Test of Missing Data Mechanism

• Activity Number: 634
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Keiji Takai*+
• Companies: Kansai University
• Keywords: missing data ; missing-data mechanism ; asymptotic theory

Abstract

Missing data often occur when we collect data. In analysis of the missing data, it is important to distinguish the missing-data mechanisms generating the data. In this talk, we propose tests of missing-data mechanisms based on the likelihood function. The null hypothesises are ignorability and MCAR. We develop the tests based on asymptotic theory of maximum likelihood estimator constructed from missing data and derive tests to test ignorability and MCAR, under the some assumptions in addition to the so-called regularity conditions. The first and second tests are constructed in the analogous way. Computation of the two tests can be made with the EM and the SEM algorithms and the Bootstrap method. We give numerical examples to show performance of the tests.