Carolyn J. Anderson, University of Illinois Urbana-Champaign

Log-multiplicative association models for categorical data have been derived from different psychological theories. Even though psychological theories differ, the same set of statistical assumptions have been used to derive essentially the same model for data.  Since underlying theories lead to the same model for data, fitting models to data does not provide sufficient evidence in support of one theory or another.  Additional assumptions are required, and the context in which the models are used needs to be taken into consideration. Emphasis is placed on underlying theories and the corresponding parameterizations of the models that include multivariate normality, multivariate nominal item response models (and special cases), sets of (conditional) multinomial logistic regression models, discrete choice models, scaling models, the Ising model, and others.

about the speaker

Carolyn received her BS with a double major in Economics and Psychology from the University of California Berkeley, her MS in statistics from the University of Illinois Urbana-Champaign, and her Ph.D. in quantitative psychology from the University of Illinois Urbana-Champagin.  I won dissertation awards from both the Psychometric Society and the American Psychological Association (Div 5).  I was a professor at the University of Illinois from 1993 to 2022 in the Department and Educational Psychology with affiliated appointments in the Departments of Psychology and Statistics.  I am currently professor emeritus in the College of Education at the University of Illinois Urbana-Champaign.  

The major focus of my research has been on modeling multivariate categorical data which encompasses psychometrics, graphical models, statistical models for categorical data, and model estimation.  I have published papers and chapters on decision making, social network models, multinomial logistic regression, multilevel modeling, and association models for categorical data.  I made the connection between graphical models for discrete and continuous variable, log-multiplicative association models, and equivalence to IRT models.  My early work on graphical models for categorical data (especially papers with Jeroen Vermunt and Hsiu-Ting Yu) led to the development of what is now coined network psychometrics.  I also collaborated with colleagues on applied papers on a variety of different subjects.  The Psychometric Society has been my main professional home, and based on years of experience, I encourage you to always go back to original sources and attend sessions at IMPS outside your area of expertise.