Communication via Data Science

Lecture 10

Katie Solarz

Duke University
STA 199 Summer 2026: Session I

May 8, 2026

Wrap-Up AE-09

Project

The bottom line, at the top

Cohesive, thoughtful analysis of a dataset of your team’s choosing (subject to instructor / TA approval) drawing on & expanding upon the techniques / methods learned in this course

  • Goal: Create a reproducible written report that introduces your research question, explains the methodology, showcases results, and discusses the implications of your work; present your work / findings to your peers and teaching team with engagig, concise slides
  • Put differently… write an in-depth article that might appear in the popular press (NYT, Chronicle, etc)…
  • For an audience that is intelligent, but non-technical and unfamiliar with the domain
  • Emphasize clear communication, attractive and informative graphics, and carefully chosen numerical summaries and / or statistical methods
  • R code and raw output should not appear in your final writeup (i.e., you will need to suppress your code chunks via Quarto settings)
  • All figures and tables should be labeled with captions

Telling a story

Setup

library(tidyverse)
library(tidymodels)
library(ggrepel)
library(ggthemes)
library(palmerpenguins)

Multiple ways of telling a story

  • Sequential reveal: Motivation, then resolution

  • Instant reveal: Resolution, and motivation hidden within

Simplicity vs. complexity

When you’re trying to show too much data at once you may end up not showing anything.

  • Never assume your audience can rapidly process complex visual displays

  • Don’t add variables to your plot that are tangential to your story

  • Don’t jump straight to a highly complex figure; first show an easily digestible subset (e.g., show one facet first)

  • Aim for memorable, but clear

Project note: Make sure to leave time to iterate on your plots after you practice your presentation. If certain plots or outputs are getting too wordy to explain, take time to simplify them!

Consistency vs. repetitiveness

Be consistent but don’t be repetitive.

  • Use consistent features throughout plots (e.g., same color represents same level on all plots)

  • Aim to use a different type of summary or visualization for each distinct analysis

  • Reading a report with ALL boxplots is like walking into an ice cream shoppe that only sells versions of vanilla (e.g., Madagascar, Vanilla Bean (the best), French Vanilla, Old Fashioned, etc…) when I want a scoop of coffee and a scoop of cinnamon!

Designing effective visualizations

Data

d <- tribble(
  ~category,                     ~value,
  "Cutting tools"                , 0.03,
  "Buildings and administration" , 0.22,
  "Labor"                        , 0.31,
  "Machinery"                    , 0.27,
  "Workplace materials"          , 0.17
)
d
# A tibble: 5 × 2
  category                     value
  <chr>                        <dbl>
1 Cutting tools                 0.03
2 Buildings and administration  0.22
3 Labor                         0.31
4 Machinery                     0.27
5 Workplace materials           0.17

Keep it simple

Judging relative area

Use color to draw attention



Play with themes for a non-standard look

Go beyond ggplot2 themes – ggthemes

Tell a story

Leave out non-story details

Order matters

Clearly indicate missing data

Reduce cognitive load

Use descriptive titles

Annotate figures

Plot sizing and layout

Sample plots

p_hist <- ggplot(mtcars, aes(x = mpg)) +
  geom_histogram(binwidth = 2)

p_text <- mtcars |>
  rownames_to_column() |>
  ggplot(aes(x = disp, y = mpg)) +
  geom_text_repel(aes(label = rowname)) +
  coord_cartesian(clip = "off")

Small fig-width

For a zoomed-in look

```{r}
#| fig-width: 3
#| fig-asp: 0.618

p_hist
```

Large fig-width

For a zoomed-out look

```{r}
#| fig-width: 6
#| fig-asp: 0.618

p_hist
```

fig-width affects text size

Multiple plots on a slide

First, ask yourself, must you include multiple plots on a slide? For example, is your narrative about comparing results from two plots?

  • If no, then don’t! Move the second plot to to the next slide!

  • If yes, use columns and sequential reveal.

Quarto

Writing your project report with Quarto

  • Figure sizing: fig-width, fig-height, etc. in code chunks.

  • Figure layout: layout-ncol for placing multiple figures in a chunk.

  • Further control over figure layout with the patchwork package.

  • Chunk options around what makes it in your final report: message, echo, etc.

  • Cross referencing figures and tables.

  • Adding footnotes and citations.

