0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

因果推論の知識マップ

Posted at

まとめツリー (2025時点)

因果推論の各トピック名が何に該当するものかなどを確認するための、理論・手法・応用・倫理を体系的に整理した知識マップです。

因果推論(Causal Inference)体系 (2025時点)
│
├─ 【1】研究設計(Design)
│    ├─ 実験的設計:RCT / A-Bテスト / クラスタランダム化 / ステップド・ウェッジ(Stepped-wedge)
│    ├─ ターゲットトライアルエミュレーション(Target Trial Emulation)
│    ├─ 準実験的設計:
│    │    ├─ Difference-in-Differences(DiD) / Event study / Dynamic DiD / Staggered adoption
│    │    ├─ Synthetic control / Regression discontinuity(RDD) / Regression kink design(RKD)
│    │    └─ Interrupted time series(ITS)
│    ├─ 観察研究デザイン:
│    │    ├─ マッチング / 傾向スコア / ネガティブコントロール
│    │    ├─ Instrumental variables設計(IV / LATE)
│    │    ├─ Proximal causal design(Proxy / Confounding bridge)
│    │    └─ Transportability design / Selection diagram
│    ├─ サンプルサイズ・検出力分析(Power analysis)
│    ├─ 事前登録・事前分析計画(Pre-registration / Pre-analysis plan)
│    └─ 倫理・実験制約
│
├─ 【2】理論基盤(Foundations)
│    ├─ 因果モデル:
│    │    ├─ Potential Outcomes(Neyman–Rubin)
│    │    ├─ Structural Causal Model(SCM, DAG, Pearl)
│    │    ├─ NPSEM / SWIG(統合的表現)
│    │    └─ 因果階層(Association → Intervention → Counterfactual)
│    ├─ 仮定・前提:
│    │    ├─ Ignorability / SUTVA(無干渉+一貫性)/ Positivity / Consistency
│    │    ├─ Confounding / Collider / M-bias / Selection bias
│    │    └─ Interference(干渉)・Partial interference
│    ├─ 理論的拡張:
│    │    ├─ Transportability / Domain adaptation
│    │    ├─ Mechanistic causation(機構的因果)
│    │    └─ Causal discoveryとの理論接続
│    └─ 哲学的基礎:
│         ├─ 反事実的推論 / 因果vs相関
│         └─ 説明可能性(Mechanistic vs Statistical causation)
│
├─ 【3】識別理論(Identification Theory)
│    ├─ グラフィカル識別:Back-door / Front-door / do-calculus
│    ├─ 操作変数識別:IV条件 / Instrumental sets / Weak IV対応(LIML, Anderson–Rubin, CLRなど)
│    ├─ Proximal識別:Bridge function / Proximal g-formula / Proxy outcome & exposure
│    ├─ 媒介識別:Sequential ignorability / Cross-world independence / Path-specific effects
│    ├─ 非パラメトリック識別:ID / IDC / ID*アルゴリズム
│    └─ 部分識別:
│         ├─ Manski bounds / Monotonicity / Partial identification under violations
│         ├─ Bounds under monotonicity(Imbens-Angrist / Lee bounds)
│         └─ Principal stratification bounds
│
├─ 【4】因果構造学習(Causal Discovery)
│    ├─ 制約ベース:PC / FCI / RFCI
│    ├─ スコアベース:GES / NOTEARS / DAG-GNN / GIES(介入データ対応)
│    ├─ 関数因果モデル:LiNGAM / ANM
│    ├─ 反変動性:Invariant Causal Prediction(ICP)
│    ├─ 潜在交絡付き学習:FCI-Latent / LiNGAM-Latent
│    ├─ Distribution shift下の構造発見
│    ├─ Active learning for causal discovery
│    ├─ 因果表現学習:
│    │    ├─ CausalVAE / Multimodal causal learning
│    │    ├─ IRM(Invariant Risk Minimization)
│    │    └─ Disentangled representation learning
│    └─ 時系列構造探索:
│         ├─ Granger causality / Vector Autoregression(VAR)
│         ├─ PCMCI / VARLiNGAM
│         └─ Non-stationary causal discovery
│
