Statistics and Probability – HL Only | 统计与概率（仅限高级水平）

📚 Statistics and Probability – HL Only | 统计与概率（仅限高级水平）

In the IB Mathematics Higher Level course, the Statistics and Probability topic extends far beyond the foundations taught at Standard Level. It introduces sophisticated tools such as probability generating functions, transformations of continuous random variables, unbiased estimators, maximum likelihood estimation, and a deeper treatment of confidence intervals and hypothesis testing. This article covers all the core HL-only content, with detailed bilingual explanations designed to reinforce your conceptual understanding and exam technique.

在IB数学高级课程中，统计与概率专题在标准水平的基础上进行了大幅拓展。它引入了概率生成函数、连续随机变量的变换、无偏估计量、极大似然估计等高级工具，并深入探讨了置信区间和假设检验。本文涵盖所有核心的HL专属内容，以双语详细讲解，旨在强化你的概念理解和考试技巧。

---

1. Probability Generating Functions | 概率生成函数

PGF) is defined for a discrete non-negative integer-valued random variable X as G(t) = E(t^X) = Σ pₓ tˣ, where pₓ = P(X=x). The series converges for |t| y identifies the distribution and provides a compact way to derive moments. For example, G'(1) = E(X) and G''(1) = E(X(X-1)), from which Var(X) follows.

对于取非负整数值的离散随机变量 X，概率生成函数（PGF）定义为 G(t) = E(t^X) = Σ pₓ tˣ，其中 pₓ = P(X=x)。该级数在 |t| ≤ 1 时收敛。PGF 能唯一地确定分布，并为推导矩提供了紧凑的方式。例如，G'(1) = E(X)，G''(1) = E(X(X-1))，由此可求方差 Var(X)。

If X ~ B(n, p), its PGF is (q + pt)ⁿ, where q = 1-p. If X ~ erty is extremely useful in deriving distributions of sums and in solving distributional problems.

若 X ~ B(n, p)，其 PGF 为 (q + pt)ⁿ，其中 q = 1-p。若 X ~ Po(λ)，其 PGF 为 exp(λ(t-1))。独立随机变量之和的 PGF 等于各变量 PGF 的乘积。因此，若 X₁, X₂, ..., Xₖ 独立且 PGF 分别为 Gᵢ(t)，则和 S = Σ Xᵢ 的 PGF 为连乘 Π Gᵢ(t)。这一性质在推导和的分布以及解决分布问题时极为有用。

---

2. Transformations of Continuous Random Variables | 连续随机变量的变换

with pdf fₓ(x) is transformed by a strictly monotonic function Y = g(X), the pdf of Y is f_y(y) = fₓ(g⁻¹(y)) × |d/dy g⁻¹(y)| chnique enables us to find the distribution of a function of a random variable without relying on moment/cumulant generating functions at this level.

当连续随机变量 X 的 pdf 为 fₓ(x)，通过严格单调函数 Y = g(X) 变换后，Y 的 pdf 为 f_y(y) = fₓ(g⁻¹(y)) × |d/dy g⁻¹(y)|。对于非单调变换，必须将 X 的取值范围划分为若干单调区间，然后将各区间贡献相加。这一技巧使我们能求出随机变量函数的分布，而无需依赖 IB 课程中不涉及的矩母函数或累积量生成函数。

A common application is finding the distribution of Y = X² when X ~ chi-squared distribution with 1 degree of freedom. Another classic is Y = aX + b, where f_y(y) = (1/|a| ys check the support of Y and ensure the transformed pdf integrates to 1 over its support.

常见的应用是当 X ~ N(0,1) 时求 Y = X² 的分布，所得为自由度为 1 的卡方分布。另一个经典变换是 Y = aX + b，此时 f_y(y) = (1/|a|) fₓ((y-b)/a)。务必检查 Y 的支撑集，并确保变换后的 pdf 在其支撑集上积分为 1。

---

3. Unbiased Estimators and Their Variances | 无偏估计量及其方差

An estimator θ̂ = u(X₁,...,Xₙ) for a parameter θ is unbiased if E(θ̂) = θ for all θ. The sample mean x̄ biased estimator of the population mean μ. For variance, the sample variance s² = Σ(xᵢ - x̄ its variance; among unbiased estimators, the one with the smallest variance is preferred.

