ggformula: Another Option for Teaching Graphics in R to Beginners

David Robinson’s excellent blog entry “Teach the tidyverse to beginners” (http://varianceexplained.org/r/teach-tidyverse) argues that a tidyverse approach is the better of two prevalent options for teaching beginners:

As his title suggests, David prefers the second option and makes a strong case for it. Indeed, if those were the only two options, I’d probably choose tidyverse as well.

A third way

But these are not the only options. Nick Horton’s “Options for teaching R to beginners: a false dichotomy?”(http://sas-and-r.blogspot.com/2017/07/options-for-teaching-r-to-beginners.html) describes a third alternative that also “get[s] students doing powerful things quickly”, as David desires. This third way is based on a formula interface provided by the combination of

the lattice package for graphics,
several functions from the stats package for modeling (e.g., lm(), t.test()), and
the mosaic package for numerical summaries and for smoothing over edge cases and inconsistencies in the other two components.

Important in this approach is the syntactic similarity that the following “formula template” brings to all of these operations.

goal ( y ~ x , data = mydata, … )

Many important data analysis operations can be executed by filling in the four boxes with the appropriate information for the desired task. This allows students to become fluent quickly with a powerful, coherant toolkit for data analysis.

Nick’s post illustrates this with an investigation of how the price of diamonds depends on (among other things) color. The similarity among the commands to compute mean price for each of two colors, to create side-by-side boxplots, to run a two-sample t test, and to fit a linear model are what make this approach so compelling.

# diamonds2 was called `recoded' in Nick's post
library(dplyr)
diamonds2 <- diamonds %>% 
  filter(color == "D" | color == "J") %>%
  mutate(col = as.character(color))

  mean(price ~ col, data = diamonds2)

## Warning in mean.default(price ~ col, data = diamonds2): argument is not numeric
## or logical: returning NA

bwplot(price ~ col, data = diamonds2)
t.test(price ~ col, data = diamonds2)
    lm(price ~ col, data = diamonds2)

This “Less Volume, More Creativity” approach is outlined in more detail in a recent R Journal article and has worked well for a growing number of instructors in first (and subsequent) courses (see for example Wang et al., “Data Viz on Day One: bringing big ideas into intro stats early and often” (2017), TISE).

Trouble in paradise

But as Nick hinted in his blog post, the use of lattice has some drawbacks. While basic graphs like histograms, boxplots, scatterplots, and quantile-quantile plots are simple to make with lattice, it is challenging to combine these simple plots into more complex plots or to plot data from multiple data sources. Splitting data into subgroups and either overlaying with multiple colors or separating into sub-plots (facets) is easy, but the labeling of such plots is not as convenient (and takes more space) that the equivalent plots made with ggplot2. And in our experience, students generally find the look of ggplot2 graphics more appealing.

On the other hand, introducing ggplot2 into a first course is challenging. The syntax is more verbose, so it takes up more of the limited space on projected images and course handouts. More importantly, the syntax is entirely unrelated to the syntax used for other aspects of the course. For those adopting a “Less Volume, More Creativity” approach, ggplot2 is tough to justify.

ggformula: The third-and-a half way

Danny Kaplan and I recently introduced ggformula, a package that provides a formula interface to ggplot2 graphics. Our hope is that this provides the best aspects of lattice (the formula interface and lighter syntax) and ggplot2 (modularity, layering, and better visual aesthetics).

For simple plots, the only thing that changes is the name of the plotting function. Each of these functions begins with gf. Here are two examples, either of which could replace the side-by-side boxplots made with lattice in Nick’s post.

gf_boxplot(price ~ col, data = diamonds2)
gf_violin(price ~ col, data = diamonds2)

If we like, we can even overlay these two types of plots to see how they compare. To do so, we simply place the then operator (%>%, also called a pipe) between the two layers and adjust the transparency so we can see both where they overlap.

gf_boxplot(price ~ col, data = diamonds2, alpha = 0.05) %>%
gf_violin(price ~ col, data = diamonds2, alpha = 0.3, fill = "navy", col = NA)

Setting and mapping attributes of a plot

In the example above, we set certain attributes (fill, color, opacity) to constants ("navy", NA, 0.3). Often we want the values of these attributes to be determined by the data, this is called mapping rather than setting and is indicated by an additional ~. One can read fill = "navy" as “fill is (or equals) navy” and fill = ~ col as "fill is determined by col. (Which colors get used for which values is controled by the fill scale for the plot, and this scale can be modified by the user who doesn’t like the default scale choice.)

gf_boxplot(price ~ col, data = diamonds2, alpha = 0.05) %>%
  gf_violin(price ~ col, data = diamonds2, alpha = 0.3, fill = ~ col)

