术语表与译注
《因果推断第一课》中文译本
原著：Peng Ding, A First Course in Causal Inference（arXiv:2305.18793v2）
译风约定
- 温情优先：像愿意陪读者慢慢想明白的朋友在说话；少生硬断语，多一点体贴与呼吸感。
- 通俗：宁可多一句人话，也不假装读者已背过定义。
- 准确：术语首次出现时给中英对照；公式、命题、代码尽量忠实。
- 轻想象：允许短喻与轻幽默，但不抢戏，也不牺牲技术含义。
进度
常用术语表（随译随补）
| English | 中文 |
|---|---|
| causal inference | 因果推断 |
| association / correlation | 关联 / 相关 |
| potential outcomes | 潜在结果 |
| Yule–Simpson Paradox | 尤尔–辛普森悖论 |
| confounder / confounding | 混杂变量 / 混杂 |
| risk difference / ratio | 风险差 / 风险比 |
| odds ratio | 优势比 |
| randomized experiment / RCT | 随机化实验 / 随机对照试验 |
| observational study | 观察性研究 |
| propensity score | 倾向得分 |
| instrumental variable | 工具变量 |
| SUTVA | 稳定单元处理值假设 |
| Science Table | 科学表 |
| average causal effect (ACE) | 平均因果效应 |
| treatment assignment mechanism | 处理分配机制 |
| counterfactual | 反事实 |
| no interference / consistency | 无干扰 / 一致性 |
| completely randomized experiment (CRE) | 完全随机化实验 |
| Fisher Randomization Test (FRT) | Fisher 随机化检验 |
| sharp null hypothesis | 尖锐原假设 |
| randomization distribution | 随机化分布 |
| difference-in-means | 均值差 |
| studentized statistic | 学生化统计量 |
| blocking / replication | 区组 / 重复 |
| Neymanian inference | Neyman 推断 / 重复抽样推断 |
| weak null hypothesis | 弱原假设 |
| conservative variance estimator | 保守方差估计 |
| Eicker–Huber–White (EHW) | Eicker–Huber–White 稳健方差 |
| Bernoulli randomized experiment (BRE) | Bernoulli 随机化实验 |
| stratified randomized experiment (SRE) | 分层随机化实验 |
| post-stratification | 事后分层 |
| propensity score (in strata) | 层内倾向得分 |
| randomized block design | 随机区组设计 |
| rerandomization (ReM) | 再随机化（基于 Mahalanobis） |
| regression adjustment | 回归调整 |
| ANCOVA | 协方差分析 |
| Lin’s estimator | Lin 估计量 |
| difference-in-difference / gain score | 双重差分 / 增益分数 |
| matched-pairs experiment (MPE) | 配对实验 |
| Wilcoxon sign-rank | Wilcoxon 符号秩 |
| McNemar’s test | McNemar 检验 |
| strong / weak null | 强 / 弱原假设 |
| studentization | 学生化 |
| design-based / randomization-based inference | 基于设计 / 基于随机化的推断 |
| internal / external validity | 内部效度 / 外部效度 |
| superpopulation | 超总体 |
| nonparametric identification | 非参数可识别 |
| selection bias | 选择偏倚 |
| ignorability / unconfoundedness | 可忽略性 / 无混杂 |
| strong ignorability | 强可忽略性 |
| selection on observables | 基于可观测变量的选择 |
| CATE | 条件平均因果效应 |
| g-formula | g-公式 |
| prima facie causal effect | prima facie 因果效应 |
| outcome regression | 结果回归 |
| inverse propensity score weighting (IPW) | 逆倾向得分加权 |
| Horvitz–Thompson (HT) estimator | Horvitz–Thompson 估计量 |
| Hajek / Hájek estimator | Hajek 估计量 |
| overlap / positivity | 重叠 / 正值性 |
| propensity score stratification | 倾向得分分层 |
| balancing score | 平衡得分 |
| truncation / trimming | 截断 / 修剪 |
| extreme counterfactual | 极端反事实 |
| design stage (observational study) | 观察性研究的设计阶段 |
| doubly robust estimator | 双重稳健估计量 |
| augmented IPW (AIPW) | 增广逆倾向得分加权 |
| working model | 工作模型 |
| double machine learning | 双重机器学习 |
