7 Heatmaps

This chapter demonstrates how to create heatmaps to display patterns across groups and time. We use Instagram posting data from Democratic presidential candidates during 2019-2020 to illustrate the technique.

7.1 Setup and Data

library(tidyverse)
library(lubridate)

We load Instagram posting data for US presidential candidates. This dataset contains individual posts with timestamps, engagement metrics, and account information.

insta_pres <- read_csv("data/instagram_pres_candidates_2019_to_July162020.csv")

Let’s examine the structure of the data:

glimpse(insta_pres)

Rows: 7,034
Columns: 17
$ Account                <chr> "Juli<U+00E1>n Castro", "Bill de Blasio", "Beto O'Rour~
$ `User Name`            <chr> "juliancastrotx", "billdeblasio", "betoorourke"~
$ `Followers at Posting` <chr> "43273", "5097", "828160", "2451338", "476561",~
$ Created                <chr> "2019-06-27 00:33:23 EDT", "2019-09-19 17:20:43~
$ Type                   <chr> "Video", "Photo", "Photo", "Album", "IGTV", "Vi~
$ Likes                  <dbl> 136773, 925, 350099, 350694, 22729, 222839, 170~
$ Comments               <dbl> 4313, 8667, 18820, 7506, 1727, 14914, 661, 2643~
$ Views                  <dbl> 490703, 0, 0, 0, 134223, 754254, 18292, 0, 9553~
$ URL                    <chr> "https://www.instagram.com/p/BzMy6_hHrLp/", "ht~
$ Link                   <chr> "https://www.instagram.com/p/BzMy6_hHrLp/", "ht~
$ Photo                  <chr> "https://scontent.cdninstagram.com/v/t51.2885-1~
$ Title                  <chr> NA, NA, NA, NA, "<U+201C>There is no one else like you~
$ Description            <chr> "Me llamo Juli<U+00E1>n Castro y me estoy postulando p~
$ `Image Text`           <chr> NA, NA, NA, NA, NA, NA, "We are moving from the~
$ `Sponsor Id`           <lgl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA,~
$ `Sponsor Name`         <lgl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA,~
$ `Overperforming Score` <dbl> 71.91, 61.10, 27.59, 19.86, 17.44, 17.33, 16.41~

7.2 Data Preparation

The key to creating effective heatmaps is aggregating data to the right level. For our time-based analysis, we need to extract month-year information from the posting dates.

insta_pres <- insta_pres %>%
  mutate(
    date = as.Date(Created),
    monthyear = floor_date(date, "month")
  )

We create a monthyear variable by flooring each date to the first of its month. This gives us a consistent grouping variable for aggregation.

7.3 Heatmaps with `geom_tile()`

Heatmaps display values in a matrix format where color intensity represents the magnitude of a variable. They’re excellent for showing patterns across two categorical dimensions—in our case, candidates and time periods.

7.3.1 Basic Heatmap: Selected Candidates

We start by filtering to five major candidates and counting posts per month:

insta_pres %>% 
  filter(`User Name` %in% c("elizabethwarren", "joebiden", "berniesanders", 
                            "pete.buttigieg", "amyklobuchar")) %>%
  group_by(`User Name`, monthyear) %>%
  summarize(n = n(), .groups = "drop") %>% 
  ggplot(aes(monthyear, `User Name`, fill = n)) +
  geom_tile(color = "white", size = 0.1) + 
  theme_minimal() + 
  scale_fill_gradient(low = "white", high = "purple") +
  labs(x = "", y = "Instagram account", fill = "Number\nof posts",
       subtitle = "Instagram use of selected Democratic candidates")

Key elements of this visualization:

geom_tile() creates rectangular tiles, one per candidate-month combination
color = “white” adds thin borders between tiles for visual separation
scale_fill_gradient() maps the count to a continuous color scale from white (low) to purple (high)

7.3.2 Expanding to All Candidates

To show all candidates, we need to create a meaningful ordering. Here we order candidates by their total posting volume:

# Create ordered list of accounts by total posts
account_order <- insta_pres %>%
  count(`User Name`) %>%
  arrange(desc(n)) %>%
  pull(`User Name`)

Now we can visualize all candidates:

insta_pres %>% 
  group_by(`User Name`, monthyear) %>%
  summarize(n = n(), .groups = "drop") %>% 
  mutate(account = factor(`User Name`, levels = rev(account_order))) %>% 
  ggplot(aes(monthyear, account, fill = n)) +
  geom_tile(color = "white", size = 0.1) + 
  theme_minimal() + 
  scale_fill_gradient(low = "white", high = "navy") +
  labs(x = "", y = "Instagram account", fill = "Number\nof posts",
       subtitle = "Instagram use of Democrats who ran for president in 2019-2020")

7.3.3 Categorical Color Scales

Sometimes a continuous color scale can be difficult to interpret, especially when the distribution is skewed. Converting counts to categories can make patterns more readable.

library(RColorBrewer)

We use cut() to bin the count into meaningful categories:

insta_pres %>% 
  filter(monthyear >= "2019-01-01") %>%
  group_by(`User Name`, monthyear) %>%
  summarize(n = n(), .groups = "drop") %>% 
  mutate(
    countfactor = cut(n, 
                      breaks = c(-1, 0, 5, 10, 20, 40, 60, 80, Inf),
                      labels = c("0", "1-5", "6-10", "11-20", "21-40", "41-60", "61-80", ">80"))
  ) %>%
  mutate(account = factor(`User Name`, levels = rev(account_order))) %>% 
  ggplot(aes(monthyear, account, fill = countfactor)) +
  geom_tile(color = "white", size = 0.1) + 
  theme_minimal() +
  scale_fill_manual(values = brewer.pal(8, "Purples"), na.value = "grey90") +
  labs(x = "", y = "Instagram account", fill = "Number\nof posts",
       subtitle = "Instagram use of Democrats who ran for president in 2019-2020")

Key differences in this approach:

cut() creates bins with custom breakpoints—we chose bins that match how people naturally think about posting frequency
scale_fill_manual() with brewer.pal() provides a sequential color palette designed for ordered categories
na.value = “grey90” handles any missing month-candidate combinations

7.4 Summary

This chapter covered heatmaps using geom_tile():

Best for showing values across two categorical dimensions (e.g., candidates and time periods)
Use continuous color scales (scale_fill_gradient()) for fine-grained differences
Use categorical color scales (via cut()) for easier interpretation of ranges
Order categories meaningfully rather than alphabetically to reveal patterns
Add white borders between tiles for visual separation