Cédric Scherer

30DayChartChallenge

Mon, 01 Apr 2024 00:00:00 +0000

Cédric Scherer

Mon, 01 Apr 2024 00:00:00 +0000

DataViz

Mon, 01 Apr 2024 00:00:00 +0000

DataWrapper

Mon, 01 Apr 2024 00:00:00 +0000

Exploring "The Simpsons" with DataWrapper for the 30DayChartChallenge 2024

Mon, 01 Apr 2024 00:00:00 +0000

What’s the Challenge About?

The #30DayChartChallenge is a data visualization challenge with the aim to create a data visualization on a certain topic for each day of April. Anyone is welcome to contribute, no matter which data source or tool is used to create the visualizations. This year, Dominic and I are hosting the fourth edition with a new set of prompts:

Find out more about the history of the challenge and explore inspirational contributions in my Nightingale article from last year.

I will not participate on every day this year as there are too many other responsibilities I am involved in currently. But for the few prompts I am participating in, my goal is to explore the possibilities to create interactive, responsive visualizations with DataWrapper. And to keep it simple so I’ve decided to set some constraints for my contributions.

Self-Imposed Rules

Data: visualize a “The Simpsons” data set (1, 2, and 3)
Tool: use DataWrapper to create the charts (and resist to use ggplot2 👀)
Colors: pick colors from “The Simpsons” (1 and 2)
Fun: choose a suitable quote from “The Simpsons” for the title

Day 1: Part-to-Whole

Day 2: Neo*

* The prompt was often interpreted as "new" (or "Neo" from "Matrix") but Dominic's original idea was "neon colors" :)

Day 4: Waffle

Day 5: Diverging

Day 9: Major/Minor

Day 10: Physical

Note: While trying to create a lollipop chart, I ended up with this column chart with gradient fills—it's actually a range plot

Day 11: Mobile-friendly

Stacked Bars for Members of the “No Pie Chart Gang”

Day 13: Family

Note: The family tree is build with a scatter plot with lots of custom annotation lines.

Day 14: Heatmap

Note: The chart is actually a choropleth map to ensure that the heatmap looks good independent of the screen width.

Day 19: Dinosaurs

Day 20: Correlation

Day 22: Mobility

Day 28: Trend

Day 29: Black’n’White (or Day 27: Good/Bad)

Posts

Mon, 01 Apr 2024 00:00:00 +0000

R

Mon, 01 Apr 2024 00:00:00 +0000

Annotations

Thu, 26 Oct 2023 00:00:00 +0000

Ggplot2

Thu, 26 Oct 2023 00:00:00 +0000

Tidyverse

Thu, 26 Oct 2023 00:00:00 +0000

Tutorial

Thu, 26 Oct 2023 00:00:00 +0000

Yet Another How-to on Labelling Bar Graphs in ggplot2

Thu, 26 Oct 2023 00:00:00 +0000

Header visualization from “maischberger” (see my note below)

Table of Content

Introduction

Yes, I have written about creating bar charts with {ggplot2} before. As one of the most common chart types, creating bar charts is a task that thousands of people likely face every day. In an old blog post I’ve shown various ways

how to calculate the percentage values,
how to position the percentage labels inside, and
how to color the bars using different colors.

Inspired by a question by one of my clients, I am now extending that list by showcasing

how to place the category labels above the bars.

Data Preparation

I am using the diamonds data set from the {ggplot2} package to generate shares of diamonds for five different categories describing the quality of the cut. In a first step, I am calculating the shares per quality and turn the categories into a factor ordered by that metric.

library(dplyr)
library(ggplot2)

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, prop))

## # A tibble: 5 × 2
##   cut         prop
##   <ord>      <dbl>
## 1 Ideal     0.400 
## 2 Premium   0.256 
## 3 Good      0.0910
## 4 Very Good 0.224 
## 5 Fair      0.0298

There are multiple other ways to calculate the shares, including diamonds |> mutate(n = n()) |> summarize(prop = n() / unique(n), .by = cut). Instead of using the experimental .by argument you can also group your data first with group_by(cut) before summarizing per cut quality.

The last step is not needed in our example case here as the ranking by shares follows the defined order of the cut qualities. However, in most other cases you likely have to sort your categories on your own.

Create a Basic Bar Chart

Now, I can easily pass the summarized data set to ggplot() and create a simple horizontal bar graph:

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col()

The default horizontal bar chart, ranked by shares.

Alternatively, you can transform the complete data set on the fly instead of calculating shares first:

ggplot(diamonds, aes(y = cut, x = after_stat(count / sum(count)))) +
  geom_bar()

The same default horizontal bar chart, this time created with geom_bar() and after_stat().

Style the Visualization

If you know me a bit, you know that before moving on I have to modify the theme and fix the grid lines (read: remove them all together in this case).

