Layers & Aesthetics

PSY 410: Data Science for Psychology

Dr. Sara Weston

2026-04-22

Setup

library(tidyverse)
library(nycflights13)
library(showtext)

showtext_auto()  # lets ggplot render Unicode symbols like ◀ ▶

set.seed(422)
some_flights <- flights |>
  filter(carrier %in% c("9E", "B6", "HA")) |>
  filter(!is.na(air_time)) |>
  group_by(carrier) |>
  slice_sample(n = 100)

min_time <- min(some_flights$air_time)
max_time <- max(some_flights$air_time)

This first section (~25 min total) opens with a layered progression — each slide adds one thing to the previous figure. Walk through them without narrating the code; the code is folded on each slide so students can expand it later if curious. Every function (stat_summary, geom_jitter, annotate, scale_fill_manual, etc.) is taught later in the deck. Right now, the point is the visual story: what each layer reveals and what a barer version was hiding.

Base figure

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  stat_summary(fun = "mean", geom = "bar")

Remove legend and turn sideways

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  stat_summary(fun = "mean", geom = "bar") +
  guides(fill = "none") +
  coord_flip()

Boxplot

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_boxplot() +
  guides(fill = "none") +
  coord_flip()

Points

some_flights |>
  ggplot(aes(x = carrier, y = air_time)) +
  geom_point() +
  coord_flip()

Jitter

some_flights |>
  ggplot(aes(x = carrier, y = air_time)) +
  geom_jitter(width = .2) +
  coord_flip()

Colored

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  coord_flip()

Cleaner theme

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  coord_flip() +
  theme_minimal()

Better colors, remove legend

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  scale_fill_manual(values = c("9E" = "#15616d", "B6" = "#ff7d00", "HA" = "#78290f")) +
  guides(fill = "none") +
  coord_flip() +
  theme_minimal()

Clean up labels

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  scale_y_continuous(labels = scales::label_number(suffix = " min")) +
  scale_fill_manual(values = c("9E" = "#15616d", "B6" = "#ff7d00", "HA" = "#78290f")) +
  guides(fill = "none") +
  labs(x = NULL, y = NULL, title = "Air time by carrier") +
  coord_flip() +
  theme_minimal()

Reference lines at min and max

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  geom_hline(yintercept = min_time, linetype = "dashed") +
  geom_hline(yintercept = max_time, linetype = "dashed") +
  scale_y_continuous(labels = scales::label_number(suffix = " min")) +
  scale_fill_manual(values = c("9E" = "#15616d", "B6" = "#ff7d00", "HA" = "#78290f")) +
  guides(fill = "none") +
  labs(x = NULL, y = NULL, title = "Air time by carrier") +
  coord_flip() +
  theme_minimal()

Annotate the extremes

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  geom_hline(yintercept = min_time, linetype = "dashed", alpha = .4) +
  geom_hline(yintercept = max_time, linetype = "dashed", alpha = .4) +
  annotate("text", x = 3.5, y = min_time, hjust = 0,
           label = paste0("◀ shortest: ", min_time, " min"),
           color = "grey30", size = 4) +
  annotate("text", x = 3.5, y = max_time, hjust = 1,
           label = paste0("longest: ", max_time, " min ▶"),
           color = "grey30", size = 4) +
  scale_x_discrete(expand = expansion(add = c(0.6, 1))) +
  scale_y_continuous(labels = scales::label_number(suffix = " min")) +
  scale_fill_manual(values = c("9E" = "#15616d", "B6" = "#ff7d00", "HA" = "#78290f")) +
  guides(fill = "none") +
  labs(x = NULL, y = NULL, title = "Air time by carrier") +
  coord_flip(clip = "off") +
  theme_minimal()

