• Aim
  • Create some intuition for neat and clean graphical visualization
  • Introducing the concept of Grammar of Graphics
  • Working with ggplot2 package
  • With a focus on static graphics for publication
    
    
    
• Requirements
  • Elementary coding skills in R
  • Basics in statistics for data analysis

1. Introduction to data visualization
2. The Grammar of Graphics
3. Graphics with ggplot2
   • Plotting geometric objects
   • Mapping variables to aesthetics
   • Scales for XY-Position, Color, etc.
   • Faceting
   • Themes, labs and saving
4. Final remarks on ggplot2

• "…is that which gives the viewer the greatest number of ideas,

• in the stortest time,

• with the least ink,

• the smallest space,

• and which tells the truth about data."

• (Tufte 2001)

• Balance between importance and graphical attention
• Clear and well-structured representation
• Data-Inc-Ratio should be 1
• Use consistant and full axis
• Use gridlines and ticks, but gently
• Check for grey print-out and color blindness
• Proportionality between numbers and their graphical representation

• Proportionality betwen numbers and their graphical representation

• Proportionality betwen numbers and their graphical representation

es_graph <- (5.3 - 0.6) / 0.6  #Effect size in graphic (inches)
es_data <- (27.5 - 18) / 18   #Effect size in data (miles per gallon)
lie_factor <- es_graph / es_data;   
lie_factor  #should be around 1

## [1] 14.84211

• Subjective sensation \(w(I)\) = Physical stimulus I\(^\beta\)
  • \(I\) Magnitude of the physical stimulus
  • \(w(I)\) Subjective magnitude of the sensation evoked by the stimulus
  • \(\beta\) > 1 … perceived over-estimation (e.g. salty taste, electric shock)
  • \(\beta\) < 1 … perceived under-estimation (e.g. smell, brightness)

• Sensation \(\frac{w(I_1)}{w(I_2)}\) = Intensity \((\frac{I_1}{I_2})^\beta\)

• Do not …
  • use 3D charts
  • use more than 6 colors
  • overload chart with information
  • use rainbow or too intensive colors
  • narrow context, e.g. very small time interval
• But have a look at the 20 imperatives of information design

• Visualisation concept created by Wilkinson (1999)
  • to define the basic elements of a statistical graphic
    
    
    
• Adapted for R by Wickham (2009)
  • who created the ggplot2 package
  • consistent and compact syntax to describe statistical graphics
  • highly modular as it breaks up graphs into semantic components
    
    
    
• Is not a guide which graph to choose and how to convey information best!

A statistical graphic is a …

• mapping of data
• to aesthetic attributes (color, size, xy-position)
• using geometric objects (points, lines, bars)
• with data being statistically transformed (summarised, log-transformed)
• and mapped onto a specific facet and coordinate system

• Which data is used as an input?
• What geometric objects are chosen for visualization?
• What variables are mapped onto which attributes?
• What type of scales are used to map data to aesthetics?
• Are the variables statistically transformed before plotting?

• Which data is used as an input?
• What geometric objects are chosen for visualization?
• What variables are mapped onto which attributes?
• What type of scales are used to map data to aesthetics?
• Are the variables statistically transformed before plotting?

• Which data is used as an input?
• What geometric objects are chosen for visualization?
• What variables are mapped onto which attributes?
• What type of scales are used to map data to aesthetics?
• Are the variables statistically transformed before plotting?

• Which data is used as an input?
• What geometric objects are chosen for visualization?
• What variables are mapped onto which attributes?
• What type of scales are used to map data to aesthetics?
• Are the variables statistically transformed before plotting?
• Is any form of facetting applied?

library("ggplot2")
packageDescription("ggplot2")

