Patches Johnson

Department of Statistics

University of Florida


Non-Linear Path Models with Continuous and Discrete Variables

Path models and their analyses are well-developed for linear systems of equations with continuous variables. However, literature and methodology are very scarce for recursive path systems that include discrete or dichotomous variables, or non-linear equations. My talk will focus on the development of methodology to analyze these types of systems. In particular, methodology to analyze systems of three variables when all variables are dichotomous will be shown. Extensions to systems of more than three variables. Methodology to study systems with non-linear functions will also be discussed. These path models are very useful in a variety of applications. Two applications of interest to be discussed include: 1) Maternal and Child Health Care- studying the relationships between race (discrete), low birth weight (intermediate continuous variable) and infant mortality (dichotomous outcome variable). 2) Clinical Trials - relationships between treatment (continuous or discrete), blood pressure (continuous "potential surrogate" variable for long-term outcome) and ultimate outcome variable (stroke - dichotomous).


Back to Colloquium Series