|Friday, February 22|
|CS02 Theme 2: Data Modeling and Analysis #1||
Fri, Feb 22, 9:00 AM - 10:30 AM
Multigraph Representation of Loglinear Models (302419)
Keywords: mathematical graph, multigraph, loglinear model, conditional independence, categorical variable
In the last 30 years graph theory has been used, with great success, to analyze and interpret loglinear models (LLMs). In short, the goal of a LLM is to identify the structural associations among a set of categorical variables, a task that can be daunting as the number of variables increases. In a landmark 1980 paper, Markov fields and chordal graphs were used to analyze LLMs, and since then many researchers have cultivated this marriage between graph theory and statistics. In this presentation, the generator multigraph will be used to analyze and interpret the LLM, thus allowing the researcher to identify all conditional independencies among the categorical variables, and all collapsibility conditions, in an organized and comprehensive way. Actual client data sets will be used to illustrate the technique.