Carrier labels inline

carrier_labels <- some_flights |>
  group_by(carrier) |>
  summarize(mean_time = mean(air_time)) |>
  mutate(name = c("9E • Endeavor Air", "B6 • JetBlue", "HA • Hawaiian Airlines"))

some_flights |>
  ggplot(aes(x = carrier, y = air_time, fill = carrier)) +
  geom_jitter(width = .2, shape = 21, color = "black", stroke = .4) +
  geom_hline(yintercept = min_time, linetype = "dashed", alpha = .4) +
  geom_hline(yintercept = max_time, linetype = "dashed", alpha = .4) +
  annotate("text", x = 3.5, y = min_time, hjust = 0,
           label = paste0("◀ shortest: ", min_time, " min"),
           color = "grey30", size = 4) +
  annotate("text", x = 3.5, y = max_time, hjust = 1,
           label = paste0("longest: ", max_time, " min ▶"),
           color = "grey30", size = 4) +
  geom_text(data = carrier_labels,
            aes(x = carrier, y = mean_time, label = name),
            inherit.aes = FALSE,
            hjust = 0.5, nudge_x = -0.35,
            fontface = "bold", size = 4.2) +
  geom_text(data = carrier_labels,
            aes(x = carrier, y = mean_time,
                label = paste0("avg: ", round(mean_time), " min")),
            inherit.aes = FALSE,
            hjust = 0.5, nudge_x = -0.6,
            size = 3.5, color = "grey30") +
  scale_x_discrete(expand = expansion(add = c(0.8, 1)), labels = NULL) +
  scale_y_continuous(labels = scales::label_number(suffix = " min")) +
  scale_fill_manual(values = c("9E" = "#15616d", "B6" = "#ff7d00", "HA" = "#78290f")) +
  guides(fill = "none") +
  labs(x = NULL, y = NULL, title = "Air time by carrier") +
  coord_flip(clip = "off") +
  theme_minimal()

Choosing the right geom

Geom selection guide

Your data	Good geom	Avoid
One continuous variable	geom_histogram(), geom_density()	geom_bar()
One categorical variable	geom_bar()	geom_histogram()
Two continuous	geom_point(), geom_smooth()	—
One continuous + one categorical	geom_boxplot(), geom_violin()	pie charts
Two categorical	geom_count(), geom_tile()	—
Change over time	geom_line()	geom_point() alone

This table is a reference slide students will come back to. Don’t read every row aloud — highlight the “Avoid” column. The most common mistake is using geom_bar() for continuous data. Also note that pie charts are absent from the “Good geom” column entirely. Students will ask about pie charts; the short answer is that position along a common axis (bar chart) is easier for the human eye to compare than angle (pie chart).

One variable: distributions

flights_sample |>
  ggplot(aes(x = dep_delay)) +
  geom_histogram(
    binwidth = 10,
    fill = "steelblue",
    color = "white") +
  labs(
    title = "Distribution of Departure Delays",
    x = "Departure delay (min)"
  ) +
  theme_minimal(base_size = 14)

Walk through the code line by line. Emphasize binwidth — students will forget it and get the default “30 bins” warning. A good rule of thumb: try a few binwidths and see what reveals the shape without being too choppy or too smooth. The right skew is a great teaching moment — real-world delay data almost always looks like this. Connect it back to psychology: reaction time distributions look similar.

One variable: distributions

histogram vs density

Histogram = actual counts. Density = smoothed estimate of the shape.

Quick comparison — don’t belabor this. The key point is that density plots are better for comparing distributions across groups (you can overlay them), while histograms are better for seeing exact counts. Students sometimes get confused by the y-axis on density plots; reassure them that the area under the curve sums to 1, but they rarely need to interpret the y-axis value itself.

One categorical variable

flights_sample |>
  ggplot(aes(x = origin)) +
  geom_bar(fill = "steelblue") +
  labs(
    title = "Flights per NYC Airport",
    x = "Origin airport",
    y = "Count"
  ) +
  theme_minimal(base_size = 14)

One categorical variable

One continuous + one categorical

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay, fill = origin)) +
  geom_boxplot(alpha = 0.7) +
  labs(
    title = "Departure Delay by Airport",
    x = "Origin airport",
    y = "Departure delay (min)"
  ) +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

One continuous + one categorical

boxplot vs violin

Violin shows the full distribution shape. Boxplot shows summary stats. Both have their place.

Students may not have seen violin plots before. Point out that a violin is essentially a mirrored density plot — it shows the full distribution shape, which a boxplot hides. The right-skewed tail of delays is much more obvious in the violin than the boxplot. Mention that we’ll see an even better option (raincloud plots) in the next session.

