Principles of multilevel analysis and its relevance to studies of antimicrobial resistance

Akke Vellinga, Kathleen Bennett, Andrew W. Murphy, Martin Cormican

When studying antimicrobial resistance it is clear that individuals do not exist in isolation and are often clustered into groups. Data within groups are generally not independent, but standard statistical approaches assume independence of observations. When data are clustered (e.g. students in schools, patients in general practices, etc.) multilevel analysis can be used. The overall idea of multilevel analysis is that the clustering is taken into account in the analysis and provides additional information on the interactions between individuals and groups. The lowest level is often the individual and additional levels are formed by clustering in groups (the higher levels). This article introduces the principles behind multilevel modelling. The approach is to provide readers with sufficient information to understand outcomes in which this statistical technique is used, without expecting the reader to be able to perform such an analysis. As multilevel modelling can be seen as an extension of linear regression analysis, this is the starting point of the article. Other concepts and terms are introduced throughout, resulting in the explanation of the accompanying article on antimicrobial prescribing and resistance in Irish general practice (Vellinga A, Tansey S, Hanahoe B et al. J Antimicrob Chemother 2012; 67: 2523-30).