「気候トークか気候アクションか？企業気候コミットの信頼性の行動的検証」複製パッケージ
Replication package for "Climate Talk or Climate Action? Behavioural Validation of Corporate Climate Commitment Credibility” (原題)
张蓉
🤖 gxceed AI 要約
日本語
企業の気候コミットの信頼性を検証する研究の複製パッケージ。20社・51文書のコーパスを対象に、CREDコードブックによる人手コーディングとStage-1 AIスコアを比較し、AIスコアと裁定済み複合指標の相関（ρ=0.269）を再現可能な形で公開する。二重盲検再裁定プロトコルやSHA-256固定コーパスなど透明性の高い手続きを含む。
English
A replication package for a study validating corporate climate commitment credibility. It releases a frozen 51-document, 20-firm corpus, dual-coder CRED codebooks, blind re-adjudication records, and a consolidated analysis script reproducing all statistics, including the Stage-1 AI score vs adjudicated composite correlation (rho = 0.269). Emphasis is on transparency and reproducibility of qualitative-to-quantitative climate disclosure coding.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
SSBJ・有報での気候関連開示が拡大する中、企業の「言葉」と「行動」の乖離をどう測るかは日本企業の開示信頼性評価に直結する。AIスコアと人手裁定の乖離は、国内でのAI活用型ESG評価の限界を示す示唆となる。
In the global GX context
As ISSB/CSRD and greenwashing scrutiny intensify, methods to test whether corporate climate talk matches action are central to disclosure scholarship. The package's transparent coding and AI-vs-human comparison offer a reproducible template for credibility assessment.
👥 読者別の含意
🔬研究者:Provides a fully reproducible protocol and dataset for validating climate commitment credibility and AI-assisted coding.
🏢実務担当者:Highlights that stated climate commitments may diverge from action, informing disclosure credibility and greenwashing risk management.
🏛政策担当者:Offers a replicable methodology for assessing whether corporate climate disclosures reflect genuine action.
📄 Abstract（原文）
/├── README.md ← this file├── analysis.py ← consolidated reproduction script (all statistics)├── requirements.txt ← Python dependencies├── data/│ ├── firm_level_dataset.csv ← Table 3 grid: 20 firms × (Region, D1–D4, CRED, DV1–DV3, ACTION)│ ├── coding_round1_coderA.csv ← Round-1 independent coding, coder A (D1–D4)│ ├── coding_round1_coderB.csv ← Round-1 independent coding, coder B (D1–D4) [see note 4.3]│ ├── coding_round2_coderA.csv ← Round-2 blind re-coding, coder A (D1, D4)│ ├── coding_round2_coderB.csv ← Round-2 blind re-coding, coder B (D1, D4)│ ├── corpus_manifest_sha256.csv ← frozen 51-document corpus manifest with SHA-256 hashes│ ├── stage1_ai_scores.csv ← Stage-1 AI first-pass scores (audit-trail extract; reproduces ρ = 0.269 vs the adjudicated composite)│ └── F01-F20_firm_code_mapping_key.md ← F-code ↔ firm de-anonymisation key├── protocol/│ ├── CRED_coding_manual_v2.0_reconstructed.md ← codebook v2.0 (Round-1 basis; reconstructed 2026-10-20 from the frozen coding protocol §4, anchors verified against all Round-1 adjudication rule citations)│ ├── CRED_coding_manual_v2.1_frozen_2026-09-15.md ← codebook v2.1 (Round-2 basis)│ ├── CRED_evidence_citation_edge-case_rules_v1.0.md ← evidence citation & edge-case rules│ └── SI_blind_readjudication_protocol_2026-10-13.md ← blind re-adjudication protocol & perimeter├── adjudication/│ ├── CRED_adjudication_full_record_2026-09-15.md ← narrative adjudication record (R1 + R2)│ ├── SI_cell_level_records_blind_pack_2026-10-13.xlsx ← all 37 contested/corrected cells│ └── blind_pack_for_adjudicator_sent_2026-10-20.xlsx ← outbound blind pack (transmittal evidence)├── appendices/│ ├── OnlineAppendix_AC_consolidated_2026-10-25.md ← consolidated Appendices A & C│ ├── OnlineAppendix_A_assembled_2026-10-20.md ← corpus windows & outcome evidence (as filed)│ └── SI_S1_blind_readjudication_2026-10-20.md ← SI Section S1 (re-adjudication results)└── results/ └── analysis_output.txt ← captured output of analysis.py (reference run)pip install -r requirements.txt # numpy, scipy, krippendorffpython3 analysis.py # prints all statistics; rewrites results/analysis_output.txt
🔗 Provenance — このレコードを発見したソース
- scidb https://doi.org/10.57760/sciencedb.017lhfirst seen 2026-11-05 06:39:15
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。