Workshop 3

Multiple Imputation of Missing Data in Simple and More Complex Settings


Nicole Erler

Department of Biostatistics, Erasmus Medical Center, Rotterdam
 

Abstract

Missing values can occur in almost any type of study but are not always appropriately handled. This one-day workshop gives an introduction into the challenges faced when working with incomplete data and discusses methods and their implementation in the analysis of datasets with missing values in covariates.

Different approaches to multiple imputation (MI) are introduced, with a focus on the currently most popular method of multiple imputation using chained equations (MICE)[1], (also known as fully conditional specification). The use of the R package “mice”[2] to perform such imputation is demonstrated and practical sessions allow the participants to test it out themselves.

Even though MICE works well in many settings, there are data structures in which its underlying assumptions are violated. Such more complex settings include the analysis of longitudinal or otherwise clustered data and survival analysis, but also models in which non-linear associations between outcome and covariates or interaction effects between covariates are assumed.

Particularly in these settings where standard methods may fail, the use of Bayesian methods has been demonstrated to be advantageous[4]. One of the Bayesian approaches that allows analysis of incomplete data with the afore mentioned features is available in the R package “JointAI”[4]. The theory behind this approach will be outlined and the use of the R package will be demonstrated and explored in practical sessions.

 

Die Workshops finden am Sonntag, den 15.09.2019 jeweils von 9:00 bis 17:00 Uhr statt. Die Anmeldung zu den Workshops erfolgt bei der Anmeldung über ConfTool.

 

[1] van Buuren, S. (2018). Flexible imputation of missing data. CRC Press. Retrieved from https://stefvanbuuren.name/fimd/

[2] van Buuren, S., & Groothuis-Oudshoorn, K. (2011). MICE: Multivariate imputation by chained equations in R. Journal of Statistical Software, 45(3), 1–67. doi: 10.18637/jss.v045.i03

[3] Erler, N. S. et al. (2016). Dealing with missing covariates in epidemiologic studies: a comparison between multiple imputation and a full Bayesian approach. Statistics in Medicine, 35(17), 2955-2974. doi: 10.1002/sim.6944

[4] Erler, N. S. (2019). JointAI: Joint Analysis and Imputation of Incomplete Data (Version 0.4.0). Retrieved from https://nerler.github.io/JointAI