Hierarchical Modeling

Jan de Leeuw
UCLA
Statistics

In this presentation we outline the hierarchical modeling or multilevel approach to the
analysis of fMRI data. In this approach we have preprocessed fMRI data collected under
various conditions on on various groups of individuals. Each group or condition gets its
own dedicated general linear model, in which the regression coefficients for the various
groups are the outcomes of a between-group regression model that ties the separate
within-group regression models together.



We give a general introduction to linear two-level modeling, and then extend the
discussion to generalized linear models, to multivariate responses, to more than two
levels, and to more complicated error structures. Estimation methods and computer
software are discussed briefly, and some of the available software is applied to
preprocessed fMRI data.


Back to Graduate Summer School: Mathematics in Brain Imaging