Clustered Standard Errors When the World Is Nested

Clustering adjusts inference when residuals correlate within groups. The right level often follows from assignment: randomize schools, cluster at school; randomize individuals within classrooms, think harder. Small cluster counts strain asymptotic approximations. Wild cluster bootstrap or permutation ideas may help; at minimum, report sensitivity to alternative clustering schemes when data allow. This note does not replace statistical consultation for your exact design — it offers reporting habits that survive scrutiny: state the cluster variable, report counts, and show robustness where feasible.