Understanding Statistical Power in Hypothesis Testing | 假设检验中的统计功效解析

📚 Understanding Statistical Power in Hypothesis Testing | 假设检验中的统计功效解析

In A‑Level Mathematics, particularly within the Edexcel specification, hypothesis testing plays a central role in statistical inference. Among the key concepts, the idea of 'power' often causes confusion yet is essential for evaluating the quality of a test. This article explores the different aspects of statistical power, from its definition and relationship with errors to the factors that influence it, illustrated with examples from binomial and normal distributions. By the end, you should be able to interpret power curves and understand how to design more effective tests.

在 A‑Level 数学中，尤其是 Edexcel 考纲里，假设检验是统计推断的核心内容。在关键概念中，“功效”这一概念常常令人困惑，但对于评价一个检验的质量却至关重要。本文将从功效的定义、与错误类型的关系，到影响功效的各种因素，并结合二项分布与正态分布的实例，全面解析统计功效的不同侧面。读完后，你将能够解读功效曲线，并理解如何设计更有效的检验。

1. What Is Statistical Power? | 什么是统计功效？

Statistical power is the probability that a hypothesis test correctly rejects the null hypothesis (H₀) when the alternative hypothesis (H₁) is true. In other words, it measures the test's ability to detect an effect or difference when one genuinely exists. A test with high power is more likely to produce a significant result when H₀ is false, reducing the risk of missing a real finding.

统计功效是指当备择假设 (H₁) 为真时，假设检验正确地拒绝原假设 (H₀) 的概率。换句话说，它衡量的是检验在真实效应或差异存在时能够检测出该效应的能力。功效高的检验在 H₀ 不成立时更有可能产生显著结果，从而减少错过真实发现的风险。

2. Type I and Type II Errors | 第一类错误与第二类错误

To fully grasp power, we must first clarify the two types of errors that can occur in hypothesis testing. A Type I error happens when we reject a true H₀; its probability is denoted by α (the significance level). A Type II error occurs when we fail to reject a false H₀; its probability is represented by β. Power is defined as 1 − β, the complement of the Type II error rate.

要完全理解功效，我们必须先厘清假设检验中可能发生的两类错误。第一类错误发生在 H₀ 为真却被拒绝时，其概率用 α（显著性水平）表示。第二类错误发生在 H₀ 为假却未被拒绝时，其概率用 β 表示。功效定义为 1 − β，即第二类错误概率的补数。

The following table summarises the two error types and their relationship with power:

下表总结了两类错误及其与功效的关系：

Decision	H₀ True	H₀ False
Reject H₀	Type I error (α)	Correct decision (Power = 1−β)
Do not reject H₀	Correct decision (1−α)	Type II error (β)

---

广告 / Sponsored

3. The Relationship between Power and Type II Error | 功效与第二类错误的关系

ces β will increase power. Type II errors often occur when the sample size is too small, the effect size is subtle, or the test is not sensitive enough. Understanding this relationship helps us choose an appropriate sample size and significance level to control both α and β, balancing the risks of false positives and false negatives.

因为功效 = 1 − β，任何减少 β 的措施都会提高功效。第二类错误通常发生在样本量过小、效应规模细微或检验不够灵敏的情况下。理解这一关系有助于我们选择合适的样本量和显著性水平，以控制 α 和 β，在误报与漏报的风险之间取得平衡。

4. Factors Affecting Statistical Power | 影响统计功效的因素

Several factors directly influence the power of a test: the significance level α, the sample size n, the true population effect size, and the nature of the test (one‑tailed versus two‑tailed). Each factor changes the shape and position of the power function, which plots power against different values of the parameter under H₁. Let us examine each in turn.

有几个因素直接影响检验的功效：显著性水平 α、样本量 n、真实的总体效应规模以及检验的类型（单尾或双尾）。每个因素都会改变功效函数的形状和位置，功效函数绘出了对应 H₁ 下不同参数取值时的功效。我们逐一进行考察。

5. Effect of Sample Size on Power | 样本大小对功效的影响

Increasing the sample size generally raises statistical power. Larger samples provide more information about the population, reduce the standard error of the test statistic, and make it easier to detect departures from H₀. For instance, with a fixed significance level and effect size, doubling the sample size may dramatically improve the probability of rejecting a false null hypothesis.

增大样本量通常会提高统计功效。更大的样本提供了更多关于总体的信息，降低了检验统计量的标准误，从而更容易检测到偏离 H₀ 的情况。例如，在固定显著水平和效应规模的情况下，将样本量翻倍可能会显著提升拒绝错误原假设的概率。

6. Effect Size and Power | 效应大小与功效

Effect size refers to the magnitude of the difference or relationship specified by the alternative hypothesis. Power is higher when the true effect size is large because the test statistic is more likely to fall in the rejection region. Conversely, if the true effect is tiny, even a well‑designed test may lack sufficient power to detect it without a very large sample.

