Beautiful plotting in R: A ggplot2 cheatsheet

Even the most experienced R users need help creating elegant graphics. The ggplot2 library is a phenomenal tool for creating graphics in R but even after many years of near-daily use we still need to refer to our Cheat Sheet. Up until now, we’ve kept these key tidbits on a local PDF. But for our own benefit (and hopefully yours) we decided to post the most useful bits of code.

* Last updated January 20, 2016 (with ggplot2 2.0 replace vjust with margin for title text)

You may also be interested in these other ggplot2-related posts:

Under the hood of ggplot2 graphics in R
Mapping in R using the ggplot2 package
A new data processing workflow for R: dplyr, magrittr, tidyr and ggplot2

We start with the the quick setup and a default plot followed by a range of adjustments below.

Quick-setup: The dataset

We’re using data from the National Morbidity and Mortality Air Pollution Study (NMMAPS). To make the plots manageable we’re limiting the data to Chicago and 1997-2000. For more detail on this dataset, consult Roger Peng’s book Statistical Methods in Environmental Epidemiology with R.

You can also download the data we’re using in this post here.

library(ggplot2)
nmmaps<-read.csv("chicago-nmmaps.csv", as.is=T)
nmmaps$date<-as.Date(nmmaps$date)
nmmaps<-nmmaps[nmmaps$date>as.Date("1996-12-31"),]
nmmaps$year<-substring(nmmaps$date,1,4)
head(nmmaps)
##      city       date death temp dewpoint   pm10     o3 time season year
## 3654 chic 1997-01-01   137 36.0    37.50 13.052  5.659 3654 winter 1997
## 3655 chic 1997-01-02   123 45.0    47.25 41.949  5.525 3655 winter 1997
## 3656 chic 1997-01-03   127 40.0    38.00 27.042  6.289 3656 winter 1997
## 3657 chic 1997-01-04   146 51.5    45.50 25.073  7.538 3657 winter 1997
## 3658 chic 1997-01-05   102 27.0    11.25 15.343 20.761 3658 winter 1997
## 3659 chic 1997-01-06   127 17.0     5.75  9.365 14.941 3659 winter 1997

A default plot in ggplot2

g<-ggplot(nmmaps, aes(date, temp))+geom_point(color="firebrick")
g
image1sm

Back to table of contents

Working with the title

Add a title (ggtitle() or labs())

g<-g+ggtitle('Temperature')
g
image2sm

Alternatively, you can use g+labs(title='Temperature')

Back to table of contents

Make title bold and add a little space at the baseline (face, margin)

In ggplot2 versions before 2.0 I used the vjust argument to move the title away from the plot. With 2.0 this no longer works and a blog comment (below) helped me identify an alternative using this link. The margin argument uses the margin function and you provide the top, right, bottom and left margins (the default unit is points).

g+theme(plot.title = element_text(size=20, face="bold", 
    margin = margin(10, 0, 10, 0)))
image3sm

Back to table of contents

Use a non-traditional font in your title (family)

Note that you can also use different fonts. It’s not as easy as it seems here, check out this post if you need to use different fonts. This may not work on a Mac (send me a note and let me know).

If you are having trouble with this you might take a look at this StackOverflow discussion.

library(extrafont)
g+theme(plot.title = element_text(size=30,lineheight=.8, 
  vjust=1,family="Bauhaus 93"))
image4sm

Back to table of contents

Change spacing in multi-line text (lineheight)

You can use the lineheight argument to change the spacing between lines. In this example, I’ve squished the lines together a bit (lineheight < 1).

g<-g+ggtitle("This is a longer\ntitle than expected")
g+theme(plot.title = element_text(size=20, face="bold", vjust=1, lineheight=0.6))
image5sm

Back to table of contents

Working with axes

Add x and y axis labels (labs(), xlab())

The easiest is:

g<-g+labs(x="Date", y=expression(paste("Temperature ( ", degree ~ F, " )")), title="Temperature")
g
image6sm

Back to table of contents

Get rid of axis ticks and tick text (theme(), axis.ticks.y)

I wouldn’t normally do this but demonstration purposes:

g + theme(axis.ticks.y = element_blank(),axis.text.y = element_blank())
image7sm

Back to table of contents

Change size of and rotate tick text (axis.text.x)

Go ahead, try to say ‘tick text’ three times fast.

