📚 3. The Presidency: Binomial, Normal & Hypothesis Testing in Election Polls | 3. 总统职位：选举民调中的二项分布、正态分布与假设检验
In A‑Level Mathematics, real‑world contexts help bring statistical distributions to life, and few scenarios are as rich as a presidential election. Polling data, candidate approval ratings, and exit surveys all rely on binomial models, normal approximations, hypothesis testing and confidence intervals. This article walks you through the core Edexcel specification points by exploring how statisticians analyse voter behaviour and make inferences about an entire electorate from a sample.
在 A‑Level 数学中，真实情境能让统计分布变得生动，而总统选举正是极其丰富的场景。民调数据、候选人支持率、出口调查都依赖于二项模型、正态近似、假设检验和置信区间。本文通过探讨统计学家如何分析选民行为并基于样本推断全体选民，带你梳理 Edexcel 考纲的核心知识点。
1. The Role of Statistics in Presidential Elections | 统计在总统选举中的作用
Presidential elections generate enormous quantities of data. Polling organisations randomly select voters and ask who they intend to support. Because each response is either 'support candidate A' or 'not', the number of supporters in a sample follows a binomial distribution, provided the sample is random and independent. Statistical inference then allows analysts to judge whether a candidate truly leads, or if apparent advantages can be explained by sampling variability.
总统选举产生海量数据。民调机构随机选择选民，询问他们支持哪位候选人。由于每个回答要么是“支持候选人 A”要么是“不支持”，只要样本是随机且独立的，样本中支持者的数量就服从二项分布。然后统计推断让分析师能够判断候选人是否真的领先，还是表面优势仅仅源于抽样波动。
2. Binomial Distribution for Voter Support | 选民支持率的二项分布
they will vote for a particular candidate. If the true population proportion is p, then X ~ ers is given by P(X = k) = nCk pᵏ (1-p)ⁿ⁻ᵏ. This model underpins all polling calculations.
令 X 表示容量为 n 的随机样本中声称将投票给某候选人的选民人数。如果真实的总体比例为 p，则 X ~ B(n, p)。观察到恰好 k 名支持者的概率为 P(X = k) = nCk pᵏ (1-p)ⁿ⁻ᵏ。这个模型是所有民调计算的基础。
P(X = k) = ⁿCₖ pᵏ (1 - p)ⁿ⁻ᵏ
The conditions for a binomial distribution are crucial: a fixed number of trials n, each trial independent, only two outcomes (support or not), and constant probability p. In real polling, pollsters use techniques like random digit dialling to approximate independence. Deviations from these conditions, such as non-response bias, must be carefully managed.
二项分布的条件至关重要：固定试验次数 n、每次试验独立、只有两种结果（支持或不支持）、概率 p 恒定。在实际民调中，调查者使用随机数字拨号等技术来近似独立性。与这些条件的偏离，如无回应偏差，必须小心处理。
3. Setting Up Hypotheses for Candidate Preference | 候选人偏好假设的建立
Suppose an election analyst claims that the incumbent president has exactly 50% support. A rival analyst suspects the true proportion is less than 50%. The null hypothesis H₀: p = 0.5 is tested against the alternative H₁: p < 0.5 (one‑tailed) or H₁: p ≠ 0.5 (two‑tailed). The choice depends on the research question; a two‑tailed test is used if we only want to detect any change.
假设一位选举分析师声称现任总统恰好拥有 50% 的支持率。另一位分析师怀疑真实比例低于 50%。零假设 H₀: p = 0.5 相对于备择假设 H₁: p < 0.5（单尾）或 H₁: p ≠ 0.5（双尾）进行检验。选择取决于研究问题；若只想检测任何变化，则使用双尾检验。
H₀: p = 0.5
H₁: p < 0.5 or p ≠ 0.5
The significance level α, often 5%, is the probability of rejecting H₀ when it is actually true. For a one‑tailed test, the critical region is determined by the lower tail of the binomial distribution if we are testing for a decrease. Edexcel exams frequently ask you to find the critical value and state the rejection region in the context of the problem.
