## Detail buku

No Buku T.LN.05.28 Chiba University, JAPAN Ninik Anisa APPLICATION OF BOOTSTRAP METHOD TO IMPUTATION IN SURVEY DATA Professor DR.Eng.Masaaki TAGURI ABSTRAK INGGRIS : One problem in survey sampling is that unit or item non response occur frequently. Compensating to this problem, imputation methods are often employed using auxiliary data. Not only deterministic imputation, such as ratio and regression imputation, but also random imputation such as ratio random and regression random imputation can be used. However, treating the imputed data set and directly applying existing methods especially in bootstrap method, it does not produce the valid result because bootstrap method does not account for the effect of missing data. The application of bootstrap method to the variance estimation and statistical inference, which will give asymptotically valid result, should take into consideration sampling design, imputation methods and type of statistics used in inference. If there are missing data the use of naive bootstrap estimator makes underestimate, unless the imputation of bootstrap data set is done in the same way as original data sets. The objective of this article is the comparison of the imputation methods make use of auxiliary data observed for all data. The viewpoints for the comparison are relative bias, variance, efficiency and distribution function. As for the imputation methods, we use ratio and regression methods for deterministic imputation and ratio and regression random methods for random imputation and the considered parameter is the population total. From simulation results, the relative bias and the variance increase as the response rate and the correlation coefficient decrease. The deterministic imputation method dependent on the condition of the auxiliary variable, when the auxiliary variable is not equal to 1 then the deterministic imputation is the best imputation method. Although the random imputation is not the best imputation method but independent on the condition of the auxiliary variable therefore we may say that the random imputation can handle in any situation. In random imputation we can say that regression random imputation may be the best method to impute missing values. This fact had proved that this method can reduce the variance compared to ratio random imputation. The superiority of regression random imputation to preserve the distribution function can be shown because the bias of parameter. Thesis