g + theme(axis.text.x=element_text(angle=50, size=20, vjust=0.5))
image8sm

Back to table of contents

Move the labels away from the plot (and add color) (theme(), axis.title.x)

I find that the labels are too close to the plot in the default settings so, similar to with the title, I’m using the vjust argument.

g + theme(
      axis.title.x = element_text(color="forestgreen", vjust=-0.35),
      axis.title.y = element_text(color="cadetblue" , vjust=0.35)   
          )
image9sm

Back to table of contents

Limit an axis to a range (ylim(), scale_x_continuous(), coord_cartesian())

Again, this plot is for demonstration purposes:

g + ylim(c(0,60))
image10sm

Alternatively: g+scale_x_continuous(limits=c(0,35)) or g+coord_cartesian(xlim=c(0,35)). The former removes all data points outside the range and second adjusts the visible area.

Back to table of contents

If you want the axes to be the same (coord_equal())

There must be a better way than this. In the example, I’m plotting temperature against temperature with some random noise (for demonstration purposes) and I want both axes to be the same scale/same range.

ggplot(nmmaps, aes(temp, temp+rnorm(nrow(nmmaps), sd=20)))+geom_point()+
  xlim(c(0,150))+ylim(c(0,150))+
  coord_equal()
image11sm

Back to table of contents

Use a function to alter labels (label=function(x){})

Sometimes it’s handy to alter your labels a little, perhaps adding units or percent signs without adding them to your data. You can use a function in this case. Here is an example:

ggplot(nmmaps, aes(date, temp))+
  geom_point(color="grey")+
  labs(x="Month", y="Temp")+
  scale_y_continuous(label=function(x){return(paste("My value is", x, "degrees"))})
image12sm

Not pretty here, but this can come in handy.

Back to table of contents

Working with the legend

We will color code the plot based on season. You can see that by default the legend title is what we specified in the color argument.

g<-ggplot(nmmaps, aes(date, temp, color=factor(season)))+geom_point()
g
image13sm

Back to table of contents

Turn off the legend title (legend.title)

g+theme(legend.title=element_blank())
image14sm

Back to table of contents

Change the styling of the legend title (legend.title)

g+theme(legend.title = element_text(colour="chocolate", size=16, face="bold"))
image15sm

Back to table of contents

Change the title of the legend (name)

To change the title of the legend you would use the name argument in your scale function. If you don’t use a `scale` function you will need to change the data itself so that it has the right format.

g+theme(legend.title = element_text(colour="chocolate", size=16, face="bold"))+
  scale_color_discrete(name="This color is\ncalled chocolate!?")
image16sm

Back to table of contents

Change the background boxes in the legend (legend.key)

I have mixed feelings about those boxes. If you want to get rid of them entirely use fill=NA.

g+theme(legend.key=element_rect(fill='pink'))
image17sm

Back to table of contents

Change the size of the symbols in the legend only (guides(), guide_legend)

Points in the legend get a little lost, especially without the boxes. To override the default try:

g+guides(colour = guide_legend(override.aes = list(size=4)))
image18sm

Back to table of contents

Leave a layer off the legend (show_guide)

Let’s say you have a point layer and you add label text to it. By default, both the points and the label text end up in the legend like this (again, who would make a plot like this? It’s for demonstration purposes):

g+geom_text(data=nmmaps, aes(date, temp, label=round(temp)), size=4)
image19sm

You can use show_guide=FALSE to turn a layer off in the legend. Useful!

g+geom_text(data=nmmaps, aes(date, temp, label=round(temp), size=4), show_guide=FALSE)
image20sm

Back to table of contents

Manually adding legend items (guides(), override.aes)

ggplot2 will not add a legend automatically unless you map aethetics (color, size etc) to a variable. There are times, though, that I want to have a legend so that it’s clear what you’re plotting. Here is the default:

# here there is no legend automatically
ggplot(nmmaps, aes(x=date, y=o3))+geom_line(color="grey")+geom_point(color="red")
image21sm

We can force a legend by mapping to a “variable”. We are mapping the lines and the points using aes and we are mapping not to a variable in our dataset but to a single string (so that we get just one color for each).

ggplot(nmmaps, aes(x=date, y=o3))+geom_line(aes(color="Important line"))+
  geom_point(aes(color="My points"))
image22sm

We’re getting close but this is not what I want. I wanted grey and red. To change the color, we use scale_colour_manual().