## Package: ggplot2
## Type: Package
## Title: An implementation of the Grammar of Graphics
## Version: 1.0.0
## Author: Hadley Wickham <h.wickham@gmail.com>, Winston Chang
##        <winston@stdout.org>
## Maintainer: Hadley Wickham <h.wickham@gmail.com>
## Description: An implementation of the grammar of graphics in R. It
##        combines the advantages of both base and lattice graphics:
##        conditioning and shared axes are handled automatically, and
##        you can still build up a plot step by step from multiple
##        data sources. It also implements a sophisticated
##        multidimensional conditioning system and a consistent
##        interface to map data to aesthetic attributes. See the
##        ggplot2 website for more information, documentation and
##        examples.
## Depends: R (>= 2.14), stats, methods
## Imports: plyr (>= 1.7.1), digest, grid, gtable (>= 0.1.1),
##        reshape2, scales (>= 0.2.3), proto, MASS
## Suggests: quantreg, Hmisc, mapproj, maps, hexbin, maptools,
##        multcomp, nlme, testthat, knitr, mgcv
## VignetteBuilder: knitr
## Enhances: sp
## License: GPL-2
## URL: http://ggplot2.org, https://github.com/hadley/ggplot2
## BugReports: https://github.com/hadley/ggplot2/issues
## LazyData: true
## Collate: 'aaa-.r' 'aaa-constants.r' 'aes-calculated.r' .....
## Packaged: 2014-05-20 22:12:23 UTC; hadley
## NeedsCompilation: no
## Repository: CRAN
## Date/Publication: 2014-05-21 15:36:28
## Built: R 3.1.0; ; 2014-05-22 09:40:09 UTC; unix
## 
## -- File: /Library/Frameworks/R.framework/Versions/3.1/Resources/library/ggplot2/Meta/package.rds

data("diamonds")
# Prices of 50,000 round cut diamonds

# A dataset containing the prices and other attributes of almost 54,000 diamonds. 
# The variables are price in USD, carat, cut quality,...
# help(diamonds)
head(diamonds)

##   carat       cut color clarity depth table price    x    y    z
## 1  0.23     Ideal     E     SI2  61.5    55   326 3.95 3.98 2.43
## 2  0.21   Premium     E     SI1  59.8    61   326 3.89 3.84 2.31
## 3  0.23      Good     E     VS1  56.9    65   327 4.05 4.07 2.31
## 4  0.29   Premium     I     VS2  62.4    58   334 4.20 4.23 2.63
## 5  0.31      Good     J     SI2  63.3    58   335 4.34 4.35 2.75
## 6  0.24 Very Good     J    VVS2  62.8    57   336 3.94 3.96 2.48

qplot(data, ...) #close to plot() fct with compressed functionality and lots of defaults
ggplot(data, mapping = aes(), ...) #the main plotting function

• data: the data frame employed
• mapping: list of asthethic assignments
  • aes(x, y, color, size, fill, shape)

ggplot(data = diamonds, mapping = aes(x=x, y=price))
# Warning: No layers in plot 

• ggplot() itself …
  • is not a plotting layer but initializes a ggplot object
  • declares the input data and some common aesthetics
• Add layers by using the + operator

ggplot(data = diamonds, mapping = aes(x=x, y=price)) + geom_point()

geom_point(mapping = NULL, data = NULL, stat, ...)

• mapping: list of asthethic assignments aes() for geom object
• stat: statistical transformation required for geom object
• NULL: inhibit values from ggplot()
• ... other arguments,
  • often aesthetics you want to set unconditionally, e.g. color="red"

• Excercise:
  1. Load library(ggplot2)
  2. Load data(diamonds) from the ggplot2 package
  3. Create a scatterplot of price and carat
     • Use ggplot(data = ..., mapping = aes(x=..., y=...)) to initiate an ggplot object
     • Use + geom_point() to create a scatterplot layer

ggplot(data = diamonds, mapping = aes(x=carat, y=price)) + geom_point()

ggplot(diamonds, aes(x=carat, y=price)) + geom_point()

Examples for basic geom_ functions

geom_point(mapping = NULL, data = NULL,
stat = "identity", position = "identity", ...)

geom_line(mapping = NULL, data = NULL,
stat = "identity", position = "identity", ...)

geom_boxplot(mapping = NULL, data = NULL,
stat = "boxplot", position = "dodge", outlier.color = "black", 
                      outlier.shape = 16, outlier.size = 2, ...)

