📚 Deep Analysis of Past Papers for Cambridge Year 13 Statistics | 剑桥A2统计历年真题深度解析
Cambridge International A Level Statistics, typically taken in Year 13, builds on foundational probability and introduces powerful inferential techniques including hypothesis testing, Poisson processes, continuous random variables and the t‑distribution. Past examination papers reveal consistent patterns, common pitfalls and the precise standards of the marking scheme. This article provides a deep, topic‑by‑topic analysis of real exam‑style questions, unpacking the reasoning behind every step and highlighting strategies to maximise marks.
剑桥国际 A Level 统计课程（通常在 Year 13 学习）在基础概率之上引入了强有力的推断工具，包括假设检验、泊松过程、连续随机变量及 t 分布。历年真题试卷展现了稳定的题型分布、常见失分点以及评分方案的精确要求。本文按专题深度解析典型真题，逐步拆解解题逻辑，并强调获取高分的关键策略。
1. Poisson Distribution – Core Calculations | 泊松分布核心计算
A common past‑paper question provides a mean rate for one time interval but asks for a probability over a different interval. For example: ‘The number of emails received per hour follows a Poisson distribution with mean 4. Find the probability that more than 6 emails arrive in a 90‑minute period.’ The first step is to scale the mean proportionally: λ = 4 × 1.5 = 6. The required probability is P(X > 6) = 1 − P(X ≤ 6). Candidates often lose marks by forgetting to adjust λ or by misreading ‘more than’ as ‘at least’. Using cumulative Poisson tables or the formula P(X = k) = exp(−λ) λᵏ / k! efficiently is essential.
历年真题中常给出某一时间间隔的均值但要求计算另一时段的概率。例如：’每小时收到邮件数服从均值为 4 的泊松分布，求 90 分钟内收到超过 6 封邮件的概率。’ 第一步是按比例调整均值：λ = 4 × 1.5 = 6。所求概率为 P(X > 6) = 1 − P(X ≤ 6)。考生常因忘记缩放 λ 或将 ‘more than’ 误作 ‘at least’ 而失分。熟练使用累积泊松表或公式 P(X = k) = exp(−λ) λᵏ / k! 至关重要。
2. Poisson Approximation to the Binomial | 二项分布的泊松近似
Examiners frequently test the condition for approximation. A typical item: ‘A manufacturer finds that 2% of light bulbs are defective. A random sample of 200 bulbs is taken. Using a suitable approximation, find the probability that at most 3 are defective.’ Here n = 200 is large and p = 0.02 is small, so np = 4. The binomial B(200, 0.02) is approximated by Po(4). Then P(X ≤ 3) is read from tables. The mark scheme awards a method mark for stating the condition np < 5 (or n large, p small). Many answers incorrectly apply a continuity correction – the Poisson approximation does not use one.
考官频繁考查近似适用条件。典型题目：’某厂家发现灯泡缺陷率为 2%，随机抽取 200 只，使用合适近似求最多 3 只缺陷的概率。’ 此处 n = 200 很大，p = 0.02 很小，故 np = 4。二项分布 B(200, 0.02) 用 Po(4) 近似。之后由表读取 P(X ≤ 3)。评分方案为写明条件 np < 5（或 n 大、p 小）赋予方法分。许多答案错误地引入连续性校正——泊松近似无需校正。
3. Normal Approximation and Continuity Correction | 正态近似与连续校正
When the normal distribution is used to approximate a binomial or Poisson variable, the continuity correction is almost always required. A past paper might ask: ‘For X ~ B(100, 0.35), use a normal approximation to estimate P(30 < X ≤ 45).’ The parameters are μ = 35, σ² = 100 × 0.35 × 0.65 = 22.75. The adjusted bounds become P(30.5 ≤ X ≤ 45.5). The rescaled statistic is (45.5 − 35)/√22.75. Without the ±0.5 adjustment the approximation can be inaccurate and marks are deducted. The same logic applies to Poisson sums: P(X ≤ 12) with λ = 10 becomes Φ((12.5 − 10)/√10).