| average causal effect on the treated (\(\tau_{\mathrm{T}}\)) | 处理组上的平均因果效应 |
| average causal effect on the control (\(\tau_{\mathrm{C}}\)) | 对照组上的平均因果效应 |
| overlap population / overlap weights | 重叠总体 / 重叠权重 |
| one-sided ignorability | 单侧可忽略性 |
| weighted least squares (WLS) | 加权最小二乘 |
| Frisch–Waugh–Lovell (FWL) | Frisch–Waugh–Lovell 定理 |
| overlap weight | 重叠权重 |
| predictive / projective estimator | 预测型 / 投影型估计量 |
| matching (observational) | 匹配 |
| matching with / without replacement | 有放回 / 无放回匹配 |
| variable-ratio matching | 可变比例匹配 |
| bias-corrected matching estimator | 偏倚校正匹配估计量 |
| curse of dimensionality | 维数灾难 |
| causal diagram | 因果图 |
| unmeasured confounder | 未测混杂变量 |
| negative outcome / exposure / control | 阴性结局 / 阴性暴露 / 阴性对照 |
| over-adjustment | 过度调整 |
| pretreatment criterion | 预处理准则 |
| M-bias / Z-bias | M 偏倚 / Z 偏倚 |
| collider | 对撞因子 |
| omitted-variable bias | 遗漏变量偏倚 |
| E-value | E-value（证据值） |
| latent ignorability | 潜在可忽略性 |
| sensitivity analysis | 敏感性分析 |
| Cornfield inequality | Cornfield 不等式 |
| Bradford Hill criteria | 希尔因果准则 |
| partial identification | 部分识别 |
| worst-case bounds | 最坏情形界 |
| Manski bounds | Manski 界 |
| sensitivity parameter | 敏感性参数 |
| Rosenbaum sensitivity model | Rosenbaum 敏感性分析模型 |
| worst-case \(p\)-value | 最坏情形 \(p\) 值 |
| \(\Gamma\) (sensitivity parameter) | 敏感性参数 \(\Gamma\) |
| strict overlap | 严格重叠 |
| regression discontinuity (RD) | 断点回归 |
| sharp / fuzzy regression discontinuity | 清晰 / 模糊断点回归 |
| running variable | 驱动变量 |
| local average causal effect at cutoff | 断点处的局部平均因果效应 |
| local complier average causal effect | 局部依从者平均因果效应 |
| bandwidth | 带宽 |
| local randomization | 局部随机化 |
| incumbency advantage | 现任优势 |
| local linear regression | 局部线性回归 |
| McCrary test | McCrary 密度检验 |
| encouragement design | 鼓励设计 |
| noncompliance | 不依从 |
| treatment assigned / received | 分配的处理 / 接受的处理 |
| intention-to-treat (ITT) | 意向性治疗 |
| always taker / complier / defier / never taker | 总是接受者 / 依从者 / 违抗者 / 从不接受者 |
| monotonicity | 单调性 |
| exclusion restriction | 排除限制 |
| complier average causal effect (CACE) | 依从者平均因果效应 |
| local average treatment effect (LATE) | 局部平均处理效应 |
| Wald estimator | Wald 估计量 |
| weak IV | 弱工具变量 |
| Fieller–Anderson–Rubin (FAR) | Fieller–Anderson–Rubin 置信区间 |
| instrumental variable inequalities | 工具变量不等式 |
| as-treated analysis | 按实际接受的处理分析 |
| per-protocol analysis | 符合方案分析 |
| Bloom estimator | Bloom 估计量 |
| natural experiment | 自然实验 |
| ordinary least squares (OLS) | 普通最小二乘 |
| endogenous / exogenous regressor | 内生 / 外生回归元 |
| endogeneity / exogeneity | 内生性 / 外生性 |
| just-identified / over-identified / under-identified | 恰好识别 / 过度识别 / 识别不足 |
| two-stage least squares (TSLS) | 两阶段最小二乘 |
| indirect least squares (ILS) | 间接最小二乘 |
| structural form / reduced form | 结构式 / 约简式 |
| control function | 控制函数 |
