Imputation Methods for Missing Data高级医学统计学2完整课件.ppt

* Example Two: Comparation of Imputation Methods reg Y X, both standard distributed continuous variables. Y = 1*X + 1*error n = 50 i = 3%, 8%, 13%…. 68% of X are set missing. for each i: 200 replications were made. y x Examples * Complete-Case Analysis Mean substitution Substitution by regression Hotdeck MI Example Two: Comparation of Imputation Methods Multiple Imputation Works ok but waste of information,  particularly in multivariate analyses Complete-Case Analysis Percent missings in x Standard deviation for beta Estimate for beta ± sd Quite stable estimate,  stronger increase in sd than in complete case analysis Mean substitution Overestimation of the effect when response is included  Substitution by regression Considerably more variance due to imputation,  break-down at about 50 % missings (m = 5, 4 variables) Hotdeck Stable estimates with small variances (m = 5, 4 variables) Multiple Imputation on Chained Equations - ICE Analysis of complete cases not bad when only few variables Imputation of means not bad for continous variables  take the mean for categorical variables, too Regression imputation don’t include response into model Hotdeck Stata’s version is not recomendable Multiple Imputation by Chained Equations very good Example Two: Comparation of Imputation Methods * Outline Background Traditional Approaches Modern Approaches Examples Summary * Multiple imputation is a method of replacing missing values which has some theoretical advantages over other methods. Software is becoming more common to handle multiple imputation and the code is relatively simple. Summary * Do something about the missing data! Many results suggest that in general there is benefit from imputation over the commonly used complete-case analysis. Data analysis must: Identify pattern amount of missing data Assess why it is missing Determine what to do about it * How many imputations are needed ? Rubin (1987) shows that t