Replication begins by identifying the entry program and sample, executing the code, and comparing specified outputs. This case uses Does Compulsory School Attendance Affect Schooling and Earnings? to illustrate computational replication with a defined scope.
Start with the replication guide: materials, execution, numerical checks, and reporting
What does this run cover?
Only the education coefficient, standard error and sample size in Table IV columns 1 (OLS) and 2 (2SLS) are checked, with first-stage diagnostics as a supplement. Reference values come from the authors’ public Table IV run log. Other columns, tables and full-paper replication are outside this example.
NEW7080.rar was freshly obtained from the authors’ MIT archive and rerun with independently organized Stata code in a separate directory. The sample is men born in 1920–1929 in the 1970 census, N=247,199. Birth-year fixed effects are retained; quarter of birth and its birth-year interactions supply excluded instruments for 2SLS. Conventional standard errors match the reference specifications.
Does this run match the reference output?
Both columns match at the reference log’s seven-decimal display precision. The downloadable comparison retains this run’s higher-precision values and differences; unprinted reference digits cannot be inferred. Agreement establishes corresponding computational results for these materials and specifications, not validity of causal identification.
What does the first stage add?
This first stage uses 30 excluded instruments, with F approximately 4.599 and partial R² approximately 0.000558. The full log retains weak-instrument diagnostics and critical values. Similar OLS and 2SLS estimates do not establish instrument validity; strength, exclusion restrictions and interpretation require separate assessment.
Download this run’s artifacts
- Complete Stata entry program
- This run’s estimates (CSV) · Reference comparison (CSV)
- Actual log and first-stage diagnostics
- Data access and rerun instructions
The package contains original rerun code, this run’s results, comparison, and instructions. Source data, third-party programs/logs and the paper PDF are not redistributed. Obtain original materials from the official source and follow its access and reuse conditions.
How to rerun from the beginning
- Obtain NEW7080.rar from the authors’ MIT archive, check attributed teaching/research conditions, and extract NEW7080.dta.
- Extract this package into a fresh practice directory, create data/, and place the dataset at data/NEW7080.dta.
- With licensed local Stata, start a fresh practice session, set that directory as the working directory, and run do reproduce.do.
- Check the new log’s completion marker, errors, N=247,199 and the two estimates, then compare them with the supplied reference values.
This run used Stata 19.5 on macOS with version 17 compatibility and no third-party commands. Windows hardware was not tested. clear replaces data in memory and replace overwrites named outputs in the practice directory; save existing work and use a fresh directory and session.
Apply the checking method to your paper
Specify the tables, columns and variables to check, preserve source materials, and inspect the sample, model, standard errors and execution order separately. Investigate differences from actual logs; record unavailable data and uncovered outputs explicitly. Use this example’s code-to-result-to-reference mapping to organize a report appropriate to your research.
Return to the replication guide · Check Stata usage conditions