Home WakeSpace Scholarship › Electronic Theses and Dissertations

THE ROBUSTNESS OF EBIC GLASSO TO DISTRIBUTIONAL MISSPECIFICATION

Electronic Theses and Dissertations

Item Files

Item Details

abstract
Network analysis is gaining in popularity as a method for modeling psychopathology. EBIC glasso is a common method of estimating such networks. An assumption of EBIC glasso is that the residuals follow a normal distribution. Thusfar, no one has assessed the robustness of EBIC glasso to distributional misspecification: assuming the data is normal when it is actually nonnormal. Using a 5 x 4 x 4 simulation design with 500 replications, the robustness of EBIC glasso is tested against such misspecifications. The conditions are compared via standardized bias and multilevel metamodels. The current study finds that misspecification in the case of assuming data is normal when it, in fact, is not, does have an effect of the parameter's deviation from the true values of the model in that underestimation of the true values occurs. This effect varies across condition and centrality measure. The paper suggests great care being taken when using nonnormal data with EBIC glasso.
subject
EBIC glasso
network analysis
robustness
contributor
Slipetz, Lindley Rose (author)
Cole, Veronica T (committee chair)
Furr, Mike (committee member)
Garrison, Mason (committee member)
Henry, Teague R (committee member)
date
2021-06-03T08:36:24Z (accessioned)
2021-06-03T08:36:24Z (available)
2021 (issued)
degree
Psychology (discipline)
identifier
http://hdl.handle.net/10339/98851 (uri)
language
en (iso)
publisher
Wake Forest University
title
THE ROBUSTNESS OF EBIC GLASSO TO DISTRIBUTIONAL MISSPECIFICATION
type
Thesis

Usage Statistics