Day 2 – The workshop
Day 2 is the workshop. It runs in three parts across the four sessions, because the day before has already covered the credibility argument and the toolchain, and the graph work in the first part is not to be hurried. The five numbered tasks and the two submissions to the class chart are the same as on the one-day course.
| Work through | What you do | Timing |
|---|---|---|
| Slides, day 2, Part 1 – The claim and its estimand | The paper in full; two studies set beside our five analysts, and the four estimands in the red-card study, which show that its teams were not answering the same question; Lundberg’s three steps; the causal-graph crash course; then our own claim, where everyone writes a one-sentence estimand and holds it until the whole class sends at once. | Session 1, 90 minutes |
R/dag_starter.R in the workspace, or Browser lab Task 2; then slides, day 2, Part 2 – Reproduction and exploration, and your working script or Task 3 |
Finish the graph: take the # off the arrows you would defend, re-run adjustmentSets(), write two sentences on what fixed effects adjust for. Then the three Multi100 instructions verbatim and what they assume. Fit mcosmo ~ unemp_c with country and year fixed effects and reproduce t = −3.804, the value recorded by Multi100 for analyst C6HJR, then type your t without sending it, and send it when the facilitator says go. |
Session 2, 90 minutes |
| Specification menu, Task 4 in the workspace or on the Browser lab, then Multiverse | Why everyone in the class obtained the same t while five analysts did not; the axes, the three levels, what Task 2 of Multi100 left open, the claim-alignment convention. Then try_spec() three or four times, my_experiments(), and submit_exploration(), which puts your dot on the chart as an open circle. Read the class chart against the five analysts. |
Session 3, 90 minutes |
Slides, day 2, Part 3 – Multiverse and preregistration, with try_spec() on the Browser lab page for your percentile, and Task 5 |
Bet on which choice moves estimates most, then read the decomposition, the placement of the analysts, the sign flips by estimator and sample, and your own percentile in the grid of 840. The critics, what preregistration settles, and then Task 5: complete the preregistration block before you run anything, translate it into syntax, run it, and submit with report_result(). |
Session 4, 90 minutes |
By the end of the day you have reproduced t = −3.804, submitted one exploratory dot and one declared dot to the class chart, and written the report that records them.