使用正态分布近似二项或泊松变量时，几乎总是需要连续校正。一份真题可能要求：’对于 X ~ B(100, 0.35)，用正态近似估计 P(30 < X ≤ 45)。’ 参数为 μ = 35, σ² = 100 × 0.35 × 0.65 = 22.75。调整后界限变为 P(30.5 ≤ X ≤ 45.5)。标准化统计量为 (45.5 − 35)/√22.75。若缺失 ±0.5 调整，近似精度下降且会被扣分。同样逻辑适用于泊松求和：P(X ≤ 12) 且 λ = 10 时，使用 Φ((12.5 − 10)/√10)。
4. Continuous Random Variables – From PDF to Median | 连续随机变量：从概率密度函数到中位数
A favourite exam task presents a probability density function such as f(x) = kx(5 − x) for 0 ≤ x ≤ 5 and zero otherwise. Part (a) asks for k, part (b) for the cumulative distribution function F(x), and part (c) for the median. Integration yields ∫₀⁵ kx(5 − x) dx = k [ (5x²/2) − (x³/3) ]₀⁵ = k (125/2 − 125/3) = k (125/6) = 1, so k = 6/125. The CDF is built piecewise: for 0 ≤ x ≤ 5, F(x) = ∫₀ˣ (6/125) t(5 − t) dt, giving a cubic expression. The median m satisfies F(m) = 0.5, solved by equating the cubic to 0.5 and testing which root lies in the support. Lost marks often arise from algebraic errors in integrating and from failing to state F(x) = 0 for x < 0 and F(x) = 1 for x > 5.
考试中常见的一类题目是给出概率密度函数，如 f(x) = kx(5 − x)，0 ≤ x ≤ 5，其余为零。(a) 求 k，(b) 求累积分布函数 F(x)，(c) 求中位数。通过积分 ∫₀⁵ kx(5 − x) dx = 1 得 k = 6/125。CDF 分段建立：0 ≤ x ≤ 5 时 F(x) = ∫₀ˣ (6/125) t(5 − t) dt，得到三次表达式。中位数 m 满足 F(m) = 0.5，令三次函数等于 0.5 并检查根是否在支撑集内。失分常源于积分代数错误，以及忘了写出 x < 0 时 F(x) = 0、x > 5 时 F(x) = 1。
5. Linear Combinations – Expectation and Variance | 线性组合的期望与方差
Deep questions combine independent normal variables. For instance: ‘The weights of apples are normally distributed with mean 120 g and standard deviation 15 g. Oranges have mean 100 g and standard deviation 10 g. Find the probability that the total weight of 5 apples and 3 oranges exceeds 1000 g.’ Let A ~ N(120, 15²) and O ~ N(100, 10²). The total T = 5A + 3O is normal with E(T) = 5×120 + 3×100 = 900 g, Var(T) = 5²×15² + 3²×10² = 5625 + 900 = 6525. Standardising, P(T > 1000) = 1 − Φ((1000 − 900)/√6525). A classic mistake is to add standard deviations rather than variances, or to treat the number of items as a multiplier for the mean only. Watch for non‑independent combinations where covariance must be included.
深度题常将独立正态变量结合。例如：’苹果重量服从正态分布，均值 120 g，标准差 15 g；橙子均值 100 g，标准差 10 g。求 5 个苹果和 3 个橙子总重量超过 1000 g 的概率。’ 设 A ~ N(120, 15²)，O ~ N(100, 10²)。总重 T = 5A + 3O 仍为正态，E(T) = 5×120 + 3×100 = 900 g，Var(T) = 5²×15² + 3²×10² = 6525。标准化后 P(T > 1000) = 1 − Φ((1000 − 900)/√6525)。典型错误是将标准差而非方差相加，或仅将物品数量作为均值的乘数。还需注意非独立组合时必须纳入协方差。
6. Hypothesis Test for a Mean – Variance Known | 总体均值检验（方差已知）
A standard z‑test appears regularly. A 2019 paper asked: ‘A machine fills bottles with a volume that is normally distributed with standard deviation 4 ml. The manufacturer claims the mean is 500 ml. A quality inspector takes a sample of 25 bottles and finds a sample mean of 497 ml. Test at the 5% significance level whether the mean has decreased.’ Here H₀: μ = 500, H₁: μ < 500. The test statistic is z = (497 − 500) / (4/√25) = −3.75. The critical value for a one‑tailed test at 5% is −1.645. Since −3.75 < −1.645, reject H₀. Equivalently, p‑value = Φ(−3.75) ≈ 0.00009, which is less than 0.05. Examiners want the conclusion stated in context: ‘There is sufficient evidence to suggest the mean volume has decreased.’
