"Much Ado about Nothing?": Dealing with Missing Values in Hierarchical Data

Free software simplifies statistical process for multiple Imputation of missing data

In empirical educational research, students’ compiled data are frequently not complete, instead they exhibit missing values. This is partly intended, because it makes it easier to divide up a large questionnaire among several people. Such an incomplete set of data can result in less accurate or even distorted conclusions, if the missing values are not treated properly.

The multiple imputation procedure constitutes a general approach to handling missing data, in which different plausible substitutions for missing values can be generated based on a statistical models. The problem at hand is: If the statistical model incorrectly specified, then the multiple imputation will also result in distortions. This problem is especially relevant if the data is hierarchically structured (e.g. students in schools), because both the structure of the data as well as the complexity of the issues have to be taken into consideration in the applied model.

The article is free of charge and is published in SAGE Open:

Grund, S., Lüdtke, O., & Robitzsch, A. (2016). Multiple imputation of multilevel missing data: An introduction to the R package pan. SAGE Open, 6(4), 1–17. doi: 10.1177/2158244016668220
http://sgo.sagepub.com/content/6/4/2158244016668220

The software packet is also free of charge from either CRAN or gitHub:

https://cran.r-project.org/package=mitml
https://github.com/simongrund1/mitml