Quality of Tests in A-Level Statistics | A-Level 统计中检验的质量

📚 Quality of Tests in A-Level Statistics | A-Level 统计中检验的质量

In Edexcel A-Level Statistics, the quality of a hypothesis test refers to how well the test controls errors and detects true effects. A good test has a low probability of making a wrong decision and a high chance of rejecting a false null hypothesis. This article explains the key ideas: significance level, critical region, p-value, Type I and Type II errors, power, and how to judge whether a test is good.

在 Edexcel A-Level 统计中，假设检验的质量是指检验控制错误和发现真实效应的能力。一个好的检验应当有较低的犯错概率，并且有较高的概率拒绝错误的原假设。本文讲解关键概念：显著性水平、拒绝域、p 值、第一类错误和第二类错误、功效，以及如何判断一个检验是否良好。

1. Hypothesis Testing Basics | 假设检验基础

A hypothesis test is a decision-making procedure for testing a claim about a population parameter, such as a binomial probability p or a normal mean μ. The null hypothesis H₀ is the default assumption, while the alternative hypothesis H₁ is what we seek evidence for. Tests can be one-tailed or two-tailed depending on H₁.

假设检验是一种决策程序，用于检验关于总体参数（如二项概率 p 或正态均值 μ）的声明。原假设 H₀ 是默认假设，备择假设 H₁ 是我们寻找证据支持的假设。根据 H₁ 的不同，检验可以是单尾或双尾。

2. Significance Level and Size of a Test | 显著性水平与检验尺度

The significance level, denoted by α, is the maximum acceptable probability of rejecting H₀ when it is actually true. Common values are α = 0.05 or α = 0.01. For discrete distributions such as the binomial, the actual size of the test may be less than α because the test statistic can only take integer values.

显著性水平，记为 α，是当 H₀ 实际为真时拒绝 H₀ 的最大可接受概率。常用取值为 α = 0.05 或 α = 0.01。对于二项分布等离散分布，检验的实际尺度可能小于 α，因为检验统计量只能取整数值。

The size of a test is the actual probability of a Type I error achieved by the chosen critical region. A test is considered valid if its size does not exceed α. If the size is much smaller than α, the test is described as conservative.

检验的尺度是所选拒绝域实际达到的第一类错误概率。如果检验的尺度不超过 α，则该检验是有效的。如果尺度远小于 α，则该检验被称为保守检验。

3. Critical Region and Critical Value | 拒绝域与临界值

The critical region is the set of sample outcomes that lead to rejection of H₀. Its boundary is called the critical value. For a one-tailed test, the critical region lies entirely in one tail; for a two-tailed test, it is split between both tails, usually with α/2 in each tail.

拒绝域是导致拒绝 H₀ 的样本结果集合，其边界称为临界值。对于单尾检验，拒绝域完全位于一个尾部；对于双尾检验，它被拆分到两个尾部，通常每侧概率为 α/2。

For example, testing H₀: p = 0.3 against H₁: p > 0.3 with n = 10 and α = 0.05 might give a critical region X ≥ 6 if P(X ≥ 6 | p = 0.3) ≤ 0.05. The value 6 is the critical value.

例如，检验 H₀: p = 0.3 对 H₁: p > 0.3，其中 n = 10 且 α = 0.05，如果 P(X ≥ 6 | p = 0.3) ≤ 0.05，则拒绝域为 X ≥ 6。数值 6 就是临界值。

4. p-Value and Decision Rule | p 值与决策规则

The p-value is the probability of obtaining a test statistic at least as extreme as the observed one, assuming H₀ is true. If the p-value is less than or equal to α, we reject H₀. A smaller p-value indicates stronger evidence against H₀.

p 值是在 H₀ 为真的前提下，得到至少与观测值一样极端的检验统计量的概率。如果 p 值小于或等于 α，我们就拒绝 H₀。p 值越小，表明反对 H₀ 的证据越强。

Using the p-value method is particularly useful in Edexcel exam questions because it allows a direct comparison with the stated significance level and avoids the need to state critical values explicitly.

在 Edexcel 考试题中，使用 p 值方法特别有用，因为它可以直接与给定的显著性水平进行比较，无需明确写出临界值。

Decision rule: Reject H₀ if p-value ≤ α

决策规则：若 p 值 ≤ α，则拒绝 H₀

5. Type I Error | 第一类错误

A Type I error occurs when H₀ is true but the test rejects it. The probability of a Type I error is exactly the size of the critical region. By choosing a small α, we control the chance of this error, but we cannot eliminate it entirely.

