I am involved in research which aims at comparing health outcomes of pediatric hospitals units. The specific problem in pediatrics is that both the sample of hospitals and the sample size of patients within each hospital are very small. Methods such as hierarchical modelling do not work. The literature I have reviewed suggests two solutions which are not really satisfactory: (1) to drop the hospitals below a certain patient sample size; (2) to pool the data over a certain number of years. Does anybody know of other suggestions or solutions?