# Angrist 与 Krueger (1991):Table IV 两列复现 对象:Does Compulsory School Attendance Affect Schooling and Earnings? 本包覆盖 Table IV 第1列 OLS、第2列2SLS;第一阶段为补充诊断,不表示整篇论文或全部设定已复现。 ## 数据来源与获取 作者 MIT 存档明确允许署名用于教学或研究: https://economics.mit.edu/people/faculty/josh-angrist/angrist-data-archive 在 Angrist and Krueger (1991) 部分取得 NEW7080.rar,解压 NEW7080.dta。本包不再分发原数据、作者程序或作者原日志,也不包含论文 PDF。访问与使用条件以原来源为准;公开下载不等于可以任意再分发。 ## 运行 1. 将本包解压到新练习目录,并建立 data 子目录。 2. 将从官方来源取得的 NEW7080.dta 放到 data/NEW7080.dta。 3. 使用合法本机 Stata,在新练习会话中将工作目录切换到练习目录,运行 `do reproduce.do`。 4. 核对 replication.log 中的完成标记、样本量和错误;将本次 estimates.csv 与 comparison.csv 的作者日志参考值对照。 需要本机合法 Stata;本次在 macOS 的 Stata 19.5、version 17兼容模式实算。没有第三方命令依赖;不需要模型服务或账号。Windows 真机未测试。 comparison.csv 和 result.json 是本页这次运行的记录。复做后以新 replication.log 和 estimates.csv 为准,手动对照参考值;不能把随包记录当作自己新运行的通过证据。 程序会替换内存数据,并覆写练习目录内 replication.log 和 estimates.csv。请先保存原工作,再使用新会话和新目录。数据文件只读,不改写。 ## 样本与设定 1970年普查,1920–1929年出生男性,247199个观测。v4为教育年限、v9为周工资对数、v16为普查年、v18为出生季度、v27为出生年。控制出生年固定效应;2SLS以出生季度和季度×出生年项作为排除工具。原两列采用常规标准误,不改成稳健或聚类口径。不要用手工两次OLS的标准误代替IV估计器。 comparison.csv 的参考值来自作者 QOB_IV.log,第1个 OLS和第1个2SLS输出;结果按其小数点后7位显示精度核对。相符仅说明指定材料/设定的计算结果一致,不证明识别假设或因果结论有效。 ## 文件 - reproduce.do:自产完整入口。 - replication.log:本次实际执行日志,运行路径已替换为通用练习目录。 - estimates.csv:本次系数、标准误、95%置信区间和样本量。 - comparison.csv:作者参考值、本次结果、差值和显示精度一致性。 - result.json:范围、运行版本、第一阶段返回值与来源摘要。 作者原程序:https://economics.mit.edu/sites/default/files/inline-files/QOB%20Table%20IV.do 作者原日志:https://economics.mit.edu/sites/default/files/inline-files/QOB_IV.log 案例:https://marx-max.aidea-labs.com/cases/angrist-krueger-1991/ # English This package checks only columns 1 (OLS) and 2 (2SLS) of Table IV in Angrist and Krueger (1991), with first-stage diagnostics as a supplement. It is not full-paper replication. Obtain NEW7080.rar from the author's MIT archive and extract NEW7080.dta into data/ under a new practice directory. The archive permits attributed teaching/research use; follow its conditions. Source data, third-party programs/logs and the paper PDF are not redistributed in this package. The supplied comparison.csv and result.json document the website run; check your new log and estimates against the references after rerunning. They do not certify your new run. With licensed local Stata, start a fresh practice session, set the package directory as the working directory, and run `do reproduce.do`. The run replaces data in memory and overwrites replication.log and estimates.csv in the practice directory, so save existing work first. Input data is read-only. This run used Stata 19.5 on macOS with version 17 compatibility; no third-party commands, model service, or account are required. Windows hardware has not been tested. Both specifications use the 1970 census sample of men born 1920–1929, N=247199, with birth-year fixed effects. The outcome is log weekly wage, education is the endogenous regressor, and quarter of birth and its birth-year interactions supply the excluded instruments. Preserve conventional standard errors used in the reference columns. The comparison uses the first OLS and first 2SLS education estimates in the author's public QOB_IV.log, matching seven-decimal display precision. Numerical agreement is not validation of causal identification. The package contains original rerun code, this run's log and estimates, a comparison table, scope/diagnostic metadata, and these instructions.