Pre-U CAIE Statistics: Paper Writing Framework and Model Answers | Pre-U CAIE 统计：论文写作框架与范文

📚 Pre-U CAIE Statistics: Paper Writing Framework and Model Answers | Pre-U CAIE 统计：论文写作框架与范文

Mastering the art of writing a statistics paper under the Cambridge Pre-U specification goes beyond number crunching – it demands a structured argument, clear communication of statistical reasoning, and the ability to interpret results in context. This guide breaks down the essential framework for crafting high-scoring responses, from planning and hypothesis formulation to reporting p-values and drawing conclusions. Model paragraphs and annotated examples illustrate how to meet the rigorous standards of the CAIE Pre-U Statistics examiners.

掌握剑桥 Pre-U 统计学的论文写作艺术，不仅仅在于数字运算——它要求结构清晰、统计推理论述明白，并能在实际背景下解读结果。本指南拆解了高分答卷的关键框架，从规划与假设构建，到 p 值汇报与结论得出，一应俱全。通过范文段落和注释示例，阐释如何满足 CAIE Pre-U 统计学考官的高标准要求。

1. Understanding the Pre-U Statistics Paper Structure | 理解 Pre-U 统计论文的结构

Pre-U Statistics papers typically include both short-answer questions and longer, essay-style investigations. The essay component usually presents a real-world scenario with a dataset or summary statistics, requiring you to choose an appropriate test, carry out calculations, interpret findings, and discuss limitations. Even when the question appears open-ended, examiners look for a logical flow: define the problem, state hypotheses, describe the model, perform analysis, check assumptions, and conclude.

Pre-U 统计试卷通常包含简答题和较长的论文式探究题。论文部分往往给出一个真实场景，附带数据集或汇总统计量，要求你选择合适的检验方法、进行计算、解读结果并讨论局限性。即使题目看似开放，考官追求的仍然是一条逻辑主线：界定问题、陈述假设、描述模型、执行分析、检验假定并得出结论。

A typical high-mark question might ask: ‘Assess whether there is evidence of a difference in the mean lifetimes of two brands of battery, using a 5% significance level. Comment on the validity of your analysis.’ The expected structure mirrors a mini statistical report, not just a mechanic’s output.

一道典型的高分题可能会问：“利用 5% 显著性水平，评估是否有证据表明两种品牌电池的平均寿命存在差异。并评论你分析的有效性。”预期的答案结构类似于一份微型统计报告，而非机械化的演算产物。

2. Pre-Writing: Interpreting the Question and Planning | 写作前：解读题目与规划

Begin by highlighting command words: ‘test’, ‘estimate’, ‘compare’, ‘assess’. Identify the variable type (continuous, discrete, categorical) and the nature of the investigation (one-sample, two-sample, paired, association). This will steer you toward the correct test – t, z, chi-squared, Wilcoxon, or a non-parametric alternative. Sketch a brief plan on the question paper: (1) Hypotheses in symbols and words; (2) Test statistic and its distribution under H₀; (3) Calculations; (4) p-value or critical region; (5) Conclusion in context.

先圈出指令词：“检验”、“估算”、“比较”、“评估”。识别变量类型（连续、离散、分类）和研究性质（单样本、双样本、配对、关联）。这将引导你选择正确的检验方法——t 检验、z 检验、卡方检验、Wilcoxon 检验或某种非参数替代方法。在试卷上草拟一个简要提纲：(1) 用符号和文字表述假设；(2) 检验统计量及其在 H₀ 下的分布；(3) 计算过程；(4) p 值或临界域；(5) 结合实际背景得出结论。

For example, if you encounter ‘Is there a link between gender and preference for three types of coffee?’, you know chi-squared test for independence is appropriate. Quickly note: rows = gender, columns = coffee type, expected frequencies under H₀, degrees of freedom = (2-1)×(3-1)=2.