参数 θ 的估计量 θ̂ = u(X₁,...,Xₙ) 若满足 E(θ̂) = θ 对所有 θ 成立，则称为无偏的。样本均值 x̄ 是总体均值 μ 的无偏估计量。对于方差，样本方差 s² = Σ(xᵢ - x̄)²/(n-1) 是 σ² 的无偏估计量，而除以 n 的版本是有偏的。估计量的有效性可通过其方差来判断；在无偏估计量中，方差最小者更优。

estimators. For instance, if observations Xᵢ are independent with variance σ², then Var(x̄ riance might have a lower mean squared error (MSE) than an unbiased one, since MSE = Var(θ̂) + [Bias(θ̂)]².

在 IB HL 中，你必须会验证无偏性并计算估计量的方差。例如，若观测值 Xᵢ 相互独立且方差为 σ²，则 Var(x̄) = σ²/n。还需比较估计量；一个有偏但方差极小的估计量，其均方误差 (MSE) 可能低于无偏估计量，因为 MSE = Var(θ̂) + [Bias(θ̂)]²。

---

4. Maximum Likelihood Estimation | 极大似然估计

function of θ. For independent observations, it factorizes as Π f(xᵢ; θ). To find the maximum likelihood estimator (MLE), we usually maximize the log-likelihood ℓ(θ) = ln L(θ) by solving dℓ/dθ = 0 and checking the second derivative.

极大似然方法旨在寻找使观测到给定样本的概率最大的参数值。似然函数 L(θ) = f(x₁,...,xₙ; θ) 被视作 θ 的函数。对于独立观测，它可分解为 Π f(xᵢ; θ)。为求极大似然估计量 (MLE)，我们通常最大化对数似然 ℓ(θ) = ln L(θ)，通过求解 dℓ/dθ = 0 并检查二阶导数来获得。

For a Poisson distribution with unknown λ, the MLE is simply the sample mean λ̂ = x̄. For a normal distribution N(μ, σ²), the MLE of μ is x̄, and the MLE of σ² is Σ(xᵢ - x̄ ber that regularity conditions must hold for the MLE to be asymptotically unbiased and efficient.

对于未知参数 λ 的泊松分布，MLE 就是样本均值 λ̂ = x̄。对于正态分布 N(μ, σ²)，μ 的 MLE 为 x̄，σ² 的 MLE 为 Σ(xᵢ - x̄)²/n（有偏，但常用作点估计）。要准备好推导二项分布、指数分布和几何分布的 MLE。同时需注意，只有满足正则条件时，MLE 才具有渐近无偏性和有效性。

---

5. Central Limit Theorem | 中心极限定理

ndom sample X₁,...,Xₙ from any distribution with mean μ and variance σ², the sample mean x̄ 30). This theorem underpins the vast majority of statistical inference, allowing us to construct confidence intervals and perform hypothesis tests even when the population distribution is unknown.

中心极限定理（CLT）指出，对于来自均值为 μ、方差为 σ² 的任意分布的随机样本 X₁,...,Xₙ，当 n 充分大（通常 n ≥ 30）时，样本均值 x̄ 近似服从 N(μ, σ²/n)。这一定理为绝大多数统计推断提供了基础，使我们即使在总体分布未知的情况下也能构建置信区间并进行假设检验。

The standardized version (x̄ - μ) / (σ/√n (0,1). When σ is unknown and estimated by the sample standard deviation s, for large n, (x̄ - μ) / (s/√n al population, we use the t-distribution. The CLT also extends to sums: Σ Xᵢ ≈ N(nμ, nσ²).

其标准化形式为 (x̄ - μ) / (σ/√n) ≈ N(0,1)。当 σ 未知且用样本标准差 s 估计时，对于大样本，(x̄ - μ) / (s/√n) 仍近似正态。然而，对于来自正态总体的小样本，我们使用 t 分布。CLT 同样适用于总和：Σ Xᵢ ≈ N(nμ, nσ²)。

---

6. Confidence Intervals for a Population Mean | 总体均值的置信区间

he population parameter, based on the sample. For a normal population with known σ, the 95% CI for μ is x̄ ± z* × σ/√n known and the sample size n < 30, we use the t-distribution with n-1 degrees of freedom: x̄ ± t* × s/√n, where t* is the critical value from the t-table.

置信区间基于样本给出了总体参数的可能取值范围。对于已知 σ 的正态总体，μ 的 95% 置信区间为 x̄ ± z* × σ/√n，其中 z* = 1.96。当 σ 未知且样本量 n < 30 时，我们使用自由度为 n-1 的 t 分布：x̄ ± t* × s/√n，其中 t* 为 t 分布的临界值。

The interpretation is crucial: "We are 95% confident that the interval contains the true μ" means that if we repeated sampling many times, 95% of constructed intervals would capture μ. It does not mean there is a 95% probability that μ lies in the interval; that is a Bayesian interpretation not used in IB. The width of the interval increases with confidence level and decreases with larger n.

