📚 OxfordAQA 9665 FM03 Exam Report: Key Further Statistics Concepts | 牛津AQA 9665 FM03 考试报告：高等统计核心概念精讲
This article draws directly on the detailed examiner feedback from the OxfordAQA 9665 FM03 January 2023 examination series. The report highlighted several areas where candidates can significantly improve, particularly in applying probability generating functions, handling Chi-squared assumptions, and interpreting p-values correctly. By revisiting these essential Further Statistics concepts, we aim to turn common mistakes into marks at A-Level.
本文直接源自牛津AQA 9665 FM03 2023年1月考季的考官详细反馈。报告指出了考生可以显著提升的几个领域，尤其是在概率生成函数的应用、卡方检验假设的处理以及正确解释p值方面。通过重新梳理这些高等统计的核心概念，我们希望帮助同学们把A-Level考试中的常见错误转化为得分点。
1. Probability Generating Functions: Definition and First Principles | 概率生成函数：定义与基本原理
The probability generating function (P.G.F.) of a discrete non-negative integer-valued random variable X is defined as GX(t) = E(tX) = Σ tx P(X = x). In the exam, candidates often lost marks by not stating this definition explicitly when required, or by writing G(t) = Σ P(X = x) tx without explaining t. Emphasise that the P.G.F. exists for |t| ≤ 1, and that GX(1) = 1.
离散非负整值随机变量 X 的概率生成函数 (P.G.F.) 定义为 GX(t) = E(tX) = Σ tx P(X = x)。考试中，考生常因未明确写出该定义，或只写 G(t) = Σ P(X = x) tx 而不解释 t 而失分。要强调 P.G.F. 在 |t| ≤ 1 时存在，且 GX(1) = 1。
The examiner’s report noted that weaker answers confused GX(t) with the moment generating function, or failed to appreciate that the coefficient of tx directly gives P(X = x). Another typical error was writing G'(1) = E(X) without expanding the derivative correctly.
考官报告指出，较弱的答案混淆了 P.G.F. 与矩母函数，或者未能认识到 tx 的系数直接给出 P(X = x)。另一个典型错误是未正确展开导数就直接写 G'(1) = E(X)。
- Always begin ‘Let GX(t) = Σ tx p(x)’ for full marks.
务必以 ‘令 GX(t) = Σ tx p(x)’ 开头才能拿满分。 - Check that GX(1) = 1 to verify your expression.
验证你的表达式满足 GX(1) = 1 以作检查。
2. Using P.G.F.s to Find Mean and Variance | 利用 P.G.F. 求均值和方差
The expectation of X is E(X) = G’X(1). The variance can be found using Var(X) = G”X(1) + G’X(1) − [G’X(1)]2. A common mistake in the exam was to stop after finding G”X(1) and assume this alone gives the variance, neglecting the extra terms. Many candidates also struggled when the P.G.F. was given in a composite form, requiring careful differentiation of (at + b)n.
X 的期望为 E(X) = G’X(1)。方差可用 Var(X) = G”X(1) + G’X(1) − [G’X(1)]2 求得。考试中常见错误是求出 G”X(1) 后就以为这等于方差，漏掉了后面的项。许多考生在处理复合形式的 P.G.F.（需对 (at + b)n 求导）时也遇到困难。
When the P.G.F. is given as GX(t) = (0.3t + 0.7)4, remember to apply the chain rule: G’X(t) = 4 × 0.3 × (0.3t + 0.7)3, then evaluate at t = 1. The second derivative follows similarly.
当已知 GX(t) = (0.3t + 0.7)4 时，要记住用链式法则：G’X(t) = 4 × 0.3 × (0.3t + 0.7)3，然后代入 t = 1 求值。二阶导数类似处理。
Var(X) = G″(1) + G′ (1) − [G′(1)]2
3. P.G.F. of the Sum of Independent Variables | 独立变量之和的 P.G.F.
If X and Y are independent, GX+Y(t) = GX(t) × GY(t). This property is frequently used to derive the distribution of a total score or to identify the distribution of a sum of Poissons or binomials. The examiner observed that some candidates multiplied P.G.F.s even when the variables were not independent, or they incorrectly multiplied the P.G.F. by a constant.
若 X 和 Y 独立，则 GX+Y(t) = GX(t) × GY(t)。该性质常用于推导总得分的分布，或识别多个泊松或二项分布之和的分布。考官观察到，部分考生在变量不独立时仍将 P.G.F. 相乘，或错误地用一个常数乘以 P.G.F.。
In one report question, candidates needed to recognise that the sum of independent Poisson variables is Poisson, and therefore the P.G.F. is eλ(t−1) eμ(t−1) = e(λ+μ)(t−1). Points were lost by not explicitly stating the independence condition.
