribution.
计算第二类错误时，将参数替换为备择值，并重新计算在对应分布下接受域的概率。
Do not confuse “do not reject H₀” with “accept H₀”. The former is always the safer interpretation.
不要把“不拒绝 H₀”与“接受 H₀”混为一谈。前者永远是一种更稳妥的表述。
Check whether the test is one-tailed or two-tailed before writing the critical region.
在写出拒绝域前，先确认检验是单侧还是双侧。

11. Conclusion | 总结

Type I and Type II errors are fundamental concepts in hypothesis testing. Type I error is the risk of false alarm, controlled by the significance level α; Type II error is the risk of missing a real effect, quantified by β and closely tied to the power of the test. Mastering these ideas allows you not only to solve exam questions accurately but also to interpret statistical results with genuine understanding.

第一类错误与第二类错误是假设检验中的基本概念。第一类错误是“虚报”的风险，由显著性水平 α 控制；第二类错误是“漏报”真实效应的风险，用 β 量化并且与检验功效密切相关。掌握这些概念，不仅有助于你准确解答考试题目，更能帮助你以真正深刻的理解去解读统计结果。

Published by TutorHao | Further Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

A-Level Further Mathematics: Type I and Type II Errors in Hypothesis Testing | 进阶数学：假设检验中的第一类与第二类错误