|
Abstract:
|
Regression models are useful statistical tools to predict future outcomes based current available data. However, missing values occur frequently on important predictors with real data. Multiple imputation method, which provides valid statistical inference in many situations, can be costly in computation with large data set and large number of predictors. Missing indicator method, which can generate biased estimates for regression coefficients, is simple to implement and computationally inexpensive. In this paper, we present simulation studies to study the bias and mean square error in predicting outcome using both methods under various missing data mechanism, and discuss its implication in handing missing data with large data sets.
|
