The Application Of Multiple Imputation In Compositional Data Analysis

Compositional data mainly presents the relative information of the whole instead of the absolute information. Recently, it is widely applied in the social structure, economic development, engineering. However, in the process of data collection and data storage, the missing data is inevitable because of a variety of known or unknown factors leading to the missing data situation. Missing data will increase the complexity of analysis, resulting in bias of the results, decreasing the efficiency of statistical work, so the handling of missing data becomes extremely important. The imputation for the missing data make the complete data set which is analyzed by statistics which can reflect the sample information better and the results reflected is more real. This paper introduce mainly about the imputation for compositional data with missing values, we do some study cases to introduce the application of the method. This theses is divided into four parts.The first chapter is introduction of the problem and the arguments of the background and the practical significance. In this chapter, we review the research of compositional data with missing values, and make a brief description of the work of this paper. Chapter two presents the introduction of the resolution with missing values and we review in detail the mean imputation and regression imputation method in theory and has been the handling of missing values. In chapter three we study the imputation methods of compositional data on the basis of the chapter two and proposed the multiple imputation for compositional data. The fourth chapter, we apply the SAS statistical software to the example of regional urban households in2004per capita annual consumption expenditure structure to get the reasonable results.