例如，当你看到“性别与三种咖啡偏好之间是否存在关联？”时，你明白适用卡方独立性检验。迅速记下：行 = 性别，列 = 咖啡种类，H₀ 下的期望频数，自由度 = (2-1)×(3-1)=2。

3. Crafting a Strong Opening: Defining the Problem and Hypotheses | 写好开头：界定问题与设定假设

Your first paragraph should state the objective in your own words and formally present hypotheses. Use clear notation: H₀: μ₁ = μ₂ vs H₁: μ₁ ≠ μ₂, or H₀: ρ = 0 vs H₁: ρ > 0. Always define parameters explicitly – ‘where μ₁ is the population mean lifetime of Brand A batteries’. This shows the examiner you understand what is being tested.

第一段应用自己的话陈述研究目的，并正式提出假设。使用清晰符号：H₀: μ₁ = μ₂ vs H₁: μ₁ ≠ μ₂，或 H₀: ρ = 0 vs H₁: ρ > 0。永远明确定义参数——“其中 μ₁ 为品牌 A 电池的总体平均寿命”。这向考官表明你理解正在检验的对象。

If the question involves a one-tailed test, justify the direction. For instance, ‘Previous research suggests that Brand A lasts longer on average, so we test H₁: μ₁ > μ₂.’ Avoid vague language; stick to parameter-centric formulations. This opening signals a well-structured answer.

如果题目涉及单侧检验，需说明方向性理由。例如：“前期研究表明品牌 A 的平均寿命更长，因此我们检验 H₁: μ₁ > μ₂。”避免模糊用语，紧扣以参数为核心的表述。这样的开头预示着一份结构良好的答案。

4. Model and Assumptions: Setting the Statistical Scene | 模型与假定：搭建统计场景

Explicitly state the statistical model you are using. For a two-sample t-test: ‘Assume that the two samples are independent, randomly drawn from normally distributed populations with equal variance.’ Name the test statistic and its distribution under the null hypothesis, for example: ‘Under H₀, T = (x̄₁ – x̄₂) / (s_p√(1/n₁ + 1/n₂)) follows a t-distribution with n₁+n₂−2 degrees of freedom.’ For chi-squared tests, mention expected frequencies and the condition that no more than 20% are below 5.

明确陈述你使用的统计模型。对于双样本 t 检验：“假设两个样本独立且随机抽取自等方差的正态总体。”写出原假设下的检验统计量及其分布，例如：“在 H₀ 下，T = (x̄₁ – x̄₂) / (s_p√(1/n₁ + 1/n₂)) 服从自由度为 n₁+n₂−2 的 t 分布。”对于卡方检验，要提及期望频数以及不超过 20% 的格值低于 5 的条件。

Check assumptions critically. If the sample size is small, mention normality assumption and perhaps refer to boxplots or Q-Q plots if provided. If variances seem unequal, you might choose Welch’s t-test. Acknowledge these decisions – examiners reward awareness of model limitations, not blind application of formulas.

批判性地检验假定。若样本量小，可提及正态性假设，如果题目提供了箱形图或 Q-Q 图，可加以引用。若方差不相等，宜选用 Welch t 检验。确认这些决策——考官欣赏对模型局限性的认识，而非盲目套用公式。

5. Performing Calculations: Step-by-Step with Transparency | 执行计算：步步清晰，过程透明

Show your working systematically. When calculating a test statistic, write down the formula first, substitute values, then give the result to appropriate precision (usually 3 significant figures). For example: ‘s_p² = [(n₁−1)s₁² + (n₂−1)s₂²] / (n₁+n₂−2) = [(19×2.3² + 23×2.1²)/42] = 4.79, so s_p ≈ 2.19.’ This allows the examiner to follow your logic and award method marks even if an arithmetic slip occurs.

系统展示计算过程。计算检验统计量时，先写出公式，代入数值，然后给出合适精度（通常 3 位有效数字）的结果。例如：“s_p² = [(n₁−1)s₁² + (n₂−1)s₂²] / (n₁+n₂−2) = [(19×2.3² + 23×2.1²)/42] = 4.79，故 s_p ≈ 2.19。”这使考官能跟上你的逻辑，即使出现算术疏漏也能给予方法分。

For p-values, use distribution tables intelligently. State: ‘From t-tables with 42 d.f., the critical value at 5% (two-tailed) is approximately 2.021. Our observed t = 2.54 exceeds this, so p < 0.05.' If you have access to exact p-values via calculator, report e.g. 'p = 0.0147'. Avoid over-rounding – a p-value of 0.000 is never correct; write p < 0.001.