解释至关重要：“我们有 95% 的把握认为该区间包含真实的 μ”，这意味着如果我们多次重复抽样，所构造的区间中有 95% 会捕获 μ。这并非表示 μ 落入该区间的概率为 95%；那是贝叶斯解释，IB 课程中不采用。区间宽度随置信水平提高而增加，随样本量 n 增大而减小。

---

广告 / Sponsored

7. Hypothesis Tests for a Single Mean | 单个均值的假设检验

nst a null hypothesis H₀. For a test about a population mean, the test statistic is Z = (x̄ - μ₀) / (σ/√n) if σ is known, or T = (x̄ - μ₀) / (s/√n) with n-1 df if σ is unknown. The p-value is the probability of obtaining a test statistic at least as extreme as the observed one, assuming H₀ is true. If p < significance level α, we reject H₀.

假设检验用于评估样本数据是否提供了足够的证据来拒绝原假设 H₀。对于关于总体均值的检验，若 σ 已知，检验统计量为 Z = (x̄ - μ₀) / (σ/√n)；若 σ 未知，则使用自由度为 n-1 的 T = (x̄ - μ₀) / (s/√n)。p 值是在 H₀ 为真的条件下，获得至少与观测值同样极端的检验统计量的概率。若 p < 显著性水平 α，则拒绝 H₀。

d test at α=0.05 with normal data and known σ, the critical values are ±1.96; reject H₀ if |Z| arly: H₀: μ = μ₀, H₁: μ ≠ μ₀ (or >, <). Link the conclusion to the context of the problem.

单尾检验检查某个方向的偏离；双尾检验检查两个方向。对于正态数据、已知 σ 且 α=0.05 的双尾检验，临界值为 ±1.96；若 |Z| > 1.96，则拒绝 H₀。务必清晰陈述假设：H₀: μ = μ₀, H₁: μ ≠ μ₀（或 >, <）。将结论与问题背景联系起来。

---

8. Hypothesis Tests for the Difference of Two Means | 两均值差的假设检验

ndent samples of sizes n₁ and n₂ from normal populations with known variances, use Z = ((x̄₁ - x̄₂) - 0) / √σ₁²/n₁ + σ₂²/n₂. If variances are unknown but assumed equal, use the pooled two-sample t-test with s_p² = ((n₁-1)s₁² + (n₂-1)s₂²) / (n₁+n₂-2) and degrees of freedom n₁+n₂-2.

当比较两个独立总体时，通常检验 H₀: μ₁ = μ₂，对立假设可为单侧或双侧。对于来自方差已知的正态总体、样本量分别为 n₁ 和 n₂ 的独立样本，使用 Z = ((x̄₁ - x̄₂) - 0) / √σ₁²/n₁ + σ₂²/n₂。若方差未知但假设相等，则使用合并双样本 t 检验，合并方差 s_p² = ((n₁-1)s₁² + (n₂-1)s₂²) / (n₁+n₂-2)，自由度为 n₁+n₂-2。

If variances are not assumed equal, Welch's t-test is more appropriate and is allowed within IB HL. The test statistic is t = (x̄₁ - x̄₂) / √s₁²/n₁ + s₂²/n₂ with degrees of freedom approximated by the Welch–Satterthwaite formula. When using technology, the output automatically applies this correction. Always check assumptions: normality and independence.

若方差不假定相等，Welch t 检验更为合适，且 IB HL 允许使用。检验统计量为 t = (x̄₁ - x̄₂) / √s₁²/n₁ + s₂²/n₂，自由度由 Welch–Satterthwaite 公式近似。使用技术计算时，输出会自动应用这一校正。务必检查假设：正态性和独立性。

---

9. Paired t-test | 配对样本 t 检验

₀: μ_d = 0. Assuming the differences are normally distributed, the test statistic is t = d̄ / (s_d/√n) with n-1 degrees of freedom, where d̄ is the mean difference and s_d is the sample standard deviation of the differences.

配对 t 检验适用于对同一实验单元进行两次测量，或存在自然配对的情况。检验针对差值 dᵢ = X₁ᵢ - X₂ᵢ 进行。原假设为 H₀: μ_d = 0。假设差值服从正态分布，检验统计量为 t = d̄ / (s_d/√n)，自由度为 n-1，其中 d̄ 为差值均值，s_d 为差值的样本标准差。

This design reduces variability caused by unit-to-unit differences, often resulting in a more powerful test than the independent two-sample test. Examples include pre-test/post-test scores, or left-hand/right-hand gripping strength. The paired test should not be confused with a two-sample test; using the wrong test can lead to incorrect conclusions.

