The Poisson Distribution: A Complete Guide for AQA A-Level Further Maths — 泊松分布完全指南：AQA 进阶数学统计篇

📚 The Poisson Distribution: A Complete Guide for AQA A-Level Further Maths | 泊松分布完全指南：AQA 进阶数学统计篇

一、泊松分布是什么：稀有事件计数的概率模型 | What Is the Poisson Distribution: A Probability Model for Counting Rare Events

在 AQA A-Level 进阶数学的 Paper 3 统计部分，泊松分布（Poisson distribution）是最常考的概率模型之一。它描述的是：在一段固定的时间、面积或体积内，某个”稀有事件”恰好发生指定次数的概率。所谓稀有事件，指的是单独一次发生概率很小、但总体发生次数可观的事件，例如电话客服中心每分钟接到的来电数、某路口一周内发生的事故数、放射性物质在单位时间内衰变的粒子数，或者一本书每一页上出现的印刷错误数。

The Poisson distribution is one of the most frequently examined probability models in the Statistics paper (Paper 3) of AQA A-Level Further Mathematics. It describes the probability that a given number of “rare events” occur within a fixed interval of time, area or volume. A rare event has a small probability of occurring on any single trial, yet a noticeable total number of occurrences overall: examples include the number of phone calls a call centre receives per minute, the number of accidents at a junction per week, the number of particles emitted by a radioactive source per unit time, and the number of printing mistakes on a page of a book.

为什么需要专门引入一个”新”的分布？因为当我们用二项分布 B(n, p) 去模拟这类问题时，会遇到一个尴尬的处境：事件发生的总次数 n 非常大（例如一分钟内理论上可能来电的次数），而每次发生的概率 p 又非常小（每一个瞬间接到来电的概率极低）。n 很大、p 很小，二项分布的阶乘计算会变得极其繁琐，甚至超出计算器的精度范围。泊松分布正是数学家们为了处理”n 大、p 小”这类极限情形而推导出来的模型，它只需要一个参数 λ，就能把整条概率分布刻画出来。

Why do we need a separate distribution at all? When we try to model such problems with the binomial distribution B(n, p), we run into a difficulty: the total number of trials n is huge (for example, the number of instants at which a call could theoretically arrive in one minute), while the success probability p on each trial is tiny. With large n and small p, the factorial calculations in the binomial formula become enormously tedious and can even exceed the precision of a calculator. The Poisson distribution was derived precisely to handle this “large n, small p” limiting case: it requires only one parameter, λ (lambda), which summarises the entire probability distribution.

在进阶数学的考试中，泊松分布经常与假设检验、正态近似、二项近似等知识点结合出题，一道大题往往涵盖多个小问。因此，彻底理解泊松分布的定义、条件、性质以及计算技巧，是拿下 AQA Paper 3 高分的关键一步。本文将从定义出发，逐步讲解它的适用条件、公式与性质、表格与计算器使用、三种分布之间的近似关系，以及考试中最常见的题型与陷阱。

In the Further Mathematics examination, the Poisson distribution is frequently combined with hypothesis testing, normal approximation and binomial approximation in multi-part questions. A thorough command of its definition, conditions, properties and calculation techniques is therefore essential for scoring well on AQA Paper 3. This article starts from the definition and works step by step through the conditions of use, the formula and its properties, tables and calculator skills, the approximation links between three distributions, and the most common question types and traps in the examination.

二、泊松分布的适用条件：独立性、恒定发生率与不重叠 | The Four Conditions: Independence, Constant Mean Rate and No Overlap

泊松分布并不是”看起来像计数问题就能用”的万能工具。考试中经常会出现一道判断型小问，给出一个现实情境，要求考生判断泊松分布是否适用，并说明理由。要答好这类问题，必须牢记泊松模型的四条核心假设。

The Poisson distribution is not a universal tool that can be applied whenever a counting situation appears. Examinations frequently include a short judgement question that presents a real-world context and asks candidates to decide whether the Poisson distribution is appropriate and to justify their answer. To answer such questions well, you must remember the four core assumptions of the Poisson model.

第一条：事件相互独立。一个事件的发生不会影响另一个事件发生的概率。例如，”某一秒内接到来电”与”下一秒内接到来电”应当互不影响。如果来电之间存在连锁效应（比如一个人打电话占线导致另一个人稍后重拨），独立性就被破坏了，泊松模型不再适用。第二条：平均发生率 λ 在考察的时间段内保持恒定。如果 λ 随时间变化 – 例如客服中心在工作高峰时段来电率明显高于深夜 – 那么整体数据就不服从单一的泊松分布。

First, the events must be independent: the occurrence of one event does not affect the probability of another event. For example, “a call arrives in this second” and “a call arrives in the next second” should not influence each other. If there is a chain effect between calls (for example, one caller finding the line busy and redialling later), independence breaks down and the Poisson model no longer applies. Second, the average rate λ must remain constant over the period being studied. If λ changes with time, such as a call centre receiving calls far more frequently during peak hours than late at night, the data as a whole does not follow a single Poisson distribution.

第三条：两个事件不可能在同一瞬间同时发生。泊松分布假设事件是”逐点”发生的，同一时刻至多发生一个事件。如果情境中允许两件或更多事件同时出现（例如同一辆车同时载着多名乘客抵达），就需要谨慎。第四条（隐含条件）：事件相对”稀有”。虽然教材通常只强调前三条，但严格来说，泊松分布是二项分布在”n 很大、p 很小”下的极限，因此事件本身的单次发生概率应当很小。

Third, two events cannot occur at exactly the same instant. The Poisson distribution assumes that events occur “one point at a time”, with at most one event at any given moment. If the context allows two or more events to occur simultaneously (for example, one bus arriving carrying many passengers at once), caution is needed. Fourth (an implicit condition): the events should be relatively rare. Although textbooks usually emphasise only the first three conditions, strictly speaking the Poisson distribution is the limit of the binomial distribution as n becomes very large and p very small, so the probability of a single event occurring should be small.

