A Natural Approach for Addressing Multiplicity Involving Neuroscience Functional Data

• Activity Number: 307
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Junshui Ma*+ and Svetnik Vladimir
• Companies: Merck Research Laboratories and Merck Research Laboratories
• Keywords: Clinical Data Analysis ; Neuroscience Data Analysis ; Functional Data Analysis ; Multiple Comparison ; Family Wise Error Rate ; High Dimensional Integration

Abstract

Many neuroscience modalities, e.g. fMRI, and EEG, output functional data in the time, spatial, or frequency domain. Multiplicity issue arises when inferences are simultaneously conduced at discretized domain points. The nature of functional data suggests that the domain should be treated as continuous, and the correlation among the inferences should be considered. The approaches that address multiplicity by adjusting p-values are not applicable. A natural solution based on functional data analysis is proposed. The essence of the proposed approach is to estimate the joint distribution of the simultaneous inferences, and directly calculate the FWER from the joint distribution. When the functional data, along with other factors (e.g. treatment), are modeled with a set of continuous basis functions, and the inferences are represented as linear transformation of the basis functions and model coefficients, the covariance matrix of the inferences can be estimated from the fitted model. Thus, the joint distribution of the inferences is obtained under Gaussian assumption. Calculating FWER from the joint distribution requires high dimensional integration, which recently becomes feasible.