A bar graph and its look - should I add titles, where should values go etc

Question

I didn't use r but I have recently decided to use it to plot graphs because of its great capability to do so. 'd like to plot a graph that shows voter turnout in elections. I know little about proper (i.e. correct in regard of their look) graphs so I found in the Internet one that looks (to me) fine. Here's this graph:

Here's my graph. You can see that it is far from being good (screenshot is from R-Fiddle and I think the graph may look different when compiled from desktop R).

What should I do to make it better (e.g. readible)? Specifically, do I need titles for x and y axis? Does it look better if the values are on top of bars (like in my graph) or within bar (the graph found in the Internet)?

My code is:

# Load packages
library(ggplot2)
library(scales)
# Create dataset
dat <- data.frame(years = c("1991", "1993", "1997", "2001", "2005", "2007", "2011", "2015"),
freq = c(43.20, 52.13, 47.93, 46.29, 40.57, 53.88, 48.92, 50.92))
# Create graph bar
ggplot(dat, aes(years, freq)) +   geom_bar(stat = "identity", width=0.55)
+ geom_text(aes(label=comma(freq), y=freq+1.1))  
+ scale_y_continuous(breaks = seq(0,50,10)) +  theme_classic()

EDIT:

I tried to incorporate as many suggestions in comments and answers as I could. I've come up with the following:

What do you think of it?

Answer 1

I agree with EdM's point that "bar plots simply have too much ink for the information conveyed." Here's a ggplot2 version of his answer:

library(ggplot2)

df <- data.frame(years=c(1991, 1993, 1997, 2001, 2005, 2007, 2011, 2015),
                 freq=c(43.20, 52.13, 47.93, 46.29, 40.57, 53.88, 48.92, 50.92))

p <- (ggplot(df, aes(x=years, y=freq)) +
      geom_line(size=1.25, color="#999999") + geom_point(size=3.5, color="black") +
      theme_bw() +
      theme(panel.border=element_blank(), panel.grid.minor=element_blank(),
            axis.title.y=element_text(vjust=1.25)) +
      scale_x_continuous("", breaks=seq(1990, 2015, 5), minor_breaks=NULL) +
      scale_y_continuous("percentage turnout", limits=c(36, 59),
                         breaks=seq(40, 55, 5), minor_breaks=NULL))
p
ggsave("percentage_turnout_over_time.png", p, width=10, height=8)

Which produces this:

Edit: here's a version with numbers on the graph:

p <- (ggplot(df, aes(x=years, y=freq, label=freq)) +
      geom_line(size=1.25, color="#999999") + geom_point(size=3.5, color="black") +
      geom_text(vjust=c(2, -1, -1.5*sign(diff(diff(df$freq))) + 0.5)) +
      theme_bw() +
      theme(panel.border=element_blank(), panel.grid.minor=element_blank(),
            axis.title.y=element_text(vjust=1.25)) +
      scale_x_continuous("", breaks=seq(1990, 2015, 5), minor_breaks=NULL) +
      scale_y_continuous("percentage turnout", limits=c(36, 59),
                         breaks=seq(40, 55, 5), minor_breaks=NULL))
p
ggsave("percentage_turnout_over_time_with_text.png", p, width=10, height=8)

Nick Cox's comment under the original post is convincing:

I see no harm in showing numbers too. People often want to read numbers off graphs just as they (should) want to read numbers off tables. Also, offering graph PLUS table in a paper would often be rejected by reviewers as too much space devoted to the same information, so hybridising graph and table is perfectly defensible.

Answer 2

Maybe "Tufte`rize" your plot:

ggplot(dat, aes(years, freq)) + 
  geom_bar(stat = "identity", width=0.55, fill="grey")+ 
  scale_y_continuous(breaks = seq(0,50,10)) + 
  geom_hline(yintercept= seq(0,50,10), col="white") + 
  theme_classic(base_size = 16) + theme(axis.ticks=element_blank()) + 
  labs(x=NULL, y=NULL) +
  ggtitle("freq per year")

or

ggplot(dat, aes(years, freq)) + 
  geom_bar(stat = "identity", width=0.55, fill="grey64")+ 
  scale_y_continuous(breaks = seq(0,65,10)) + 
  theme_classic(base_size = 18) + 
  theme(axis.text.y=element_blank(), axis.ticks=element_blank(), axis.line.y=element_blank()) + 
  labs(x=NULL, y=NULL) + 
  geom_text(aes(label=format(freq,decimal.mark = ",")), vjust=-.3, size = 4)

Answer 3

This is not a good application of a bar plot. Tufte would not be pleased, even with the allegedly "Tufte`rized" bar plots recommended in another answer. Bar plots simply have too much ink used for the information conveyed. See Tufte's website and books, and if possible attend one of his seminars on how to display data effectively.

Displaying these data requires no more than a line plot of turnout versus year. The line plot, with equal spacing of actual years along the x-axis, will also remove the implication contained in the bar plot that the observations are equally spaced in time.

Here's an example that emphasizes the changes over the range of observed values:

I find that the connecting lines between the points help to keep track of the time relationships, even though there are only 8, unequally spaced, observations. Nick Cox rightly noted in a comment that this might tend to suggest linear changes in between dates. The lines are gray, with black dots at the observations, in an attempt to de-emphasize such a suggestion.

If you are more interested in changes with respect to a baseline of 0% turnout you could adjust the y-axis limits accordingly. But have you ever had anything close to 0% turnout for an election? Also, figures are not good places to show results down to 2 decimal places. For that, use a table.

Code is all from R base graphics. You can probably do something more elegant with ggplot but I have little experience with that. First, change your "years" into numeric from text:

dat$years <- as.numeric(as.character(dat$years))

Then for the plot:

plot (freq~years,data=dat,xlim=c(1990,2016),xlab="Year",ylab="Percentage Turnout",type="l",axes=FALSE,col="gray")
points (freq~years,data=dat,pch=19)
axis(1,at=seq(1990,2020,10))
axis(2,at=seq(42,54,6))

Standard R graphics put a potentially distracting box around the entire plot, which is omitted here by the axes=FALSE specification, in the plot command that draws the gray lines. The points command then places the points. The separate specifications with axis allows control over where the tick marks are placed and labeled; R may tend to over-label a bit to some people's taste.