📚 Testing for the Mean of a Poisson Distribution | 检验泊松分布的均值
Hypothesis testing for the mean of a Poisson distribution is a core topic in Edexcel A‑Level Mathematics. Understanding how to test whether a Poisson mean has changed or differs from a claimed value requires careful formulation of null and alternative hypotheses, calculation of probabilities using the Poisson model, and interpretation of statistical significance. This article will walk you through every step, from setting up hypotheses to using critical regions, p‑values, and Poisson tables, with worked examples and exam tips.
对泊松分布均值的假设检验是 Edexcel A‑Level 数学的核心考点。要检验泊松均值是否发生变化或是否与某声称值不同，需要准确把握原假设与备择假设的设定、利用泊松模型计算概率，以及理解统计显著性的含义。本文将逐步解析从建立假设到使用临界域、p 值及泊松分布表的过程，并提供例题与应试技巧。
1. Introduction to Poisson Distribution and Its Mean | 泊松分布及其均值简介
constant average rate. It is defined by a single parameter λ, which represents both the mean and the variance of the distribution. If a random variable X follows a Poisson distribution, we write X ~ Po(λ). The probability mass function is given by P(X = x) = (λˣ e⁻λ) / x! for x = 0, 1, 2, ...
泊松分布用于对固定时间或空间区间内事件发生的次数建模，要求事件独立且以恒定的平均速率发生。该分布由单一参数 λ 定义，λ 既是分布的均值也是方差。若随机变量 X 服从泊松分布，记为 X ~ Po(λ)，其概率质量函数为 P(X = x) = (λˣ e⁻λ) / x!，其中 x = 0, 1, 2, …
The parameter λ can be interpreted as the expected number of occurrences per unit interval. For example, if a call centre receives an average of 4 calls per hour, the number of calls in a given hour can be modelled as Po(4). In hypothesis testing, we usually test whether the true mean λ equals a specified value λ₀.
参数 λ 可解释为单位区间内事件发生的期望次数。例如，若某呼叫中心平均每小时接到 4 通电话，则任一小时内接到的电话数可用 Po(4) 建模。在假设检验中，我们通常检验真实的均值 λ 是否等于某个指定值 λ₀。
2. The Concept of Hypothesis Testing | 假设检验的概念
Hypothesis testing is a statistical method used to make decisions about a population parameter based on sample data. It begins with two competing statements: the null hypothesis (H₀) and the alternative hypothesis (H₁). The null hypothesis is a statement of no effect or no change, which we assume to be true until evidence suggests otherwise. The alternative hypothesis represents what we are trying to find evidence for.
假设检验是一种基于样本数据对总体参数做出决策的统计方法。它从两个对立的陈述出发：原假设 (H₀) 和备择假设 (H₁)。原假设是“无效应”或“无变化”的陈述，在没有足够反面证据之前我们假定其为真。备择假设则代表我们希望寻找证据支持的结论。
After collecting data, we calculate a test statistic and determine how extreme the observed result would be if H₀ were true. If the probability of obtaining such an extreme result is very small (less than a pre‑set significance level α), we reject H₀ in favour of H₁. Otherwise, we do not have sufficient evidence to reject H₀. In the context of a Poisson mean, the test statistic is usually the observed count itself, and we use the Poisson distribution under H₀ to compute probabilities.
收集数据后，我们计算检验统计量并判断：若 H₀ 为真，得到当前观测结果或更极端结果的概率有多大。如果该概率非常小（小于预先设定的显著性水平 α），我们就拒绝 H₀ 而接受 H₁；否则，我们没有足够证据拒绝 H₀。在检验泊松均值时，检验统计量通常是观察到的计数本身，我们根据 H₀ 下的泊松分布来计算概率。
3. Setting Up Hypotheses for Poisson Mean | 设定泊松均值的假设
tion mean λ equals a specific value λ₀: H₀ : λ = λ₀. The alternative hypothesis can be one‑tailed (lower or upper) or two‑tailed, depending on the question being asked. For a lower‑tail test, H₁ : λ < λ₀ (we suspect a decrease). For an upper‑tail test, H₁ : λ > λ₀ (we suspect an increase). For a two‑tailed test, H₁ : λ ≠ λ₀ (we are testing for any difference).
检验泊松均值时，原假设总是总体均值 λ 等于某指定值 λ₀：H₀ : λ = λ₀。备择假设可以是单侧的（左侧或右侧）或双侧的，具体取决于研究问题。左侧检验时，H₁ : λ < λ₀（怀疑均值减小）；右侧检验时，H₁ : λ > λ₀（怀疑均值增大）；双侧检验时，H₁ : λ ≠ λ₀（检验是否存在任何差异）。
Choosing the correct alternative hypothesis is crucial. The wording of the problem often gives clear direction: phrases like 'has decreased', 'lower than', or 'less than' indicate a lower‑tail test; 'increased', 'higher than', or 'more than' indicate an upper‑tail test; and 'changed', 'different from', or 'not equal to' suggest a two‑tailed test. Always define the parameter λ clearly in context before writing the hypotheses.