• Add, combine and edit layers like a toolbox
• Extensive list of all ggplot2 objects can be found at
  • docs.ggplot2.org
  • including many examples at the end of each topic

• Besides mapping onto x- and y-position
  • variables can be assigned to geom aesthetics

Examples:

geom_point(aes(x=carat, y=price, size = carat ))#: point size varies with `carat`
aes(..., color = carat)#: color varies with `carat`
aes(..., fill = carat)#: fill color varies with `carat`
aes(..., linetype = carat)#: linetype varies with `carat`

Setting mappings for geom extends or replaces ggplot() mappings

ggplot(diamonds, aes(x=carat, y=price)) + geom_point(aes(color = cut))

But you can also state universal mappings within ggplot() objects

ggplot(diamonds, aes(x=carat, y=price, color = cut)) + geom_point()

Including some additional manipulations of variables

ggplot(diamonds, aes(x=carat, y=price, size=carat^2, alpha=carat^2)) + geom_point() 

• Excercise:
  1. Create a boxplot of y=price and x=carat
  2. Look at docs.ggplot2.org to find the right geom-object for a boxplot
  3. Map the variable cut onto the fill aesthetic: aes(fill=cut)

ggplot(diamonds, aes(x=cut, y=price, fill=cut)) + geom_boxplot() 

• data and aes inside geom replace ggplot() attributes

avgprice <- data.frame(avg = mean(diamonds$price))

ggplot(diamonds, aes(x=cut, y=price, fill=cut)) + geom_boxplot()  +
  geom_hline(data=avgprice, aes(yintercept = avg))

avgprice <- data.frame(avg = mean(diamonds$price))

ggplot() + 
  geom_boxplot(data=diamonds, aes(x=cut, y=price, fill=cut))  +
  geom_hline(data=avgprice, aes(yintercept = avg))

• you can also save a ggplot object and later add layers

p <- ggplot(diamonds, aes(x=cut, y=price, fill=cut)) + geom_boxplot()  
p

ggplot(diamonds, aes(x=carat)) + geom_bar()

geom_bar(mapping = NULL, data = NULL, stat = "bin", position = "stack",
  ...)

• stat argument statistically transforms input data (bin means bin and count)
• position argument: dodge for side-by-side bars or stack for additive bars

ggplot(diamonds, aes(x=carat)) + geom_bar(stat="bin")

ggplot(diamonds, aes(x=carat)) + geom_bar(stat="bin", binwidth=1)

- binwidth sets the width of the bin

• Excercise:
  1. Do the transformation stat="bin" by hand
  2. Cut carat into groups (0,1] (1,2] (2,3] (3,4] (4,5]
     • cut(diamonds$carat, breaks=seq(0,5,1))
  3. Count observation by group
     • as.data.frame(table(diamonds$caratcut))
  4. Use ggplot() + geom_bar()
     • what stat argument do you need?

diamonds$caratcut <- cut(diamonds$carat, breaks=seq(0,5,1))
diamondbin <- as.data.frame(table(diamonds$caratcut))
diamondbin

##    Var1  Freq
## 1 (0,1] 36438
## 2 (1,2] 15613
## 3 (2,3]  1857
## 4 (3,4]    27
## 5 (4,5]     4

ggplot(diamondbin, aes(x=Var1, y=Freq)) + 
  geom_bar(stat="bin")  # Not right, why?

ggplot(diamondbin, aes(x=Var1, y=Freq)) + 
  geom_bar(stat="identity")  # Here we go!