| Mendelian randomization | 孟德尔随机化 |
| single-nucleotide polymorphism (SNP) | 单核苷酸多态性 |
| reverse causality | 反向因果 |
| genome-wide association study (GWAS) | 全基因组关联研究 |
| pleiotropy | 多效性 |
| summary statistics | 汇总统计量 |
| fixed-effect estimator | 固定效应估计量 |
| Fisher weighting | Fisher 加权 |
| Egger regression | Egger 回归 |
| InSIDE | 工具强度独立于直接效应 |
| invalid IV | 无效工具变量 |
| principal stratification | 主分层 |
| principal stratum | 主层 |
| principal score | 主层得分 |
| principal ignorability | 主层可忽略性 |
| truncation by death | 因死亡截断 |
| survivor average causal effect (SACE) | 存活者平均因果效应 |
| surrogate endpoint | 替代终点 |
| dissociative / associative effects | 分离效应 / 关联效应 |
| causal necessity / sufficiency | 因果必要性 / 因果充分性 |
| Heckman selection model | Heckman 选择模型 |
| strong monotonicity | 强单调性 |
| mediation analysis | 中介分析 |
| nested potential outcomes | 嵌套潜在结果 |
| natural direct / indirect effect (NDE / NIE) | 自然直接效应 / 自然间接效应 |
| composition assumption | 复合假定 |
| cross-world counterfactual | 跨世界反事实 |
| a priori counterfactual | 先验反事实 |
| sequential ignorability | 序贯可忽略性 |
| mediation formula | 中介公式 |
| Baron–Kenny method | Baron–Kenny 方法 |
| Sobel’s test | Sobel 检验 |
| product / difference method | 乘积法 / 差分法 |
| controlled direct effect (CDE) | 控制直接效应 |
| generalized propensity score | 广义倾向得分 |
| multi-valued treatment | 多值处理 |
| time-varying treatment / confounding | 随时间变化的处理 / 混杂 |
| g-null paradox | g-零悖论 |
| plug-in / recursive estimation | 代入估计 / 递推估计 |
| marginal structural model (MSM) | 边际结构模型 |
| structural nested model (SNM) | 结构嵌套模型 |
| g-methods | g-方法 |
| generalized method of moments (GMM) | 广义矩方法 |
| Pearson / squared multiple correlation | Pearson 相关系数 / 平方复相关系数 |
| tower property / Eve’s Law | 塔性质 / 夏娃定律 |
| delta method | delta 方法 |
| nuisance parameter | 多余参数 |
| unbiasedness / consistency | 无偏性 / 一致性 |
| coverage / over-coverage | 覆盖 / 过度覆盖 |
| type one / type two error | 第一类 / 第二类错误 |
| Wald interval / test | Wald 型区间 / Wald 检验 |
| duality (CI and tests) | 置信集与检验的对偶 |
| Fisher’s exact test | Fisher 精确检验 |
| Behrens–Fisher problem | Behrens–Fisher 问题 |
| Fieller–Creasy problem | Fieller–Creasy 问题 |
| Monte Carlo | Monte Carlo |
| sampling with / without replacement | 有放回抽样 / 无放回抽样 |
| (nonparametric) bootstrap | （非参数）自助法 |
| population / sample OLS | 总体 / 样本普通最小二乘 |
| linear projection | 线性投影 |
| restricted mean model | 受限均值模型 |
| hat matrix / leverage scores | 帽子矩阵 / 杠杆值 |
| weighted least squares (WLS) | 加权最小二乘 |
| link function | 连接函数 |
| maximum likelihood estimate (MLE) | 最大似然估计 |
| case-control study | 病例对照研究 |
| simple random sampling | 简单随机抽样 |
| inclusion indicator | 入样指示 |
| finite population correction | 有限总体校正因子 |
| finite population CLT | 有限总体中心极限定理 |
说明
全书约 490 页、29 章正文 + 附录 A–C。前言至第 6 章已做轻润色（统一口气与术语，未改公式结构）。参考文献不译。
网页阅读：仓库根目录用 Quarto book 生成站点。本地 quarto preview；推到 GitHub 后用 Pages 打开。步骤见根目录 README.md。