答题模板值得记下来：判断类小问的标准写法是”该情境（不）适合用泊松分布建模，因为事件（不）独立、平均发生率（不）恒定、事件（不）会同时发生”，然后结合题目给出的具体情境各补一句话。只要把假设和情境一一对应，这类 2 分小问就能稳稳拿到。

A standard answer template is worth memorising: the model answer for a judgement question is “The situation (is / is not) suitable for modelling with a Poisson distribution, because the events (are / are not) independent, the mean rate (is / is not) constant, and events (can / cannot) occur simultaneously”, followed by one sentence linking each assumption to the given context. As long as you match each assumption to the context, this type of two-mark question is guaranteed.

三、泊松分布的公式与符号：X ~ Po(λ) | The Formula and Notation: X ~ Po(λ)

如果一个随机变量 X 表示”固定区间内稀有事件发生的次数”，且满足上文的四条假设，那么 X 服从参数为 λ 的泊松分布，记作 X ~ Po(λ)。这里的 λ 读作 lambda，表示该区间内事件发生的平均次数，例如”平均每小时接到 8 通电话”写作 λ = 8，单位是”每区间”而不是”每单位时间”。

If a random variable X counts the number of rare events in a fixed interval, and the four assumptions above are satisfied, then X follows a Poisson distribution with parameter λ, written X ~ Po(λ). The Greek letter λ (lambda) denotes the mean number of events in that interval, for example “an average of 8 calls per hour” is written λ = 8. Note that λ is measured “per interval”, not “per unit time”.

泊松分布的概率质量函数（PMF）是：P(X = x) = e^−λ · λ^x / x!，其中 x = 0, 1, 2, 3, …。字母 e 是自然常数，约等于 2.71828，x! 表示 x 的阶乘，即 x! = x × (x − 1) × … × 2 × 1，并规定 0! = 1。这个公式看似复杂，但实际上只需要三步：先算 e^−λ，再算 λ^x，最后除以 x!。例如 λ = 2 时，P(X = 0) = e⁻² ≈ 0.1353，P(X = 1) = 2e⁻² ≈ 0.2707，P(X = 2) = 2²e⁻²/2 ≈ 0.2707。

The probability mass function (PMF) of the Poisson distribution is P(X = x) = e^−λ · λ^x / x!, where x = 0, 1, 2, 3, …. The letter e is the natural constant, approximately 2.71828, and x! denotes the factorial of x, defined as x! = x × (x − 1) × … × 2 × 1, with the convention that 0! = 1. The formula looks complicated but actually involves only three steps: compute e^−λ, compute λ^x, then divide by x!. For example, with λ = 2, P(X = 0) = e⁻² ≈ 0.1353, P(X = 1) = 2e⁻² ≈ 0.2707, and P(X = 2) = 2²e⁻²/2 ≈ 0.2707.

关于计算器：AQA 进阶数学允许使用的科学计算器大多内置了泊松概率函数（通常标记为 PoissonPD 或 poissonpdf），可以一步算出 P(X = x)。但考试要求考生能够手算小参数情形（如 λ = 0.5、λ = 1 这类数值），并且能正确读懂题目给出的泊松累积概率表。特别提醒：不要把泊松公式中的 e^−λ 和指数分布混淆，泊松分布的自变量是”次数 x”，指数分布的自变量才是”时间 t”。

About calculators: most scientific calculators permitted in AQA Further Mathematics have a built-in Poisson probability function (usually labelled PoissonPD or poissonpdf) that computes P(X = x) in one step. However, the examination expects candidates to be able to calculate small-parameter cases by hand (such as λ = 0.5 or λ = 1) and to read Poisson cumulative probability tables correctly. A word of caution: do not confuse e^−λ in the Poisson formula with the exponential distribution. The argument of the Poisson distribution is the count x, whereas the argument of the exponential distribution is the time t.

符号方面还需要区分两个容易混淆的记号：X ~ Po(λ) 表示 X 服从泊松分布，而 P(X = x) 表示”X 恰好等于 x 的概率”。考试答案中必须写清楚”设 X 为……”的定义句，例如”Let X be the number of calls received in one hour, so X ~ Po(8)”。定义随机变量这一步在评分标准中通常单独占分，漏写会被扣过程分。

On notation, two easily confused symbols must be distinguished: X ~ Po(λ) states that X follows a Poisson distribution, while P(X = x) denotes the probability that X takes exactly the value x. In examination answers you must write a clear definition sentence such as “Let X be the number of calls received in one hour, so X ~ Po(8)”. Defining the random variable is usually worth a separate method mark in the mark scheme, and omitting it loses process marks.

四、均值等于方差：泊松分布最独特的性质 | Mean Equals Variance: The Signature Property of the Poisson

泊松分布最著名、也最常被用来出题的性质是：它的期望（均值）和方差相等，都等于参数 λ。用公式表示就是 E(X) = λ，Var(X) = λ。这一点与二项分布形成鲜明对比：二项分布 B(n, p) 的均值是 np，方差是 np(1 − p)，方差总是小于均值（因为 0 < 1 − p < 1）。

The most famous property of the Poisson distribution, and the one most often used to construct examination questions, is that its expectation (mean) and variance are equal, both being the parameter λ. In symbols, E(X) = λ and Var(X) = λ. This contrasts sharply with the binomial distribution: for B(n, p) the mean is np and the variance is np(1 − p), so the variance is always smaller than the mean because 0 < 1 − p < 1.