Also, I am modifying the x axis range and labels. Instead of showing proportions, I decide to show percentages (0-100). Also, to follow good practice I am adding the percentage label to the axis using label_percent() from the {scales} package. I am also removing the padding on the left and right of the bars and adjust the limits so that the 40% label is shown as well.

theme_set(theme_minimal(base_family = "Spline Sans"))
theme_update(
  panel.grid.minor = element_blank(),
  panel.grid.major = element_blank(),
  axis.line.x = element_line(color = "grey80", linewidth = .4),
  axis.ticks.x = element_line(color = "grey80", linewidth = .4),
  axis.title.y = element_blank(),
  plot.margin = margin(10, 15, 10, 15)
)

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col() +
  scale_x_continuous(
    expand = c(0, 0), limits = c(0, .4),
    labels = scales::label_percent(),
    name = "Proportion"
  )

The same bar chart with a modified theme and a polished x axis.

Place Category Labels on the Top

The approach I take to now to move the labels to the top of the bars is: faceting!

There are multiple options including placing the labels with geom_text and shifting them upwards. But by far the fastest way (and also likely the one that breaks last when the number of bars changes) is using the facet functionality of {ggplot2}.

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col() +
  facet_wrap(~ cut) +
  scale_x_continuous(
    expand = c(0, 0), limits = c(0, .4),
    labels = scales::label_percent(),
  )

Creating small multiples based on the variable mapped to the y axis leads to a set of mostly empty panels with redundant labels by default.

It doesn’t work “out of the box”, however. But that’s a quick fix if you know about the ncol and the scales arguments in the facet_wrap() function! The trick is that we force all small multiples in a single column (so that bars share a common baseline again) by setting ncol = 1. By default, the axis ranges are kept constant across small multiples. By setting scales = "free_y" we can free the axis range which removes redundant, empty groups and all the resulting white space.

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, -prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col() +
  facet_wrap(~ cut, ncol = 1, scales = "free_y") +
  scale_x_continuous(
    name = "Proportion", expand = c(0, 0), 
    limits = c(0, .4), labels = scales::label_percent()
  )

Now, our facets looks like a regular bar chart. However, we have redundant labels which we remove in the next step.

Note that we also have to flip the order of our categories as now they’re ordered top to bottom, not bottom to top anymore.

The final step is cleaning up the labels. First, let’s remove the category names on the y axis by passing guide = "none" in scale_y_discrete().

To modify the new labels, the so-called strip texts, we address the text element strip.text via theme(). The margin of zero on the left ensures that, together with the horizontal justification (hjust = 0) that the strip text labels are full left-aligned with the baseline of the bars. The small margin at the top and the bottom ensure that the labels are not clipped (e.g. that the descender of y is shown completely).

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, -prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col() +
  facet_wrap(~ cut, ncol = 1, scales = "free_y") +
  scale_x_continuous(
    name = "Proportion", expand = c(0, 0), 
    limits = c(0, .4), labels = scales::label_percent()
  ) +
  scale_y_discrete(guide = "none") +
  theme(strip.text = element_text(
    hjust = 0, margin = margin(1, 0, 1, 0), 
    size = rel(1.1), face = "bold"
  ))

The polished new y axis labels, originally strip text of small multiples, replace the original axis labels.

To add some spacing between the last bar and the axis line, one can adjust the vertical padding of each panel by passing expansion(add = c(.8, .6) to the expand argument in scale_y_discrete().

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, -prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col() +
  facet_wrap(~ cut, ncol = 1, scales = "free_y") +
  scale_x_continuous(
    name = "Proportion", expand = c(0, 0), 
    limits = c(0, .4), labels = scales::label_percent()
  ) +
  scale_y_discrete(
    guide = "none", expand = expansion(add = c(.8, .6))
  ) +
  theme(strip.text = element_text(
    hjust = 0, margin = margin(1, 0, 1, 0), 
    size = rel(1.1), face = "bold"
  ))

The final version with polished category labels by adjusting the strip text of the facets.

Bonus: Style the Bars

Let’s merge this new approach with some of the tricks from my previous blog post. We add direct labels and highlight the top-ranked category.

Highlight Top-Ranked Category

By mapping the cut variable to fill, bars would be colored by categories. To color only the first, top ranked bar, I am making use of the rank which is equal to the factor level. Thus, mapping the fill to as.numeric(cut) == 1) returns TRUE for “Ideal” and FALSE otherwise. To customize the fill colors, we add scale_fill_manual() to pass a vector of two custom colors. As we don’t need a legend, we also set guide = "none".

p <- 
  diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, -prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col(aes(fill = as.numeric(cut) == 1)) +
  facet_wrap(~ cut, ncol = 1, scales = "free_y") +
  scale_x_continuous(
    name = "Proportion", expand = c(0, 0), 
    limits = c(0, .4), labels = scales::label_percent()
  ) +
  scale_y_discrete(guide = "none", expand = expansion(add = c(.8, .6))) +
  scale_fill_manual(values = c("grey50", "#1D785A"), guide = "none") +
  theme(strip.text = element_text(
    hjust = 0, margin = margin(1, 0, 1, 0), 
    size = rel(1.1), face = "bold"
  ))

p

Drawing attention to the top-ranked category by using a different fill color.

