A0574
Title: Testing missing completely at random for partially observed functional data
Authors: Siegfried Hoermann - Graz University of Technology (Austria)
David Kraus - Masaryk University (Czech Republic)
Dominik Liebl - University Bonn (Germany)
Maximilian Ofner - Graz University of Technology (Austria) [presenting]
Abstract: The statistical analysis of incompletely observed functional data, referred to as partially observed functional data, has gained considerable attention recently. Corresponding data can result from issues like malfunctioning measuring devices that fail to record the signals over certain periods. Current methods for analysing such data typically rely on a missing completely at random (MCAR) assumption, meaning the missingness mechanism is independent of the data itself. However, limited focus has been given to verifying this assumption. This gap is addressed by proposing novel procedures to test MCAR for different types of random functions. In addition, asymptotic distributions are established, and the consistency of the tests is discussed under various alternatives. The finite sample properties of the methods are then evaluated through a simulation study. Finally, a real-data application illustrates the practical utility of the methodology, highlighting its potential not only for justifying statistical methods but also for identifying the cause of missingness.
