Modern Multivariate Analysis

Term: 
Winter
Credits: 
2.0
Course Description: 

Advanced methods for ALL tracks.

This course gives an introduction to some of the hottest topics in multivariate analysis. The approaches discussed do no try to build a simplistic model of how some variables may affect other ones, rather ask, whether, with an appropriate definition of these concepts, associations and effects (evidential or causal), exist. The principles and methods developed can be applied equally to survey and organic data.

Topics to be covered:

Effects and associations, Simpson's paradox Latent class analysis Data collection design and inference Causal analysis from observational studies Propensity score based matching Graphical models of causality Bayesian networks Analysis of higher order interactions Marginal models and path models

Learning Outcomes: 

This course gives an introduction to some of the hottest topics in multivariate analysis. The approaches discussed do no try to build a simplistic model of how some variables may affect other ones, rather ask, whether, with an appropriate definition of these concepts, associations and effects (evidential or causal), exist, or perhaps can be explained away with other variables. Approaches to the definition, measurement and detection of effects will play a central role. The topics to be covered include Simpson’s paradox, modern theories of causality, Bayesian networks, path analysis. A unifying technical approach (marginal modeling) will be used. The principles and methods developed can be applied equally to survey and organic data. The class will emphasize understanding of the relevant concepts and the students will be encouraged to explore the various software implementations.

Assessment: 

Grading and student requirements
To earn credit in this class, students will have to
write a final test (30%)
submit a research report at the end of the term (70%)
Students who audit the class will have to successfully write the final test

Prerequisites: 

The class is offered at the PhD level. Some knowledge of multivariate statistical methods at about the level covered in the Multivariate Statistics class at CEU is assumed. Students are assumed to be able to work with software on their own. The particular software platform is not specified, but R has the most coverage of the material of the class.