A-Level Maths: How to Determine a Two-Tailed Test | A-Level 数学：双尾检验的判定方法

📚 A-Level Maths: How to Determine a Two-Tailed Test | A-Level 数学：双尾检验的判定方法

In A-Level Mathematics, hypothesis testing is a core topic in statistics. A two-tailed test is used when we want to detect a change in a parameter in either direction, without specifying whether the change is an increase or a decrease. This article explains how to identify, set up, and interpret two-tailed tests, with clear criteria to help you avoid common mistakes.

在 A-Level 数学中，假设检验是统计学的核心内容。当我们想要检测某个参数在任意方向上发生变化，而不预先指定是增大还是减小时，就要使用双尾检验。本文将讲解如何识别、建立和解释双尾检验，并提供清晰的判定方法，帮助你避免常见错误。

---

1. What Is a Two-Tailed Test? | 什么是双尾检验？

A two-tailed test is a hypothesis test where the alternative hypothesis states that the population parameter is not equal to a specified value. The critical region is split into two tails of the sampling distribution, one on the left and one on the right.

双尾检验是一种假设检验，其备择假设表明总体参数不等于某个指定值。临界区域被分成采样分布的两个尾部：一个在左侧，一个在右侧。

For example, if a coin is suspected of being biased but we do not know whether it favours heads or tails, we would use a two-tailed test. We are testing for bias in either direction.

例如，如果怀疑一枚硬币有偏倚，但我们不知道它偏向正面还是反面，就应使用双尾检验。我们是在检验任何方向的偏倚。

---

2. When to Use a Two-Tailed Test | 何时使用双尾检验

Use a two-tailed test when the claim or research question does not specify a direction. Key phrases include “different from”, “changed”, “not equal to”, “has an effect”, or “there is a difference”.

当研究问题没有指明方向时，应使用双尾检验。关键提示词包括“不同于”、“改变了”、“不等于”、“有效果”或“存在差异”。

• English: “A new drug changes the mean reaction time.” → Two-tailed because “changes” is not directional.
• 中文：“一种新药改变了平均反应时间。”→ 双尾，因为“改变”没有方向。
• English: “The proportion of voters has changed since last year.” → Two-tailed.
• 中文：“选民比例自去年以来发生了变化。”→ 双尾。

If the wording says “greater than”, “less than”, “increased”, or “decreased”, then a one-tailed test is appropriate. Direction matters.

如果措辞是“大于”、“小于”、“增加”或“减少”，则应使用单尾检验。方向是关键。

---

3. Null and Alternative Hypotheses | 原假设与备择假设

For a two-tailed test, the null hypothesis always includes an equality, and the alternative hypothesis uses the not-equal sign.

对于双尾检验，原假设总是包含等号，而备择假设使用不等号。

H₀ : p = p₀   vs   H₁ : p ≠ p₀

Here p₀ is the claimed value of the population proportion or parameter. The alternative H₁ covers both p < p₀ and p > p₀, so the critical region must be split.

其中 p₀ 是总体比例或参数的声称值。备择假设 H₁ 同时覆盖 p < p₀ 和 p > p₀，因此临界区域必须分开。

For a mean μ, the hypotheses would be:

对于均值 μ，假设为：

H₀ : μ = μ₀   vs   H₁ : μ ≠ μ₀

---

4. Significance Level and the Split | 显著性水平与分割

In a two-tailed test, the significance level α is divided equally between the two tails. If α = 0.05, then each tail has probability 0.025.

在双尾检验中，显著性水平 α 被平均分配到两个尾部。如果 α = 0.05，则每个尾部的概率为 0.025。

This is a common source of error: students sometimes use the full α in one tail. Remember: for two-tailed tests, always split α in half.

这是常见的错误来源：学生有时会在一个尾部使用完整的 α。请记住：对于双尾检验，始终将 α 对半分割。

Test Type | 检验类型	Significance Level α | 显著性水平 α	Probability in Each Tail | 每个尾部的概率
One-tailed | 单尾	0.05	0.05 in one tail | 一个尾部 0.05
Two-tailed | 双尾	0.05	0.025 in each tail | 每个尾部 0.025
Two-tailed | 双尾	0.01	0.005 in each tail | 每个尾部 0.005

---

5. Critical Region for Discrete Distributions | 离散分布下的临界区域

When working with binomial distributions, the critical region for a two-tailed test is found by identifying the smallest and largest values of the test statistic that are sufficiently unlikely. The total probability of the critical region should be less than or equal to α, with each tail having probability at most α/2.

