05 regression interpreter
Skill lianxhcn/PXa2026a/skills/core/05-regression-interpreter
Lecture
npx -y skills add lianxhcn/PXa2026a --skill 05-regression-interpreterAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use this skill when reading, interpreting, or writing up regression results. Given a regression table and variable definitions, it explains each coefficient in plain language with correct direction, magnitude and units; reads log-level, log-log and standardized coefficients correctly (semi-elasticity / elasticity / standard-deviation); checks whether a results write-up overstates correlation as causation and rewrites causal wording ("causes/raises/improves") into comparison/conditional wording ("is associated with / higher / after controlling for"); distinguishes statistical from economic significance; and adds outline (**#) comments to bare regression do-files. Emphasizes comparison-not-effect framing and human review; never fabricates coefficients, data, or results.
SKILL.md
4.8 KB, as published. Nobody here has run it
regression-interpreter(回归结果解读员)
会跑回归之后,最琐碎、也最容易出错的一步是把系数写成人话、且不写过头。这个技能帮你把一张回归表解读成准确的自然语言,核对对数/标准化系数的读法,并守住「相关不等于因果」这道护栏。判断权始终在你手里——AI 起草,你回到变量定义和数据核对。对应第 4 讲「AI 协作」的四项训练。
何时用
- 拿到一张回归表,想快速、准确地把核心系数解释成人话;
- 因变量或自变量取过对数、做过标准化,拿不准系数该按百分比、弹性还是标准差来读;
- 写完结果说明,想检查有没有把「相关」写成「导致 / 提升 / 影响」;
- 分不清「统计显著」与「经济显著」,想说清一个系数在现实里到底多大;
- 有一段光秃秃的回归 do 代码,想加上分层大纲注释、方便课堂导航与复看。
使用方式(提示词)
逐个解释核心系数:
你是一位计量经济学助教。下面是一张回归表和变量定义。请逐个解释【核心系数】:(1) 方向、大小和单位;(2) 若因变量/自变量取了对数,按半弹性/弹性正确解读;(3) 用「比较」而非「因果」的措辞;(4) 每个系数一句话,通俗但准确。 【变量定义】… 【回归表】<粘贴 esttab 输出>
检查相关 vs 因果:
请审查下面这段实证结果表述,找出把【相关关系】写成【因果关系】的地方。对每一处:(1) 指出是哪个词(如"导致""提升""使得");(2) 说明为何超出回归能支持的范围;(3) 给出「比较/条件」措辞的改写版本。不要改动数字与结论强度以外的内容。 【待审段落】<粘贴>
核对对数、标准化的系数读法:
下面是变量字典和我对系数的解读草稿。请核对解读与变量【变换方式】是否一致:因变量为 ln(·)→按百分比/半弹性读;两端都是 ln(·)→按弹性读;变量标准化→按标准差读。指出不一致处并改正。 【变量字典】… 【我的解读草稿】<粘贴>
回归代码加注释版:
请为下面这段 Stata 回归代码加注释:(1) 用
**#/**##多级大纲注释标出每个分析步骤;(2) 关键命令后加一句行内注释,说明在做什么、为什么;(3) 不改动代码逻辑。 【原始代码】<粘贴 do 代码>
输出约定
- 系数解读必须说清方向、大小、单位,并按变换方式(水平/对数/标准化)给正确读法;
- 一律用「比较/条件」措辞(相关、更高、在控制…之后),不下因果结论,除非用户另给识别策略;
- 区分统计显著(星号、$p$ 值)与经济显著(用实际单位说清效应多大);
- 加注释时不得顺手改动设定(控制变量、标准误选项等);
- 一切以用户提供的回归表、变量定义、代码为准,不编造系数、不虚构数据与结论;拿不准处显式标出。
最小使用示例
用 regression-interpreter 帮我解释这张表:因变量 lwage=ln(时薪),自变量 educ=受教育年限,educ 系数 0.109、括号 0.014、三颗星。这个系数怎么读?能说"多读一年书让工资涨 11%"吗?
人工检查清单
- AI 给的每一句解读,回到变量定义与数据处理代码核对——尤其对数系数是否被当成绝对值、百分比有没有搞反;
- 因果措辞审查可能矫枉过正,把本就恰当的表述也标成问题,由你判断;
- 变换方式以你的代码为准,别让 AI 凭变量名猜;
- AI 辅助、不替代判断:能否下因果结论,取决于识别策略而非回归系数本身,最终由你把关。
来源与许可
- 原创:本课程为「连享会 2026 暑期班·初级班」编写,Agent 无关的开放 Skill,MIT(随仓库
LICENSE-CODE)。 - 最后验证日期:<!-- TODO: T7 彩排 Claude Code 与 Codex 双端实测后填写 -->