13-02 · Consulting and reproducibility recap

Begin with what researchers did. In the lamb study, researchers assign supplements to whole paddocks, then weigh six lambs in each. Counting more lambs or repeat weights does not add paddocks assigned to supplements. In the cow study, researchers assign an order of supplements to each cow and measure that cow repeatedly. Explain those actions before choosing a model.

Define the outcome before comparing groups. A count of mastitis episodes can include several episodes in the same cow. It is not a count of cows affected. Episode rates also need the amount of time the cows were observed. State exactly what is measured, the units, period and animals or groups the answer concerns.

Trace every claim. The data description, calculation, model, comparison, interval, figure and paragraph must answer the same question. State which group is subtracted from which. Synthetic teaching observations are not published research findings.

A report that runs can still be wrong. A successful render can preserve a wrong outcome, omitted source-group difference, reversed comparison or unsupported causal claim. Check both the calculation and its scientific interpretation. State what you could not verify.

Checks should affect decisions. A model check looks for evidence against an assumption. A sensitivity comparison changes one plausible assumption and asks whether the answer changes. Explain what you checked, why it matters and whether it changes your trust in the result. Do not try alternatives merely to find a favourable answer.

Plan a new study from a useful difference. State what change would matter, how researchers would assign treatments, three feasible study sizes and two plausible sets of assumptions. Use simglm for the simulation. Explain failed fits and the uncertainty caused by using only a limited number of simulated experiments. More simulations refine the numerical calculation; they do not add animals or pens to the planned study.

Speak to the researcher. Start with the supported answer, explain uncertainty and limits plainly, and name the next information or design change that would help.

After class, spend 1–2 hours total on the final report, slides and rehearsal. Presentations are on Wednesday 2 December, 12:30–16:00. Suggested project formats are learning guidance; confirmed examination requirements will be communicated separately.