Fig. 2.

Schematic representation of observations within clusters differing in density and sampling strategy. Notes: Panel (a) shows a balanced sampling pattern with 10 dots sampled by design within each of 10 randomly selected clusters, a situation often handled through the use of weights or so-called “fixed effects” (dummy indicator variables for all clusters except for an omitted reference cluster). Panel (b) shows a sparse, unbalanced pattern of 100 dots arranged randomly across 100 clusters, resulting in few observations per cluster (including some clusters with zero observations). Panel (c) shows an unbalanced pattern 1000 dots within 100 clusters.