Applying the MIDAS Touch: An Accurate and Scalable Approach to Imputing Missing Data

12 February 2020, Version 1
This content is an early or alternative research output and has not been peer-reviewed at the time of posting.

Abstract

We propose an accurate, fast, and scalable approach to multiply imputing missing political science data, which we call MIDAS (Multiple Imputation with Denoising Autoencoders). MIDAS employs a class of unsupervised neural networks known as denoising autoencoders, which were recently developed to optimize the task of dimensionality reduction by corrupting a subset of input data and attempting to reconstruct it through a series of nonlinear transformations. We repurpose denoising autoencoders for multiple imputation by treating missing values as a special case of corrupted data and drawing imputed values from a model trained to minimize the reconstruction error on an additional portion of originally observed values. Systematic tests involving real as well as simulated data demonstrate that MIDAS produces accurate imputed values and parameter estimates and scales more efficiently to long and wide datasets than leading existing multiple imputation algorithms. We provide open-source software for implementing MIDAS in the Python programming environment.

Comments

Comments are not moderated before they are posted, but they can be removed by the site moderators if they are found to be in contravention of our Commenting Policy [opens in a new tab] - please read this policy before you post. Comments should be used for scholarly discussion of the content in question. You can find more information about how to use the commenting feature here [opens in a new tab] .
This site is protected by reCAPTCHA and the Google Privacy Policy [opens in a new tab] and Terms of Service [opens in a new tab] apply.