PhD defense: Small sample solutions for Structural Equation Modeling

Summary

Structural equation modeling (SEM) is a statistical modeling
procedure that is frequently used in social and behavioral
sciences to study the relationships among observable and
unobservable (latent) variables. Due to a combination of
factors, a prominent disadvantage of the technique is that
its recommended use is constrained to settings where there
is an ample amount of data available. One of the main
reasons for this is that the default estimator for continuous
data is Maximum Likelihood, which is prone to various issues
when applied to small to moderate samples. While some of
the issues are fairly easily identifiable (e.g., flagged by
software warnings/errors or solutions in the output that are
simply impossible from a theoretical perspective), other
issues are more likely to pass unnoticed for applied users
(e.g., biased point estimates, inaccurate confidence
intervals, or ill-behaving fit statistics). Given that in many
settings, small samples are simply reality, the overall goal of
this PhD project was to improve the usability of SEM in small
samples by identifying, implementing, and evaluating
different approaches that could potentially mitigate typical
small sample issues.

Please confirm your attendance before Monday,
January 15 via email to Sara.Dhaene@UGent.be