Confidence Intervals and Tests Using the t-Distribution | 使用 t 分布的置信区间与假设检验

📚 Confidence Intervals and Tests Using the t-Distribution | 使用 t 分布的置信区间与假设检验

In A-Level Statistics, the normal distribution often serves as the foundation for confidence intervals and hypothesis tests about a population mean. However, this approach requires knowing the population standard deviation σ. In practice, σ is rarely known and must be estimated by the sample standard deviation s. When we replace σ with s, the sampling distribution of the mean follows a t-distribution rather than a normal distribution, provided the underlying population is normal or the sample size is reasonably large. This article covers all the t-based procedures required for Edexcel A-Level Mathematics: one-sample t intervals and tests, paired t procedures, and two-sample t intervals and tests assuming equal variances. You will learn how to construct confidence intervals, perform hypothesis tests, interpret p-values and critical values, and check the necessary assumptions.

在 A-Level 统计学中，正态分布通常是关于总体均值的置信区间和假设检验的基础。然而，这种方法需要知道总体标准差 σ。在实际中，σ 鲜为人知，必须用样本标准差 s 来估计。当用 s 代替 σ 时，均值的抽样分布服从 t 分布而非正态分布，前提是总体呈正态或样本量足够大。本文涵盖了 Edexcel A-Level 数学要求的所有基于 t 的程序：单样本 t 区间和检验、配对 t 程序以及假设方差相等的双样本 t 区间和检验。您将学会构建置信区间、进行假设检验、解释 p 值与临界值，以及检查必要的假设条件。

---

1. Introduction to the t-Distribution | t 分布简介

The t-distribution (also called Student's t) is a continuous probability distribution that is symmetric and bell-shaped like the normal distribution, but has heavier tails. It arises naturally when we standardise a sample mean using the sample standard deviation s instead of the true σ. The additional uncertainty from estimating σ makes extreme values more likely, so the t-distribution provides wider and more accurate confidence intervals and more conservative tests when σ is unknown.

t 分布（亦称学生 t 分布）是一种对称的钟形连续概率分布，类似于正态分布，但尾部更重。当我们用样本标准差 s 而非真实的 σ 对样本均值进行标准化时，该分布自然出现。由估计 σ 带来的额外不确定性使得极端值更有可能出现，因此当 σ 未知时，t 分布能够提供更宽、更准确的置信区间以及更保守的检验。

1. As the degrees of freedom increase, the t-distribution approaches the standard normal distribution. For large n (typically n > 30), the two distributions are virtually indistinguishable, which is why many textbooks suggest using z-procedures for large samples even when σ is unknown. However, the Edexcel specification expects you to use the t-distribution whenever σ is estimated, regardless of sample size.

t 分布的形状取决于一个称为自由度的参数，通常用 ν 或 df 表示。对于容量为 n 的单样本，df = n − 1。随着自由度的增加，t 分布趋近于标准正态分布。当 n 较大时（通常 n > 30），两种分布几乎无法区分，这就是许多教科书建议在大样本且 σ 未知时使用 z 程序的原因。但 Edexcel 考纲要求，只要 σ 需要估计，就必须使用 t 分布，无论样本大小。

---

2. Degrees of Freedom and Shape | 自由度与形状

The term 'degrees of freedom' squared deviations by (n − 1) instead of n. This loses one degree of freedom because the deviations are constrained to sum to zero once x̄ has been computed. Consequently, the t-distribution for a sample of size n has df = n − 1.

“自由度”一词与用于估计参数时所使用的独立信息数量有关。在计算样本方差 s² 时，我们用 (n − 1) 而非 n 去除偏差平方和。这就损失了一个自由度，因为一旦算出 x̄，偏差的总和就必须为零。因此，容量为 n 的样本的 t 分布自由度为 df = n − 1。

r tails, meaning a larger critical value is needed to capture the same central probability. For example, the 95% nding t-value with 5 df is approximately 2.571 and with 20 df is about 2.086. The table of critical values provided in the Edexcel formula booklet gives t-values for various df and tail probabilities.

