50. You assigned a new workflow to forty offices and collected thousands of employee responses. What determines the uncertainty of the treatment comparison?
HardStatistics & experiments
Technical
Show model answer
Offices were randomized, so employees within an office do not represent independent treatment assignments. Shared conditions can make their outcomes correlated; thousands of responses cannot be analyzed as thousands of independently randomized units.
Use a design-appropriate cluster analysis or randomization-based method, considering office sizes and the number of clusters. Report the limited assignment count and avoid claiming precision based only on the employee response total.
What interviewers listen for
- Analyze the randomized clusters
- Account for within-office dependence
- Report the actual assignment count