Practicum
Every lecture has one practicum with three coordinated parts. The replication lab reproduces a bounded published result from its original package and traces the decisions behind it. The Stata problem set asks you to construct, estimate, diagnose, and interpret the method yourself. The browser lab exposes a misconception through a prediction you commit to before moving a control, and ends by handing an object to Stata.
Each practicum page contains the replication target, the problem set, the lab handoff, and the submission package. The lab project is a self-contained folder with starter do-files, data with provenance, automated checks, and a README; the instructor build is the executed version with its Stata output.
| # | Practicum | Replication target | Lecture | Status |
|---|---|---|---|---|
| P01 | Recover the basic LP estimator | Levels-LP component of Jordà–Taylor’s Example 1 simulation | 1 | in development |
| P02 | Complete the levels / long-differences comparison | Both persistence settings of Example 1 | 2 | in development |
| P03 | Reconstruct an empirical specification | Cumulative shelter-price response, Inoue–Jordà–Kuersteiner Figure 4, August 2024 author archive | 3 | in development |
| P04 | A baseline fiscal multiplier | Ramey–Zubairy linear cumulative multipliers | 4 | in development |
| P05 | Compare coverage procedures | Newey–West versus lag-augmented bands, Inoue–Jordà–Kuersteiner Figure 2 | 5 | in development |
| P06 | Pointwise versus simultaneous bands | Inoue–Jordà–Kuersteiner Figure 3 | 6 | in development |
| P07 | A bounded bias–variance study | Selected Li–Plagborg-Møller–Wolf designs, ported to Stata | 7 | in development |
| P08 | Raw versus restricted responses | Jordà–Taylor Figure 6 unrestricted versus Gaussian-basis | 8 | in development |
| P09 | Fiscal responses across states | Ramey–Zubairy slack versus non-slack multipliers | 9 | in development |
| P10 | Recover causal weights | Kolesár–Plagborg-Møller shock-weight figure | 10 | in development |
| P11 | Reconstruct historical response paths | Normal versus financial recessions, When Credit Bites Back | 11 | in development |
| P12 | Banking deregulation and the labor share | Dube–Girardi–Jordà–Taylor application | 12 | in development |
| P13 | Reproduce and scrutinize a counterfactual | Jordà–Taylor Example 8 | 13 | in development |
| P14 | Independent reproduction audit | A course teaching package built on the Ramey–Zubairy benchmark, then another team’s capstone | 14 | in development |
| ★ | Capstone | A published result, one extension, a peer audit | 4–14 | spec live |
Fixed conventions across all practica.
- Stata 19.5 (StataNow/SE). Core tasks use transparent commands:
regress,newey,ivregress 2sls,gmm. Packaged commands (lpirf,lpdid) are checks, not prerequisites. See the setup page. - One entry point. Each lab project has a
master.dothat rebuilds every output from the documented starting data, with relative paths and a log. - Provenance travels with the data. Every dataset carries its source, version, vintage, and redistribution terms in a
PROVENANCE.md. - Three labels, never confused. An exact replication matches published numbers to a stated tolerance; a statistical reproduction reruns a simulation at reduced replications; a conceptual reproduction rebuilds the design with different inputs. Every practicum says which it is.
- Seeds are recorded. Randomized exercises fix and report their seeds; browser and Stata comparisons share the same input observations, not just the same seed.