|
Activity Number:
|
74
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #305243 |
|
Title:
|
Inference and Diagnostics in Nonlinear Structural Equation Models with Applications in Neuroimaging
|
|
Author(s):
|
Klaus K. Holst*+ and Esben Budtz-Jørgensen
|
|
Companies:
|
University of Copenhagen and University of Copenhagen
|
|
Address:
|
Øster Farimagsgade 5, opg B., Copenhagen, International, 1014K, Denmark
|
|
Keywords:
|
structural equation model ; model diagnostics ; nonlinear model ; neuroimaging
|
|
Abstract:
|
Structural equation models (SEM) are an important tool for causal modeling of complex and high dimensional data structures with the potential of capturing important features in data that would be concealed in a standard univariate analysis. The widely used classical SEM formulation assumes strictly linear relationships between variables. However the validity of these strong assumptions is often neglected due to lack of sophisticated model diagnostics tools in available software tools. In the talk recent developments in model diagnostics for detecting misspecifications of the mean structure will be discussed. The main focus will be on techniques based on aggregates of residuals. Furthermore different estimation approaches for handling possible non-linear relationships between latent variables will be compared and demonstrated on data from a PET brain imaging study.
|
