From last time

Line graphs
Various techniques and considerations

Working with time
lubridate package

Updates

Homework
due on Tuesday
instructions!

Today

Scatter plots
Considerations, overplotting, line of best fit

Relational data
Practicing joins with dplyr

The purpose of a scatter plot

To study a relationship between two numeric variables
can also view by a group (categorical variable)
and sometimes with a third numeric variable


Line graphs (from Wednesday) are just a special kind of scatter plot
with a chronological variable (or proxy of one) on the x
and lines connecting the points to emphasize trends

Today's data

friends_info
friends_emotions
friends

Questions to examine

How does the relationship between viewers and IMDB rating look by:

season

predominant emotion of the episode

focus character

Today's data

friends_info

Basic scatter plot

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating)) +
  geom_point()

Basic scatter plot

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating)) +
  geom_point(alpha = 0.5, color = "red", size = 2)

Basic scatter plot

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating)) +
  geom_jitter()

Basic scatter plot

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, color = season)) +
  geom_jitter()

Basic scatter plot

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, 
             color = as.character(season))) +
  geom_jitter()

Basic scatter plot

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, color = season)) +
  geom_jitter(size = 2) +
  scale_colour_gradient(low = "#fafafa", high = "#191970",
                        breaks = seq(1, 10, 1))

Scatter plot with best-fit line

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating)) +
  geom_jitter(size = 2) +
  geom_smooth(method = "lm")

Scatter plot with best-fit line

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating)) +
  geom_jitter(size = 2) +
  geom_smooth(method = "lm", se = FALSE)

Scatter plot with best-fit line

friends_info %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating)) +
  geom_jitter(size = 2) +
  geom_smooth(method = "lm", level = 0.99,
              color = "purple", fill = "#DCD0FF")

Questions to examine

How does the relationship between viewers and IMDB rating look by:

season

predominant emotion of the episode

focus character

Today's data

friends_emotions

Relational data

The collective term for multiple tables of (related) data
can easily be combined thanks to joins (from dplyr)


Mutating joins: adds new variables (columns) to a data frame based on matching observations in another
possible through keys: variables that uniquely identify observations

Today's data

Today's data

friends_emotions

Data manipulation

friends_joyful_sad <- friends_emotions %>% 
  group_by(season, episode, emotion) %>%
  summarize(count = n()) %>%
  add_count(wt = count) %>% 
  mutate(percent = count / n) %>% 
  filter(emotion %in% c("Joyful","Sad")) %>% 
  select(-c(count, n)) %>% 
  pivot_wider(names_from = emotion, values_from = percent) %>% 
  mutate(Sad = replace_na(Sad, 0))

Data manipulation

Data manipulation

friends_joyful_sad <- friends_emotions %>% 
  group_by(season, episode, emotion) %>% 
  summarize(count = n()) %>% 
  add_count(wt = count) %>%
  mutate(percent = count / n) %>%
  filter(emotion %in% c("Joyful","Sad")) %>% 
  select(-c(count, n)) %>% 
  pivot_wider(names_from = emotion, values_from = percent) %>% 
  mutate(Sad = replace_na(Sad, 0))

Data manipulation

Data manipulation

friends_joyful_sad <- friends_emotions %>% 
  group_by(season, episode, emotion) %>% 
  summarize(count = n()) %>% 
  add_count(wt = count) %>% 
  mutate(percent = count / n) %>% 
  filter(emotion %in% c("Joyful","Sad")) %>%
  select(-c(count, n)) %>%
  pivot_wider(names_from = emotion, values_from = percent) %>% 
  mutate(Sad = replace_na(Sad, 0))

Data manipulation

Data manipulation

friends_joyful_sad <- friends_emotions %>% 
  group_by(season, episode, emotion) %>% 
  summarize(count = n()) %>% 
  add_count(wt = count) %>% 
  mutate(percent = count / n) %>% 
  filter(emotion %in% c("Joyful","Sad")) %>%
  select(-c(count, n)) %>% 
  pivot_wider(names_from = emotion, values_from = percent) %>%
  mutate(Sad = replace_na(Sad, 0))

Data manipulation

Data manipulation

friends_joyful_sad <- friends_emotions %>% 
  group_by(season, episode, emotion) %>% 
  summarize(count = n()) %>% 
  add_count(wt = count) %>% 
  mutate(percent = count / n) %>% 
  filter(emotion %in% c("Joyful","Sad")) %>%
  select(-c(count, n)) %>% 
  pivot_wider(names_from = emotion, values_from = percent) %>% 
  mutate(Sad = replace_na(Sad, 0))

Data manipulation

Joining data

friends_info %>% 
  left_join(friends_joyful_sad, by = c("episode","season"))

Basic scatter plot

friends_info %>% 
  left_join(friends_joyful_sad, by = c("episode","season")) %>% 
  filter(season <= 4) %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, color = Joyful)) +
  geom_jitter()

Basic scatter plot

friends_info %>% 
  left_join(friends_joyful_sad, by = c("episode","season")) %>% 
  filter(season <= 4) %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, color = Joyful)) +
  geom_jitter() +
  scale_colour_gradient(low = "#fafafa", high = "#191970", 
                        breaks = seq(0.1, 0.4, 0.1))