较低的自由度会使 t 分布具有明显更厚的尾部，意味着需要更大的临界值才能覆盖相同的中央概率。例如，标准正态的 95% 临界值是 z* = 1.96，而自由度为 5 的对应 t 值约为 2.571，自由度为 20 时约为 2.086。Edexcel 公式册中提供的临界值表列出了不同自由度和尾部概率下的 t 值。

---

3. When to Use the t-Distribution | 何时使用 t 分布

You must use the t-distribution when constructing a confidence interval or performing a hypothesis test for a population mean (or for a difference between means) and the population standard deviation σ is unknown. This is the rule for Edexcel A-Level. Even if the sample size is large enough to invoke the Central Limit Theorem, the exam board still expects you to use t-procedures unless σ is given.

当为总体均值（或均值差）构建置信区间或进行假设检验，且总体标准差 σ 未知时，必须使用 t 分布。这是 Edexcel A-Level 的规定。即使样本量足够大而适用中心极限定理，考试局依然期望你使用 t 程序，除非给出了 σ。

The three main scenarios examined in Edexcel are: (i) one-sample inference for a population mean μ; (ii) paired data, where the difference d is analysed as a single sample of differences; (iii) two independent samples, assuming the two populations have a common variance σ² (pooled estimate). All three rely on the t-distribution with appropriate degrees of freedom.

Edexcel 考试中涉及的三种主要情况为：(i) 单一总体均值 μ 的推断；(ii) 配对数据，此时将差值 d 作为单一的差值样本进行分析；(iii) 两个独立样本，并假设两个总体具有共同方差 σ²（合并估计）。这三种情况均依赖于具有适当自由度的 t 分布。

---

4. One-Sample t Confidence Interval | 单样本 t 置信区间

For a random sample of size n drawn from a normal population with unknown mean μ and unknown standard deviation, a 100(1 − α)% confidence interval for μ is given by:

对于来自均值 μ 和标准差均未知的正态总体的容量为 n 的随机样本，μ 的 100(1 − α)% 置信区间为：

x̄ ± t_{n−1, α/2} × (s / √n)

where x̄ is the sample mean, s is the sample standard deviation, and t_{n−1, α/2} om such that the area in each tail is α/2. Equivalently, the interval can be written as [x̄ − t* × s/√n, x̄ + t* × s/√n].

其中 x̄ 为样本均值，s 为样本标准差，t_{n−1, α/2} 是自由度为 (n − 1) 的 t 分布的临界值，满足每个尾部面积为 α/2。等价地，该区间可写为 [x̄ − t* × s/√n, x̄ + t* × s/√n]。

To find the critical value, you can use the t-tables supplied in the exam. For a 95% responds to a two-tail probability of 0.05 with df = n − 1. The margin of error is t* × s/√n. Notice that as the sample size grows, s/√n decreases and the t-critical value approaches z*, so the interval narrows.

要找到临界值，可以使用考试中提供的 t 值表。对于 95% 置信区间，α = 0.05，查找对应双尾概率 0.05 且 df = n − 1 的值。误差范围是 t* × s/√n。注意，随着样本量增大，s/√n 减小，t 临界值趋近 z*，因此区间变窄。

---

5. One-Sample t Hypothesis Test | 单样本 t 假设检验

ve can be two-tailed (H₁: μ ≠ μ₀) or one-tailed (H₁: μ < μ₀ or μ > μ₀). The test statistic is:

单样本 t 检验用于判断总体均值 μ 是否与假设值 μ₀ 存在差异。原假设为 H₀: μ = μ₀，备择假设可为双尾 (H₁: μ ≠ μ₀) 或单尾 (H₁: μ < μ₀ 或 μ > μ₀)。检验统计量为：

t = (x̄ − μ₀) / (s / √n)

ou then compare the computed t with the critical value from tables or find the p-value. If |t| e significance level α, you reject H₀. Otherwise, there is insufficient evidence to reject H₀.