在报告中某题里，考生需认识到独立泊松变量之和仍为泊松分布，因此 P.G.F. 为 eλ(t−1) eμ(t−1) = e(λ+μ)(t−1)。未明确陈述独立条件导致失分。
4. Hypothesis Testing with the Poisson Distribution | 泊松分布下的假设检验
The Poisson hypothesis test typically involves a null hypothesis H0: λ = λ0 against a one- or two-tailed alternative. Candidates must find P(X ≥ k) or P(X ≤ k) and compare with the significance level. The exam report stressed that many students used an incorrect critical region or misidentified when the observed value was in the tail.
泊松假设检验通常涉及原假设 H0: λ = λ0，对单侧或双侧备择。考生须计算 P(X ≥ k) 或 P(X ≤ k) 并与显著性水平比较。考试报告强调，许多学生使用了错误的拒绝域，或错误判断观测值是否落入尾部。
It is essential to define the critical region clearly before looking at the data. For a two-tailed test at the 5% level, both tails should contain probability ≤ 0.025. A persistent error was to double the observed one-tailed p-value without checking symmetry of the distribution.
在查看数据前明确定义拒绝域至关重要。对 5% 水平的双侧检验，每个尾部概率应 ≤ 0.025。一个持续出现的错误是在不检验分布对称性的情况下直接将单侧 p 值乘以 2。
When using p-value approach: if test statistic is x, find probability of at least as extreme under H0, and reject H0 if p < significance level.
使用 p 值方法时：若检验统计量为 x，计算在原假设下至少达到如此极端的概率，若 p 小于显著性水平则拒绝 H0。
5. Type I and Type II Errors | 第一类错误和第二类错误
A Type I error occurs when H0 is rejected when it is true; its probability is exactly the significance level. A Type II error is failing to reject H0 when H1 is true. In the January 2023 FM03 paper, many candidates could state the definitions but struggled to calculate the actual probability of a Type II error for a specific alternative λ.
第一类错误发生在 H0 为真却被拒绝；其概率恰好为显著性水平。第二类错误是 H1 为真却未能拒绝 H0。在 2023 年 1 月 FM03 试卷中，许多考生能陈述定义，但在计算特定备择 λ 下的第二类错误概率时出现困难。
Calculating the probability of a Type II error requires using the distribution under H1. For example, with critical region X ≥ 7 under H0: λ = 4, and true λ = 6, the probability of Type II error is P(X ≤ 6 | λ = 6). This was frequently miscomputed as P(X ≤ 6 | λ = 4).
计算第二类错误概率需使用 H1 下的分布。例如，在 H0: λ = 4 下拒绝域为 X ≥ 7，而真实 λ = 6，则第二类错误概率为 P(X ≤ 6 | λ = 6)。这常被错算为 P(X ≤ 6 | λ = 4)。
Always clearly indicate which parameter you are using in each probability statement to avoid this confusion.
务必在每个概率表达式中清楚标明所使用的参数，以避免混淆。
6. Chi-Squared Goodness-of-Fit Test | 卡方拟合优度检验
This test assesses whether observed frequencies match an expected distribution. The test statistic is χ2 = Σ (Oi − Ei)2 / Ei. The examiner noted that students often lost marks by failing to check that all expected frequencies are at least 5. When an Ei is below 5, categories must be combined, and the degrees of freedom recalculated accordingly.
该检验评估观测频数是否匹配预期分布。检验统计量为 χ2 = Σ (Oi − Ei)2 / Ei。考官指出，学生常因未检查所有预期频数是否至少为 5 而失分。当某个 Ei 小于 5 时，必须合并分类并相应重新计算自由度。
Degrees of freedom (ν) for a goodness-of-fit test are (number of categories after pooling − 1 − number of estimated parameters). A subtle error was to subtract 1 instead of the correct number of parameters when the expected distribution was specified using sample estimates, such as estimating p from a binomial model.
拟合优度检验的自由度 (ν) 为（合并后的分类数 − 1 − 估计参数个数）。当期望分布使用样本估计（如从二项模型中估计 p）时，一个易犯错误是只减 1 而未减去正确的参数个数。
Example: testing a binomial distribution Bin(5, p) where p is estimated from the data, then ν = (number of pooled classes − 2). Always state ν explicitly before consulting tables.
示例：检验二项分布 Bin(5, p) 且 p 由数据估计，则 ν =（合并后分类数 − 2）。在查表前务必明确给出 ν。
7. Chi-Squared Test for Association in Contingency Tables | 列联表中的卡方独立性检验
For an r × c contingency table, expected frequencies are calculated as Eij = (row total × column total) / grand total. The degrees of freedom are (r−1)(c−1). The report revealed that candidates often used the formula for goodness-of-fit expected frequencies by mistake, leading to completely wrong test statistics.
对 r × c 列联表，预期频数计算公式为 Eij =（行总和 × 列总和）/ 总和。自由度为 (r−1)(c−1)。报告显示，考生常误用拟合优度的预期频数公式，导致检验统计量完全错误。
Another point of confusion was the interpretation of the conclusion: a significant result indicates an association, not a direct causation. Additionally, many students attempted to conduct the test when more than 20% of expected frequencies were below 5, which violates the approximation’s validity.