显著性水平 α（通常为 5%）是当 H₀ 为真时拒绝它的概率。对于单尾检验，若检验支持率下降，拒绝域由二项分布的下尾确定。Edexcel 考试常要求你找出临界值，并在问题情境中陈述拒绝域。
4. Test Statistic and Critical Regions | 检验统计量与拒绝域
of obtaining a result at least as extreme as the observed number of supporters. If this p‑ n a sample of 100 voters, if 40 support the president and H₁: p < 0.5, we compute P(X ≤ 40 | p = 0.5). If that probability is less than 0.05, the result is significant.
直接使用二项分布，我们计算得到至少与观察到的支持人数同样极端结果的概率。如果这个 p 值低于 α，则拒绝 H₀。例如，在 100 名选民的样本中，若 40 人支持总统且 H₁: p < 0.5，我们计算 P(X ≤ 40 | p = 0.5)。如果该概率小于 0.05，结果就是显著的。
alues can be found from binomial tables or the cumulative distribution function. For a two‑tailed test at 5% In polling language, this would suggest the true proportion is unlikely to be exactly 0.5.
临界值可以通过二项分布表或累积分布函数找到。对于 n = 100、p = 0.5 下 5% 显著性水平的双尾检验，下临界值可能是 c₁ = 40，上临界值 c₂ = 60，即当 X ≤ 40 或 X ≥ 60 时拒绝 H₀。用民调语言来说，这意味着真实比例不太可能恰好是 0.5。
5. Normal Approximation to the Binomial | 二项分布的正态近似
ing, sample sizes of 1000 or more make this approximation very accurate, provided np > 5 and n(1-p) > 5.
当 n 较大且 p 不太接近 0 或 1 时，二项分布 B(n, p) 可用均值为 μ = np、方差为 σ² = np(1-p) 的正态分布近似。在总统民调中，1000 或更大的样本量使这种近似非常精确，前提是 np > 5 且 n(1-p) > 5。
X ~ B(n, p) ≈ N(np, np(1-p))
he normal approximation simplifies the test statistic to z = (X - np) / √(np(1-p)). This z‑ pared against critical values from the standard normal distribution, such as ±1.96 for a 5% two‑tailed test. The approximation allows quick calculation of p‑values and confidence intervals without summing many binomial probabilities.
使用正态近似可将检验统计量简化为 z = (X - np) / √(np(1-p))。将这个 z 值与标准正态分布的临界值（如 5% 双尾检验的 ±1.96）进行比较。该近似可以快速计算 p 值和置信区间，而无需累加大量二项概率。
6. Continuity Correction in Practice | 实践中的连续性校正
ity correction improves accuracy. When calculating P(X ≤ k), we use P(Y < k + 0.5) where Y ~ k schemes typically require the continuity correction when a normal approximation is used.
由于二项分布是离散的而正态是连续的，连续性校正可以提高准确性。当计算 P(X ≤ k) 时，我们使用 P(Y < k + 0.5)，其中 Y ~ N(np, np(1-p))。在我们的民调示例中，P(X ≤ 40) 变为 P(Y < 40.5)。Edexcel 的评分方案通常要求在使用正态近似时采用连续性校正。
To test H₀: p = 0.5 against H₁: p < 0.5 with an observed 420 supporters out of 1000, the z‑statistic becomes z = (420 + 0.5 - 500) / √1000×0.5×0.5 correction factor +0.5 is added because we are approximating P(X ≤ 420) with P(Y < 420.5).
为了检验 H₀: p = 0.5 对 H₁: p < 0.5，观察到 1000 人中有 420 名支持者，z 统计量为 z = (420 + 0.5 - 500) / √1000×0.5×0.5 = -79.5 / 15.811 ≈ -5.03。这远远超出任何临界值，提供了反对 H₀ 的强有力的证据。注意连续性校正因子 +0.5 是因为我们正用 P(Y < 420.5) 近似 P(X ≤ 420)。
7. Confidence Intervals for the True Proportion | 真实比例的置信区间
pollsters often report a confidence interval for the true proportion p. An approximate 95% confidence interval uses the sample proportion p̂ = x / n and the formula p̂ ± z* √(p̂(1-p̂) / n), where z* = 1.96. For a poll of 1000 voters with 520 supporters, p̂ = 0.52, giving 0.52 ± 1.96 × √0.52×0.48 / 1000 ≈ (0.489, 0.551).