选择正确的备择假设非常重要。题目中的措辞通常会给出明确方向：“减少了”、“低于”、“小于” 表明左侧检验；“增加了”、“高于”、“大于” 表明右侧检验；“发生了变化”、“不同于”、“不等于” 则意味着双侧检验。在写下假设之前，务必在上下文中明确定义参数 λ。
4. Test Statistic and Its Distribution | 检验统计量及其分布
s the observed count X from the sample. Under the null hypothesis H₀ : λ = λ₀, we assume X ~ ean is small; we work directly with the Poisson probability function or cumulative tables.
在对泊松均值的检验中，检验统计量为样本观测计数 X。在原假设 H₀ : λ = λ₀ 下，我们假定 X ~ Po(λ₀)，并基于该分布进行所有概率计算。在均值较小时不需要标准化，可以直接使用泊松概率函数或累积分布表。
The test statistic is simply the number of events recorded in the given interval. For example, if a factory averages 2 accidents per month and we observe 0 accidents in a particular month, the test statistic is 0. We then ask: how likely is it to observe 0 or fewer accidents if the true mean were 2? This probability guides our conclusion.
检验统计量就是在给定区间内记录到的事件数。例如，某工厂平均每月发生 2 起事故，而某月观察到 0 起事故，则检验统计量为 0。我们要问：如果真实均值为 2，观察到 0 起或更少事故的概率有多大？这一概率将指导我们得出结论。
5. Critical Regions and Significance Levels | 临界区域与显著性水平
The significance level α (typically 0.05 or 5%) is the maximum probability of making a Type I error — For a Poisson test, the critical region is a set of counts that have a total probability ≤ α under H₀, chosen to be as close as possible to α without exceeding it.
显著性水平 α（通常为 0.05 或 5%）是犯第一类错误——当 H₀ 为真时拒绝 H₀——的最大允许概率。临界区域（或拒绝域）由所有会导致拒绝 H₀ 的检验统计量取值组成。对于泊松检验，临界区域是一组计数，其在 H₀ 下的总概率 ≤ α，且尽可能接近 α 但不超过。
For a lower‑tail test, the critical region is X ≤ c, where c is the largest integer such that P(X ≤ c | λ = λ₀) ≤ α. For an upper‑tail test, the critical region is X ≥ c, where c is the smallest integer such that P(X ≥ c | λ = λ₀) ≤ α. In a two‑ is not required. Critical values can be found using cumulative Poisson tables or software.
对左侧检验，临界区域为 X ≤ c，其中 c 是满足 P(X ≤ c | λ = λ₀) ≤ α 的最大整数。对右侧检验，临界区域为 X ≥ c，其中 c 是满足 P(X ≥ c | λ = λ₀) ≤ α 的最小整数。在双侧检验中，显著性水平通常被平均分配到两个尾部，尽管泊松分布并非对称。临界值可利用累积泊松分布表或软件获得。
6. One‑Tailed vs Two‑Tailed Tests | 单侧检验与双侧检验
The choice between a one‑tailed and a two‑tailed test affects the critical region and the interpretation of p‑values. A one‑tailed test is used when we have a specific direction of interest (increase or decrease). A two‑tailed test is used when we are only interested in whether the parameter has changed, regardless of direction.
单侧检验与双侧检验的选择会影响临界区域和 p 值的解读。当我们有明确的方向性关注（增大或减小）时使用单侧检验；当我们只关心参数是否发生变化，不问方向时，则使用双侧检验。
In Edexcel exam questions, you will often be told explicitly whether to conduct a one‑tailed or two‑tailed test. However, you should also be able to deduce the correct form from the wording of the problem. Remember that a two‑tailed test at significance level α can be conducted by halving α for each tail in the Poisson context, but because the distribution is discrete, the exact total probability may be slightly less than α.
