Circular barplot and distortion

A collection of common dataviz caveats by Data-to-Viz.com




The issue

Circular barplots are a good alternative to classic barplots. It is basically the same thing with each bar displayed along a circle instead of a line, resulting in a more eye-catching figure.


However, circular barplots are less efficient in representing data accurately. Indeed, it is harder to compare bar lengths since they are not on the same line anymore. Moreover, the inner base of the bar is often thinner than the outer part. This results in a bias: group with large values appear even larger than they are in reality.


This bias is even more important if:

  • left - the number of bars is low.
  • right - the inner radius of the circle is too small.
# package
library(tidyverse)
library(viridis)
library(patchwork)

# Create dataset
data=data.frame(
  individual=paste( "Mister ", seq(1,9), sep=""),
  group=c( rep('A', 3), rep('C', 3), rep('D', 3)) ,
  value=sample( seq(10,100), 9, replace=T)
)
data = data %>% arrange(group, value)

# Set a number of 'empty bar' to add at the end of each group
empty_bar=1
to_add = data.frame( matrix(NA, empty_bar*nlevels(data$group), ncol(data)) )
colnames(to_add) = colnames(data)
to_add$group=rep(levels(data$group), each=empty_bar)
data=rbind(data, to_add)
data=data %>% arrange(group)
data$id=seq(1, nrow(data))

# Get the name and the y position of each label
label_data=data
number_of_bar=nrow(label_data)
angle= 90 - 360 * (label_data$id-0.5) /number_of_bar     # I subtract 0.5 because the letter must have the angle of the center of the bars. Not extreme right(1) or extreme left (0)
label_data$hjust<-ifelse( angle < -90, 1, 0)
label_data$angle<-ifelse(angle < -90, angle+180, angle)


# Make the plot
p1 = ggplot(data, aes(x=as.factor(id), y=value, fill=group)) +       # Note that id is a factor. If x is numeric, there is some space between the first bar

  geom_bar(aes(x=as.factor(id), y=value, fill=group), stat="identity", alpha=0.5) +
  scale_fill_viridis(discrete=TRUE) +
  ylim(-80,120) +
  theme_minimal() +
  theme(
    legend.position = "none",
    axis.text = element_blank(),
    axis.title = element_blank(),
    panel.grid = element_blank(),
    plot.margin = unit(rep(-1,4), "cm")
  ) +
  coord_polar() +
  geom_text(data=label_data, aes(x=id, y=value+10, label=individual, hjust=hjust), color="black", fontface="bold",alpha=0.6, size=2.5, angle= label_data$angle, inherit.aes = FALSE )



# Create dataset 2
data=data.frame(
  individual=paste( "Mister ", seq(1,60), sep=""),
  group=c( rep('A', 20), rep('C', 24), rep('D', 16)) ,
  value=sample( seq(10,100), 60, replace=T)
)
data = data %>% arrange(group, value)

# Set a number of 'empty bar' to add at the end of each group
empty_bar=1
to_add = data.frame( matrix(NA, empty_bar*nlevels(data$group), ncol(data)) )
colnames(to_add) = colnames(data)
to_add$group=rep(levels(data$group), each=empty_bar)
data=rbind(data, to_add)
data=data %>% arrange(group)
data$id=seq(1, nrow(data))

# Get the name and the y position of each label
label_data=data
number_of_bar=nrow(label_data)
angle= 90 - 360 * (label_data$id-0.5) /number_of_bar     # I subtract 0.5 because the letter must have the angle of the center of the bars. Not extreme right(1) or extreme left (0)
label_data$hjust<-ifelse( angle < -90, 1, 0)
label_data$angle<-ifelse(angle < -90, angle+180, angle)


# Make the plot
p2 <- ggplot(data, aes(x=as.factor(id), y=value, fill=group)) +       # Note that id is a factor. If x is numeric, there is some space between the first bar

  geom_bar(aes(x=as.factor(id), y=value, fill=group), stat="identity", alpha=0.5) +
  scale_fill_viridis(discrete=TRUE) +
  ylim(-10,120) +
  theme_minimal() +
  theme(
    legend.position = "none",
    axis.text = element_blank(),
    axis.title = element_blank(),
    panel.grid = element_blank(),
    plot.margin = unit(rep(-1,4), "cm")
  ) +
  coord_polar() +
  geom_text(data=label_data, aes(x=id, y=value+10, label=individual, hjust=hjust), color="black", fontface="bold",alpha=0.6, size=2.5, angle= label_data$angle, inherit.aes = FALSE )


p1 + p2

Solving the issue

Circular barcharts can be a good option in very specific conditions. You need:

  • a large number of bars
  • a clear pattern in the data: if all groups have almost the same value, a circular version is not accurate enough for comparison.
  • to keep a large inner circle
# package
library(tidyverse)
library(viridis)

# Create dataset
data=data.frame(
  individual=paste( "Mister ", seq(1,60), sep=""),
  group=c( rep('A', 20), rep('C', 24), rep('D', 16)) ,
  value=sample( seq(10,100), 60, replace=T)
)
data = data %>% arrange(group, value)

# Set a number of 'empty bar' to add at the end of each group
empty_bar=1
to_add = data.frame( matrix(NA, empty_bar*nlevels(data$group), ncol(data)) )
colnames(to_add) = colnames(data)
to_add$group=rep(levels(data$group), each=empty_bar)
data=rbind(data, to_add)
data=data %>% arrange(group)
data$id=seq(1, nrow(data))

# Get the name and the y position of each label
label_data=data
number_of_bar=nrow(label_data)
angle= 90 - 360 * (label_data$id-0.5) /number_of_bar     # I subtract 0.5 because the letter must have the angle of the center of the bars. Not extreme right(1) or extreme left (0)
label_data$hjust<-ifelse( angle < -90, 1, 0)
label_data$angle<-ifelse(angle < -90, angle+180, angle)


# Make the plot
p = ggplot(data, aes(x=as.factor(id), y=value, fill=group)) +       # Note that id is a factor. If x is numeric, there is some space between the first bar

  geom_bar(aes(x=as.factor(id), y=value, fill=group), stat="identity", alpha=0.5) +
  scale_fill_viridis(discrete=TRUE) +
  ylim(-80,120) +
  theme_minimal() +
  theme(
    legend.position = "none",
    axis.text = element_blank(),
    axis.title = element_blank(),
    panel.grid = element_blank(),
    plot.margin = unit(rep(-1,4), "cm")
  ) +
  coord_polar() +
  geom_text(data=label_data, aes(x=id, y=value+10, label=individual, hjust=hjust), color="black", fontface="bold",alpha=0.6, size=2.5, angle= label_data$angle, inherit.aes = FALSE )

p

Note: don’t forget to add a legend or annotation for the colors and a scale for the bar height. See more in the circular barplot section.


Dataviz decision tree

Data To Viz is a comprehensive classification of chart types organized by data input format. Get a high-resolution version of our decision tree delivered to your inbox now!


High Resolution Poster
 

A work by Yan Holtz for data-to-viz.com