除假设检验外，民调机构常常报告真实比例 p 的置信区间。一个近似的 95% 置信区间使用样本比例 p̂ = x / n 和公式 p̂ ± z* √(p̂(1-p̂) / n)，其中 z* = 1.96。对于 1000 名选民中有 520 名支持者的民调，p̂ = 0.52，得到 0.52 ± 1.96 × √0.52×0.48 / 1000 ≈ (0.489, 0.551)。
p̂ ± z* √(p̂(1−p̂) / n)
This interval means we are 95% confident that the true proportion of supporters lies between 48.9% and 55.1%. If a candidate's confidence interval does not overlap 50%, the candidate might indeed be in the lead. In Edexcel exams, you may be asked to interpret such intervals in context and discuss factors like sample size and confidence level.
这个区间意味着我们有 95% 的把握认为真实支持率介于 48.9% 与 55.1% 之间。如果某候选人的置信区间未覆盖 50%，该候选人可能确实领先。在 Edexcel 考试中，你可能需要结合情境解释此类区间，并讨论样本量和置信水平等因素。
8. Determining Required Sample Size | 确定所需样本量
y voters to sample to achieve a desired margin of error. The margin of error is E = z* √(p̂(1-p̂) / n). Solving for n yields n = (z*² × p̂(1-p̂)) / E². Without prior information, the worst‑case scenario is p̂ many national polls aim for around 1000 respondents, yielding a margin of error around ±3%.
在开展民调前，分析师要确定需要抽样多少选民以达到期望的误差幅度。误差幅度为 E = z* √(p̂(1-p̂) / n)。求解 n 得到 n = (z*² × p̂(1-p̂)) / E²。若没有先验信息，最坏情况是 p̂ = 0.5，使乘积最大为 0.25。因此许多全国性民调将受访人数定为约 1000 人，产生的误差幅度约为 ±3%。
n = (z*² × p̂(1-p̂)) / E²
For example, to estimate a president's approval rating within 2 percentage points with 95% confidence, assuming p̂ xam question: calculate the minimum sample size and comment on practical constraints such as cost and time.
例如，要以 95% 置信水平将总统支持率的估计误差控制在 2 个百分点以内，假设 p̂ = 0.5 且 z* = 1.96，所需的样本量为 n = (1.96² × 0.25) / 0.02² ≈ 2401。这是典型的考题：计算最小样本量，并评论成本和时效等实际限制。
9. Type I and Type II Errors in Polls | 民调中的第I类与第II类错误
A Type I error occurs when the poll leads the analyst to reject H₀ (e.g. declare a change in support) when in fact H₀ is true. The probability of this error is exactly α, the significance level. In a presidential context, a Type I error might mean claiming a candidate has fallen below 50% when their support is still solidly at half.
当民调导致分析师在 H₀ 实际为真时拒绝 H₀（例如，声称支持率发生了变化），即发生第 I 类错误。这一错误的概率恰好是 α，即显著性水平。在总统选举情境下，第 I 类错误可能意味着声称某候选人的支持率已跌至 50% 以下，而实际上其支持率仍稳固在 50%。
false null hypothesis. The probability, β, depends on the true value of p. If a candidate's support has genuinely dropped to 45% is the probability of correctly detecting such a drop. Larger sample sizes increase power.
当我们未能拒绝一个错误的零假设，即发生第 II 类错误。其概率 β 取决于真实 p 值。假如某候选人的支持率确实已降至 45%，但我们的检验效能不足，未能拒绝 H₀: p = 0.5，便犯了第 II 类错误。检验效能 = 1 - β 是正确检出此等下降的概率。扩大样本量能提高检验效能。
10. Interpreting Results and Margin of Error | 结果解读与误差边际
When a news outlet reports that 'the president leads 52% to 48% with a margin of error of ±3%', this refers to the 95% requires students to interpret findings in lay terms while maintaining statistical rigour.
当新闻媒体报道“总统以 52% 对 48% 领先，误差幅度为 ±3%”时，这指的是 95% 置信区间。如果区间大幅重叠，该领先在统计上并不显著。分析师必须判断观察到的差异是否有实质意义，而不仅仅是随机抽样波动。Edexcel 要求学生在保持统计严谨性的同时以通俗语言解读结果。
A common mistake is to treat a non‑significant result as proof of 'no difference'. Absence of evidence is not evidence of absence. If a poll of 500 voters shows 51% support, the confidence interval might easily straddle 50%, meaning the president could be above or below half. More data would be needed to make a definitive call.