对于 p 值，灵活使用分布表。可以陈述：“查自由度为 42 的 t 分布表，5%（双侧）临界值约为 2.021。我们观测到的 t = 2.54 大于此值，故 p < 0.05。”若能用计算器获得精确 p 值，报告如“p = 0.0147”。避免过度四舍五入——p 值写成 0.000 永远不正确；应写 p < 0.001。

6. Interpreting Results in Context | 结合实际背景解读结果

This is where many students stumble: they stop at ‘reject H₀’ without translating the finding into plain English. Always link back to the original problem. Write: ‘There is sufficient evidence at the 5% level to conclude that the mean lifetime of Brand A batteries is greater than that of Brand B. The estimated difference is 3.5 hours (95% CI: 0.8 to 6.2 hours).’

这正是许多学生失分的地方：他们停留在“拒绝 H₀”，而没有将发现转化为平实的语言。永远要回扣原问题。可以写：“在 5% 水平上，有充分证据推断品牌 A 电池的平均寿命大于品牌 B。估计差异为 3.5 小时（95% 置信区间：0.8 至 6.2 小时）。”

If a confidence interval is given, interpret it precisely: ‘We are 95% confident that the true difference in population means lies between 0.8 and 6.2 hours. Since the interval does not contain 0, this aligns with the hypothesis test result.’ Avoid saying ‘there is a 95% probability that the true difference is in this interval’ – it is a confidence statement about the method, not a probability about the parameter.

若给出置信区间，精确解读：“我们有 95% 的把握认为总体均值之差介于 0.8 与 6.2 小时之间。由于该区间不包含 0，这与假设检验结果一致。”避免说“真实差异有 95% 的概率落在此区间内”——这是关于方法的置信度陈述，而非关于参数的概率陈述。

7. Discussion of Limitations and Diagnostic Checks | 讨论局限性与诊断检验

Examiners expect a critical evaluation. Mention potential violations of assumptions: ‘The sample sizes were moderate (n₁=20, n₂=24), so the t-test is reasonably robust to mild non-normality, but we should check for outliers or skewness.’ Refer to any provided plots. If variances were pooled, comment on the assumption of equal variance: ‘The ratio of sample variances is 1.2, which is not extreme, but Levene’s test could formally verify homogeneity.’

考官期待批判性评估。提及可能违反假设的情况：“样本量中等（n₁=20, n₂=24），所以 t 检验对轻度非正态具有一定稳健性，但我们应当检查离群值或偏态。”如有图示，加以引用。若合并了方差，评论方差相等的假设：“样本方差之比为 1.2，不算极端，但 Levene 检验可正式验证方差齐性。”

For chi-squared tests, discuss whether any expected frequencies were low and how that might affect validity. For regression, comment on residual plots, influence points, and R² interpretation. Acknowledging limitations demonstrates deeper statistical understanding and distinguishes top-scoring scripts.

对于卡方检验，讨论是否有任何期望频数偏低以及这可能如何影响有效性。对于回归，评论残差图、影响点以及 R² 的解读。承认局限性展示了更深层的统计理解，是高分答卷的分野。

8. Model Answer: Full Worked Example (Two-Sample t-Test) | 范文：完整示例（双样本 t 检验）

Question: ‘A farmer wishes to compare the yields of two wheat varieties, A and B, grown in adjacent plots. The yields (tonnes per hectare) from 12 plots of Variety A and 10 plots of Variety B are summarised as follows: Variety A – x̄₁=8.4, s₁=0.62; Variety B – x̄₂=7.9, s₂=0.58. Stating any assumptions, test at the 1% significance level whether there is a difference in mean yields.’