ggplot(nmmaps, aes(x=date, y=o3))+geom_line(aes(color="Important line"))+
   geom_point(aes(color="Point values"))+
  scale_colour_manual(name='', values=c('Important line'='grey', 'Point values'='red'))
image23sm

Tantalizingly close! But we don’t want a line with a point for both. Line=grey and point=red. The final step is to override the aesthetics in the legend. The guide() function allows us to control guides like the legend:

ggplot(nmmaps, aes(x=date, y=o3))+geom_line(aes(color="Important line"))+
   geom_point(aes(color="Point values"))+
  scale_colour_manual(name='', values=c('Important line'='grey', 'Point values'='red'), guide='legend') +
  guides(colour = guide_legend(override.aes = list(linetype=c(1,0)
                                                      , shape=c(NA, 16))))
image24sm

Voila!

Back to table of contents

Working with the background colors

There are ways to change the entire look of your plot with one function (see below) but if you want to simply change the background color of the panel you can use the following:

Change the panel color (panel.background)

ggplot(nmmaps, aes(date, temp))+geom_point(color="firebrick")+
  theme(panel.background = element_rect(fill = 'grey75'))
image25sm

Back to table of contents

Change the grid lines (panel.grid.major)

ggplot(nmmaps, aes(date, temp))+geom_point(color="firebrick")+
  theme(panel.background = element_rect(fill = 'grey75'),
  panel.grid.major = element_line(colour = "orange", size=2),
  panel.grid.minor = element_line(colour = "blue"))
image26sm

Back to table of contents

Change the plot background (not the panel) color (plot.background)

ggplot(nmmaps, aes(date, temp))+geom_point(color="firebrick")+
  theme(plot.background = element_rect(fill = 'grey'))
image27sm

Back to table of contents

Working with margins

Changing the plot margin (plot.margin)

I sometimes find that I need to add a little space to one margin of my plot. Similar to the previous examples we can use an argument to the theme() function. In this case the argument is plot.margin. In order to illustrate I’m going to add a background color using plot.background so you can see the default:

# the default
ggplot(nmmaps, aes(date, temp))+
  geom_point(color="darkorange3")+
  labs(x="Month", y="Temp")+
  theme(plot.background=element_rect(fill="darkseagreen"))
image28sm

Now let’s add extra space to both the left and right. The argument, plot.margin, can handle a variety of different units (cm, inches etc) but it requires the use of the function unit from the package grid to specify the units. Here I’m using a 6 cm margin on the right and left.

library(grid)
ggplot(nmmaps, aes(date, temp))+
  geom_point(color="darkorange3")+
  labs(x="Month", y="Temp")+
  theme(plot.background=element_rect(fill="darkseagreen"),
        plot.margin = unit(c(1, 6, 1, 6), "cm")) #top, right, bottom, left
image29sm

Again, not a pretty plot!

Back to table of contents

Creating multi-panel plots

The ggplot2 package has two nice functions for creating multi-panel plots. They are related but a little different facet_wrap creates essentially a ribbon of plots based on a single variable while facet_grid can take two variables.

Create a single row of plots based on one variable (facet_wrap())

ggplot(nmmaps, aes(date,temp))+geom_point(color="aquamarine4")+facet_wrap(~year, nrow=1)
image30sm

Back to table of contents

Create a matrix of plots based on one variable (facet_wrap())

ggplot(nmmaps, aes(date,temp))+geom_point(color="chartreuse4")+
    facet_wrap(~year, ncol=2)
image31sm

Back to table of contents

Allow scales to roam free (scales)

The default for multi-panel plots in ggplot2 is to use equivalent scales in each panel. But sometimes you want to allow a panel’s own data to determine the scale. This is not often a good idea since it may give your user the wrong impression about the data but to do this you can set scales="free" like this:

ggplot(nmmaps, aes(date,temp))+geom_point(color="chartreuse4")+
  facet_wrap(~year, ncol=2, scales="free")
image32sm

Back to table of contents

Create a grid of plots using two variables (facet_grid())

ggplot(nmmaps, aes(date,temp))+geom_point(color="darkgoldenrod4")+
    facet_grid(year~season)
image33sm

Back to table of contents

Put two (potentially unrelated) plots side by side (pushViewport(), grid.arrange())

I find that doing this is not nearly as straightforward as traditional (base) graphics. Here are two approaches:

myplot1<-ggplot(nmmaps, aes(date, temp))+geom_point(color="firebrick")
myplot2<-ggplot(nmmaps, aes(temp, o3))+geom_point(color="olivedrab")

library(grid)
pushViewport(viewport(layout = grid.layout(1, 2)))
print(myplot1, vp = viewport(layout.pos.row = 1, layout.pos.col = 1))
print(myplot2, vp = viewport(layout.pos.row = 1, layout.pos.col = 2))

# alternative, a little easier
library(gridExtra)
grid.arrange(myplot1, myplot2, ncol=2)

image34sm

To change from row to column arrangement you can change facet_grid(season~year) to facet_grid(year~season).

