Assignment 5 - Plotting with ggplot2

library(plyr)
library(ggplot2)
library(lattice)
library(xtable)

html_print <- function(x, ..., digits = 0, include.rownames = FALSE, 
                       header=names(x)){
  # print html table with customized header, 
  # without permanently changing column names of the dataframe
  origHeader <- names(x)
  names(x) <- header
  print(xtable(x, digits = digits, ...), 
        type = 'html', include.rownames = include.rownames)
  names(x) <- origHeader
}

gDat <- read.table('data/gapminderDataFiveYear.txt',
                   sep='\t', quote='"', header=TRUE)
gDat <- droplevels(subset(gDat, continent!='Oceania'))

Historical Life Expectancy by Continents

pExp <- ggplot(gDat, aes(x=year, y=lifeExp, color=continent))
pExp + geom_point(alpha=0.3) + geom_smooth(method='loess')

xyplot(lifeExp ~ year, gDat, group=continent, 
       type=c('p','smooth'),
       auto.key=TRUE)

Across different continents, the life expectancy trend line over time are very distinct.

Similarities

• Lattice and ggplot2 are equally capable of producing this plot with simple syntax.
• In both case, one has to specify the data.frame, the variable going on x and y axis, the grouping used to superimpose plots and the including of data points and line fitting.

Differences

• The lattice plot is enclosed in a single statement, while the ggplot2 plot is defined in incremental steps, gradually adding features to the plot. I think the implication is that ggplot2 makes it easier to reuse components of a plot, potentially at the cost of hard to trace the effect of a line of code on finished a plot.
• the group variable in ggplot2 is indirectly defined through the colour variable, which is not as intuitive
• ggplot2 creates a legend by default, while lattice requires auto.key=TRUE
• The use of transparency and colour seem to be more straightforward with ggplot2

Proportion of countries with low life expectancy over time by continent

Code in this section is adapted from https://gist.github.com/ArephB/6667983#file-stat545a-2013-hw03_bolandnazar-moh-rmd

Average of Life Expectancy over time is 59.2623. And this value is used as the cutoff for low life expectancy countries.

# split a vector by a threshold value, and count the lower and upper tier
count_splits <- function(c, threshold){
  tierSizes <- c(lower=sum(c < threshold),
                 upper=sum(c >= threshold))
  return(tierSizes)
}
lifeRatio <- ddply(gDat, ~continent + year, function(x){
  count <- count_splits(x$lifeExp, benchmark)
  ratio <- count/length(x$lifeExp)
  tier <- c('below', 'above')
  return(data.frame(tier, count, ratio))
})

pLowExpR <- ggplot(subset(lifeRatio, tier=='below'), aes(x=year, y=ratio, color=continent))
pLowExpR + geom_point() + geom_line()

xyplot(ratio ~ year, lifeRatio, group=continent, 
       subset=tier=='below',
       auto.key=list(column=4), 
       type=c('o'), 
       xlab="Year", ylab="Ratio of Low Life Expectancy",
       main="Trends in Low Life Expectancy Ratio in Different Continents",
       )

The bar chart below captures roughly the same information as the one above. But the bar plot provides

• a better vision representation for the proportion
• the option to include the exact count of countries of low life expectancy (through the axis label)

The cons are costing more space and paint.

barchart(as.character(year)~count|continent, lifeRatio, group=tier,
         stack=TRUE,
         auto.key=list(column=2, title='Life Expectancy Tier'), 
         xlab="Life Expectancy Tier Count", ylab="Year",
         scales=list(x = list(relation="free")),
         main="Trends in Low Life Expectancy Ratio in Different Continents",
         )

Same plot done with ggplot2:

pLowExpR2 <- ggplot(lifeRatio, aes(x=year, y=count, fill=tier))
pLowExpR2 + scale_fill_manual(values = c("below" = "orange","above" = "skyblue")) +
  geom_bar(stat="identity") +
  facet_wrap(~continent, scales='free') 

Similarity

• Panel conditioning and the use of facets ease the way of making of similar and comparable plots.

Difference

• Lattice bar chart by default uses the height of the bar to represent a pre-computed count. ggplot2 by default performs the count by category on the provided variable, but its default behaviour can be changed by geom_bar(stat=“identity”) and specifying the y variable in the aesthetics.
• It seems that I cannot make the ggplot2 bar chart horizontal when I have scales='free'.

Historical Trends in GDP per Capita by Most Extreme Case of a Continent

data aggregation code is modified from Jenny's example

extremGdp <- ddply(gDat, ~ continent+year, function(x) {
  gdpPercap <- range(x$gdpPercap)
  return(data.frame(gdpPercap, stat = c("min", "max")))
})
#html_print(extremGdp)

xyplot(gdpPercap ~ year|stat, extremGdp, groups=continent,
       type=c('o','g'),
       scales=list(y = list(relation="free")),
       layout=c(1,2),
       xlab='Year', ylab='Extreme GDP per Capital',
       auto.key=list(column=4))

xyplot(gdpPercap ~ year|continent, extremGdp, groups=stat,
       type=c('o','g'),
       layout=c(4,1),
       xlab='Year', ylab='Extreme GDP per Capital',
       auto.key=list(column=2))

minGdpPlt <- ggplot(extremGdp, 
                    aes(x=year, y=gdpPercap, color=continent, linetype=stat))
minGdpPlt + geom_point() + geom_line()

I intend to show trends of GDP per capita given by the top and bottom countries of different continents. So I can see the magnitude of difference and continuously widening gap between the two collections of trend lines.

I have not found a simple solution in lattice. But ggplot2 certainly has a neat solution for it.

Difference

• Initially I felt using the colour keyword to define group in ggplot2 was unintuitive. Now I've realized how succinct it is. Colour, size, linetype, and shape can be used both independently and combined to define groups in the data and how to visually represent these group in a plot.

Other plots

Here the distribution of GDP per capital by continent is plotted. GDP per capita of Africa, Europe and America all have distinct mean from each other. GDP per capita of Asian countries seem most variable.

ggplot(gDat, aes(x=gdpPercap, fill=continent)) + 
  geom_density(alpha=0.3) + 
  scale_x_log10()