Clustered Data Inference When the Cluster Size is Potentially Infomative

Somnath Datta
University of Louisville, USA

We discuss how to extend parametric and nonparametric inference procedures when the classical assumption of independence is violated due to clustering. Clustered data arise in a number of practical applications where observations belonging to different clusters are independent but observations within the same cluster are dependent. While making adjustments for possible cluster dependence, one should also be aware of the “informative cluster size” phenomenon which occurs when the size of the cluster is a random variable that is correlated to the outcome distribution within a cluster, often through a cluster specific latent factor. We demonstrate the correct inference procedures under various scenarios.