Year: 2019
Type: article
Abstract: Multiple imputation is a tool for parameter estimation and inference with partially observed data, which is used increasingly widely in medical and social research.When the data to be imputed are correlated or have a multilevel structure -repeated observations on patients, school children nested in classes within schools within educational districts -the imputation model needs to include this structure.Here we introduce our joint modelling package for multiple imputation of multilevel data, jomo, which uses a multivariate normal model fitted by Markov Chain Monte Carlo (MCMC).Compared to previous packages for multilevel imputation, e.g.pan, jomo adds the facility to (i) handle and impute categorical variables using a latent normal structure, (ii) impute level-2 variables, and (iii) allow for cluster-specific covariance matrices, including the option to give them an inverse-Wishart distribution at level 2. The package uses C routines to speed up the computations and has been extensively validated in simulation studies both by ourselves and others.
Source: The R Journal
DOI: 10.32614/rj-2019-028
Authors: Matteo Quartagno, Simon Grund, James Carpenter
Institution: London School of Hygiene & Tropical Medicine
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W2970547260
