Making the Most of Clustered Data in Laboratory Animal Research Using Multi-Level Models

In the following review article, I address the fitting of multi-level models for the analysis of hierarchical data in laboratory animal medicine. Using an example of paternal dietary effects on the weight of offspring in a mouse model, this review outlines the reasons and benefits of using a multi-level modeling approach. To start, the concept of clustered/autocorrelated data is introduced, and the implications of ignoring the effects of clustered data on measures of association/model coefficients and their statistical significance are discussed. The limitations of other methods compared with multi-level modeling for analyzing clustered data are addressed in terms of statistical power, control of potential confounding effects associated with group membership, proper estimation of associations and their statistical significance, and adjusting for multiple levels of clustering. In addition, the benefits of being able to estimate variance partition coefficients and intra-class correlation coefficients from multi-level models is described, and the concepts of more complex correlation structures and various methods for fitting multi-level models are introduced. The current state of learning materials including textbooks, websites, and software for the nonstatistician is outlined to describe the accessibility of multi-level modeling approaches for laboratory animal researchers.