Showing individual points: geom_jitter()

When your x-axis is categorical, geom_point() would stack every observation on the same vertical line. geom_jitter() adds a small random horizontal wiggle so each point is visible.

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay, color = origin)) +
  geom_jitter(width = 0.2, alpha = 0.4) +
  labs(
    title = "Every Flight, Plotted",
    x = "Origin airport",
    y = "Departure delay (min)"
  ) +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

width = 0.2 keeps points within each category’s “lane.” alpha = 0.4 makes them semi-transparent so dense areas look darker. Flag both settings — students will want to tune them. We’ll see geom_jitter() come back as the “raw data” layer under every summary plot in the rest of the deck.

Showing individual points: geom_jitter()

stat_summary()

The psychology staple: means + error bars

In psych papers, you’ll see this constantly: a bar chart showing group means with error bars. stat_summary() is how you make it.

Transition: “So far we’ve been plotting raw data. But in psychology, you’ll almost always need to show summary statistics — means, error bars, confidence intervals. That’s what stat_summary() is for.” This is the most psych-relevant section of the deck. Spend ~8 minutes on stat_summary() total. Students will use this constantly in assignments and the final project.

stat_summary() basics

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay)) +
  stat_summary(
    fun = mean,
    geom = "bar",
    fill = "steelblue", width = 0.5) +
  stat_summary(
    fun.data = mean_cl_normal,
    geom = "errorbar",
    width = 0.3) +
  labs(
    title = "Mean Departure Delay by Airport",
    x = "Origin airport",
    y = "Departure delay (min)"
  ) +
  theme_minimal(base_size = 14)

stat_summary() basics

Breaking down stat_summary()

# The bar (mean)
stat_summary(fun = mean, geom = "bar")

# The error bars (95% confidence interval)
stat_summary(fun.data = mean_cl_normal, geom = "errorbar")

• fun = the function to calculate (mean, median, etc.)
• fun.data = a function that returns ymin, y, ymax (like mean_cl_normal)
• geom = what shape to use to show the result

The fun vs fun.data distinction trips students up. Emphasize: fun returns one number (e.g., the mean), while fun.data returns three (ymin, y, ymax) — which is what you need for error bars. Students don’t need to memorize this; they need to know the pattern: one stat_summary() for the bar, another for the error bars.

What are error bars showing?

This matters! Always state it in your figure caption.

Error bar type	What it means	How to get it
SE (standard error)	Precision of the mean	mean_se
SD (standard deviation)	Spread of the data	Write your own
95% CI	Confidence interval	mean_cl_normal

# SE
stat_summary(fun.data = mean_se, geom = "errorbar")

# 95% CI (most common in psych)
stat_summary(fun.data = mean_cl_normal, geom = "errorbar")

Why does this matter?

The APA Publication Manual (7th ed.) recommends showing individual data points alongside summary statistics whenever possible.

Bar charts with error bars are ubiquitous in psychology — but they hide the shape of the data. Outliers, skew, and bimodality all disappear behind a rectangle.

This is a great moment to connect to the replication crisis discussion from Session 1. One reason problematic results went undetected for so long is that bar-and-error-bar plots hid the underlying data. If reviewers had seen the individual data points, they might have noticed impossible patterns (like the LaCour data). The delay data makes the same point — the means look nearly identical, but the underlying distributions are wildly right-skewed.

Adding individual data points

Error bars alone hide the data. Show the points too:

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay, color = origin)) +
  geom_jitter(width = 0.15, alpha = 0.4, size = 2) +
  stat_summary(fun = mean, geom = "point", size = 4, color = "black") +
  stat_summary(fun.data = mean_cl_normal, geom = "errorbar", width = 0.2, color = "black") +
  labs(
    title = "Departure Delay by Airport",
    subtitle = "Error bars = 95% CI. Individual flights shown.",
    x = "Origin airport",
    y = "Departure delay (min)"
  ) +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

Walk through the layering order: jittered points go first (behind), then the summary point and error bars on top. Emphasize that alpha = 0.4 makes individual points semi-transparent so they don’t overwhelm the summary. This is the pattern students should use for most psych figures going forward: show the data AND the summary.

