The purpose of this practical is to use some of R’s packages to solve a missing data problem in a ‘real’ dataset. This can be either the dataset you’ve brought to this course yourself or the ‘selfreport’ data from the mice package. The ‘selfreport’ data is used in Section 7.3 of the first edition of the book Flexible Imputation of Missing Data (van Buuren, S. (2012). Flexible Imputation of Missing Data. CRC/Chapman & Hall, FL: Boca Raton).

Suggested packages:

library(mice) # for visualisation & pmm
library(VIM) # visualisation & several hotdeck imputation methods (fast implementation)
library(simputation) # common interface for several imputation methods
library(magrittr) # for pipe operator
library(naniar) # additional package for practical
library(visdat) # visualisation of datasets
library(UpSetR) # Upset plots
library(ggplot2) # plots

Selfreport data

The selfreport data contains data from two studies, one from the mgg study, the survey dataset, where height and weight are selfreported measures (hr and wr) and one from the krul study, the calibration dataset, where the height and weight measures are both selfreported and measured (variables hm and wm). The goal for the self-report data is to obtain correct estimates of the prevalence of obesity (Body Mass Index > 32) in the survey dataset. Note that the Body Mass Index (bmi) is computed as follows:

\[\text{bmi} = \frac{\text{weight in kg}}{(\text{height in cm})^2} \]

  1. Check missing data patterns.
  1. Check the relation between measured and self reported bmi in the calibration dataset.
  1. Impute missing data.
  1. What is the estimated prevalence (selfreported and imputed data) of obesity for different age and sex in the survey dataset?

End of practical.