第一类错误发生在 H₀ 为真但检验却拒绝了它。第一类错误的概率恰好是拒绝域的尺度。通过选择较小的 α，我们可以控制这种错误的发生概率，但无法完全消除它。

In discrete tests, the actual Type I error probability is often lower than α because the critical region is chosen as the largest set of outcomes with probability not exceeding α. This is an important point when commenting on the quality of a test.

在离散检验中，实际第一类错误概率通常低于 α，因为拒绝域被选为概率不超过 α 的最大结果集合。在评论检验质量时，这一点很重要。

Decision	H₀ is true	H₀ is false
Reject H₀	Type I error (α)	Correct decision (power = 1 − β)
Do not reject H₀	Correct decision	Type II error (β)

Table: Types of error in hypothesis testing

表：假设检验中的两类错误

6. Type II Error | 第二类错误

A Type II error occurs when H₀ is false but the test fails to reject it. The probability of a Type II error is denoted by β. Unlike α, the value of β depends on the true value of the parameter under H₁, so it is not a single fixed number unless H₁ specifies a simple alternative value.

第二类错误发生在 H₀ 为假但检验未能拒绝它。第二类错误的概率记为 β。与 α 不同，β 的值取决于 H₁ 下参数的真实值，因此除非 H₁ 指定了简单备择值，否则它不是一个固定数值。

A test with a high probability of Type II error is weak because it often misses a real effect. In exam questions, you may need to calculate β for a given alternative value, such as H₁: p = 0.7, by finding the probability that X falls outside the critical region under that alternative.

第二类错误概率高的检验较弱，因为它经常错过真实效应。在考试题中，你可能需要针对给定的备择值（如 H₁: p = 0.7）计算 β，即求出在该备择值下 X 落在拒绝域之外的概率。

7. Power of a Test | 检验的功效

The power of a test is the probability of correctly rejecting H₀ when H₁ is true. It is equal to 1 − β. High power is a desirable quality of a test because it means the test is sensitive to real departures from H₀.

检验的功效是当 H₁ 为真时正确拒绝 H₀ 的概率，等于 1 − β。高功效是检验的理想质量，因为它意味着检验对真实偏离 H₀ 的情况很敏感。

Power = 1 − β = P(Reject H₀ | H₁ is true)

功效 = 1 − β = P(拒绝 H₀ | H₁ 为真)

Power depends on the true value of the parameter. For example, if H₀: p = 0.4 and the true p = 0.8, the test will usually have higher power than if the true p = 0.5, because the distributions under H₀ and H₁ are further apart.

功效取决于参数的真实值。例如，若 H₀: p = 0.4 而真实 p = 0.8，检验通常比真实 p = 0.5 时有更高的功效，因为 H₀ 和 H₁ 下的分布相隔更远。

8. Factors Affecting Power | 影响功效的因素

Several factors influence the power of a test:

影响检验功效的因素有以下几个：

Sample size n — larger samples give more information and increase power.
样本量 n — 样本量越大，信息越多，功效越高。
Significance level α — larger α increases power but also increases the Type I error rate.
显著性水平 α — α 越大，功效越高，但第一类错误率也会上升。
Effect size — the distance between the true parameter and the null value; larger effects are easier to detect.
效应量 — 真实参数与原假设值之间的距离；效应越大越容易被检测到。
One-tailed vs two-tailed — a one-tailed test has more power than a two-tailed test at the same α because all of α is concentrated in one tail.
单尾与双尾 — 在相同的 α 下，单尾检验比双尾检验功效更高，因为全部 α 集中在一个尾部。

In a given exam problem, you can improve power by increasing n or by using a one-tailed test when the alternative is directional, provided this is justified by the context.

在具体考题中，你可以通过增大 n 来提高功效，或者在备择假设有方向性时使用单尾检验（前提是上下文支持），从而提高功效。

9. Quality of a Good Test | 良好检验的质量

A high-quality test should have a Type I error probability no greater than the stated α and as high a power as possible for plausible alternatives. In Edexcel questions, you may be asked to explain why a test is or is not good, often by referencing its actual size or power.