在使用二项分布时，双尾检验的临界区域通过找到足够不可能发生的检验统计量的最小值和最大值来确定。临界区域的总概率应小于或等于 α，每个尾部的概率至多为 α/2。

For example, let X ~ B(20, 0.4) and test H₀ : p = 0.4 vs H₁ : p ≠ 0.4 at α = 0.05. We need to find c₁ and c₂ such that P(X ≤ c₁) ≤ 0.025 and P(X ≥ c₂) ≤ 0.025.

例如，设 X ~ B(20, 0.4)，在 α = 0.05 下检验 H₀ : p = 0.4 对 H₁ : p ≠ 0.4。我们需要找到 c₁ 和 c₂，使得 P(X ≤ c₁) ≤ 0.025 且 P(X ≥ c₂) ≤ 0.025。

Using binomial tables: P(X ≤ 3) = 0.0160, P(X ≤ 4) = 0.0510. So the lower tail is X ≤ 3. For the upper tail, P(X ≥ 12) = 1 − P(X ≤ 11) = 1 − 0.9435 = 0.0565, and P(X ≥ 13) = 1 − P(X ≤ 12) = 1 − 0.9790 = 0.0210. So the upper tail is X ≥ 13.

使用二项分布表：P(X ≤ 3) = 0.0160，P(X ≤ 4) = 0.0510。因此下尾部为 X ≤ 3。对于上尾部，P(X ≥ 12) = 1 − P(X ≤ 11) = 1 − 0.9435 = 0.0565，而 P(X ≥ 13) = 1 − P(X ≤ 12) = 1 − 0.9790 = 0.0210。因此上尾部为 X ≥ 13。

Critical region: X ≤ 3 or X ≥ 13

临界区域：X ≤ 3 或 X ≥ 13

---

6. p-Value Method for Two-Tailed Tests | 双尾检验的 p 值方法

The p-value for a two-tailed test is calculated differently depending on the observed test statistic. For a symmetric distribution like the normal distribution, the p-value is twice the tail probability beyond the observed value.

双尾检验的 p 值根据观察到的检验统计量以不同方式计算。对于正态分布等对称分布，p 值是观测值之外尾部概率的两倍。

If the observed test statistic is z, then:

如果观测到的检验统计量为 z，则：

p-value = 2 × P(Z > |z|)   or   p-value = 2 × P(Z < −|z|)

p 值 = 2 × P(Z > |z|) 或 p 值 = 2 × P(Z < −|z|)

For discrete distributions, the p-value is the sum of probabilities for all outcomes that are as extreme or more extreme than the observed value, in both tails.

对于离散分布，p 值是所有与观测值一样极端或更极端的（两个尾部的）结果的概率之和。

Decision rule: if p-value ≤ α, reject H₀. Otherwise, do not reject H₀.

判定规则：如果 p 值 ≤ α，则拒绝 H₀。否则，不拒绝 H₀。

---

7. Symmetry and Doubling the Tail | 对称性与尾部加倍

In a two-tailed test using a normal or t-distribution, the symmetry of the distribution allows us to double the one-tailed p-value. However, for binomial or Poisson distributions, the distribution may be skewed, so we cannot simply double a one-tailed probability.

在使用正态分布或 t 分布的双尾检验中，分布的对称性允许我们将单尾 p 值加倍。然而，对于二项分布或泊松分布，分布可能偏斜，因此我们不能简单地加倍单尾概率。

For skewed distributions, the p-value is the sum of the probability of the observed value and all values less likely (in either direction) under H₀. This requires careful table work or software.

对于偏斜分布，p 值是观测值的概率与在 H₀ 下所有更不可能发生的值（两个方向）的概率之和。这需要仔细查表或使用软件。

Always check whether the distribution is symmetric before doubling. This is a frequent trap in exams.

在加倍之前，始终检查分布是否对称。这是考试中常见的陷阱。

---

8. Example: Proportion of Voters | 示例：选民比例

A political researcher claims that the proportion of voters supporting a policy has changed from the previous value of 0.6. A random sample of 100 voters finds 52 supporters. Test at the 5% significance level.

一位政治研究员声称，支持某项政策的选民比例已从此前的 0.6 发生了变化。随机抽取 100 名选民，其中 52 人支持。在 5% 显著性水平下进行检验。

Step 1: Hypotheses. H₀ : p = 0.6, H₁ : p ≠ 0.6. This is two-tailed.

步骤 1：建立假设。H₀ : p = 0.6，H₁ : p ≠ 0.6。这是双尾检验。

Step 2: Test statistic. The sample proportion is 0.52. Using the normal approximation:

步骤 2：计算检验统计量。样本比例为 0.52。使用正态近似：

z = (0.52 − 0.60) / √(0.60 × 0.40 / 100) = (−0.08) / 0.0490 = −1.633

Step 3: Critical values at α = 0.05, two-tailed: z = ±1.96. Since −1.633 lies between −1.96 and +1.96, it is not in the critical region.