Back to table of contents

Working with themes

You can change the entire look of the plots by using custom theme. As an example, Jeffrey Arnold has put together the library ggthemes with several custom themes. For a list you can visit the ggthemes site. Here is an example:

Use a new theme (theme_XX())

library(ggthemes)
ggplot(nmmaps, aes(date, temp, color=factor(season)))+
  geom_point()+ggtitle("This plot looks a lot different from the default")+
  theme_economist()+scale_colour_economist()
image35sm

Back to table of contents

Change the size of all plot text elements

Personally, I find default size of the tick text, legends and other elements to be a little too small. Luckily it’s incredibly easy to change the size of all the text elements at once. If you look below at the section on creating a custom theme you’ll notice that the sizes of all the elements are relative (rel()) to the base_size. As a result, you can simply change the base_size and you’re done. Here is the code:

theme_set(theme_gray(base_size = 30))
ggplot(nmmaps, aes(x=date, y=o3))+geom_point(color="red")
image36sm

Back to table of contents

Tip on creating a custom theme

If you want to change the theme for an entire session you can use theme_set as in theme_set(theme_bw()). The default is called theme_gray. If you wanted to create your own custom theme, you could extract the code directly from the gray theme and modify. Note that the rel() function change the sizes relative to the base_size.

theme_gray

function (base_size = 12, base_family = "") 
{
  theme(
    line = element_line(colour = "black", size = 0.5, linetype = 1, lineend = "butt"), 
    rect = element_rect(fill = "white", colour = "black", size = 0.5, linetype = 1), 
    text = element_text(family = base_family, face = "plain", colour = "black", size = base_size, hjust = 0.5, vjust = 0.5, angle = 0, lineheight = 0.9), 
    
    axis.text = element_text(size = rel(0.8), colour = "grey50"), 
    strip.text = element_text(size = rel(0.8)), 
    axis.line = element_blank(), 
    axis.text.x = element_text(vjust = 1), 
    axis.text.y = element_text(hjust = 1), 
    axis.ticks = element_line(colour = "grey50"), 
    axis.title.x = element_text(), 
    axis.title.y = element_text(angle = 90), 
    axis.ticks.length = unit(0.15, "cm"), 
    axis.ticks.margin = unit(0.1, "cm"), 
    
    legend.background = element_rect(colour = NA), 
    legend.margin = unit(0.2, "cm"), 
    legend.key = element_rect(fill = "grey95", colour = "white"), 
    legend.key.size = unit(1.2, "lines"), 
    legend.key.height = NULL, 
    legend.key.width = NULL, 
    legend.text = element_text(size = rel(0.8)), 
    legend.text.align = NULL, 
    legend.title = element_text(size = rel(0.8), face = "bold", hjust = 0), 
    legend.title.align = NULL, 
    legend.position = "right", 
    legend.direction = NULL, 
    legend.justification = "center", 
    legend.box = NULL, 

    panel.background = element_rect(fill = "grey90", colour = NA), 
    panel.border = element_blank(), 
    panel.grid.major = element_line(colour = "white"), 
    panel.grid.minor = element_line(colour = "grey95", size = 0.25), 
    panel.margin = unit(0.25, "lines"), 
    panel.margin.x = NULL, 
    panel.margin.y = NULL, 

    strip.background = element_rect(fill = "grey80", colour = NA), 
    strip.text.x = element_text(), 
    strip.text.y = element_text(angle = -90), 
    
    plot.background = element_rect(colour = "white"), 
    plot.title = element_text(size = rel(1.2)), 
    plot.margin = unit(c(1, 1, 0.5, 0.5), "lines"), complete = TRUE)
}

Working with colors

For simple applications working with colors is straightforward in ggplot2 but when you have more advanced needs it can be a challenge. For a more advaned treatment of the topic you should probably get your hands on Hadley’s book which has nice coverage. There are a few other good sources including the R Cookbook and the ggplot2 online docs. Tian Zheng at Columbia has created a useful PDF of R colors.

