Hierarchical Data Structures and Multilevel Modeling

The statistical methods most familiar to readers of the biomedical or social-behavioral literature usually assume independence of sampling. The Student t  test, 1-way ANOVA, their nonparametric equivalents, as well as the χ2 and basic multivariable linear and logistic regression, all assume that the observations analyzed are independent of one another. These methods will produce invalid results when this assumption does not hold, for example when d ata are in a hierarchical arrangement that lends shared characteristics to groups of observations.
Source: The Journal of Pediatrics - Category: Pediatrics Authors: Tags: Statistics, Explained Source Type: research