% Tutorial 6: Visualizations % DPI R Bootcamp % Jared Knowles

Overview

In this lesson we hope to learn:

• What is data visualization and why does it matter?
• How to draw diagnostic plots in base graphics
• Colors
• ggplot2'
• Basic geoms
• Layering and faceting plots
• Putting it together

Graphics Matter

• Graphics are a huge part of R and R encourages its users to use visualization to aid and present analysis
• Graphics are hard to get right, but tools can help make it easier
• Lots of great resources to help available for this online and linked to at the end of this document

Visualization Philosophy

• Visualization is about thinking about what objects in our dataset map with what visual cues for the viewer
  • liberal use of names(object) helps
• What is the best way to show discrete variables, continuous variables, and connected variables
• We also need to be unafraid about trying transformations of variables, recoding variables, and reshaping variables to tell the story we are looking for
• What is a plot?

Our data for this exercise

• As we now know, student data is quite fascinating
• We have many discrete attribute variables
  • Race, gender, economics, etc.
• We have some continuous variables
  • Test scores, attendance
• We have some grouping variables
  • Schools, districts, students, teachers
• Rich ground to make a number of visualizations that are useful

Base graphics

• R has base graphics which are useful for quickly doing some plots

hist(df$readSS)

Base graphics are simple

• You can use them to understand your data and do quick diagnostics
• The commands are pretty easy to remember
• You can do complex things with them if you like, but the syntax can be confusing

Base graphics has some limitations

• What if we try to do a scatterplot of lots of data?

plot(df$readSS, df$mathSS)

• To add a line, we have to use another line of code

plot(df$readSS, df$mathSS)
lines(lowess(df$readSS ~ df$mathSS), col = "red")

Basic Plot

• Don't use R base graphics
• ggplot2 is pretty much the new standard in R
• Create beautiful, clear plots with a nice consistent set of language conventions

library(ggplot2)
qplot(readSS, mathSS, data = df)

Now, how?

• Apply best visualization principles, particularly the “Grammar of Graphics”
• ggplot2 an R package does just this by breaking plots into a few basic components
  • aesthetics, geoms, layers, and scales
• Making high quality graphics is just a matter of applying the right data to the right aesthetics in an appropriate geom, adding layer for detail, and ensuring the scale is appropriate
• Let's talk about how to do that next

What are the basic plot types?

What are some advanced plot types?

Your Turn: What are some examples of interesting visualizations we could use?

Understanding Grammar of Graphics through A Scatterplot

qplot(readSS, mathSS, data = df, alpha = I(0.3)) + theme_dpi()

• What is a scatterplot?
• A 2d representation of data as coordinates, in this case the readSS is the x coordinate and mathSS is the y coordinate for each observation in our data
• Each observation is a point (geom)
• Both axes scale in a linear fashion (scales)

Grammar of Graphics

• In one way of thinking about this, each data visualization has four components
• Geometries that represent data (points, bars, lines)
• Statistics that represent information about the data (identity, mean, median, deviance)
• Scales that map the geometries and statistics to space (linear, quadratic, logrithmic)
• A coordinate system and canvas to put all on

Geoms

• Geoms are the way data is represented, you can think of it like a chart type in another programming language
• We have seen a number of examples, and geoms can be combined in unique ways to convey more data
• Quiz: Represent df$mathSS using 3 separate geoms

Geom Quiz

qplot(mathSS, readSS, data = df) + theme_dpi()

qplot(mathSS, data = df) + theme_dpi()

qplot(factor(grade), mathSS, data = df, geom = "line", group = stuid, alpha = I(0.2)) + 
    theme_dpi()

Aesthetics

• geoms allow us to only display a couple of data elements at once, to do more we need to map to other visual representations
• This is what aesthetics are
• aesthetics are colors, glyphs (shapes), and sizes of graph objects mapped to visual cues
• ggplot2 has an extended syntax that makes this obvious

ggplot(df, aes(x = readSS, y = mathSS)) + geom_point()

# Identical to: qplot(readSS,mathSS,data=df)

• aes says we are specifying aesthetics, here we specified x and y to make a two dimensional graphic

Examples of Aesthetics

data(mpg)
qplot(displ, cty, data = mpg) + theme_dpi()

qplot(displ, cty, data = mpg, size = cyl) + theme_dpi()

qplot(displ, cty, data = mpg, shape = drv, size = I(3)) + theme_dpi()

qplot(displ, cty, data = mpg, color = class) + theme_dpi()

Experiment with Aesthetics

Draw some plots with different aesthetics using our student level dataset

Some Considerations with Aesthetics

• aesthetics are very sensitive to whether a variable is continuous or discrete or ordered
• R isn't always so worried about this!
• Why does it matter? Let's see a few examples

Aesthetics Considerations (ordered)

qplot(mathSS, readSS, data = df[1:100, ], size = race, alpha = I(0.8)) + theme_dpi()

• Does this make sense?