In order to use color with your data, most importantly, you need to know if you’re dealing with a categorical or continuous variable.

Categorical variables: manually select the colors (scale_color_manual())

ggplot(nmmaps, aes(date, temp, color=factor(season)))+
  geom_point() + 
  scale_color_manual(values=c("dodgerblue4", "darkolivegreen4",
                              "darkorchid3", "goldenrod1"))
image37sm

Back to table of contents

Categorical variables: try a built-in palette (based on colorbrewer2.org) (scale_color_brewer()):

ggplot(nmmaps, aes(date, temp, color=factor(season)))+
  geom_point() + 
  scale_color_brewer(palette="Set1")
image38sm

How about using the Tableau colors (but you need the library ggthemes):

library(ggthemes)
ggplot(nmmaps, aes(date, temp, color=factor(season)))+
  geom_point() + 
  scale_colour_tableau()
image39sm

Back to table of contents

Color choice with continuous variables (scale_color_gradient(), scale_color_gradient2())

In our example we will change the color variable to ozone, a continuous variable that is strongly related to temperature (higher temperature = higher ozone). The function scale_color_gradient() is a sequential gradient while scale_color_gradient2() is diverging.

Here is a default continuous color scheme (sequential color scheme):

ggplot(nmmaps, aes(date, temp, color=o3))+geom_point()
image40sm
# this code produces an identical plot
#ggplot(nmmaps, aes(date, temp, color=o3))+geom_point()+scale_color_gradient()

Manually change the low and high colors (sequential color scheme):

ggplot(nmmaps, aes(date, temp, color=o3))+geom_point()+
  scale_color_gradient(low="darkkhaki", high="darkgreen")
image41sm

The temperature data is normally distributed so how about a diverging color scheme (rather than sequential). For diverging color you can use the scale_color_gradient2 function.

mid<-mean(nmmaps$o3)
ggplot(nmmaps, aes(date, temp, color=o3))+geom_point()+
scale_color_gradient2(midpoint=mid,
                      low="blue", mid="white", high="red" )
image42sm

Back to table of contents

Working with annotation

Add text annotation in the top-right, top-left etc. (annotation_custom() and textGrob())

I often have trouble figuring out the best way to add text to a plot in say, the top left corner, without using hard-coded coordinates. With ggplot2 you can set annotation coordinates to Inf but I find this only moderately useful. Here is an example (based on code from here) using the library grid that allows you to specify the location based on scaled coordinates where 0 is low and 1 is high.

The grobTree function (from grid) creates a grid graphical object and textGrob creates the text graphical object. The annotation_custom() function comes from ggplot2 and is designed to use a grob as input.

library(grid)

my_grob = grobTree(textGrob("This text stays in place!", x=0.1,  y=0.95, hjust=0,
  gp=gpar(col="blue", fontsize=15, fontface="italic")))

ggplot(nmmaps, aes(temp, o3))+geom_point(color="firebrick")+
  annotation_custom(my_grob)
image43sm

‘Big deal’ you say!? It is a big deal. The value here is particularly evident when you have multiple plots with different scales. In the plot below you see that the axis scales vary yet the same code as above can be used to put the annotation is the same place on each facet. Nice!

library(grid)

my_grob = grobTree(textGrob("This text stays in place!", x=0.1,  y=0.95, hjust=0,
  gp=gpar(col="blue", fontsize=12, fontface="italic")))

ggplot(nmmaps, aes(temp, o3))+geom_point(color="firebrick")+facet_wrap(~season, scales="free")+
  annotation_custom(my_grob)
image44sm

Working with coordinates

Flip a plot on it’s side (coord_flip())

It is incredibly easy to flip your plot on it’s side. Here I’ve added the coord_flip() which is all you need to flip the plot.

ggplot(nmmaps, aes(x=season, y=o3))+geom_boxplot(fill="chartreuse4")+coord_flip()
image45sm

Working with plot types

Alternatives to the box plot (geom_jitter() and geom_violin())

Start with a box plot

Box plots are great, but they can be so incredibly boring. There are alternatives, first – a box plot:

g<-ggplot(nmmaps, aes(x=season, y=o3))
g+geom_boxplot(fill="darkseagreen4")
image46sm

Effective, yes. Interesting, no.