该设计减少了由个体间差异引起的变异性，通常比独立双样本检验具有更高的功效。例子包括前测/后测成绩，或者左手/右手握力。切勿将配对检验与双样本检验混淆；使用错误的检验可能导致错误结论。

---

10. Hypothesis Tests for Proportions | 比例的假设检验

For a single proportion p, the test statistic is Z = (p̂ - p₀) / √(p₀(1-p₀)/n), where p̂ pproximation requires np₀ ≥ 5 and n(1-p₀) ≥ 5. Confidence intervals for a proportion are p̂ ± z* √(p̂(1-p̂)/n), though these can be inaccurate for small n or extreme p̂ ods like the Wilson interval, but IB mainly expects the Wald-based method with conditions.

对于单个比例 p，检验统计量为 Z = (p̂ - p₀) / √(p₀(1-p₀)/n)，其中 p̂ 为样本比例。正态近似要求 np₀ ≥ 5 且 n(1-p₀) ≥ 5。比例的置信区间为 p̂ ± z* √(p̂(1-p̂)/n)，尽管对于小样本或极端 p̂ 这可能不准确；此时技术手段可能采用 Wilson 区间等方法，但 IB 主要考察基于 Wald 方法并满足条件的情况。

For comparing two proportions, the pooled estimate p̂_pooled = (x₁+x₂)/(n₁+n₂) is used under H₀: p₁ = p₂. The test statistic is Z = (p̂₁ - p̂₂) / √(p̂_pooled(1-p̂ al approximation, and remember that the null hypothesis specifies equality of proportions.

在比较两个比例时，原假设 H₀: p₁ = p₂ 下使用合并估计 p̂_pooled = (x₁+x₂)/(n₁+n₂)。检验统计量为 Z = (p̂₁ - p̂₂) / √(p̂_pooled(1-p̂_pooled)(1/n₁+1/n₂))。务必验证正态近似条件，并记住原假设指定比例相等。

---

11. Type I and II Errors and Power | 第一类/第二类错误与检验功效

A Type I error occurs when H₀ is true but we reject it; its probability is α (the significance level). A Type II error happens when H₀ is false but we fail to reject it; its probability is β. The power of a test is 1 - β, the probability of correctly rejecting a false null hypothesis. Power increases with larger sample size, larger effect size, and higher α.

当 H₀ 为真却被拒绝时，发生第一类错误；其概率为 α（显著性水平）。当 H₀ 为假却未被拒绝时，发生第二类错误；其概率为 β。检验的功效为 1 - β，即正确拒绝错误原假设的概率。功效随样本量增大、效应量增大以及 α 提高而增加。

for a mean with known variance. For a specific alternative μ = μ₁, β = P(do not reject H₀ | tribution. Understanding power is crucial for designing studies with adequate sample size.

IB HL 要求计算简单情形下的第二类错误概率和功效，例如已知方差的单尾 Z 检验。对于特定备择值 μ = μ₁，β = P(未拒绝 H₀ | μ = μ₁)。这需要计算在真实的备择分布下，检验统计量落入接受域的概率。理解功效对于设计具有足够样本量的研究至关重要。

---

12. Summary and Interconnections | 总结与各概念之间的联系

The HL-only topics in Statistics and Probability are deeply interconnected. The Central Limit Theorem justifies the normal approximations used in hypothesis tests and confidence intervals. Maximum likelihood estimators often coincide with sample means and proportions that we test. Probability generating functions give precise distributions for discrete sums, while transformation techniques handle continuous cases. Unbiasedness and variance of estimators link estimation theory to the efficiency of tests. Mastering these connections will not only boost your IB exam performance but also lay solid foundations for university-level statistics.

统计与概率的 HL 专属内容之间有着深刻的联系。中心极限定理为假设检验和置信区间中所用的正态近似提供了依据。极大似然估计量常与我们检验的样本均值和比例一致。概率生成函数给出了离散求和的精确分布，而变换技术则处理连续情形。估计量的无偏性和方差将估计理论与检验的有效性联系起来。掌握这些联系不仅能提升你在 IB 考试中的表现，还能为你学习大学统计课程奠定扎实基础。

Published by TutorHao | Statistics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）