Added value of multilevel models

Hierarchies are very common in the social and the behavioural sciences and often occur naturally: e.g., pupils in classes, classes in schools; employees in departments, departments in firms; suspects in courts; offspring within families. Less obvious examples of hierarchies are observations nested within subjects (repeated measurements) or observations nested in studies (meta-analysis). In the field of traffic safety nested data structures can be seen in data on roadside surveys (drivers nested within police checks or locations, police checks or locations nested within regions); on accidents (drivers and passengers in vehicles, vehicles in accidents, accidents in regions; on repeated measurements; or on meta-analysis.

Multilevel analyses are capable of dealing with the issue of dependence of observations thus calculate correct standard errors, taking account of the degree of dependence of the observations in the sample under study. Moreover, research problems in social and behavioural science often involve complex relationships between variables that belong to different levels of aggregation. Those complex problems simply cannot be solved with analyses at either the aggregated or the disaggregated level. Multilevel models overcome these obstacles in an elegant and productive way by allowing the researcher to analyse those different levels synchronically and study how they interact. It is therefore possible to translate a research problem into a design reproducing a lot of the nuances at stake and without giving in too drastically towards simplifying the nature of the issue under evaluation.

Given the wide range of research problems involving multiple levels, there are also a large number of different multilevel models including repeated measurements, binomial regression, Poisson regression, multivariate models. These models are extensions of the more classical analysis techniques tailored to the research situation. Which model is appropriate depends on the research design (2-samples, correlations, repeated-measures) and the type of the dependent variable (continuous measurements, proportion data, counts). The multiple levels models described by SafetyNet should be viewed as generalisations of these already known analyses.