├─ 【5】処置(Treatment Types)
│    ├─ 二値・連続・多値処置
│    ├─ 時間依存処置(Time-varying treatment)
│    ├─ 潜在・未知処置(Latent treatment)
│    └─ 高次元・マルチモーダル処置(画像・テキスト・行動系列など)
│
├─ 【6】欠測データと選択バイアス(Missing Data & Selection)
│    ├─ 欠測メカニズム:MCAR / MAR / MNAR
│    ├─ 欠測データ下の因果推論:
│    │    ├─ Multiple imputation / IPW・IPCW
│    │    ├─ Pattern mixture models / Selection models
│    │    └─ Sensitivity analysis for MNAR
│    ├─ Selection bias:
│    │    ├─ Selection diagram / Transportability
│    │    ├─ Heckman correction / Two-stage models
│    │    └─ Truncation by death / Principal stratification
│    └─ 測定誤差対応:SIMEX / Misclassification adjustment / Validation sub-study
│
├─ 【7】データ構造別の因果推論(Data-Structure-Specific Methods)
│    ├─ パネルデータ:FE/RE / Event study / Staggered adoption / Sun–Abraham / Callaway–Sant'Anna
│    ├─ 反復横断データ:Pseudo-panel / Cohort analysis
│    ├─ 時系列データ:CausalImpact / Interrupted time series / Policy shock analysis
│    ├─ 生存・時間軸データ:RMST / Competing risks / Multi-state / AFT models / Principal stratification(Truncation by death)
│    ├─ ネットワークデータ:Interference / Spillover / Network exposure mapping
│    ├─ 空間データ:Spatial DID / Geographic RDD / Spatial lag・error models
│    └─ 階層・クラスターデータ:Multilevel causal models / Cluster RCT
│
├─ 【8】推定手法(Estimation Methods)
│    ├─ 観測可能交絡調整:
│    │    ├─ 回帰調整 / 傾向スコア / マッチング
│    │    ├─ Doubly robust / TMLE / AIPW
│    │    ├─ 重み診断・Positivity対応(共通サポート / trimming / ESS / Overlap weighting)
│    │    └─ 共変量選択(DAGベース / Collider回避)
│    ├─ 非観測交絡対応:
│    │    ├─ 操作変数法(IV / LATE / JIVE / Weak IV対応)
│    │    ├─ Shift-share IV / Bartik IV / Judge IV
│    │    ├─ Mendelian randomization / Proxy変数法 / Negative control
│    │    └─ Proximal causal inference(Confounding bridge / Proxy-based g-formula)
│    ├─ 時間依存性交絡(g-methods):MSM / g-formula / SNM / Sequential g-estimation
│    ├─ 機械学習系:
│    │    ├─ Causal Forest / Meta-learners(S/T/X/R)/ DML / BART
│    │    ├─ DragonNet / TEDVAE / CausalGAN
│    │    └─ Neural causal models / Deep IV
│    ├─ ベイズ因果推論:Bayesian networks / MCMC / Bayesian TMLE / Hierarchical causal models / Bayesian adaptive trials
│    └─ スケーラブル推論:Federated causal inference / Streaming causal inference / Differential privacy因果
│
├─ 【9】効果推定と異質性(Effect Estimation & Heterogeneity)
│    ├─ 効果の型:ATE / ATT / CATE / ITE / QTE / LATE
│    ├─ 確率的介入:Stochastic interventions / Modified treatment policies / Shifted interventions
│    ├─ 非線形・交互作用:Dose-response / Threshold / Interaction effects
│    ├─ サブグループ効果・HTE・ITR
│    ├─ 動的・個別化効果(Dynamic HTE / DTR)
│    ├─ 媒介・経路効果(NDE / NIE / Path-specific effects, with sequential ignorability)
│    ├─ Spillover効果分解(直接vs間接)
│    ├─ Principal stratification(Compliance / Truncation by death)
│    └─ 可視化:ICE plot / PDP / CATE heatmap / forest plot
│
├─ 【10】政策学習・介入最適化(Policy Learning & Intervention Optimization)