在 H₀ 下，该统计量服从自由度为 n − 1 的 t 分布。然后将计算出的 t 值与表中的临界值进行比较，或求出 p 值。若 |t| 超过临界值（双尾），或 p 值小于显著性水平 α，则拒绝 H₀。否则，没有足够证据拒绝 H₀。

ons, for a two-tailed test at level α, the rejection region is t < −c or t > c where c = t_{n−1, α/2}. For a one-tailed test, all of α is placed in one tail. The interpretation must be stated in the context of the problem, for example, "There is evidence at the 5% level that the true mean differs from 50."

在使用临界域时，对于显著性水平为 α 的双尾检验，拒绝域为 t < −c 或 t > c，其中 c = t_{n−1, α/2}。对于单尾检验，所有的 α 都被置于单尾中。结论必须结合问题背景陈述，例如：“在 5% 水平上，有证据表明真实均值与 50 存在差异。”

---

广告 / Sponsored

6. Example: One-Sample t Procedures | 示例：单样本 t 程序

heir masses: 198, 202, 199, 197, 203, 200, 196, 201, 195, 204 (grams). The sample mean is x̄ = 199.5 g and the sample standard deviation is s = 3.03 g. Construct a 95% confidence interval for μ and test whether the true mean is 200 g at the 5% significance level.

假设某制造商声称一包饼干平均重 200 克。一个消费者组织随机抽取了 10 包饼，记录质量：198, 202, 199, 197, 203, 200, 196, 201, 195, 204（克）。样本均值 x̄ = 199.5 g，样本标准差 s = 3.03 g。构建 μ 的 95% 置信区间，并在 5% 显著性水平下检验真实均值是否为 200 g。

For n = 10, df = 9. From t-tables, the two-tailed 5% critical value t_{9, 0.025} = 2.262. The standard error is s/√n = 3.03 / √10 ≈ 0.958. The margin of error is 2.262 × 0.958 ≈ 2.166. The 95% = (197.3, 201.7) g. Since 200 lies well inside this interval, we do not reject H₀ at the 5% level.

n = 10，df = 9。查 t 表，双尾 5% 临界值 t_{9, 0.025} = 2.262。标准误 s/√n = 3.03/√10 ≈ 0.958。误差范围 2.262 × 0.958 ≈ 2.166。95% 置信区间为 (199.5 − 2.166, 199.5 + 2.166) = (197.3, 201.7) g。因为 200 完全落在该区间内，我们在 5% 水平上不拒绝 H₀。

Formally, the test statistic is t = (199.5 − 200) / 0.958 ≈ −0.522. Since |−0.522| ter than 0.05. We conclude that there is insufficient evidence to dispute the manufacturer' nterval and the hypothesis test lead to the same conclusion; this is an important duality.

正式的检验统计量为 t = (199.5 − 200) / 0.958 ≈ −0.522。由于 |−0.522| < 2.262，结果不显著。p 值（查表或计算器）大于 0.05。我们得出结论，没有足够证据质疑厂商的说法。置信区间和假设检验得出相同结论；这是一个重要的对偶关系。

---

7. Paired t-Test | 配对 t 检验

A paired t-test is used when we have two measurements on the same individual or matched pairs. Examples include before-and-after studies, or tests where twins are compared. Instead of treating the two sets of observations as independent, we calculate the differences d_i = x_i − y_i for each pair, and then perform a one-sample t-test on the mean difference μ_d. Typically, H₀: μ_d = 0, and H₁ may be two-sided or one-sided.

当对同一个体进行两次测量或存在匹配配对时，使用配对 t 检验。例如前后对比研究，或比较双胞胎的测试。我们并不将两组观测视为独立，而是计算每对的差值 d_i = x_i − y_i，然后对差值均值 μ_d 进行单样本 t 检验。通常 H₀: μ_d = 0，H₁ 可为双尾或单尾。

The test statistic is t = (d̄ − μ_d₀) / (s_d / √n), where n is the number of pairs, d̄ is the mean of the differences, and s_d ibution with df = n − 1. The critical values are found exactly as for a one-sample t-test.

