← Back to AI Math Mind Map
📐 AI MathID: hypothesis-testing

Hypothesis Testing

假设检验
🎯Core Definition
Hypothesis testing makes a binary decision between the null H0H_0 and the alternative H1H_1: reject H0H_0 when the test statistic falls in the rejection region. Two error types: Type I (falsely rejecting H0H_0) has probability α\alpha, Type II (missing a real effect) has probability β\beta, and power =1−β= 1-\beta; the p-value is the probability of observing the current or more extreme result under H0H_0 — not the probability that H0H_0 is true.
💡Use Cases
significance in A/B testing, feature screening, paper claims; interviews typically probe the α\alpha/β\beta trade-off, p-value misconceptions, and Bonferroni vs FDR-BH corrections under multiple testing.
⚡Key Problems Solved
running kk independent tests inflates false positives — the family-wise error rate approaches 1−(1−α)k≈kα1-(1-\alpha)^k \approx k\alpha; Bonferroni lowers each threshold to α′=α/k\alpha' = \alpha/k, strictly bounding the family-wise error at α\alpha but losing power;FDR-BH instead controls the false discovery rate FDR=E[V/R]\text{FDR} = \mathbb{E}[V/R] (the fraction of false rejections), which keeps far more power when kk is in the thousands, e.g. genomics or feature screening.
🎯Core Interview Exam Points
Fundamentals · Production Scenarios · Deep Dive
Core Def
Explain the core definition, underlying mechanism/math, and key principles of Hypothesis Testing.
Use Cases
In production engineering, what are the primary use cases for Hypothesis Testing and what core pain points does it solve?
🔥 High-Frequency Deep-Dive Questions
1
Define Type I error α\alpha, Type II error β\beta and power 1−β1-\beta. Why does lowering α\alpha raise β\beta?
2
Precise definition of a p-value? Why is it not the probability that H0H_0 is true?
3
How does the family-wise error accumulate over kk tests? Principle and cost of Bonferroni α′=α/k\alpha'=\alpha/k?
4
FWER vs FDR? The concrete steps of the Benjamini–Hochberg procedure?
5
How do α\alpha, effect size and sample size affect power? How to do power analysis / sample-size planning?
Updated 2026-08-12
🎯
Verify Mastery: Practice Exam Questions for "Hypothesis Testing"
Take pitfall multiple-choice questions with derivation options; mistakes are automatically logged.
🚀 Start Card Practice ➔
← Previous CardPoint EstimationNext Card →Confidence Intervals

🔗 More AI Math Knowledge Cards

引用出处: TalentMe 450+ 工业级 AI 架构考点知识库 (Tech Vault) · 遵循 CC-BY-4.0 知识共享协议
URL: https://talentme.airsota.com/resources/tech/roadmap/mindmap/Math/card/hypothesis-testing
中文速览与面试要点

假设检验 (Hypothesis Testing)

定义: 假设检验 (Hypothesis Testing) 在零假设 H_0 与备择假设 H_1 间做二选一决策: 当检验统计量落入拒绝域时拒绝 H_0。两类错误: 第一类错误 (误拒 H_0) 概率为 , 第二类错误 (漏报真效应) 概率为 , 功效 = 1- ;p 值 = 在 H_0 为真时观察到当前或更极端结果的概率, 不是 H_0 为真的概率。 使用场景: A/B 测试显著性判断、特征筛选、论文假设验证;面试常考两类错误的权衡 (降低 会升高 )、p 值误区, 以及多重比较下的 Bonferroni 与 FDR-BH 校正。 解决的问题: 同时做 k 个独立检验时假阳性累积, 总体第一类错误率约为 1-(1- )^k k ;Bonferroni 校正把每个检验的阈值降为 ' = /k, 严格保证总体 (family-wise) 错误率 , 但功效损失大;FDR-BH 改为控制错误发现率 FDR = E[V/R] (拒绝中假阳性占比), 在 k 达数千的基因/特征筛选中显著提升功效, 是大规模多重比较的主流选择。

核心要点

  • 第一类错误 、第二类错误 与检验功效 1- 的定义?降低 为什么增大 ?
  • p 值的精确定义是什么?为什么不能说 p 值是 H_0 为真的概率?
  • 做 k 次检验时总体假阳性率如何累积?Bonferroni 校正 ' = /k 的原理与代价?
  • FWER 与 FDR 的区别?Benjamini–Hochberg 流程的具体步骤?
  • 显著性水平、效应量、样本量如何影响功效?如何做功效分析与样本量规划?