Another Aesthetics Concern (ordered)

df$proflvl2 <- factor(df$proflvl, levels = c("advanced", "basic", "proficient", 
    "below basic"))
df$proflvl2 <- ordered(df$proflvl2)
qplot(mathSS, readSS, data = df[1:100, ], color = proflvl2, size = I(3)) + scale_color_brewer(type = "seq") + 
    theme_dpi()

Aesthetics Concern 2 (discrete and continuous)

• What aesthetics can we map continuous variables like mathSS to, and what can we map discrete characteristics like race to?

qplot(factor(grade), readSS, data = df[1:100, ], color = mathSS, geom = "jitter", 
    size = I(3.2)) + theme_dpi()

• What's wrong?

Aesthetics Concern 2

qplot(factor(grade), readSS, data = df[1:100, ], color = dist, geom = "jitter", 
    size = I(3.2)) + theme_dpi()

• What's wrong?

Thinking about Aesthetics

• One concern is discrete v. continuous variables

Aesthetic Discrete Continuous

Color Disparate colors Sequential or divergent colors Size Unique size for each value mapping to radius of value Shape A shape for each value does not make sense

Another is ordered v. unordered

Aesthetic Ordered Unordered

Color Sequential or divergent colors Rainbow Size Increasing or decreasing radius does not make sense Shape does not make sense A shape for each value

Scales

• Scales are the way the numeric/categorical data is mapped to a visual representation
• They transform the geoms and aesthetics
• Scales preserve the mapping, but allow us to explore reshaping the data

Scales Caveats

• Despite not changing the mapping of data to space, scales can dramatically influence the interpretation of our data
• Rescaling data should be done thoughtfully and with care
• Picking a good scale can be really difficult, and sometimes, a good scale just won't exist without reshaping or subsetting the data!

Scales also apply to color and fill

• The colors we choose can carry important meaning, as we have discussed
• So can the shapes we use for points, or for lines
• It is important to keep this in mind

Layers

• Exactly what they sound like, each plot is a simple series of layers
• One way to do layers is to break plots up into small multiples (see Tufte)

qplot(readSS, mathSS, data = df) + facet_wrap(~grade) + theme_dpi(base_size = 12) + 
    geom_smooth(method = "lm", se = FALSE, size = I(1.2))

We can also facet across more attributes

qplot(readSS, mathSS, data = df) + facet_grid(ell ~ grade) + theme_dpi(base_size = 12) + 
    geom_smooth(method = "lm", se = FALSE, size = I(1.2))

Visualizing Categorical Data

• Visualization is not just limited to continuous data using scatterplots
• Sometimes we want to look at the density of data in different categories
• Sometimes we want to look at the size of groups compared to an expected group size
• What are some other examples?

Structural Plots

library(vcd)
df$proflvl <- factor(df$proflvl, levels = c("advanced", "proficient", "basic", 
    "below basic"))
a <- structable(proflvl ~ race, data = df)
mosaic(a, shade = TRUE)

How do we interpret these plots?

• We use size and color to express proportionality and expected values
• This can be applied to other areas as well

Another example

library(vcd)
df$proflvl <- factor(df$proflvl, levels = c("advanced", "proficient", "basic", 
    "below basic"))
a <- structable(female ~ race, data = df)
mosaic(a, shade = TRUE)

A few pro tips

• Don't be afraid to manipulate the data
  • Aggregate, combine, rescale variables to make the story easier to understand
• Don't use too much color
  • People print your work, color blindness is more prevalent than you think
  • Color is fickle. Monitors and projectors reproduce it differently.
• Don't overwhelm the user
  • Make your visual depictions as easy as possible to immediately understand that data
• Don't plot everything all at once

Colors in R

• ggplot2 default colors are based on maximizing space between colors using HCL
• Read more here: A tutorial on ggplot2 colors; HCL color

The R Colorspace is Huge

##  [1] "plum"        "violet"      "darkmagenta" "magenta4"    "magenta3"   
##  [6] "magenta2"    "magenta"     "magenta1"    "orchid4"     "orchid"

##  [1] "salmon1"       "darksalmon"    "orangered4"    "orangered3"   
##  [5] "coral"         "orangered2"    "orangered"     "orangered1"   
##  [9] "lightsalmon2"  "lightsalmon"   "peru"          "tan3"         
## [13] "darkorange2"   "darkorange4"   "darkorange3"   "darkorange1"  
## [17] "linen"         "bisque3"       "bisque1"       "bisque2"      
## [21] "darkorange"    "antiquewhite3" "antiquewhite1" "papayawhip"   
## [25] "moccasin"      "orange2"       "orange"        "orange1"      
## [29] "orange4"       "wheat4"        "orange3"       "wheat"        
## [33] "oldlace"

