Statistics and Its Interface

Volume 6 (2013)

Number 3

A semi-parametric approach for imputing mixed data

Pages: 399 – 412

DOI: http://dx.doi.org/10.4310/SII.2013.v6.n3.a11

Authors

Irene B. Helenowski (Feinberg School of Medicine, Department of Preventive Medicine. Northwestern University, Chicago, Illinois, U.S.A.)

Hakan Demirtas (School of Public Health, University of Illinois at Chicago, Chicago, Illinois, U.S.A.)

Abstract

In this work, we present a semi-parametric method for imputing mixed data which allows us to relax assumptions of the general location model. This approach involves transforming continuous and binary variables to normally distributed data, imputing the data via joint modeling under the normality assumption, and back-transforming the data to their original scale. Transformation and backtransformation of the data comprise the nonparametric portion, and multiple imputation under the normality assumption constitutes the parametric portion of our method. Simulations involving generated mixed data with binary variables and with continuous variables following normal, $t$, Gamma, and mixture Gamma distributions and real data applications indicate promising results, leading us to recommend our approach as a possible avenue for imputing mixed data by semi-parametric means.

Full Text (PDF format)