这个性质最直接的用途是”反推参数”：当题目只给出样本数据而不直接给出 λ 时，可以用样本均值来估计 λ。例如，一家书店统计了 100 个星期中每天售出的某畅销书数量，算出平均每天卖出 3.2 本，那么就可以设 X ~ Po(3.2)。由于方差等于均值，还可以进一步用样本方差来检验数据是否真的服从泊松分布：如果样本方差明显大于或小于样本均值，说明数据很可能不满足泊松假设。

The most direct use of this property is to recover the parameter: when a question provides sample data but not λ itself, you can estimate λ with the sample mean. For example, a bookshop records the number of copies of a bestseller sold per day over 100 weeks and finds an average of 3.2 copies per day; you may then set X ~ Po(3.2). Because the variance equals the mean, you can also use the sample variance to check whether data really follow a Poisson distribution: if the sample variance is clearly larger or smaller than the sample mean, the data probably do not satisfy the Poisson assumptions.

考试中还有一种经典考法：给出 E(X) 和 Var(X) 的数值（例如 E(X) = 3，Var(X) = 3），要求判断 X 是否可能服从泊松分布。答案就是”是，因为泊松分布的均值等于方差”。反之，如果题目给出 E(X) = 3、Var(X) = 5，那么 X 不可能服从泊松分布，原因同样是均值不等于方差。这类 1 分判断小问送分题，关键是答出”mean = variance”这个核心理由，并指出具体数值相等或不相等。

There is also a classic examination format: give the values of E(X) and Var(X) (for example E(X) = 3 and Var(X) = 3) and ask whether X could follow a Poisson distribution. The answer is “yes, because the mean of a Poisson distribution equals its variance”. Conversely, if the question gives E(X) = 3 and Var(X) = 5, then X cannot follow a Poisson distribution, for exactly the same reason. For these one-mark judgement gifts, the key is to state the core reason “mean = variance” and to point out whether the given numbers are equal or not.

均值等于方差还隐含了另一个考点：λ 必须是正数，且通常不是整数。λ 是”平均次数”，可以是 2.5、0.7 这样的非整数，而 X 的取值永远是整数 0, 1, 2, …。很多同学会误把 λ 当成 X 的可能取值，这是概念性错误：λ 是分布的参数，X 才是随机变量。在画概率分布图时，横轴是整数 x，纵轴是对应的概率，图形呈右偏（正偏）形态，且随着 λ 增大越来越接近对称。

Mean equals variance also implies another subtle point: λ must be positive and is usually not an integer. λ is an “average count” and can be a non-integer such as 2.5 or 0.7, whereas X always takes integer values 0, 1, 2, …. Many students mistakenly treat λ as a possible value of X; this is a conceptual error: λ is a parameter of the distribution, while X is the random variable. When sketching the probability distribution, the horizontal axis shows the integer values of x and the vertical axis shows the corresponding probabilities. The graph is right-skewed (positively skewed), becoming more symmetric as λ grows.

五、累积概率计算：P(X ≤ k) 与统计表的使用 | Cumulative Probabilities: P(X ≤ k) and Statistical Tables

考试中真正高频的是累积概率问题：求 P(X ≤ k)、P(X ≥ k) 或 P(a ≤ X ≤ b)。这些都可以从 P(X ≤ k) 出发换算。核心换算公式有三条：P(X ≤ k) 直接查表或按计算器；P(X > k) = 1 − P(X ≤ k)；P(X ≥ k) = 1 − P(X ≤ k − 1)。最后一条特别容易出错，因为”大于等于 k”的补事件是”小于等于 k − 1″，而不是”小于等于 k”。

The genuinely high-frequency questions in examinations concern cumulative probabilities: finding P(X ≤ k), P(X ≥ k) or P(a ≤ X ≤ b). All of these can be converted from P(X ≤ k). There are three essential conversion formulas: P(X ≤ k) is read directly from tables or the calculator; P(X > k) = 1 − P(X ≤ k); and P(X ≥ k) = 1 − P(X ≤ k − 1). The last one is particularly error-prone, because the complement of “at least k” is “at most k − 1”, not “at most k”.

例如，设 X ~ Po(2.5)，求 P(X ≥ 3)。正确做法：P(X ≥ 3) = 1 − P(X ≤ 2) = 1 − (P(X=0) + P(X=1) + P(X=2))。查表或计算得 P(X ≤ 2) ≈ 0.5438，所以 P(X ≥ 3) ≈ 0.4562。如果误写成 P(X ≥ 3) = 1 − P(X ≤ 3)，就会得到 1 − 0.7576 = 0.2424，答案相差甚远。另一个常见换算：P(1 ≤ X ≤ 4) = P(X ≤ 4) − P(X ≤ 0)。

For example, let X ~ Po(2.5) and find P(X ≥ 3). The correct approach is P(X ≥ 3) = 1 − P(X ≤ 2) = 1 − (P(X = 0) + P(X = 1) + P(X = 2)). From tables or a calculator, P(X ≤ 2) ≈ 0.5438, so P(X ≥ 3) ≈ 0.4562. If you mistakenly write P(X ≥ 3) = 1 − P(X ≤ 3), you obtain 1 − 0.7576 = 0.2424, a very different answer. Another common conversion is P(1 ≤ X ≤ 4) = P(X ≤ 4) − P(X ≤ 0).

使用统计表时要注意表格的格式：AQA 提供的泊松累积分布表通常给出 P(X ≤ x) 的值，行是 λ 的取值，列是 x 的取值。查表前先确认 λ 精确对应表格中的行；如果 λ 不在表中（例如 λ = 2.47），需要使用计算器而不是强行取近似值。另外，表头一定要看清楚是 P(X ≤ x) 还是 P(X = x)，很多同学因为看错表头导致整道大题全部算错。

When using statistical tables, pay attention to the format: the Poisson cumulative distribution table provided by AQA usually gives values of P(X ≤ x), with rows for λ and columns for x. Before reading the table, confirm that λ matches a row exactly; if λ is not in the table (for example λ = 2.47), use a calculator rather than forcing an approximation. Also, check carefully whether the table header says P(X ≤ x) or P(X = x): many students misread the header and consequently get the whole multi-part question wrong.