│    ├─ 政策評価:Off-policy evaluation(IPS / DR / MRDR) / Value function estimation
│    ├─ 政策学習:Contextual bandits / Thompson sampling / Q-learning / A-learning / RL for DTR / Causal RL
│    ├─ 介入最適化:Optimal policy / Welfare maximization / Uplift modeling / Fairness-constrained optimization / Active assignment
│    └─ 適応的実験設計:Bayesian adaptive trials / Multi-armed bandits / Sequential design
│
├─ 【11】信頼性・一般化(Robustness & Generalizability)
│    ├─ 感度分析・部分同定:E-value / Rosenbaum bounds / Oster法 / Bayesian sensitivity analysis
│    ├─ 多重比較:FDR / FWER(Bonferroni / Holm / BH) / Pre-specified vs data-driven
│    ├─ 外的妥当性:Transportability / Domain adaptation / Distribution shift
│    └─ 干渉・ネットワーク効果:Partial interference / Spillover robustness / Network effect estimation
│
├─ 【12】実務・診断・再現性(Practice & Diagnostics)
│    ├─ 事前診断:DAG検証 / Common support / Testable implications
│    ├─ 事後診断:バランス検証 / プラシーボテスト / Falsification / Negative control outcomes
│    ├─ 因果判断基準:Bradford Hill基準 / 疫学的因果推論の実務ガイドライン
│    ├─ 不確実性評価:クラスタロバストSE / ブートストラップ / Permutation / ベイズ信頼区間
│    ├─ 自動化:AutoCausal / CausalML pipelines / Automated covariate selection
│    ├─ ソフトウェア:R / Python / Stata / Julia
│    ├─ 再現性:OSF / GitHub / Docker / Quarto
│    ├─ 報告基準:CONSORT / STROBE / RECORD / TRIPOD-CAUSAL / ROBINS-I / GRADE
│    └─ 結果の伝達:可視化(forest / DAG / effect plots) / Stakeholder communication
│
├─ 【13】分野別応用(Domain Applications)
│    ├─ 疫学・公衆衛生:Vaccine / Pharmacoepidemiology / MR / Target trial
│    ├─ 計量経済学:IV / DID / RDD / RKD / Synthetic control / Bunching
│    ├─ マーケティング・ビジネス:CLV / Attribution / Uplift / Large-scale A/B
│    ├─ 教育:Value-added / Peer effects / Policy evaluation
│    ├─ 環境・気候:Pollution / Climate intervention / Environmental policy
│    ├─ 法学・政策:Legal interventions / Judicial reform / Regulatory impact
│    ├─ AI・機械学習:Fairness / Causal XAI / AI alignment / Model evaluation
│    └─ 社会科学:Political science / Criminology / Sociology
│
├─ 【14】倫理・社会的側面(Ethics & Society)
│    ├─ 因果クレームの倫理・誤用防止
│    ├─ 公平性:Demographic parity / Equalized odds / Counterfactual fairness / Individual fairness
│    ├─ 説明可能性:Causal XAI / Counterfactual explanations
│    ├─ プライバシー保護:Differential privacy / Anonymization
│    ├─ フェデレーション倫理:Data sovereignty / Federated fairness
│    └─ 社会応用倫理:医療・経済政策・AI因果責任
│
└─ 【15】因果推論のフロンティア(Emerging Frontiers)
     ├─ LLMと因果推論:Causal reasoning / Counterfactual data generation / Causal extraction from text
     ├─ Neurosymbolic因果推論:Differentiable causal models / Neural-symbolic integration
     ├─ Quantum causality:Quantum causal models / Non-classical causation
     ├─ Causal world models:Model-based RL / Counterfactual planning / Causal imagination
     └─ 新興応用領域:Causal genomics / Systems biology / Causal simulators / Network meta-analysis / Meta-science
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?