Old friends

Create the ggmice equivalent of mice plots

How to re-create the output of the plotting functions from mice with ggmice. In alphabetical order of the mice functions.

First load the ggmice, mice, and ggplot2 packages, some incomplete data and a mids object into your workspace.

# load packages
library(ggmice)
library(mice)
library(ggplot2)
# load incomplete dataset from mice
dat <- boys
# generate imputations
imp <- mice(dat, method = "pmm", printFlag = FALSE)

bwplot

Box-and-whisker plot of observed and imputed data.

# original plot
mice::bwplot(imp, hgt ~ .imp)

# ggmice equivalent
ggmice(imp, aes(x = .imp, y = hgt)) +
  geom_boxplot() +
  labs(x = "Imputation number")

# extended reproduction with ggmice
ggmice(imp, aes(x = .imp, y = hgt)) +
  stat_boxplot(geom = "errorbar", linetype = "dashed") +
  geom_boxplot(outlier.colour = "grey", outlier.shape = 1) +
  labs(x = "Imputation number") +
  theme(legend.position = "none")

densityplot

Density plot of observed and imputed data.

# original plot
mice::densityplot(imp, ~hgt)

# ggmice equivalent
ggmice(imp, aes(x = hgt, group = .imp)) +
  geom_density()

# extended reproduction with ggmice
ggmice(imp, aes(x = hgt, group = .imp, size = .where)) +
  geom_density() +
  scale_size_manual(
    values = c("observed" = 1, "imputed" = 0.5),
    guide = "none"
  ) +
  theme(legend.position = "none")

fluxplot

Influx and outflux plot of multivariate missing data patterns.

# original plot
fluxplot(dat)

# ggmice equivalent
plot_flux(dat)

md.pattern

Missing data pattern plot.

# original plot
md <- md.pattern(dat)

# ggmice equivalent
plot_pattern(dat)

# extended reproduction with ggmice
plot_pattern(dat, square = TRUE) +
  theme(
    legend.position = "none",
    axis.title = element_blank(),
    axis.title.x.top = element_blank(),
    axis.title.y.right = element_blank()
  )

plot.mids

Plot the trace lines of the MICE algorithm.

# original plot
plot(imp, hgt ~ .it | .ms)

# ggmice equivalent
plot_trace(imp, "hgt")

stripplot

Stripplot of observed and imputed data.

# original plot
mice::stripplot(imp, hgt ~ .imp)

# ggmice equivalent
ggmice(imp, aes(x = .imp, y = hgt)) +
  geom_jitter(width = 0.25) +
  labs(x = "Imputation number")

# extended reproduction with ggmice (not recommended)
ggmice(imp, aes(x = .imp, y = hgt)) +
  geom_jitter(
    shape = 1,
    width = 0.1,
    na.rm = TRUE,
    data = data.frame(
      hgt = dat$hgt,
      .imp = factor(rep(1:imp$m, each = nrow(dat))),
      .where = "observed"
    )
  ) +
  geom_jitter(shape = 1, width = 0.1) +
  labs(x = "Imputation number") +
  theme(legend.position = "none")

xyplot

Scatterplot of observed and imputed data.

# original plot
mice::xyplot(imp, hgt ~ age)

# ggmice equivalent
ggmice(imp, aes(age, hgt)) +
  geom_point()

# extended reproduction with ggmice
ggmice(imp, aes(age, hgt)) +
  geom_point(size = 2, shape = 1) +
  theme(legend.position = "none")

Extensions

Interactive plots

To make ggmice visualizations interactive, the plotly package can be used. For example, an interactive influx and outflux plot may be more legible than a static one.

# load packages
library(plotly)
# influx and outflux plot
p <- plot_flux(dat)
ggplotly(p)

Plot multiple variables

You may want to create a plot visualizing the imputations of multiple variables as one object. To visualize multiple variables at once, the variable names are saved in a vector. This vector is used together with the functional programming package purrr and visualization package patchwork to map() over the variables and subsequently wrap_plots to create a single figure.

# load packages
library(purrr)
library(patchwork)
# create vector with variable names
vrb <- names(dat)

Display box-and-whisker plots for all variables.

# original plot
mice::bwplot(imp)