When Nesting is Nuisance in Cluster Sampled Longitudinal Research

Cluster sampling is an efficient means of recruiting and retaining an optimal sample size for a longitudinal study, but results in sample data nested at three levels (repeated measures [level 1] collected from participants [level 2] sampled from within cluster units [level 3]) when research questions often only involve the first two. Researchers have three options when analyzing cluster sampled longitudinal data to avoid inferential errors when the third level of nesting is nuisance: 1) model it, 2) control it, or 3) a hybrid combination of both. The purpose of this paper is twofold. First, a smaller-scale Monte Carlo simulation was conducted to determine if options to control for nuisance nesting can achieve comparable parameter estimate accuracy, confidence interval coverage, and statistical power levels obtainable when nuisance nesting is modeled. Second, we demonstrate the three methods to address nuisance nesting using data from a previously published multi-site longitudinal study. The R and rblimp syntax scripts used to impute missingness and analyze the example data in keeping with a “nesting is nuisance” approach are included.