在 Edexcel 考试题中，通常会明确指出进行单侧还是双侧检验，但考生也应能从题意推断正确的检验形式。注意，在泊松情况下进行显著性水平为 α 的双侧检验时，可以将 α 平分到两侧，但由于分布的离散性，实际总概率可能略低于 α。
7. Calculating p‑Values | 计算 p 值
The p‑ at least as extreme as the one observed, in the direction(s) specified by H₁. For a lower‑tail test, p‑value = P(X ≤ observed value | λ = λ₀). For an upper‑tail test, p‑value = P(X ≥ observed value | λ = λ₀). For a two‑ ty or sum probabilities in both tails that are ≤ the observed tail probability, though the 'double the smaller tail' approach is widely taught at A‑Level. If the p‑value ≤ α, we reject H₀.
p 值是在 H₀ 为真的前提下，得到检验统计量至少与观测值同样极端（在 H₁ 指定的方向上）的概率。对左侧检验，p 值 = P(X ≤ 观测值 | λ = λ₀)。对右侧检验，p 值 = P(X ≥ 观测值 | λ = λ₀)。对双侧检验，常见做法是将较小一侧的尾部概率加倍，或累加两个尾部中概率不超过观测侧概率的部分。在 A‑Level 中广泛采用“较小尾部概率加倍”的方法。如果 p 值 ≤ α，则拒绝 H₀。
Using p‑values instead of critical regions can be more informative, as it gives a measure of the strength of evidence against H₀. A‑level questions may ask you to state the p‑value and draw a conclusion. It is essential to show the probability calculation clearly, stating the distribution used under H₀.
使用 p 值而非临界区域能提供更多信息，因为它衡量了反对 H₀ 的证据强度。A‑Level 考题可能要求陈述 p 值并得出结论。关键是要清晰地展示概率计算过程，明确指出在 H₀ 下所使用的分布。
8. Worked Example: One‑Tailed Test | 例题：单侧检验
A company claims that the number of defects per metre of fabric follows a Poisson distribution with mean 2.5. A quality inspector suspects the defect rate has decreased and counts defects in a randomly selected 10‑metre length, finding a total of 15 defects. Test, at the 5% significance level, whether there is evidence that the mean number of defects per metre has decreased.
某公司声称每米织物上的疵点数服从均值为 2.5 的泊松分布。质检员怀疑疵点率降低了，于是随机选取 10 米长的织物进行检验，发现总共有 15 处疵点。在 5% 的显著性水平下，检验是否有证据表明每米疵点的均值降低了。
.5 × 10 = 25. We set H₀ : λ = 2.5 (per metre, i.e. λ₁₀ = 25 for 10 m), H₁ : λ < 2.5 (lower‑tail). The observed test statistic is X = 15. We calculate p‑value = P(X ≤ 15 | Po(25) ≈ 0.0223. Since 0.0223 < 0.05, we reject H₀. There is sufficient evidence at the 5% level to conclude that the mean number of defects per metre has decreased.
首先注意，检验对象是每米的均值。在 10 米长度下，若 H₀ 为真，总疵点数应服从 Po(25)，因为 10 米的 λ = 2.5 × 10 = 25。设立 H₀ : λ = 2.5（每米，对应 10 米 λ₁₀ = 25），H₁ : λ < 2.5（左侧检验）。观测到的检验统计量为 X = 15。计算 p 值 = P(X ≤ 15 | λ = 25)。查泊松分布表或使用工具，Po(25) 下 P(X ≤ 15) ≈ 0.0223。由于 0.0223 < 0.05，拒绝 H₀。在 5% 显著性水平下，有充分证据表明每米疵点均值降低了。
9. Worked Example: Two‑Tailed Test | 例题：双侧检验
A biologist records the number of rare birds spotted per hour at a nature reserve. It is believed that sightings follow a Poisson distribution with mean 3.2 per hour. After recent habitat changes, the biologist wants to know if the rate has changed. She observes for 5 hours and records 10 sightings. Test at the 1% significance level whether there is evidence that the mean per hour has changed.
一位生物学家记录自然保护区每小时观测到的珍稀鸟类数量。据信，观测数服从均值为 3.2 的泊松分布。近期栖息地发生变化，她想知道这个比率是否改变。她观察了 5 个小时，记录到 10 次观测。在 1% 显著性水平下，检验是否有证据表明每小时均值发生了变化。
Here we have a two‑ailed test. Under H₀, the hourly mean λ = 3.2, so the expected count over 5 hours is 16: X ~ H₀ : λ = 3.2 vs H₁ : λ ≠ 3.2. Since the observed value is 10, which is below 16, the lower‑tail probability P(X ≤ 10 | λ=16) is the primary tail. Using Po(16) tables, P(X ≤ 10) ≈ 0.0774. Because this is a two‑tailed test, we double the smaller tail: p‑value = 2 × 0.0774 = 0.1548. At the 1% level (α = 0.01), p‑value > 0.01, so we do not reject H₀. There is insufficient evidence at the 1% level to say the mean has changed.