标准 z 检验频繁出现。2019 年真题：’一台机器装瓶，体积服从正态分布，标准差 4 ml。制造商声称均值为 500 ml。质检员抽取 25 瓶得样本均值 497 ml。以 5% 显著性水平检验均值是否下降。’ 设 H₀: μ = 500，H₁: μ < 500。检验统计量 z = (497 − 500) / (4/√25) = −3.75。单侧 5% 临界值为 −1.645。因 −3.75 < −1.645，拒绝 H₀。等价地，p 值 = Φ(−3.75) ≈ 0.00009 < 0.05。考官要求结论置于背景中：‘有充分证据表明平均体积已下降。’
7. t‑Test for a Population Mean – Small Sample | 总体均值的 t 检验（小样本）
When the population variance is unknown and the sample size is small, the t‑distribution is essential. A past paper scenario: ‘The lifetime of a component is assumed to be normally distributed. Nine components are tested and their lifetimes in hours are recorded. The sample mean is 1520 and the sample standard deviation is 90. Test at 1% significance whether the mean life differs from 1600 hours.’ With 8 degrees of freedom, H₀: μ = 1600, H₁: μ ≠ 1600 (two‑tailed). t = (1520 − 1600) / (90/√9) = −80 / 30 = −2.667. The two‑tailed critical value from t₈ is ±3.355. Since |−2.667| < 3.355, we do not reject H₀. A common error is using the z‑critical value or miscounting degrees of freedom (n−1).
当总体方差未知且样本量小时必须使用 t 分布。真题情境：’某元件寿命假定服从正态分布。测试 9 个元件，记录其寿命（小时），得样本均值 1520，样本标准差 90。在 1% 显著性水平下检验平均寿命是否不同于 1600 小时。’ 自由度 8，H₀: μ = 1600，H₁: μ ≠ 1600（双尾）。t = (1520 − 1600) / (90/√9) = −2.667。t₈ 双尾临界值为 ±3.355。因 |−2.667| < 3.355，不拒绝 H₀。常见错误是使用 z 临界值，或算错自由度（应为 n−1）。
8. Type I & II Errors and Power – Deep Interpretation | 第 I 类与第 II 类错误及功效
Higher‑tier questions examine error probabilities. A specimen problem: ‘A test of H₀: μ = 50 vs H₁: μ < 50 is conducted using a sample of size 25 from a normal distribution with σ = 8. The rejection region is x̄ < 47. Find the probability of Type I error. If the true mean is 46, find the probability of Type II error and the power of the test.’ Type I error: α = P(reject H₀ | H₀ true) = P(x̄ < 47 | μ = 50) = Φ((47 − 50)/(8/√25)) = Φ(−1.875) ≈ 0.0304. Type II error: β = P(accept H₀ | μ = 46) = P(x̄ ≥ 47 | μ = 46) = 1 − Φ((47 − 46)/1.6) = 1 − Φ(0.625) ≈ 0.266. Power = 1 − β ≈ 0.734. Mark schemes insist on clear definition of each error type and a diagram often helps.
高阶题目考查错误概率。一道样题：’由正态总体（σ = 8）抽取 n = 25，检验 H₀: μ = 50 vs H₁: μ < 50，拒绝域为 x̄ < 47。求第 I 类错误概率。若真实均值为 46，求第 II 类错误概率及检验功效。’ 第 I 类错误：α = P(拒绝 H₀ | H₀ 真) = P(x̄ < 47 | μ = 50) = Φ((47 − 50)/1.6) = Φ(−1.875) ≈ 0.0304。第 II 类错误：β = P(接受 H₀ | μ = 46) = P(x̄ ≥ 47 | μ = 46) = 1 − Φ((47 − 46)/1.6) ≈ 0.266。功效 = 1 − β ≈ 0.734。评分方案要求清晰定义两类错误，借助示意图更佳。
9. Poisson Hypothesis Testing | 泊松分布的假设检验
Unlike the normal case, Poisson tests rely on exact probabilities. In a past question: ‘A factory records an average of 0.8 accidents per week. After a safety campaign, 2 accidents occur in a 4‑week period. Test at 5% level whether the rate has reduced.’ Null hypothesis: λ = 0.8 × 4 = 3.2. The observed x = 2. Compute P(X ≤ 2 | λ = 3.2) using exp(−3.2)(1 + 3.2 + 3.2²/2) ≈ 0.379. This p‑value (0.379) > 0.05, so there is insufficient evidence to conclude the rate has reduced. Always scale λ to the observation window and remember the alternative direction – if testing for an increase, use P(X ≥ observed value).
