General Methods and Algorithms for Modeling and Imputing Discrete Data under a Variety of Constraints

Loglinear modeling methods have become quite straightforward to apply to discrete data X. The models for missing data involve minor extensions of hot-deck methods (Little and Rubin 2002). Edits are structural zeros that forbid certain patterns. Winkler (2003) provided the theory for connecting edit with imputation. In this paper, we give methods and algorithms for modeling/edit/imputation under linear and convex constraints. The methods can be used for statistical matching (D’Orazio, Di Zio, and Scanu 2006), edit/imputation in which models are also controlled for external constraints such as in benchmark data, and for creating synthetic data with significantly reduced risk of re-identification (Winkler 2007).