Alternative to a box plot: plot of points

What if we plot the points themselves?

g+geom_point()
image47sm

Not only boring but uninformative, you could add transparency to deal with overplotting, but this is not good either. Let’s try something else.

Alternative to a box plot: jitter the points (geom_jitter())

Try adding a little jitter to the data. I like this for in-house visualization but be careful using jittering because you’re purposely adding noise to your data and this can result in misinterpretation of your data.

g+geom_jitter(alpha=0.5, aes(color=season),position = position_jitter(width = .2))
image48sm

This is better and I think since we’re working with season, a variable everyone will be familiar with, the extra noise will not likely lead to confusion.

Alternative to a box plot: violin plot perhaps (geom_violin())

Violin plots, similar to box plots except you’re using a kernel density to show where you have the most data, are a useful visualization.

g+geom_violin(alpha=0.5, color="gray")
image49sm

What if we rotated and added the jittered points.

g+geom_violin(alpha=0.5, color="gray")+geom_jitter(alpha=0.5, aes(color=season),
      position = position_jitter(width = 0.1))+coord_flip()
image50sm

This is nice and I like it. But be wary of using unusual plot types, they take more time for your users to understand. Sometimes the simplest and most conventional plot type is your best bet when sharing with others. Box plots may be boring but people know how to interpret them immediately.

Add a ribbon to your plot (geom_ribbon())

This is not the perfect dataset for this, but using ribbon can be useful. In this example we will create a 30-day running average using the filter() function so that our ribbon is not too noisy.

# add a filter
nmmaps$o3run<-as.numeric(filter(nmmaps$o3, rep(1/30,30), sides=2))
ggplot(nmmaps, aes(date, o3run))+geom_line(color="lightpink4", lwd=1)
image51sm

How does it look if we fill in the area below the curve using the geom_ribbon() function?

# add a filter

ggplot(nmmaps, aes(date, o3run))+geom_ribbon(aes(ymin=0, ymax=o3run), fill="lightpink3", color="lightpink3")+
  geom_line(color="lightpink4", lwd=1)
image52sm

Above is not really the conventional way to use geom_ribbon(). Instead, why don’t we draw a ribbon that gives us one standard deviation above and below our data:

nmmaps$mino3<-nmmaps$o3run-sd(nmmaps$o3run, na.rm=T)
nmmaps$maxo3<-nmmaps$o3run+sd(nmmaps$o3run, na.rm=T)

ggplot(nmmaps, aes(date, o3run))+geom_ribbon(aes(ymin=mino3, ymax=maxo3), fill="steelblue2", color="steelblue2")+
  geom_line(color="steelblue4", lwd=1)
image53sm

Create a tiled correlation plot (geom_tile())

I’ll admit that I find creating tiled correlation plots a bit cumbersome, I always have to copy and paste code from a previous project. Nevertheless, it’s a useful plot type so I’m posting the code here.

First step is to create the correlation matrix. I’m using Pearson because all the variables are fairly normally distributed – you may want to consider Spearman if your variables follow a different pattern. Note that since a correlation matrix has redundant information I’m setting half of it to NA.

#careful, I'm sorting the field names so that the ordering in the final plot is correct
thecor<-round(cor(nmmaps[,sort(c("death", "temp", "dewpoint", "pm10", "o3"))], method="pearson", use="pairwise.complete.obs"),2)
thecor[lower.tri(thecor)]<-NA
thecor
##          death dewpoint    o3 pm10  temp
## death        1    -0.47 -0.24 0.00 -0.49
## dewpoint    NA     1.00  0.45 0.33  0.96
## o3          NA       NA  1.00 0.21  0.53
## pm10        NA       NA    NA 1.00  0.37
## temp        NA       NA    NA   NA  1.00

Now I’m going to put it in “long” format using the melt function from the reshape2 package and I’ll drop the records with NA values.

library(reshape2)

thecor<-melt(thecor)
thecor$Var1<-as.character(thecor$Var1)
thecor$Var2<-as.character(thecor$Var2)
thecor<-na.omit(thecor)
head(thecor)
##        Var1     Var2 value
## 1     death    death  1.00
## 6     death dewpoint -0.47
## 7  dewpoint dewpoint  1.00
## 11    death       o3 -0.24
## 12 dewpoint       o3  0.45
## 13       o3       o3  1.00

Now for the plot. I’m using geom_tile but if you have a lot of data you might consider geom_raster which can be much faster.

ggplot(thecor, aes(Var2, Var1))+
 geom_tile(data=thecor, aes(fill=value), color="white")+
 scale_fill_gradient2(low="blue", high="red", mid="white", 
  midpoint=0, limit=c(-1,1),name="Correlation\n(Pearson)")+
 theme(axis.text.x = element_text(angle=45, vjust=1, size=11, hjust=1))+
 coord_equal()
image54sm

Back to table of contents

Working with smooths

You’ve likely already learned how amazingly easy it is to add a smooth to your data using ggplot2. You can simply use stat_smooth() which will add a LOESS smooth if you have fewer than 1000 points or a GAM otherwise. Since we have more than 1000 points the smooth is a GAM.

