13 R - ggplot2

13.1 Make a Plot You Can Explain

The ggplot2 package builds charts in layers. You start with a dataset, map variables to visual properties (such as the x-axis or colour), and add a geometric object—a geom—to say what should be drawn. This makes it easy to revise a plot without starting from scratch.

Install the package once with install.packages("ggplot2") and load it in each new R session:

library(ggplot2)

campaigns <- data.frame(
  week = rep(1:3, times = 2),
  channel = rep(c("Email", "Search"), each = 3),
  spend = c(100, 130, 150, 200, 240, 260),
  sales = c(330, 390, 440, 460, 510, 570)
)
campaigns

Each row represents one channel in one week. That matters for interpretation: a point in the plot below is not a customer or an individual order.

Start with points

ggplot(campaigns, aes(x = spend, y = sales)) +
  geom_point()

ggplot(campaigns, ...) chooses the dataset. Inside aes(), spend is mapped to horizontal position and sales to vertical position. geom_point() draws one point for each row. The result shows an association in six illustrative observations; it does not establish that spending caused sales to rise.

To distinguish channels, map colour to a variable:

ggplot(campaigns, aes(x = spend, y = sales, colour = channel)) +
  geom_point(size = 3) +
  labs(x = "Campaign spend", y = "Weekly sales",
       colour = "Channel", title = "Spend and sales by channel")

Because colour = channel is inside aes(), each channel receives a colour and a legend. size = 3 is outside aes() because every point should have the same size. Try moving it inside aes() and compare the results: the distinction between mapping and setting is one of the most common sources of confusing plots.

Choose a geom that fits the question

To see how sales change over time, use a line for each channel:

ggplot(campaigns, aes(x = week, y = sales, colour = channel)) +
  geom_line(linewidth = 1) +
  geom_point() +
  scale_x_continuous(breaks = 1:3) +
  labs(x = "Week", y = "Sales", title = "Weekly sales by channel")

Mapping colour to channel also tells geom_line() which observations belong to the same line. If you had only one observation per channel, a line would not be meaningful. For comparisons of totals across channels, calculate those totals first and then draw columns:

channel_totals <- aggregate(sales ~ channel, data = campaigns, FUN = sum)

ggplot(channel_totals, aes(x = channel, y = sales)) +
  geom_col(fill = "steelblue") +
  labs(x = "Channel", y = "Total sales across three weeks")

geom_col() uses the already calculated values in channel_totals. geom_bar() has a different default: it counts rows. Choosing the wrong geom could turn a plot of sales into a plot of how many rows each channel has.

Check the Data Behind the Plot

Before interpreting a graphic, check what a row represents, whether missing values were omitted, and whether the displayed measure is a count, average, or total. A polished plot cannot correct a join that duplicated observations upstream.

You can add layers such as facet_wrap(~ channel) when separate panels make comparisons clearer. Start with a simple graph that answers one question, label it for its intended reader, and keep the R code so you can rebuild it when data change.