Add Percentages as Direct Labels

Similarly, we can pass an expression to color and hjust inside the geom_text() component that we use to add the direct labels. As TRUE is encoded as 1, all group that have a share lower than 5% are right-aligned while all others are left-aligned (as FALSE = 0). To move the labels a bit more inside and outside, respectively, I am cheating by adding some spaces before and after the label.

p +
  geom_text(
    aes(label = paste0("  ", sprintf("%2.1f", prop * 100), "%  "), 
        color = prop > .05, hjust = prop > .05),
    size = 4, fontface = "bold", family = "Spline Sans"
  ) +
  scale_color_manual(values = c("black", "white"), guide = "none")

Now, the bars are labelled directly including a rule that automatically places the labels inside the bars as long as they are wide enough to fit the label.

Alternatively, you can pass the value for hjust directly by using an ifelseor if_else condition: hjust = if_else(prop > .05, 1.2, -.2):

p +
  geom_text(
    aes(label = paste0(sprintf("%2.1f", prop * 100), "%"), 
        color = prop > .05, hjust = if_else(prop > .05, 1.2, -.2)),
    size = 4, fontface = "bold", family = "Spline Sans"
  ) +
  scale_color_manual(values = c("black", "white"), guide = "none")

The same logic applies when we want to control the text color, which is recommended here to increase the contrast. With the final scale_color_manual() I change the text color to white in case the label is placed inside the bar and black otherwise.

Another way to style the labels would be scales::label_percent(accuracy = .1, prefix = " ", suffix = "% ")(prop) (or make use of the superseded scales::percent()) but that’s rather long and also not that easy to remember.

One could of course also remove the x axis as the values are now shown as direct labels.

p +
  geom_text(
    aes(label = paste0("  ", sprintf("%2.1f", prop * 100), "%  "), 
        color = prop > .05, hjust = prop > .05),
    size = 4, fontface = "bold", family = "Spline Sans"
  ) +
  scale_x_continuous(guide = "none", name = NULL, expand = c(0, 0)) +
  scale_color_manual(values = c("black", "white"), guide = "none")

A version in which the x axis has been removed as it shows redundant information.

Alternative Approach

Here is an approach using geom_text(). The trick here is to (i) reduce the width (read: height in our case) of the bars to allow for space for the labels and (ii) add the labels with geom_text() in combination with a custom vjust or nudge_y setting.

diamonds |> 
  summarize(prop = n() / nrow(diamonds), .by = cut) |> 
  mutate(cut = forcats::fct_reorder(cut, prop)) |> 
  ggplot(aes(prop, cut)) +
  geom_col(width = .5) +
  geom_text(
    aes(label = cut, x = 0),
    family = "Spline Sans", fontface = "bold",
    hjust = 0, vjust = -1.7, size = 4.5
  ) +
  scale_x_continuous(
    expand = c(0, 0), limits = c(0, .4),
    labels = scales::label_percent(),
    name = "Proportion"
  ) +
  scale_y_discrete(guide = "none")

An example using geom_text() to place the category labels in combination with vjust.

That’s a great solution, too. I see some potential issues coming up here, for example problems in case the labels become larger (can be fixed by removing the clipping and adding some margin) or the number of bars increases (and that may be especially a problem in an automated workflow). In the latter case, the space between bars may become too small and/or the placement of the labels, adjusted via vjust or nudge_y, is not perfectly above the bars anymore.

Conclusion

To illustrate the different behavior of the two approaches, let’s run the exact same codes on a new data set with more categories:

p1 <- 
  mpg |> 
  filter(year == "2008") |> 
  summarize(prop = n() / nrow(mpg), .by = manufacturer) |> 
  mutate(manufacturer = forcats::fct_reorder(stringr::str_to_title(manufacturer), -prop)) |> 
  ggplot(aes(prop, manufacturer)) +
  geom_col() +
  facet_wrap(~ manufacturer, ncol = 1, scales = "free_y") +
  scale_x_continuous(
    name = "Proportion", expand = c(0, 0), 
    limits = c(0, .1), labels = scales::label_percent()
  ) +
  scale_y_discrete(
    guide = "none", expand = expansion(add = c(.8, .6))
  ) +
  theme(strip.text = element_text(
    hjust = 0, margin = margin(1, 0, 1, 0), 
    size = rel(1.1), face = "bold"
  ))

p2 <- 
  mpg |> 
  filter(year == "2008") |> 
  summarize(prop = n() / nrow(mpg), .by = manufacturer) |> 
  mutate(manufacturer = forcats::fct_reorder(stringr::str_to_title(manufacturer), prop)) |> 
  ggplot(aes(prop, manufacturer)) +
  geom_col(width = .5) +
  geom_text(
    aes(label = manufacturer, x = 0),
    family = "Spline Sans", fontface = "bold",
    hjust = 0, vjust = -1.7, size = 4.5
  ) +
  scale_x_continuous(
    name = "Proportion", expand = c(0, 0), 
    limits = c(0, .1), labels = scales::label_percent()
  ) +
  scale_y_discrete(guide = "none")

library(patchwork)
p1 + p2

Applying both codes to a different data set illustrates nicely the differences of the two approaches to place labels on top. The facet approach (left) ensures that labels are placed above, while the bars get thinner. The geom approach (right) uses a fixed bar width and thus the labels overlap at some point.