常见的错误是将不显著的结果视为“没有差异”的证据。没有证据并不等于证据不存在。如果一项 500 名选民的民调显示 51% 的支持率，其置信区间很可能跨越 50%，意味着总统支持率可能高于也可能低于一半。要得出确切结论需要更多数据。
11. Worked Example: A Presidential Approval Poll | 实例解析：总统支持率民调
A poll of 1200 randomly selected likely voters finds 552 approve of the president's performance. Test at the 5% significance level whether the president's approval rating differs from 50%.
一项针对 1200 名随机挑选的可能选民的民调发现，552 人认可总统的表现。在 5% 显著性水平下检验总统支持率是否不等于 50%。
Solution: H₀: p = 0.5, H₁: p ≠ 0.5 (two‑tailed). Sample proportion p̂ = 552/1200 = 0.46. Conditions: np₀ = 600 > 5, n(1-p₀) = 600 > 5, so normal approximation is appropriate. Test statistic: z = (p̂ - p₀) / √(p₀(1-p₀) / n) = (0.46 - 0.5) / √0.25/1200 = -0.04 / 0.01443 ≈ -2.77. al values: ±1.96. Since -2.77 < -1.96, we reject H₀. There is sufficient evidence at the 5% level to conclude the president's approval rating differs from 50%. In fact, the p‑ conclusion. If the question asked for a confidence interval, we could report 0.46 ± 1.96 × √0.46×0.54/1200 ≈ (0.432, 0.488).
解答： H₀: p = 0.5, H₁: p ≠ 0.5（双尾）。样本比例 p̂ = 552/1200 = 0.46。 条件：np₀ = 600 > 5, n(1-p₀) = 600 > 5，因此正态近似适用。 检验统计量：z = (p̂ - p₀) / √(p₀(1-p₀) / n) = (0.46 - 0.5) / √0.25/1200 = -0.04 / 0.01443 ≈ -2.77。 临界值：±1.96。由于 -2.77 < -1.96，我们拒绝 H₀。在 5% 水平下有充分证据表明总统支持率与 50% 不同。事实上，p 值约为 0.0056，强化了这一结论。若问题要求置信区间，我们可以报告 0.46 ± 1.96 × √0.46×0.54/1200 ≈ (0.432, 0.488)。
12. Common Mistakes and Exam Tips | 常见错误与考试技巧
Students often forget the continuity correction when using the normal approximation, or they use an incorrect formula for the standard error. Remember that the standard error under H₀ uses p₀, not p̂, in hypothesis tests, whereas confidence intervals use p̂. Also, always state your conclusion in context: 'There is evidence to suggest the president's true approval rating has changed from 50%.'
学生经常在使用正态近似时忘记连续性校正，或使用错误的标准误差公式。请注意，在假设检验中，零假设下的标准误差使用 p₀ 而非 p̂，而置信区间则使用 p̂。此外，务必在情境中陈述结论：“有证据表明总统的真实支持率已从 50% 发生变化”。
- Check the conditions np > 5 and n(1-p) > 5 before approximating.
- Always define X and p clearly at the start.
- For two‑tailed tests, remember to double the tail probability or compare with α/2.
- Draw a normal curve and shade the rejection region to visualise the problem.
- 在近似前检查 np > 5 和 n(1-p) > 5 的条件。
- 一开始就要清晰定义 X 和 p。
- 对于双尾检验，记得将尾部概率加倍或与 α/2 比较。
- 绘制正态曲线并标出拒绝域以可视化问题。
Another typical exam pitfall is misinterpreting a confidence interval: a 95% interval means that if we repeated the sampling many times, 95% of such intervals would capture p. It does not mean there is a 95% chance that p lies in the specific interval. Being precise with phrasing will secure method marks.
另一个典型考试陷阱是误解置信区间：95% 置信区间意味着如果重复抽样多次，其中 95% 的区间会包含真实的 p，而不是指 p 有 95% 的概率落在该特定区间内。措辞精确才能拿到方法分。
Published by TutorHao | Edexcel A‑Level Maths 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