Grip strength of severely malnourished children during nutritional rehabilitation in the Jimma hospital of Ethiopia

Abstract

The frequent occurence of missing data in scientific studies is not uncommon. In longitudinal clinical studies for instance, one would expect, in a 'dream land' scenario, a complete profile for each study subject. Unfortunately, this is rarely the case. Plagued therefore with the issue of missingness, it becomes vital to understand the mechanism that led to the missing data to be able to perform analyses that will lead to valid inference. Under the missing at random (MAR) assumption, likelihood based methods which admit ignoring the missing process as well as Multiple Imputation are valid. On the other hand the non-ignorable missing not at random (MNAR) process necessitates the need for models that explicitly incorporate the dropout mechanism. Unfortunately these models are usually based on very strong and unverifiable assumptions which are difficult to implement, and sensitive to misspecifications. The objective of this thesis was to compare the effectiveness of recommended methods