此处为双侧检验。在 H₀ 下，每小时均值 λ = 3.2，因此 5 小时的期望计数为 16：H₀ 下 X ~ Po(16)。实际观测 X = 10。设立 H₀ : λ = 3.2 vs H₁ : λ ≠ 3.2。观测值 10 低于 16，因此左侧尾部概率 P(X ≤ 10 | λ=16) 为主要尾部。查 Po(16) 表，P(X ≤ 10) ≈ 0.0774。因为是双侧检验，将较小尾部加倍：p 值 = 2 × 0.0774 = 0.1548。在 1% 水平下（α = 0.01），p 值 > 0.01，故不拒绝 H₀。没有足够证据在 1% 水平下认为均值发生了变化。
10. Using Poisson Probability Tables | 使用泊松概率表
he complement: P(X ≥ x) = 1 − P(X ≤ x−1). Always be careful with the discrete nature: P(X ≤ 5) is directly read; for P(X < 5), this equals P(X ≤ 4). These simple adjustments are crucial for correct calculations in exams.
Edexcel 在公式册中提供累积泊松分布表，表中给出不同 λ 值下 P(X ≤ x) 的值。要计算 P(X ≥ x)，使用补集：P(X ≥ x) = 1 − P(X ≤ x−1)。务必注意离散特性：P(X ≤ 5) 可直接查表；而 P(X < 5) 等于 P(X ≤ 4)。这些简单调整对考试中正确计算至关重要。
a critical value for a given α, look down the column corresponding to λ = λ₀. For a lower‑tail critical value c, find the largest x such that P(X ≤ x) ≤ α. For an upper‑ α, i.e. P(X ≤ x−1) ≥ 1−α. Practice with tables will speed up your work and reduce errors.
用表格查找给定 α 的临界值时，需在 λ = λ₀ 对应的列中查找。对于左侧临界值 c，找出最大的 x 使 P(X ≤ x) ≤ α。对于右侧临界值，找出最小的 x 使 P(X ≥ x) ≤ α，即 1 − P(X ≤ x−1) ≤ α，也即 P(X ≤ x−1) ≥ 1−α。多加练习使用表格能提升速度并减少错误。
11. Assumptions and Conditions | 假设条件与注意事项
For a hypothesis test on a Poisson mean to be valid, the data must meet the assumptions of the Poisson distribution: events must occur independently, at a constant average rate, and the probability of more than one event in an infinitesimally small interval is negligible. In practice, independence and constant rate are the main concerns. For example, if you are counting arrivals at a shop, external factors like a sale event could violate the constant rate assumption.
要使泊松均值的假设检验有效，数据必须满足泊松分布的假设：事件须独立发生，具有恒定的平均速率，且在无穷小的时间区间内发生多于一个事件的概率可忽略不计。实践中，独立性和恒定速率是主要关注点。例如，统计店铺顾客到达数时，促销活动等外部因素可能破坏恒定速率假设。
Furthermore, the test relies on the quality of the sample. The observation interval must be clearly defined and fixed before data collection. Avoid 'cherry‑picking' an interval that gives a desired result. Also, when combining intervals (like scaling up to a larger period), ensure the rate remains constant across the combined interval. Always state these assumptions in your exam answers if asked to comment on the validity of a test.
此外，检验依赖于样本质量。观测区间必须在数据收集前就明确定义并固定下来。避免“挑选”能带来期望结果的区间。当合并区间（如扩大到更长时间段）时，要确保速率在整个合并区间内保持恒定。若考题要求评价检验的有效性，务必在答案中陈述这些假设。
12. Connection to Normal Approximation (Large Mean) | 与正态近似的联系（均值较大时）
on distribution can be approximated by a normal distribution with mean λ and variance λ: X ~ N(λ, λ) approximately. This allows the use of a z‑ ollows a standard normal distribution. A continuity correction is often applied to improve accuracy because we are approximating a discrete distribution with a continuous one.
当泊松均值 λ 较大时（通常 λ > 15 或 20），泊松分布可用均值为 λ、方差为 λ 的正态分布来近似：X 近似服从 N(λ, λ)。这使得我们可以使用 z 检验。检验统计量为 Z = (X − λ) / √λ，在 H₀ 下近似服从标准正态分布。由于是用连续分布近似离散分布，通常会使用连续性校正来提高精度。
This approximation is particularly useful when the value of λ₀ is not in the Poisson table provided or when calculations by hand are tedious. However, Edexcel questions often explicitly ask for a test based on the exact Poisson distribution unless stated otherwise. If you use the normal approximation, remember to apply correction and state that it is an approximation. In many real‑world contexts, the Poisson exact test is preferred for smaller means, while the normal approximation is acceptable for larger ones.
当 λ₀ 值不在所提供的泊松分布表中，或手工计算过于繁琐时，该近似尤其有用。不过，除非明确说明，Edexcel 考题通常要求基于精确泊松分布进行检验。若使用正态近似，记得进行校正并说明这是近似方法。在许多实际场景中，对较小均值宜采用精确泊松检验，而较大均值时正态近似是可接受的。
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