Both approaches have their pros and cons. In circumstances, where you can tweak the exact setting of bar widths, font sizes, and vertical justification, the geom_text() approach might be easier to set up.

Using the facet_wrap() approach ensures that the labels are always above the bars and that the labels are not clipped by the panel or plot border. This is especially powerful in case the data changes and charts need to be updated without any further modifications. Or if you want to apply a function to multiple data sets without the need to include further arguments to modify the widths and spacing. At the same time, the thinner bars make it more difficult to place labels inside the bars. However, the same issue would pop up when adjusting the widths and font sizes in the geom_text() example.

Finally, I should note that also the facet approach will break at some point: if the figure height is not sufficient, no bars are visible at all. But scaling the figure height based on the number of categories is something one can easy automate as well.

Both approaches, the facet approach on the left and the geom approach on the right, need sufficient figure height to work nicely.

Note on the Header Image

I’ve seen this bar chart on the german TV talk show “maischberger”, airing on September 27, 2023. A wonderful revival of 3D-bars, combined with a glossy, transparent gradient style. It shows the number of newly constructed apartments per year over time.

R Session Info

## R version 4.3.0 (2023-04-21)
## Platform: aarch64-apple-darwin20 (64-bit)
## Running under: macOS Ventura 13.2.1
## 
## Matrix products: default
## BLAS:   /Library/Frameworks/R.framework/Versions/4.3-arm64/Resources/lib/libRblas.0.dylib 
## LAPACK: /Library/Frameworks/R.framework/Versions/4.3-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.11.0
## 
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
## 
## time zone: Europe/Berlin
## tzcode source: internal
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] patchwork_1.1.2   ggplot2_3.4.3     dplyr_1.1.0       systemfonts_1.0.4
## 
## loaded via a namespace (and not attached):
##  [1] gtable_0.3.4      jsonlite_1.8.7    highr_0.10        compiler_4.3.0    tidyselect_1.2.0  stringr_1.5.0     jquerylib_0.1.4   scales_1.2.1     
##  [9] textshaping_0.3.6 yaml_2.3.7        fastmap_1.1.1     R6_2.5.1          labeling_0.4.2    generics_0.1.3    knitr_1.42        forcats_1.0.0    
## [17] tibble_3.2.1      bookdown_0.35     munsell_0.5.0     bslib_0.5.1       pillar_1.9.0      rlang_1.1.1       utf8_1.2.3        stringi_1.7.12   
## [25] cachem_1.0.8      xfun_0.40         sass_0.4.7        cli_3.6.1         withr_2.5.0       magrittr_2.0.3    digest_0.6.33     grid_4.3.0       
## [33] rstudioapi_0.15.0 lifecycle_1.0.3   vctrs_0.6.3       evaluate_0.20     glue_1.6.2        farver_2.1.1      blogdown_1.18     ragg_1.2.5       
## [41] fansi_1.0.4       colorspace_2.1-0  rmarkdown_2.20    tools_4.3.0       pkgconfig_2.0.3   htmltools_0.5.6

Automation

Wed, 05 Jul 2023 00:00:00 +0000

Efficiency

Wed, 05 Jul 2023 00:00:00 +0000

Efficiency and Consistency: Automate Subset Graphics with ggplot2 and purrr

Wed, 05 Jul 2023 00:00:00 +0000

Header image by Purrple Cat

During a consulting session a few weeks ago, we discussed automated plot generation with {ggplot2}. What does that mean? If you have a bunch of variables, you may want to create a set of explorative charts for different variables. Or you may want to create a set of explanatory charts, one for each category of your data set. For both use cases, this requires redundant work as the plots use almost the same code to generate the visualizations.

Instead of copying and pasting the same code and adjusting the variables, iterate over a vector of groups (variables, categories, numeric ranges) to generate the same visual for different data sets by using a custom function.

There are several use cases why you may want to use a functional programming approach to generate the same chart for different data subsets:

explore distributions or relationships of different variables
communicate results for different groups
generate charts for custom reports using various data (sub)sets

There is a great blogpost by Ariel Muldoon on automating explorative plots by iterating over different variables. This blog post here focuses more on the generation of polished charts to visualize subsets of the same data set. In the end, I will also share some examples for working with variables and some shortcuts for exploring data sets visually.

Table of Content

The Setup

We are going to visualize relationships for different numeric variables of the mpg data set which features fuel economy data of popular car models for different years, manufacturers, and car types. In this tutorial, we are using only data from 2008.

For some data-wrangling steps, we make use of the {dplyr} packages. To visualize the data, we use the packages {ggplot2} and {patchwork}. We will also use some functions of other packages, namely {tidyr}, {stringr}, {prismatic} and {ggforce} which we address via the namespace.

