A flexible and efficient algorithm for joint imputation of general data

TL;DR

The paper introduces GERBIL, a novel algorithm for data imputation that efficiently handles complex data structures by leveraging a latent joint normal model, outperforming existing methods in accuracy and speed.

Contribution

It develops a new joint modeling-based imputation method, GERBIL, that is flexible, efficient, and suitable for high-dimensional, mixed-structure data.

Findings

GERBIL performs well compared to FCS-based methods.

GERBIL is computationally more efficient.

GERBIL effectively handles high-dimensional, mixed-structure data.

Abstract

Imputation of data with general structures (e.g., data with continuous, binary, unordered categorical, and ordinal variables) is commonly performed with fully conditional specification (FCS) instead of joint modeling. A key drawback of FCS is that it does not invoke an appropriate data augmentation mechanism and as such convergence of the resulting Markov chain Monte Carlo procedure is not assured. Methods that use joint modeling lack these drawbacks but have not been efficiently implemented in data of general structures. We address these issues by developing a new method, the so-called GERBIL algorithm, that draws imputations from a latent joint multivariate normal model that underpins the generally structured data. This model is constructed using a sequence of flexible conditional linear models that enables the resulting procedure to be efficiently implemented on high dimensional datasets in practice. Simulations show that GERBIL performs well when compared to those that utilize FCS. Furthermore, the new method is computationally efficient relative to existing FCS procedures.