Bad Visualizations

Bad Visualizations

Bad Visualizations

Younger adults are large percentage of coronavirus hospitalizations in United States, according to new CDC data

Source: The Washington Post

Bad Visualizations

Source: Roeder K (1994) DNA fingerprinting: A review of the controversy (with discussion). Statistical Science 9:222-278, Figure 4

Want More?

Visit Karl Broman's Website

Thoughtful Visualization

• Start with a thoughtful question about your data.

• Think of several alternative ways to show the same relationship (e.g., vary plot type, color, size, facetting, etc.)

• Motivate every option in function

• Prioritize reader-friendliness (e.g., are you using color-blind friendly colors?)

• Attention to detail (titles, labels, axes, ticks, grids, etc.)

Asking Questions about the Data

Make a list of questions you think you can answer by analyzing the data on terrorist attacks. Write down as many as you can.

Example 1:

What is the relationship between armed assaults and bombings? Are they complements or substitutes?

• two continuous variables--> a scatterplot

ggplot2: Grammar of Graphics

A plot consists of:

1. mappings (aes): variables that are mapped to graphical elements

2. layers: graphical elements (geoms such as points, lines, rectangles, text, etc.) and statistical transformations (stats such as identity, counts, bins, etc.)

3. scales: map values in the data space to values in the aesthetic space (color, size, shape, position)

4. coordinate system (coord): Cartesian, but can be others

5. facetting: subsetting and arranging data

6. theme: fine-tuning the result, e.g. font, background, margins

Load the data on global terrorist attacks

library(classdata)
data("terr_attacks.wide")
str(terr_attacks.wide)

## Classes 'tbl_df', 'tbl' and 'data.frame':    2015 obs. of  16 variables:
##  $ country                            : chr  "Afghanistan" "Albania" "Algeria" "Angola" ...
##  $ ccode                              : num  700 339 615 540 160 371 900 305 373 692 ...
##  $ cabb                               : chr  "AFG" "ALB" "DZA" "AGO" ...
##  $ year                               : int  2001 2001 2001 2001 2001 2001 2001 2001 2001 2001 ...
##  $ GDPpc                              : num  NA 2454 3617 2214 7776 ...
##  $ population                         : num  20531160 3060173 31590320 15562791 37471535 ...
##  $ tradeofgdp                         : num  NA 57.4 58.7 150.3 21.9 ...
##  $ polity2                            : int  NA 5 -3 -3 8 5 10 10 -7 -8 ...
##  $ Armed Assault                      : num  2 0 80 22 0 0 0 0 2 0 ...
##  $ Bombing/Explosion                  : num  10 1 17 5 2 0 0 0 1 0 ...
##  $ Facility/Infrastructure Attack     : num  2 0 2 5 0 0 1 0 1 0 ...
##  $ Assassination                      : num  0 0 4 0 0 2 0 0 0 0 ...
##  $ Hostage Taking (Kidnapping)        : num  0 0 3 3 0 0 0 0 0 0 ...
##  $ Unarmed Assault                    : num  0 0 0 0 0 0 1 0 0 0 ...
##  $ Hijacking                          : num  0 0 0 0 0 0 0 0 0 0 ...
##  $ Hostage Taking (Barricade Incident): num  0 0 0 0 0 0 0 0 0 0 ...

library(ggplot2)
ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`, y=`Bombing/Explosion`))+geom_point()
ggplot(data=terr_attacks.wide, aes(x=log(`Armed Assault`+1), y=log(`Bombing/Explosion`+1)))+geom_point()
ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1, y=`Bombing/Explosion`+1))+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))

Interpreting Scatterplots

• General patterns

  • Form and direction
  • Strength

• Localized patterns

• Deviations from the pattern

  • Outliers

Interpreting Scatterplots: Example

1. Form: Roughly linear, several distinct groups
2. Strength: pretty strong. Data points form a line.
3. Direction: Positively Associated.
4. Outliers: AK, DC, NY, CA.

Looking at Conditional Relationships

• Can use color, size, shape, etc. to show additional information

• E.g., can color observations based on regime type. (Why?)

  • Data Management Tool #1: Recoding

terr_attacks.wide$dem<-"Autocracy"
terr_attacks.wide$dem[terr_attacks.wide$polity2>7]<-"Democracy"

Color by Regime Type

library(ggplot2)
terr_attacks.wide$dem<-"Autocracy"
terr_attacks.wide$dem[terr_attacks.wide$polity2>7]<-"Democracy"
ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1, y=`Bombing/Explosion`+1, colour=dem))+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+  scale_colour_discrete(name = "Regime Type")

Printer-Friendly Plot

If intended for print, consider using a gray scale.

ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1, y=`Bombing/Explosion`+1, colour=dem))+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+  scale_colour_grey(name = "Regime Type")

Or change marker type rather than colors

ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1, y=`Bombing/Explosion`+1, shape=dem,))+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_shape_manual(name = "Regime Type",values=c(1,3))

Colorblind-Friendly Colors

• Colorblind-Friendly Colors

cbPalette <- c("#999999", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2", "#D55E00", "#CC79A7")
cbbPalette <- c("#000000", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2", "#D55E00", "#CC79A7")
ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1, y=`Bombing/Explosion`+1, colour=dem))+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+  scale_colour_manual(name = "Regime Type", values=cbPalette)

Set Theme Options

• Control the overall looks of your plot

?theme

#Set theme options:
theme_set(theme_grey() + #set the theme to the canned theme named "theme_bw"
            theme(panel.background = element_rect(fill = NA, color = 'black'))+ #remove the background color (fill), make the axes black
            theme(axis.text=element_text(size=10), axis.title=element_text(size=12,face="bold")))

Scale by GDP/cap

ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1, y=`Bombing/Explosion`+1, colour=dem,size=GDPpc))+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))

Facetting

Can facet to display plots of subsets of data: facet_wrap, facet_grid

• E.g., can facet by region, year, etc.

• facet_wrap optimizes the number of rows and columns

• facet_grid allows you to decide to facet by column, row, or both (if conditioning on two variables). Works best if the conditioning variables have 2-4 categories.

More Recoding

• Code the region variable from ccode.

terr_attacks.wide$region<-NA
terr_attacks.wide$region[terr_attacks.wide$ccode<200]<- "The Americas"
terr_attacks.wide$region[terr_attacks.wide$ccode>=200 & terr_attacks.wide$ccode<401]<- "Europe"

Your Turn

Code the rest of the values of region using the following coding scheme:

Facet by Region

ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1,y=`Bombing/Explosion`+1,colour=dem,size=GDPpc))+facet_wrap(~region)+geom_point()+scale_y_log10("Bombing/Explosion", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+scale_x_log10("Armed Assault", breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))

ggplot(data=terr_attacks.wide, aes(x=`Armed Assault`+1,y=`Bombing/Explosion`+1,size=GDPpc))+facet_grid(region~dem)+
  geom_point()+ scale_y_log10("Bombing/Explosion",breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))+
  scale_x_log10("Armed Assault",breaks=c(1,11,101,1001,10001),labels=c("0","10","100","1000","10000"))

What We Learned

• Intro to thoughtful graphics

• Scatterplots using ggplot2

  • Color, size, facetting

• Using logged scales

• Titles, breaks, and labels

• Recoding variables