与正态分布不同，泊松检验依赖精确概率。真题例如：’某工厂平均每周发生 0.8 起事故。安全活动后，4 周内发生 2 起事故。在 5% 水平下检验事故率是否下降。’ 原假设下 λ = 0.8 × 4 = 3.2，观测值 x = 2。计算 P(X ≤ 2 | λ = 3.2) ≈ 0.379。该 p 值 > 0.05，故证据不足以支持事故率降低。务必按观测时段换算 λ，并注意备择方向——若检验增大，应使用 P(X ≥ 观测值)。
10. Combined Problems – Conditional Probability and Distributions | 综合难题：条件概率与分布结合
Long‑mark questions weave several topics together. A typical exam final part: ‘20% of seeds from supplier A and 5% from supplier B fail to germinate. A gardener buys 70% of seeds from A. If 8 out of a batch of 100 fail to germinate, find the posterior probability that the batch came from B.’ First compute prior probabilities: P(fail | A) = 0.2, P(fail | B) = 0.05. Using Bayes’ theorem, the probability that the batch is from B given 8 failures in 100 is calculated from binomial likelihoods. Let F ~ Bin(100, p). P(data | B) = C(100,8) (0.05)⁸ (0.95)⁹² and similarly for A. The posterior combines these with priors 0.3 and 0.7. This tests both conditional probability manipulation and binomial probability evaluation. Such questions reward clear proportional reasoning and accurate calculator use.
高分值题目常综合多个知识点。典型压轴题：’供应商 A 的种子 20% 不发芽，供应商 B 的种子 5% 不发芽。园丁购买的种子 70% 来自 A。若一批 100 粒中有 8 粒不发芽，求该批种子来自 B 的后验概率。’ 首先明确先验：P(不发芽|A) = 0.2，P(不发芽|B) = 0.05。利用贝叶斯定理，给定 100 粒中 8 粒失败，来自 B 的概率由二项似然计算。设 F ~ Bin(100, p)。P(数据|B) = C(100,8) (0.05)⁸ (0.95)⁹²，A 类似。后验概率结合先验 0.3 和 0.7。此题既考察条件概率处理，又检验二项概率计算。清晰的比率推理与准确的运算可获高分。
11. Common Pitfalls in Year 13 Statistics Exams | A2 统计考试常见陷阱
An analysis of examiner reports reveals repeated mistakes. (1) Forgetting to check the conditions for an approximation – always state n large, p small for binomial‑to‑Poisson, or np and n(1‑p) > 5 for normal approximation. (2) Misapplying the continuity correction – using 0.5 when +0.5 is correct, or applying it for a Poisson‑to‑Poisson approximation. (3) Confusing standard deviation and variance in linear combinations – Var(aX ± bY) requires squared coefficients. (4) Using the normal critical value instead of a t‑value for small samples. (5) Reporting a ‘significant’ result without a contextual conclusion. (6) Mishandling two‑tailed tests by halving the significance level incorrectly in the p‑value comparison. Each pitfall is illustrated in official marking guidelines.
考官报告分析揭示了反复发生的错误。（1）忘记验证近似条件——二项近似泊松须写明 n 大 p 小，正态近似须 np 与 n(1−p) 均 > 5。（2）连续校正误用——应加 0.5 时却减了，或对泊松间近似画蛇添足。（3）线性组合时混淆标准差与方差——Var(aX ± bY) 需系数平方。（4）小样本下用正态临界值代替 t 值。（5）仅给出“显著”结论
Published by TutorHao | Year 13 统计 Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
屏轩国际教育cambridge primary/secondary checkpoint, cat4, ukiset,ukcat,igcse,alevel,PAT,STEP,MAT, ibdp,ap,ssat,sat,sat2课程辅导，国外大学本科硕士研究生博士课程论文辅导