另一个混淆点是对结论的解释：显著结果意味着存在关联，而非直接因果关系。此外，许多学生在超过 20% 的预期频数小于 5 时仍进行检验，这违反了近似的有效性。
The formula χ2 = Σ (O−E)2/E is the same, but the E’s are computed differently. Practice distinguishing the two scenarios.
χ2 = Σ (O−E)2/E 公式相同，但 E 的计算方式不同。练习区分两种情形。
8. Continuous Distributions: Using PDF and CDF | 连续分布：PDF 与 CDF 的应用
The FM03 paper included tasks where a continuous random variable had a given probability density function (p.d.f.) f(x) on [a, b]. Candidates had to find the cumulative distribution function F(x) = ∫ax f(t) dt and use it to determine median or quartiles by solving F(m) = 0.5. Errors often appeared in the piecewise definition of F(x), with many forgetting to set F(x) = 0 before a and 1 after b.
FM03 试卷中包含给定区间 [a, b] 上概率密度函数 f(x) 的连续随机变量。考生需求累积分布函数 F(x) = ∫ax f(t) dt，并通过解 F(m) = 0.5 来确定中位数或四分位数。错误常出现在 F(x) 的分段定义中，许多人忘记在 a 之前设 F(x) = 0、在 b 之后设 1。
The median m satisfies F(m) = 0.5. Ensure that your integration yields the correct constant term. A common slip was solving ∫0m f(x) dx = 0.5 but forgetting that the lower limit is a, not always 0. Always verify that 0 ≤ F(x) ≤ 1 is maintained.
中位数 m 满足 F(m) = 0.5。要确保积分得出正确的常数项。常见失误是解 ∫0m f(x) dx = 0.5 却忘记积分下限为 a，不总是 0。请始终验证是否保持 0 ≤ F(x) ≤ 1。
9. Normal Approximation to Poisson and Binomial | 泊松与二项的正态近似
When λ is large (e.g., λ > 15), the Poisson(λ) can be approximated by N(λ, λ). For binomial, use N(np, np(1−p)) when np and n(1−p) are both > 5. The continuity correction is mandatory when moving from a discrete to a continuous distribution. Many FM03 scripts omitted the correction, producing inaccurate probabilities.
当 λ 较大（如 λ > 15）时，泊松(λ) 可用 N(λ, λ) 近似。对于二项分布，当 np 与 n(1−p) 均大于 5 时使用 N(np, np(1−p))。从离散到连续分布时，连续性校正是强制的。许多 FM03 答卷省去了校正，导致概率不准确。
Example: for X ~ Po(25), P(X < 20) ≈ P(Y < 19.5) with Y ~ N(25, 25), then standardise. Without the −0.5 adjustment, the approximation is less reliable. The examiner stressed writing the correction step explicitly.
例如：X ~ Po(25)，P(X < 20) ≈ P(Y < 19.5) 其中 Y ~ N(25, 25)，然后标准化。若未进行 −0.5 调整，近似可靠性降低。考官强调必须明确写出校正步骤。
| Discrete | Correct Normal Limit |
| X ≤ k | X ≤ k + 0.5 |
| X ≥ k | X ≥ k − 0.5 |
10. Interpreting p-Values and Statistical Significance | 理解 p 值与统计显著性
The examiner report highlighted a recurring misconception: “p-value is the probability that H0 is true”. This is incorrect. The p-value is the probability of obtaining a test statistic at least as extreme as the one observed, assuming H0 is true. It is not the error probability of H0 itself. Stating the interpretation correctly is often required for the final mark.
考官报告强调了一个反复出现的误解：“p 值是 H0 为真的概率”。这是错误的。p 值是在假设 H0 成立的条件下，得到至少与观测值一样极端的检验统计量的概率。它不是 H0 本身的错误概率。正确陈述解释往往是拿到最后一分的关键。
A related error was concluding ‘accept H0‘ when the p-value was large. The accepted phrase is ‘do not reject H0‘ or ‘insufficient evidence to reject H0‘, because we never accept the null hypothesis as true. Practice phrasing your conclusions in context of the problem.
相关错误是当 p 值较大时得出结论 ‘接受 H0‘。应使用的表述是 ‘不拒绝 H0‘ 或 ‘证据不足以拒绝 H0‘，因为我们从不接受原假设为真。练习在题目情境下措辞结论。
A clear sentence like: “Since p = 0.032 < 0.05, there is sufficient evidence at the 5% level to reject H0 and conclude that the mean number of defects has increased.” gains full credit.
明确的句子如：“由于 p = 0.032 < 0.05，在 5% 水平下有足够证据拒绝 H0，并认为平均缺陷数已增加。”可拿满分。
Published by TutorHao | Further Mathematics 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