What is the intra-class correlation coefficient? What is the estimated between-cluster and within-cluster (or between- and within-level) variability in the data points? Do you have a suffient number of clusters/levels to justify a multi-level analysis? Or are there so few "levels" that you are better off simply adjusting for a dummy variable representing cluster or level?