##  [1] "snow1"       "snow2"       "rosybrown"   "rosybrown1"  "rosybrown2" 
##  [6] "rosybrown3"  "rosybrown4"  "lightcoral"  "indianred"   "indianred1" 
## [11] "indianred3"  "brown"       "brown4"      "brown1"      "brown3"     
## [16] "brown2"      "firebrick"   "firebrick1"  "chocolate"   "chocolate4" 
## [21] "saddlebrown" "seashell3"   "seashell2"   "seashell4"   "sandybrown" 
## [26] "peachpuff2"  "peachpuff3"

Some Practical Advice

• Use ColorBrewer palettes: Available by default in ggplot2
• Designed for geographic mapping purposes to be easy to distinguish, easy to interpret across a diverse array of users, and sensitive to color-blindness
• As easy in ggplot2 as adding any color scale: +scale_color_brewer(palette"X")
• You can choose from sequential palettes, diverging palettes, qualitative palettes,and many more–depending on how many colors you need and what makes sense in your data
• Let's think about colors a little more together

Above and Beyond

Scary R Code

library(grid)
p1<-qplot(readSS,..density..,data=df,fill=race,
      position='fill',geom='density')+scale_fill_brewer(
        type='qual',palette=2)

p2<-qplot(readSS,..fill..,data=df,fill=race,
      position='fill',geom='density')+scale_fill_brewer(
        type='qual',palette=2)+ylim(c(0,1))+theme_bw()+
          opts(legend.position='none',
               axis.text.x=theme_blank(),
               axis.text.y=theme_blank(),
               axis.ticks=theme_blank(),
               panel.margin=unit(0,"lines"))+ylab('')+
                 xlab('')

vp<-viewport(x=unit(.65,"npc"),y=unit(.73,"npc"),
             width=unit(.2,"npc"),height=unit(.2,"npc"))
print(p1)
print(p2,vp=vp)

Exercises

1. Embed one plot in another plot in R using two different data elements from our data set. For example, plot a histogram of readSS inside a scatterplot of readSS and mathSS

2. Explore some examples on the ggplot2 website. What are some ways to overlay more than 3 dimensions of data in a single plot?

3. What types of data work best for what types of visualizations?

References

1. Hadley Wickham's JSM 2012 Presentation
2. Hadley Wickam's ggplot2 Intro Presentation
3. The ggplot2 Homepage
4. ggplot2 Documentation
5. Quick R: Basic Graphs
6. Quick R: Advanced Graphs

Session Info

It is good to include the session info, e.g. this document is produced with knitr version 0.9.6. Here is my session info:

print(sessionInfo(), locale = FALSE)

## R version 2.15.2 (2012-10-26)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## 
## attached base packages:
## [1] splines   grid      stats     graphics  grDevices utils     datasets 
## [8] methods   base     
## 
## other attached packages:
##  [1] R2SWF_0.4        snow_0.3-10      gbm_1.6-3.2      reshape_0.8.4   
##  [5] caret_5.15-048   foreach_1.4.0    cluster_1.14.3   reshape2_1.2.2  
##  [9] lme4_0.999999-0  Matrix_1.0-10    lattice_0.20-10  xtable_1.7-0    
## [13] vcd_1.2-13       colorspace_1.2-0 MASS_7.3-22      Hmisc_3.10-1    
## [17] survival_2.37-2  sandwich_2.2-9   quantreg_4.94    SparseM_0.96    
## [21] gridExtra_0.9.1  mgcv_1.7-22      eeptools_0.1     mapproj_1.2-0   
## [25] maps_2.3-0       proto_0.3-10     plyr_1.8         stringr_0.6.2   
## [29] ggplot2_0.9.3    lmtest_0.9-30    zoo_1.7-9        knitr_0.9.6     
## 
## loaded via a namespace (and not attached):
##  [1] codetools_0.2-8    compiler_2.15.2    dichromat_1.2-4   
##  [4] digest_0.6.0       evaluate_0.4.3     formatR_0.7       
##  [7] gtable_0.1.2       iterators_1.0.6    labeling_0.1      
## [10] markdown_0.5.3     munsell_0.4        nlme_3.1-106      
## [13] RColorBrewer_1.0-5 scales_0.2.3       stats4_2.15.2     
## [16] tools_2.15.2

Attribution and License

This work (R Tutorial for Education, by Jared E. Knowles), in service of the Wisconsin Department of Public Instruction, is free of known copyright restrictions.