Default – adding LOESS or GAM (stat_smooth())

Here it is at its simplest – no formula required:

ggplot(nmmaps, aes(date, temp))+geom_point(color="firebrick")+
  stat_smooth()
image55sm

Specifying the formula (stat_smooth(formula=))

But ggplot2 allows you to specify the model you want it to use. For example, let’s say you want to increase the GAM dimension (add some additional wiggles to the smooth):

ggplot(nmmaps, aes(date, temp))+
  geom_point(color="grey")+
  stat_smooth(method="gam", formula=y~s(x,k=10), col="darkolivegreen2", se=FALSE, size=1)+
  stat_smooth(method="gam", formula=y~s(x,k=30), col="red", se=FALSE, size=1)+
  stat_smooth(method="gam", formula=y~s(x,k=500), col="dodgerblue4", se=FALSE, size=1)
image56sm

Adding a linear fit (stat_smooth(method="lm"))

Though the default is a smooth, it is also easy to add a standard linear fit:

ggplot(nmmaps, aes(temp, death))+geom_point(color="firebrick")+
  stat_smooth(method="lm", se=FALSE)
image57sm

Note that the same could be achieved using the more cumbersome:

lmTemp<-lm(death~temp, data=nmmaps)
ggplot(nmmaps, aes(temp, death))+geom_point(col="firebrick")+
  geom_abline(intercept=lmTemp$coef[1], slope=lmTemp$coef[2])

Convert your plots to online, interactive graphics using Plot.ly

Plot.ly is a great tool for easily creating online, interactive graphics directly from your ggplot2 plots. The process is surprisingly easy, and can be done from within R, but there are enough steps that I describe how to create graphics like the one below in a separate post.

Note that I have been using a free plot.ly account for this one plot which limits to 500 views per day. After 500 views plot.ly sends an incredible number of e-mails about the exceedance (sometimes more than 15 per day). So for now I have turned off the plot below and you’ll have to imagine it!


This post created in RStudio using knitr. R version 3.0.2 (2013-09-25) and ggplot2 version 0.9.3.1.

56 responses

  1. Hi Zev,

    Awesome job! One of the most comprehensive tutorials on ggplot2 I’ve ever come through. By the way, how can you increase the size of x- & y-axis legend? Sorry if I missed it in your post. Thank you so much!

    Cheers,

  2. Hi, this looks really good, but I am not sure how to get hold of the data you are using. Any hints apart from consulting Dr Peng’s book?

  3. A great post; but a small mistake:

    It should be: stat_smooth(method=”lm”, se=False) instead of: stat_smooth(method=”lm”, se=FALSE)

    in the section adding a linear fit.

    • No “FALSE” is correct, not “False” try it:

      x<-data.frame(a=1:10, b=1:10) ggplot(x, aes(a, b))+geom_point(color="firebrick")+ stat_smooth(method="lm", se=FALSE)

  4. I have ~ 14,000 points for my XY scatter, so I have to necessarily use low alpha=0.025 or it will become too dark. However, when I want to display just 7 points of interest, they are not visible because of the low alpha. I am assuming that it is possible to retain alpha for the XY scatter points, and change alpha for just my points of interest? I just dont know how though, could you help please? Thanks!
    My syntax so far, using the awesome examples on your website, looks like this:
    >input=read.table(“PfamAseeds_SeqIdnty_Stats_R_Annotated_ScatterPlot.txt”,sep=”\t”,header=TRUE)
    > df=as.data.frame(input)
    > summary(df)
    PfamA_ID SeqIdentityMean SeqIdentityStdDev
    :14823 Min. :0.0718 Min. :0.00000
    PF00646.28: 1 1st Qu.:0.2510 1st Qu.:0.02910
    .
    .
    .
    (Other) : 3
    > trial trial

    • I’m not going to be able to give you complete code for this right now given time constraints but it would probably be best to do something like this. Keep in mind this is pseudo-code…

      
      alldata < - ...
      just7points <- ...
      ggplot()+
      geom_point(data=alldata, aes(...), alpha= 0.25)+
      geom_point(data=just7points, aes(...), alpha=1)
      

      Once you have the code worked out, send it to me and I'll post.

