# Week 12 worksheet: a separate teaching example, then your own project

Four groups of three: driver, outcome translator, uncertainty auditor. Rotate roles after the break. All cress observations and conditions are invented; they provide practice, not assessed-case evidence.

## Opening recall (12:30–12:45)

Recall the calf study: researchers assign one diet to each calf and record five weights from it. What would change if diet were instead assigned to a whole pen? Then name the animals or groups researchers treat or follow in your own study, what they measure, and which measurements share an animal or environment.

## Teaching target (13:15–13:40)

Researchers choose pulse or mist watering for each tray and count sprouts out of 25 seeds on day 10. Trays may have higher or lower sprouting chances than their watering and batch predict; we call that extra difference a tray deviation. Compare mist minus pulse with that deviation set to zero, and give the six batch probabilities equal weight. State the units and time. Read data/README.md before running estimates-student.R.

1. Researchers choose a watering method for each tray, and all 25 seeds in that tray receive it. How many trays receive each method? Why does counting more seeds in those trays not add more tray assignments?
2. Explain the data generating process: tray conditions change the chance of sprouting, then a count is observed. Why do seeds in one tray share a tray term?
3. Read the model coefficient and spread of tray differences on its internal log-odds scale, then use the printed probabilities and mist-minus-pulse comparison. Odds are chance of sprouting divided by chance of not sprouting. Convert the probability difference and interval to percentage points. Which result sets tray deviation to zero? Why does that differ from averaging over trays with different deviations?
4. Does a probability interval describe the distribution of individual seed outcomes? A seed either sprouts or does not; the interval concerns estimation of a probability. A new tray count has additional tray and binomial variability.
5. Contrast hypothetical probabilities .40 and .55: give difference in percentage points, risk ratio and relative change. These illustrative values are separate from the fitted teaching results.

## One figure, one paragraph (13:55–14:30)

Using the cress teaching result, make or annotate one figure and write 100–150 words: question and design; model and the trays researchers assigned to each watering method; estimated contrast and interval; practical interpretation against an invented worthwhile change of 15 percentage points, chosen before seeing the results; and one limitation. Explain why an interval can exclude zero yet span the practical threshold. Label day 10, comparison direction, units and interval, and say that the records are invented.

## Peer review (14:30–14:55)

Swap teaching-example drafts. Recover the question and conclusion without an oral explanation. Check comparison direction, units, time, conditions, interval type, allocation unit and wording. Give one change that improves the scientific explanation and one feature to preserve; authors revise immediately.

## Your project workshop (15:10–15:45)

Return to your own approved dataset or raw fallback. Write and justify your own scientific question, main outcome, how researchers treated or followed the animals or groups, the comparison to estimate, a candidate model and checks of its assumptions. The cress model and conditional target need not suit your project. Choose which predictions/contrasts answer your question, explain what is conditioned on or averaged, and record an alternative worth comparing. Consult the instructor about an unresolved design/analysis issue. Do not copy the teaching script and simply rename variables.

## Exit (15:45–16:00)

Individually explain: why crossing zero does not prove no effect; why precision does not establish causality; and one decision you must justify for your own project.