最后提醒一个易错点：”至少一个”问题的速算公式。P(X ≥ 1) = 1 − P(X = 0) = 1 − e^−λ。这个公式在 λ 较小时非常实用，例如 λ = 0.05 时，P(X ≥ 1) = 1 − e^−0.05 ≈ 0.0488。类似的还有 P(X = 0) = e^−λ 这一”零事件概率”，它在推导泊松过程、可靠性问题（如”某设备在一年内不出故障的概率”）中频繁出现，务必熟练掌握。

One final reminder about an easy-to-miss point: the quick formula for “at least one” questions. P(X ≥ 1) = 1 − P(X = 0) = 1 − e^−λ. This formula is very practical for small λ; for example, when λ = 0.05, P(X ≥ 1) = 1 − e^−0.05 ≈ 0.0488. Similarly, P(X = 0) = e^−λ, the “probability of no events”, appears frequently in derivations of Poisson processes and in reliability problems (such as “the probability that a device does not fail within one year”). Make sure you can use it fluently.

六、二项分布逼近泊松分布：n 大 p 小时的极限 | Approximating the Binomial by the Poisson: The Large n, Small p Limit

泊松分布与二项分布之间有一条重要的桥梁：当 n 很大、p 很小时，二项分布 B(n, p) 可以用泊松分布 Po(np) 来近似。直觉上，二项分布描述”n 次独立重复试验中成功的次数”，如果每次成功的概率 p 非常小，那么成功事件本身就成了”稀有事件”，恰好落入泊松模型的适用范围。

There is an important bridge between the Poisson and binomial distributions: when n is large and p is small, the binomial distribution B(n, p) can be approximated by the Poisson distribution Po(np). Intuitively, the binomial distribution describes the number of successes in n independent repeated trials; if the success probability p on each trial is very small, then success itself becomes a “rare event”, which falls exactly within the scope of the Poisson model.

教材给出的经验规则是：当 n ≥ 50、p ≤ 0.1，且 np ≤ 5（有些教材放宽到 np ≤ 10）时，近似效果足够好。实际操作时，用 λ = np 代入泊松公式即可。例如，某产品的次品率为 2%，随机抽取 100 件产品，问恰好有 3 件次品的概率。精确计算要用 B(100, 0.02)，而近似计算用 X ~ Po(2)，P(X = 3) = e⁻² × 8 / 6 ≈ 0.1804，与精确值 0.1823 非常接近，误差不到 1%。

The rule of thumb given in textbooks is that the approximation is good when n ≥ 50, p ≤ 0.1 and np ≤ 5 (some textbooks relax this to np ≤ 10). In practice, simply substitute λ = np into the Poisson formula. For example, if a product has a defect rate of 2% and 100 items are sampled, find the probability that exactly 3 are defective. The exact calculation uses B(100, 0.02), while the approximation uses X ~ Po(2): P(X = 3) = e⁻² × 8 / 6 ≈ 0.1804, very close to the exact value of 0.1823, with an error of less than 1%.

考试中这类题通常会直接给出指令：”Use a Poisson approximation to find the probability that…”。看到”Poisson approximation”字样，第一步就是把 λ 算出来（λ = np），并写明”Since n is large and p is small, X ~ B(n, p) is approximately Po(np)”。这个说明句在评分标准中通常占一个方法分，千万不要省略。

In examinations, such questions usually give a direct instruction: “Use a Poisson approximation to find the probability that…”. When you see the words “Poisson approximation”, the first step is to compute λ = np and to write “Since n is large and p is small, X ~ B(n, p) is approximately Po(np)”. This explanatory sentence usually carries a method mark in the mark scheme, so never omit it.

反方向的近似同样存在：当 λ 很小（比如 λ ≤ 5）时，泊松分布 P(X ≤ k) 的值也可以反过来用于近似某些复杂二项概率。不过 AQA 考纲中更常考的是”二项 → 泊松”这一方向，同学们只需牢牢掌握正向近似即可。常见配合考点：先判断是否满足近似条件，再完成计算，最后用”the approximation is appropriate because…”补一句理由。

The reverse approximation also exists: when λ is small (say λ ≤ 5), Poisson cumulative values can be used to approximate certain complicated binomial probabilities. However, the AQA specification more commonly examines the “binomial to Poisson” direction, so students only need to master the forward approximation firmly. A typical combined format is: first judge whether the approximation conditions are satisfied, then perform the calculation, and finally add a sentence of justification such as “the approximation is appropriate because…”.

七、泊松分布逼近正态分布：λ 大时的连续性修正 | Approximating the Poisson by the Normal: Continuity Correction for Large λ

当 λ 足够大时（教材惯例是 λ ≥ 10，AQA 考纲通常以 λ ≥ 15 作为安全线），泊松分布的形状会越来越接近钟形，因此可以用正态分布 N(λ, λ) 来近似。这里的逻辑是：泊松分布作为独立稀有事件计数之和，由中心极限定理可知，当 λ 增大时它趋于正态分布，且均值和方差都等于 λ。

When λ is sufficiently large (the textbook convention is λ ≥ 10, and the AQA specification usually takes λ ≥ 15 as a safe line), the shape of the Poisson distribution becomes increasingly bell-shaped, so it can be approximated by a normal distribution N(λ, λ). The logic is that the Poisson distribution, as a sum of counts of independent rare events, tends towards a normal distribution as λ grows by the central limit theorem, with both its mean and variance equal to λ.