检验统计量为 t = (d̄ − μ_d₀) / (s_d / √n)，其中 n 为配对数量，d̄ 为差值均值，s_d 为差值的标准差。在 H₀ 下，t 服从自由度为 n − 1 的 t 分布。临界值的查找与单样本 t 检验完全相同。

Pairing helps to remove variability between subjects, often increasing the power of the test. It is vital to recognise when data are paired. A common mistake is to treat paired data as two independent samples, which can lead to incorrect standard errors and loss of statistical power.

配对有助于消除受试者之间的变异性，常能提高检验的功效。关键是要能识别配对数据。常见的错误是将配对数据当作两个独立样本处理，这将导致标准误计算错误并损失统计功效。

---

8. Confidence Interval for Paired Mean Difference | 配对均值差的置信区间

A 100(1 − α)% confidence interval for the true mean difference μ_d is constructed analogously to the one-sample case:

真实差值均值 μ_d 的 100(1 − α)% 置信区间与单样本情况类似，公式为：

d̄ ± t_{n−1, α/2} × (s_d / √n)

This interval provides a range of plausible values for the population mean difference. If zero lies within the interval, then at the corresponding confidence level, there is no significant difference between the two measurements. If the entire interval is positive (or negative), that suggests a significant directional effect. This interval technique is often required in exam questions involving paired data.

该区间提供了总体均值差的合理范围。若零包含在区间内，则在相应的置信水平下，两次测量之间无显著差异。若整个区间为正（或为负），则表明存在显著的方向性效应。涉及配对数据的考题常要求使用这种区间技术。

---

9. Two-Sample t-Test (Equal Variances) | 双样本 t 检验（等方差）

-test. The null hypothesis is H₀: μ₁ = μ₂, and the alternative can be one-tailed or two-tailed. The assumption of equal variances, σ²₁ = σ²₂, is part of the standard A-Level model. You are not required to test this assumption formally, but you should state it.

当比较两个方差未知但假设相等、且相互独立的总体的均值（μ₁ 和 μ₂）时，Edexcel 使用合并方差的双样本 t 检验。原假设为 H₀: μ₁ = μ₂，备择假设可为单尾或双尾。等方差假设 σ²₁ = σ²₂ 是标准 A-Level 模型的一部分。你不必对该假设进行正式检验，但应当说明。

The pooled estimate of the common standard deviation, s_p, is given by:

共同标准差的合并估计值 s_p 由下式给出：

s_p² = [(n₁ − 1)s₁² + (n₂ − 1)s₂²] / (n₁ + n₂ − 2)

The test statistic is then:

于是检验统计量为：

t = (x̄₁ − x̄₂) / [s_p × √1/n₁ + 1/n₂]

d p-values are obtained using this df. The pooled approach yields a more precise estimate of variability than treating the samples separately, provided the equal-variance assumption holds.

在 H₀ 下，t 服从自由度为 n₁ + n₂ − 2 的 t 分布。临界值和 p 值得据此自由度获取。只要等方差假设成立，合并方法就能给出比分别处理更精确的变异性估计。

---

10. Confidence Interval for Difference of Two Means | 两均值差的置信区间

A 100(1 − α)% confidence interval for μ₁ − μ₂ using the pooled method is:

使用合并方法，μ₁ − μ₂ 的 100(1 − α)% 置信区间为：

(x̄₁ − x̄₂) ± t_{ν, α/2} × s_p × √1/n₁ + 1/n₂

terpreted in the same way: if zero is inside the interval, the difference is not significant at the chosen level. The interval also provides information about the magnitude of the difference, which is more informative than a simple test result.

其中 ν = n₁ + n₂ − 2。该区间的解读方式相同：若零落在区间内，则差值在所选水平上不显著。区间同时提供了差值大小的信息，比单纯的检验结果更具参考价值。

---

11. Assumptions and Model Checking | 假设条件与模型检验

All t-procedures assume that the data are a random sample from a normal distribution. For paired tests, the differences should be normally distributed. For two-sample tests with pooled variance, the two populations are assumed independent, normal, and with equal variance. Moderate departures from normality are acceptable, especially when sample sizes are reasonably large because the

Published by TutorHao | A-Level Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）