Basic scatter plot

friends_info %>% 
  left_join(friends_joyful_sad, by = c("episode","season")) %>% 
  filter(season <= 4) %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, color = Sad)) +
  geom_jitter() +
  scale_colour_gradient(low = "#fafafa", high = "#191970", 
                        breaks = seq(0, 0.25, 0.05))

Questions to examine

How does the relationship between viewers and IMDB rating look by:

season

predominant emotion of the episode

focus character

Today's data

friends

Today's data

Today's data

Data manipulation

friends_top_actor <- friends %>% 
  group_by(season, episode, speaker) %>%
  summarize(count = n()) %>%
  add_count(wt = count) %>%
  mutate(percent = count / n) %>%
  filter(speaker %in% c("Chandler Bing","Joey Tribbiani",
                        "Monica Geller","Phoebe Buffay",
                        "Rachel Green","Ross Geller")) %>% 
  filter(percent == max(percent)) %>% 
  select(-c(count, n, percent))

Data manipulation

Data manipulation

friends_top_actor <- friends %>% 
  group_by(season, episode, speaker) %>% 
  summarize(count = n()) %>% 
  add_count(wt = count) %>% 
  mutate(percent = count / n) %>% 
  filter(speaker %in% c("Chandler Bing","Joey Tribbiani",
                        "Monica Geller","Phoebe Buffay",
                        "Rachel Green","Ross Geller")) %>%
  filter(percent == max(percent)) %>%
  select(-c(count, n, percent))

Data manipulation

Joining data

friends_info %>% 
  left_join(friends_top_actor, by = c("episode","season")) %>%

Basic scatter plot

friends_info %>% 
  left_join(friends_top_actor, by = c("episode","season")) %>% 
  ggplot(aes(x = us_views_millions, y = imdb_rating, color = speaker)) +
  geom_jitter(size = 2)

Overplotting

When the sample size is large, overplotting can disguise trends

Techniques:
use smaller dots and/or transparency

add color by group

add jittering

add a rug plot

Overplotting

txhousing %>% 
  filter(month == 1 & listings < 20000) %>% 
  ggplot(aes(x = median, y = listings)) +
  geom_point()

Overplotting

txhousing %>% 
  filter(month == 1 & listings < 20000) %>% 
  ggplot(aes(x = median, y = listings)) +
  geom_point() +
  geom_rug(color = "purple", alpha = 0.1, size = 2)

Overplotting

When the sample size is large, overplotting can disguise trends

Techniques:
use smaller dots and/or transparency

add color by group

add jittering

add a rug plot

add a marginal distribution

Overplotting

Overplotting

When the sample size is large, overplotting can disguise trends

Techniques:
use smaller dots and/or transparency

add color by group

add jittering

add a rug plot

add a marginal distribution

create a hexbin

Overplotting

txhousing %>% 
  filter(month == 1 & listings < 20000) %>% 
  ggplot(aes(x = median, y = listings)) +
  geom_hex(bins = 30)

Overplotting

storms %>% 
  ggplot(aes(x = wind, y = pressure)) +
  geom_hex(bins = 30)

Bubble chart

Adding a third quant. variable to size

Not generally recommended!

Why?

- encoding the same type of variable (numeric) on two different scales: position and size
- hard to compare the strengths of different associations
- much easier to perceive differences when encoded by position rather than size
- hard to see small differences in size
- difficult to match scale of circle size to scale of difference

Bubble chart

penguins %>% 
  filter(species != "Gentoo") %>% 
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point()

Bubble chart

penguins %>% 
  filter(species != "Gentoo") %>% 
  ggplot(aes(x = flipper_length_mm, y = body_mass_g, 
             size = bill_length_mm)) +
  geom_point()

Bubble chart

penguins %>% 
  filter(species != "Gentoo") %>% 
  ggplot(aes(x = flipper_length_mm, y = body_mass_g, 
             size = bill_length_mm)) +
  geom_point(alpha = 0.5, color = "red") +
  scale_size(range = c(0.1, 7), breaks = c(35, 40, 45, 50, 55)) +
  theme(legend.position = "top")

Bubble chart

penguins %>% 
  filter(species != "Gentoo") %>% 
  mutate(label = case_when(flipper_length_mm == 192 & 
                             body_mass_g == 2700 ~ "Chinstrap",
                           flipper_length_mm == 184 & 
                             body_mass_g == 4650 ~ "Adelie")) %>% 
  ggplot(aes(x = flipper_length_mm, y = body_mass_g, 
             size = bill_length_mm, color = species)) +
  geom_point(alpha = 0.5) +
  scale_size(range = c(0.1, 7), breaks = c(35, 40, 45, 50, 55), 
             name = "Bill Length (mm)") +
  geom_label_repel(aes(x = flipper_length_mm, y = body_mass_g, 
                       color = species, label = label), 
                   inherit.aes = FALSE) +
  scale_color_discrete(guide = "none") +
  theme(legend.position = "top")

Bubble chart

Upcoming

Graphic critique due before midterm
Details on syllabus

Homework 2 due Tuesday June 8

Lecture 7 on Monday June 7