用正态分布近似离散分布时，必须使用连续性修正（continuity correction）。核心规则是：把离散的整数边界”平移半格”。具体来说：P(X ≤ k) 近似为 P(Y ≤ k + 0.5)；P(X < k) 近似为 P(Y ≤ k − 0.5)；P(X ≥ k) 近似为 P(Y ≥ k − 0.5)；P(X > k) 近似为 P(Y ≥ k + 0.5)。其中 Y ~ N(λ, λ)。

When approximating a discrete distribution with a normal one, the continuity correction is essential. The core rule is to shift the discrete integer boundary by half a unit. Specifically: P(X ≤ k) is approximated by P(Y ≤ k + 0.5); P(X < k) by P(Y ≤ k − 0.5); P(X ≥ k) by P(Y ≥ k − 0.5); and P(X > k) by P(Y ≥ k + 0.5), where Y ~ N(λ, λ).

举例：设 X ~ Po(20)，求 P(X ≤ 16)。近似为 Y ~ N(20, 20)，P(Y ≤ 16.5) = P(Z ≤ (16.5 − 20)/√20) = P(Z ≤ −0.7826) ≈ 0.2168。如果忘记连续性修正，直接算 P(Y ≤ 16) = P(Z ≤ −0.8944) ≈ 0.1856，误差明显。可见”±0.5″这一步虽然小，却直接决定答案对错，也是评分标准中专门设置的一个方法分。

Example: let X ~ Po(20) and find P(X ≤ 16). Approximate with Y ~ N(20, 20): P(Y ≤ 16.5) = P(Z ≤ (16.5 − 20)/√20) = P(Z ≤ −0.7826) ≈ 0.2168. If you forget the continuity correction and compute P(Y ≤ 16) = P(Z ≤ −0.8944) ≈ 0.1856 directly, the error is significant. The “±0.5” step is small but determines whether the answer is correct, and it carries a dedicated method mark in the mark scheme.

正态近似的考点常与假设检验结合：当 λ 很大时，用泊松分布直接算检验概率会非常繁琐，此时将检验统计量标准化为 Z 值、查标准正态表即可。做题时先确认 λ ≥ 15（或题目给出的阈值），写出近似分布 N(λ, λ)，再谨慎处理连续性修正，最后别忘了把 Z 值保留到合适的小数位数（通常 2 到 3 位）。

The normal approximation is often combined with hypothesis testing: when λ is very large, computing test probabilities directly from the Poisson distribution is extremely tedious, so you standardise the test statistic to a Z value and read the standard normal table. When solving, first confirm λ ≥ 15 (or the threshold given in the question), state the approximating distribution N(λ, λ), handle the continuity correction carefully, and finally keep the Z value to a suitable number of decimal places (usually 2 to 3).

八、泊松假设检验：单侧与双侧检验 | Hypothesis Testing with the Poisson: One-Tailed and Two-Tailed Tests

假设检验是 AQA 进阶数学 Paper 3 的重头戏，而”基于泊松分布的假设检验”几乎是每年的必考题型。检验的对象是参数 λ：原假设 H₀: λ = λ₀ 表示”平均发生率没有变化”，备择假设则根据题意取 λ > λ₀（单侧右尾）、λ < λ₀（单侧左尾）或 λ ≠ λ₀（双侧）。

Hypothesis testing is a centrepiece of AQA Further Mathematics Paper 3, and “hypothesis testing with a Poisson distribution” is almost a guaranteed topic every year. The object of the test is the parameter λ: the null hypothesis H₀: λ = λ₀ states that “the mean rate has not changed”, while the alternative hypothesis is λ > λ₀ (one-tailed upper), λ < λ₀ (one-tailed lower) or λ ≠ λ₀ (two-tailed), depending on the wording of the question.

标准解题步骤（务必按顺序书写）：第一步，定义随机变量 X 为”区间内事件次数”，写出 X ~ Po(λ₀)（在原假设下）。第二步，写出假设 H₀: λ = λ₀，H₁: λ > λ₀（或相应方向）。第三步，确定显著性水平 α（常见 5% 或 1%）。第四步，计算在原假设成立时观测值 x 对应的尾部概率，例如 P(X ≥ x)。第五步，比较：若尾部概率小于 α，则拒绝 H₀；否则不拒绝 H₀。第六步，用情境语言下结论，例如”有充分证据表明平均来电率显著上升”。

The standard solution steps (write them in order) are: first, define the random variable X as the number of events in the interval and write X ~ Po(λ₀) under the null hypothesis. Second, state H₀: λ = λ₀ and H₁: λ > λ₀ (or the appropriate direction). Third, note the significance level α (commonly 5% or 1%). Fourth, compute the tail probability corresponding to the observed value x under the null hypothesis, for example P(X ≥ x). Fifth, compare: if the tail probability is less than α, reject H₀; otherwise do not reject H₀. Sixth, conclude in the language of the context, for example “there is sufficient evidence that the mean call rate has increased significantly”.

单侧检验的方向判断是关键失分点。看到”has increased / more than / exceeds”取右尾 P(X ≥ x)；看到”has decreased / fewer than / less than”取左尾 P(X ≤ x)。双侧检验则要求把显著性水平对半分：临界值 c₁ 和 c₂ 分别满足 P(X ≤ c₁) ≤ α/2 且 P(X ≥ c₂) ≤ α/2，当观测值落入任一临界区域时拒绝 H₀。双侧检验中”观测值恰好在边界上”的情形要特别小心，按”小于等于临界概率才拒绝”的严格规则处理。

The direction of a one-tailed test is a key source of lost marks. For “has increased / more than / exceeds” use the upper tail P(X ≥ x); for “has decreased / fewer than / less than” use the lower tail P(X ≤ x). A two-tailed test requires splitting the significance level in half: the critical values c₁ and c₂ satisfy P(X ≤ c₁) ≤ α/2 and P(X ≥ c₂) ≤ α/2 respectively, and you reject H₀ when the observed value falls in either critical region. Be especially careful when the observed value lies exactly on the boundary in a two-tailed test: apply the strict rule that rejection requires the tail probability to be no greater than the threshold.