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar() 

• Scale is a realization of data values in terms of asthetical/physical values
  
  
• control the mapping of data (domain) to aesthetics (range)
• each aesthethic has its own (default) scale
• scale depends on the variable type:
  • discrete (factor, logical, character)
  • continuous (numeric)

• Scale specifications have the form scale_AESTHETIC_SCALENAME()
  • AESTHETIC: x, y, color, size or shape
  • SCALENAME: grey, gradient, hue, manual, continuous

- to adjust the ticks of a (continuous) y-axis you would set
+ scale_y_continuous(breaks = seq(0,9000,1000))

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar() +
  scale_y_continuous(breaks = seq(0,9000,1000))

- to adjust the name of the fill mapping you would set
+ scale_fill_hue(name = "Cut quality")

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar() +
  scale_fill_hue(name="Cut quality")

- to switch from hue color scale to grey scale
+ scale_fill_grey()

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar() +
  scale_fill_grey()

• Excercise:
  1. Use the previous barplot
     • ggplot(diamonds, aes(x=carat, fill=cut)) + geom_bar()
  2. Set the name of the fill aesthetic to "Cut quality"
     • scale_fill_discrete(…)
  3. Set the name of the x aesthetic to "Carat in gramm"
     • scale_x_continuous(…)
  4. Set the breaks of the y aesthetic to seq(0, 9000, 500)
  5. Set the label of the fill aesthetic to c(1:5)
     • docs.ggplot2.org/current/discrete_scale.html to see what arguments are allowed in scales

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar() +
  scale_fill_discrete(name = "Cut quality", label = c(1:5)) + 
  scale_x_continuous(name = "Carat in gramm") + 
  scale_y_continuous(breaks = seq(0,9000,500))

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar() +  scale_fill_hue(
  name = expression(paste("Age \ngroup \nnot", integral(f(x) * dx^{2}, a, b)))
  ) 

#for more mathematical expressions see ?plotmath

Every plot has two position scales - horizontal and vertical

data(economics) #US economic time series
head(economics)

##         date   pce    pop psavert uempmed unemploy
## 1 1967-06-30 507.8 198712     9.8     4.5     2944
## 2 1967-07-31 510.9 198911     9.8     4.7     2945
## 3 1967-08-31 516.7 199113     9.0     4.6     2958
## 4 1967-09-30 513.3 199311     9.8     4.9     3143
## 5 1967-10-31 518.5 199498     9.7     4.7     3066
## 6 1967-11-30 526.2 199657     9.4     4.8     3018

?economics #monthly data with consumption, population and unemployment 

p <- ggplot(economics, aes(x=date, y=unemploy)) + geom_line()
p

To change the limits of a scale (and remove data outside limits)

p + scale_y_continuous(limits=c(6000,9000)) # removes observations outside [6000,9000]

Alternatively

p + ylim(6000,9000) # removes observations outside [6000,9000]

p + coord_cartesian(ylim=c(6000,9000)) # to zoom in and not remove data

When working with time series data, the date scale is useful:

p + scale_x_date(limits = as.Date(c("2000-10-01", "2005-01-01"))) # for date values

When working with time series data, the date scale is useful:

p + scale_x_date(limits = as.Date(c("2000-10-01", "2005-01-01")),
                 breaks= "6 months")

• Position scales (scale_x_continous and scale_y_continous)
  • have an attribute trans="identity" to transform the scale

p + scale_y_continuous(trans="identity", breaks=seq(0,15000,2500), limits=c(0,15000))

• Position scales (scale_x_continous and scale_y_continous)
  • have an attribute trans="log" to transform the scale

p + scale_y_continuous(trans="log", breaks=c(1,100,1000,15000), limits=c(1,15000))

• Position scales (scale_x_continous and scale_y_continous)
  • have an attribute trans="identity" to transform the scale
• taken from Wickham (2009)

Alternatively, one can use the functions

              p1 <- p + scale_y_log10(limits=c(1,15000))
                                p2 <- p + scale_y_reverse()
                                                  p3 <- p + scale_y_sqrt()

• Excercise
  1. Load data set data(economics)
  2. Create a time series plot of consumption expenditures pce
     • aes(x=date, y=pce)
     • geom_line()
  3. Limit the time frame to 1970 to 2005
     • scale_x_date(limits = as.Date(c(..., ...)))
  4. Log-transform the y scale
     • scale_y_continuous()
  5. Provide proper limits and ticks for y axis
     • scale_y_continuous(breaks = c(1,...), limits = c(1,...))

