Background on Multiply Imputed Synthetic Datasets

Abstract

In 1993, the Journal of Official Statistics published a special issue on data confidentiality. Two articles in this volume laid the foundation for the development of multiply imputed synthetic datasets (MISDs). In his discussion “Statistical Disclosure Limitation,” Rubin (1993) for the first time suggested generating synthetic datasets based on his ideas of multiple imputation for missing values (Rubin, 1987). He proposed to treat all the observations from the sampling frame that are not part of the sample as missing data and to impute them according to the multiple imputation framework. Afterwards, simple random samples from these fully imputed datasets should be released to the public. Because the released dataset does not contain any real data, disclosure of sensitive information is very difficult. On the other hand, if the imputation models are selected carefully and the predictive power of the models is high, most of the information contained in the original data will be preserved. This approach is now called generating fully synthetic datasets in the literature.