最后，检验结论必须与情境结合，不能只写”reject H₀”。AQA 评分标准要求结论句包含三个要素：证据强度（sufficient / insufficient）、统计动作（reject / do not reject H₀）、情境含义（例如”at the 5% level, there is sufficient evidence that the mean number of defects has increased”）。此外，若题目要求”find the critical region”，需要列出临界值的完整范围（例如 X ≥ 7），而不是只给一个数。

Finally, the conclusion must be tied to the context; writing only “reject H₀” is not enough. The AQA mark scheme requires the conclusion sentence to contain three elements: the strength of evidence (sufficient / insufficient), the statistical action (reject / do not reject H₀), and the contextual meaning (for example “at the 5% level, there is sufficient evidence that the mean number of defects has increased”). Also, if the question asks you to “find the critical region”, you must list the full range of critical values (for example X ≥ 7), not just a single number.

九、区间缩放与”至少一个”题型 | Scaling the Interval and “At Least One” Questions

泊松分布中 λ 与区间大小成正比，这是解决”换区间”类题目的核心原理。如果 X ~ Po(λ) 表示”单位区间内的平均事件数”，那么长度变为原来的 k 倍时，新的 λ’ = kλ。例如平均每小时收到 5 通电话，则每 2 小时平均收到 10 通，每 30 分钟平均收到 2.5 通。注意：缩放的是 λ，而不是概率本身。

In the Poisson distribution, λ is proportional to the size of the interval; this is the core principle for solving “change of interval” questions. If X ~ Po(λ) describes the mean number of events per unit interval, then when the interval is multiplied by a factor of k, the new parameter is λ’ = kλ. For example, if calls arrive at an average of 5 per hour, then the average is 10 per 2 hours and 2.5 per 30 minutes. Note that what scales is λ, not the probabilities themselves.

典型例题：某加油站平均每 10 分钟有 3 辆车进站，车辆到达数服从泊松分布。求 (a) 5 分钟内没有车辆进站的概率；(b) 15 分钟内至少有两辆车进站的概率。第 (a) 问先把 λ 缩放到 5 分钟：λ = 3 × 5/10 = 1.5，然后 P(X = 0) = e^−1.5 ≈ 0.2231。第 (b) 问缩放到 15 分钟：λ = 3 × 15/10 = 4.5，P(X ≥ 2) = 1 − P(X ≤ 1) = 1 − (e^−4.5 + 4.5e^−4.5) ≈ 1 − 0.0611 = 0.9389。

Typical example: a petrol station receives an average of 3 cars every 10 minutes, and arrivals follow a Poisson distribution. Find (a) the probability that no car arrives in 5 minutes; (b) the probability that at least two cars arrive in 15 minutes. For part (a), first scale λ to 5 minutes: λ = 3 × 5/10 = 1.5, then P(X = 0) = e^−1.5 ≈ 0.2231. For part (b), scale to 15 minutes: λ = 3 × 15/10 = 4.5, then P(X ≥ 2) = 1 − P(X ≤ 1) = 1 − (e^−4.5 + 4.5e^−4.5) ≈ 1 − 0.0611 = 0.9389.

这道题的两个小问完美展示了”区间缩放”的两条常见路线：题目给的是”每 10 分钟 3 辆”，问的是”5 分钟”和”15 分钟”，都需要先乘上比例系数。另一个常见变体是给”每小时”问”每天”，或者给”每 100 米”问”每 250 米”。无论区间变大还是变小，逻辑都一样：新 λ = 原 λ × (新区间 ÷ 原区间)。

These two parts perfectly illustrate the two common routes of “interval scaling”: the question gives “3 cars per 10 minutes” but asks about “5 minutes” and “15 minutes”, so both require multiplying by a scale factor first. Another common variant gives “per hour” and asks about “per day”, or gives “per 100 metres” and asks about “per 250 metres”. Whether the new interval is larger or smaller, the logic is the same: new λ = old λ × (new interval ÷ old interval).

“至少一个”与”至多一个”是这一节的送分题类型。P(X ≥ 1) = 1 − e^−λ，P(X = 0) = e^−λ，P(X ≤ 1) = e^−λ(1 + λ)。把这三个式子背熟，遇到”none / at least one / at most one”的英文表述就能秒反应。还要注意英文题干的用词差异：”no more than 2″ 是 P(X ≤ 2)，”fewer than 2″ 是 P(X ≤ 1)，”at least 2″ 是 P(X ≥ 2)，”more than 2″ 是 P(X ≥ 3)。

“At least one” and “at most one” are the gift questions of this section. P(X ≥ 1) = 1 − e^−λ, P(X = 0) = e^−λ, and P(X ≤ 1) = e^−λ(1 + λ). Memorise these three formulas and you will respond instantly to the English phrasings “none”, “at least one” and “at most one”. Also be alert to the wording differences in English questions: “no more than 2” means P(X ≤ 2), “fewer than 2” means P(X ≤ 1), “at least 2” means P(X ≥ 2), and “more than 2” means P(X ≥ 3).

