Multilevel models

Multilevel models (also called hierarchical or mixed models) have been developed to answer issues raised by data structured by several levels, typically when some individuals share a common context that may affect the considered behaviour. This is for instance the case for pupils in one school, employees in one firm, patients in one hospital... The clas-sic questions that are adressed by multilevel models are to highlight the existence of these "contextual effects", to quantify in which measure they contribute to explain heterogeneity between individuals and/or simply obtain unbiased estimates of the impact of some individ-ual variables we are interested in. This document presents a first practical introduction of these models. It insists on the details of their concrete implementation by usual statistical softwares (Sas, R, Stata) and on the interpretation that can be done of the results obtained by these methods. It shows two concrete examples corresponding on a variable of interest respectively continous and binary.