题目：“某农场主希望比较小麦品种 A 与 B 的产量，二者种植于相邻地块。品种 A 的 12 块地及品种 B 的 10 块地的产量（吨/公顷）汇总如下：品种 A – x̄₁=8.4, s₁=0.62；品种 B – x̄₂=7.9, s₂=0.58。列出任何假设，并在 1% 显著性水平下检验平均产量是否存在差异。”

Model Response | 范文回答

Objective: We wish to determine whether the population mean yields of Variety A and Variety B differ. Let μ₁ be the mean yield of Variety A, μ₂ be the mean yield of Variety B. Hypotheses: H₀: μ₁ = μ₂; H₁: μ₁ ≠ μ₂ (two-tailed test).

目标：我们希望判断品种 A 与品种 B 的总体平均产量是否存在差异。设 μ₁ 为品种 A 的平均产量，μ₂ 为品种 B 的平均产量。假设：H₀: μ₁ = μ₂；H₁: μ₁ ≠ μ₂（双侧检验）。

Model and assumptions: We use a two-sample t-test assuming independent random samples, approximate normality of yields (plausible for crop data), and equal population variances. The pooled variance estimate is:

s_p² = [(n₁−1)s₁² + (n₂−1)s₂²] / (n₁+n₂−2) = [11×0.62² + 9×0.58²] / 20

Calculating: 11×0.3844 = 4.2284; 9×0.3364 = 3.0276; sum = 7.256; s_p² = 7.256/20 = 0.3628, so s_p ≈ √0.3628 ≈ 0.6023.

模型与假设：我们使用双样本 t 检验，假定样本独立随机，产量近似正态（农业数据通常合理），且总体方差相等。合并方差估计为：

s_p² = [(n₁−1)s₁² + (n₂−1)s₂²] / (n₁+n₂−2) = [11×0.62² + 9×0.58²] / 20

计算：11×0.3844 = 4.2284; 9×0.3364 = 3.0276; 合计 = 7.256; s_p² = 7.256/20 = 0.3628，故 s_p ≈ √0.3628 ≈ 0.6023。

Test statistic: Under H₀, T = (x̄₁ − x̄₂) / [s_p √(1/n₁ + 1/n₂)] ~ t(20). Compute denominator: s_p √(1/12 + 1/10) = 0.6023 × √(0.08333+0.1) = 0.6023 × √0.18333 ≈ 0.6023 × 0.4282 ≈ 0.2579. Then T = (8.4 − 7.9) / 0.2579 = 0.5 / 0.2579 ≈ 1.939.

检验统计量：在 H₀ 下，T = (x̄₁ − x̄₂) / [s_p √(1/n₁ + 1/n₂)] ~ t(20)。计算分母：s_p √(1/12 + 1/10) = 0.6023 × √(0.08333+0.1) = 0.6023 × √0.18333 ≈ 0.6023 × 0.4282 ≈ 0.2579。于是 T = (8.4 − 7.9) / 0.2579 = 0.5 / 0.2579 ≈ 1.939。

Critical value and decision: From t-tables, the two-tailed 1% critical value with 20 d.f. is 2.845. Since |T| = 1.939 < 2.845, we do not reject H₀. The exact p-value is approximately 0.066 (via technology), which exceeds 0.01.

临界值与决策：查 t 表，自由度为 20 时双侧 1% 临界值为 2.845。由于 |T| = 1.939 < 2.845，我们不拒绝 H₀。精确 p 值约为 0.066（通过技术手段），大于 0.01。

Conclusion in context: There is insufficient evidence at the 1% significance level to claim a difference between the mean yields of the two wheat varieties. The observed difference of 0.5 tonnes per hectare could plausibly be due to sampling variability. A 99% confidence interval for (μ₁ − μ₂) is 0.5 ± 2.845×0.2579 ≈ (−0.23, 1.23), which includes zero, consistent with the test.

结合背景的结论：在 1% 显著性水平下，没有足够证据声称两个小麦品种的平均产量存在差异。观察到的 0.5 吨/公顷的差异可能合理地源自抽样变异。针对 (μ₁ − μ₂) 的 99% 置信区间为 0.5 ± 2.845×0.2579 ≈ (−0.23, 1.23)，包含 0，与检验结果一致。

Limitations: The sample sizes are small, limiting the power of the test. The assumption of equal variances might be questioned, though the ratio s₁²/s₂² ≈ 1.14 is close to 1. We also have not seen diagnostic plots to verify normality; in practice, a check for outliers would be prudent.