ggplot(economics, aes(x=date, y=pce)) + geom_line() +
  scale_x_date(limits = as.Date(c("1970-01-01", "2005-01-01"))) +
  scale_y_continuous(trans="log", breaks=c(1,100,1000,10000), limits=c(1,10000))

p <- ggplot(diamonds, aes(x=carat, y=price, color=carat)) + geom_point()
p + scale_color_gradient()

Input via #RRGGBB hexadecimal-code

p + scale_color_gradient(low="#FF0000" , high="black", 
                              na.value = "grey50", limits=c(0,5))

• via hsv: hue, saturation, and value/brightness, each in the range [0, 1].
• via hcl: hue [0,360], chroma [0,100], and luminance [0,100]

p + scale_color_gradient2(midpoint=3, low=hsv(0.3,0.5,1) , mid="blue", 
                               high=hcl(20,80,40))

via color palettes, e.g., rainbow, terrain.colors, topo.colors

p + scale_color_gradientn(colours=rainbow(3), breaks=c(2,3,4))

• via _grey: grey color scale

p <- ggplot(diamonds, aes(x=carat, fill=cut)) +
              geom_bar(position = "dodge", stat="bin")
p + scale_fill_grey()  

• via _hue: qualitative color scale with evenly spaced hues
  • chroma=60, luminance=100 and hue moves in steps of 15°

• via _hue: qualitative color scale with evenly spaced hues
  • chroma=60, luminance=100 and hue moves in steps of 15°

p + scale_fill_hue()  

via color brewer and RColorBrewer package

p +  scale_fill_brewer(type="qual", palette="Set1") 

via color brewer and RColorBrewer package

RColorBrewer::display.brewer.all() #for an overview of all palletes

• Wickham (2009) recommends:
  • "Set1" and "Dark2" for categorical data
  • "Set2", "Pastel1", "Pastel2" and "Accent" for areas

via _manual: either set each color by hand or via secondary sources

library(wesanderson)
p + scale_fill_manual(values = wes_palette("Cavalcanti",5))

• Excercise:
  1. Use ggplot(diamonds, aes(x=carat, fill=cut)) + geom_bar()
  2. Change the position of the bar plot to dodge
  3. Have a look at docs.ggplot2.org/current/scale_brewer.html
  4. Set the fill color according to brewer type
     • seq (sequential), div (diverging) or qual (qualitative)
     • scale_fill_brewer(type="")
  5. Run RColorBrewer::display.brewer.all()
  6. Set the fill colors according to brewer set PuRd
     • scale_fill_brewer(palette="")

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar(position = "dodge") +
 scale_fill_brewer(type="seq") 

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar(position = "dodge") +
 scale_fill_brewer(type="div") 

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar(position = "dodge") +
 scale_fill_brewer(type="qual") 

ggplot(diamonds, aes(x=carat, fill=cut)) +  geom_bar(position = "dodge") +
 scale_fill_brewer(palette="PuRd") 

• Consult the online documentation for an overview:
  • docs.ggplot2.org/current/theme.html
• For a information on color (and fill)
  • SAPE Reference ggplot2 on color
  • Modern color models
  • Cookbook R ggplot2 colors
  • Hexadecimal Color chart
  • A Wes Anderson Palette Generator

• Faceting (also called conditioning, latticing or trellising)
  • divides the data into subsets according to categorical variables
  • then plots in multiple panels
  • a bit like creating a contingency table

facet_grid(facets = . ~ ., scales = "fixed", ...) # e.g.: margins=TRUE for totals
facet_wrap(facets = . ~ ., scales = "fixed", ...) # e.g.: ncol=3 ,nrow=3

• facets: formula how to split facet (row ~ col)
• scales: axis scale be fixed, free, free_x, free_y
• margins: add row/column totals as additional plot panels
• ncol, nrow: set number of panel columns and rows

• facet_grid() (left) is fundamentally 2d
  • being made up of two independent components
• facet_wrap() (right) is 1d, but wrapped into 2d to save space.

