A General Solution to Functional Data Modeling and Inference Using R

• Activity Number: 179
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Junshui Ma*+
• Companies: Merck
• Keywords: Functional Data Analysis ; Linear Mixed Effect Model ; Neuroscience Data ; Multiple Comparison ; R programming

Abstract

Functional data generated by various image and bio-signal modalities, e.g. fMRI and EEG, are increasingly observed in pre-clinical experiments and clinical trials. Although functional data analysis (FDA) has been a well-studied area, data analysts are frequently reluctant to analyze these datasets using FDA. One major reason can be that FDA has not been a widely familiar area and some statisticians can find the different views in the publications confusing. A quick solution is to propose a recipe-like procedure using a popular and freely available statistical language like R. In this study, a general FDA framework based on linear mixed effect models is proposed. It is capable of modeling a variety of functional data, including 1-dimensional bio-signals, and 2- or 3-dimensional images. At the inference stage, as a seamless extension to the modeling framework, a multiple comparison adjustment procedure is presented to generate powerful adjusted p-Values and simultaneous confidence intervals. This paper focuses more on how to implement the proposed procedure in R, less on the theories behind it. Two clinical functional datasets are analyzed to demonstrate its efficacy.