Unfold to explore setup steps

library(ggplot2)   ## for plotting
library(purrr)     ## for iterative tasks
library(dplyr)     ## for data wrangling
library(patchwork) ## for multi-panel plots

## customize plot style
theme_set(theme_minimal(base_size = 15, base_family = "Anybody"))
theme_update(
  axis.title.x = element_text(margin = margin(12, 0, 0, 0), color = "grey30"),
  axis.title.y = element_text(margin = margin(0, 12, 0, 0), color = "grey30"),
  panel.grid.minor = element_blank(),
  panel.border = element_rect(color = "grey45", fill = NA, linewidth = 1.5),
  panel.spacing = unit(.9, "lines"),
  strip.text = element_text(size = rel(1)),
  plot.title = element_text(size = rel(1.4), face = "bold", hjust = .5),
  plot.title.position = "plot"
)

## adjust data set
mpg <-
  ggplot2::mpg |> 
  filter(year == 2008) |> 
  mutate(manufacturer = stringr::str_to_title(manufacturer))

The Use Case

Let’s visualize the 2008 car fuel data and explore the relationship of displacement and highway miles per gallon per manufacturer.

g <- 
  ggplot(mpg, aes(x = hwy, y = displ)) +
  scale_x_continuous(breaks = 2:8*5) +
  labs(x = "Highway miles per gallon", y = "Displacement in litres", color = NULL)

g + geom_point(aes(color = manufacturer), alpha = .5, size = 3)

A (too) colorful scatterplot of displacement versus highway miles per gallon with points colored by manufacturer.

Two issues arise here:

too many categories: the use of color encoding is not useful given the large number of manufacturers (a usual recommended limit of categorical colors is 5-8)
too many data points: the number of observations, especially with identical value combinations, leads to* overplotting and color mixing

A common solution to circumvent these issues are small multiples:

 g + 
  geom_point(alpha = .5, size = 2) +
  facet_wrap(~ manufacturer, ncol = 3)

Small multiples visualizing the relationship of displacement versus highway miles per gallon per manufaturer as a grid of scatter plots.

While it solves the mentioned issues, the resulting small multiple is likely too dense to effectively communicate the relationships for each manufacturer. Due to the large number of manufacturers, each plot also becomes rather small.

To focus on a single manufacturer, we may decide to create a plot for a subset of the data. To allow for comparison, we also plot all other car models as smaller, grey circles.

g +
  ## filter for manufacturer of interest
  geom_point(data = filter(mpg, manufacturer == "Audi"), 
             color = "#007cb1", alpha = .5, size = 4) +
  ## add shaded points for other data
  geom_point(data = filter(mpg, manufacturer != "Audi"), 
             shape = 1, color = "grey45", size = 2) +
  ## add title manually
  ggtitle("Audi")

A scatterplot highlighting the car models manufactured by Audi. The grey circles represent all other car models featured in the data set.

To communicate the relationship for all manufacturers, e.g. a dedicated section for each in a report or revealing the results step by step in a presentation, we now need to repeat the same code 15 times and replace the filter conditions and title. Or we iterate the process.

The Function

The first step is to create a custom function that takes the subset condition as an input and creates the plot with the filtered data. The code inside the user-defined function below is basically the same as the one to create the Audi chart. The only difference is that we do not specify Audi but create a placeholder, here called group, to control the filtering condition and title: In the geom_point() calls, we subset our data based on group; in the labs() function we use this string as the plot title.

plot_manufacturer <- function(group) {
  
  ## check if input is valid
  if (!group %in% mpg$manufacturer) stop("Manufacturer not listed in the data set.")
  
  ggplot(mapping = aes(x = hwy, y = displ)) +
    ## filter for manufacturer of interest
    geom_point(data = filter(mpg, manufacturer %in% group), 
               color = "#007cb1", alpha = .5, size = 4) +
    ## add shaded points for other data
    geom_point(data = filter(mpg, !manufacturer %in% group), 
               shape = 1, color = "grey45", size = 2) +
    scale_x_continuous(breaks = 2:8*5) +
    ## add title automatically based on subset choice
    labs(x = "Highway gallons", y = "Displacement", 
         title = group, color = NULL)
}

Now we can run the function manually for each manufacturer featured in the data set:

## run function for specific subsets
plot_manufacturer("Audi")
plot_manufacturer("Chevrolet")

Using the custom function, we can create plots for specific manufacturers without the need to copy-paste and adjust the ggplot code manually.

The Automation

The final step is simple: we use the map() function from the {purrr} package to iterate over a vector of manufacturers and pass the elements to our plot_manufacturer() function:

groups <- unique(mpg$manufacturer)
map(groups, ~plot_manufacturer(group = .x))

All 15 scatter plots for the manufacturers featured in the 2008 car fuel data.

A More Complex Example

Let’s consider a more complex visualization which consists of multiple plots. As before, we first check the input and filter the data. Afterwards, we create three plots and combine them with the help of the {patchwork} package.

