📚 Hypothesis Testing for the Difference Between Means | 两个均值差的假设检验
In A-Level Mathematics, hypothesis testing for the difference between two population means allows us to compare two groups using sample data. Common scenarios include comparing the effectiveness of two teaching methods, the mean lifetimes of two brands of batteries, or before‑and‑after measurements in a paired design. Edexcel exam questions expect you to choose the correct test, state hypotheses clearly, compute the test statistic, refer to the appropriate distribution, and write a conclusion in context.
在A‑Level数学中，两个总体均值差的假设检验使我们能够利用样本数据比较两组对象。常见情境包括比较两种教学方法的有效性、两个品牌电池的平均寿命，或配对设计中的前后测量。Edexcel考试题目要求你选择正确的检验方法，清晰地陈述假设，计算检验统计量，引用正确的分布，并写出符合背景的结论。
1. Introduction to Two‑Sample Hypothesis Testing | 双样本假设检验简介
A two‑sample test for means investigates whether the difference between two population means, μ₁ − μ₂, takes a specific value (often zero). We collect independent random samples from each population and use their sample means, x̄₁ and x̄₂, to draw inferences. The central question is whether any observed difference between x̄₁ and x̄₂ is statistically significant or could easily have occurred by random sampling variation.
双样本均值检验研究两个总体均值之差 μ₁ − μ₂ 是否等于某个特定值（通常为零）。我们从每个总体中收集独立的随机样本，并利用样本均值 x̄₁ 和 x̄₂ 进行推断。核心问题是：x̄₁ 与 x̄₂ 之间观察到的差异是否具有统计显著性，还是可以轻易地由随机抽样波动造成。
2. Null and Alternative Hypotheses | 原假设与备择假设
means: H₀: μ₁ − μ₂ = 0, or equivalently μ₁ = μ₂. The alternative hypothesis H₁ can be two‑tailed (μ₁ ≠ μ₂), or one‑ e choice depends on the research question or the wording in the exam problem, for instance “test whether the mean for group 1 is greater than that for group 2”.
原假设 H₀ 通常表明两个总体均值没有差异：H₀: μ₁ − μ₂ = 0，或等价地 μ₁ = μ₂。备择假设 H₁ 可以是双尾的（μ₁ ≠ μ₂），也可以是单尾的（μ₁ < μ₂ 或 μ₁ > μ₂）。选择取决于研究问题或考题的措辞，例如“检验第一组的均值是否大于第二组的均值”。
3. Independent vs. Paired Samples | 独立样本与配对样本
Independent samples occur when the observations in one sample have no natural link to those in the other sample, for example measuring heights of boys and girls from separate classes. Paired samples arise when each observation in one sample is matched or naturally linked to one in the other, such as measuring the same individuals before and after a treatment. The analysis method differs: independent tests compare two sample means directly, while paired tests examine the mean of the differences.
当一个样本中的观测值与另一个样本没有自然联系时，即构成独立样本，例如分别测量两个班级中男孩与女孩的身高。当一个样本中的每个观测值与另一个样本中的观测值存在配对或自然关联时，即为配对样本，例如测量同一组个体在处理前和处理后的数据。分析方法有所不同：独立检验直接比较两个样本均值，而配对检验则考察差值序列的均值。
4. The Test Statistic for Known Variances (z‑test) | 已知方差时的检验统计量（z检验）
30) that the sample variances provide a good approximation, the sampling distribution of x̄₁ − x̄₂ is approximately normal. The z‑test statistic is:
z = (x̄₁ − x̄₂) / √σ₁²/n₁ + σ₂²/n₂
当两个总体的方差 σ₁² 和 σ₂² 已知，或者两个样本量都足够大（通常 n₁ ≥ 30 且 n₂ ≥ 30）使得样本方差提供良好近似时，x̄₁ − x̄₂ 的抽样分布近似正态。z检验统计量为：
z = (x̄₁ − x̄₂) / √σ₁²/n₁ + σ₂²/n₂
Under H₀, this z follows the standard normal distribution N(0,1). The critical value is obtained from normal tables, and the test is sometimes used as a large‑sample approximation even when variances are unknown.
