13-01 · Ask, check and communicate

Start here

Four groups of three. Bring your Week 12 main figure, results paragraph and current report, or bring your analysis plan and a labelled descriptive figure. No additional preparation is required. This class rehearses research consultation, reproducibility and explanation; all researcher messages are classroom simulations.

Open the instructor-supplied Posit Cloud project for 13-consulting-reproducibility. All files are already provided in its project root. Foundations slides Workshop slides

Case cards · Information-request worksheet · Relay checking guide · Peer-review sheet · Rehearsal sheet

Opening recall — 12:30–12:45

Show your figure to a neighbour. Explain the question, which group is subtracted from which, units, time and intended population. Locate the treatment allocation or sampling units, rather than counting rows. Identify one useful estimate and one limitation. If the report is incomplete, work from your own descriptive figure and analysis plan, stating what remains provisional.

Essential teaching — 12:45–13:05

Learn to ask a few questions that change the recommendation: what is measured and over which period; whether researchers assign diets to individual animals or whole pens, and how they choose the assignments; which observations share units or environments; which comparison was planned; what happened to missing measurements or removals; what effect would be useful. Distinguish documented facts, assumptions and unknowns. Recommendations should follow from the question and design before you choose model syntax.

Mini consulting clinic — 13:05–13:40

Use the case cards. Each group has a consultant, researcher and auditor. Consultants initially read only the opening message; researchers read their matching briefing and supply information requested. Spend eight minutes on the first case; rotate roles for an eight-minute second case; spend ten minutes writing a maximum 120-word recommendation, then share the question that changed your advice. Do not actually contact the named paper authors or any researcher.

Include an outcome, model justified by grouping/allocation, useful contrast, diagnostic, limitation and one next information request. Unknown metadata stay unknown. The cards describe distinct hypothetical practice studies, not the assessed cases or findings from the background publications. They supply requested design facts, not an analysis formula.

Break 13:40–13:55.

Reproducible-analysis relay — 13:55–14:35

In the supplied Posit Cloud project, open week-13.Rproj; make a working copy of the relay folder. From that folder, inspect report-to-audit.qmd before attempting quarto render report-to-audit.qmd --to html. The base-R report and small invented dataset deliberately contain issues to audit, including an input-path failure. Preserve originals and record each repair. This forage-drying practice study is separate from the assessed animal projects.

Use four ten-minute blocks: execution/provenance check; design/outcome/contrast audit; tracing claims and diagnostic evidence; repair or proposed repair with verification. Check whether the model, contrast, units, replication and conclusions agree. A successful render alone does not establish a defensible analysis.

If computing is unavailable, use circulated draft, source report and data. Audit source and reasoning; explicitly mark numerical reproduction as unverified.

Follow-up design review — 14:35–14:55

Exchange actual project materials with the next group: calf → broiler → pigs → horses → calf. Inspect for twelve minutes and discuss feedback for eight. Use the peer-review sheet to audit the saved analysis plan, revisions, animals or groups that received separate treatment assignments, or were selected separately for observation, reasons for the chosen outcome distribution and model, main comparison, diagnostic decision and justified alternative. Ask what evidence would change the recommendation; reward defensible reasoning rather than matching a prescribed formula. Check the proposed follow-up’s useful effect, sample sizes, two prospective assumptions and simulation simplifications. Return two concrete comments with locations, consequences and feasible actions.

The project’s launch page and case packs govern the design. Do not impose the relay or calf model on every outcome. Temperature associations in the horse case require an association interpretation; the future shelter intervention is a different question.

Break 14:55–15:10.

Repair and rehearse — 15:10–15:45

Rotate to analyst, communicator and auditor. Spend fifteen minutes repairing one consequential issue or documenting a blocker; twelve minutes rehearsing a three-minute account and answering questions; eight minutes writing remaining actions. Use the rehearsal handout for timings. Ensure everyone can explain the design and estimate.

Remaining actions and exit — 15:45–16:00

Each group names one repair and one remaining action with an owner. Individually explain why a report can render cleanly while supporting the wrong scientific conclusion. Complete presentation preparation in 1–2 hours total; presentations are on Wednesday 2 December, 12:30–16:00.