The main plot is again a scatter plot of displacement versus highway miles per gallon. This time, we do not add the other car models to our plot. To ensure that the plots cover the full range of the data, I set the axis limits based on the variable ranges of the full data set (lims_x and lims_y in scale_x_continuous() and scale_y_continuous(), respectively). The color of the points is mapped to the car type. To avoid inconsistent coloring of the points due to different car types listed per manufacturer, I create a named color vector (pal) to match the colors by class (scale_color_manual()).

To both axes, I add marginal box plots showing the 1-D summary statistics of the variables. Again, I am applying the limits based on the full data set to match the axis ranges of the scatter plot. I also remove all non-data elements (theme_void()) and the axis guides (guide = "none" in the positional scales).

Another function argument allows to control whether the plot is saved to disk (save = TRUE) or not (save = FALSE).

plot_manufacturer_marginal <- function(group, save = FALSE) {
  
  ## check if input is valid
  if (!group %in% mpg$manufacturer) stop("Manufacturer not listed in the data set.")
  if (!is.logical(save)) stop("save should be either TRUE or FALSE.")
  
  ## filter data
  data <- filter(mpg, manufacturer %in% group)
  
  ## set limits
  lims_x <- range(mpg$hwy) 
  lims_y <- range(mpg$displ)
  
  ## define colors
  pal <- RColorBrewer::brewer.pal(n = n_distinct(mpg$class), name = "Dark2")
  names(pal) <- unique(mpg$class)
  
  ## scatter plot
  main <- ggplot(data, aes(x = hwy, y = displ, color = class)) +
    geom_point(size = 3, alpha = .5) +
    scale_x_continuous(limits = lims_x, breaks = 2:8*5) +
    scale_y_continuous(limits = lims_y) +
    scale_color_manual(values = pal, name = NULL) +
    labs(x = "Highway miles per gallon", y = "Displacement") +
    theme(legend.position = "bottom")
  
  ## boxplots
  right <- ggplot(data, aes(x = manufacturer, y = displ)) +
    geom_boxplot(linewidth = .7, color = "grey45") +
    scale_y_continuous(limits = lims_y, guide = "none", name = NULL) +
    scale_x_discrete(guide = "none", name = NULL) +
    theme_void()
  
  top <- ggplot(data, aes(x = hwy, y = manufacturer)) +
    geom_boxplot(linewidth = .7, color = "grey45") +
    scale_x_continuous(limits = lims_x, guide = "none", name = NULL) +
    scale_y_discrete(guide = "none", name = NULL) +
    theme_void()
  
  ## combine plots
  p <- top + plot_spacer() + main + right + 
    plot_annotation(title = group) + 
    plot_layout(widths = c(1, .05), heights = c(.1, 1))
  
  ## save multi-panel plot
  if (isTRUE(save)) {
    ggsave(p, filename = paste0(group, ".pdf"), 
           width = 6, height = 6, device = cairo_pdf)
  }
  
  return(p)
}

Now we can apply the function to manufacturers line by line:

plot_manufacturer_marginal("Dodge")

There are 11 cars listed that are produced by Dodge: 5 minivans, 12 pickups, and 4 SUVs.

plot_manufacturer_marginal("Nissan")

The 7 Nissan cars belong either to the midsize or the SUV class.

Or we iterate over the vector of manufacturers with the {purrr} package.

map(groups, ~plot_manufacturer_marginal(.x))

If you wish to only save the visualizations but not plot them, use the walk() function:

walk(groups, ~plot_manufacturer_marginal(.x, save = TRUE))

An Example with Variables as Inputs

To wrap up, let’s consider the use case of exploring the data set. We create a general function that works with any data set and two numeric variables. Based on these three inputs, the function generates a scatter plot with a linear fitting.

In addition, we allow to define a variable to encode points by color as well as to control the size and transparency of the points. If the user passes a column for color encoding, we either use (i) a categorical palette and linear fittings for each group for qualitative variables or (ii) a sequential palette with a single smoothing line for quantitative variables.

plot_scatter_lm <- function(data, var1, var2, pointsize = 2, transparency = .5, color = "") {
  
  ## check if inputs are valid
  if (!exists(substitute(data))) stop("data needs to be a data frame.")
  if (!is.data.frame(data)) stop("data needs to be a data frame.")
  if (!is.numeric(pull(data[var1]))) stop("Column var1 needs to be of type numeric, passed as string.")
  if (!is.numeric(pull(data[var2]))) stop("Column var2 needs to be of type numeric, passed as string.")
  if (!is.numeric(pointsize)) stop("pointsize needs to be of type numeric.")
  if (!is.numeric(transparency)) stop("transparency needs to be of type numeric.")
  if (color != "") { if (!color %in% names(data)) stop("Column color needs to be a column of data, passed as string.") }
  
  g <- 
    ggplot(data, aes(x = !!sym(var1), y = !!sym(var2))) +
    geom_point(aes(color = !!sym(color)), size = pointsize, alpha = transparency) +
    geom_smooth(aes(color = !!sym(color), color = after_scale(prismatic::clr_darken(color, .3))), 
                method = "lm", se = FALSE) +
    theme_minimal(base_family = "Roboto Condensed", base_size = 15) +
    theme(panel.grid.minor = element_blank(),
          legend.position = "top")
   