在 H₀ 下，该 z 值服从标准正态分布 N(0,1)。临界值由正态分布表查得。即使方差未知，该检验也常在样本量较大时作为近似方法使用。
5. The Two‑Sample t‑Test with Pooled Variance | 合并方差的双样本t检验
ut the common variance is unknown, we estimate σ² using the pooled variance sₚ². The formula for sₚ² is:
sₚ² = [(n₁ − 1)s₁² + (n₂ − 1)s₂²] / (n₁ + n₂ − 2)
若假定两个总体具有相等的方差（σ₁² = σ₂² = σ²），但共同方差未知，则使用合并方差 sₚ² 来估计 σ²。sₚ² 的公式为：
sₚ² = [(n₁ − 1)s₁² + (n₂ − 1)s₂²] / (n₁ + n₂ − 2)
The test statistic then becomes a t‑statistic with degrees of freedom df = n₁ + n₂ − 2:
t = (x̄₁ − x̄₂) / (sₚ √1/n₁ + 1/n₂)
检验统计量变为自由度为 df = n₁ + n₂ − 2 的 t 统计量：
t = (x̄₁ − x̄₂) / (sₚ √1/n₁ + 1/n₂)
This pooled t‑test is the most common method in Edexcel papers when the question states or implies equal variances and provides summary statistics for two samples.
当考题明确或隐含方差相等，并给出两个样本的汇总统计量时，合并方差的双样本t检验是Edexcel试卷中最常用的方法。
6. Welch’s t‑Test for Unequal Variances | 不等方差下的Welch t检验
When we cannot assume equal population variances, Welch’s t‑test provides a more robust alternative. The test statistic is:
t = (x̄₁ − x̄₂) / √s₁²/n₁ + s₂²/n₂
当不能假定总体方差相等时，Welch t检验是一种更稳健的替代方法。检验统计量为：
t = (x̄₁ − x̄₂) / √s₁²/n₁ + s₂²/n₂
The degrees of freedom are approximated using the Satterthwaite formula:
df ≈ (s₁²/n₁ + s₂²/n₂)² / [ (s₁²/n₁)²/(n₁−1) + (s₂²/n₂)²/(n₂−1) ]
自由度由 Satterthwaite 公式近似得出：
df ≈ (s₁²/n₁ + s₂²/n₂)² / [ (s₁²/n₁)²/(n₁−1) + (s₂²/n₂)²/(n₂−1) ]
Edexcel questions may direct you to use Welch’s test when sample variances are markedly different or when the assumption of equal variances is not justified. The computed df is usually rounded down to the nearest integer for table lookup.
当样本方差差异明显或等方差假设不成立时，Edexcel题目可能会指导学生使用Welch检验。计算出的自由度通常向下取整以便查表。
7. Paired t‑Test | 配对t检验
For paired data, we reduce the two‑sample problem to a single‑sample t‑ fferences dᵢ = xᵢ − yᵢ for each pair. We then test H₀: μ_d = 0 against H₁: μ_d ≠ 0 (or one‑sided). The test statistic is:
t = d̄ / (sd / √n)
where d̄ is the mean of the differences, sd is their standard deviation, and n is the number of pairs. The statistic follows a t‑distribution with df = n − 1.
对于配对数据，我们将双样本问题转化为对每对差值 dᵢ = xᵢ − yᵢ 的单样本t检验。检验 H₀: μ_d = 0 对 H₁: μ_d ≠ 0（或单侧）。检验统计量为：
t = d̄ / (sd / √n)
其中 d̄ 为差值的均值，sd 为它们的标准差，n 为配对数量。该统计量服从自由度为 n − 1 的 t 分布。
8. Critical Values and Rejection Regions | 临界值与拒绝域
Once the test statistic is computed, compare it with the critical value from the relevant distribution (z or t). For a two‑tailed test at significance level α, the rejection region is |statistic| > critical value. For a one‑tailed test, the rejection region lies entirely in the direction of H₁. Insist on drawing a clear sketch showing the acceptance and rejection areas whenever possible, as Edexcel often awards marks for a correctly labelled diagram.
