Data sparseness in contextual population health research: effects of small group size and cluster analysis on linear and non-linear multilevel models

The current use of multilevel models to examine the effects of surrounding contexts on health outcomes attest to their value as a statistical method for analyzing grouped data. But the use of multilevel modeling with data from population-based surveys is often limited by the small number of cases per level-2 unit, prompting a recent trend in the neighborhood literature to apply cluster analysis techniques to address the problem of data sparseness. In this paper we use Monte Carlo simulations to investigate the effects of marginal group sizes and cluster analysis techniques on the validity of parameter estimates in both linear and non-linear multilevel models.