Table 2 Summary of simulation study results and recommendations on their use

Method

Use equal weighting of clusters and clusters minus cluster-level parameters degrees of freedom

Use restricted pseudo-likelihood and clusters minus cluster-level parameters degrees of freedom

Use Fay and Graubard standard errors, clusters minus cluster-level parameters degrees of freedom, and an independent working correlation matrix

Valid results ^a

Cluster size > 10

Or

Common outcome (prevalence > 10%)

All scenarios

Cluster size ≤ 50

Or

CV cluster size ≤ 0.5

Competitive power ^b

Common cluster size

Or

High ICC (ICC > 0.05)

Varying cluster size

Or

20 + clusters

Low ICC ≤ 0.01