十、AQA Paper 3 考试技巧：常见陷阱与四步解题法 | AQA Paper 3 Exam Technique: Common Traps and the Four-Step Method

综合多年真题，泊松分布相关题目最容易丢分的五个陷阱如下。陷阱一：忘记缩放 λ。题目给”每 20 分钟 4 个”却问”每 5 分钟”，直接用 λ = 4 计算，全错。陷阱二：尾部概率方向搞反。P(X ≥ k) 写成 1 − P(X ≤ k) 而不是 1 − P(X ≤ k − 1)。陷阱三：正态近似漏掉连续性修正，直接拿整数边界查表。陷阱四：假设检验结论只写统计术语、不写情境。陷阱五：把 λ 当整数处理，或把 λ 与 x 混为一谈。

Based on years of real examination papers, the five most common traps in Poisson-related questions are as follows. Trap one: forgetting to scale λ. The question gives “4 per 20 minutes” but asks about “per 5 minutes”, and you use λ = 4 directly, losing everything. Trap two: reversing the tail probability direction, writing P(X ≥ k) as 1 − P(X ≤ k) instead of 1 − P(X ≤ k − 1). Trap three: omitting the continuity correction in a normal approximation and reading the table with the raw integer boundary. Trap four: concluding hypothesis tests in statistical jargon only, without the context. Trap five: treating λ as an integer, or confusing λ with x.

应对大题，推荐使用”四步解题法”。第一步，Define：写清”Let X be the number of … in …”, 并写出分布 X ~ Po(λ)，注明 λ 的数值和单位区间。第二步，Compute：根据小问类型选择公式 – 单点概率用 PMF，区间概率用累积表，近似题写近似分布。第三步，Convert：把题目语言翻译成概率符号（at least → P(X ≥ k)，no more than → P(X ≤ k)）。第四步，Conclude：假设检验用情境语言下结论，普通计算题把答案保留到合适精度（概率通常保留 3 到 4 位小数）。

For multi-part questions, use the “four-step method”. Step one, Define: write “Let X be the number of … in …” and state the distribution X ~ Po(λ), noting the value of λ and the unit interval. Step two, Compute: choose the formula according to the type of part, using the PMF for single-point probabilities, cumulative tables for interval probabilities, and the approximating distribution for approximation questions. Step three, Convert: translate the wording into probability symbols (at least → P(X ≥ k), no more than → P(X ≤ k)). Step four, Conclude: draw conclusions in contextual language for hypothesis tests, and give numerical answers to a suitable precision for ordinary calculations (probabilities usually to 3 or 4 decimal places).

时间管理上，统计大题一般建议 10 到 15 分钟完成。如果某个小问卡住超过 3 分钟，先跳过做后面的部分，因为泊松大题的小问之间通常相互独立，后面的小问不依赖前面的答案。例如第 (a) 问求 P(X = 2)，第 (b) 问做假设检验，二者可以完全独立作答，没必要在一棵树上吊死。

On time management, a statistics multi-part question should take roughly 10 to 15 minutes. If a part stalls for more than 3 minutes, skip it and move on, because the parts of a Poisson question are usually independent of each other: for instance, part (a) might ask for P(X = 2) while part (b) runs a hypothesis test, and the two can be answered completely independently. There is no need to waste time on one part.

十一、综合例题演练：改编自真题的完整解答 | Worked Example: A Full Solution Adapted from a Real Exam Question

下面这道综合题改编自 AQA 进阶数学真题的典型结构，涵盖了本文讲到的所有核心考点。题目：某银行网点的客户到达数服从泊松分布，平均每 10 分钟到达 4.5 位客户。(a) 求 10 分钟内恰好有 3 位客户到达的概率；(b) 求 5 分钟内至少有 1 位客户到达的概率；(c) 用正态近似求 30 分钟内到达客户数不超过 10 的概率；(d) 该网点声称平均到达率仍为每 10 分钟 4.5 位，某日随机观察 10 分钟发现来了 9 位客户，在 5% 显著性水平下检验该声称是否成立。

The following integrated question is adapted from the typical structure of real AQA Further Mathematics papers and covers every core point in this article. Question: customer arrivals at a bank branch follow a Poisson distribution with a mean of 4.5 customers per 10 minutes. (a) Find the probability that exactly 3 customers arrive in 10 minutes. (b) Find the probability that at least 1 customer arrives in 5 minutes. (c) Using a normal approximation, find the probability that no more than 10 customers arrive in 30 minutes. (d) The branch claims the mean arrival rate is still 4.5 per 10 minutes; on one day, 9 customers arrive in a randomly observed 10-minute period. Test this claim at the 5% significance level.

第 (a) 问解答：设 X 为 10 分钟内到达的客户数，X ~ Po(4.5)。P(X = 3) = e^−4.5 × 4.5³ / 3! ≈ 0.1687。用计算器的 PoissonPD 功能核对结果一致。第 (b) 问解答：先把 λ 缩放到 5 分钟，λ = 4.5 × 5/10 = 2.25。设 Y 为 5 分钟内到达的客户数，Y ~ Po(2.25)。P(Y ≥ 1) = 1 − P(Y = 0) = 1 − e^−2.25 ≈ 1 − 0.1054 = 0.8946。

Solution to part (a): let X be the number of customers arriving in 10 minutes, so X ~ Po(4.5). Then P(X = 3) = e^−4.5 × 4.5³ / 3! ≈ 0.1687, which matches the PoissonPD function on a calculator. Solution to part (b): first scale λ to 5 minutes: λ = 4.5 × 5/10 = 2.25. Let Y be the number of customers arriving in 5 minutes, so Y ~ Po(2.25). Then P(Y ≥ 1) = 1 − P(Y = 0) = 1 − e^−2.25 ≈ 1 − 0.1054 = 0.8946.