步骤 3：在 α = 0.05 双尾下的临界值为 z = ±1.96。由于 −1.633 位于 −1.96 和 +1.96 之间，未落入临界区域。

Step 4: Conclusion. Do not reject H₀. There is insufficient evidence that the proportion has changed.

步骤 4：结论。不拒绝 H₀。没有充分证据表明比例发生了变化。

---

9. Common Mistakes and How to Avoid Them | 常见错误与避免方法

• English: Using the full significance level in one tail instead of splitting it. Always divide α by 2.
• 中文：在单侧尾部使用完整的显著性水平，而不是对半分割。始终将 α 除以 2。
• English: Writing H₁ with a directional sign (< or >) when the question asks for a two-tailed test. Always use ≠.
• 中文：当题目要求双尾检验时，却在 H₁ 中写了方向符号（< 或 >）。始终使用 ≠。
• English: Doubling a p-value for a skewed binomial distribution without thinking. Check symmetry first.
• 中文：对于偏斜的二项分布，不经思考就直接将 p 值加倍。先检查对称性。
• English: Confusing “critical region” with “acceptance region”. The critical region leads to rejection of H₀.
• 中文：混淆“临界区域”和“接受区域”。临界区域导致拒绝 H₀。

When in doubt, return to the wording of the problem. If no direction is stated, choose two-tailed.

如有疑问，回到题目措辞。如果没有指明方向，就选择双尾检验。

---

10. Relationship with Confidence Intervals | 与置信区间的关系

A two-tailed hypothesis test at significance level α is equivalent to checking whether the null value lies inside a (1 − α) × 100% confidence interval. If the null value is outside the interval, we reject H₀.

显著性水平为 α 的双尾检验等价于检查原假设值是否落在 (1 − α) × 100% 置信区间内。如果原假设值在区间之外，则拒绝 H₀。

For the voter example, a 95% confidence interval for the proportion is approximately:

对于选民示例，比例的 95% 置信区间约为：

0.52 ± 1.96 × √(0.52 × 0.48 / 100) = 0.52 ± 0.0980 = (0.422, 0.618)

Since 0.6 lies inside this interval, we do not reject H₀. This matches the conclusion from the critical region method.

由于 0.6 位于该区间内，我们不拒绝 H₀。这与临界区域法的结论一致。

---

11. Exam-Style Workflow Summary | 考试式解题流程总结

Follow these steps for any two-tailed test in the exam:

在考试中，任何双尾检验都遵循以下步骤：

1. English: Identify from the question whether direction is specified. If not, use a two-tailed test.
2. 中文：从题目中判断是否指明了方向。如果没有，使用双尾检验。
3. English: Write down H₀ and H₁, with H₁ using ≠.
4. 中文：写出 H₀ 和 H₁，其中 H₁ 使用 ≠。
5. English: State the significance level and divide α by 2 for each tail.
6. 中文：说明显著性水平，并将 α 除以 2 用于每个尾部。
7. English: Find the critical values or critical region using tables or a calculator.
8. 中文：使用查表或计算器找到临界值或临界区域。
9. English: Compute the test statistic from the sample data.
10. 中文：根据样本数据计算检验统计量。
11. English: Compare and conclude with a contextual sentence.
12. 中文：比较并给出结合语境的结论句。

Always include a conclusion that refers to the original context, not just “reject H₀” or “do not reject H₀”.

始终写出结合原始语境的结论，而不仅仅是“拒绝 H₀”或“不拒绝 H₀”。

---

12. Quick Decision Guide: One-Tailed or Two-Tailed? | 快速判断指南：单尾还是双尾？

Wording in Question | 题目措辞	Type of Test | 检验类型
is greater than / increased / higher | 大于 / 增加 / 更高	One-tailed (upper) | 单尾（上尾）
is less than / decreased / lower | 小于 / 减少 / 更低	One-tailed (lower) | 单尾（下尾）
has changed / different / not equal | 已改变 / 不同 / 不等于	Two-tailed | 双尾
has an effect / affects | 有效果 / 影响	Two-tailed | 双尾

Remember that “changed” can be an increase or a decrease, so it is always two-tailed unless a direction is explicitly given.

请记住，“改变”可能是增加也可能是减少，因此除非明确给出了方向，否则始终是双尾检验。

By mastering these criteria, you can confidently determine and perform two-tailed tests in A-Level exams. Practice with past papers and always check your hypotheses for the correct sign.

掌握这些判定标准后，你就能在 A-Level 考试中自信地判断并完成双尾检验。多练习历年真题，并始终检查你的假设符号是否正确。

Published by TutorHao | Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）