Data

To explore visualizations, we will generate some new data to play with.

One of the best things about R is its ability to simulate realistic data through randomization and factorization.

We will simulate a lakes experiment testing 3 treatments at 22 lakes across four sample rounds through a growing season.

library(ggplot2)
library(data.table)

# First, we create vectors of factor levels
trt <- 1:3
lakes <- 200:221
sample_round <- 1:4

# Second, use expand.grid to make identifier frame
dat <- data.table(expand.grid(trt, lakes, sample_round))
class(dat)
str(dat)
setnames(dat, c('trt','lake', 'sample_round'))
# make lake and trt and sample rounds factors
dat[, lake := factor(lake)]
dat[, sample_round := factor(sample_round)]
dat[, trt := factor(trt)]

# summary table of observations
dat[, .N, .(lake, trt)]

# Finally, we generate some random data.
# Chlorophyll A measured by a spectrophotometer:
dat[, chlA := rnorm(.N, mean = (150 + (10*trt)), sd = 15)]

# Algal species richness measured by microscopic ID:
dat[, algal_sp := rpois(.N, lambda = 4 + (-trt))]

Now that we have our data, let’s visualize!

Basics

• ggplot() arguments:
  • default dataset - what data are we working with
  • set of mappings
    • ‘Aesthetics’ from variables
    • what columns should we use for different aspects of the plot
• Add components of figures with layers and ‘+’ sign
  • geom_histogram()

First we specify an ‘aesthetic’. The plot is based on the chosen data, and columns. The plot places an x axis that spans the range of the algal species data, but no actual data are plotted.

p <- ggplot(dat, aes(algal_sp))
p

After we have created a ggplot object, we can then show the data. This is called by a “+” following immediately after each object.

How the data are displayed, are described by geometries (geom_ functions).

For instance, geom_bar creates a filled barplot layer:

p + geom_bar()

Do Exercise 2 - Chlorophyll A histogram.

Nearly immediately in any dataset, a question is how one variable changes with another. Especially with continuous and factor variables, this is a good way to check for outliers.

In ggplot2 syntax, the definition of the x and y variables are done in the ggplot() function, aes() argument:

p <- ggplot(dat, aes(x=trt, y=chlA)) 
p + geom_point()

Here we use the geom_point layer to view all the data. To summarize the data statistically we can use a boxplot:

p + geom_boxplot()

By default the quartiles of the data are shown, along with lines extending to 90% of the normalized data distribution, with points for values outside that 90%.

A ‘violin plot’ combines the best of the histogram with the boxplot schematic:

p + geom_violin()

Multiple layers can be added on top of one another:

p +
	geom_violin() +
	geom_point()

Do Exercise 3 Scatterplot.

Grouping

• Group on a single graph: are there lake differences?

in the aesthetic argument in a ggplot function, a group can be specified. This can be a color:

p <- ggplot(dat, aes(x=algal_sp, y=chlA, col = lake))
p + geom_point()

But with lots of lakes, the default color palette doesn’t separate so well.

or it can be a shape:

p <- ggplot(dat, aes(x=algal_sp, y=chlA, shape = trt))
p + geom_point()

or it can be a size:

p <- ggplot(dat, aes(x=algal_sp, y=chlA, size = trt))
p + geom_point()

Groupings can be mixed or crossed to explore interactions:

p <- ggplot(dat, aes(x=algal_sp, y=chlA, shape = trt, col = lake))
p + geom_point()

Again, there are too many lakes for this to be meaningful. In fact some default ggplot2 behavior attempts to limit the poor decision making here.

Do Exercise 4, Plotting groups.

Another way to plot things by lake would be to put all the lakes on the x-axis, and chlorophyll on the y axis, and use color to indicate the richness:

p <- ggplot(dat, aes(x = lake, y = chlA, size = algal_sp))
p + geom_point()

And finally, using filled bar plots is not advised for showing individual observations, but can give an idea of cumulative totals if those are important:

p <- ggplot(dat, aes(x = lake, y = chlA))
p + geom_bar(stat = 'identity')
# IDENTITY is to tell ggplot not to count occurrences but sum them

This is marginally more useful when treatments are indicated:

p <- ggplot(dat, aes(x = lake, y = chlA, fill = trt))
p + geom_bar(stat = 'identity', position = 'dodge')

When using cumulative totals for the same categories across groups, a fractional approach can be helpful:

p <- ggplot(dat, aes(x = lake, y = chlA, fill = trt))
p + geom_bar(stat = 'identity', position = 'stack')

Facet specification

When there is too much information to grasp happening on one plot, we can take advantage of the facet_ family of functions:

p <- ggplot(dat, aes(x=algal_sp, y=chlA, col = trt))
p + geom_point() + 
  facet_wrap(~lake)

Do Exercise 5, Faceting.

The facet plot gives an enormous advantage when data checking. In one visualization you can see:

• completeness of information by group (i.e., site)
• range of data across grouns
• differences by factor within groups (i.e., treatments)

Facet plots should nearly always encourage you to use a multilevel model (i.e. “mixed effects”) - but that is a different class.

Trendlines

The ggplot tool is meant to step you from data visualization into modeling, though it is not a modeling tool. Still, it can visualize the results of models and relationships.

In fact, there is a relationship that we used to create the data:

p <- ggplot(dat, aes(x = algal_sp, y = chlA))
p + geom_point() + 
   stat_smooth(method = 'lm')

Plotting dates

Dates provide special challenges for the data managing biologist. They are universally used, easily understandable, and with the potential to be formatted more ways than can be imagined.

Dates are horrible to manage.

Luckily visualizing data arranged by date, or a time series, is one way to start to diagnose potential problems.

Some data to play with:

# the seq.Date function generates a sequence of dates
datedat <- data.table(date = seq.Date(as.Date('2018-01-01'), Sys.Date(), by = 1))
# the sin function of the rownumber (.I)
datedat[, y := sin(.I)]
# need a second date column to show ggplot quirks
datedat[, datechar := as.character(date)]

The ggplot functionality nicely formats dates, but only if they are specified in a particular way:

# the two columns *look* identical
datedat[, .(date, datechar)]


p <- ggplot(datedat, aes(x = datechar, y = y ))
p + geom_point()

p <- ggplot(datedat, aes(x = date, y = y ))
p + geom_point()

• Flagging discontinuous time series

Sometimes there is a break in the data that will be important to know about, and visualizing by day is the fastest way to find it:

# remove some data and replot:
datedat[sample(1:.N, 3), date := NA]
p <- ggplot(datedat, aes(x = date, y = y ))
p + geom_line()

But the geom_line function will connect across data observations, hiding the break. Another function (geom_path) which more obviously demonstrates the breaks because it tries to plot one value for each x.

p <- ggplot(datedat, aes(x = date, y = y ))
p + geom_path()

Additional information

• Geometric object
  • geom_point()
  • geom_line()
• Statistical visualization
  • geom_smooth()
  • geom_bar()
  • geom_histogram()
  • geom_boxplot()
• Dataset and aesthetic adjustments
  • scale_continuous()
  • scale_manual()
  • lims()
  • labs()
  • guide_legend()
  • theme(), theme_bw(), theme_classic()
• Grouping related data
  • Same plot
    • Assign grouping variables as default or layer aes()
      • group
      • color
      • shape
  • Multiple plots
    • facet_grid()
    • facet_wrap()

Saving plots as new files

ggsave(“acacia_by_treatment.jpg”)

• Lots of optional arguments
  • Location
  • Type
  • Size

ggsave(“figures/acacia_by_treatment.pdf”, height = 5, width = 5)