Adding individual data points

Pair coding break

Your turn: 10 minutes

Give 10 minutes. Circulate actively — the most common issues will be: (1) forgetting to use stat_summary() twice (once for bar, once for error bars), (2) putting geom_jitter() after the bar so the points are hidden behind it, and (3) not knowing where to put the caption in labs(). If pairs finish early, challenge them to add time_of_day as a facet or fill variable. Reveal the solution after time is up. Students should create their own flights_sample using the setup code shown on this slide (or you can share it in the chat).

Using the flights dataset (filter to !is.na(dep_delay) first):

1. Create a bar chart with error bars showing mean dep_delay by origin
2. Add individual data points (jittered) behind the bars
3. Color the bars by origin
4. Add a caption noting that error bars show the 95% CI

Tip

You’ll need stat_summary() twice — once for the bar, once for the error bars. Look at the examples from the last few slides.

Solution: Mean departure delay by airport

flights |>
  filter(!is.na(dep_delay)) |>
  ggplot(aes(x = origin, y = dep_delay, fill = origin)) +
  geom_jitter(width = 0.15, alpha = 0.2) +
  stat_summary(fun = mean, geom = "bar", alpha = 0.6, width = 0.5) +
  stat_summary(fun.data = mean_cl_normal, geom = "errorbar", width = 0.2) +
  labs(
    title = "Mean Departure Delay by NYC Airport",
    x = "Origin airport",
    y = "Departure delay (min)",
    caption = "Error bars represent 95% confidence intervals."
  ) +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

Solution: output

Position & scales

This second lecture block (~20 min) covers position adjustments, scales, coordinate systems, and the polished figure. This is more about refinement than new geoms — students already know the building blocks, now they learn how to control them precisely.

Position adjustments

When geoms overlap, position adjustments fix it:

Position	What it does	When to use
"dodge"	Side by side	Grouped bar charts
"stack"	Stacked on top	Stacked bars
"fill"	Stacked to 100%	Comparing proportions
"jitter"	Random wiggle	Overplotted points

dodge: grouped bar charts

This is the trickiest position adjustment. The key gotcha: when you dodge both the bars and the error bars, they need to dodge by the same width. That’s why we use position_dodge(0.6) for the error bars to match width = 0.6 on the bars. If these don’t match, the error bars float between the bars. Students will hit this bug — it’s a good one to flag proactively.

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay, fill = time_of_day)) +
  stat_summary(fun = mean, geom = "bar", position = "dodge", width = 0.6) +
  stat_summary(fun.data = mean_cl_normal, geom = "errorbar",
               position = position_dodge(0.6), width = 0.2) +
  labs(
    title = "Departure Delay by Airport and Time of Day",
    x = "Origin airport",
    y = "Mean delay (min)",
    fill = "Time of day"
  ) +
  theme_minimal(base_size = 14)

dodge: grouped bar charts

stack and fill

Scales: controlling axes and colors

Spend ~3 minutes on this and the next slide together. Students don’t need to memorize every scale function — they need to know the naming pattern: scale_ + the aesthetic (fill, color, x, y) + _ + the type (manual, continuous, discrete). Once they see the pattern, they can guess function names and use autocomplete.

Scales translate data values into visual properties:

# Control axis breaks (tick marks)
scale_x_continuous(breaks = seq(0, 180, by = 30))

# Set specific colors
scale_fill_manual(values = c("EWR" = "lightblue", "JFK" = "coral", "LGA" = "lightgreen"))

scale_fill_manual()

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay, fill = origin)) +
  geom_boxplot(alpha = 0.7) +
  scale_fill_manual(values = c("EWR" = "#3498db", "JFK" = "#e74c3c", "LGA" = "#f1c40f")) +
  labs(title = "Custom colors", x = "Origin airport", y = "Delay (min)") +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

Use a website like HTML color codes to find the appropriate HEX code for your color.

scale_fill_manual()

Coordinate systems

The coord_cartesian() vs scale_y_continuous(limits = ...) distinction is critical and students will confuse them. The key: coord_cartesian() zooms the view (like pinching to zoom on a photo), while scale_y_continuous(limits = ...) filters the data first and then plots. This means boxplot whiskers, smooth lines, and error bars can all change when you use scale limits because outliers are removed before the calculation. Always use coord_cartesian() for zooming.

