Module 02 Activity

Scenario

A dataset has arrived for a comparison. Establish what one row is and whether the table can support the test at all.

What you build

An analysis table with a stated grain, declared dtypes and a passing check dictionary.

Steps

1. Write down the outcome, predictor, grouping, unit and time window before touching the data.
2. Set the dtype of every column deliberately — Categorical for groups, ordered=True for ordinal scales, nullable integers for counts.
3. Count missing values by group with count in the aggregation, and report whether missingness is balanced.
4. Compare len(df) with df[unit].nunique(). If units repeat, aggregate before testing.
5. Set the reference level explicitly with C(var, Treatment(reference=...)).
6. Run a dictionary of boolean checks and treat every failure as a question for the data owner.

Evidence to hand in

• The five decisions, written before the analysis.
• The dtype of every column, with the reason.
• The missingness table by group.
• Rows versus distinct units, and any aggregation.
• The reference level you chose.
• The check results, with what each failure led to.

Review checklist

• The unit of analysis is stated and the table's grain matches it.
• Missingness is counted by group and its balance reported.
• Repeated measurements are aggregated, or the dependence is addressed.
• Every grouping variable is Categorical with a deliberately set reference.
• No failed check was silently relaxed.

Mark complete