Effects of Model Misspecification of Synthetic Dataon Estimation in a Matrix-Variate Multiple Linear Regression Model

  • John A. Zylstra Department of Mathematics and Statistics and Department of Statistics, University of Maryland, Baltimore County (UMBC), Baltimore, Maryland, USA
Keywords: imputation, disclosure control, privacy protection, Synthetic data


Consequences of model misspecification of multiply-imputed synthetic data generated from a matrix-variate multiple linear regression model via posterior predictive sampling are explored. Through case analysis across combinations of fully- or under-specified models imposed on the actual and synthetic data, accuracy of variance estimates from the synthetic data literature is evaluated when the synthetic data user’s point estimate is unbiased. The accuracy of variance estimates is a function of prior parameters and order relations are explored for informative parameter values.


Download data is not yet available.