If n observations come in clusters of m with intraclass correlation ρ, the variance of their mean is σ2/n × [1 + (m − 1)ρ]. The bracket is the design effect, and neff = n/[1 + (m − 1)ρ] is the number of independent observations that would give the same precision. The poll view repeats the book's experiment (12 people, polled at random or as 4 groups of next-door neighbours on a street sorted by opinion). The technical replicates view shows what happens when the clustering is ignored in a test: m measurements of each of n mice treated as nm independent ones give far too many false positives.
code/ch13_experimental_design.R (poll_once, Figure 13.12).