📖 《把 AI 当助手——经济实证研究的诚信、方法与工具》双语网页版:每章中文部在前,English Edition 在后,内容与最新合订 PDF 一致。
附录 B · S.X.Y 学术诚信检查点速查表（23 个）
表 B.1 S.X.Y 学术诚信检查点速查表（23 个）。
| 编号 | 章 | 检查点内容 |
| S0.0 | 0 | 读前震慑：AI不得署名 / 七种剽窃 / 五类AI不端 |
| S1.1 | 1 | 自学环境正版（不跑盗版Stata） |
| S1.2 | 1 | AI辅助代码不伪造结果（每个系数必须自己跑出） |
| S1.3 | 1 | 结论不复用AI原话（AI腔与学术语言差异） |
| S2.1 | 2 | AI辅助理论构图不虚构理论关系（逐箭头核实文献） |
| S3.1 | 3 | AI推荐文献必用Zotero核实元数据 |
| S3.2 | 3 | 不虚构二次文献（AI幻觉引用是五类AI不端之首） |
| S4.1 | 4 | 假说必有文献支撑——AI辅助设计不虚构理论关系 |
| S5.1 | 5 | 借用量表注明出处与信效度证据（不把借用说成自编） |
| S5.2 | 5 | 抽样方案如实描述（非概率抽样不得冒充概率抽样） |
| S6.1 | 6 | 数据出处与版本留痕 |
| S6.2 | 6 | 爬虫合规三原则（robots.txt/频率控制/版权声明） |
| S7.1 | 7 | 调查伦理：知情同意 + 数据脱敏（涉人研究须过伦理审查） |
| S7.2 | 7 | 问卷数据不得事后修改（原始数据只读，修订留痕） |
| S8.1 | 8 | 数据处理全留痕可复现（do+data+log三件套） |
| S9.1 | 9 | 不p-hacking（规格曲线自证，不剔样凑显著） |
| S9.2 | 9 | 稳健性检验如实报告（主检验+替换+子样本三组） |
| S9.3 | 9 | 可复现三件套（do+data+log，data/raw只读） |
| S9.4 | 9 | AI辅助计量不编造系数（AI只解释不生成数字） |
| S10.1 | 10 | 访谈数据严禁AI生成/虚构（含“用AI补访谈细节”红线） |
| S10.2 | 10 | AI转写编码须披露方法与人机一致性κ≥0.7 |
| S11.1 | 11 | AI使用声明三版齐全且真实（开题/投稿/答辩） |
| S11.2 | 11 | 署名（ICMJE四条）与投稿诚信（不一稿多投/不拆分发表） |
增补的 6 类检查点（来自方法论要求，加粗行即此 6 条）：
表 B.2 增补编号 / 方法论主题 / 为什么新增速查表。
L3|L3|L3
增补编号 & 方法论主题 & 为什么新增
S5.1 & 构念的测量 & 量表开发/借用的学术规范
S5.2 & 抽样方法 & 抽样诚实是调查研究的底线
S7.1 & 研究道德 & 涉人研究的伦理审查要求
S7.2 & 调查研究 & 问卷数据完整性保障
S9.1 & 推断统计 & p-hacking是实证研究最常见不端
S10.2 & 定性分析 & AI辅助编码的质量控制与披露\
English Edition
Appendix B · Quick Reference to the 23 S.X.Y Academic-Integrity Checkpoints
Table B.1 Quick reference to the 23 S.X.Y academic-integrity checkpoints.
| Code | Ch. | Checkpoint content |
| S0.0 | 0 | Pre-reading shock: AI may never be an author / seven kinds of plagiarism / five categories of AI misconduct |
| S1.1 | 1 | Licensed self-study environment (never run pirated Stata) |
| S1.2 | 1 | AI-assisted code must not fabricate results (every coefficient must be produced by your own run) |
| S1.3 | 1 | Conclusions must not reuse AI phrasing (the gap between AI voice and academic language) |
| S2.1 | 2 | AI-assisted theory diagrams must not invent theoretical links (verify every arrow against the literature) |
| S3.1 | 3 | Every AI-recommended reference must be verified in Zotero |
| S3.2 | 3 | No fabricated secondary sources (AI-hallucinated citations are the first of the five AI misconducts) |
| S4.1 | 4 | Every hypothesis must be literature-backed—AI-assisted design must not invent theoretical links |
| S5.1 | 5 | Borrowed scales must cite their source and reliability/validity evidence (never present a borrowed scale as self-developed) |
| S5.2 | 5 | Sampling design described truthfully (non-probability sampling must not masquerade as probability sampling) |
| S6.1 | 6 | Data provenance and version trail |
| S6.2 | 6 | Three crawler-compliance principles (robots.txt / rate limiting / copyright notice) |
| S7.1 | 7 | Survey ethics: informed consent + de-identification (human-subjects research requires ethics review) |
| S7.2 | 7 | Questionnaire data must not be altered post hoc (raw data read-only, revisions logged) |
| S8.1 | 8 | All data processing traceable and reproducible (do + data + log triad) |
| S9.1 | 9 | No p-hacking (the specification curve certifies you; never drop observations to buy significance) |
| S9.2 | 9 | Robustness checks reported faithfully (baseline + alternative variables + subsamples, all three) |
| S9.3 | 9 | Reproducibility triad (do + data + log, with data/raw read-only) |
| S9.4 | 9 | AI-assisted econometrics must not invent coefficients (AI explains, never generates numbers) |
| S10.1 | 10 | Interview data must never be AI-generated or fabricated (including the red line of “using AI to fill in interview details”) |
| S10.2 | 10 | AI transcription and coding must disclose the method and human–machine agreement |
| S11.1 | 11 | All three AI-use declarations complete and truthful (proposal / submission / defence) |
| S11.2 | 11 | Authorship (the four ICMJE criteria) and submission integrity (no multiple submission / no salami slicing) |
The six added checkpoints (drawn from methodological requirements; the bold rows above are exactly these six):
Table B.2 Added codes, methodological theme, and why they were added.
L3|L3|L3
Added code & Methodological theme & Why it was added
S5.1 & Measurement of constructs & Academic norms for scale development and borrowing
S5.2 & Sampling methods & Honest sampling is the baseline of survey research
S7.1 & Research ethics & Ethics review requirements for human-subjects research
S7.2 & Survey research & Safeguarding the integrity of questionnaire data
S9.1 & Inferential statistics & p-hacking is the most common misconduct in empirical research
S10.2 & Qualitative analysis & Quality control and disclosure for AI-assisted coding\