  if (color != "") { 
    if (is.numeric(pull(data[color]))) {
      g <- g + scale_color_viridis_c(direction = -1, end = .85) +
        guides(color = guide_colorbar(
          barwidth = unit(12, "lines"), barheight = unit(.6, "lines"), title.position = "top"
        ))
    } else {
      g <- g + scale_color_brewer(palette = "Set2")
    }
  }
    
  return(g)
}

Because I want (better have to) pass the aesthetics as strings with {purrr}, I first turn the input into symbols by using the sym() function and then those get evaluated with !! (so-called “bang-bang”). Note that, in case you just want to use the function without iteration, you can also use {{ }} (so-called embracing) to pass unquoted variables.

Unfold to see {{ }} example

plot_scatter_lm_embraced <- function(data, var1, var2, color = NULL) {
  
  v1 <- deparse(substitute(var1))
  v2 <- deparse(substitute(var2))
  v3 <- deparse(substitute(color))
  
  ## check if inputs are valid
  if (!exists(substitute(data))) stop("data needs to be a data frame.")
  if (!is.data.frame(data)) stop("data needs to be a data frame.")
  if (!v1 %in% names(data)) stop("Column var1 needs to be a column of data.")
  if (!v2 %in% names(data)) stop("Column var2 needs to be a column of data.")
  if (!is.numeric(pull(data[v1]))) stop("Column var1 needs to be of type numeric.")
  if (!is.numeric(pull(data[v2]))) stop("Column var2 needs to be of type numeric.")
  if (!v3 %in% c(names(data), "NULL")) stop("Column color needs to be a column of data.")
  
  g <- 
    ggplot(data, aes(x = {{ var1 }}, y = {{ var2 }})) +
    geom_point(aes(color = {{ color }}), alpha = .5) +
    geom_smooth(aes(color = {{ color }}), method = "lm", se = FALSE) +
    theme_minimal(base_family = "Roboto Condensed", base_size = 15) +
    theme(panel.grid.minor = element_blank(),
          legend.position = "top")
   
  if (v3 != "NULL") {
    if (is.numeric(pull(data[v3])))  {
      g <- g + scale_color_viridis_c(direction = -1, end = .85) +
        guides(color = guide_colorbar(
          barwidth = unit(12, "lines"), barheight = unit(.6, "lines"), title.position = "top"
        ))
    } else {
      g <- g + scale_color_brewer(palette = "Set2")
    }
  }
  
  return(g)
}

plot_scatter_lm_embraced(data = mpg, var1 = displ, var2 = hwy)

An example using the function with embracing without passing a color column.

plot_scatter_lm_embraced(data = mpg, var1 = displ, var2 = hwy, color = cyl)

An example using the function with embracing, passing a quantitative variable to color.

plot_scatter_lm_embraced(data = mpg, var1 = displ, var2 = hwy, color = class)

An example using the function with embracing, passing a qualitative variable to color.

To iterate over the function, we have two options:

we can fix one variable and pass the other as vector with map()
we can pass two variables as vectors to var1 and var2 with map2()

Let’s consider an example in which we vary both positional variables, using the 2008 car fuel data:

map2(
  c("displ", "displ", "hwy"), 
  c("hwy", "cty", "cty"),
  ~plot_scatter_lm(
    data = mpg, var1 = .x, var2 = .y, 
    color = "cyl", pointsize = 3.5
  )
)

Scatter plot showing the linear trends for different combinations of displ, cty, and hwy for the 2008 car fuel data. As the number of cylinders is encoded as numeric, the sequential green-purple viridis scale is applied to the points.

Now we are going to apply our function to a different data set, the Palmer penguins to visualize x-y relationships per species. We automatically iterate over all combinations of a set of chosen numeric variables: we first generate a vector containing the column names of interest (names) and then create a data frame with all possible combinations (names_set) with the help of expand_grid() from the {tidyr} package.

## set up variables of interest
names <- c("bill_length_mm", "bill_depth_mm", "flipper_length_mm", "body_mass_g")
## ... and create all possible combinations
names_set <- tidyr::expand_grid(names, names)

Using another mapping function from the {purrr} package called pmap(), we can map over multiple arguments simultaneously:

library(palmerpenguins)

pmap(
  names_set, ~plot_scatter_lm(
    data = penguins, 
    var1 = .x, var2 = .y, color = "species"
  )
)

Scatter plots for each combination of four numeric variables of the Palmer penguins data set. Passing the categorical variable species to color in our custom function returns linear smoothing lines for each of the three groups which are encoded with a categorical palette from the ColorBrewer project.

Shortcuts for Explorative Plots

`plot()`

I am not a fan of base R plots but one big advantage is the behavior when applying the plot() function to a full data set. The output is a grid of plots showing the relationship of all variable combinations:

plot(penguins)

Running plot() on a data frame returns a grid of scatter plots visualizing the x-y relationships of all possible combinations of variables.