第 (c) 问解答：30 分钟对应 λ = 4.5 × 3 = 13.5，且 λ ≥ 15 的近似条件不满足，严格来说题目应使用 λ ≥ 15 的情境；为展示方法，这里按 λ = 13.5 演示流程。设 W 为 30 分钟内到达的客户数，W ~ Po(13.5)，用 Y ~ N(13.5, 13.5) 近似。P(W ≤ 10) ≈ P(Y ≤ 10.5) = P(Z ≤ (10.5 − 13.5)/√13.5) = P(Z ≤ −0.8165) ≈ 0.2071。注意”no more than 10″对应 P(W ≤ 10)，连续性修正用 +0.5。

Solution to part (c): 30 minutes gives λ = 4.5 × 3 = 13.5, which does not strictly satisfy the λ ≥ 15 approximation condition; to demonstrate the method we still run the procedure with λ = 13.5. Let W be the number of customers arriving in 30 minutes, W ~ Po(13.5), approximated by Y ~ N(13.5, 13.5). Then P(W ≤ 10) ≈ P(Y ≤ 10.5) = P(Z ≤ (10.5 − 13.5)/√13.5) = P(Z ≤ −0.8165) ≈ 0.2071. Note that “no more than 10” corresponds to P(W ≤ 10), and the continuity correction adds 0.5.

第 (d) 问解答：设 X 为 10 分钟内到达的客户数。原假设 H₀: λ = 4.5，备择假设 H₁: λ ≠ 4.5（双侧，因为”是否成立”没有方向）。观测值 x = 9。计算 P(X ≥ 9) = 1 − P(X ≤ 8)。查表或计算器得 P(X ≤ 8) ≈ 0.9597，所以 P(X ≥ 9) ≈ 0.0403。双侧检验要求尾部概率与 α/2 = 0.025 比较：0.0403 > 0.025，因此不能拒绝 H₀。结论：在 5% 显著性水平下，没有充分证据表明平均到达率发生变化，该网点的声称可以接受。

Solution to part (d): let X be the number of customers arriving in 10 minutes. The null hypothesis is H₀: λ = 4.5 and the alternative is H₁: λ ≠ 4.5 (two-tailed, because “whether the claim holds” has no direction). The observed value is x = 9. Compute P(X ≥ 9) = 1 − P(X ≤ 8). From tables or a calculator, P(X ≤ 8) ≈ 0.9597, so P(X ≥ 9) ≈ 0.0403. For a two-tailed test, compare this tail probability with α/2 = 0.025: since 0.0403 > 0.025, we do not reject H₀. Conclusion: at the 5% significance level there is insufficient evidence that the mean arrival rate has changed, so the branch’s claim is accepted.

这道综合题完整覆盖了：定义随机变量、PMF 计算、区间缩放、正态近似加连续性修正、双侧假设检验五个考点。建议同学们合上答案，把 (a) 到 (d) 独立重做一遍，再对照评分标准自查每一步的过程分是否齐全，特别是 (b) 问的 λ 缩放和 (d) 问的双侧比较。

This integrated question fully covers five examination points: defining the random variable, PMF calculation, interval scaling, normal approximation with continuity correction, and two-tailed hypothesis testing. I recommend closing the answer, redoing parts (a) to (d) independently, and then checking against the mark scheme whether every method mark is present, especially the λ scaling in part (b) and the two-tailed comparison in part (d).

Summary | 总结

泊松分布是 AQA A-Level 进阶数学 Paper 3 的核心模型，本文围绕它梳理了七条必背要点。第一，适用条件：事件独立、平均发生率恒定、不同时发生、事件稀有。第二，定义与公式：X ~ Po(λ)，P(X = x) = e^−λλ^x/x!。第三，核心性质：E(X) = Var(X) = λ，这是判断数据是否服从泊松分布的依据。第四，累积概率换算：P(X ≥ k) = 1 − P(X ≤ k − 1)。第五，二项逼近：n 大 p 小时 B(n, p) ≈ Po(np)。第六，正态逼近：λ 大时 Po(λ) ≈ N(λ, λ)，务必加连续性修正。第七，假设检验：定义 H₀ 与 H₁、计算尾部概率、与 α（双侧为 α/2）比较、用情境语言下结论。

The Poisson distribution is the core model of AQA A-Level Further Mathematics Paper 3, and this article has organised seven essential points around it. First, the conditions of use: events are independent, the mean rate is constant, events do not occur simultaneously, and events are rare. Second, definition and formula: X ~ Po(λ) with P(X = x) = e^−λλ^x/x!. Third, the signature property: E(X) = Var(X) = λ, which is the criterion for judging whether data follow a Poisson distribution. Fourth, cumulative probability conversions: P(X ≥ k) = 1 − P(X ≤ k − 1). Fifth, the binomial approximation: when n is large and p is small, B(n, p) ≈ Po(np). Sixth, the normal approximation: when λ is large, Po(λ) ≈ N(λ, λ), always with the continuity correction. Seventh, hypothesis testing: state H₀ and H₁, compute the tail probability, compare with α (or α/2 for two-tailed tests), and conclude in contextual language.

在实战层面，务必养成”先定义、再缩放、后换算、终结论”的答题习惯：每次动笔前先写清随机变量和分布，遇到区间变化先缩放 λ，遇到 at least / no more than 先翻译成概率符号，最后用情境语言收尾。只要把七条要点和四步流程吃透，并配合近五年真题反复演练，泊松分布在 AQA Paper 3 中就是稳定的得分点。

At the practical level, develop the answering habit of “define first, then scale, then convert, and finally conclude”: before writing anything, state the random variable and its distribution; whenever the interval changes, scale λ first; whenever the wording says “at least” or “no more than”, translate it into probability symbols first; and finally close with contextual language. Once you master the seven essential points and the four-step procedure, and practise with the last five years of real papers, the Poisson distribution will be a reliable source of marks in AQA Paper 3.

更多咨询请联系16621398022（同微信）