An Analysis Of Factors Affecting The Multiple Imputation Based On Discrimination

In statistical investigations, due to various reasons there always exists missing data invarying degree in survey. Especially in recent years, with the development of big data and thedatabase wide-ranging, both the increase of magnitude of sample size and the amplification ofvariate dimensionality as well as the instantaneity of recording time have increased theprobability of generating of the missing data, the generation of missing data is becoming moreand more difficult to avoid. However, the missing data has brought different deep impacts onstatistical analysis. On the one hand, It increased the difficulty of the process of statisticalanalysis. As the classical analysis methods are based on a assumption that the collection iscomplete. On the other hand, the results of statistical analysis probably lost the value ofapplication as unrepresentative. The research of missing value becomes more and moreimportant. At present, the higher attention on the missing data imputation has been receivedboth in the domestic and overseas. Especially as the thought of multiple imputation was putforward, it showed great advantages in practice which makes it became the focus of thevarious researchers.In data investigations, questionnaire content may often lead that missing data of surveyshows different forms. The missing mechanism of data, missing pattern, the degree of datamissing and the relationships between the other variables will have influence on theimputation methods and effect. The article focuses on the multiple imputation in singlevariable random missing and multivariate at random missing pattern. It gives attentions to themultiple imputation based on discrimination model with different sample sizes, missingdegrees, auxiliary variables and so on.The whole article is divided into five chapters. In the first one, the article givesintroduction to the topic selection background and research significance. Besides, itsummarizes the domestic and overseas research status and the research achievements aboutmultiple imputation, especially multiple imputation. In the second chapter, it expounds thecauses of missing data and the loss mechanism. As well it points that as imputation can makethe shortfall of a incomplete collection, it is necessary to control the data quality ahead of time. In Chapter3, it detailed introduces not only the related basis theory about themultiple imputation but also the comparison and evaluation of related methods in details. InChapter4, it studies that empirical analysis of the missing data imputation effect that based ondifferent sample size, ratio of loss, auxiliary variables in the case of univariate andmultivariate. Chapter5is a summary of the article and prospect of the whole article.