506 – Differential and/or Biased Missingness: Myths, Methods, and Manifestations
Functional Response Models for Addressing Outliers and Beyond
Tian Chen
University of Toledo, Toledo, Ohio
R. Chen
Amazon.com, Inc., Seattle, Washington
P. Wu
Christiana Care Health System, Newark, Delaware
N. Lu
Huazhong University of Science and Technology, Wuhan, China
H. He
Tulane University, New Orleans, Louisiana
H. Zhang
St. Jude Children’s Research Hospital, Memphis, Tennessee
J. Kowalski
Emory University, Atlanta, Georgia
Xin Tu
University of Rochester, Rochester, New York
Functional response models (FRM) are widely used to address limitations of classic statistical models as well as timely issues arising in research and practice. In this paper, we discusses two most recent applications of this class of models: one in robust regression analysis against outliers and the other on modeling human interaction. Popular semi-parametric regression models such as the generalized estimating equations (GEE) improve robustness of inference over parametric models. However, such models are not robust against outlying observations. The rank regression (RR) provides more robust estimates over the GEE. Unfortunately, the RR and its recent extensions to longitudinal data do not sufficiently address missing data. We discuss a new FRM-based approach to address outliers in longitudinal studies with missing data following the missing at random (MAR) mechanism. The second application focuses on modeling between-subject attributes, a new data type arising in modeling human interaction such as friendship between individuals. Such attributes are both conceptually and analytically differerent from whtin-subject attributes in conventional statistical models. The FRM is uniquely positioned to model this new data type.