Modeling Multilevel Data with Small Sample Sizes

The conference is preceded by a one-day workshop on April 11 on “Modeling Multilevel Data with Small Sample Sizes

With multilevel data structures, large samples on the cluster level are particularly difficult to obtain due to  financial constraints (e.g., it is prohibitively costly to sample many higher level units), geographic constraints (e.g., the problem being studied only pertains to a few countries), or population constraints (e.g., there are few people who qualify for inclusion in the study). In these contexts, it is often infeasible or impossible to collect more data and small sample analyses can be unavoidable. However, many common methods for performing multilevel analysis are unstable or provide poor estimates with smaller sample sizes. This workshop will discuss strategies that can be applied to multilevel data with small samples to yield estimates that are as unbiased as possible while also maintaining as much power as possible. The workshop will feature strategies from both Bayesian and frequentist perspectives and will cover model types such as two-level models and latent growth models.


Rens van de Schoot
Associate Professor – Methodology and Statistics
Utrecht University

Dan McNeish

Assistant Professor – Center for Developmental Science and the Quantitative Psychology program at the University of North Carolina, Chapel Hill.