`facet_matrix()`

A similar, and more flexible take on this idea is the facet_matrix() functionality from the {ggforce} package. Instead of passing variable names in the aesthetics, we specify placeholders called .panel_x and .panel_y. We then create our ggplot as usual by adding layers and scales.

Finally, we specify the variables to use in the facet inside the facet_matrix() component. You can also specify specific layers for different areas (upper, diagonal, lower) inside facet_matrix():

ggplot(penguins, aes(x = .panel_x, y = .panel_y)) +
  geom_point(aes(color = species), alpha = .5) +
  geom_smooth(aes(color = species), method = "lm") +
  ggforce::geom_autodensity(aes(color = species, fill = after_scale(color)), alpha = .7) +
  scale_color_brewer(palette = "Set2", name = NULL) +
  ggforce::facet_matrix(vars(names), layer.lower = 2, layer.diag = 3)

facet_matrix() from the {ggforce} package offers a great way to generate similar grids with {ggplot2}, allowing for the usual flexibility in terms of fine-tuning and polishing. By specifying different layers for the three areas , we can combine all kind of chart types to visualize the relationships of the variables.

R Session Info

## R version 4.2.3 (2023-03-15)
## Platform: x86_64-apple-darwin17.0 (64-bit)
## Running under: macOS Big Sur ... 10.16
## 
## Matrix products: default
## BLAS:   /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRlapack.dylib
## 
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] palmerpenguins_0.1.1 patchwork_1.1.2      dplyr_1.1.2          purrr_1.0.1          ggplot2_3.4.2       
## 
## loaded via a namespace (and not attached):
##  [1] tidyselect_1.2.0   xfun_0.39          bslib_0.5.0        splines_4.2.3      lattice_0.20-45    colorspace_2.1-0   vctrs_0.6.2        generics_0.1.3    
##  [9] htmltools_0.5.4    viridisLite_0.4.1  yaml_2.3.7         mgcv_1.8-42        utf8_1.2.3         rlang_1.1.1        jquerylib_0.1.4    pillar_1.9.0      
## [17] glue_1.6.2         withr_2.5.0        tweenr_2.0.2       RColorBrewer_1.1-3 lifecycle_1.0.3    stringr_1.5.0      munsell_0.5.0      blogdown_1.18     
## [25] gtable_0.3.1       ragg_1.2.5         evaluate_0.20      labeling_0.4.2     knitr_1.42         fastmap_1.1.1      fansi_1.0.4        highr_0.10        
## [33] Rcpp_1.0.10        scales_1.2.1       cachem_1.0.7       jsonlite_1.8.4     farver_2.1.1       systemfonts_1.0.4  textshaping_0.3.6  ggforce_0.4.1     
## [41] digest_0.6.31      stringi_1.7.12     bookdown_0.34      polyclip_1.10-4    grid_4.2.3         cli_3.6.0          tools_4.2.3        magrittr_2.0.3    
## [49] sass_0.4.5         tibble_3.2.1       tidyr_1.3.0        pkgconfig_2.0.3    MASS_7.3-58.2      Matrix_1.5-3       rmarkdown_2.20     rstudioapi_0.14   
## [57] R6_2.5.1           prismatic_1.1.1    nlme_3.1-162       compiler_4.2.3

Tips

Wed, 05 Jul 2023 00:00:00 +0000

Workflow

Wed, 05 Jul 2023 00:00:00 +0000

Challenge

Tue, 28 Mar 2023 00:00:00 +0000

Community

Tue, 28 Mar 2023 00:00:00 +0000

Cédric Scherer

30DayChartChallenge

Cédric Scherer

DataViz

DataWrapper

Exploring "The Simpsons" with DataWrapper for the 30DayChartChallenge 2024

What’s the Challenge About?

Self-Imposed Rules

Day 1: Part-to-Whole

Day 2: Neo*

Day 4: Waffle

Day 5: Diverging

Day 9: Major/Minor

Day 10: Physical

Day 11: Mobile-friendly

Stacked Bars for Members of the “No Pie Chart Gang”

Day 13: Family

Day 14: Heatmap

Day 19: Dinosaurs

Day 20: Correlation

Day 22: Mobility

Day 28: Trend

Day 29: Black’n’White (or Day 27: Good/Bad)

Posts

R

Tags

Annotations

Ggplot2

Tidyverse

Tutorial

Yet Another How-to on Labelling Bar Graphs in ggplot2

Table of Content

Introduction

Data Preparation

Create a Basic Bar Chart

Style the Visualization

Place Category Labels on the Top

Bonus: Style the Bars

Highlight Top-Ranked Category

Add Percentages as Direct Labels

Alternative Approach

Conclusion

Note on the Header Image

Automation

Efficiency

Efficiency and Consistency: Automate Subset Graphics with ggplot2 and purrr

Table of Content

The Setup

The Use Case

The Function

The Automation

A More Complex Example

An Example with Variables as Inputs

Shortcuts for Explorative Plots

plot()

facet_matrix()

Tips

Workflow

Challenge

Community

`plot()`

`facet_matrix()`