  5. Wow… This is just wonderful, you must be from another planet because it takes patience and skills to explain the way you did. Thank you, i really appreciate this !!!

  6. The vjust and hjust arguments are deprecated. What are the updates to increase spacing between the axes and axis text?
    Thank you

    • If you look at the help for scale_color_grey() you’ll see that there are arguments for start and end. So scale_color_grey(start=0.8, end=0.2) will give you light to dark. Regarding scale_color_brewer() those are pre-built scales so if you want to drop a color you’d likely need to use scale_color_manual() after figuring out the colors. To figure out the colors:

      
      library(RColorBrewer)
      brewer.pal(5, "Blues")
      
  7. Helpful, thanks. I find it useful to look at the default text margins to take as a starting point when adjusting. For example, the default plot.title margin (see theme_grey function) is margin = margin(b = half_line * 1.2), where half_line is half the base_size font, or in short the bottom title margin is .5*11*1.2 = 6.6 pt.

  8. Is there any way to customize where the tick marks are along the x-axis? I have a time series, and I’d like to have major tick marks every January, then minor tick marks every quarter (every April, July and October). Preferably, I’d also like the labels to read “Jan 2012”, “Apr 2012”, etc instead of just the numeric year.

    Thanks!

  9. Hi there- Anyone else experience the problem on a Mac wherein the plots rendered always have grey lines in the background– that aren’t ggplot2 gridlines??
    How to get rid of these?
    thanks

  10. Hi!

    Unfortunatelly, I just found this site. It is very helpful.

    However, I hope someone here, can help me out. I want to annotate differing text to fix position on all facetted plots.

    I used this code with + annotate_custom()
    grob <- grobTree(textGrob(label = paste("N = ", out2[, 6], " (", round(out2[, 7], 2), "%)", sep = ""), x = 0.01, y = 0.75, hjust = 0, gp = gpar(col = "black", fontsize = 13, fontface = "bold")))

    However, it always annotates the first element to all plots.

    Any suggestions?

    Thanks in advance,
    Laszlo

  11. Amazing cheatsheet. I’m new to ggplot2 and it would’ve made my day a lot easier if I had started here. Thanks for this post!

  12. I inserted minor ticks in the y-axis of my ggplot graph without labels and I wanted it to be shorter than the tick marks with labels. Wondering how this is done in r or I am looking for a r code to do it. Thanks a lot.

    • Off-hand I’m not sure. Perhaps feed a vector to the theme/axis.ticks.length. Why don’t you post code to stackoverflow and send the link.

  13. I have a small dataset (2 variables x and y, 15 instances), and would like a matrix of scatterplots y vs. x, y vs. log(x), y vs. sqrt(x), y vs. I(1/x), log(y) vs. x, log(y) vs. sqrt(x), …, I(1/y) vs. I(1/x). The ‘pairs’ command in basic R gives too much (I only want y or some function of y on the vertical axis). How can I get this matrix of 16 scatterplots using ggplot2? Such a plot is useful in searching for transformations of x and y so that the relationship is linear with constant variance (for an analysis using simple linear regression). Thanks!

  14. Question about the annotation with facet_wrap. Let’s say you didn’t want them all to say the same thing, but rather something specific per each grid. Can you do that?

    • Yes, how to do it depends on what you’re trying to do. Probably the easiest thing is to have a new dataset that has the same variable you’re using to break on facets with the text you want. But, again, really depends on what you’re trying to do. At this link (https://gist.github.com/kdauria/524eade46135f6348140) is an example of how to add a custom linear regression formula.

  15. This is an excellent resource. Thank you for posting. I have a question about facet_grid(). I am using it to plot 43 histograms. Everything works nicely, but all the histograms are printed next to each other in one long line. Is there a way to get them to print in 2 or 3 rows.

    • Not with facet_grid() — the number of rows and columns is determined by your facetting variables. Try facet_wrap() instead with the nrow, ncol arguments.

Leave a Reply to Bob Cancel reply

Your email address will not be published. Required fields are marked *