Topics in Analysis and Sensitivity Analysis of Incomplete Longitudinal data

Abstract

Research studies that involve the collection of data are often, though perhaps not intentionally, subject to some form of incompleteness, thus necessitating the application of modified analysis techniques. As a consequence, methodology for handling missing data has advanced significantly over the recent decades, and can be foreseen to continue doing so in the coming years. In the past, limitations – both methodological, as well as computational – have prohibited the use of more sophisticated approaches, and analyses of incomplete data have historically revolved around simple methods such as complete case analysis. Recent developments, however, have broadened the possibilities available to the researcher – a number of which have been examined in this thesis – and the consideration of such should supersede the use of simple techniques that have been shown to yield substandard results. At best, these na¨ıve approaches might be used as preliminary steps or supplements to more appropriate methods. Despite, however, the availability of a wide array of courses of analysis and corresponding computational routines for their implementation, the modeling of incomplete data remains less than straightforward. This is primarily due to the fact that the inferential validity of applicable techniques is hinged on unverifiable assumptions regarding the underlying missing data mechanism. A large part of this thesis has been devoted to emphasize this, underscoring the importance of sensitivity analysis in modeling incomplete data. The different methodologies presented in this thesis have been considered to high-light either (or both) of two objectives. On the one hand, a number of chapters (Chapters 5, 6, 8 and 9) serve to demonstrate, review and/or propose modifications to existing approaches, geared towards providing useful and important information regarding their application. Chapters 4, 7, and 8, on the other hand, introduce techniques that might serve as tools within the context of a sensitivity analysis, thereby broadening the possibilities under such. A brief overview of the resulting conclusions for the pertinent chapters is now presented. ...