Model Averaging For Regression With Fragmented Data

Modern scientific research and applications very often encounter data from different data sources,and for each data source,various covariates can be generated for statistical analysis.Unfortunately,these different data sources as well as the associated covariates are usually not available for every subject.Such kind of "fragmented data" is now becoming more and more popular and calls for new statistical analysis methodologies.In this pa-per,we propose a novel method based on model averaging that fits each candidate model by all available data and selects weights is rigorously proved.The finite sample perfor-mance of the proposed method is future confirmed by extensive Monte Carlo simulation studies.A real example for personal income prediction based on data from a leading e-community of wealth management in Shanghai is also presented to illustrate the practical usage of our method.