Cross referencing figures

As seen in Figure 1, there is a positive and relatively strong relationship between body mass and flipper length of penguins.

ggplot(penguins, aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point()
Warning: Removed 2 rows containing missing values or values outside the scale
range (`geom_point()`).
Figure 1: The relationship between body mass and flipper length of penguins. 
As seen in @fig-penguins, there is a positive and relatively strong relationship between body mass and flipper length of penguins.

```{r}
#| label: fig-penguins

ggplot(penguins, aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point()
```

Cross referencing tables

Table 1 displays summaries of flipper length by species.

penguins |>
  group_by(species) |>
  summarize(
    Mean = mean(flipper_length_mm, na.rm = TRUE),
    Median = median(flipper_length_mm, na.rm = TRUE),
    SD = sd(flipper_length_mm, na.rm = TRUE)
  ) |>
  knitr::kable(digits = 3)
Table 1: Flipper length summaries by species
species Mean Median SD
Adelie 189.954 190 6.539
Chinstrap 195.824 196 7.132
Gentoo 217.187 216 6.485 
@tbl-penguins displays summaries of flipper length by species.

```{r}
#| label: tbl-penguins
#| tbl-cap: Flipper length summaries by species

penguins |>
  group_by(species) |>
  summarize(
    Mean = mean(flipper_length_mm, na.rm = TRUE),
    Median = median(flipper_length_mm, na.rm = TRUE),
    SD = sd(flipper_length_mm, na.rm = TRUE)
  ) |>
  knitr::kable(digits = 3)
```

Take A Sad Plot & Make It Better

Going the extra mile

ae-10-effective-dataviz

Data prep

Code
library(tidyverse)
library(scales)

staff <- read_csv("data/instructional-staff.csv")

staff_long <- staff |>
  pivot_longer(
    cols = -faculty_type, names_to = "year",
    values_to = "percentage"
  ) |>
  mutate(
    percentage = as.numeric(percentage),
    faculty_type = fct_relevel(
      faculty_type,
      "Full-Time Tenured Faculty",
      "Full-Time Tenure-Track Faculty",
      "Full-Time Non-Tenure-Track Faculty",
      "Part-Time Faculty",
      "Graduate Student Employees"
    ),
    year = as.numeric(year),
    faculty_type_color = if_else(faculty_type == "Part-Time Faculty", "firebrick1", "gray40")
  )

Pick a purpose

Code
p <- ggplot(
  staff_long,
  aes(
    x = year,
    y = percentage,
    color = faculty_type_color, group = faculty_type
    )
  ) +
  geom_line(linewidth = 1, show.legend = FALSE) +
  labs(
    x = NULL,
    y = "Percent of Total Instructional Staff",
    color = NULL,
    title = "Trends in Instructional Staff Employment Status, 1975-2011",
    subtitle = "All Institutions, National Totals",
    caption = "Source: US Department of Education, IPEDS Fall Staff Survey"
  ) +
  scale_y_continuous(labels = label_percent(accuracy = 1, scale = 1)) +
  scale_color_identity() +
  theme(
    plot.caption = element_text(size = 8, hjust = 0),
    plot.margin = margin(0.1, 0.6, 0.1, 0.1, unit = "in")
  ) +
  coord_cartesian(clip = "off") +
  annotate(
    geom = "text",
    x = 2012, y = 41, label = "Part-Time\nFaculty",
    color = "firebrick1", hjust = "left", size = 5
  ) +
  annotate(
    geom = "text",
    x = 2012, y = 13.5, label = "Other\nFaculty",
    color = "gray40", hjust = "left", size = 5
  ) +
  annotate(
    geom = "segment",
    x = 2011.5, xend = 2011.5,
    y = 7, yend = 20,
    color = "gray40", linetype = "dotted"
  )

p

Use labels to communicate the message

Code
p +
  labs(
    title = "Instruction by part-time faculty on a steady increase",
    subtitle = "Trends in Instructional Staff Employment Status, 1975-2011\nAll Institutions, National Totals",
    caption = "Source: US Department of Education, IPEDS Fall Staff Survey",
    y = "Percent of Total Instructional Staff",
    x = NULL
  )

Simplify

Code
p +
  labs(
    title = "Instruction by part-time faculty on a steady increase",
    subtitle = "Trends in Instructional Staff Employment Status, 1975-2011\nAll Institutions, National Totals",
    caption = "Source: US Department of Education, IPEDS Fall Staff Survey",
    y = "Percent of Total Instructional Staff",
    x = NULL
  ) +
  theme(panel.grid.minor = element_blank())

Summary

  • Represent percentages as parts of a whole
  • Place variables representing time on the x-axis when possible
  • Pay attention to data types; e.g., represent time as time on a continuous scale, not years as levels of a categorical variable
  • Prefer direct labeling over legends
  • Use accessible colors
  • Use color to draw attention
  • Pick a purpose and label, color, annotate for that purpose
  • Communicate your main message directly in the plot labels
  • Simplify before you call it done (a.k.a. “Before you leave the house, look in the mirror and take one thing off”)