¹⁶Since multilevel analysis involves two or more levels, questions are often asked about optimal sample sizes. Hox (2002) mentions Kreft’s 30/30 rule, which suggests 30 groups with at least 30 individuals in each. This could be sufficient for the estimation of the regression coefficients but inadequate for other purposes. If it is cross-level interactions that are of interest, Hox recommends the 50/20 rule: 50 groups with 20 or more in each group. If there is strong interest in the random part, the advice is 100 groups with a minimum of ten in each: http://essedunet.nsd.uib.no/cms/topics/multilevel/ch3/5.html. A slightly different take is offered by Rabe-Hesketh, Skrondal (2008, p. 62): “It is often said that the random-effects approach should only be used if there is a sufficient number of clusters in the sample, typically more than 10 or 20. However, if a random-effects approach is used merely to make appropriate inferences regarding β, a smaller number of clusters may suffice. Regarding cluster sizes, these should be large in the fixed-effects approach if the α_j are of interest. However, in random-effects models, it is only required that there are a good number of clusters of size 2 or more. It does not matter if there are also ‘clusters’ of size 1”.