# Flip x and y (great for long category labels)
coord_flip()

# Zoom in without dropping data
coord_cartesian(ylim = c(-10, 30))
# vs
scale_y_continuous(limits = c(-10, 30))  # This DROPS data outside the range!

coord_flip() in action

flights_sample |>
  ggplot(aes(y = origin, x = dep_delay, fill = origin)) +
  geom_boxplot(alpha = 0.7) +
  labs(title = "Flipped axes — better for category labels", y = "Origin airport", x = "Delay (min)") +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

coord_flip() in action

Putting it together

A polished figure

This is the “bringing it all together” slide. Walk through the code slowly, pointing out each layer and why it’s there: jittered points for raw data, boxplot for summary, stat_summary diamond for the mean, manual colors for distinction, and informative labels. This is roughly the quality you’d expect in a journal submission. Students should aim for this level on Assignment 4 and the final project.

flights_sample |>
  ggplot(aes(x = origin, y = dep_delay, fill = origin, color = origin)) +
  geom_jitter(width = 0.2, alpha = 0.35, size = 1.5) +
  # outlier.shape = NA hides boxplot outliers because geom_jitter already shows every point
  geom_boxplot(alpha = 0.5, width = 0.4, outlier.shape = NA) +
  # shape = 18 is a solid diamond — draws attention to the mean
  stat_summary(fun = mean, geom = "point", shape = 18, size = 5, color = "black") +
  scale_fill_manual(values = c("EWR" = "#3498db", "JFK" = "#e74c3c", "LGA" = "#f1c40f")) +
  scale_color_manual(values = c("EWR" = "#2980b9", "JFK" = "#c0392b", "LGA" = "#b7950b")) +
  labs(
    title = "Departure Delays Are Similar Across NYC Airports",
    subtitle = "Individual flights, box summaries, and mean (diamond) shown",
    x = "Origin airport",
    y = "Departure delay (min)",
    caption = "N = 500 sampled flights. Diamond = mean."
  ) +
  theme_minimal(base_size = 14) +
  theme(legend.position = "none")

A polished figure

Get a head start

Assignment 4 preview

This is the head-start exercise (~10 min remaining). Students don’t need to finish this in class — it’s a bridge to Assignment 4. Encourage them to have fun making the “bad” version deliberately awful. If students seem finished early, suggest they try adding facet_wrap(~time_of_day) to any of today’s plots as a preview of faceting.

Assignment 4 will ask you to create “bad” and “good” versions of figures. Start experimenting:

1. Take any plot from today and make it deliberately bad — wrong geom, missing labels, confusing colors
2. Then fix it. What did you change and why?
3. Try creating the same data with three different geoms. Which one communicates best?

Wrapping up

Today’s toolkit

Tool	What it does
geom_histogram()	Distribution of one continuous variable
geom_density()	Smooth distribution estimate
geom_boxplot()	Summary of continuous by categorical
geom_violin()	Full distribution by categorical
geom_jitter()	Individual points with horizontal wiggle
stat_summary()	Calculate and display summary stats
position = "dodge"	Side-by-side grouped plots
scale_fill_manual()	Custom colors
coord_flip()	Swap axes

Before next class

📖 Read:

• Supplementary: Visual perception principles (will be posted)
• Optional: Knaflic, Storytelling with Data, Ch 1–3

✅ Practice:

• Create a bar chart with error bars for a dataset of your choice
• Try boxplot vs violin on the same data
• Experiment with coord_flip() and scale_fill_manual()

Key takeaways

1. Match your geom to your data — the wrong choice misleads
2. stat_summary() is powerful — means, error bars, custom functions
3. Position adjustments handle overlap (dodge, stack, fill, jitter)
4. Scales control axes, colors, and legends
5. coord_cartesian() zooms; scale limits drop data — know the difference

The one thing to remember

The difference between a chart and a figure is intention — every layer should earn its place.

Next time: Perception & Design

End on this line and let it land. Next session builds directly on this one — we go from “which geom” to “which color, which theme, which design decisions.” Students should have R4DS Ch 9 fresh in their minds and the optional perception reading done before Monday.