效应大小指的是备择假设所指定的差异或关系的量级。当真实效应较大时，检验统计量更有可能落入拒绝域，因此功效更高。相反，若真实效应非常微小，即便检验设计良好，如果没有非常大的样本，也可能缺乏足够的功效来检测它。

7. Choice of Significance Level | 显著性水平的选择

A smaller α (e.g., 0.01 instead of 0.05) makes it harder to reject H₀, thus reducing power for a given sample size and effect. While a smaller α lowers the chance of a Type I error, it simultaneously increases β. In practice, researchers must weigh the consequences of both types of errors when setting α.

较小的 α（例如 0.01 而非 0.05 ）使得拒绝 H₀ 更难，从而在给定样本量和效应下降低功效。虽然较小的 α 减少了第一类错误的机会，但它同时也增大了 β。在实践中，研究人员必须在设定 α 时权衡两类错误的后果。

8. One‑tailed vs Two‑tailed Tests | 单尾与双尾检验

A one‑tailed test concentrates the rejection region entirely in one tail of the distribution, which provides greater power to detect an effect in the specified direction than a two‑tailed test of the same α level. However, a one‑tailed test forfeits the ability to detect effects in the opposite direction. Choosing between the two involves assumptions about the direction of the effect and affects the resulting power.

单尾检验将拒绝域完全集中在分布的一侧尾部，这使得在相同 α 水平下，它比双尾检验具备更高效力来检测特定方向上的效应。然而，单尾检验牺牲了检测相反方向效应的能力。两者之间的选择涉及对效应方向的假设，并影响最终的检验功效。

9. Distribution of Test Statistic: Binomial and Normal | 检验统计量的分布：二项与正态

In Edexcel A‑Level, you encounter hypothesis tests based on the binomial distribution (for proportions) and the normal distribution (for means). The power calculations differ because the critical values and probability calculations depend on the distribution. For a binomial test with a given sample size n and hypothesised probability p₀, the power for a specific alternative p₁ can be found by summing the probabilities of the critical region under the true p₁. For a normal test, power is computed using the non‑central Z‑distribution or standard normal tables after adjusting for the true mean.

在 Edexcel A‑Level 中，你会遇到基于二项分布（用于比例）和正态分布（用于均值）的假设检验。功效计算的差异在于临界值和概率计算取决于所使用的分布。对于给定样本量 n 和假设概率 p₀ 的二项检验，在特定备择概率 p₁ 下的功效可以通过在真实 p₁ 下对拒绝域的概率求和而得到。对于正态检验，功效可利用非中心 Z 分布或通过调整真实均值后查标准正态表来计算。

10. Simple Example of Calculating Power | 计算功效的简单例子

.05. The critical region is X ≥ 10. If the true proportion p₁ = 0.5, the power is P(X ≥ 10 | is probability can be found from tables or a calculator: about 0.588, meaning roughly 58.8% = 0.7, the power rises to about 0.965, illustrating how power increases with effect size.

考虑一个二项检验：H₀: p = 0.3 对 H₁: p > 0.3，其中 n = 20，α ≈ 0.05。拒绝域为 X ≥ 10。如果真实比例 p₁ = 0.5，功效即为 P(X ≥ 10 | p = 0.5)，使用 B(20, 0.5)。该概率可从表或计算器得到：大约 0.588，意味着正确拒绝 H₀ 的机会约为 58.8%。若 p₁ = 0.7，功效升至约 0.965，直观显示了功效如何随效应大小而提高。

11. Power Function | 功效函数

The power function K(θ) gives the probability of rejecting H₀ as a function of the true parameter θ. For θ in the region of H₀, K(θ) equals the probability of a Type I error (usually α). For θ moving away from H₀, K(θ) ideally increases towards 1. Plotting K(θ) against θ yields a power curve, which is a valuable visual tool for comparing different tests or sample sizes.

功效函数 K(θ) 给出了拒绝 H₀ 的概率随真实参数 θ 变化的函数。当 θ 处于 H₀ 区域时，K(θ) 等于第一类错误的概率（通常为 α）。随着 θ 偏离 H₀，理想的 K(θ) 应趋向于 1。将 K(θ) 对 θ 作图就得到功效曲线，这是比较不同检验或样本量时非常有用的可视化工具。

12. How to Increase Power in Practice | 如何在实际中提高功效

To boost power, you can increase the sample size, relax α slightly (though carefully), choose a one‑tailed test if directional, or improve the measurement precision to reduce variability. In experimental design, ensuring a large enough effect size through stronger interventions also helps. Always bear in mind that power analysis is a crucial step before data collection to guarantee meaningful results.

要提高功效，你可以增大样本量，稍微放宽 α（但需谨慎），如果方向明确就选择单尾检验，或改进测量精度以减少变异性。在实验设计中，确保通过更强的干预得到足够大的效应规模也有帮助。请始终牢记，功效分析是数据收集之前确保结果有意义的关键步骤。

Published by TutorHao | Edexcel A‑Level Maths Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）