|
Abstract:
|
Missing data is a common problem for sample surveys that, if ignored, results in increased variance and likely bias for survey estimates. Data values based upon small samples or that have large error may also adversely affect the variance for survey estimates. Imputation, in which values are assigned for missing and "weak" data, is one approach commonly taken to reduce the variance and bias for survey estimates. Imputation for price data is typically carried out through the use of models reflecting change in prices over time. However, if no historical data are available (e.g., initial collection, for rare items), alternative models are required. Approaches for modeling prescription drug prices using related prices, both from sample and from administrative sources, will be presented, performance of the models will be reviewed, and implications for future direction will be discussed.
|
