AP 统计学推断知识点全总结
点估计 · 区间估计 · 假设检验 · 斜率推断 · 卡方检验
一、核心总览:三类推断的根本区别
| 推断类型 | 核心任务 | 典型关键词 | 输出形式 |
|---|---|---|---|
| 点估计 | 用样本统计量估计一个总体参数值 | estimate, best guess, single value | 单个数值 |
| 区间估计 | 构造置信区间,给出参数的合理范围 | confidence interval, margin of error, plausible | 统计量 ± 误差边界 |
| 假设检验 | 判断总体参数是否显著异于某一值 | convincing evidence, significant, test claim, p-value | 结论 + p值 |
二、点估计(所有场景汇总)
核心规则:样本统计量 = 总体参数的点估计
| 推断对象 | 点估计公式 | 适用场景 |
|---|---|---|
| 单总体比例 | $\hat{p}$ | 调查支持率、合格率、占比 |
| 单总体均值 | $\bar{x}$ | 平均身高、成绩、时长 |
| 两总体比例差 | $\hat{p}_1-\hat{p}_2$ | 两组占比差异 |
| 两总体均值差 | $\bar{x}_1-\bar{x}_2$ | 两组均值差异 |
| 配对均值差 | $\bar{d}$ | 前后测、同体双测、配对数据 |
| 回归斜率 | $b$ | 用 $x$ 预测 $y$ 的线性关系强度 |
真题示例
A sample of 200 students found 76 own a car.
What is the point estimate for the proportion of all students who own a car?
答案:$\hat{p}=76/200=0.38$
三、区间估计(置信区间)全公式
通用结构
1. 单比例置信区间
- 临界值:$z^*$(正态分布)
- 条件:$n\hat{p}\ge10,\ n(1-\hat{p})\ge10$;随机、独立(10%条件)
2. 单均值置信区间($\sigma$ 未知)
- 临界值:$t^*$,自由度 $df=n-1$
- 条件:随机、独立、近似正态($n\ge30$ 或正态总体)
3. 两比例差置信区间
4. 配对均值差置信区间
- 自由度:$df=n-1$
5. 线性回归斜率置信区间
- $SE_b$:斜率标准误(题目直接给出)
- 自由度:$df=n-2$
- 用途:估计总体真实斜率 $\beta_1$
四、假设检验全体系
通用步骤
- 设定 $H_0,\ H_a$
- 验证条件
- 计算检验统计量
- 求 p值
- 下结论
1. 单比例 z 检验
- $H_0:p=p_0$
- 标准误用原假设比例 $p_0$,非 $\hat{p}$
2. 单均值 t 检验
3. 两比例 z 检验(混合比例)
4. 两独立样本 t 检验
5. 配对 t 检验
- 关键词:same day, before/after, matched pairs
6. 回归斜率 t 检验
- 常用 $H_0:\beta_1=0$(无线性关系)
- $H_a:\beta_1\neq0\ /\ >0\ /\ <0$
五、卡方检验($\boldsymbol{\chi^2}$)专项
AP 共3 类卡方检验,统计量公式相同,自由度不同
通用统计量
1. 卡方拟合优度检验
- 用途:检验单分类变量是否符合已知分布
- 自由度:$df=k-1$($k$=类别数)
- 关键词:specified distribution, model fit, expected counts
2. 卡方独立性检验
- 用途:检验两个分类变量是否相关
- 自由度:$df=(r-1)(c-1)$
- 关键词:associated, related, independent, contingency table
3. 卡方正齐性检验
- 用途:检验多个总体的分类分布是否相同
- 自由度:$df=(r-1)(c-1)$
- 关键词:same distribution, homogeneous, multiple groups
卡方共同条件
- 随机样本
- 所有期望频数 $E\ge5$
- 观测独立
六、线性回归斜率推断(专项总结)
1. 点估计
样本斜率:$b$
2. 置信区间
3. 假设检验
- $H_0:\beta_1=0$(无线性相关)
- $H_a:\beta_1\neq0$(有线性相关)
关键词
linear relationship, slope, predict, regression line
七、终极题型判定表(AP 真题一眼区分)
| 题目问法关键词 | 直接判定 | 对应方法 |
|---|---|---|
| estimate, single value | 点估计 | 样本统计量直接作答 |
| confidence interval, margin of error | 区间估计 | 统计量±临界值×标准误 |
| convincing evidence, significant, test claim | 假设检验 | 算统计量+p值+结论 |
| associated, contingency table, distribution fit | 卡方检验 | $\chi^2=\sum\frac{(O-E)^2}{E}$ |
| regression, slope, predict | 斜率推断 | t检验 / t区间 |
| before/after, same subject, matched pairs | 配对 t | 对差值做单样本 t |
| two independent groups, no pairing | 两独立样本 t/z | 双样本公式 |
| proportion, percentage, rate | 比例类 | z分布 |
| mean, average, length/score | 均值类 | t分布 |
八、历年真题典型问法对照
1. 点估计真题
What is the point estimate for the true mean difference?
→ 直接答 $\bar{d}$
2. 区间估计真题
Construct and interpret a 95% confidence interval for the slope of the regression line.
→ 斜率区间:$b\pm t^*SE_b$
3. 假设检验真题
Is there convincing evidence that the proportion is greater than 0.3?
→ 单比例 z 检验,右尾
4. 卡方真题
Is there an association between gender and preference?
→ 卡方独立性检验
5. 斜率推断真题
Is there evidence of a linear relationship between x and y?
→ 斜率 t 检验,$H_0:\beta_1=0$
九、易错点终极提醒
- 比例检验用 $p_0$,区间用 $\hat{p}$
- 均值几乎都用 t 分布,除非总体 $\sigma$ 已知
- 卡方永远是右尾检验,p值为 $P(\chi^2>统计量)$
- 配对≠两独立样本,看到配对直接用差值
- 斜率检验自由度 = n-2,非常容易错
- 区间包含0 ↔ 检验不拒绝H0(双向等价)