局限性：样本量较小，限制了检验的功效。方差相等的假设可能被质疑，尽管 s₁²/s₂² ≈ 1.14 接近 1。我们也没有看到诊断图来验证正态性；在实际操作中，谨慎起见应检查离群值。

9. Model Paragraph: Chi-Squared Test for Independence | 范文段落：卡方独立性检验

Scenario: ‘A survey of 200 people examines the relationship between age group (under 30, 30-50, over 50) and preference for organic food (Yes/No). The observed frequencies are given. Test for independence at the 5% level.’

场景：“一项针对 200 人的调查研究了年龄组（30 岁以下、30-50 岁、50 岁以上）与有机食品偏好（是/否）之间的关系。给出了观测频数。在 5% 水平下检验独立性。”

H₀: Age group and organic food preference are independent. H₁: They are not independent. Expected frequencies calculated under independence: for each cell, E = (row total × column total) / grand total. For example, expected count for ‘under 30, Yes’ = (60×80)/200 = 24.

H₀：年龄组与有机食品偏好相互独立。H₁：它们不独立。在独立性假设下计算期望频数：每个格子的 E = (行合计 × 列合计) / 总计。例如，“30 岁以下、是”的期望频数 = (60×80)/200 = 24。

The chi-squared statistic: χ² = Σ (O−E)²/E. Computing each cell: (28−24)²/24 = 0.667; … (summing all six cells) yields χ² = 7.82. Degrees of freedom = (3−1)×(2−1) = 2. From tables, critical value at 5% with 2 d.f. is 5.991. Since 7.82 > 5.991, we reject H₀. p-value ≈ 0.020.

卡方统计量：χ² = Σ (O−E)²/E。计算每个格子：(28−24)²/24 = 0.667；……（加总所有六个格子）得到 χ² = 7.82。自由度 = (3−1)×(2−1) = 2。查表知 5% 水平下自由度为 2 的临界值为 5.991。由于 7.82 > 5.991，我们拒绝 H₀。p 值 ≈ 0.020。

Conclusion: There is significant evidence of an association between age group and preference for organic food. Inspection of contributions shows the under-30 group prefers organic more than expected (28 vs 24), while the over-50 group prefers it less (18 vs 24). Comment on validity: one expected frequency was 12, which is above 5, so the approximation is acceptable, though we should be cautious with borderline cells.

结论：有显著证据表明年龄组与有机食品偏好之间存在关联。观察贡献值发现，30 岁以下组喜欢有机食品的人数多于期望（28 vs 24），而 50 岁以上组则少于期望（18 vs 24）。有效性评论：所有期望频数均大于 5，最低为 12，因此近似是可接受的，但对边界格值应保持谨慎。

10. Writing About Correlation and Regression | 撰写相关与回归分析

When analysing bivariate data, frame the investigation clearly. ‘We examine the relationship between hours of revision (x) and test score (y) for 15 students.’ Compute Pearson’s r: r = S_xy / √(S_xx S_yy). Then test H₀: ρ = 0 using t = r√(n−2) / √(1−r²) with n−2 d.f. Always include the scatterplot description: ‘The scatterplot shows a moderate positive linear association with no obvious outliers.’

分析二元数据时，要清楚地构建研究框架。“我们研究 15 名学生的复习时数 (x) 与测验分数 (y) 之间的关系。”计算 Pearson 相关系数 r：r = S_xy / √(S_xx S_yy)。然后使用 t = r√(n−2) / √(1−r²)，自由度为 n−2，检验 H₀: ρ = 0。务必要包含散点图描述：“散点图显示中等程度的正线性相关，无明显的离群值。”

For regression, present the least squares line: y = a + bx, where b = S_xy / S_xx and a = ȳ − b x̄. Interpret coefficients: ‘For every additional hour of revision, the test score is predicted to increase by 2.4 marks.’ Always comment on the coefficient of determination R²: ‘R² = 0.64, so 64% of the variation in test scores can be explained by revision hours.’