高质量检验应使第一类错误概率不超过给定的 α，并尽可能提高对合理备择假设的功效。在 Edexcel 考题中，可能会要求你解释某个检验为何好或不好，通常要参考其实际尺度或功效。

If two tests have the same significance level, the one with greater power is better. If a test has an actual size well below α, it may be conservative, but its power could be improved by adjusting the critical region or increasing the sample size.

如果两个检验具有相同的显著性水平，功效更大的那个更好。如果一个检验的实际尺度远低于 α，它可能较保守，但其功效可以通过调整拒绝域或增大样本量来提高。

10. Worked Example | 例题解析

Example: A coin is tossed 20 times to test H₀: p = 0.5 against H₁: p > 0.5 at α = 0.05. Using the binomial distribution B(20, 0.5), the critical region is X ≥ 15 because P(X ≥ 15 | p = 0.5) = 0.0207 < 0.05, while P(X ≥ 14) = 0.0577 > 0.05. If 16 heads are observed, H₀ is rejected.

例题：抛硬币 20 次，检验 H₀: p = 0.5 对 H₁: p > 0.5，α = 0.05。使用二项分布 B(20, 0.5)，拒绝域为 X ≥ 15，因为 P(X ≥ 15 | p = 0.5) = 0.0207 < 0.05，而 P(X ≥ 14) = 0.0577 > 0.05。如果观察到 16 次正面，则拒绝 H₀。

The actual size of this test is 0.0207, so the test is conservative. Its power against the alternative p = 0.7 can be found by calculating P(X ≥ 15 | p = 0.7) = 0.4164. This means the test correctly rejects H₀ only about 41.6% of the time when p = 0.7, which is relatively low power for n = 20.

该检验的实际尺度为 0.0207，因此它是一个保守检验。它对备择假设 p = 0.7 的功效可通过计算 P(X ≥ 15 | p = 0.7) = 0.4164 得到。这意味着当 p = 0.7 时，检验正确拒绝 H₀ 的概率仅为约 41.6%，对于 n = 20 来说功效较低。

Actual size = 0.0207, Power at p = 0.7 = 0.4164

实际尺度 = 0.0207，p = 0.7 时的功效 = 0.4164

11. Common Misconceptions | 常见误区

Do not say that α is the probability that H₀ is true. Also, failing to reject H₀ does not prove H₀ is true; it may simply mean the test had low power. The p-value is not the probability that H₀ is true, nor is it the probability that the observed data occurred by chance alone.

不要说 α 是 H₀ 为真的概率。此外，未能拒绝 H₀ 并不证明 H₀ 为真；它可能仅仅意味着检验功效较低。p 值不是 H₀ 为真的概率，也不是观测数据仅由偶然发生的概率。

A significant result at α = 0.05 does not mean there is a 95% chance that H₁ is true. It only means that, if H₀ were true, such an extreme result would occur less than 5% of the time.

在 α = 0.05 下结果显著，并不意味着 H₁ 为真的概率是 95%。它只表示，如果 H₀ 为真，如此极端的结果出现的概率不到 5%。

12. Exam Tips | 考试提示

In Edexcel questions on quality of tests, always state the actual probability of the critical region for discrete tests, compare it with α, and comment on Type I and Type II errors. Define power clearly and use correct notation H₀, H₁, α, β, p.

在 Edexcel 关于检验质量的题目中，对于离散检验，务必给出拒绝域的实际概率，与 α 比较，并评论第一类和第二类错误。明确定义功效，并使用正确符号 H₀、H₁、α、β、p。

When calculating power, clearly state the alternative value under which you are working, and use the correct distribution. In binomial questions, always check whether you are using B(n, p) under H₀ or under H₁.

计算功效时，要明确说明你所使用的备择值，并使用正确的分布。在二项分布题目中，一定检查你使用的是 H₀ 下的 B(n, p) 还是 H₁ 下的 B(n, p)。

If a question asks whether a test is good, refer to both size and power. A test can be valid but have poor power, so say “the test is valid but may lack power” rather than simply “the test is bad”.

如果题目问一个检验是否良好，要同时提及尺度和功效。一个检验可能有效但功效差，因此应说“该检验有效但可能缺乏功效”，而不是简单地说“该检验不好”。

Published by TutorHao | Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

Quality of Tests in A-Level Statistics | A-Level 统计中检验的质量