• aligns vertical scales which is useful for wide screens

diamondssub <- subset(diamonds, 
                        cut %in% c("Good", "Premium") & 
                        clarity %in% c("VS2", "IF"))

ggplot(diamondssub, aes(x=carat)) + geom_histogram() + 
  facet_grid(cut ~ .)

• single column useful to compare distributions

ggplot(diamondssub, aes(x=carat)) + geom_histogram() + 
  facet_grid(cut ~ ., scales="free_y")

• same plot but with free y scales

ggplot(diamondssub, aes(x=carat, y=price)) + geom_point() + 
  facet_grid(cut ~ clarity)

• Excercise
  1. Create a smoothed line of x=carat and y=price
     • geom_smooth()
  2. Use facet_grid to explore the relationship for levels of cut
     • facet_grid(. ~ .)
  3. Add points with same xy-variables to the plot
     • + geom_point()
  4. Set the alpha value of the points to 0.4

ggplot(diamondssub, aes(x=carat, y= price)) + geom_smooth() + 
  facet_grid(cut ~ .) + 
  geom_point(alpha=0.4)

facet_wrap(facets = . ~ ., scales = "fixed", ...) # e.g.: ncol=3 ,nrow=3

• most useful when a single variable has many levels
• construct a long ribbon of panels

ggplot(diamonds, aes(x=carat, y=price)) + geom_point() + 
  facet_wrap(~clarity, nrow=2)

• faceting by level of clarity

• Faceting (using facet_XXX) when
  • need to broadly disentangle variables
  • avoid overlaps when plotting
  • allows two dimensions to compare two variables at ones
  • scales can vary across panels
    
    
• Grouping (using asthetics) when
  • want to explore small differences
  • want to have groups all in one plot

• Themes control all aspects of non-data display
  • Don't affect how data is rendered by geoms or scale
  • Control over
    • title, axis labels, axis ticks labels,
    • legend labels, legend key labels
    • as well as the color of ticks, grid lines and backgrounds

p + theme_grey() # see plot to the left
p + theme_bw() #see plot to the right 
p + theme() # to set options

ggplot(diamondssub, aes(x=carat, y=price, linetype=cut)) +  geom_smooth() + 
  theme(legend.position = "bottom", legend.direction = "horizontal", 
        title= element_text(face = "bold"),
        panel.background = element_rect(fill = "#999922"), 
        panel.grid.minor = element_line(color="red", linetype="dotted"))

Labs control the labelling of titles and axes

p + labs(title="Diamond bar plot", x="Carat in gramm", y="Price in USD")

Alternatively

p + ggtitle("Diamond bar plot") + xlab("Carat in gramm") + ylab("Price in USD")

• ggplot has implemented ggsave() for exporting graphs
• recognises the extensions:
  • eps/ps, tex (pictex), pdf, jpeg, tiff, png, bmp, svg and wmf

ggsave(filename="diamondplot.eps", plot=last_plot(), 
       width=10, height=5, dpi=300, units="cm")

ggsave(filename="diamondplot.pdf", 
       width=11, height=8.5, dpi=600, units="in")

• Weaknesses of ggplot2
  • Since Feb 2014 only maintenance mode = no new features
    • However, ggvis is supposed to become the new ggplot
  • ggplot2 is slower than some other graphic systems (e.g. lattice)
  • Does not really support 3d-graphs

• Why to use ggplot2
  • Large community as one of the most popular R packages
  • Uses sensible and attractive decisions about
    • dimensions, scales and colors by default
  • Additional packages in-the-same-vain like ggplot2 for
    • geographical information (ggmap),
    • genomic data (ggbio),
    • Markov Chain Monte Carlo simulations (ggmcmc)
    • and interactive graphics (ggvis)

• ggplot2 docs with many examples at the end of each topic
• R cookbook website provides solutions for frequent problems
• ggplot2 Handout Summary from the Basel Biometric Section.
  
  Thank you for your attention!
  
  If you have any questions or remarks don't hesitate to contact me!
  tim.winke@gmail.com