A more interesting use of mapping colors occurs when elements of each color overlap. For example, a scatterplot with overlaid regression fits shows that although the price of J diamonds tends to be higher overall, after taking into consideration the number of carats, the D color diamonds tend to sell for more than the J color diamonds. (It also shows that the simple linear fit is poor for the largest diamonds.)

xyplot(price ~ carat, groups = col, data = diamonds2,
       auto.key = TRUE, type = c("p", "r"), alpha = 0.5) 
gf_point(price ~ carat, color = ~ col, data = diamonds2, alpha = 0.5) %>%
   gf_lm(price ~ carat, color = ~ col, data = diamonds2, alpha = 0.5)

In lattice, this requires us to manually turn on a legend with auto.key = TRUE and to introduce the type argument and the c() function to get both layers. In ggplot2, we simply put one layer over the other (so no new syntax to learn). The (improved) legend is generated automatically, and the portion of the plot devoted to the data is substantially larger than for the lattice plot. So the ggformula plot is both technically better and syntactically simpler than its lattice counterpart.

Getting a little help

Everyone forgets some details from time to time. The ggformula functions provide a mechanism for obtaining a bit of help without going to the full help page for the function: simply execute the function with no arguments.

gf_violin()

## gf_violin() uses 
##     * a formula with shape y ~ x. 
##     * geom:  violin 
##     * stat:  ydensity 
##     * position:  dodge 
##     * key attributes:  alpha, color, fill, group, linetype, size, weight,
##                    draw_quantiles = NULL, trim = TRUE,
##                    scale = "area", bw, adjust = 1, kernel
##                    = "gaussian"
## 
## For more information, try ?gf_violin

This terse help describes how the formula is used. (For those already familiar with ggplot2, the formula shape shows how the formula is converted into ggplot2 aesthetics.) Also listed are the geom (type of mark drawn), any non-identity stats and positions, and the main attributes that can be mapped or set to refine the plot. Just this much help lets us quickly (a) thinken the lines, (b) make the violins less (or more) smooth, (c) add lines at the quartiles of the data, and scale the area of the violins according the number of observations represented.

gf_violin(price ~ col, data = diamonds2,
          color = ~ col, fill = ~col, alpha = 0.2,
          scale = "count",
          draw_quantiles = c(0.25, 0.5, 0.75), size = 0.8,
          adjust = 1/2)

Facets

Sometimes it is preferable to create multiple plot arranged in a grid rather than to overlay subgroups in the same space. ggformula provides two ways to create these facets. The first uses | very much like lattice does.

gf_point(price ~ carat | col, data = diamonds2, alpha = 0.2) %>%
  gf_lm()
gf_point(price ~ carat | col ~ clarity, data = diamonds2, alpha = 0.2) %>%
  gf_lm()

Notice that the gf_lm() layer inherits information from the the gf_points() layer in these plots, saving some typing when the information is the same in multiple layers.

The second way to adds facets with gf_facet_wrap() or gf_facet_grid(). The commands below create the same plots as those above.

gf_point(price ~ carat, data = diamonds2, alpha = 0.2, size = 0.5) %>%
  gf_lm(alpha = 0.5) %>%
  gf_facet_wrap( ~ color)

gf_point(price ~ carat, data = diamonds2, alpha = 0.2, size = 0.5) %>%
  gf_lm(alpha = 0.5) %>%
  gf_facet_grid( color ~ clarity)

Refinements

The full power to modify plot limits, titles, scales, theme elements, etc. is inherited by ggformula from ggplot2. For the most common of these we offer the functions gf_lims(), gf_labs(), and gf_theme().
For the rest, we offer gf_refine() which allows us to make any ggplot2 refinements without resorting to the + syntax used by ggformula.

gf_point(price ~ carat, data = diamonds2, 
         color = ~ col, alpha = 0.2, size = 0.5) %>%
  gf_lm(alpha = 0.5) %>%
  gf_facet_wrap( ~ color) %>%
  gf_labs(title = "Price vs Size", subtitle = "(2 colors of diamonds)",
          caption = "source: ggplot2::diamonds",
          x = "size (carat)", y = "price (US dollars)"
  ) %>%
  gf_refine(scale_color_manual(values = c("red", "navy"), guide = "none")) %>%
  gf_theme(theme_minimal())

Although this example uses a lot of customization, it is completely modular. Any line of the customization can be omitted, or additional features can be customized one line at a time.

Fitting into the tidyverse work flow

ggformala also fits into a tidyverse-style workflow (arguably better than ggplot2 itself does). Data can be piped into the initial call to a ggformula function and there is no need to switch between %>% and + when moving from data transformations to plot operations.

diamonds %>%
  filter(color %in% c("D", "G", "J")) %>%
  mutate(col = as.character(color)) %>%
  gf_bar( ~ col)