Brian Habing ColloquiumThursday, March 30, 2017 - 2:45pm
Statistics Department Colloquium
Where: LeConte College, Room 210
Speaker: Brian Habing
Affiliation: University of South Carolina, Department of Statistics
Title: Some Implications of Model Choice
The most common implementations of item response theory can be mathematically thought of as non-linear factor analysis or as non-linear mixed modeling. The non-linear part typically involves using a logistic link to model the probability an examinee will correctly answer a test question, and the latent factor or random effect is the examinee ability. As in many branches of statistics, the specifics of the model (such as the logistic link) and estimation method are often based on ease of implementation. This talk will discuss how using the obvious choices can lead to some non-intuitive or unexpected difficulties, and will use examples from the research of current and former Ph.D. students Tammiee Dickenson,Yin Fu, Leslie Hendrix, Han Lee, Maureen Petkewich, Jessalyn Smith, Na Yang, and Xinchu Zhao.