Examining of Multiple Imputation Method in Two Missing Observation Mechanisms

Abstract

This study contains an examination of the missing data structures, occurring in many fields, especially in livestock. It also examines the processes to obtain the solution for the missing data. For this purpose, linolenic acid measurements obtained from four different anatomic regions of two animal species were taken as dependent variables. For the dependent variable, the observations were deleted at the ratio of 10% and 20%, creating the missing structures of missing completely at random (MCAR) and missing at random (MAR). Subsequently, these data sets were completed using multiple imputation (MI) method. Generalized Estimating Equation (GEE) and mixed model methods were used in the missing data structures and for the purpose of evaluating the data completed with MI. The study were obtained almost same results obtained from GEE and mixed model in the missing data structures. At the same time, there was not found difference between the methods in completed data using MI method. As a result, it is stated that valid results obtained in missing data structures by used GEE and mixed model analysis. When these results are also compared, it can be concluded that multiple imputation (with these ratio of missing) is not necessary before GEE and mixed model.