计算出检验统计量后，将其与相关分布（z或t）的临界值进行比较。对于显著性水平 α 下的双尾检验，拒绝域为 |统计量| > 临界值。对于单尾检验，拒绝域完全位于 H₁ 所指示的方向上。应尽量绘制清晰的示意图，标明接受域和拒绝域，因为Edexcel常对正确标注的图形给予分数。
9. p‑values and Conclusions | p值与结论
The p‑value is the probability of obtaining a test statistic at least as extreme as the one observed, assuming H₀ is true. If the p‑value is less than α, we reject H₀. Conclude in the context of the problem: state whether there is sufficient evidence to support the alternative hypothesis. Avoid generic phrases like “reject H₀” without interpretation; instead write, for example, “There is significant evidence at the 5% level that the mean score for Method A is higher than that for Method B.”
p值是在H₀成立的前提下，观察到至少与当前统计量同样极端的结果的概率。若p值小于α，则拒绝H₀。需结合问题背景给出结论：说明是否有充分证据支持备择假设。避免使用“拒绝H₀”这样缺乏解释的笼统表述；应写作例如：“在5%的显著性水平下，有显著证据表明方法A的平均得分高于方法B。”
10. Underlying Assumptions | 基本假设
All these tests rest on important assumptions. For the independent two‑sample t‑test (pooled variance), we assume:
- Each sample is randomly drawn from its population and the observations are independent.
- 每个样本从其总体中随机抽取，且观测值相互独立。
- Both populations are normally distributed, or sample sizes are sufficiently large for the Central Limit Theorem to apply.
- 两个总体均服从正态分布，或者样本量足够大，以便中心极限定理适用。
- The population variances are equal (for the pooled test).
- 总体方差相等（针对合并方差检验）。
所有这些检验都依赖于重要假设。对于独立双样本t检验（合并方差），我们假定：每个样本从其总体中随机抽取且观测值独立；两个总体正态或样本量足够大以应用中心极限定理；总体方差相等。Welch t检验则无需等方差假设。配对t检验要求差值近似服从正态分布。答题时，即使简要提及，也应当检查这些假设。
11. Worked Example (Independent t‑test) | 示例（独立t检验）
andard deviation 8; Group B (n₂ = 25) has mean 68 with standard deviation 7. Test at the 5% e level whether there is a difference in mean scores, assuming equal population variances.
两组学生分别采用不同的复习方案。A组（n₁ = 20）的测试平均分为72，标准差为8；B组（n₂ = 25）的平均分为68，标准差为7。假设总体方差相等，在5%显著性水平下检验平均分是否存在差异。
H₀: μ₁ = μ₂, H₁: μ₁ ≠ μ₂ (two‑ (1216 + 1176)/43 = 2392/43 ≈ 55.6279. sₚ ≈ 7.458. The t statistic: t = (72 − 68) / (7.458 √1/20 + 1/25) = 4 / (7.458 × 0.3) = 4 / 2.2374 ≈ 1.788. Degrees of freedom = 43. The two‑tailed critical value for t₄₃ at 5% is approximately 2.017. Since |1.788| do not reject H₀. There is insufficient evidence to conclude a difference in mean scores.
H₀: μ₁ = μ₂，H₁: μ₁ ≠ μ₂（双尾）。合并方差 sₚ² = [19×64 + 24×49] / 43 = (1216 + 1176)/43 = 2392/43 ≈ 55.6279，sₚ ≈ 7.458。t统计量 = (72 − 68) / (7.458 √1/20 + 1/25) = 4 / 2.2374 ≈ 1.788。自由度 = 43。5%双尾临界值 t₄₃ ≈ 2.017。由于 |1.788| < 2.017，不拒绝H₀。没有足够证据表明平均分数有差异。
12. Common Mistakes and Exam Tips | 常见错误与考试技巧
Many students confuse independent and paired designs: a paired test is only valid when observations are naturally linked. Forgetting to state hypotheses in symbols and words can cost marks. Miscalculating the pooled standard deviation or using the wrong degrees of freedom is frequent. Always verify whether the question implies equal variances; if the sample standard deviations differ substantially, consider Welch’s test. Finally, round the test statistic and critical values appropriately, but do not round excessively during intermediate steps.
许多学生混淆独立设计与配对设计：仅当观测值自然成对时，配对检验才有效。忘记用符号和文字陈述假设会失分。计算合并标准差出错或使用错误的自由度也屡见不鲜。务必判断题目是否暗示方差相等；若样本标准差差异较大，应考虑Welch检验。最后，适当对检验统计量和临界值进行舍入，但在中间步骤中避免过度舍入。
Published by TutorHao | Mathematics Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图