Linking, Pooling, and now Fusion:
does this lead to conFusion?
Presented by
Bob Baskin, CFACT
Date/Location March 31, 2014
10:30 a.m. – Noon
Rock Creek Conference Room
National surveys such as the Medical Expenditure Panel Survey are designed to produce national and regional level estimates but analytical questions are often at the level of small subdomains for which the surveys were not designed to produce estimates. In order to address subdomain questions three approaches have been used. First, data from the Medical Expenditure Panel Survey has been linked to other data through the National Health Interview Survey. Second, multiple years of the Medical Expenditure Panel Survey have been pooled. Third, modelling and simulation have been employed. For example, some state level estimates have been produced from the Medical Expenditure Panel Survey by combining pooling and modelling. In recent work, “Out of Sample Fusion in Risk Prediction”, a method called data fusion was applied to biological data to improve confidence intervals for threshold levels. The current work attempts to apply the fusion method to the Medical Expenditure Panel Survey. This method of fusion can be used with one MEPS year as base year and other MEPS years as fusion data and the results compared with straight pooling. Also single years of MEPS data can be fused with simulated data. Eventually the object of this approach would be to apply fusion of MEPS with external sources such as Medicare Current Benefits Survey or claims data.