对于回归分析，给出最小二乘直线：y = a + bx，其中 b = S_xy / S_xx，a = ȳ − b x̄。解读系数：“复习时间每增加 1 小时，测验分数预计增加 2.4 分。”务必对决定系数 R² 加以评论：“R² = 0.64，即测验分数的变异中有 64% 可由复习时数解释。”

Avoid extrapolation warnings: ‘Using this model to predict scores for 50 hours of revision is unreliable as it lies far beyond the observed range (x_max = 20).’ Also, note that association does not imply causation.

给出外推警告：“使用此模型预测复习 50 小时的分数不可靠，因为该值远远超出观测范围 (x_max = 20)。”此外，要指明关联不代表因果。

11. Common Pitfalls and How to Avoid Them | 常见失分点及应对策略

Pitfall / 陷阱	Solution / 对策
Not defining parameters, e.g., ‘μ’ without description	Always write: ‘μ is the population mean…’
Using a one-tailed test without justification	State why direction is expected, or default to two-tailed
Confusing p-value with α; e.g., ‘p = 0.03 means H₀ is false’	Say: ‘p < 0.05 provides evidence against H₀'
Omitting degrees of freedom or distribution of test statistic	State: ‘T ~ t(18) under H₀’ before calculations
No contextual conclusion: ‘Reject H₀’ only	Add: ‘Hence there is evidence that the new drug reduces blood pressure…’
Treating p-values as fixed probabilities about parameters	Avoid: ‘Probability that H₀ is true’; use ‘given H₀, probability of observing…’
Not checking assumptions and test validity	Mention sample size, normality, independence, expected frequencies
Poor presentation of calculations (missing steps)	Show formula → substitution → result

By internalising this checklist, you turn potential weaknesses into deliberate strengths, directly addressing the examiner’s mark scheme.

通过内化这份检核清单，你将潜在的弱点转化为有意识的强项，从而直击评卷官的给分方案。

12. Final Tips for Top Marks | 冲刺高分终极建议

Practice writing under timed conditions. Use past papers to craft full essay responses, then compare with mark schemes to see where interpretation marks are awarded. A well-structured answer that communicates statistical reasoning in plain English almost always scores higher than one with perfect arithmetic but disconnected commentary.

在限时条件下练习写作。利用历年真题撰写完整的论文式回答，然后与评分方案对照，查看解读分的授予点。一份结构良好、能用平实英语传达统计推论的答卷，几乎总是比运算完美但评述脱节的答卷得分更高。

Remember the golden thread: Problem → Hypotheses → Model → Calculations → Decision → Interpretation → Limitations. Every sentence should advance this narrative. Vary your vocabulary: instead of repeatedly saying ‘Null hypothesis’, use ‘H₀’, ‘no difference assumption’, ‘baseline model’. This shows fluency.

记住这条黄金主线：问题 → 假设 → 模型 → 计算 → 决策 → 解读 → 局限性。每一句话都应推进这一叙述。措辞要多样化：不要总是重复“零假设”，可使用“H₀”、“无差异假设”、“基准模型”，这显示你的语言娴熟度。

Finally, precision matters: report test statistics to three significant figures, p-values sensibly, and confidence intervals consistently. Never write ‘p = 0.000’; instead ‘p < 0.001'. These small touches signal a mature statistician, exactly what Pre-U examiners value most. Approach every paper as an opportunity to demonstrate not just what you have computed, but what you have understood.

最后，精确性至关重要：检验统计量报告三位有效数字，p 值合理呈现，置信区间前后一致。绝不要写“p = 0.000”，而要写“p < 0.001”。这些细微之处标志着一个成熟的统计学思维者，也正是 Pre-U 考官最为看重的品质。把每份试卷都看作一次机会，去展示你不仅计算了什么，更理解了什么。

Published by TutorHao | Statistics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

Pre-U CAIE Statistics: Paper Writing Framework and Model Answers | Pre-U CAIE 统计：论文写作框架与范文