ABSTRACT
The rationale for applying multilevel models to hierarchical data is well established (Skrondal and Rabe-Hesketh, 2004; Snijders and Bosker, 1999). When lower level units are nested within one or more higher level strata, conventional single level regression analysis is not appropriate since observations are no longer independent: pupils in the same schools, or households in the same communities, tend to be more similar to one another than pupils in different schools or households in different communities. Such dependency means standard errors are downwardly biased if the nesting is ignored, and spurious inferences regarding predictor effects may be made (Hox, 2002).
